Experimental approaches for measuring single-cell gene expression can observe each cell at only one time point, requiring computational approaches for reconstructing the dynamics of gene expression during cell fate transitions. RNA velocity is a promising computational approach for this problem, but existing inference methods fail to capture key aspects of real data, limiting their utility. To address these limitations, we developed VeloVAE, a Bayesian model for RNA velocity inference. VeloVAE uses variational Bayesian inference to estimate the posterior distribution of latent time, latent cell state, and kinetic rate parameters for each cell. Our approach can incorporate prior distributions on rate parameters and time points; model lineage bifurcations using branching differential equations; and directly model discrete count data. We show that VeloVAE significantly outperforms previous approaches in terms of data fit, accuracy of inferred differentiation directions, and transcription rate estimation. These improvements allow VeloVAE to accurately model gene expression dynamics in complex biological systems, including hematopoiesis, induced pluripotent stem cell reprogramming, the developing mouse brain, and the entire mouse embryo. We find that the latent time automatically inferred using all cells can even outperform pseudotime inferred using manually chosen cell subsets and root cells. Our work provides important new tools for modeling sequential changes in gene expression from single-cell expression data.

Rapidly evolving wireless channel conditions and communication standards demand adaptable forward error correction (FEC) decoders. Existing rigid architectures, designed for a single standard, exhibit limited adaptability in terms of throughput and/or coding gain, hindering the timely deployment of new applications. To overcome these limitations, we propose QFEC (quad-mode FEC decoder), a unified and highly configurable FEC decoder. QFEC enables a wide range of throughput vs. coding gain tradeoffs across QC-LDPC, Turbo, Polar, and Convolutional codes (CC) by providing full configurability for existing standards and proprietary systems. This ensures communication reliability under varying channel conditions and seamless support for both legacy and emerging communication protocols. Our hardware-unified approach leverages a shared memory and computation unit architecture that exploits the inherent commonalities in the iterative message-passing dataflow of all four code types. We attain outstanding flexibility at high data rates through an innovative combination of a fully customizable interconnect and a multi-mode computation datapath. The QFEC chip, fabricated in GF 12 nm FinFET technology, achieves 9.97 Gb/s throughput for the Optical Communication Terminal (OCT) standard and 6.52 Gb/s for 5G BG1, while consuming normalized energy efficiency (NEE) of 1.04 pJ/b and 1.53 pJ/b, respectively. This design can reach a maximum of 25.4 Gb/s using a proprietary QC-LDPC configuration. This design significantly surpasses existing solutions in flexibility by offering the broadest support for standards and parameters with a unified, efficient architecture. To the best of our knowledge, this is the first chip implementation of a fully flexible quad-mode FEC decoder for QC-LDPC, Polar, Turbo, CC codes.

We introduce NBLoc, a narrowband frequency hopping, long-range localization system designed for low-power Internet of Things (IoT) devices. Traditional high-accuracy localization systems typically require wide-bandwidth and power-demanding radio frequency (RF) circuits, leading to limitations such as short operational range and high power consumption to achieve decimeter-level accuracy. NBLoc overcomes these challenges by using narrowband symbols with a frequency-hopping mechanism across a wide bandwidth, enabling the localization of low-power tags over large areas. NBLoc features a novel custom-designed RF analog frontend (AFE) integrated circuit (IC), that eliminates the need for a conventional phase-locked loop, significantly reducing the cost and power consumption of the receiving tag. This advancement is enabled by NBLoc's thoughtful waveform design and specialized signal processing algorithms, which mitigate phase noise and uncertainty. Compared to previous solutions, NBLoc achieves lower power consumption and extended operational range due to its narrowband symbols while maintaining high localization accuracy by leveraging a 100 MHz localization bandwidth through frequency hopping. In NBLoc, system anchors transmit narrowband orthogonal symbols, hopping across the localization bandwidth in a predetermined pattern known to the tag. The tag, equipped with the custom low-power RF AFE IC, dynamically tunes its local oscillator (LO) frequency to match the hop pattern and capture these symbols, which are then used to estimate the channel impulse response (CIR). The tag calculates the time difference of arrival (TDOA) for each anchor pair from the CIRs, and determines its 2D location via multilateration. The system was implemented and tested using the low-power RF AFE IC in both line-of-sight (LOS) and non-line-of-sight (NLOS) environments, achieving decimeter-level accuracy across areas as large as 269 × 125 m$^{2}$.

This paper presents ParaBase, a configurable baseband processing architecture that efficiently handles parallel sample streams targeting ultra-wide bandwidth inter-satellite optical communications for the target data rates exceeding 100 Gbps. We propose a parallelogram-style systolic accelerator, specifically designed for parallel processing for correlation kernels, preserving the hardware efficiency inherent in the systolic-array architecture. ParaBase supports end-to-end baseband processing through a set of heterogeneous configurable accelerators customized to their respective parallel processing requirements. It outperforms the SIMD-style architecture and surpasses previous baseband processors by ~7.0X in terms of energy efficiency for FIR filtering. The overall architecture reports energy efficiency of up to 2.9 TOPS/W and 121.8 Gbits/J while supporting a wide range of data rates from 1~128 Gbps via fast reconfiguration for various modulation schemes.

Stream processing, which involves real-time computation of data as it is created or received, is vital for various applications, specifically wireless communication. The evolving protocols, the requirement for high-throughput, and the challenges of handling diverse processing patterns make it demanding. Traditional platforms grapple with meeting real-time throughput and latency requirements due to large data volume, sequential and indeterministic data arrival, and variable data rates, leading to inefficiencies in memory access and parallel processing. We present Canalis, a throughput-optimized framework designed to address these challenges, ensuring high-performance while achieving low energy consumption. Canalis is a hardware-software co-designed system. It includes a programmable spatial architecture, Flux Stream Processing Unit (FluxSPU), proposed by this work to enhance data throughput and energy efficiency. FluxSPU is accompanied by a software stack that eases the programming process. We evaluated Canalis with eight distinct benchmarks. When compared to CPU and GPU in mobile SoC to demonstrate the effectiveness of domain specialization, Canalis achieves an average speedup of 13.4 \(\times\) and 6.6 \(\times\) , and energy savings of 189.8 \(\times\) and 283.9 \(\times\) , respectively. In contrast to equivalent ASICs of the benchmarks, the average energy overhead of Canalis is within 2.4 \(\times\) , successfully maintaining generalizations without incurring significant overhead.

Recent advances in model pruning have enabled sparsity-aware deep neural network accelerators that improve the energy-efficiency and performance of inference tasks. We introduce SONA, a novel transform-domain neural network accelerator in which convolution operations are replaced by element-wise multiplications with sparse-orthogonal weights. SONA employs an output stationary dataflow coupled with an energy-efficient memory organization to reduce the overhead of sparse-orthogonal transform-domain kernels that are concurrently processed without any conflicts. Weights in SONA are non-uniformly quantized with bit-sparse canonical-signed-digit representations to reduce multiplications to simple additions. Moreover, for sparse fully-connected layers (FCLs), SONA introduces column-based-block structured pruning, which is integrated into the same architecture that maintains full multiply-and-accumulate (MAC) array utilization. Compared to prior dense and sparse neural networks accelerators, SONA can reduce inference energy by$5.1\times$and$2.4 \times$and increase performance by$5.2\times$and$2.1\times$, respectively, for convolution layers. For sparse FCLs, SONA can reduce inference energy by$2.4\times$and increase performance by$2\times$compared to prior work.

The combination of pre-trained models and task-specific fine-tuning schemes, such as BERT, has achieved great success in various natural language processing (NLP) tasks. However, the large memory and computation costs of such models make it challenging to deploy them in edge devices. Moreover, in real-world applications like chatbots, multiple NLP tasks need to be processed together to achieve higher response credibility. Running multiple NLP tasks with specialized models for each task increases the latency and memory cost latency linearly with the number of tasks. Though there have been recent works on parameter-shared tuning that aim to reduce the total parameter size by partially sharing weights among multiple tasks, computation remains intensive and redundant despite different tasks using the same input. In this work, we identify that a significant portion of activations and weights can be reused among different tasks, to reduce cost and latency for efficient multi-task NLP. Specifically, we propose TaskFusion, an efficient transfer learning software-hardware co-design that exploits delta sparsity in both weights and activations to boost data sharing among tasks. For training, TaskFusion uses ℓ1 regularization on delta activation to learn inter-task data redundancies. A novel hardware-aware sub-task inference algorithm is proposed to exploit the dual delta sparsity. We then designed a dedicated heterogeneous architecture to accelerate multi-task inference with an optimized scheduling to increase hardware utilization and reduce off-chip memory access. Extensive experiments demonstrate that TaskFusion can reduce the number of floating point operations (FLOPs) by over 73% in multi-task NLP with negligible accuracy loss, while adding a new task at the cost of only < 2% parameter size increase. With the proposed architecture and optimized scheduling, Task-Fusion can achieve 1.48--2.43× performance and 1.62--3.77× energy efficiency than those using state-of-the-art single-task accelerators for multi-task NLP applications.

Genomics is playing an important role in transforming healthcare. Genetic data, however, is being produced at a rate that far outpaces Moore's Law. Many efforts have been made to accelerate genomics kernels on modern commodity hardware such as CPUs and GPUs, as well as custom accelerators (ASICs) for specific genomics kernels. While ASICs provide higher performance and energy efficiency than general-purpose hardware, they incur a high hardware design cost. Moreover, in order to extract the best performance, ASICs tend to have significantly different architectures for different kernels. The divergence of ASIC designs makes it difficult to run commonly used modern sequencing analysis pipelines due to software integration and programming challenges.

Daily tracking of small objects or animals anywhere on earth for long time-periods is a long sought-after goal. Recently, the emergence of low earth orbit (LEO) satellites offers a unique pathway to achieve this goal. However, to date, LEO trackers have not achieved cm-size. While the integrated chip can be readily scaled to sub-cm size, the size of trackers remains limited by their battery and antenna size. To address these two fundamental size limiting factors, this paper presents a LEO satellite localization system that is specifically optimized to reduce antenna size and transmit power, thereby reducing battery size. To reduce power, a new cooperative waveform is designed which enhances the localization accuracy, combined with an increased packet length to enable low transmit power while maintaining packet energy. However, this long packet length introduces a intra-packet Doppler shift which we address by proposing a localization algorithm that includes a Doppler shift correction. The final result is a 50 kHz periodic BPSK signal with 23 dBm equivalent isotropic radiation power (EIRP), and 120 ms packet length (> 10 × longer than conventional), at a 60 s interval. The proposed solution enables 7 months operation on a 2.5 × 1.2 cm LiPo battery within a North American search area. To address the antenna size, the optimal transmit frequency was studied and a 1 cm loop antenna with 65% radiation efficiency was designed with internal matching to 50 Ohm. Using the proposed techniques, three satellite flyover experiments were performed to confirm the accuracy of the proposed tracking system and localization algorithms using a USRP-X310, a custom 1 cm-size antenna, and a commercial satellite cluster. The measured average localization error is 320 - 840 m depending on satellite trajectories, demonstrating an improved accuracy in real life measurements compared to prior art with experimental result while simultaneously achieving 15 -- 26 dB lower transmit power and > 3 × lower packet energy.

Despite recent improvements in nanopore basecalling accuracy, germline variant calling of small insertions and deletions (INDELs) remains poor. Although precision and recall for single nucleotide polymorphisms (SNPs) now exceeds 99.5%, INDEL recall remains below 80% for standard R9.4.1 flow cells. We show that read phasing and realignment can recover a significant portion of false negative INDELs. In particular, we extend Needleman-Wunsch affine gap alignment by introducing new gap penalties for more accurately aligning repeated n-polymer sequences such as homopolymers ([Formula: see text]) and tandem repeats ([Formula: see text]). At the same precision, haplotype phasing improves INDEL recall from 63.76 to [Formula: see text] and nPoRe realignment improves it further to [Formula: see text].

With generational gains from transistor scaling, GPUs have been able to accelerate traditional computation-intensive workloads. But with the obsolescence of Moore's Law, single GPU systems are no longer able to satisfy the computational and memory requirements of emerging workloads. To remedy this, prior works have proposed tightly-coupled multi-GPU systems. However, multi-GPU systems are hampered from efficiently utilizing their compute resources due to the Non-Uniform Memory Access (NUMA) bottleneck. In this paper, we propose DualOpt, a lightweight hardware-only solution that reduces the remote memory access latency by delivering optimizations catered to a workload's locality profile. DualOpt uses the spatio-temporal locality of remote memory accesses as a metric to classify workloads as cache insensitive and cache-friendly. Cache insensitive workloads exhibit low spatio-temporal locality, while cache-friendly workloads have ample locality that is not exploited well by the conventional cache subsystem of the GPU. For cache insensitive workloads, DualOpt proposes a fine-granularity transfer of remote data instead of the conventional cache line transfer. These remote data are then coalesced so as to efficiently utilize inter-GPU bandwidth. For cache-friendly workloads, DualOpt adds a remote-only cache that can exploit locality in remote accesses. Finally, a decision engine automatically identifies the class of workload and delivers the corresponding optimization, which improves overall performance by 2.5× on a 4-GPU system, with a small hardware overhead of 0.032%.

Systolic arrays have been successful to accelerate dense linear algebra for deep neural networks (DNNs), but cannot handle sparse computations efficiently. Though early attempts have been made to perform sparse matrix operations on weight-pruned DNNs, handling highly sparse matrices with skewed nonzero distribution commonly seen in real-world graph analytics remains challenging. In this paper, we propose FlexTPU framework to repurpose tensor processing units (TPUs) to execute sparse matrix-vector operations (SpMV). First, we propose a lightweight Z-shape mapping of sparse matrices onto the systolic array to eliminate the processing of zeros as much as possible, regardless of the sparsity and nonzero distribution. On top of the mapping, we devise an SpMV dataflow executed by an array of PEs, which are a slightly modified version of the conventional TPU PE. Second, in contrast to the excess preprocessing mandatory for prior attempts, the Z-shape mapping facilitates on-the-fly matrix condensing from the widely-used compressed sparse matrix (e.g. CSR) representation. This is accomplished by a proposed sparse data loader that includes an on-chip row decoder and parallel nonzero loaders. We evaluate FlexTPU on a broad set of synthetic and real-world sparse matrices. The experimental result shows that FlexTPU achieves 3.55× speedup and 3.27× energy saving over a state-of-the-art design, Sparse-TPU. It performs even better on sparse matrices with power-law distributions. Compared to state-of-the-art library implementations on a CPU and a GPU, FlexTPU also achieves an average speedup of 2.4× and 4.3×, and energy saving of 130.4× and 495.3×, respectively. FlexTPU is also evaluated against a recent re configurable (chip multi-processor) CMP machine, Transmuter. FlexTPU outperforms Transmuter by achieving 5.12× speedup and 2.65× energy saving.

A key problem in computational biology is discovering the gene expression changes that regulate cell fate transitions, in which one cell type turns into another. However, each individual cell cannot be tracked longitudinally, and cells at the same point in real time may be at different stages of the transition process. This can be viewed as a problem of learning the behavior of a dynamical system from observations whose times are unknown. Additionally, a single progenitor cell type often bifurcates into multiple child cell types, further complicating the problem of modeling the dynamics. To address this problem, we developed an approach called variational mixtures of ordinary differential equations. By using a simple family of ODEs informed by the biochemistry of gene expression to constrain the likelihood of a deep generative model, we can simultaneously infer the latent time and latent state of each cell and predict its future gene expression state. The model can be interpreted as a mixture of ODEs whose parameters vary continuously across a latent space of cell states. Our approach dramatically improves data fit, latent time inference, and future cell state estimation of single-cell gene expression data compared to previous approaches.

Near-memory processing has been extensively studied to optimize memory intensive workloads. However, none of the proposed designs address sparse matrix transposition, an important building block in sparse linear algebra applications. Prior work shows that sparse matrix transposition does not scale as well as other sparse primitives such as sparse matrix vector multiplication (SpMV) and hence has become a growing bottleneck in common applications. Sparse matrix transposition is highly memory intensive but low in computational intensity, making it a promising candidate for near-memory processing. In this work, we propose MeNDA, a scalable near-DRAM multi-way merge accelerator that eliminates the off-chip memory interface bottleneck and exposes the high internal memory bandwidth to improve performance and reduce energy consumption for sparse matrix transposition. MeNDA adopts a merge sort based algorithm, exploiting spatial locality, and proposes a near-memory processing unit (PU) featuring a high-performance hardware merge tree. Because of the wide application of merge sort in sparse linear algebra, MeNDA is an extensible solution that can be easily adapted to support other sparse primitives such as SpMV. Techniques including seamless back-to-back merge sort, stall reducing prefetching and request coalescing are further explored to take full advantage of the increased system memory bandwidth. Compared to two state-of-the-art implementations of sparse matrix transposition on a CPU and a sparse library on a GPU, MeNDA is able to achieve a speedup of 19.1X, 12.0X, and 7.7x, respectively. MeNDA also shows an efficiency gain of 3.8x over a recent SpMV accelerator integrated with HBM. Incurring a power consumption of only 78.6 mW, a MeNDA PU can be easily accommodated by commodity DIMMs.

Graph Pattern Mining (GPM) algorithms mine structural patterns in graphs. The performance of GPM workloads is bottlenecked by control flow and memory stalls. This is because of data-dependent branches used in set intersection and difference operations that dominate the execution time.

RF system development is traditionally constrained by a restrictive trade-off between power efficiency and programmatic flexibility. We outline a path towards achieving both, thereby enabling a range of new system concepts that better utilize limited resources. As an example, for many future applications, we consider RF convergence – reusing the same spectrum and waveforms to achieve multiple distributed system functions and goals, simultaneously. To enable this next step in processing, we develop a novel framework that includes both software and the system-on-chip (SoC) design.

Convolutional neural networks (CNNs) are built with convolution layers which account for most of their computation time. The differences in the convolution kernel types (2D, point-wise, depth-wise), and input sizes lead to significant differences in their computation and memory demands. In this work, we exploit run-time reconfiguration to adapt to the differences in the characteristics of different convolution kernels on a low-power reconfigurable architecture, Transmuter. The architecture consists of light-weight cores interconnected by caches and crossbars that support run-time reconfiguration between different cache modes - shared or private, different dataflow modes - systolic or parallel, and different computation mapping schemes. To achieve run-time reconfiguration, we propose a decision-tree-based engine that selects the optimal Transmuter configuration at a low cost. The proposed method is evaluated on commonly-used CNN models such as ResNetl8, VGGII, AlexNet and MobileNetV3. Simulation results show that run-time reconfiguration helps improve the energy efficiency of Transmuter in the range of 3.1$\times-13.7\times$ across all networks.

Forward error correction (FEC) is a critical component in communication systems as the errors induced by noisy channels can be corrected using the redundancy in the coded message. This paper introduces a novel multi-mode FEC decoder accelerator that can decode Turbo, LDPC, and Polar codes using a unified architecture. The proposed design explores the similarities in these codes to enable energy efficient decoding with minimal overhead in the total area of the unified architecture. Moreover, the proposed design is highly reconfigurable to support various existing and future FEC standards including 3GPP LTE/5G, and IEEE 802.11n WiFi. Implemented in GF 12nm FinFET technology, the design occupies 8.47mm2 of chip area attaining 25% logic and 49% memory area savings compared to a collection of single-mode designs. Running at 250MHz and 0.8V, the decoder achieves per-iteration throughput and energy efficiency of 690Mb/s and 44pJ/b for Turbo; 740Mb/s and 27.4pJ/b for LDPC; and 950Mb/s and 45.8pJ/b for Polar.

A variety of complex systems, including social and communication networks, financial markets, biology, and neuroscience are modeled using temporal graphs that contain a set of nodes and directed timestamped edges. Temporal motifs in temporal graphs are generalized from subgraph patterns in static graphs in that they also account for edge ordering and time duration, in addition to the graph structure. Mining temporal motifs is a fundamental problem used in several application domains. However, existing software frameworks offer suboptimal performance due to high algorithmic complexity and irregular memory accesses of temporal motif mining.This paper presents $\mathsf{Mint}$—a novel accelerator architecture and a programming model for mining temporal motifs efficiently. We first divide this workload into three fundamental tasks: search, book-keeping, and backtracking. Based on this, we propose a task-centric programming model that enables decoupled, asynchronous execution. This model unlocks massive opportunities for parallelism, and allows storing task context information on-chip. To best utilize the proposed programming model, we design a domain-specific hardware accelerator using its data path and memory subsystem design to cater to the unique workload characteristics of temporal motif mining. To further improve performance, we propose a novel optimization called search index memoization that significantly reduces memory traffic. We comprehensively compare the performance of $\mathsf{Mint}$ with state-of-the-art temporal motif mining software frameworks (both approximate and exact) running on both CPU and GPU, and show $9\times-2576\times$ benefit in performance.

Read alignment is a time-consuming step in genome sequencing analysis. The most widely used software for read alignment, BWA-MEM, and the recently published faster version BWA-MEM2 are based on the seed-and-extend paradigm for read alignment. The seeding step of read alignment is a major bottleneck contributing ~40% to the overall execution time of BWA-MEM2 when aligning whole human genome reads from the Platinum Genomes dataset. This is because both BWA-MEM and BWA-MEM2 use a compressed index structure called the FMD-Index, which results in high bandwidth requirements, primarily due to its character-by-character processing of reads. For instance, to seed each read (101 DNA base-pairs stored in 37.8 bytes), the FMD-Index solution in BWA-MEM2 requires ~68.5 KB of index data.We propose a novel indexing data structure named Enumerated Radix Tree (ERT) and design a custom seeding accelerator based on it. ERT improves bandwidth efficiency of BWA-MEM2 by 4.5× while guaranteeing 100% identical output to the original software, and still fitting in 64 GB DRAM. Overall, the proposed seeding accelerator implemented on AWS F1 FPGA (f1.4xlarge) improves seeding throughput of BWA-MEM2 by 3.3×. When combined with seed-extension accelerators, we observe a 2.1× improvement in overall read alignment throughput over BWA-MEM2. The software implementation of ERT is integrated into BWA-MEM2 (ert branch: https://github.com/bwa-mem2/bwa-mem2/tree/ert) and is open sourced for the benefit of the research community.

Over the last decade, advances in high-throughput sequencing and the availability of portable sequencers have enabled fast and cheap access to genetic data. For a given sample, sequencers typically output fragments of the DNA in the sample. Depending on the sequencing technology, the fragments range from a length of 150-250 at high accuracy to lengths in few tens of thousands but at much lower accuracy. Sequencing data is now being produced at a rate that far outpaces Moore's law and poses significant computational challenges on commodity hardware. To meet this demand, software tools have been extensively redesigned and new algorithms and custom hardware have been developed to deal with the diversity in sequencing data. However, a standard set of benchmarks that captures the diverse behaviors of these recent algorithms and can facilitate future architectural exploration is lacking. To that end, we present the GenomicsBench benchmark suite which contains 12 computationally intensive data-parallel kernels drawn from popular bioinformatics software tools. It covers the major steps in short and long-read genome sequence analysis pipelines such as basecalling, sequence mapping, de-novo assembly, variant calling and polishing. We observe that while these genomics kernels have abundant data level parallelism, it is often hard to exploit on commodity processors because of input-dependent irregularities. We also perform a detailed microarchitectural characterization of these kernels and identify their bottlenecks. GenomicsBench includes parallel versions of the source code with CPU and GPU implementations as applicable along with representative input datasets of two sizes - small and large.

The MinION is a recent-to-market handheld nanopore sequencer. It can be used to determine the whole genome of a target virus in a biological sample. Its Read Until feature allows us to skip sequencing a majority of non-target reads (DNA/RNA fragments), which constitutes more than 99% of all reads in a typical sample. However, it does not have any on-board computing, which significantly limits its portability.

Each fall, millions of monarch butterflies across the northern US and Canada migrate up to 4,000 km to overwinter in the exact same cluster of mountain peaks in central Mexico. To track monarchs precisely and study their navigation, a monarch tracker must obtain daily localization of the butterfly as it progresses on its 3-month journey. And, the tracker must perform this task while having a weight in the tens of milligram (mg) and measuring a few millimeters (mm) in size to avoid interfering with monarch's flight. This paper proposes mSAIL, 8 × 8 × 2.6 mm and 62 mg embedded system for monarch migration tracking, constructed using 8 prior custom-designed ICs providing solar energy harvesting, an ultra-low power processor, light/temperature sensors, power management, and a wireless transceiver, all integrated and 3D stacked on a micro PCB with an 8 × 8 mm printed antenna. The proposed system is designed to record and compress light and temperature data during the migration path while harvesting solar energy for energy autonomy, and wirelessly transmit the data at the overwintering site in Mexico, from which the daily location of the butterfly can be estimated using a deep learning-based localization algorithm. A 2-day trial experiment of mSAIL attached on a live butterfly in an outdoor botanical garden demonstrates the feasibility of individual butterfly localization and tracking.

With the end of Dennard scaling and Moore's law, it is becoming increasingly difficult to build hardware for emerging applications that meet power and performance targets, while remaining flexible and programmable for end users. This is particularly true for domains that have frequently changing algorithms and applications involving mixed sparse/dense data structures, such as those in machine learning and graph analytics. To overcome this, we present a flexible accelerator called Transmuter, in a novel effort to bridge the gap between General-Purpose Processors (GPPs) and Application-Specific Integrated Circuits (ASICs). Transmuter adapts to changing kernel characteristics, such as data reuse and control divergence, through the ability to reconfigure the on-chip memory type, resource sharing and dataflow at run-time within a short latency. This is facilitated by a fabric of light-weight cores connected to a network of reconfigurable caches and crossbars. Transmuter addresses a rapidly growing set of algorithms exhibiting dynamic data movement patterns, irregularity, and sparsity, while delivering GPU-like efficiencies for traditional dense applications. Finally, in order to support programmability and ease-of-adoption, we prototype a software stack composed of low-level runtime routines, and a high-level language library called TransPy, that cater to expert programmers and end-users, respectively.

Much of the recent work on domain-specific architectures has focused on bridging the gap between performance/efficiency and programmability. We consider one such example architecture, Transformer, consisting of light-weight cores interconnected by caches and crossbars that supports run-time reconfiguration between shared and private cache mode operations. We present customized implementation of a select set of linear algebra kernels, namely, triangular matrix solver, LU decomposition, QR decomposition and matrix in-version, on Transformer. The performance of the kernel algorithms is evaluated with respect to execution time and energy efficiency. Our study shows that each kernel achieves high performance for a certain cache mode and that this cache mode can change when the matrix size changes, making a case for run-time reconfiguration.

Recent work on neural network architectures has focused on bridging the gap between performance/efficiency and programmability. We consider implementations of three popular neural networks, ResNet, AlexNet and ASGD weight-dropped Recurrent Neural Network (AWD RNN) on a low power programmable architecture, Transformer. The architecture consists of light-weight cores interconnected by caches and crossbars that support run-time reconfiguration between shared and private cache mode operations. We present efficient implementations of key neural network kernels and evaluate the performance of each kernel when operating in different cache modes. The best-performing cache modes are then used in the implementation of the end-to-end network. Simulation results show superior performance with ResNet, AlexNet and AWD RNN achieving 188.19 GOPS/W, 150.53 GOPS/W and 120.68 GOPS/W, respectively, in the 14 nm technology node.

Innovations in genome sequencing techniques are enabling remarkably fast and low cost production of raw genome data. As Moore's law tapers off, bottlenecks in genome sequencing are shifting to computational resources for mapping reads to reference DNA. This paper presents SeedEx, a read-alignment accelerator focused on the seed-extension step. SeedEx is based on the observation that only a small fraction of reads require large edit distance for alignment, hence an area efficient narrow-band seed-extension accelerator can suffice in practice. However, due to the highly error-sensitive nature of genomic workloads, guaranteeing optimality of alignment result is of cardinal importance. Towards this end, we propose a speculation-and-test based framework by using strict but powerful optimality checking mechanisms. We demonstrate SeedEx by an implementation on a cloud FPGA. SeedEx achieves 6.0× iso-area throughput speedup when compared to a banded Smith-Waterman baseline, and achieving 43.9 M seed extentions/s on AWS f1.2xlarge instance. Integration with BWA-MEM2 improves the execution time by 2.3×.

We present the world's first autonomous mixed-signal SoC framework, driven entirely by user constraints, along with a suite of automated generators for analog blocks. The process-agnostic framework takes high-level user intent as inputs to generate optimized and fully verified analog blocks using a cell-based design methodology. Our approach is highly scalable and silicon-proven by an SoC prototype which includes 2 PLLs, 3 LDOs, 1 SRAM, and 2 temperature sensors fully integrated with a processor in a 65nm CMOS process. The physical design of all blocks, including analog, is achieved using optimized synthesis and APR flows in commercially available tools. The framework is portable across different processes and requires no-human-in-the-Ioop, dramatically accelerating design time.

The Internet of Things (IoT) is a rapidly evolving application space. One of the fascinating new fields in IoT research is mm-scale sensors, which make up the Internet of Tiny Things (IoT2). With their miniature size, these systems are poised to open up a myriad of new application domains. Enabled by the unique characteristics of cyber-physical systems and recent advances in low-power design and bare-die 3D chip stacking, mm-scale sensors are rapidly becoming a reality. In this paper, we will survey the challenges and solutions to 3D-stacked mm-scale design, highlighting low-power circuit issues ranging from low-power SRAM and miniature neural network accelerators to radio communication protocols and analog interfaces. We will discuss system-level challenges and illustrate several complete systems and their merging application spaces.

Accurate, low-latency, and energy-efficient optical flow estimation is a fundamental kernel function to enable several real-time vision applications on mobile platforms. This paper presents neighbor-guided semi-global matching (NG-fSGM), a new low-complexity optical flow algorithm tailored for low-power mobile applications. NG-fSGM obtains high accuracy optical flow by aggregating local matching costs over a semiglobal region, successfully resolving local ambiguity in textureless and occluded regions. The proposed NG-fSGM aggressively prunes the search space based on neighboring pixels' information to significantly lower the algorithm complexity from the original fSGM. As a result, NG-fSGM achieves 17.9× reduction in the number of computations and 8.37× reduction in memory space compared to the original fSGM without compromising its algorithm accuracy. A multicore architecture for NG-fSGM is implemented in hardware to quantify algorithm complexity and power consumption. The proposed architecture realizes NG-fSGM with overlapping blocks processed in parallel to enhance throughput and to lower power consumption. The eightcore architecture achieves 20 M pixel/s (66 frames/s for VGA) throughput with 9.6 mm2area at 679.2-mW power consumption in 28-nm node.

An energy efficient State-of-Charge (SOC) indication algorithm and integrated system for small IoT batteries are introduced in this paper. The system is implemented in a 180-nm CMOS technology. Based on a key finding that small Li-ion batteries exhibit a linear dependence between battery voltage and load current, we propose an instantaneous linear extrapolation (ILE) algorithm and circuit allowing on-demand estimation of SOC. Power consumption is 42nW and maximum SOC indication error is 1.7%.

This paper presents a new energy-efficient ring oscillator collapse-based comparator, which is called edge-pursuit comparator (EPC) and demonstrated it in a 15-bit SAR ADC. The comparator automatically adjusts the performance according to its input difference without any control, eliminating unnecessary energy spent on coarse comparisons. The employed SAR ADC supplements a 10-bit differential main CDAC with a 5-bit common-mode CDAC which uses common to differential gain tuning to improves linearity by reducing the effect of switch parasitic capacitance. A test chip fabricated in 40nm CMOS shows 74.12 dB SNDR and 173.4 dB FOMs. The comparator consumes 104 nW with the full ADC consuming 1.17 μW.

Sparse matrices are widely used in graph and data analytics, machine learning, engineering and scientific applications. This paper describes and analyzes OuterSPACE, an accelerator targeted at applications that involve large sparse matrices. OuterSPACE is a highly-scalable, energy-efficient, reconfigurable design, consisting of massively parallel Single Program, Multiple Data (SPMD)-style processing units, distributed memories, high-speed crossbars and High Bandwidth Memory (HBM). We identify redundant memory accesses to non-zeros as a key bottleneck in traditional sparse matrix-matrix multiplication algorithms. To ameliorate this, we implement an outer product based matrix multiplication technique that eliminates redundant accesses by decoupling multiplication from accumulation. We demonstrate that traditional architectures, due to limitations in their memory hierarchies and ability to harness parallelism in the algorithm, are unable to take advantage of this reduction without incurring significant overheads. OuterSPACE is designed to specifically overcome these challenges. We simulate the key components of our architecture using gem5 on a diverse set of matrices from the University of Florida's SuiteSparse collection and the Stanford Network Analysis Project and show a mean speedup of 7.9× over Intel Math Kernel Library on a Xeon CPU, 13.0× against cuSPARSE and 14.0× against CUSP when run on an NVIDIA K40 GPU, while achieving an average throughput of 2.9 GFLOPS within a 24 W power budget in an area of 87 mm2.

This paper presents the Neural Cache architecture, which re-purposes cache structures to transform them into massively parallel compute units capable of running inferences for Deep Neural Networks. Techniques to do in-situ arithmetic in SRAM arrays, create efficient data mapping and reducing data movement are proposed. The Neural Cache architecture is capable of fully executing convolutional, fully connected, and pooling layers in-cache. The proposed architecture also supports quantization in-cache. Our experimental results show that the proposed architecture can improve inference latency by 8.3× over state-of-art multi-core CPU (Xeon E5), 7.7× over server class GPU (Titan Xp), for Inception v3 model. Neural Cache improves inference throughput by 12.4× over CPU (2.2× over GPU), while reducing power consumption by 50% over CPU (53% over GPU).

Genomics can transform health-care through precision medicine. Plummeting sequencing costs would soon make genome testing affordable to the masses. Compute efficiency, however, has to improve by orders of magnitude to sequence and analyze the raw genome data. Sequencing software used today can take several hundreds to thousands of CPU hours to align reads to a reference sequence. This paper presents GenAx, an accelerator for read alignment, a time-consuming step in genome sequencing. It consists of a seeding and seed-extension accelerator. The latter is based on an innovative automata design that was designed from the ground-up to enable hardware acceleration. Unlike conventional Levenshtein automata, it is string independent and scales quadratically with edit distance, instead of string length. It supports critical features commonly used in sequencing such as affine gap scoring and traceback. GenAx provides a throughput of 4,058K reads/s for Illumina 101 bp reads. GenAx achieves 31.7× speedup over the standard BWA-MEM sequence aligner running on a 56-thread dualsocket 14-core Xeon E5 server processor, while reducing power consumption by 12× and area by 5.6×.

We introduce Harmonium , a novel ultra wideband (UWB) RF localization architecture that achieves decimeter-scale accuracy indoors. Harmonium strikes a balance between tag simplicity and processing complexity to provide fast and accurate indoor location estimates. Harmonium uses only commodity components and consists of a small, inexpensive, lightweight, and FCC-compliant UWB transmitter or tag , fixed infrastructure anchors with known locations, and centralized processing that calculates the tag’s position. Anchors employ a new frequency-stepped narrowband receiver architecture that rejects narrowband interferers and extracts high-resolution timing information without the cost or complexity of traditional UWB approaches. In a complex indoor environment, 90% of position estimates obtained with Harmonium exhibit less than 31 cm of error with an average of 9 cm of inter-sample noise. In non-line-of-sight conditions (i.e., through-wall), 90% of position error is less than 42 cm. The tag draws 75 mW when actively transmitting, or 3.9 mJ per location fix at the 19 Hz update rate. Tags weigh 3 g and cost $4.50 USD at modest volumes. Furthermore, VLSI-based design concepts are identified for a simple, low-power realization of the Harmonium tag to offer a roadmap for the realization of Harmonium concepts in future integrated systems. Harmonium introduces a new design point for indoor localization and enables localization of small, fast objects such as micro quadrotors, devices previously restricted to expensive optical motion capture systems.

We explore SRAM-based last-level caches (LLCs) as a substrate for automata processing that are faster and integrated on processor dies. We propose and evaluate two architectures and mapping policies, one optimized for performance and the other optimized for space, across a set of 20 diverse benchmarks from ANMLZoo [5] and Regex [1] suites. We also demonstrate acceleration of parsing activities in browser front-end as a case study where FSA computations are a bottleneck taking up to 40% of the loading time of web pages [3]. The performance optimized and space optimized designs provide a speedup of 15× and 9× over Micron's AP respectively.

This paper presents the Compute Cache architecture that enables in-place computation in caches. Compute Caches uses emerging bit-line SRAM circuit technology to re-purpose existing cache elements and transforms them into active very large vector computational units. Also, it significantly reduces the overheads in moving data between different levels in the cache hierarchy. Solutions to satisfy new constraints imposed by Compute Caches such as operand locality are discussed. Also discussed are simple solutions to problems in integrating them into a conventional cache hierarchy while preserving properties such as coherence, consistency, and reliability. Compute Caches increase performance by 1.9× and reduce energy by 2.4× for a suite of data-centric applications, including text and database query processing, cryptographic kernels, and in-memory checkpointing. Applications with larger fraction of Compute Cache operations could benefit even more, as our micro-benchmarks indicate (54× throughput, 9× dynamic energy savings).

This paper investigates the feasibility of a unified processor architecture to enable error coding flexibility and secure communication in low power Internet of Things (IoT) wireless networks. Error coding flexibility for wireless communication allows IoT applications to exploit the large tradeoff space in data rate, link distance and energy-efficiency. As a solution, we present a light-weight Galois Field (GF) processor to enable energy-efficient block coding and symmetric/asymmetric cryptography kernel processing for a wide range of GF sizes (2m, m = 2, 3, ..., 233) and arbitrary irreducible polynomials. Program directed connections among primitive GF arithmetic units enable dynamically configured parallelism to efficiently perform either four-way SIMD 5- to 8-bit GF operations, including multiplicative inverse, or a wide bit-width (e.g., 32-bit) GF product in a single cycle. To illustrate our ideas, we synthesized our GF processor in a 28nm technology. Compared to a baseline software implementation optimized for a general purpose ARM M0+ processor, our processor exhibits a 5-20 x speedup for a range of error correction codes and symmetric/asymmetric cryptography applications. Additionally, our proposed GF processor consumes 431μW at 0.9V and 100MHz, and achieves 35.5pJ/b energy efficiency while executing AES operations at 12.2Mbps. We achieve this within an area of 0.01mm2.

A spin synapse with analog programmability using all charge current is proposed. Compared with using spin current, the proposed all-charge-current synapses can be placed in a larger cross-bar array to form a denser and larger neural network. Using the current summation, DOT product can be realized. We further employ a compact racetrack converter as the neuron to implement a rectified-linear neural network, saving area by 67% and energy by 69% compared with a spin binary-threshold neural network while achieving similar accuracy with MNIST digit recognition benchmark. Storing the domain wall motion in a time-based fashion, a recurrent neural network can be realized for time-involved inference tasks.

Finite State Automata are widely used to accelerate pattern matching in many emerging application domains like DNA sequencing and XML parsing. Conventional CPUs and compute-centric accelerators are bottlenecked by memory bandwidth and irregular memory access patterns in automata processing.

Long-range low-power localization is a key technology that enables a host of new applications of wireless sensor nodes. We present RF-Echo, a new low-power RF localization solution that achieves decimeter accuracy in long range indoor non-line-of-sight (NLOS) scenarios. RF-Echo introduces a custom-designed active RF reflector ASIC (application specific integrated circuit) fabricated in a 180nm CMOS process which echoes a frequency-shifted orthogonal frequency division multiplexing (OFDM) signal originally generated from an anchor. The proposed technique is based on time-of-flight (ToF) estimation in the frequency domain that effectively eliminates inter-carrier and inter-symbol interference in multipath-rich indoor NLOS channels. RF-Echo uses a relatively narrow bandwidth of $\leq$80 MHz which does not require an expensive very high sampling rate analog-to-digital converter (ADC). Unlike ultra-wideband (UWB) systems, the active reflection scheme is designed to operate at a relatively low carrier frequency that can penetrate building walls and other blocking objects for challenging NLOS scenarios. Since the bandwidth at lower frequencies (2.4 GHz and sub-1 GHz) is severely limited, we propose novel signal processing algorithms as well as machine learning techniques to significantly enhance the localization resolution given the bandwidth constraint of the proposed system. The newly fabricated tag IC consumes 62.8 mW active power. The software defined radio (SDR) based anchor prototype is rapidly deployable without the need for accurate synchronization among anchors and tags. Field trials conducted in a university building confirm up to 85 m operation with decimeter accuracy for robust 2D localization.

Current-induced domain wall (DW) motion in spintronic racetrack memory promises energy-efficient analog computation using compact magnetic nanowires. This paper explores the feasibility of analog-to-digital converter (ADC) based on current-induced DW motion and introduces an n-bit ADC using n racetrack magnetic nanowires. With each magnetic nanowire having a different configuration granularity, an n-bit binary or gray code is generated simultaneously. The proposed ADC structure achieves 21 fJ/conversion-step at 20 MHz with an area of about 10 μm2. The racetrack ADC is suitable for applications requiring dense ADC arrays, such as image sensors. This paper describes one ultrahigh speed digital pixel sensor imaging system benefiting from the racetrack ADC.

In recent years, operating at near-threshold supply voltages has been proposed to improve energy efficiency in circuits, yet decreased efficacy of dynamic voltage scaling has been observed in recent planar technologies. However, foundries have introduced a shift from planar to FinFET fabrication processes. In this paper, we study 7nm FinFET's ability to voltage scale and compare it to planar technologies across three dynamic voltage scaling scenarios. The switch to FinFET allows for a return to strong voltage scalability. We find up to 8.6× higher energy efficiency at NT compared to nominal supply voltage (vs. 4.8× gain in 20nm planar).

In this paper, we define a configurable Software Defined Radio (SDR) baseband processor design for the Internet of Things (IoT). We analyzed the fundamental algorithms in communications systems on IoT devices to enable a microarchitecture design that supports many IoT standards and custom nonstandard communications. Based on this analysis, we propose a custom SIMD execution model coupled with a scalar unit. We introduce several architectural optimizations to this design: streaming registers, variable bit width datapath, dedicated ALUs for critical kernels, and an optimized flexible reduction network. We employ voltage scaling and clock gating to further reduce the power, while more than a 100% time margin has been reserved for reliable operation in the near-threshold region. Together our architectural enhancements lead to a 71× power reduction compared to a classic general purpose SDR SIMD architecture. Our IoT SDR datapath has sub-mW power consumption based on SPICE simulation, and is placed and routed to fit within an area of 0.074mm2in a 28nm process. We implemented several essential elementary signal processing kernels and combined them to demonstrate two end-to-end upper bound systems, 802.15.4-OQPSK and Bluetooth Low Energy. Our full SDR baseband system consists of a configurable SIMD with a control plane MCU and memory. For comparison, the best commercial wireless transceiver consumes 23.8mW for the entire wireless system (digital/RF/ analog). We show that our digital system power is below 2mW, in other words only 8% of the total system power. The wireless system is dominated by RF/analog power comsumption, thus the price of flexibility that SDR affords is small. We believe this work is unique in demonstrating the value of baseband SDR in the low power IoT domain.

This paper presents Neighbor-Guided SemiGlobal Matching (NG-fSGM), a new method for optical flow. It is based on SGM, a popular dynamic programming algorithm for stereo vision, where the disparity of each pixel is calculated by aggregating local matching costs over the entire image to resolve local ambiguity in texture-less and occluded regions. Unlike conventional SGM, NG-fSGM operates on a subset of the search space that has been aggressively pruned based on neighboring pixels' information. Our proposed method achieves a fast approximation of SGM with significantly simpler cost aggregation and flow computation. Compared to a prior SGM extension for optical flow, the proposed NG-fSGM provides about 9x reduction in the number of computations and 5x reduction in the memory requirement with only 0.17% accuracy degradation when evaluated with Middlebury benchmark test cases.

This paper presents an energy-autonomous wireless communication system for ultra-small Internet-of-Things (IoT) platforms. In the proposed system, all necessary components, including the battery, energy-harvesting solar cells, and the RF antenna, are fully integrated within a millimeter-scale form factor. Designing an energy-optimized wireless communication system for such a miniaturized platform is challenging because of unique system constraints imposed by the ultra-small system dimension. The proposed system targets orders of magnitude improvement in wireless communication energy efficiency through a comprehensive system-level analysis that jointly optimizes various system parameters, such as node dimension, modulation scheme, synchronization protocol, RF/analog/digital circuit specifications, carrier frequency, and a miniaturized 3-D antenna. We propose a new protocol and modulation schemes that are specifically designed for energy-scarce ultra-small IoT nodes. These new schemes exploit abundant signal processing resources on gateway devices to simplify design for energy-scarce ultra-small sensor nodes. The proposed dynamic link adaptation guarantees that the ultra-small IoT node always operates in the most energy efficient mode for a given operating scenario. The outcome is a truly energy-optimized wireless communication system to enable various classes of new applications, such as implanted smart-dust devices.

Wireless sensor nodes (WSNs) are essential elements for today's structural health monitoring (SHM). As the design technology for WSNs evolves, there have been continuous efforts to address challenges for WSNs such as short lifetime, high power consumption, and bulky volume. Recent trends show energy harvesting becoming a popular solution for extending the lifetime of WSNs; even implementing energy-autonomous systems is an option. Smaller WSN form factors have been developed for volume-limited applications and sensor nodes as small as a few mm3were created with custom-designed integrated circuits (ICs). Custom IC-based sensor nodes enable energy-efficient implementation of WSNs for SHM. Ultralow power circuits with operation power on the order of nanowatts are introduced in this paper. Ultralow power energy harvesters, timers, wakeup receivers, sensing modalities, and microprocessors are expected to significantly extend sensor node lifetime or achieve sensor node operation with ambient energy harvesting.

Most server-grade systems provide Chipkill-Correct error protection at the expense of power and performance. In this paper we present a low overhead solution to improving the reliability of commodity DRAM systems with no change in the existing memory architecture. Specifically, we propose five erasure and error correction (E-ECC) schemes that provide at least Chipkill-Correct protection for x4 (Schemes 1, 2 and 3), x8 (Scheme 4) and x16 (Scheme 5) DRAM systems. All schemes have superior error correction performance due to the use of strong symbol-based codes. Synthesis results in 28 nm node show that the decoding latency of these codes is negligible compared to the DRAM access latency. In addition, we make use of erasure codes to extend the lifetime of the DRAM systems. Specifically, once a chip is marked faulty due to persistent errors, all E-ECC schemes correct erasures due to that faulty chip and also correct an additional random error in a second chip. Evaluation with SPEC2006 workloads show that compared to x4 Chipkill-Correct schemes, Scheme 5 has the highest IPC improvement (mean of 7 percent) and Scheme 4 has the largest power reduction (mean of 18 percent) and the largest increase in energy efficiency (mean of 25 percent).

In this paper, approximate SRAMs are explored in the context of error-tolerant applications, in which energy is saved at the cost of the occurrence of read/write errors (i.e., signal quality degradation). This analysis investigates variation-resilient techniques that enable dynamic management of the energy-quality tradeoff down to the bit level. In these techniques, the different impacts of errors on quality at different bit positions are explicitly considered as key enabler of energy savings that are far larger than a simple voltage scaling. The analysis is based on the experimental results in an energy-quality scalable 28-nm SRAM and the extrapolation to a wide range of conditions through the models that combine the individual energy contributions. Results show that the joint adoption of multiple bit-level techniques provides substantially larger energy gains than individual techniques. Compared with the simple voltage scaling at isoquality, the joint adoption of these techniques can provide more than $2\times $ energy reduction at negligible area penalty. Energy savings turn out to be highly sensitive to the choice of joint techniques, thus showing the crucial importance of dynamic energy-quality management in approximate SRAMs.

As we show in this paper, I/O has become the limiting factor in scaling down size and power toward the goal of invisible computing. Achieving this goal will require composing optimized and specialized---yet reusable---components with an interconnect that permits tiny, ultra-low power systems. In contrast to today's interconnects which are limited by power-hungry pull-ups or high-overhead chip-select lines, our approach provides a superset of common bus features but at lower power, with fixed area and pin count, using fully synthesizable logic, and with surprisingly low protocol overhead.

Current-induced domain wall motion in racetrack memory promises energy-efficient analog computation using compact magnetic nanowires. This paper explores the feasibility of data converters based on current-induced domain wall motion and introduces an n-bit ADC using n racetrack magnetic nanowires. With each magnetic nanowire having a different configuration granularity, an n-bit binary or gray code is generated simultaneously. The proposed ADC structure achieves 21fJ/conversion-step at 20MHz with area under 10μm2. The racetrack ADC is suitable for applications requiring dense ADC arrays, such as image sensors. This paper describes one ultra-high speed DPS imaging system benefiting from the racetrack ADC.

Communication in multi-processor systems-on-chip requires guaranteed throughput and latency. If the network is unaware of ongoing communication patterns, applications may not receive their necessary bandwidth or may suffer high network latencies. Many techniques have been proposed to provide quality-of-service (QoS) in the network by regulating network traffic; however, as network sizes have increased, the complexity of these techniques has grown as well, particularly in the case of multi-hop networks.

Separable allocators in on-chip routers perform switch allocation in two stages that often make uncoordinated decisions resulting in sub-optimal switch allocation. We propose Virtual Input Crossbars (VIX), where more than one virtual channel (VC) of an input port is connected to the crossbar. VIX improves switch allocation by allowing more than one input VC of an input port to transmit flits in the same cycle. Also, more input VCs can participate in the output arbitration, reducing the chances of uncoordinated decisions. VIX improves network throughput by more than 15% for the topologies studied without affecting the router critical path.

An ARM-based sensing platform powered entirely by small-scale benthic microbial fuel cells (MFCs) for oceanic sensing applications is presented. The ultra-low power chip featuring an ARM Cortex-M0 processor, 3kB of SRAM, and power management unit (PMU) with energy harvesting from MFCs is designed to consume 11nW in sleep mode for perpetual sensing operation. A small-scale micro-MFC with 21.3cm2anode surface area was connected to the on-chip PMU to charge a thin film battery of 1mAh capacity. A 49.3-hour long-term experiment with 8-min sleep interval and 1 sec wake-up time demonstrated the sustainability of chip-on-mud concept. During sleep mode, the system charges the 4V battery at 380nA from the micro-MFC generating 5.4μW of power, which can support up to 20mA of active mode current.

In this paper, we present switched analog circuit (SAC), a new circuit architecture, to implement an energy-efficient mixed-signal dot product (DP) kernel for machine learning and signal processing applications. SAC operates by fast switching the analog inputs to output via variable width digital pulses. The output accuracy and energy consumption of SAC is analyzed and verified for an average and Gaussian blur filter. Simulations in a commercial 130 nm process for a 120 x 120 image show energy savings of 19x-to-32x compared to a digital implementation for signal-to-noise ratios (SNRs) of 30 dB-to-24 dB, respectively.

This paper proposes a novel 3D switch, called 'Hi-Rise', that employs high-radix switches to efficiently route data across multiple stacked layers of dies. The proposed interconnect is hierarchical and composed of two switches per silicon layer and a set of dedicated layer to layer channels. However, a hierarchical 3D switch can lead to unfair arbitration across different layers. To address this, the paper proposes a unique class-based arbitration scheme that is fully integrated into the switching fabric, and is easy to implement. It makes the 3D hierarchical switch's fairness comparable to that of a flat 2D switch with least recently granted arbitration. The 3D switch is evaluated for different radices, number of stacked layers, and different 3D integration technologies. A 64-radix, 128-bit width, 4-layer Hi-Rise evaluated in a 32nm technology has a throughput of 10.65 Tbps for uniform random traffic. Compared to a 2D design this corresponds to a 15% improvement in throughput, a 33% area reduction, a 20% latency reduction, and a 38% energy per transaction reduction.

Deeply scaled CMOS circuits are increasingly susceptible to transient faults and soft errors; emerging post-CMOS devices can be more vulnerable, sometimes exhibiting erratic errors of arbitrary duration. Applying timing and supply voltage margin is wasteful and becoming ineffective, and conventional checking and sparing techniques provide only a limited error coverage against widely varying errors. We propose a confidence-driven computing (CDC) model for an adaptive protection against nondeterministic errors. The CDC model employs fine-grained temporal redundancy and confidence checking for a faster adaptation and tunable reliability. The CDC model can be extended to deeply scaled CMOS circuits that are mainly affected by transient faults and soft errors, where an early checking (EC) technique can be used to perform independent error checking for more flexibility and better performance. To evaluate the CDC model, we apply a sample-based field-programmable gate array emulation along with real-time error injection. The CDC model is shown to adapt to fluctuating error rates and enhance the system reliability by effectively trading off performance. To evaluate the EC technique at a finer time scale, we create a new event-based simulation to capture path delay distribution, error model, and their interactions. The EC technique improves the system reliability by more than four orders of magnitude when errors are of short duration. Both the CDC model and the EC technique are synthesized in a 45-nm CMOS technology for cost estimates: 1) the area overhead is as low as 12% and 2) energy overhead can be limited to 19%.

In this paper, we explore the challenges in scaling on-chip networks towards kilo-core processors. Current low-radix topologies optimize for fast local communication, but do not scale well to kilo-core systems because of the large number of routers required. These increase both power and hop count. In contrast, symmetric high-radix topologies optimize for global communication with fewer hop counts, but degrade local communication with their large, slow routers. To address both local and global communication optimizations independently, we decouple the interconnect design using asymmetric high-radix topologies. By setting a design goal of matching rauter speed with wire speed, our praposed topologies use fast medium-radix rauters to optimize for local communication and a few slow high-radix rauters that reduce hop count to optimize for global communication. Our asymmetric high-radix designs are enabled by recently praposed SwizzleSwitches, which allow us to achieve peiformance scalability within realistic power budgets. We prapose and evaluate two asymmetric high-radix topologies: Super-Star (asymmetric folded Clos) and Super-StarX (asymmetric folded Clos with superimposed mesh). Our evaluations show that the best performing asymmetric high-radix topology improves average network latency over a mesh by 45% while reducing the power consumption by 40%. When compared to symmetric high-radix topologies network thraughput is improved by 2.9× while still praviding similar latency benefits and power ejficiency.

This paper proposes an energy-efficient VGA full-frame feature extraction processor design. It is based on the SURF algorithm and makes various algorithmic modifications to improve efficiency and reduce hardware overhead while maintaining extraction performance. Low clock frequency and deep parallelism derived from a one-sample-per-cycle matched-throughput architecture provide significantly larger room for voltage scaling and enables full-frame extraction. The proposed design consumes 4.7mW at 400mV and achieves 72% higher energy efficiency than prior work.

We present a self-adapting power management unit (PMU) for ultra-low power wireless sensor nodes. The PMU uses 1.03nF of on-chip MIM capacitance in a reconfigurable switched-capacitor network (SCN) that automatically adapts to different battery voltages for down-conversion and different harvesting sources/harvesting conditions for up-conversion. The PMU achieves 63.8% / 60.7% down-conversion efficiency at 17.9μW active mode / 12.8nW sleep mode power loading. With the adaptive down-conversion ratio, load power range is improved by 3.76× and 5.48× in sleep and active mode, respectively. We show how the proposed adaptation method enables harvesting with solar, microbial fuel cell, and thermal energy sources, increases harvesting efficiency by 1.92× and achieves the peak extraction efficiency of 99.8% for solar cell.

The power target for exascale supercomputing is 20MW, with about 30% budgeted for the memory subsystem. Commodity DRAMs will not satisfy this requirement. Additionally, the large number of memory chips (>10M) required will result in crippling failure rates. Although specialized DRAM memories have been reorganized to reduce power through 3D-stacking or row buffer resizing, their implications on fault tolerance have not been considered. We show that addressing reliability and energy is a co-optimization problem involving tradeoffs between error correction cost, access energy and refresh power---reducing the physical page size to decrease access energy increases the energy/area overhead of error resilience. Additionally, power can be reduced by optimizing bitline lengths. The proposed 3D-stacked memory uses a page size of 4kb and consumes 5.1pJ/bit based on simulations with NEK5000 benchmarks. Scaling to 100PB, the memory consumes 4.7MW at 100PB/s which, while well within the total power budget (20MW), is also error-resilient.

Oxide breakdown has become an increasingly pressing reliability issue in modern very large scale integration design with ultrathin oxides. The conventional guard-band methodology assumes uniformly thin oxide thickness, resulting in overly pessimistic reliability estimation that severely degrades system performance. In this paper, we present the use of limited post-fabrication measurements of oxide thicknesses from on-chip sensors to aid in the chip-level oxide breakdown reliability management. A key challenge, which is the focus of this paper, is precisely predicting and managing the reliability condition of each chip with a limited number of measurements and quantifying the tradeoff between reliability margin and system performance. Given the post-fabrication measurements, chip oxide breakdown reliability can be formulated as a conditional distribution that allows one to achieve a significantly more accurate chip lifetime estimation. The estimation is then used to individually tune the supply voltage of each chip for performance maximization while maintaining or improving the reliability. Experimental results show that, by using 25 measurements, the proposed method can achieve an average of 19% performance improvement, and a 27% maximum for a design with up to 50 million devices, with an average operation time of approximately 0.4 s per chip.

The theoretical lower limit of subthreshold swing in mosfets (60 mV/decade) significantly restricts low-voltage operation since it results in a low ON -to- OFF current ratio at low supply voltages. This paper investigates extremely low-power circuits based on new Si/SiGe heterojunction tunneling transistors (HETTs) that have a subthreshold swing of . Device characteristics, as determined through technology computer aided design tools, are used to develop a Verilog-A device model to simulate and evaluate a range of HETT-based circuits. We show that an HETT-based ring oscillator (RO) shows a 9-19 times reduction in dynamic power compared to a CMOS RO. We also explore two key differences between HETTs and traditional mosfets, namely, asymmetric current flow and increased Miller capacitance, analyze their effect on circuit behavior, and propose methods to address them. HETT characteristics have the most dramatic impact on static random access memory (SRAM) operation and we propose a novel seven-transistor HETT-based SRAM cell topology to overcome, and take advantage of, the asymmetric current flow. This new HETT SRAM design achieves 7-37 times reduction in leakage power compared to CMOS.

With multi-core processors now mainstream, the shift to many-core processors poses a new set of design challenges. In particular, the scalability of coherence protocols remains a significant challenge. While complex Network-on-Chip interconnect fabrics have been proposed and in some cases implemented, most of industry has slowly evolved existing coherence solutions to meet the needs of a growing number of cores. Industries' slow adoption of Network-on-Chip designs is in large part due to the significant effort needed to design and verify the system. However, simply scaling bus-based coherence is not straightforward either because of increased contention and latency on the bus for large core counts.

Designing an ultra-low power sensor node requires careful consideration of the system-level energy budget. Depending on applications, various components can dominate total energy. In this paper, we review three different system energy budget scenarios where any of the microprocessor, memory, and timer of a sensor node can dominate the energy budget. The design space and corresponding trade-offs for these three components are explored to suggest guidelines for the design of ultra-low power sensor nodes.

Supply voltage scaling has stagnated in recent technology nodes, leading to so-called "dark silicon." In this paper, we investigate the limit of voltage scaling together with task parallelization to maintain task completion latency. When accounting for parallelization overheads, minimum task energy is obtained at "near threshold" supply-voltages across 6 commercial technology nodes and provides 4X improvement in overall CMP performance.

A scalable architecture to design high radix switch fabric is presented. It uses circuit techniques to re-use existing input and output data buses and switching logic for fabric configuration and supporting multiple arbitration policies. In addition, it integrates a 4-level message-based priority arbitration for quality of service. Fine grain clock gating, tiled fabric topology and self-regenerating bit-line repeaters enable scaling the router to 8k wires. A 64x64(128b data) switch fabric fabricated in 45nm SOI CMOS spans 4.06mm2 and achieves a throughput of 4.5Tb/s at 3.4Tb/s/W at 1.1V with a peak measured efficiency of 7.4Tb/s/W at 0.6V.

Near-threshold operation has emerged as a competitive approach for energy-efficient architecture design. In particular, a combination of near-threshold circuit techniques and parallel SIMD computations achieves excellent energy efficiency for easy-to-parallelize applications. However, near-threshold operations suffer from delay variations due to increased process variability. This is exacerbated in wide SIMD architectures where the number of critical paths are multiplied by the SIMD width. This paper provides a systematic in-depth study of delay variations in near-threshold operations and shows that simple techniques such as structural duplication and supply voltage/frequency margining are sufficient to mitigate the timing variation problems in wide SIMD architectures at the cost of marginal area and power overhead.

In this work we expose the challenges of interfacing both conventional and new systems with an extremely resource constrained platform. We find that even when attempts are made to utilize an industry standard protocol (I2C), necessary protocol modifications for ultra-low power design means that interfacing remains non-trivial.

We propose an adaptive WWL pulse width and voltage modulation architecture for low voltage bit-interleaved 8T SRAMs. The 8T bitcell offers improved read/write margins but suffers from write and half select concerns when bit-interleaved [1]. Also, low voltage operation leads to a long-tailed write time distribution, requiring large timing margins and limiting Vmin. To minimize timing margins and reduce Vmin, both WWL pulse width and voltage level are adaptively modulated by monitoring written values through the read path. In a 65nm CMOS prototype chip, Vmin is lowered from 700mV to 500mV using this technique, providing 2.55× leakage power reduction and 2.4× active power reduction.

Aggressive transistor scaling continues to drive increasingly complex digital designs. The large number of transistors available today enables the development of chip multiprocessors that include many cores on one die communicating through an on-chip interconnect. As the number of cores increases, scalable communication platforms, such as networks-on-chip (NoCs), have become more popular. However, as the sole communication medium, these interconnects are a single point of failure so that any permanent fault in the NoC can cause the entire system to fail. Compounding the problem, transistors have become increasingly susceptible to wear-out related failures as their critical dimensions shrink. As a result, the on-chip network has become a critically exposed unit that must be protected. To this end, we present Vicis, a fault-tolerant architecture and companion routing protocol that is robust to a large number of permanent failures, allowing communication to continue in the face of permanent transistor failures. Vicis makes use of a two-level approach. First, it attempts to work around errors within a router by leveraging reconfigurable architectural components. Second, when faults within a router disable a link's connectivity, or even an entire router, Vicis reroutes around the faulty node or link with a novel, distributed routing algorithm for meshes and tori. Tolerating permanent faults in both the router components and the reliability hardware itself, Vicis enables graceful performance degradation of networks-on-chip.

This paper investigates the optimization of sleep mode energy consumption for ultra-low VddCMOS circuits, which is motivated by our findings that minimization of sleep mode energy holds great potential for reducing total energy consumption. We propose a unique approach of using a power gating switch (PGS) in ultra-low Vddregimes. Unlike the conventional manner of using PGSs, our optimization suggests using minimal-sized PGSs with a slightly higher Vddto compensate for voltage drop across the PGS. In SPICE simulations, this reduces total energy consumption by ~125× compared to conventional approaches. The effectiveness of the proposed optimization is also confirmed by measurements taken from an ultra-low power microprocessor. Additionally, the feasibility of using minimal PGSs in ultra-low Vddregimes is investigated using SPICE simulations and silicon measurements.

We designed two compact in situ NBTI and oxide degradation sensors with digital outputs in 130 nm CMOS. The 308 μm2NBTI sensor and the 150 μm2oxide degradation sensor provide digital frequency outputs and are compatible with a cell-based design methodology without requiring analog supplies. The sensors enable high-volume data collection and monitoring of degradation mechanisms to guide dynamic control schemes and warn of impending device failure. Large scale data-collection permits improved modeling and the potential for insight into the underlying reliability mechanisms. The oxide degradation sensor monitors the change in gate leakage under stress conditions and is the first proposed of its kind. The NBTI sensor is 110× smaller than previous work and is designed to compensate for temperature variations during measurement. A maximum error of 2.2% is observed for the NBTI sensor under process, voltage, and temperature variations. It provides ΔVthmeasurement with 3σ accuracy of 1.23 mV from 40° C-110° C.

This paper investigates pipelining methodologies for the ultra low voltage regime. Based on an analytical model and simulations, we propose a pipelining technique that provides higher energy efficiency and performance than conventional approaches to ultra low voltage design. Two-phase latch based design and sequential circuit optimizations are also proposed to further improve energy efficiency and performance. Silicon results demonstrate a 16b multiplier using the approaches in 65nm CMOS improve energy efficiency by 30% and performance by 60%.

We propose an adaptive reliability enhancement structure for deeply-scaled CMOS and future devices that exhibit nondeterministic behavior. This structure forms the basis of a confidence-driven computing model that can be implemented in either a rollback recovery or an iterative dual modular redundancy method incorporating synchronous handshake schemes. The performance and cost of the computing model are estimated using a 45 nm CMOS technology and the functionality is verified by FPGA-based emulation. The confidence-driven computing model is demonstrated using a 16-bit, 12-stage CORDIC processor operating under random, transient errors. The confidence-driven computing model adapts to the fluctuating error rates at the device substrate level to guarantee the reliability of computation at the system level. This computing model costs 4.2 times smaller area and 2.7 times less energy overhead than triple modular redundancy to guarantee a system-level mean time to failure of two years.

Driven by continued scaling of Moore's Law, the number of processing elements on a die are increasing dramatically. Recently there has been a surge of wide single instruction multiple data architectures designed to handle computationally intensive applications like 3D graphics, high definition video, image processing, and wireless communication. A limit of the SIMD width of these types of architectures is the scalability of the interconnect network between the processing elements in terms of both area and power. To mitigate this problem, we propose the use of a new interconnect topology, XRAM, which is a low power high performance matrix style crossbar. It re-uses output buses for control programming, and stores multiple swizzle configurations at the cross points using SRAM cells, significantly reducing routing congestion and control signaling. We show that compared to conventionally implemented crossbars, the area scales with the product of inputx output ports while consuming almost 50% less energy. We present an application case study, color-space conversion, utilizing XRAM and show a 1.4× gain in performance while consuming 1.5-2.5× less power.

This paper proposes an ultra low energy FFT processor suitable for sensor applications. The processor is based on R4MDC but achieves full utilization of computational elements. It has two parallel datapaths that increase throughput by a factor of 2 and also enable high memory utilization. The proposed design is implemented in 65nm CMOS technology and post-layout simulation including parasitic capacitances shows it achieves 9.25× higher energy efficiency than state-of-the-art FFT processors and high throughput relative to past subthreshold circuit implementations.

We present a 45 nm half-differential 6T SRAM (HD-SRAM) with differential write and single-ended read, enabling asymmetric sizing and VTHselection. The HD-SRAM bitcell uses SRAM physical design rules to achieve the same area as a commercial differential 6T SRAM (D-SRAM). We record measurements from 80 32 kb SRAM arrays. HD-SRAM is 18% lower energy and 14% lower leakage than D-SRAM. It has a 72 mV-lower VMIN, demonstrating higher stability.

A low leakage memory is an indispensable part of any sensor application that spends significant time in standby (sleep) mode. Although using high Vth (HVT) devices is the most straightforward way to reduce leakage, it also limits operation speed during active mode. In this paper, a low leakage 10T SRAM cell, which compensates for operation speed using a readily available secondary supply, is proposed in a 0.18μm CMOS process. It achieves the lowest-to-date leakage power consumption and achieves robust operation at low voltage without sacrificing operation speed. The 10T SRAM has a bit cell area of 17.48μm2 and is measured to consume 1.85fW per bit at 0.35V.

In this paper, we propose a stratification+hybrid quasi Monte Carlo (SH-QMC) approach to improve the efficiency of Monte Carlo-based statistical static timing analysis (SSTA) using sample size reduction. Sample size reduction techniques proposed in the literature exhibit a tradeoff between accuracy of the Monte Carlo estimate with fewer samples and their ability to handle large number of variables in multidimensional space. This paper proposes to target several such techniques to different sets of process variation variables by using information about the importance of these variables to the circuit delay, and the capability of the techniques to handle multiple dimensions. Simulations on benchmark circuits up to 90 K gates show that the proposed method requires up to 224 samples for varying levels of process variation to achieve accurate timing estimates. Results also show that when SH-QMC is performed with multiple parallel threads on a quad-core processor, the approach is faster than traditional SSTA with comparable accuracy. When the proposed SH-QMC technique is supplemented with a graph pruning method the runtime is further reduced by 46-48% on average. The technique is also extended to include an incremental approach to recompute a percentile delay metric after engineering change order.

Gate oxide breakdown (OBD) is a key factor limiting the useful lifetime of an integrated circuit. Unfortunately, the conventional approach for full chip OBD reliability analysis assumes a uniform oxide thickness and worst-case temperature for all devices. In practice, however, gate oxide thickness varies from die-to-die and within-die and hence may cause different reliability for different devices even chips. Moreover, due to the increased across-die temperature variation, such difference may be exacerbated. Thus, as the precision of variation control worsens, an alternative reliability analysis approach is needed. In this paper, we propose a statistical framework for chip-level gate OBD reliability analysis while considering both die-to-die and within-die components of thickness variations as well as the across-die temperature variation. The thickness of each device is modeled as a distinct random variable and thus the full chip reliability estimation problem is defined on a huge sample space of several million devices. We observe that the chip-level OBD reliability function is independent of the relative location of the individual devices. This enables us to transform the problem such that the resulting representation can be expressed in terms of much fewer random variables. Using this transformation, we present a computationally efficient and accurate approach for estimating the full chip reliability while considering spatial correlations of gate oxide thickness as well as temperature variation. We show that, compared to Monte Carlo simulation, the proposed method incurs an error of only around 1% while improving the runtime by more than three orders of magnitude.

In this paper, we propose a new circuit technique for on-chip communication, the edge encoding technique, to reduce the energy consumption in multi-cycle interconnects. Both average and worst-case energy are reduced by desynchronizing the edges of rising and falling transitions. In a 1.2 V 65-nm CMOS technology, the proposed approach achieves up to 34% energy reduction with no latency overhead over optimally designed conventional busses due to coupling capacitance reductions. The technique further reduces energy consumption by 39% with iso-throughput at the expense of one-cycle latency. Energy savings are shown to be both larger and more robust to process, voltage, and temperature variations than previous techniques.

Suppresing device leakage while maximizing drive current is the prime focus of semiconductor industry. Rapid Thermal Annealing (RTA) drives process development on this front by enabling fabrication steps such as shallow juction formation that require a low thermal budget. However, decrease in junction anneal time for more aggressive device scaling has reduced the characteristic thermal length to dimensions less than the typical die size. Also, the amount of heat transferred, and hence the local anneal temperature, is affected by the layout pattern dependence of optical properties in a region. This variation in local anneal temperature causes a variation in performance and leakage across the chip by affecting the threshold voltage (Vth) and extrinsic transistor resistance (Rext). In this work, we propose a new local anneal temperature variation aware analysis framework which incorporates the effect of RTA induced temperature variation into timing and leakage analysis. We solve for chip level anneal temperature distribution, and employ TCAD based device level models for drive current (Ion) and leakage current (Ioff) dependence on anneal temperature variation, to capture the variation in device performance and leakage based on its position in the layout. Experimental results based on a 45 nm experimental test chip show anneal temperature variation of up to ~10.5°C, which results in ~6.8% variation in device performance and ~2.45X variation in device leakage across the chip. The corresponding variation in inverter delay was found to be ~7.3%. The temperature variation for a 65 nm test chip was found to be ~8.65°C.

Circuits designed in aggressively scaled technologies face both stringent power constraints and increased process variability. Achieving high parametric yield is a key design objective, but is complicated by the correlation between power and performance. This paper proposes a novel design time body bias selection framework for parametric yield optimization while reducing testing costs. The framework considers both inter- and intra-die variations as well as power-performance correlations. This approach uses a feature extraction technique to explore the underlying similarity between the gates for effective clustering. Once the gates are clustered, a Gaussian quadrature based model is applied for fast yield analysis and optimization. This work also introduces an incremental method for statistical power computation to further reduce the optimization complexity. The proposed framework improves parametric yield from 39% to 80% on average for 11 benchmark circuits while runtime is linear with circuit size and on the order of minutes for designs with up to 15K gates.

Modern CMOS technologies employ process-induced stress to improve carrier mobility and increase drive current. This stress has been shown to be strongly layout dependent; however there is a lack of physical models relating potential performance variation to critical layout parameters. This paper presents compact closed-form models that capture the layout dependence of mechanical stress induced in the device channel while considering all relevant sources of stress (STI, tensile/compressive nitride liners, and embedded SiGe). The models are calibrated using ring oscillator frequency data obtained from an experimental test chip to verify their accuracy. Results indicate that the models accurately capture the layout dependence of stress and carrier mobility for a variety of layout permutations and the root mean square error in the predicted ring oscillator frequency is less than 1% for the different layout experiments. These models can help drive layout optimization and timing/power analysis without the use of technology computer-aided design (TCAD) tools, which are slow and very limited in capacity.

To exploit the benefits of throughput-optimized processors such as GPUs, applications need to be redesigned to achieve performance and efficiency. In this work, we present techniques to speed up statistical timing analysis on throughput processors. We draw upon advancements in improving the efficiency of Monte Carlo based statistical static timing analysis (MC SSTA) using techniques to reduce the sample size or smart sampling techniques. An efficient smart sampling technique, Stratification + Hybrid Quasi Monte Carlo (SH-QMC), is implemented on a GPU based on NVIDIA CUDA architecture. We show that although this application is based on MC analysis with straightforward parallelism available, achieving performance and efficiency on the GPU requires exposing more parallelism and finding locality in computations. This is in contrast with random sampling based algorithms which are inefficient in terms of sample size but can keep resources utilized on a GPU. We show that SH-QMC implemented on a Multi GPU is twice as fast as a single STA on a CPU for benchmark circuits considered. In terms of an efficiency metric, which measures the ability to convert a reduction in sample size to a corresponding reduction in runtime w.r.t a random sampling approach, we achieve 73.9% efficiency with the proposed approaches compared to 4.3% for an implementation involving performing computations on smart samples in parallel. Another contribution of the paper is a critical graph analysis technique to improve the efficiency of Monte Carlo based SSTA, leading to 2--9X further speedup.

Static noise margin analysis using butterfly curves has traditionally played a leading role in the sizing and optimization of SRAM cell structures. Heightened variability and reduced supply voltages have resulted in increased attention being paid to new methods for characterizing dynamic robustness. In this work, a technique based on vector field analysis is presented for quickly extracting both static and dynamic stability characteristics of arbitrary SRAM topologies. It is shown that the traditional butterfly curve simulation for 6T cells is actually a special case of the proposed method. The proposed technique not only allows for standard SNM ¿smallest-square¿ measurements, but also enables tracing of the state-space separatrix, an operation critical for quantifying dynamic stability. It is established via importance sampling that cell characterization using a combination of both separatrix tracing and butterfly SNM measurements is significantly more correlated to cell failure rates then using SNM measurements alone. The presented technique is demonstrated to be thousands of times faster than the brute force transient approach and can be implemented with widely available, standard design tools.

In aggressively scaled technologies, reliability concerns such as oxide breakdown have become a key issue. Dynamic reliability management (DRM) has been proposed as a mechanism to dynamically explore the trade-off between system performance and reliability margin. However, existing DRM methods are hampered by the fact that they do not accurately model spatial and temporal variations in process and temperature parameters which have a strong impact on chip reliability. In addition, they make the simplifying assumption that the future workloads are identical to the currently observed one. This makes them sensitive to sudden workload variations and outliers. In this paper, we present a novel workload-aware dynamic reliability management framework that accounts for local variations in both the process and temperature. The reliability estimation, along with the predicted remaining workload is fed to a dynamic voltage/frequency scaling module to manage the system reliability and optimize processor performance. Using a fast on-line analytical/table-look-up method we demonstrate an average error of 1% with up to 5 orders of magnitude speedup compared to Monte Carlo simulation. Experiments on an Alpha-like processor show our DRM framework fully utilizes the available margin and achieves 28.7% performance improvement on average.

Double patterning lithography (DPL) is widely considered the only lithography solution for 32nm and several subsequent technology nodes. DPL decomposes and prints the critical layout shapes in two exposures, leading to mismatch between adjacent devices due to systematic offsets between the two exposures. This results in adjacent devices with different mean critical dimension (CD), and uncorrelated CD variation. Such a mismatch can increase functional failures in SRAM cells and degrade yield. This paper analyzes the impact of DPL on functional failures in SRAM bitcells, and proposes a DPL-aware SRAM sizing scheme to effectively mitigate yield losses. Experimental results based on 45nm industrial models and test chip measurements show that DPL can significantly impact SRAM cell robustness. Using the proposed DPL-aware sizing scheme, the SRAM cell failure probability can be reduced by up to 3.6X. Also, for iso-robustness, cells optimized by the proposed approach have 7.9% lower dynamic energy as compared to non-DPL aware sizing optimization.

Increase in variability in the nanometer era has contributed to pessimistic guardbands for conventional circuit design techniques that optimize at worst-case process corners. Smart deterministic approaches have been proposed that employ statistical timing analysis to reduce pessimism in the guardbands while retaining the deterministic nature of the algorithms. Other statistical optimization techniques focus on algorithms to maximize robustness of design while being aware of variability. It is not clear how much improvement can be gained using the latter set of approaches over more simple deterministic approaches. This work presents a new lower bound to evaluate these statistical optimization techniques, drawing inspiration from recent advances in sampling based SSTA. We prove that the presented lower bound gives the minimum possible area that can be achieved for a design while meeting a particular timing yield, which is the percentage of die that meeting a specified timing constraint. We then compare several statistical design optimization approaches, including one proposed in this paper called SLOP, against the computed lower bound. We show that even the simplest statistical optimization approaches produce area results which are, on average, within 9.6% of the lower bound while the best ones performed only marginally better, reaching within 3.7% of the bound. This demonstrates that the proposed bound is a close bound. In addition, it also shows that the existing optimization methods have nearly exhausted the obtainable improvement from being statistically aware and mostly provide trade-offs in runtime speed.

Traditional process variation modeling is primarily focused on design-time analysis and optimization. However, with the advances of post-silicon techniques, accurate variation model is also highly desired in various post-silicon applications, such as post-silicon tuning, test vector generation, and reliability prediction. The accuracy of such post-silicon variation models is greatly improved by incorporating test measurements from each wafer or die. However, to limit test cost, the number of measurements must be reduced as much as possible. This paper proposes an active learning framework to dynamically extract post-silicon process variation models with tightened variance from measurements. The framework is composed of two stages, active training and model adaptation. Active training collects information and initializes the models to be used for the forthcoming wafers. Model adaptation stage then validates the models and optimally determines the test configuration for partial testing to reduce the test cost. Experimental results based on the measurements from two industrial processes show that the proposed framework can achieve variation models with variance reduction of ~80% when compared with design-time variation models. Meanwhile, the average estimation error for those untested sites is well maintained at ~2-3% using merely ~30% available test structures for two processes.

This paper describes critical circuit building blocks for emerging sensing applications, particularly those where volume, and therefore power consumption constraints are orders of magnitude below current state of the art. Developments in ultra-low power microprocessors and memories are described, along with sub-nW timekeeping circuits, pW voltage references, and efficient DC-DC voltage conversion circuits at sub-μA current loads. Taken together, these circuit design advances point to a vision of true mm3low-cost sensing nodes with hybrid power sources, i.e., scavenged and micro-battery, providing long lifetime and reliable operation.

Robust design is a critical concern in ultra-low voltage operation due to large sensitivity to process and environmental variations. In particular, clock networks need careful attention to ensure robust distribution of well-defined clock signals to avoid setup and hold time violations. In this paper, we investigate the design methodology of robust clock networks for ultra-low voltage applications. A case study shows that an optimally-chosen clock network improves skew variation by 36× and energy consumption by 49%, compared to a typical clock network. Additionally, the impact of supply voltage and technology scaling on the optimal clock network construction is investigated.

Miniature wireless sensors with long lifetimes enable new applications for medical diagnosis, infrastructure monitoring, military surveillance, and environmental sensing among many other applications in a growing field. Sensor miniaturization leads to decreased on-sensor energy capacity, and lifetime requirements further constrain the sensor's power budget. To enable millimeter-scale wireless sensors with lifetimes of months to years, a new class of low-power circuit techniques is required. Wireless sensors collect and digitize environmental data before processing and transmitting the data wirelessly to base stations or other sensor nodes. Recent low-power advances for each of these functions shed light on how ubiquitous sensing can become a reality.

Power has become the primary design constraint for chip designers today. While Moore's law continues to provide additional transistors, power budgets have begun to prohibit those devices from actually being used. To reduce energy consumption, voltage scaling techniques have proved a popular technique with subthreshold design representing the endpoint of voltage scaling. Although it is extremely energy efficient, subthreshold design has been relegated to niche markets due to its major performance penalties. This paper defines and explores near-threshold computing (NTC), a design space where the supply voltage is approximately equal to the threshold voltage of the transistors. This region retains much of the energy savings of subthreshold operation with more favorable performance and variability characteristics. This makes it applicable to a broad range of power-constrained computing segments from sensors to high performance servers. This paper explores the barriers to the widespread adoption of NTC and describes current work aimed at overcoming these obstacles.

Modeling the effect of coupling-noise on circuit delay is a key issue in static timing analysis and involves thevictim-aggressor alignmentproblem. As delay-noise strongly depends on the skew between the victim-aggressor driver input transitions, it is not possiblea prioriidentify the victim-driver input transition that results in the worst-case delay-noise. Several approaches have been proposed in literature which heuristically search for the worst-case victim-aggressor alignment. This paper presents an analytical result that obviates the need to search for theoptimalvictim-driver input transition, thereby simplifying the victim-aggressor alignment problem significantly. Using the properties of standard nonlinear complementary metal-oxide semiconductor drivers, it is shown that for monotonic input transitions the worst-case victim-driver input transition is the one that switches at thelatestpoint in its timing window. Similarly, the victim-driver input alignment at the earliest point in the timing window is optimal for early-mode analysis. Although this result has been empirically observed in the industry, to the best of our knowledge this is the first paper which provides a rigorous analysis and shows that the above result holds for both linear and nonlinear drivers. It is also shown that the latest alignment of the victim-driver input transition results in the latest victimreceiver outputarrival time even for the cases where the victim is coupled tomultipleaggressors. Finally, experimental results show that limiting the alignment of the victim to only the latest victim-driver input transition can significantly reduce the runtime of existing approaches with no loss of accuracy.

Process-induced mechanical stress is used to enhance carrier transport and achieve higher drive currents in current complementary metal-oxide-semiconductor technologies. This paper explores how to fully exploit the layout dependence of stress enhancement and proposes a circuit-level, block-based, stress-enhanced optimization algorithm that uses stress-optimized layouts in conjunction with dual-Vthassignment to achieve optimal power-performance tradeoffs. We begin by studying how channel stress and drive current depend on layout parameters such as active area length and contact placement, while considering all layout-dependent sources of mechanical stress in a 65 nm industrial process. We then investigate the three main layout properties that impact mechanical stress in this process and discuss how to improve stress-based performance enhancement in standard cell libraries. While varying the stress-altering layout properties of a number of standard cells in a 65 nm industrial library, we show that ?dual-stress? standard cell layouts (analogous to ?dual-Vth?) can be designed to achieve drive current differences up to ~ 14% while incurring less than half the leakage penalty of dual-Vth. Therefore, when the flexibility of ?dual-stress? assignment is combined with dual-Vthassignment (within the proposed joint optimization framework), simulation results for a set of benchmark circuits show that leakage is reduced by ~ 24% on average, for iso-delay, when compared to dual-Vthassignment. Since mobility enhancement does not incur the exponential leakage penalty associated withVthassignment, our optimization technique is ideal for leakage power reduction. However, our framework can also be used to achieve higher performance circuits for iso-leakage and our joint optimization framework can be used to reduce delay on average by ~ 5%. In both cases, the proposed method only incurs a small area penalty (<0.5%).

Voltage scaling is desirable in static RAM (SRAM) to reduce energy consumption. However, commercial SRAM is susceptible to functional failures whenVDDis scaled down. Although several published SRAM designs scaleVDDto 200-300 mV, these designs do not sufficiently consider SRAM robustness, limiting them to small arrays because of yield constraints, and may not correctly target the minimum energy operation point. We examine the effects on area and energy for the differential 6T and 8T bit cells asVDDis scaled down, and the bit cells are either sized and doped, or assisted appropriately to maintain the same yield as with fullVDD. SRAM robustness is calculated using importance sampling, resulting in a seven-order run-time improvement over Monte Carlo sampling. Scaling 6T and 8T SRAMVDDdown to 500 mV and scaling 8T SRAM to 300 mV results in a 50% and 83% dynamic energy reduction, respectively, with no reduction in robustness and low area overhead, but increased leakage per bit. Using this information, we calculate the supply voltage for a minimum total energy operation (VMIN) based on activity factor and find that it is significantly higher for SRAM than for logic.

We review adaptive design techniques with particular emphasis on error-tolerant techniques. We compare and contrast traditional adaptive approaches with error-tolerant techniques and analyze the margins eliminated by each of them. We discuss the applications of the latter to on-chip communication and signal-processing. Finally, we focus on a specific example of an error-tolerant technique for general-purpose computing called Razor.

Process scaling has given designers billions of transistors to work with. As feature sizes near the atomic scale, extensive variation and wearout inevitably make margining uneconomical or impossible. The ElastIC project seeks to address this by creating a large-scale chip-multiprocessor that can self-diagnose, adapt, and heal. Creating large, flexible designs in this environment naturally lends itself to the repetitive nature of network-on-chip (NoC), but the loss of a single link or router will result in complete network failure. In this work we present Vicis, an ElastIC-style NoC that can tolerate the loss of many network components due to wearout induced hard faults. Vicis uses the inherent redundancy in the network and its routers in order to maintain correct operation while incurring a much lower area overhead than previously proposed N-modular redundancy (NMR) based solutions. Each router has a built-in-self-test (BIST) that diagnoses the locations of hard fault and runs a number of algorithms to best use ECC, port swapping, and a crossbar bypass bus to mitigate them. The routers work together to run distributed algorithms to solve network-wide problems as well, protecting the networking against critical failures in individual routers. In this work we show that with stuck-at fault rates as high as 1 in 2000 gates, Vicis will continue to operate with approximately half of its routers still functional and communicating.

Crosstalk delay-noise which occurs due to the simultaneous transitions of victim and aggressor drivers is very sensitive to their mutual alignment. Hence, during static noise analysis, it is crucial to identify the worst-case victim-aggressor alignment which results in the maximum delay-noise. Although several approaches have been proposed to obtain the worst-case aggressor alignment, most of them compute only the worst-case stage delay of the victim. However, in reality it is essential to compute the worst-case combined delay of victim stage and victim receiver gate [5, 9]. We propose a heuristic approach to compute the worst-case aggressor alignment which maximizes the victim receiver output arrival time. In this work, we use a novel cumulative gate overdrive voltage (CGOV ) metric to model the victim receiver output transition. HSPICE simulations, performed on industrial nets to validate the proposed methodology, show an average error of 1.7% in delay-noise when compared to the worst-case alignment obtained by an exhaustive sweeping.

Leakage power minimization is critical to semiconductor design in nanoscale CMOS. On the other hand increasing variability with scaling adds complexity to the leakage analysis problem. In this work we seek to achieve tractability in Monte Carlo-based statistical leakage analysis. A novel approach for fast and accurate statistical leakage analysis considering inter-die and intra-die components is proposed. We show that the optimal way to select samples, to capture intra-die variation accurately, is according to the probability distribution function of total process variation. Intelligent selection of samples is performed using a Quasi Monte Carlo technique. Results are presented for benchmarks with sizes varying from approximately 5,000 to 200,000 gates. The largest benchmark with 198461 gates is evaluated in 3 minutes with the proposed approach compared to 23 hours for random sampling with comparable accuracy. Compared to a conventional analytical approach using Wilkinson's approximation, the proposed technique offers superior accuracy while maintaining efficiency. State dependence and multiple sources of variation are considered and the approach is scalable with number of process parameter variables for standard cell characterization cost. We also show reduction in sample size to meet target accuracy for computing leakage distribution due to the inter-die component only when compared to random selection of samples.

Current trends in technology scaling foreshadow worsening transistor reliability as well as greater numbers of transistors in each system. The combination of these factors will soon make long-term product reliability extremely difficult in complex modern systems such as systems on a chip (SoC) and chip multiprocessor (CMP) designs, where even a single device failure can cause fatal system errors. Resiliency to device failure will be a necessary condition at future technology nodes. In this work, we present a network-on-chip (NoC) routing algorithm to boost the robustness in interconnect networks, by reconfiguring them to avoid faulty components while maintaining connectivity and correct operation. This distributed algorithm can be implemented in hardware with less than 300 gates per network router. Experimental results over a broad range of 2D-mesh and 2D-torus networks demonstrate 99.99% reliability on average when 10% of the interconnect links have failed.

Oxide breakdown has become an increasingly pressing reliability issue in modern VLSI design with ultra-thin oxides. The conventional guard-band methodology assumes uniformly thin oxide thickness and results in overly pessimistic reliability estimation that severely degrades the system performance. In this study we present the use of limited post-fabrication measurements of oxide thicknesses from on-chip sensors to aid in the chip-level oxide breakdown reliability prediction and quantify the trade-off between reliability margin and system performance. Given the post-fabrication measurements, chip oxide breakdown reliability can be formulated as a conditional distribution that allows us to achieve a significantly more accurate chip lifetime estimation. The estimation is then used to individually tune the supply voltage of each chip for performance maximization while maintaining or improving the reliability. Experimental results show that the proposed method can achieve performance improvement of 19% on average and 27% at maximum for a design with up to 50 million devices, using merely 25 measurements per chip, while analysis time is only 0.4 second.

Process induced variations in the interconnect capacitance and resistance have resulted in significant uncertainly in the interconnect delay. In this work, we propose a new method to compute the interconnect corner considering coupling-noise due to simultaneous switching of aggressors. In prior approaches, the interconnect corners were computed under the assumption that the aggressor nets are not switching and no coupling-noise is injected on the victim net. In this paper, we first show that the interconnect corners obtained under such assumptions could in reality be much different from the true interconnect corner and could therefore result in optimistic delay analysis, particularly for fast-path analysis performed to check hold time violations. We also show that in some cases, the interconnect corner may not lie at an extreme point of the process variation range. In this work, we use the Elmore delay metric to efficiently search for the correct interconnect corner of the victim stage considering delay noise. We then show experimental results to verify the effectiveness of our proposed approach and demonstrate that the traditional approaches of computing the interconnect corners could lead to errors of up to 60% on a net by net basis.

Rising design uncertainties at advanced process nodes place conflicting demands on todays engineers. The widening safety-margins required to ensure robust designs in the face of such uncertainties lead to conservative designs with unacceptable power and performance overheads. On the other hand, low-power techniques such as dynamic voltage scaling (DVS) and clock-gating adversely affect circuit robustness. In this paper, we will review a number of techniques for adaptive design which mitigate the impact of margins by tuning system parameters (voltage and frequency) according to variations in runtime workload, environmental and process conditions. We will evaluate the margins that each of the different approaches require to guarantee correctness in the worst-case and typical operating point. We also propose a technique called Razor which is an aggressive method for eliminating all safety-margins through in situ error-detection and correction. Error-detection is achieved by a new type of Razor flip-flop that monitors critical path endpoints and flags timing errors upon detecting spurious transitions. Recovery is achieved through replay from a check-pointed state. We show the design of the transition-detecting Razor flip-flop and illustrate how it naturally detects SEU in combinational logic and inside latches. We present a 64 bit processor implementing Razor and show, on an average, 33% energy savings from our measurements on silicon.

The theoretical lower limit of subthreshold swing in MOSFETs (60 mV/decade) significantly restricts low voltage operation since it results in a low ON to OFF current ratio at low supply voltages. This paper investigates extremely-low power circuits based on new Si/SiGe HEterojunction Tunneling Transistors (HETTs) that have subthreshold swing < 60 mV/decade. Device characteristics as determined through Technology Computer Aided Design (TCAD) tools are used to develop a Verilog-A device model to simulate and evaluate a range of HETT-based circuits. We show that a HETT-based ring oscillator (RO) shows a 9−19X reduction in dynamic power compared to a CMOS RO. We also explore two key differences between HETTs and traditional MOSFETs, namely asymmetric current flow and increased Miller capacitance, analyzing their effect on circuit behavior and proposing methods to address them. Finally, HETT characteristics have the most dramatic impact on SRAM operation and hence we propose a novel 7-transistor HETT-based SRAM cell topology to overcome, and take advantage of, the asymmetric current flow. This new HETT SRAM design achieves 7−37X reduction in leakage power compared to CMOS.

Soft errors in combinational logic circuits are emerging as a significant reliability problem for VLSI designs. Technology scaling trends indicate that the soft error rates (SER) of logic circuits will be dominant factor for future technology generations. SER mitigation in logic can be accomplished by optimizing either the gates inside a logic block or the flipflops present on the block boundaries. We present novel circuit optimization techniques that target these elements separately as well as in unison to reduce the SER of combinational logic circuits. First, we describe the construction of a new class of flip-flop variants that leverage the effect of temporal masking by selectively increasing the length of the latching window thereby preventing faulty transients from being registered. In contrast to previous flip-flop designs that rely on logic duplication and complicated circuit design styles, the new variants are redesigned from the library flip-flop using efficient transistor sizing. We then propose a flip-flop selection method that uses slack information at each primary output node to determine the flip-flop configuration that produces maximum SER savings. Next, we propose a gate sizing algorithm that trades off SER reduction and area overhead. This approach first computes bounds on the maximum achievable SER reduction by resizing a gate. This bound is then used to prune the circuit graph, arriving at a smaller set of candidate gates on which we perform incremental sensitivity computations to determine the gates that are the largest contributors to circuit SER. Third, we propose a unified, co-optimization approach combining flip-flop selection with the gate sizing algorithm. The joint optimization algorithm produces larger SER reductions while incurring smaller circuit overhead than either technique taken in isolation. Experimental results on a variety of benchmarks show average SER reductions of 10.7X with gate sizing, 5.7X with flip-flop assignment, and 30.1X for the combined optimization approach, with no delay penalties and area overheads within 5-6%. The runtimes for the optimization algorithms are on the order of 1-3 minutes.

Subthreshold circuits have drawn a strong interest in recent ultralow power research. In this paper, we present a highly efficient subthreshold microprocessor targeting sensor application. It is optimized across different design stages including ISA definition, microarchitecture evaluation and circuit and implementation optimization. Our investigation concludes that microarchitectural decisions in the subthreshold regime differ significantly from that in conventional superthreshold mode. We propose a new general-purpose sensor processor architecture, which we call the Subliminal Processor. On the circuit side, subthreshold operation is known to exhibit an optimal energy point (Knin)- However, propagation delay also becomes more sensitive to process variation and can reduce the energy scaling gain. We conduct thorough analysis on how supply voltage and operating frequency impact energy efficiency in a statistical context. With careful library cell selection and robust static RAM design, the Subliminal Processor operates correctly down to 200 mV in a 0.13-mum technology, which is sufficiently low to operate at Vmin. Silicon measurements of the Subliminal Processor show a maximum energy efficiency of 2.6 pJ/instruction at 360 mV supply voltage and 833 kHz operating frequency. Finally, we examine the variation in frequency and Vminacross die to verify our analysis of adaptive tuning of the clock frequency and Vminfor optimal energy efficiency.

Increasing process variation in the nanometer regime motivates the use of statistical static timing analysis tools for timing verification. As device dimensions get smaller, signal integrity effects such as crosstalk noise become more significant. Therefore, it is necessary to accurately model the impact of crosstalk noise on the circuit delay. Process variations cause variability in the crosstalk alignment which leads to the variability in the delay noise. However, most of the existing approaches model delay noise as a worst-case deterministic quantity. In this work, we capture the variability of delay noise by first deriving the closed-form expressions of mean and standard deviation of the delay noise distribution. Next, we obtain the correlation information of the delay noise and use it to represent the delay noise distribution in canonical form. Delay noise, in canonical form, can easily be integrated with existing SSTA tools. We show experimental results that verify the accuracy of our approach.

In recent years, process-induced mechanical stress has emerged as a useful manufacturing technique that enhances carrier transport and increases drive currents. This improvement in current has helped to compensate the decline of device scaling factors in parameters such as tox, Vth, and Vdd. In this work, we propose stress as a means to achieve optimal power-performance trade-off by combining stress-based, performance-enhanced standard cell assignment with dual-Vth assignment. We study how stress-induced performance enhancements are affected by layout properties and improve standard cell layouts so that performance gains are maximized. We then develop a circuit-level, block-based, stress-enhanced optimization algorithm that includes all layout-dependent sources of mechanical stress. By combining the two performance enhancement techniques (stress-based and dual-Vth) for a set of benchmark circuits, we find that our stress-aware optimization, decreases leakage by ~24% on average, for iso-delay, when compared to dual-Vth assignment. Similarly, for iso-leakage, our optimization algorithm reduces delay on average by 5%. In both cases, the proposed method only incurs a small area penalty (< 0.5%).

Modeling and accuracy difficulties exist with traditional SSTA analysis and optimization methods. In this paper we describe methods to improve the efficiency of Monte Carlo-based statistical static timing analysis. We propose a Stratification + Hybrid Quasi Monte Carlo (SH-QMC) approach to reduce the number of samples required for Monte Carlo based SSTA. Our simulations on benchmark circuits up to 90K gates show that the proposed method requires 23.8X fewer samples on average to achieve comparable accuracy in timing estimation as a random sampling approach. Results on benchmark circuits also show that when SH-QMC is performed with multiple parallel threads on a quad core processor, the approach is faster than traditional SSTA with comparable accuracy. SH-QMC scales better than traditional SSTA with circuit size. We also propose an incremental approach to recompute a percentile delay metric after ECO. The results show that on average only 1.4% and 0.7% of original samples need to be evaluated for exact recomputation of the 95th percentile and 99th percentile delays, after sample size reduction using SH-QMC.

SSTA has received a considerable amount of attention in recent years. However, it is a general rule that any approach can only be as accurate as the underlying models. Thus, variation models are an important research topic, in addition to the development of statistical timing tools. These models attempt to predict fluctuations in parameters like doping concentration, critical dimension (CD), and ILD thickness, as well as their spatial correlations. Modeling CD variation is a difficult problem because it contains a systematic component that is context dependent as well as a probabilistic component that is caused by exposure and defocus variation. Since these variations are dependent on topology, modern-day designs can potentially contain thousands of unique CD distributions. To capture all of the individual CD distributions within statistical timing, a transistor-specific model is required. However, statistical CD models used in industry today do not distinguish between transistors contained within different standard cell types (at the same location in a die), nor do they distinguish between transistors contained within the same standard cell. In this work we verify that the current methodology is error-prone using a 90 nm industrial library and lithography recipe (with industrial OPC) and propose a new SSTA delay model that on average reduces error of standard deviation from 11.8% to 4.1% when the total variation (sigma/mu) is 4.9% - a 2.9X reduction. Our model is compatible with existing SSTA techniques and can easily incorporate other sources of variation such as random dopant fluctuation and line-edge roughness.

Gate oxide breakdown is a key factor limiting the useful lifetime of an integrated circuit. Unfortunately, the conventional approach for full chip oxide reliability analysis assumes a uniform oxide-thickness for all devices. In practice, however, gate-oxide thickness varies from die-to-die and within-die and as the precision of process control worsens an alternative reliability analysis approach is needed. In this work, we propose a statistical framework for chip level gate oxide reliability analysis while considering both die-to-die and within-die components of thickness variation. The thickness of each device is modeled as a distinct random variable and thus the full chip reliability estimation problem is defined on a huge sample space of several million devices. We observe that the full chip oxide reliability is independent of the relative location of the individual devices. This enables us to transform the problem such that the resulting representation can be expressed in terms of only two distinct random variables. Using this transformation we present a computationally efficient and accurate approach for estimating the full chip reliability while considering spatial correlations of gateoxide thickness. We show that, compared to Monte Carlo simulation, the proposed method incurs an error of only 1∼6% while improving the runtime by around three orders.

Mobility degradation and device scaling limitations have led process engineers to develop new techniques that introduce mechanical stress in MOSFET channels, which results in enhanced carrier transport. New fabrication steps strive to increase carrier mobility which, consequently, increases both Ionand Ioffin CMOS devices. However, most stress-enhancement techniques are dependent on layout parameters and their effects can be exploited within standard cell library design. In this work, we propose a new standard cell library design methodology that shares VDDand VSSsource/drain connections across standard cell boundaries. Such sharing allows for increased channel stress in both the corresponding device as well as its neighboring devices. Using an industrial 65 nm process and standard cell library, we show that our standard cell design methodology can be seamlessly integrated into current, state-of-the-art digital IC design flows. The new shared source/ drain technique improves critical path delay by 11% on average over a number of benchmarks for only a ~35% increase in leakage. Furthermore, stress-enhanced standard cell libraries offer a superior power/ delay tradeoff compared to dual-Vthacross a wide range of operating points with reduced manufacturing costs. Specifically, our stress-enhanced library (with a single Vth) consumes ~2.5X less leakage than its dual-Vthcounterpart.

Technology scaling reduces gate delays while wire delays may increase. Our work studies the interaction of this phenomenon with technology mapping and its impact on modern EDA flows. In particular, we demonstrate that the use of larger standard cells increases the number of long wires and may undermine circuit delay optimization at 65 nm and below. Experiments with 130 nm, 90 nm, 65 nm, and 45 nm industrial CMOS technology suggest that limiting the use of larger standard cells in technology mapping becomes more effective at 65 nm and 45 nm node, resulting in up to 12% improvement in critical path delay on large benchmark circuits.

Ubiquitous computing has a number of compelling applications ranging from biomedical sensing to environmental monitoring. These computing systems require low cost sensor nodes with volumes ≪1mm3and lifetimes on the order of months or years. We advocate the use of aggressively scaled supply voltages in such applications to maximize energy efficiency. This paper reviews our recent progress in mapping out the low energy design space. We explore the design and test of three low voltage systems targeting ubiquitous computing. We conclude with a survey of open research directions in the ultra-low energy design space.

Ultra Low voltage operation has recently drawn significant attention due to its large potential energy savings. However, typical design practices used for super-threshold operation are not necessarily compatible with the low voltage regime. Here, radically different guidelines may be needed since existing process technologies have been optimized for super-threshold operation. We therefore study the selection of the optimal technology in ultra low voltage designs to achieve minimum energy and minimum variability which are among foremost concerns. We investigate five industrial technologies, from 250nm to 65nm. We demonstrate that mature technologies are often the best choice in very low voltage applications, saving as much as ~1800X in total energy consumption compared to a poorly selected technology. In parallel, the effect of technology choice on variability is investigated, when operating at the energy optimal design point. The results show up to a 4X improvement in delay variation due to global process shift and mismatch when using the most advanced technologies despite their large variability at nominal Vdd.

As technology is aggressively scaled, nano-regime VLSI designs are becoming increasingly susceptible to process variations. Unlike pre-silicon optimization, post-silicon techniques can tune the individual die to better meet the power-delay constraints. This paper proposes a variation-aware methodology for the simultaneous gate sizing and clustering for post-silicon tuning with adaptive body biasing. The proposed methodology uses an accurate table look-up model and fully explores the interaction between gate sizing and optimal body bias based clustering. In addition, it is suitable for industrial test cases with tens of thousands gates. Our optimization methodology includes a body bias distribution alignment strategy to mitigate the impact of critical gates. In this way, the cluster's body bias voltage is not simply determined by only a few critical gates. We also prove the linear dependence between the mean of the body bias probability distribution and the gate size. Based on this, we further investigate a simultaneous sizing and re-clustering algorithm for better leakage savings. A circuit re-balancing and gate snapping scheme is then suggested to map the solution to a standard cell library. Compared with arecently-reported method, the proposed methodology can obtain on average 25.5% leakage saving at nearly the same run time.

Process-induced mechanical stress is used to enhance carrier transport and achieve higher drive currents in current CMOS technologies. In this paper, we study how stress-induced performance enhancements are affected by layout properties and suggest guidelines for improving layouts so that performance gains are maximized. All MOS devices in this work include STI and nitride stress liners as sources of stress. Additionally, the PMOS devices incorporate the stress effects caused by the embedded SiGe S/D layer common in today's processes. First, we study how stress and drive current depend on layout parameters such as active area length and contact placement. We develop an intuition for the drive current dependency on these parameters and propose simple guidelines to improve a layout while considering mechanical stress effects. We then use these guidelines to improve the standard cell layouts in a 65nm industrial library. Experimental results show that we can enhance NMOS and PMOS drive currents by ~5% and ~12%, respectively, while only increasing NMOS leakage current by 1.48X and PMOS leakage current by 3.78X. By applying our guidelines to a 3-input NOR gate and a 3-input NAND gate, we are able to achieve a ~13.5% PMOS drive current improvement in the NOR gate and a ~7% NMOS drive current improvement in the NAND gate, without increasing cell area in either case

Battery life is an important concern for modern embedded processors. Supply voltage scaling techniques can provide an order of magnitude reduction in energy. Current commercial memory technologies have been limited in the degree of supply voltage scaling that can be performed if they are to meet yield and reliability constraints. This has limited designers from exploring the near threshold operating regions for embedded processors. Summarizing prior work we show how proper sizing of memory cells can guarantee that the memory cell reliability in the near threshold supply voltage region matches that of a standard memory cell. However, this robustness comes with a significant area cost. We show how to employ these cells to build cache architectures that greatly reduce energy consumption. We propose an embedded processor based on these new cache architectures that operates in a low power mode, with minimal impact on full performance runtime. The proposed cache uses near threshold tolerant cache ways to reduce access energy combined with traditional cache ways to maintain performance. The access policy of the cache ways is then dynamically reconfigured to obtain energy efficient performance while minimally impacting the high performance mode runtime. Using near threshold cache architectures we show an energy reduction of 53% over a traditional filter cache. For the MIBench embedded benchmarks we show on average an 86% (7.3times) reduction in energy while in low power (10 MHz) mode with only an average 2% increase in runtime in high performance (400 MHz) mode. And for SpecInt applications we show a 77% (4.4times) reduction in energy in low power mode with only an average 4.8% increase in runtime for high performance mode. In addition we show that these trends hold from 130 nm to 45 nm technology nodes.

Static-timing analysis (STA) has been one of the most pervasive and successful analysis engines in the design of digital circuits for the last 20 years. However, in recent years, the increased loss of predictability in semiconductor devices has raised concern over the ability of STA to effectively model statistical variations. This has resulted in extensive research in the so-called statistical STA (SSTA), which marks a significant departure from the traditional STA framework. In this paper, we review the recent developments in SSTA. We first discuss its underlying models and assumptions, then survey the major approaches, and close by discussing its remaining key challenges.

Adaptive body biasing is a powerful technique that allows post-silicon tuning of individual manufactured dies such that each die optimally meets the delay and power constraints. Assigning individual bias control to each gate leads to severe overhead, rendering the method impractical. However, assigning a single bias control to all gates in the circuit prevents the method from compensating for intra-die variation and greatly reduces its effectiveness. In this paper, we propose a new variability-aware method that clusters gates at design time into a handful of carefully chosen independent body-bias groups, which are then individually tuned post-silicon for each die. We show that this allows us to obtain near-optimal performance and power characteristics with minimal overhead. For each gate, we generate the probability distribution of its post-silicon ideal body bias voltage using an efficient sampling method. We then use these distributions and their correlations to drive a statistically aware clustering technique. We study the physical design constraints and show how the area and wirelength overhead can be significantly limited using the proposed method. Compared with a fixed design-time based dual threshold voltage assignment method, we improve leakage power by 38%-68% while simultaneously reducing the standard deviation of delay by two to nine times.

Increasing levels of process variation in current technologies have a major impact on power and performance and result in parametric yield loss. In this paper, we develop an efficient gate-level approach to accurately estimate and optimize the parametric yield, defined by leakage power and delay limits, by finding their joint probability distribution function. We consider inter-die variations, as well as intra-die variations, with correlated and random components. The correlation between power and performance arises due to their dependence on common process parameters and is shown to have a significant impact on the yield, particularly in high-frequency bins. We then propose a new heuristic approach to incrementally compute the gradient of yield with respect to gate sizing and gate-length biasing in the circuit with high efficiency and accuracy. We show how this gradient information can be effectively used by a nonlinear optimizer to perform yield optimization. The proposed yield-analysis approach is compared with Monte Carlo simulations and shows high accuracy, with the yield estimates achieving an average error of 2%. The proposed optimization approach is implemented and tested, and we demonstrate an average yield increase of 40% using gate sizing (as compared to a deterministically optimized circuit). Even higher improvements are demonstrated when both gate sizing and gate-length-biasing techniques are used.

Reliability failure mechanisms, such as time-dependent dielectric breakdown (TDDB), electromigration, and negative bias temperature instability (NBTI), have become a key concern in integrated circuit (IC) design. The traditional approach to reliability qualification assumes that the system will operate at maximum performance continuously under worst case voltage and temperature conditions. In reality, due to widely varying environmental conditions and an increased use of dynamic control techniques, such as dynamic voltage scaling and sleep modes, the typical system spends a very small fraction of its operational time at maximum voltage and temperature. In this paper, we show how this results in a reliability ";slack"; that can be leveraged to provide increased performance during periods of peak processing demand. We develop a novel, real time reliability model based on workload driven conditions. Based on this model, we then propose a new dynamic reliability management (DRM) scheme that results in 20%-35 % performance improvement during periods of peak computational demand while ensuring the required reliability lifetime.

In this paper, we propose a new circuit technique called self-timed regenerator (STR) to improve both speed and power for on-chip global interconnects. The proposed circuits are placed along global wires to compensate the loss in resistive wires and to amplify the effect of wire inductance in the wires to enable transmission line like behavior. For different wire widths, the number of STR and sizing of the transistors are optimized to accelerate the signal propagation while consuming minimum power. In 90-nm CMOS technology, STR design achieved a delay improvement of 14% over the conventional repeater design. Furthermore, 20% power reduction is achieved for iso-delay, and 8% delay improvement for iso-power compared with the repeater design. The proposed technique has also been applied to a clock distribution network, reducing clock power by 26%.

Power supply noise increases the circuit delay, which may lead to performance failure of a design. Decoupling capacitance (decap) addition is effective in reducing the power supply noise, thus making the supply network more robust in presence of large switching currents. Traditionally, decaps have been allocated in order to minimize the worst-case voltage drop occurring in the power grid. In this paper, we propose an approach for timing-aware decap allocation which uses global time slacks to drive the decap optimization. Non-critical gates with larger timing slacks can tolerate a relatively higher supply voltage drop as compared to the gates on the critical paths. The decap allocation is formulated as a non-linear optimization problem using Lagrangian relaxation, and modified adjoint method is used to efficiently obtain the sensitivities of objective function to decap sizes. A fast path-based heuristic is also implemented and compared with the global optimization formulation. The two approaches have been implemented and tested on ISCAS85 benchmark circuits and with grids of different sizes. Compared to uniformly allocated decaps, the proposed approach utilizes 35.5% less total decap to meet the same delay target. For the same total decap budget, the proposed approach is shown to improve the circuit delay by 10.1% on an average.

We present, in this paper, novel algorithms to compute the set of "top-k" aggressors in a design. We show that the computation of the set of top-k aggressors is non-trivial, since we must consider all permutations of aggressors that are coupled to a critical path. Also, different sets of aggressors contribute different amounts of noise to each critical path and a brute-force enumeration to obtain the set of top-k aggressors has impractical runtime. Our proposed approach uses two key techniques to reduce the runtime complexity: Firstly, we model the delay noise propagated from a victim net to its fanout net by a so-called pseudo aggressor, which simplifies our problem formulation significantly. Secondly, we define a dominance property for aggressor sets, which imposes a partial ordering on the aggressor sets and allows us to efficiently prune the enumeration space. We then demonstrate the effectiveness of our proposed algorithm on benehmark circuits.

Subthreshold circuit design is a strong candidate for use in future low power applications. It is not clear, however, that device scaling to 45nm and beyond will be beneficial in Subthreshold circuits. We investigate the implications of device scaling on subthreshold circuits and find that the slow scaling of gate oxide thickness leads to a 60% reduction in Ion/Ioff between the 90nm and 32nm device generations. We highlight the effects of this device degradation on noise margins, delay, and energy. We subsequently propose an alternative scaling strategy and demonstrate significant improvements in noise margins, delay, and energy in sub-Vth circuits.

Subthreshold operation is a promising method for reducing power consumption in ultra-low power applications, such as active RFIDs and sensor networks. It was shown in previous works that operating at the Vmin supply voltage results in optimal energy operation, where Vmin typically falls below the threshold voltage. However, all previous subthreshold analyses ignore the leakage current in standby mode. Hence, for applications where operation at Vmin results in completion of the task well ahead of the required deadline, the energy consumption can be significantly under-estimated. In this paper, we investigate the effect of the non-zero standby energy on the optimal energy consumption in subthreshold operation. We first analyze energy consumption both with and without a cutoff technique in standby mode. Two parameters are proposed to capture the cutoff structure's effect on the energy consumption. Second, a methodology to minimize the total energy consumption is addressed. The selection of the cutoff structure is examined by comparing three different structures. Then, a co-optimization method to optimize the size of the cutoff structure concurrently with the supply voltage, is proposed. This approach reduces energy by 99.2% compared to standby-energy-unaware optimization.

Voltage scaling is desirable in SRAM to reduce energy consumption. However, commercial SRAM is prone to functional failures when Vddis scaled. Several SRAM designs scale Vddto 200-300mV to minimize energy per access, but these designs do not consider SRAM robustness, limiting them to small arrays and sensor type applications. We examine the effects on area and energy for a differential 6T, single-ended 6T with power rail collapsing and an 8T bitcell as Vddis scaled and the bitcells are sized appropriately to maintain robustness. SRAM robustness is examined using importance sampling to reduce simulation runtime. At high voltages, the differential 6T bitcell is the smallest for the same failure rate, but the 8T bitcell is smaller when Vddis scaled below 450mV. For Vddbelow Vth, bitcells must be sized greatly to retain robustness and large arrays become impractical. The differential 6T and 8T designs have the lowest dynamic energy consumption, and the single-ended 6T design has the lowest leakage. The supply voltage for minimum energy operation depends on cache configuration and can be well above Vthfor large caches with low dynamic activity.

Modeling the effect of coupling noise on circuit delay is a key issue in static timing analysis (STA) and involves the “victimaggressor alignment” problem. As delay-noise depends strongly on the skew between the victim-aggressor input transitions’, it is not possible to apriori identify the victim input transition that results in the latest arrival time at the victim. Several approaches that heuristically search for the worst-case victim-aggressor alignment have been proposed in literature. In this paper we present an analytical result that obviates the need to search for the worst-case victim input transition, thereby simplifying the victim-aggressor alignment problem significantly. Using the properties of standard nonlinear CMOS drivers, we show that regardless of the switching of the aggressors, the worst-case victim input transition is the one that switches at the latest point in its timing window. Although this result has been empirically observed in the industry, to the best of our knowledge, this is the first work that provides a rigorous analysis and shows that the result holds for both linear and non-linear drivers. We also show that limiting the alignment of the victim to only the latest victim input transition can significantly reduce the runtime of existing heuristic techniques with no loss of accuracy.

Parameter variations cause high yield losses due to their large impact on circuit delay. In this paper, we propose the use of so-called soft-edge flip-flops as an effective way to mitigate these yield losses. Soft-edge flip-flops have a small window of transparency (ranging from 0.25-3 FO4) instead of a hard edge, allowing limited cycle stealing on critical paths, and thus compensating for delay variations. By enabling time borrowing, soft-edge flip-flops allow random delay variations to average out across multiple logic stages. In addition, they address small amounts of delay unbalance between logic stages, further maximizing the frequency of operation. We develop a library of soft-edge flip-flops with varying amounts of softness. We show that the power and area overhead of soft-edge flip-flops grows directly with the amount of softness. We then propose a statistically aware flip-flop assignment algorithm that maximizes the gain in timing yield while minimizing the incurred power overhead. Experimental results on a wide range of benchmark circuits show that the proposed approach improves the mean delay by 1.9-22.3% while simultaneously reducing the standard deviation of delay by 1.9-24.1% while increasing power by a small amount (0.3-2.8%).

In this paper, we propose a new edge encoding technique to reduce the energy consumption in multi-cycle interconnects. Both average and worst-case energy are reduced by desynchronizing the edges of rising and falling transitions. In a 1.2V 65nm CMOS technology, the approach achieves up to 31% energy reduction with no latency overhead over optimally designed conventional busses due to coupling capacitance reductions. The technique further reduces energy consumption by 38% with iso-throughput at the expense of one-cycle latency. Energy savings are shown to be more robust to process variations than previous techniques.

Subthreshold circuit design has become a popular approach for building energy efficient digital circuits. One drawback is performance degradation due to the exponentially reduced driving current. This had limited subthreshold circuits to relatively low performance applications such as sensor networks. To retain the excellent energy efficiency while reducing performance loss, we propose to apply subthreshold and near-threshold techniques to chip multi-processors. We show that an architecture where several slower cores are clustered together with a shared faster L1 cache is optimal for energy efficiency, because processor cores and memory operate best at different supply and threshold voltages. In particular, SPLASH2 benchmarks show about a 53% energy improvement over the traditional CMP approach (about 70% over a single core machine).

Soft errors have emerged as an important reliability challenge for nanoscale very large scale integration designs. In this paper, we present a fast and efficient soft error rate (SER) analysis methodology for combinational circuits. We first present a novel parametric waveform model based on the Weibull function to represent particle strikes at individual nodes in the circuit. We then describe the construction of the descriptor object that efficiently captures the correlation between the transient waveforms and their associated rate distribution functions. The proposed algorithm consists of operations to inject, propagate, and merge these descriptors while traversing forward along the gates in a circuit. The parameterized waveforms enable an efficient static approach to calculate the SER of a circuit. We exercise the proposed approach on a wide variety of combinational circuits and observe that our algorithm has linear runtime with the size of the circuit. The runtimes for soft error estimation were observed to be in the order of about 1 s, compared to several minutes or even hours for previously proposed methods.

Reliability failure mechanisms, such as time dependent dielectric breakdown, electromigration, and thermal cycling have become a key concern in processor design. The traditional approach to reliability qualification assumes that the processor will operate at maximum performance continuously under worst case voltage and temperature conditions. However, the typical processor spends a very small fraction of its operational time at maximum voltage and temperature. In this paper, we show how this results in a reliability "slack" that can be leveraged to provide increased performance during periods of peak processor demand. We develop a novel, real time reliability model based on workload driven conditions. We then propose a new dynamic reliability management (DRM) scheme that results in 20-35% performance improvement during periods of peak computational demand while ensuring the required reliability lifetime.

Soft errors have emerged as an important reliability challenge for nanoscale VLSI designs. In this paper, we present a fast and efficient soft error rate (SER) computation algorithm for combinational circuits. We first present a novel parametric waveform model based on the Weibull function to represent particle strikes at individual nodes in the circuit. We then describe the construction of the SET descriptor that efficiently captures the correlation between the transient waveforms and their associated rate distribution functions. The proposed algorithm consists of operations to inject, propagate and merge SET descriptors while traversing forward along the gates in a circuit. The parameterized waveforms enable an efficient static approach to calculate the SER of a circuit. We exercise the proposed approach on a wide variety of combinational circuits and observe that our algorithm has linear runtime with the size of the circuit. The runtimes for soft error estimation were observed to be in the order of about one second, compared to several minutes or even hours for previously proposed methods

Statistical static timing analysis (SSTA) is emerging as a solution for predicting the timing characteristics of digital circuits under process variability. For computing the statistical max of two arrival time probability distributions, existing analytical SSTA approaches use the results given by Clark in [8]. These analytical results are exact when the two operand arrival time distributions have jointly Gaussian distributions. Due to the nonlinear max operation, arrival time distributions are typically skewed. Furthermore, nonlinear dependence of gate delays and non-gaussian process parameters also make the arrival time distributions asymmetric. Therefore, for computing the max accurately, a new approach is required that accounts for the inherent skewness in arrival time distributions. In this work, we present analytical solution for computing the statistical max operation.1 First, the skewness in arrival time distribution is modeled by matching its first three moments to a so-called skewed normal distribution. Then by extending Clark's work to handle skewed normal distributions we derive analytical expressions for computing the moments of the max. We then show using initial simulations results that using a skewness based max operation has a significant potential to improve the accuracy of the statistical max operation in SSTA while retaining its computational efficiency.

Statistical static timing analysis (SSTA) has become a key method for analyzing the effect of process variation in aggressively scaled CMOS technologies. Much research has focused on the modeling of spatial correlation in SSTA. However, the vast majority of these works used artificially generated process data to test the proposed models. Hence, it is difficult to determine the actual effectiveness of these methods, the conditions under which they are necessary, and whether they lead to a significant increase in accuracy that warrants their increased runtime and complexity. In this paper, we study 5 different correlation models and their associated SSTA methods using 35420 critical dimension (CD) measurements that were extracted from 23 reticles on 5 wafers in a 130nm CMOS process. Based on the measured CD data, we analyze the correlation as a function of distance and generate 5 distinct correlation models, ranging from simple models which incorporate one or two variation components to more complex models that utilize principle component analysis and Quad-trees. We then study the accuracy of the different models and compare their SSTA results with the result of running STA directly on the extracted data. We also examine the trade-off between model accuracy and run time, as well as the impact of die size on model accuracy. We show that, especially for small dies (< 6.6mm x 5.7mm), the simple models provide comparable accuracy to that of the more complex ones, while incurring significantly less runtime and implementation difficulty. The results of this study demonstrate that correlation models for SSTA must be carefully tested on actual process data and must be used judiciously.

Adaptive body biasing (ABB) is a powerful technique that allows post-silicon tuning of individual manufactured dies such that each die optimally meets the delay and power constraints. Assigning individual bias control to each gate leads to severe overhead, rendering the method impractical. However, assigning a single bias control to all gates in the circuit prevents the method from compensating for intra-die variation and greatly reduces its effectiveness. In this paper, we propose a new variability-aware method that clusters gates at design time into a handful of carefully chosen independent body bias groups, which are then individually tuned post-silicon for each die. We show that this allows us to obtain near-optimal performance and power characteristics with minimal overhead. For each gate, we generate the probability distribution of its post-silicon ideal body bias voltage using an efficient sampling method. We then use these distributions and their correlations to drive a statistically-aware clustering technique. We study the physical design constraints and show how the area and wirelength overhead can be significantly limited using the proposed method. Compared to a fixed design time based dual threshold voltage assignment method, we improve leakage power by 38-71% while simultaneously reducing the standard deviation of delay by 2-9X.

Soft errors in logic are emerging as a significant reliability problem for VLSI designs. This paper presents novel circuit optimization techniques to mitigate soft error rates (SER) of combinational logic circuits. First, we propose a gate sizing algorithm that trades off SER reduction and area overhead. This approach first computes bounds on the maximum achievable SER reduction by resizing a gate. This bound is then used to prune the circuit graph, arriving at a smaller set of candidate gates on which we perform incremental sensitivity computations to determine the gates that are the largest contributors to circuit SER. Second, we propose a flipflop selection method that uses slack information at each primary output node to determine the flipflop configuration that produces maximum SER savings. This approach uses an enhanced flipflop library that contains flipflops of varying temporal masking ability. Third, we propose a unified, cooptimization approach combining flipflop selection with the gate sizing algorithm. The joint optimization algorithm produces larger SER reductions while incurring smaller circuit overhead than either technique taken in isolation. Experimental results on a variety of benchmarks show SER reductions of 7.9X with gate sizing, 6.6X with flipflop assignment, and 28.2X for the combined optimization approach, with no delay penalties and area overheads within 5--6%. The runtimes for the optimization algorithms are on the order of 1--3 minutes.

The placement of on-die decoupling capacitors (decap) between the power and ground supply grids has become a common practice in high performance processor designs. In this paper, we present a new active decap circuit that significantly increase the effectiveness of decap in suppressing power supply fluctuations. The proposed circuit senses the supply drop and drives an amplified and inverted voltage fluctuation on the decap. The active decoupling circuit is powered by a separate power supply and we study the optimal allocation of the total C4s/pads between this second power supply and the regular supply as well as the optimal allocation of the total decoupling capacitance between actively switched and traditional static decap. Finally, we demonstrate that the overhead of the proposed method is small compared to the area of the decaps. Using the proposed method, the maximum supply drop is reduced by 45% compared to the use of only traditional decap, corresponding to an increase in the effective decap of approximately 8times.

Subthreshold design has become an important area in low-power design due to its ultra-low power consumption and high energy efficiency. This is very useful in mobile applications where battery life is crucial. For most current DVS processor designs, the voltage range is limited from full Vddto approximately half Vddat most. Subthreshold design enables wide-range dynamic voltage scaling (DVS) by allowing circuits to operate in subthreshold voltages. In this paper we give an overview of several issues in energy-efficient subthreshold design. First, from a theoretical point of view, we show that, for subthreshold supply-voltages, leakage energy becomes dominant, making "just in time completion" energy inefficient. We derive an analytical model for the minimum energy-optimal voltage and study its trends with technology scaling. Second, we evaluate different low-power approaches such as MTCMOS (multiple-threshold CMOS) and DVS and show that wide-range DVS provides the best energy efficiency. Finally, we study the impact of process variation on subthreshold operation and discuss which sources of process variation are dominant in this regime and how they can be addressed

Mobile applications with battery lifetimes on the order of thousands of days have placed stringent energy requirements on circuits. In this paper, we propose a new energy optimization technique for ultra-low energy circuits operating in the subthreshold regime. Our technique uses simultaneous gate sizing and supply voltage scaling to reduce energy. We demonstrate the effectiveness of our technique on benchmark circuits and offer insight on the roles of the timing distribution and wire capacitance in determining the achievable energy reductions.

Recent progress in the development of subthreshold circuit design techniques has created the opportunity for dramatic energy reductions in many applications. However, energy efficiency comes at the price of timing and energy variability due to process variations. We explore energy optimality in the subthreshold regime, discuss variability in this region, and highlight the energy and variability characteristics of a real subthreshold design.

In this paper, statistical models for the efficient analysis of interconnect delay and crosstalk noise in the presence of back-end process variations are developed. The proposed models enable closed-form computation of means and variances of interconnect-delay, crosstalk-noise peak, and coupling-induced-delay change for given magnitudes of variation in relevant process parameters, such as linewidth, metal thickness, metal spacing, and interlayer dielectric (ILD) thickness. The proposed approach is based on the observation that if the variations in different physical dimensions are assumed to be independent normal random variables, then the interconnect behavior also tends to have a Gaussian distribution. In the proposed statistical models, delay and noise are expressed directly as functions of changes in physical parameters. This formulation allows us to preserve all correlations and can be very useful in evaluating delay and noise sensitivities due to changes in various physical dimensions. For interconnect-delay computation, the authors express the resistance and capacitance of a line as a linear function of random variables and then use these to compute circuit moments. They show that ignoring higher order terms in the resulting variational moments does not result in a loss of accuracy. Finally, these variability-aware moments are used in known closed-form delay and slew metrics to compute interconnect-delay probability density functions (pdfs). Similarly for coupling noise and dynamic-delay analysis, the authors rely on the linearity (Gaussian) assumption, allowing us to truncate nonlinear terms and express noise and dynamic-delay pdfs as linear functions of variations in relevant geometric dimensions. They compare their approach to SPICE-based Monte Carlo simulations and report the error in mean and standard deviation of interconnect delay to be 1% and 4% on average, respectively

At current operating frequencies, inductive-coupling effects can be significant and should be included for accurate crosstalk-noise analysis. In this paper, an analytical framework to model crosstalk noise in coupled RLC interconnects is presented. The proposed model is based on transmission-line theory and captures high-frequency effects in on-chip interconnects. The new model is generic in nature and can be applied to asymmetric driver-and-line configurations for aggressor and victim wires. The model is compared against SPICE simulations and is shown to capture both the waveform shape and peak noise accurately. Over a large set of random test cases, the average error in noise-peak estimation is approximately 6.5%. A key feature of the new model is that its derivation and form enables physical insight into the total coupling-noise-waveform shape and its dependence on relevant physical-design parameters. Due to its simplicity and physical nature, the proposed model can be applied to investigate the impact of various physical-design optimizations (e.g., wire sizing and spacing, shield insertion) on total RLC coupled noise. The effectiveness of various existing noise-reduction techniques in the presence of mutual-inductance coupling is studied here. The obtained results indicate that common (capacitive) noise-avoidance techniques can behave quite differently when both capacitive and inductive coupling are considered together.

In addition to traditional constraints on frequency, leakage current has emerged as a stringent constraint in modern processor designs. Since leakage current exhibits a strong inverse correlation with circuit delay, effective parametric yield prediction must consider the dependence of leakage current on frequency. In this paper, a new chip-level statistical method to estimate the total leakage current in the presence of within-die and die-to-die variability is presented. A closed-form equation for total chip leakage that models the dependence of the leakage current distribution on different process parameters is developed. The proposed analytical expression is obtained directly from pertinent design information and includes both subthreshold and gate leakage currents. Using this model, an integrated approach to accurately estimate the yield loss when both frequency and power limits are imposed on a design is then presented. The proposed method demonstrates the importance of considering both these limiting factors while calculating the yield of a lot

Runtime leakage current, defined as circuit leakage during normal operation (i.e., nonstandby mode), has become a major concern in very advanced technologies along with traditional standby mode leakage. In this paper, we propose a new leakage reduction method that specifically targets runtime leakage current. We first observe that the state probabilities of nodes in a circuit tend to be skewed, meaning that they have either high or low values. We then propose a method that exploits these skewed state probabilities by setting only those transistors to high-$V_t$(thick-oxide) that have a high likelihood of being off (on) and, hence, contributing significantly to the total runtime leakage. Accordingly, we also propose a library specifically tailored to the proposed approach, where$V_t$and$T_ ox$assignment with favorable tradeoffs under skewed input probabilities is provided. For further leakage reduction, we also introduce circuit resynthesis using pin reordering, pin rewiring, mapping, and decomposition. The optimization algorithm shows substantial leakage improvement over probability unaware optimization using a traditional standard cell library.

In this paper, we develop an approach for statistical modeling of crosstalk noise and dynamic delay degradation in coupled RC interconnects under process variations. The proposed model enables closed-form computation of mean and variance of noise peak and worst case dynamic delay for given variabilities in physical dimensions. We compare the proposed model against HSPICE Monte Carlo simulations and report an average error in mean and standard deviation of noise peak to be 2.7% and 3.7% respectively.

In this paper, we present a novel approach to obtain any desired intermediate threshold voltage in a dual VT process. The intermediate threshold voltages are achieved by combining low and high threshold voltages in a device. We show that this combination can be easily implemented in layouts with negligible design and manufacturing overhead. Our results show that power-delay characteristics of the achieved intermediate thresholds match well with the ideal (but impractical) scenario that assumes that all intermediate thresholds are available in the technology.

The progressive trend of fabrication technologies towards the nanometer regime has created a number of new physical design challenges for computer architects. Design complex-ity, uncertainty in environmental and fabrication conditions, and single-event upsets all conspire to compromise system correctness and reliability. Recently, researchers have be-gun to advocate a new design strategy, called Better Than Worst-Case design, that couples a complex core component with a simple reliable checker mechanism. By delegating the responsibility for correctness and reliability of the design to the checker, it becomes possible to build correct-certified designs that effectively address the challenges of deep sub-micron design. In this paper, we present the concepts of Better Than Worst-Case design and highlight two exemplary designs: the DIVA checker and Razor logic. We show how this approach to system implementation relaxes design constraints on core components, which reduces the effects of physical design challenges and creates opportunities to optimize performance and power characteristics. We demonstrate the advantages of relaxed design constraints for the core components by applying typical-case optimization (TCO) techniques to an adder circuit. By analyzing the carry-propagation charac-teristics of real programs, it is possible to design an adder circuit that when incorporated into a Better Than Worst-Case design exhibits significantly reduced latency. Finally, we discuss the challenges and opportunities posed to CAD tools in the context of Better Than Worst-Case design. In particular, additional support is required for analyzing run-time characteristics of designs, and many opportunities are created to incorporate typical-case optimizations into syn-thesis, testing and verification. 1.

Leakage power is emerging as a key challenge in IC design. Leakage is increasingly exponentially with each technology generation and is expected to become the dominant part of total power. Device threshold voltage scaling, shrinking device dimensions, and larger circuit sizes are causing this dramatic increase in leakage. As leakage varies exponentially with process parameters, yield of the chip is often directly influenced by leakage. Increasing amount of leakage is also critical for power constraint ICs. Traditionally, leakage has been considered as an important design variable in handheld devices and in standby circuit operation. However, this significant increase of leakage now warrants that it be considered as the key design variable in all IC designs.This tutorial presents a comprehensive review of leakage power issues in IC design. The tutorial is organized in four major parts. The first part provides an overview of technology and scaling trends which are causing the significant increase in leakage current. The device physics that leads to sub-threshold and gate leakage will be described, along with their dependence on circuit design variables. This part of the tutorial will also cover basic transistor and circuit techniques to minimize leakage, such as the stack effect.The second part of the tutorial will focus on circuit level leakage estimation and avoidance. Use of multiple threshold voltages has been very successful in controlling the leakage of the circuit. Comprehensive description of multiple-Vt techniques for leakage avoidance will be presented along with associated leakage estimation techniques. Multiple-threshold design (MTCMOS) will be described along with its leakage benefits and performance trade-offs. Multiple oxide technology options and associated impact on gate leakage will also be discussed.Third part of the tutorial focuses on chip level effects on leakage. Leakage is heavily dependent on local and global process variations and can vary by an order of magnitude over the technology spread. Leakage estimation techniques which consider both inter and intra-die process variations will be covered. This part of the tutorial also focuses on chip-level leakage minimization techniques. Leakage minimization techniques such as Adaptive Body Bias (ABB) and power supply control will be presented.The last part of the tutorial covers system and circuit architectures for leakage avoidance. In standby mode, the leakage of the circuit can be lowered by putting it a low-leakage state. Caches and memory circuits occupy large percentage of area in model chips. The leakage of caches and memories need to be carefully controlled. This section of the tutorial will cover topics including state assignment for leakage minimization, leakage-driven memory and cache circuits and architectures.The tutorial is intended for designers and CAD engineers interested in next generation design techniques and methodologies and emerging power challenges. Basic background of VLSI and CAD is useful though not needed.

With process scaling runtime leakage current, when the circuit is operating, has become a major concern in addition to traditional standby mode leakage. In this paper we propose a new leakage reduction method that specifically targets runtime leakage current. We first observe that the state probabilities of nodes in a circuit tend to be skewed, meaning that they have either a high or a low value. We then propose a method that exploits these skewed state probabilities by setting only those transistors to high-Vt (thick-oxide) that have a high likelihood of being OFF (ON) and hence contributing significantly to the total runtime leakage. Accordingly, we also propose a library specifically tailored for the proposed approach, where Vt and Tox assignment with favorably trade-offs under skewed input probabilities are provided. The optimization algorithm performs simultaneous sizing, Vt and Tox assignment and shows substantial leakage improvement over probability-unaware optimization.

In this paper we present a second-generation sensor network processor which consumes less than one picoJoule per instruction (typical processors use 100's to 1000's of picoJoules per instruction). As in our first-generation design effort, we strive to build microarchitectures that minimize area to reduce leakage, maximize transistor utility to reduce the energy-optimal voltage, and optimize CPI for efficient processing. The new design builds on our previous work to develop a low-power subthreshold-voltage sensor processor, this time improving the design by focusing on ISA, memory system design, and microarchitectural optimizations that reduce overall design size and improve energy-per-instruction. The new design employs 8-bit datapaths and an ultra-compact 12-bit wide RISC instruction set architecture, which enables high code density via micro-operations and flexible operand modes. The design also features a unique memory architecture with prefetch buffer and predecoded address bits, which permits both faster access to the memory and smaller instructions due to few address bits. To achieve efficient processing, the design incorporates branch speculation and out-of-order execution, but in a simplified form for reduced area and leakage-energy overheads. Using SPICE-level timing and power simulation, we find that these optimizations produce a number of Pareto-optimal designs with varied performance-energy tradeoffs. Our most efficient design is capable of running at 142 kHz (0.1 MIPS) while consuming only 600 fJ/instruction, allowing the processor to run continuously for 41 years on the energy stored in a miniature 1g lithium-ion battery. Work is ongoing to incorporate this design into an intra-ocular pressure sensor.

In this paper, we propose a new sensitivity based, statistical gate sizing method. Since circuit optimization effects the entire shape of the circuit delay distribution, it is difficult to capture the quality of a distribution with a single metric. Hence, we first introduce a new objective function that provides an effective measure for the quality of a delay distribution for both ASIC and high performance designs. We then propose an efficient and exact sensitivity based pruning algorithm based on a newly proposed theory of perturbation bounds. A heuristic approach for sensitivity computation which relies on efficient computation of statistical slack is then introduced. Finally, we show how the pruning and statistical slack based approaches can be combined to obtain nearly identical results compared with the brute-force approach but with an average run-time improvement of up to 89x. We also compare the optimization results against that of a deterministic optimizer and show an improvement up to 16% in the 99-percentile circuit delay and up to 31% in the standard deviation for the same circuit area.

Process variability greatly affects power and timing of nanometer scale CMOS circuits, leading to parametric yield loss due to both timing and power constraint violations. This parametric yield loss will continue to worsen in future technologies as a result of increasing process variations [1] and the increased importance of leakage power. Hence, statistical techniques are required to maximize parametric yield under given power and frequency constraints. Recently, much progress has been reported in the area of statistical modeling of leakage power [6] and circuit timing [2-5]. These techniques are useful in analyzing the impact of process variations on performance and power in nanometer CMOS designs. In this extended abstract, we outline the need for statistical optimization methods.

Increasing levels of process variation in current technologies have a major impact on power and performance, and result in parametric yield loss. In this work we develop an efficient gate-level approach to accurately estimate the parametric yield defined by leakage power and delay constraints, by finding the joint probability distribution function (jpdf) for delay and leakage power. We consider inter-die variations as well as intra-die variations with correlated and random components. The correlation between power and performance arise due to their dependence on common process parameters and is shown to have a significant impact on yield in high-frequency bins. We also propose a method to estimate parametric yield given the power/delay jpdf that is much faster than numerical integration with good accuracy. The proposed approach is implemented and compared with Monte Carlo simulations and shows high accuracy, with the yield estimates achieving an average error of 2%.

The increased dominance of intra-die process variations has motivated the field of statistical static timing analysis (SSTA) and has raised the need for SSTA-based circuit optimization. We propose a new sensitivity based, statistical gate sizing method. Since brute-force computation of the change in circuit delay distribution to gate size change is computationally expensive, we propose an efficient and exact pruning algorithm. The pruning algorithm is based on a novel theory of perturbation bounds which are shown to decrease as they propagate through the circuit. This allows pruning of gate sensitivities without complete propagation of their perturbations. We apply our proposed optimization algorithm to ISCAS benchmark circuits and demonstrate the accuracy and efficiency of the proposed method. Our results show an improvement of up to 10.5% in the 99-percentile circuit delay for the same circuit area, using the proposed statistical optimizer and a run time improvement of up to 56/spl times/ compared to the brute-force approach.

On-chip buses are typically designed to meet performance constraints for worst-case conditions, including process corner, temperature, IR-drop, and neighboring net switching pattern. This can result in significant performance slack at more typical operating conditions. We propose a dynamic voltage scaling (DVS) technique for buses, based on a double sampling latch which can detect and correct for delay errors without the need for retransmission. The proposed approach recovers the available slack at non-worst-case operating points through more aggressive voltage scaling and track changing conditions by monitoring the error recovery rate. Voltage margins needed in traditional designs to accommodate worst-case performance conditions are therefore eliminated, resulting in a significant improvement in energy efficiency. The approach was implemented for a 6 mm memory read bus operating at 1.5 GHz (0.13 /spl mu/m technology node) and was simulated for a number of benchmark programs. Even at the worst-case process and environment conditions, energy gains of up to 17% are achieved, with error recovery rates under 2.3%. At more typical process and environment conditions, energy gains range from 35% to 45%, with a performance degradation under 2%. An analysis of optimum interconnect architectures for maximizing energy gains with this approach shows that the proposed approach performs well with technology scaling.

In this paper, we have modelled the flip-flop clock to output delay dependency on the data arrival time and introduced this phenomenon in timing analysis. Traditionally, finding the minimum clock period of a flip-flop based sequential design was based on the assumption that the setup-time and clock to output delay of a flip-flop are constant and hence each stage of the pipeline can be analyzed independently. However, it is well known that the delay of a flip-flop depends on the data arrival time at its input and hence there exists an interdependence among different pipeline stages. The problem of finding the minimum clock period of such a coupled system is a non-trivial problem. In this paper, we formu-late the problem of finding the minimum clock period of a flip-flop based sequential circuit accounting for these dependencies. We show that the problem is a non-linear convex optimization problem. We propose three different solution approaches and compare their results on ISCAS '89 sequential benchmark circuits. Modeling these data arrival time dependencies we have seen a consistent decrement of approximately 50-60ps compared to the traditional approach using constant setup-time and flip-flop delays. We also show how the analysis can be extended to account for hold time constraints for short paths in the circuit.

With the increased significance of leakage power and performance variability, the yield of a design is becoming constrained both by power and performance limits, thereby significantly complicating circuit optimization. In this paper, we propose a new optimization method for yield optimization under simultaneous leakage power and performance limits. The optimization approach uses a novel leakage power and performance analysis that is statistical in nature and considers the correlation between leakage power and performance to enable accurate computation of circuit yield under power and delay limits. We then propose a new heuristic approach to incrementally compute the gradient of yield with respect to gate sizes in the circuit with high efficiency and accuracy. We then show how this gradient information can be effectively used by a non-linear optimizer to perform yield optimization. We consider both inter-die and intra-die variations with correlated and random components. The proposed approach is implemented and tested and we demonstrate up to 40% yield improvement compared to a deterministically optimized circuit.

We present a new gate delay model for accurate modeling of difficult waveform shapes, such as those resulting from coupling capacitance noise, inductive ringing and resistive shielding. Our modeling approach uses a process of time shifting and time stretching of a set of so-called base-waveforms. These base-waveforms are selected from a large set of noisy waveform shapes that occur in interconnect structures with coupling and other types of noise, such that the delay error across all considered waveforms is minimized. Depending on the desired accuracy one or more base-waveforms can be used. This method is also used to model the gate output waveforms allowing for closure, in terms of the used base-waveforms across a circuit library. We show that the determination of the optimal set of base-waveforms under such input-to-output closure is exponential in complexity. We, therefore, propose an heuristic approach that maps the problem to the unate covering problem for which efficient solution methods are available. The new delay model can be applied in a timing analysis program with minor changes. We present results that demonstrate the accuracy of the new delay model for waveforms perturbed with noise for a large set of waveform shapes.

Power supply integrity verification has become a key concern in high performance designs. In deep submicron technologies, power supply noise can significantly increase the circuit delay and lead to performance failures. Traditional static timing analysis which applies worst-case voltage margins to compute circuit delay leads to a very conservative analysis because the worst-case drop is localized to a small area of the die. In this paper, we propose a new approach for analyzing the impact of power supply variations on circuit delay. The circuit delay maximization problem is formulated as a constrained non-linear optimization problem which takes both IR and Ldi/dt drops into account The proposed approach does not require apriori knowledge of critical paths in the circuit and can be effectively incorporated in an existing static timing analysis framework. The proposed method has been implemented and tested on ISCAS85 benchmark circuits and compared with the traditional methods for computing worst-case circuit delay under supply variations.

This paper presents a novel approach towards the simultaneous Vt-assignment and gate-sizing problem. This inherently discrete problem is formulated as a continuous problem, allowing it to be solved using any of several widely available and highly efficient non-linear optimizers. We prove that, under our formulation, the optimal solution has discrete Vts assigned to almost every gate, thus eliminating the need for a sophisticated snapping heuristic. We show that this technique performs dual-Vt assignment and gate sizing in a very efficient manner. Compared to a sensitivity based method, we achieve average leakage savings of 31% and average total power savings of 7.4% with very efficient runtimes.

Sensor network processors and their applications are a growing area of focus in computer system research and design. Inherent to this design space is a reduced processing performance requirement and extremely high energy constraints, such that sensor network processors must execute low-performance tasks for long durations on small energy supplies. In this paper, we demonstrate that subthreshold-voltage circuit design (400 mV and below) lends itself well to the performance and energy demands of sensor network processors. Moreover, we show that the landscape for microarchitectural energy optimization dramatically changes in the subthreshold domain. The dominance of leakage power in the subthreshold regime demands architectures that i) reduce overall area; ii) increase the utility of transistors; while iii) maintaining acceptable CPI efficiency. We confirm these observations by performing SPICE-level analysis of 21 sensor network processors and memory architectures. Our best sensor platform, implemented in 130nm CMOS and operating at 235 mV, only consumes 1.38 pJ/instruction, nearly an order of magnitude less energy than previously published sensor network processor results. This design, accompanied by bulk-silicon solar cells for energy scavenging, has been manufactured by IBM and is currently being tested.

Subthreshold circuit design is a compelling method for ultra-low power applications. However, subthreshold designs show dramatically increased sensitivity to process variations due to the exponential relationship of subthreshold drive current with V th variation. In this paper, we present an analysis of subthreshold energy efficiency considering process variation, and propose methods to mitigate its impact. We show that, unlike superthreshold circuits, random dopant fluctuation is the dominant component of variation in subthreshold operation. We investigate how this variability can be ameliorated with proper circuit sizing and choice of circuit logic depth. We then present a statistical analysis of the energy efficiency of subthreshold circuits considering process variations. We show that the energy optimal supply voltage increases due to process variations and study its dependence on circuit parameters. We verify our analytical models against Monte Carlo SPICE simulations and show that they accurately predict the minimum energy and energy optimal supply voltage. Finally, we use the developed statistical energy model to determine the optimal pipelining depth in subthreshold designs

Buffer insertion is an important technique used to achieve timing closure in high performance VLSI designs. As the number of buffers in ASIC designs has increased with process scaling, the power con-sumption of buffers has become a critical concern. In this paper, we present an efficient algorithm that performs van Ginneken style buffer insertion on RC trees and minimizes the total power con-sumption under a given delay constraint. Our algorithm is based on a formulation that uses a buffer library consisting of continuous buffer sizes. We construct solution candidates in the form of surfaces in the 3-D delay, capacitance and power (DCP) space and show the mecha-nisms to propagate and merge them in the interconnect tree. Instead of a single minimal power solution, the algorithm produces an entire DCP surface from which a suitable solution point can be selected. We also present a post-processing step where buffers with continu-ous (non-standard) sizes are snapped to discrete size values corre-sponding to the buffers in a given library. The proposed algorithm has a worst-case runtime complexity that is polynomial (quadratic) in the number of possible buffer locations. We implemented and tested our proposed algorithm on a number of large benchmark nets and observed that our method produces a speedup in runtime of 5-6X in comparison with previous power aware buffer insertion methods.

This paper presents a new method based on Bayesian networks (BNs) for computing the exact probability distribution of the delay of a circuit. The method is based on BNs, which allows an efficient means to factor the joint probability distributions over variables in a circuit graph. The space complexity of the method presented here is O(m/sup |C|/), where m is the number of distinct values taken by each delay variable and |C| is the number of variables in the largest clique. The maximum clique size present in a BN is shown to be much smaller than the circuit size. For large circuits, where it is not practically feasible to compute the exact distribution, methods to reduce the problem size and get a lower bound on the exact distribution are presented. Comparison of the results with Monte Carlo simulations shows that we can reduce the size of the circuit by as much as 89% while maintaining the maximum difference between the predicted and simulated 3/spl sigma/ values to be less than 3%.

Standby leakage current minimization is a pressing concern for mobile applications that rely on standby modes to extend battery life. In this paper, we propose new leakage current reduction methods in standby mode. First, we propose a combined approach of sleep-state assignment and threshold voltage (V/sub t/) assignment in a dual-V/sub t/ process for subthreshold leakage (I/sub sub/) reduction. Second, for the minimization of gate oxide leakage current (I/sub gate/) which has become comparable to I/sub sub/ in 90-nm technologies, we extend the above method to a combined sleep-state, V/sub t/ and gate oxide thickness (T/sub ox/) assignments approach in a dual-V/sub t/ and dual-T/sub ox/ process to minimize both I/sub sub/ and I/sub gate/. By combining V/sub t/ or V/sub t//T/sub ox/ assignment with sleep-state assignment, leakage current can be dramatically reduced since the circuit is in a known state in standby mode and only certain transistors are responsible for leakage current and need to be considered for high-V/sub t/ or thick-T/sub ox/ assignment. A significant improvement in the leakage/performance tradeoff is therefore achievable using such combined methods. We formulate the optimization problem for simultaneous state/V/sub t/ and state/V/sub t//T/sub ox/ assignments under delay constraints and propose both an exact method for its optimal solution as well as two practical heuristics with reasonable run time. We implemented and tested the proposed methods on a set of synthesized benchmark circuits and show substantial leakage current reduction compared to the previous approaches using only state assignment or V/sub t/ assignment alone.

On-chip L1 and L2 caches represent a sizeable fraction of the total power consumption of microprocessors. In nanometer-scale technology, the subthreshold leakage power is becoming one of the dominant total power consumption components of those caches. In this study, we present optimization techniques to reduce the subthreshold leakage power of on-chip caches assuming that there are multiple threshold voltages, V/sub T/'s, available. First, we show a cache leakage optimization technique that examines the tradeoff between access time and subthreshold leakage power by assigning distinct V/sub T/'s to each of the four main cache components-address bus drivers, data bus drivers, decoders, and static random access memory (SRAM) cell arrays with sense amplifiers. Second, we show optimization techniques to reduce the leakage power of L1 and L2 on-chip caches without affecting the average memory access time. The key results are: 1) two additional high V/sub T/'s are enough to minimize leakage in a single cache-3 V/sub T/'s if we include a nominal low V/sub T/ for microprocessor core logic; 2) if L1 size is fixed, increasing L2 size can result in much lower leakage without reducing average memory access time; 3) if L2 size is fixed, reducing L1 size may result in lower leakage without loss of the average memory access time for the SPEC2K benchmarks; and 4) smaller L1 and larger L2 caches than are typical in today's processors result in significant leakage and dynamic power reduction without affecting the average memory access time.

Power consumption, particularly runtime leakage, in long on-chip buses has grown to be an unacceptable portion of the total power budget due to heavy buffer insertion used to combat RC delays. In this paper, we propose a new bus encoding algorithm and circuit scheme for on-chip buses that eliminates capacitive crosstalk while simultaneously reducing total power. We utilize a buffer design approach with a selective use of high-threshold voltage transistors and couple this buffer design with a novel bus encoding scheme. The proposed encoding scheme significantly reduces total power by 26% and runtime leakage power by 42% while also eliminating capacitive crosstalk. In addition, the proposed encoding is specifically optimized to reduce the complexity of the encoding logic, allowing for a significant reduction in overhead which has not been considered in previous bus encoding work.

Dynamic voltage scaling (DVS) is a popular approach for energy reduction of integrated circuits. Current processors that use DVS typically have an operating voltage range from full to half of the maximum V/sub dd/. However, there is no fundamental reason why designs cannot operate over a much larger voltage range: from full V/sub dd/ to subthreshold voltages. This possibility raises the question of whether a larger voltage range improves the energy efficiency of DVS. First, from a theoretical point of view, we show that, for subthreshold supply voltages, leakage energy becomes dominant, making "just-in-time computation" energy-inefficient at extremely low voltages. Hence, we introduce the existence of a so-called "energy-optimal voltage" which is the voltage at which the application is executed with the highest possible energy efficiency and below which voltage scaling reduces energy efficiency. We derive an analytical model for the energy-optimal voltage and study its trends with technology scaling and different application loads. Second, we compare several different low-power approaches including MTCMOS, standard DVS, and the proposed Insomniac (extended DVS into subthreshold operation). A study of real applications on commercial processors shows that Insomniac provides the best energy efficiency. From these results, we conclude that extending the voltage range below V/sub dd//2 will improve the energy efficiency for many processor designs.

There is an increased dominance of intra-die process variations, creating a need for an accurate and fast statistical timing analysis. Most of the recent proposed approaches assume a Single Input Switching model. Our experiments show that SIS underestimates the mean delay of a stage by upto 20% and overestimates the standard deviation upto 26%. We also show that Multiple Input Switching has a greater impact on statistical timing, than regular static timing analysis. Hence, we propose a modeling technique for gate delay variability, considering MIS. Our model can be efficiently incorpo-rated into most of the statistical timing analysis frameworks. On average over all test cases, our approach underestimates mean delay of a stage by 0.01% and overestimates the standard deviation by only 2%, hence increasing the robustness to process variations. Our modeling technique is independent of the deterministic MIS model, and we show that its sensitivity to variations in the MIS model is small.

In this paper we develop an approach to model interconnect delay under process variability for timing analysis and physical design optimization. The technique allows for closed-form computation of interconnect delay probability density functions (PDFs) given variations in relevant process parameters such as linewidth, metal thickness, and dielectric thickness. We express the resistance and capacitance of a line as a linear function of random variables and then use these to compute circuit moments. Finally, these variability-aware moments are used in known closed-form delay metrics to compute interconnect delay PDFs. We compare the approach to SPICE based Monte Carlo simulations and report an error in mean and standard deviation of delay of 1% and 4% on average, respectively.

Power consumption, particularly runtime leakage, in long on-chip buses has grown to an unacceptable portion of the total power budget due to heavy buffer insertion to combat RC delays. In this paper, we propose a new bus encoding algorithm and circuit scheme for on-chip buses that eliminates capacitive crosstalk while simultaneously reducing total power. We introduce a new buffer design approach with selective use of high threshold voltage transistors and couple this buffer design with a novel bus encoding scheme. The proposed encoding scheme significantly reduces total power by 26% and runtime leakage power by 42% while also eliminating capacitive crosstalk. In addition, the proposed encoding is specifically optimized to reduce the complexity of the encoding logic, allowing for a significant reduction in overhead which has not been considered in previous bus encoding work.

Architectural simulation has achieved a prominent role in the system design cycle by providing designers the ability to quickly examine a wide variety of design choices. However, the recent trend in system design toward architectures that react to circuit-level phenomena has outstripped the capabilities of traditional cycle-based architectural simulators. In this paper, we present an architectural simulator design that incorporates a circuit modeling capability, permitting architectural-level simulations that react to circuit characteristics (such as latency,energy,or current draw) on a cycle-by-cycle basis. While these additional capabilities slow simulation speed, we show that the careful application of circuit simulation optimizations and simulation sampling techniques permit high levels of detail with sufficient speed to examine entire workloads.

In this paper, we address the problem of signal pruning in static timing analysis (STA). Traditionally, signals are propagated through the circuit and are pruned, such that only the signal with the latest arrival time at each node is propagated forward. This signal pruning is a key to the linear run time of STA. However, it was previously observed that a signal with the latest arrival time may not be the most critical signal, as an earlier signal with a larger transition time can result in a longer delay in the down-stream logic. Hence, arrival time based pruning can result in an optimistic delay, incorrect critical paths, and discontinuities of the delay during circuit optimization. Although algorithms were proposed to remedy this issue, they rely on propagation of multiple signals and have an exponential worst-case complexity. In this paper, we propose a new timing analysis algorithm, which uses a two pass traversal of the circuit. In the initial backward traversal, we construct delay tables which record the required time at a node as a function of the transition time at that node. This is followed by a forward traversal where signals are pruned not based on arrival times but based on slack. The proposed algorithm corrects the accuracy problems of the arrival time based pruning while at the same time maintaining the linear run time of STA. We implemented our algorithm and demonstrated its accuracy and efficiency.

Power supply integrity analysis is critical in modern high perfor-mance designs. In this paper, we propose a stochastic approach to obtain statistical information about the collective IR and LdI/dt drop in a power supply network. The currents drawn from the power grid by the blocks in a design are modelled as stochastic processes and their statistical information is extracted, including correlation infor-mation between blocks in both space and time. We then propose a method to propagate the statistical parameters of the block currents through the linear model of the power grid to obtain the mean and standard deviation of the voltage drops at any node in the grid. We show that the run time is linear with the length of the current wave-forms allowing for extensive vectors, up to millions of cycles, to be analyzed. We implemented the approach on a number of grids, including a grid from an industrial microprocessor and demonstrate its accuracy and efficiency. The proposed statistical analysis can be use to determine which portions of the grid are most likely to fail as well as to provide information for other analyses, such as statistical timing analysis.

Leakage current has become a stringent constraint in modern processor designs in addition to traditional constraints on frequency. Since leakage current exhibits a strong inverse correlation with circuit delay, effective parametric yield prediction must consider the dependence of leakage current on frequency. In this paper, we present a new chip-level statistical method to estimate the total leakage current in the presence of within-die and die-to-die variability. We develop a closed-form expression for total chip leakage that models the dependence of the leakage current distribution on a number of process parameters. The model is based on the concept of scaling factors to capture the effects of within-die variability. Using this model, we then present an integrated approach to accurately estimate the yield loss when both frequency and power limits are imposed on a design. Our method demonstrates the importance of considering both these limiters in calculating the yield of a lot.

Increasing levels of process variability in sub-100nm CMOS design has become a critical concern for performance and power constraint designs. In this paper, we propose a new statistically aware Dual-Vt and sizing optimization that considers both the variability in performance and leakage of a design. While extensive work has been performed in the past on statistical analysis methods, circuit optimization is still largely performed using deterministic methods. We show in this paper that deterministic optimization quickly looses effectiveness for stringent performance and leakage constraints in designs with significant variability. We then propose a statistically aware dual-Vt and sizing algorithm where both delay constraints and sensitivity computations are performed in a statistical manner. We demonstrate that using this statistically aware optimization, leakage power can be reduced by 15-35% compared to traditional deterministic analysis. The improvements increase for strict delay constraints making statistical optimization especially important for high performance designs.

We develop an approach to minimize total power in a dual-Vdd and dual-Vth design. The algorithm runs in two distinct phases. The first phase relies on upsizing to create slack and maximize low Vdd assignments in a backward topological manner. The second phase proceeds in a forward topological fashion and both sizes and re-assigns gates to high Vdd to enable significant static power savings through high Vth assignment. The proposed algorithm is implemented and tested on a set of combinational benchmark circuits. A comparison with traditional CVS and dual-Vth/sizing algorithms demonstrate the advantage of the algorithm over a range of activity factors, including an average power reduction of 30% (50%) at high (nominal) primary input activities.

Dynamic voltage scaling (DVS) is a popular approach for energy reduction of integrated circuits. Current processors that use DVS typically have an operating voltage range from full to half of the maximum Vdd. However, it is possible to construct designs that operate over a much larger voltage range: from full Vdd to subthreshold voltages. This possibility raises the question of whether a larger voltage range improves the energy efficiency of DVS. First, from a theoretical point of view, we show that for subthreshold supply voltages leakage energy becomes dominant, making "just in time completion" energy inefficient. We derive an analytical model for the minimum energy optimal voltage and study its trends with technology scaling. Second, we use the proposed model to study the workload activity of an actual processor and analyze the energy efficiency as a function of the lower limit of voltage scaling. Based on this study, we show that extending the voltage range below 1/2 Vdd will improve the energy efficiency for most processor designs, while extending this range to subthreshold operation is beneficial only for very specific applications. Finally, we show that operation deep in the subthreshold voltage range is never energy-efficient.

Standby leakage current minimization is a pressing concern for mobile applications that rely on standby modes to extend battery life. Also, gate oxide leakage current (I/sub gate/) has become comparable to subthreshold leakage (I/sub sub/) in 90 nm technologies. In this paper, we propose a new method that uses a combined approach of sleep-state, threshold voltage (V/sub t/ and gate oxide thickness (T/sub ox/) assignments in a dual-V/sub t/ and dual-T/sub ox/ process to minimize both I/sub sub/ and I/sub gate/. Using this method, total leakage current can be dramatically reduced since in a known state in standby mode, only certain transistors are responsible for leakage current and need to be considered for high-V/sub t/ or thick-T/sub ox/ assignment. We formulate the optimization problem for simultaneous state, V/sub t/ and T/sub ox/ assignments under delay constraints and propose two practical heuristics. We implemented and tested the proposed methods on a set of synthesized benchmark circuits. Results show an average leakage current reduction of 5a-6X and 2-3X compared to previous approaches that only use state or state+V/sub t/ assignment, respectively, with small delay penalties.

We present a sensitivity-based algorithm for total power including dynamic and subthreshold leakage power minimization using simultaneous sizing, Vdd and Vth assignment. The proposed algorithm is implemented and tested on a set of combinational benchmark circuits. A comparison with traditional CVS based algorithms demonstrates the advantage of the algorithm including an average power reduction of 37% at primary input activities of 0.1. We also investigate the impact of various low Vdd values on total power savings.

The quadratic relationship between voltage and energy has made dynamic voltage scaling (DVS) one of the most powerful techniques to reduce system power demands. Recently, techniques such as Razor DVS, voltage overscaling, and Intelligent Energy Management have emerged as approaches to further reduce voltage by eliminating costly voltage margins inserted into traditional designs to ensure always-correct operation. The degree to which a global voltage controller can shave voltage margins is limited by imbalances in pipeline stage latency. Since all pipeline stages share the same voltage, the stage exercising the longest critical path will define the overall voltage of the system, even if other stages could potentially run at lower voltages. In this paper, we evaluate two local tuning mechanisms in the context of Razor DVS, a local voltage controller scheme that allows each pipeline stages it's own voltage level, and a lower cost dynamic retiming scheme that incorporates per-stage clock delay elements to allow longer-latency pipeline stages to "borrow" time from shorter-latency stages.

This paper presents a new library-compatible approach to gate-level timing characterization in the presence of resistive/inductive/capacitive (RLC) interconnect loads. We show that for a gate driving an RLC interconnect, the driver-output waveform exhibits inflection points and, hence, the traditional approach of approximating driver output with a saturated ramp is highly inaccurate. We describe a two-ramp model based on transmission-line theory that accurately predicts both the 50% delay and waveform shape (slew rate) at the driver output when inductive effects are significant. The approach does not rely on piecewise linear Thevenin voltage sources and is compatible with existing library characterization methods. Results are compared with SPICE and demonstrate typical errors under 10% for both delay and slew rate. We also propose a new criterion for evaluating the importance of on-chip inductance by comparing rise time at the driver output with the time of flight.

In this paper, we introduce a simple metric for the slew rate of an RC circuit based on the first two circuit moments. Metrics focusing on 50% delay of RC circuits have been proposed recently that greatly improve the accuracy of the traditional Elmore delay model. However, these new models have not been applied to the determination of transition time or slew rates (e.g., 10-90% of V/sub dd/). We study how well existing approaches to 50% delay modeling translate to slew-rate modeling. We first describe a new metric called slew with two moments (S2M) that is based on Elmore's observation that the transition time of a step response is proportional to the standard deviation of the corresponding impulse response. The S2M metric modifies Elmore's original formulation by deriving a new constant of proportionality. This new constant is shown to be more accurate for general RC circuits. Next, we show that metrics relying on a simple constant multiplied by standard deviation such as S2M and Elmore do not work well for near-end nodes. To address this issue, we propose a new slew metric called scaled S2M that provides high accuracy across all types of nodes, while maintaining the advantage of a simple closed-form expression. Scaled S2M is shown to be very accurate for both near and far-end nodes. The average error for scaled S2M is approximately 2% with 96% of all nodes showing less than 5% error from a large set of industrial 0.18-/spl mu/m microprocessor nets.

With shrinking cycle times, clock skew has become an increasingly difficult and important problem for high performance designs. Traditionally, clock skew has been analyzed using case-files which cannot model intradie-process variations and hence result in a very optimistic skew analysis. In this paper, we present a statistical skew analysis method to model intradie process variations. We first present a formal model of the statistical clock-skew problem and then propose an algorithm based on propagation of joint probability density functions in a bottom-up fashion in a clock tree. The analysis accounts for topological correlations between path delays and has linear runtime with the size of the clock tree. The proposed method was tested on several large clock-tree circuits, including a clock tree from a large industrial high-performance microprocessor. The results are compared with Monte Carlo simulation for accuracy comparison and demonstrate the need for statistical analysis of clock skew.

Gate sizing is a practical and a feasible crosstalk noise correction technique in the post route design stage, especially for block level sea-of-gates designs. The difficulty in gate sizing for noise reduction is that, by increasing a driver size, noise at the driver output is reduced, but noise injected by that driver on other nets is increased. This can create cyclical dependencies between nets in the circuit with noise violations. In this paper, we propose a fast and effective heuristic postroute gate-sizing algorithm that uses a graph representation of the noise dependencies between nodes. Our method utilizes gate sizing in both directions and works in linear time as a function of the number of gates. The effectiveness of the algorithm is shown on several industrial high-performance designs.

On-chip caches represent a sizable fraction of the total power consumption of microprocessors. As feature sizes shrink, the dominant component of this power consumption will be leakage. However, during a fixed period of time, the activity in a data cache is only centered on a small subset of the lines. This behavior can be exploited to cut the leakage power of large data caches by putting the cold cache lines into a state preserving, low-power drowsey mode. In this paper, we investigate policies and circuit techniques for implementing drowsy data caches. We show that with simple microarchitectural techniques, about 80%-90% of the data cache lines can be maintained in a drowsy state without affecting performance by more than 0.6%, even though moving lines into and out of a drowsy state incurs a slight performance loss. According to our projections, in a 70-nm complementary metal-oxide-semiconductor process, drowsy data caches will be able to reduce the total leakage energy consumed in the caches by 60%-75%. In addition, we extend the drowsy cache concept to reduce leakage power of instruction caches without significant impact on execution time. Our results show that data and instruction caches require different control strategies for efficient execution. In order to enable drowsy instruction caches, we propose a technique called cache subbank prediction, which is used to selectively wake up only the necessary parts of the instruction cache, while allowing most of the cache to stay in a low-leakage drowsy mode. This prediction technique reduces the negative performance impact by 78% compared with the no-prediction policy. Our technique works well even with small predictor sizes and enables a 75% reduction of leakage energy in a 32-kB instruction cache.

In this paper we address the growing issue of gate oxide leakage current (I/sub gate/) at the circuit level. Specifically, we develop a fast approach to analyze the total leakage power of a large circuit block, considering both I/sub gate/ and subthreshold leakage (I/sub sub/). The interaction between I/sub sub/ and I/sub gate/ complicates analysis in arbitrary CMOS topologies and we propose simple and accurate heuristics based on lookup tables to quickly estimate the state-dependent total leakage current for arbitrary circuit topologies. We apply this method to a number of benchmark circuits using a projected 100-nm technology and demonstrate accuracy within 0.09% of SPICE on average with a four order of magnitude speedup. We then make several observations on the impact of I/sub gate/ in designs that are standby power limited, including the role of device ordering within a stack and the differing state dependencies for NOR versus NAND topologies. Based on these observations, we propose the use of pin reordering as a means to reduce I/sub gate/. We find that for technologies with appreciable I/sub gate/, this technique is more effective at reducing total leakage current in standby mode than state assignment, which is often used for I/sub sub/ reduction.

We develop a method to estimate the variation of leakage current due to both intra-die and inter-die gate length process variability. We derive an analytical expression to estimate the probability density function (PDF) of the leakage current for stacked devices found in CMOS gates. These distributions of individual gate leakage currents are then combined to obtain the mean and variance of the leakage current for an entire circuit. We also present an approach to account for both the inter- and intra-die gate length variations to ensure that the circuit leakage PDF correctly models both types of variation. The proposed methods were implemented and tested on a number of benchmark circuits. Comparison to Monte Carlo simulation validates the accuracy of the proposed method and demonstrates the efficiency of the proposed analysis method. Comparison with traditional deterministic leakage current analysis demonstrates the need for statistical methods for leakage current analysis.

Process variation has become a significant concern for static timing analysis. In this paper, we present a new method for path-based statistical timing analysis. We first propose a method for modeling inter- and intra-die device length variations. Based on this model, we then present an efficient method for computing the total path delay probability distribution using a combination of device length enumeration for inter-die variation and an analytical approach for intra-die variation. We also propose a simple and effective model of spatial correlation of intra-die device length variation. The analysis is then extended to include spatial correlation. We test the proposed methods on paths from an industrial high-performance microprocessor and present comparisons with traditional path analysis which does not distinguish between inter- and intra-die variations. The characteristics of the device length distributions were obtained from measured data of 8 test chips with a total of 17688 device length measurements. Spatial correlation data was also obtained from these measurements. We demonstrate the accuracy of the proposed approach by comparing our results with Monte-Carlo simulation.

The growing impact of within-die process variation has created the need for statistical timing analysis, where gate delays are modeled as random variables. Statistical timing analysis has traditionally suffered from exponential run time complexity with circuit size, due to arrival time dependencies created by reconverging paths in the circuit. In this paper, we propose a new approach to statistical timing analysis that is based on statistical bounds of the circuit delay. Since these bounds have linear run time complexity with circuit size, they can be computed efficiently for large circuits. Since both a lower and upper bound on the true statistical delay is available, the quality of the bounds can be determined. If the computed bounds are not sufficiently close to each other, we propose a heuristic to iteratively improve the bounds using selective enumeration of the sample space with additional run time. We demonstrate that the proposed bounds have only a small error and that by carefully selecting an small set of nodes for enumeration, this error can be further improved.

ABSTRACT * This paper presents a new library compatible approach to gatelevel timing characterization in the presence of RLC interconnect loads. We describe a two-ramp model based on transmission line theory that accurately predicts both the 50 % delay and waveform shape (slew rate) at the driver output when inductive effects are significant. The approach does not rely on piecewise linear Thevenin voltage sources. It is compatible with existing library characterization methods and is computationally efficient. Results are compared with SPICE and demonstrate typical errors under 10 % for both delay and slew rate. Categories and Subject Descriptors

In this paper we introduce simple metrics for the slew rate of an RC circuit based on the first two circuit moments. We develop two new slew metrics, S2M (slew with 2 moments) and scaled S2M, that provide high accuracy with the advantage of simple closed form expressions. S2M is very accurate for middle and far end nodes but it does not perform as well for near end nodes. Scaled S2M is developed to improve upon S2M for near end nodes and is shown to be highly accurate for near as well as far end nodes. For a large set of nets from an industrial 0.18 μm microprocessor, S2M matches SPICE within 2% on average with 78% of the sinks having less than 1% error. For the same test cases, the average error for scaled S2M is less than 3% with 99.4% of the nodes showing less than 5% error.

Gate sizing is a practical and a feasible crosstalk noise repair technique in the post route design stage, especially for block level sea-of-gates designs. The difficulty in gate sizing for noise reduction is that by increasing a driver size, noise at the driver output is reduced, but noise injected by that driver on other nets is increased. This can create cyclical dependencies between nets in the circuit with noise violations. In this paper, we propose a fast and effective heuristic post-route gate sizing algorithm that uses a graph representation of the noise dependencies between nodes. Our method utilizes gate sizing in both directions and works in linear time as a function of the number of gates. The effectiveness of the algorithm is shown on several industrial high performance designs.

We propose a new method that uses a combined approach of sleep-state assignment and threshold voltage (Vt) assignment in a dual-Vt process. While each of these methods has previously been used individually, their combined effect has not been leveraged to date. By combining Vt and sleep-state assignment, leakage current can be dramatically reduced since the circuit is in a known state in standby-mode and only transistors that are off need to be considered for high-Vt assignment. A significant improvement in the leakage/ performance trade-off is therefore achievable using such a combined method. We formulate the optimization problem for simultaneous state and Vt assignment under delay constraints and propose both an exact method for its optimal solution as well as a number of practical heuristics with reasonable run time. We compare our results with Vt and sleep state assignment only and demonstrate an average decrease in leakage current of 3.5X compared to previous approaches.

In this paper we address the growing issue of gate oxide leakage current (I gate) at the circuit level. Specifically, we develop a fast approach to analyze the total leakage power of a large circuit block, considering both I gate and subthreshold leakage (I sub). The interaction between I sub and I gate complicates analysis in arbitrary CMOS topologies and we propose simple and accurate heuristics based on table look-ups to quickly estimate the state-dependent total leakage current within 1 % of SPICE. We then make several observations on the impact of I gate in designs that are standby power limited, including the role of device ordering within a stack and the differing state dependencies for NOR vs. NAND topologies. Based on these observations, we propose the use of pin reordering as a means to reduce I gate due to the dependence of gate leakage on stack node voltages.

Today's rising power densities have been widely cited as the foremost challenge to continued CMOS scaling. In fact, the current power crisis is reminiscent of the final days previous technologies, such as the once popular bipolar and NMOS technologies and even vacuum tubes. How CMOS technology will respond to the current power challenge to extend CMOS scaling to sub-90nm technology is an important question for designer and CAD tool developers alike. With aggressive scaling a number of new challenges have arisen, such as leakage control, heat removal and power supply distribution, that need to be addressed using new design techniques in conjunction with new CAD solutions.This panel brings together experts in circuit design and CAD tool development to discuss the current status of low-power design and provide opinions on what new EDA capabilities are most important in the power-constrained design era. For instance, how will power be distributed in a robust fashion in sub-90nm ICs, and what are the critical EDA analysis and optimization capabilities? What are the best techniques for leakage reduction, not only in standby modes, but also in the active mode? And how far will voltage scaling take us in attacking the dynamic power consumption issue? What will a power-centric design flow look like and how will it change the way we design ICs? The objective of the panel is to explore these issues and formulate a list of critical issues that need to be addressed by the EDA community to enable successful scaling of CMOS into the sub-90nm era.

In this paper, we present an efficient method for computing switching windows in the presence of delay noise. In static timing analysis, delay noise has traditionally been modeled using a simple switch-factor based noise model and the computation of switching windows is performed using an iterative algorithm, resulting in an overall run time of O(n2), where n is the number of gates in the circuit. It has also been shown that the iterations converge to different solutions, depending on the initial assumptions, making it unclear which solution is correct. In this paper, we show that the iterative nature of the problem is due to the switching-factor based noise model and the order in which events are evaluated. We utilize a delay noise model based on superposition and propose a new algorithm with a run time that is linear with the circuit size. Since the algorithm is non iterative and does not operate with initial assumptions, it also eliminates the multiple solution problem. We tested the algorithm on a number of designs and show that it achieves significant speedup over the iterative approach.

On-chip L1 and L2 caches represent a sizeable fraction of the total power consumption of microprocessors. In deep sub-micron technology, the subthreshold leakage power is becoming the dominant fraction of the total power consumption of those caches. In this paper, we present optimization techniques to reduce the leakage power of on-chip caches assuming that there are multiple threshold voltages, VTH's, available. First, we show a cache leakage optimization technique that examines the trade-off between access time and leakage power by assigning distinct VTH's to each of the four main cache components address bus drivers, data bus drivers, decoders, and SRAM cell arrays with sense-amps. Second, we show optimization techniques to reduce the leakage power of L1 and L2 on-chip caches without affecting the average memory access time. The key results are: 1) 2 VTH's are enough to minimize leakage in a single cache; 2) if L1 size is fixed, increasing the L2 size can result in much lower leakage without reducing average memory access time; 3) if L2 size is fixed, reducing L1 size can result in lower leakage without loss of the average memory access time; and 4) smaller L1 and larger L2 caches than are typical in today's processors result in significant leakage and dynamic power reduction without affecting the average memory access time.

We propose the concept of an optimal inductance value that can substantially reduce delay of global RLC signals while maintaining good signal integrity (low ringing/overshoot). We exploit the fact that inductance results in faster transition times to improve delay of buffers in global signal lines. We observe that voltage overshoot, slew rate, and total line delay all show strong inflection points at the same value of inductance. At this optimal value of inductance significant improvements in signal transition time, and hence in overall signal delay, are obtained with negligible ringing. We propose adjusting the power grid to achieve this optimal inductance. Results show that the delay of a I cm line with 9 inserted repeaters can be reduced by 8-12% with acceptable ringing by operating at the optimal inductance point.

We develop a method to estimate the variation of leakage current due to both intra-die and inter-die gate length process variability. We derive an analytical expression to estimate the probability density function (PDF) of the leakage current for stacked devices found in CMOS gates. These distributions of individual gate leakage currents are then combined to obtain the mean and variance of the leakage current for an entire circuit. We also present an approach to account for both the inter- and intra-die gate length variations to ensure that the circuit leakage PDF correctly models both types of variation. The proposed methods were implemented and tested on a number of benchmark circuits. Comparison to Monte-Carlo simulation validates the accuracy of the proposed method and demonstrates the efficiency of the proposed analysis method. Comparison with traditional deterministic leakage current analysis demonstrates the need for statistical methods for leakage current analysis.

With increasing clock frequencies and silicon integration, power aware computing has become a critical concern in the design of embedded processors and systems-on-chip. One of the more effective and widely used methods for power-aware computing is dynamic voltage scaling (DVS). In order to obtain the maximum power savings from DVS, it is essential to scale the supply voltage as low as possible while ensuring correct operation of the processor. The critical voltage is chosen such that under a worst-case scenario of process and environmental variations, the processor always operates correctly. However, this approach leads to a very conservative supply voltage since such a worst-case combination of different variabilities is very rare. In this paper, we propose a new approach to DVS, called Razor, based on dynamic detection and correction of circuit timing errors. The key idea of Razor is to tune the supply voltage by monitoring the error rate during circuit operation, thereby eliminating the need for voltage margins and exploiting the data dependence of circuit delay. A Razor flip-flop is introduced that double-samples pipeline stage values, once with a fast clock and again with a time-borrowing delayed clock. A metastability-tolerant comparator then validates latch values sampled with the fast clock. In the event of timing error, a modified pipeline mispeculation recovery mechanism restores correct program state. A prototype Razor pipeline was designed in a 0.18 /spl mu/m technology and was analyzed. Razor energy overhead during normal operation is limited to 3.1%. Analyses of a full-custom multiplier and a SPICE-level Kogge-Stone adder model reveal that substantial energy savings are possible for these devices (up to 64.2%) with little impact on performance due to error recovery (less than 3%).

The growing impact of within-die process variation has created the need for statistical timing analysis, where gate delays are modeled as random variables. Statistical timing analysis has traditionally suffered from exponential run time complexity with circuit size, due to arrival time dependencies created by reconverging paths in the circuit. In this paper, we propose a new approach to statistical timing analysis which uses statistical bounds and selective enumeration to refine these bounds. First, we provide a formal definition of the statistical delay of a circuit and derive a statistical timing analysis method from this definition. Since this method for finding the exact statistical delay has exponential run time complexity with circuit size, we also propose a new method for computing statistical bounds which has linear run time complexity. We prove the correctness of the proposed bounds. Since we provide both a lower and upper bound on the true statistical delay, we can determine the quality of the bounds. If the computed bounds are not sufficiently close to each other, we propose a heuristic to iteratively improve the bounds using selective enumeration of the sample space with additional run time. The proposed methods were implemented and tested on benchmark circuits. The results demonstrate that the proposed bounds have only a small error, which can be further reduced using selective enumeration with modest additional run time.

One of the critical challenges in today's high-performance IC design is to take noise into account as early as possible in the design cycle. Current noise analysis tools are effective at analyzing and identifying noise in the postroute design stage when detailed parasitic information is available. However, noise problems identified at this stage of the design cycle are very difficult to correct due to the limited flexibility in design and may cause additional iterations of routing and placement which adds costly delays in its time to market. In this paper, we introduce a probabilistic preroute noise analysis approach to identify postroute noise failures before the actual detailed route is completed. We introduce new methods to estimate the RC characteristics of victim and aggressor lines, their coupling capacitances, and the aggressor transition times before routing is performed. The approach is based on congestion information obtained from a global router. Since the exact location and relative position of wires in the design are not yet available, we propose a novel probabilistic method for capacitance extraction. We present results on two high-performance microprocessors in 0.18 /spl mu/m technology that demonstrate the effectiveness of the proposed approach.

With the increase in current densities, electromigration has become a critical concern in high-performance designs. Typically, electromigration has involved the process of time-domain simulation of drivers and interconnect to obtain average, root mean square (r.m.s.), and peak current values for each wire segment. However, this approach cannot be applied to large problem sizes where hundreds of thousands of nets must be analyzed, each consisting of many thousands of RC elements. The authors propose a static electromigration analysis approach. They show that the charge transfer through wire segments of a net can be calculated directly by solving a system of linear equations, derived from the nodal formulation of the circuit, thereby eliminating the need for time domain simulation. The authors account for the different possible switching scenarios that give rise to unidirectional or bidirectional current by separating the charge transfer from the rising and falling transitions and also propose approaches for modeling multiple simultaneous switching drivers. They implemented the proposed static analysis approach in an industrial electromigration analysis tool that was used on a number of industrial circuits, including a large microprocessor.

In this paper, we propose an accurate and fast method to estimate the crosstalk noise in the presence of multiple aggressor nets for use in physical design automation tools. Since noise estimation is often part of the inner loop of optimization algorithms, very efficient closed-form solutions are needed. Previous approaches model aggressor nets one at a time, assuming that the coupling capacitance to all quiet aggressor nets are grounded. They also model the load from interconnect branches as a lumped capacitor, the value of which is the sum of interconnect and load capacitances of the branch. Finally, previous works typically use simple lumped 2-4-node circuit templates and employ a so-called dominant pole approximation to solve the template circuit. While these approximations allow for very fast analysis, they may result in significant underestimation of the noise. In this paper, we propose a new and more comprehensive fast noise estimation method. We propose a novel reduction technique for modeling quiet aggressor nets based on the concept of coupling point admittance. We also propose a reduction method to replace tree branches with effective capacitors which models the effect of resistive shielding. Furthermore, we model the simplified single aggressor net crosstalk noise problem using a 6-node template circuit and propose a new double pole approach to solve the template circuit. We have tested the proposed method on noise-prone interconnects from an industrial high-performance processor. Our results show a worst case error of 7.8% and an average error of 2.7%, while allowing for very fast analysis.

In this paper, a precorrected-fast-Fourier-transform (FFT) approach for fast and highly accurate simulation of circuits with on-chip inductance is proposed. This work is motivated by the fact that circuit analysis and optimization methods based on the partial element equivalent circuit model require the solution of a subproblem in which a dense inductance matrix must be multiplied by a given vector, an operation with a high computational cost. The grid representation enables the use of the discrete FFT for fast magnetic vector potential calculation. The precorrected-FFT method has been applied to accurately simulate large industrial circuits with up to 121 000 inductors and over 7 billion mutual inductive couplings in about 20 min. Techniques for trading off CPU time with accuracy using different approximation orders and grid constructions are also illustrated. Comparisons with a block-diagonal sparsification method are used to illustrate the accuracy and effectiveness of this method. In terms of accuracy, memory, and speed, it is shown that the precorrected-FFT method is an excellent approach for simulating on-chip inductance in a large circuit.

Noise simulators and noise avoidance tools are playing an increasingly critical role in the design of deep submicron circuits. However, noise estimates produced by these simulators are often very pessimistic. For large, high-performance industrial ICs, which can contain hundreds of thousands of nets, the worst case estimates of the noise results in thousands of reported violations, without any information about the likelihood of the possible noise violation. In this paper, we present a probabilistic approach to prioritize the violating nets based on the likelihood of occurrence of the reported noise. We derive an upper bound on the probability that the total noise injected on a given victim net by a specific set of aggressors exceeds a threshold. This is equivalent to a lower bound on the expected number of clock cycles required to realize the noise violation for the first time, i.e., mean time-to-failure. If the probability of a failure in a victim is sufficiently small, it is possible that even during the operation of the part for a number of years, the probability of failure on the net is negligible and the net can be assigned a lower priority for the application of noise avoidance strategies. We demonstrate the utility of this approach through experiments carried out on an large industrial processor design using a state-of-the-art industrial noise analysis tool.

In this paper, we present a new approach to model the impact of cross-coupling noise on interconnect delay. We introduce a new linear driver model that accurately models the noise pulse induced on a switching signal net due to cross-coupling capacitance. The proposed model effectively captures the nonlinear behavior of the victim-driver gate during its transition and has an average error below 8% whereas the traditional approach using a Thevenin model incurs an average error of 48%. The proposed linear driver model enables the use of linear superposition which allows the analysis of large interconnects and an efficient determination of the worst-case transition times of the aggressor nets. We proposed a new approach to determine the worst-case alignment of the aggressor net transitions with respect to the victim net transition, emphasizing the need to maximize not merely the delay of the interconnect alone but the combined delay of the interconnect and receiver gate. We show that in the presence of multiple aggressor nets, the worst case delay may occur when their noise peaks are not aligned, although the error incurred from aligning all peaks is small in practice. We then show that the worst-case alignment time of the combined noise pulse from all aggressors with respect to the victim transition is a nonlinear function of the receiver gate output loading, the victim transition time, and the noise pulsewidth and height. To efficiently compute the worst-case alignment time, we propose a new representation of the alignment such that it closely fits a linear function of the input variables. The worst-case alignment time is then computed for a gate using a precharacterization approach, requiring only eight sample points while maintaining a small error. The proposed methods were implemented in an industrial noise analysis tool called ClariNet. Results on industrial designs, including a large PPCmicroprocessor design, are presented to demonstrate the effectiveness of our approach.

The corruption of signals due to capacitive and inductive coupling of interconnects has become a significant problem in the design of deep submicron circuits (DSM). Noise simulators, based on worstcase assumptions, are overly pessimistic. As a result, when they are used on industrial ICs with hundreds of thousands of nets, thousands of nets are reported as having potential noise violations. There is a need to prioritize the problem nets based on the likelihood of the noise and possibly even eliminate them from further consideration if the likelihood is negligable. In this paper, a probabilistic approach is described which allows for a quantitative means to prioritize nets based on the likelihood of the reported noise violation. We derive upper bounds on the probability that the total noise injected on a given victim net by a specific set of aggressors exceeds a threshold. This bound is then used to determine a lower bound on the expected number of clock cycles (ENC) before the first violation occurs on a given net. Nets can be prioritized based on the ENC. We demonstrate the utility of this approach through experiments carried out on a large industrial processor design using a state-ofthe -art industrial noise analysis tool. A significant and interesting # Work done while on sabbatical with Advanced Design Tools group at Motorola, Austin TX. Also suported by NSF Center for Low Power Electronics (CLPE) (NSF Grant No. EEC-9523338, State of Arizona and Industrial consortium).

Noise estimation and avoidance are becoming critical, 'must have' capabilities in today's high performance IC design. An accurate yet efficient crosstalk noise model which contains as many driver/interconnect parameters as possible, is necessary for any sensitivity based noise avoidance approach. In this paper, we present a complete analytical crosstalk noise model which incorporates all physical properties including victim and aggressor drivers, distributed RC characteristics of interconnects and coupling locations in both victim and aggressor lines. We present closed-form analytical expressions for peak noise and noise width as well as sensitivities to all model parameters. We then use them model parameter sensitivities to analyze and evaluate various noise avoidance techniques such as driver sizing, wire sizing, wire spacing and layer assignment. Both our model and noise avoidance evaluations are verified using realistic circuits in 0.13/spl mu/ technology. We also present effectiveness of discussed noise avoidance techniques on a high performance microprocessor core.

A new shielding scheme, active shielding, is proposed for reducing delays on interconnects. As opposed to conventional (passive) shielding, the active shielding approach helps to speed up signal propagation on a wire by ensuring in-phase switching of adjacent nets. Results show that the active shielding scheme improves performance by up to 16% compared to passive shields and up to 29% compared to unshielded wires. When signal slopes at the end of the line are compared, savings of up to 38% and 27% can be achieved when compared to passive shields and unshielded wires, respectively.

Delay due to capacitive coupling of interconnects has become an important reliability issue in the design of nanometer circuits. In this paper we present a probabilistic approach towards analyzing the impact of capacitive coupling noise on signal delay. The variation in the delay is due to the variation in the relative arrival times of the aggressors and the victim. We derive expressions for the moments of the victim voltage in the presence of noise. From these we compute estimates of the earliest and latest possible arrival times of the victim. We compare the analytical results with Monte Carlo simulations using SPICE. Even though the analytical calculations are 200 times faster than the Monte Carlo simulations, the differences in the estimates of the mean and standard deviation of the arrival time is no more than 2.8%. In addition, the width of the timing intervals using the proposed approach is reduced by as much as 48% with a confidence level of 0.984. That is 98.4% of the Monte Carlo simulations result in an arrival time that falls within the derived interval which is 48% shorter.

This paper describes a fast method to estimate crosstalk noise in the presence of multiple aggressor nets for use in physical design automation tools. Since noise estimation is often part of the inner-loop of optimization algorithms, very efficient closed-form solutions are needed. Previous approaches have typically used simple lumped 3-4 node circuit templates. One aggressor net is modeled at a time assuming that the coupling capacitances to all quiet aggressor nets are grounded. They also model the load from interconnect branches as a lumped capacitor and use a dominant pole approximation to solve the template circuit. While these approximations allow for very fast analysis, they result in significant underestimation of the noise. In this paper, we propose a new and more comprehensive fast noise estimation model. We use a 6 node template circuit and propose a novel reduction technique for modeling quiet aggressor nets based on the concept of coupling point admittance. We also propose a reduction method to replace tree branches with effective capacitors which models the effect of resistive shielding. Finally, we propose a new double pole approach to solve the template circuit. We tested the proposed method on noise-prone interconnects from an industrial high performance processor. Our results show a worst-case error of 7.8% and an average error of 2.7%, while allowing for very fast analysis.

The simulation of on-chip inductance using PEEC-based circuit analysis methods often requires the solution of a subproblem where an extracted inductance matrix must be multiplied by a current vector, an operation with a high computational cost. This paper presents a highly accurate technique, based on a precorrected-FFT approach, that speeds up this calculation. Instead of computing the inductance matrix explicitly, the method exploits the properties of the inductance calculation procedure while implicitly considering the effects of all of the inductors in the layout. An optimized implementation of the method has been applied to accurately simulate large industrial circuits with up to 121,000 inductors and nearly 7 billion mutual inductive couplings in about 20 minutes. Techniques for trading off the CPU time with the accuracy using different approximation orders and grid constructions are also illustrated. Comparisons with a block diagonal sparsification method in terms of accuracy, memory and speed demonstrate that our method is an excellent approach for simulating on-chip inductance in a large circuit.

Dynamic voltage scaling (DVS) reduces the power consumption of processors when peak performance is unnecessary. However, the achievable power savings by DVS alone is becoming limited as leakage power increases. In this paper, we show how the simultaneous use of adaptive body biasing (ABB) and DVS can be used to reduce power in high-performance processors. Analytical models of the leakage current, dynamic power, and frequency as functions of supply voltage and body bias are derived and verified with SPICE simulation. We then show how to determine the correct trade-off between supply voltage and body bias for a given clock frequency and duration of operation. The usefulness of our approach is evaluated on real workloads obtained using real-time monitoring of processor utilization for four applications. The results demonstrate that application of simultaneous DVS and ABB results in an average energy reduction of 48% over DVS alone.

Noise analysis has become a critical concern in advanced chip designs. Traditional methods suffer from two common issues. First, noise that is propagated through the driver of a net is combined with noise injected by capacitively coupled aggressor nets using linear summation. Since this ignores the non-linear behavior of the driver gate the noise that develops on a net can be significantly underestimated. We therefore propose a new linear model that accurately combines propagated and injected noise on a net and which maintains the efficiency of linear simulation. After the propagated and injected noise are correctly combined on a victim net, it is necessary to determine if the noise can result in a functional failure. This is the second issue that we discuss in this paper. Traditionally, noise failure criteria have been based on unity gain points of the DC or AC transfer curves. However, we will show that for digital designs, these approaches can result in a pessimistic analysis in some cases, while in other cases, they allow circuit operation that is extremely close to regions that are unstable and do not allow sufficient margin for error in the analysis. In this paper, we compare the effectiveness of the discussed noise failure criteria and also present a propagation based method, which is intended to overcome these drawbacks. The proposed methods were implemented in a noise analysis tool and we demonstrate results on industrial circuits.

On-chip caches represent a sizable fraction of the total power consumption of microprocessors. Although large caches can significantly improve performance, they have the potential to increase power consumption. As feature sizes shrink the dominant component of this power loss will be leakage. However, during a fixed period of time the activity in a cache is only centered on a small subset of the lines. This behavior can be exploited to cut the leakage power of large caches by putting the cold cache lines into a state preserving, low-power drowsy mode. Moving lines into and out of drowsy state incurs a slight performance loss. In this paper we investigate policies and circuit techniques for implementing drowsy caches. We show that with simple architectural techniques, about 80%-90% of the cache lines can be maintained in a drowsy state without affecting performance by more than 1%. According to our projections, in a 0.07 um CMOS process, drowsy caches will be able to reduce the total energy (static and dynamic) consumed in the caches by 50%-75%. We also argue that the use of drowsy caches can simplify the design and control of low-leakage caches, and avoid the need to completely turn off selected cache lines and lose their state.

We describe the impact of process variation on leakage power for a 0.18mm CMOS technology. We show that variability, manifested in Ldrawn, Tox, and Nsub, can drastically affect the leakage current. We first present Monte Carlo-based simulation results for leakage current in various CMOS gates when the process parameters are varied both individually and concurrently. We then derive an analytical model to estimate the mean and standard deviation of the leakage current as a function of the process parameter distributions. We demonstrate that the results of the analytical model match well with Monte-Carlo simulations and also show the statistical mean leakage current is significantly different from the leakage predicted using a nominal case file.

On-chip caches represent a sizeable fraction of the total power consumption of microprocessors. Although large caches can significantly improve performance, they have the potential to increase power consumption. As feature sizes shrink, the dominant component of this power loss will be leakage. In our previous work we have shown how the drowsy circuit - a simple, state-preserving, low-leakage circuit that relies on voltage scaling for leakage reduction - can be used to reduce the total energy consumption of data caches by more than 50%. In this paper, we extend the architectural control mechanism of the drowsy cache to reduce leakage power consumption of instruction caches without significant impact on execution time. Our results show that data and instruction caches require different control strategies for efficient execution. To enable drowsy instruction caches, we propose a technique called cache sub-bank prediction which is used to selectively wake up only the necessary parts of the instruction cache, while allowing most of the cache to stay in a low leakage drowsy mode. This prediction technique reduces the negative performance impact by 76% compared to the no-prediction policy. Our technique works well even with small predictor sizes and enables an 86% reduction of leakage energy in a 64 K byte instruction cache.

Static timing analysis has traditionally used the PERT method for identifying the critical path of a circuit. The authors show in this paper that due to the influence of the transition time of a signal on the subsequent path delay, the traditional timing analysis approach can report an optimistic circuit delay and may identify the wrong critical path. Also, the calculated circuit delay is a discontinuous function with respect to transistor and gate sizes, posing a severe problem for circuit optimization methods. The authors also examine an alternate approach where the propagated signal is constructed by combining the latest arrival time and the slowest transition time from all signals incident on a node. While this approach remedies the problem of discontinuity, it can significantly overestimate the circuit delay and can also identify the wrong critical path. In this paper, they therefore propose a new timing analysis algorithm and prove that it computes the correct and continuous timing graph delay and the proper critical path. The proposed algorithm selectively propagates multiple signals through each timing edge in cases where there exists ambiguity regarding which arriving signal represents the critical path. They show that the algorithm propagates the sufficient and necessary set of signals for computing the delay of a general timing graph. The authors also introduce a new property of digital gates, referred to as the transition shift property, and, using this property, show that the number of propagated signals can be significantly reduced for timing graphs of digital circuits. Finally, they discuss the computation of required times and node slacks for the traditional approaches and propose corresponding algorithms for the new approaches. They show that while the traditional approach can incur both a positive or negative error in the computed slack, the proposed algorithms compute a conservative slack for off-critical nodes and the correct and continuous slack for the critical path. The proposed algorithms were implemented in an industrial static timing analysis and optimization tool, and the authors present results for a number of industrial circuits. Their results show that the traditional timing analysis method underestimates the circuit delay by as much as 39%, while the discussed alternate approach can overestimate circuit delay by as much as 17%. The proposed method computes the correct delay, while incurring only a small run time overhead in all cases.

Careful design and verification of the power distribution network of a chip are of critical importance to ensure its reliable performance. With the increasing number of transistors on a chip, the size of the power network has grown so large as to make the verification task very challenging. The available computational power and memory resources impose limitations on the size of networks that can be analyzed using currently known techniques. Many of today's designs have power networks that are too large to be analyzed in the traditional way as flat networks. In this paper, we propose a hierarchical analysis technique to overcome the aforesaid capacity limitation. We present a new technique for analyzing a power grid using macromodels that are created for a set of partitions of the grid. Efficient numerical techniques for the computation and sparsification of the port admittance matrices of the macromodels are presented. A novel sparsification technique using a 0-1 integer linear programming formulation is proposed to achieve superior sparsification for a specified error. The run-time and memory efficiency of the proposed method are illustrated on industrial designs. It is shown that even for a 60 million-node power grid, our approach allows for an efficient analysis, whereas previous approaches have been unable to handle power grids of such size.

Cross-coupled noise analysis has become a critical concern in today's VLSI designs. Typically, noise analysis makes the assumption that all aggressing nets can simultaneously switch in the same direction. This creates a worst- case noise pulse on the victim net that often leads to false noise violations. In this article we present a new approach that uses logic implications to identify the maximum set of aggressor nets that can inject noise simultaneously under the logic constraints of the circuit. We propose an approach to efficiently generate logic implications from a transistor-level description and propagate them in the circuit using ROBDD representations. We propose a new method for lateral propagation of implications and also show how tristate gates and high-impedance signal states can be handled using tristate implications. We then show that the problem of finding the worst-case logically feasible noise can be represented as a maximum weighted independent set problem and show how to efficiently solve it. Initially, we restrict our discussion to zero-delay implications, which are valid for glitch-free circuits, and then extend our approach to timed implications. The proposed approaches were implemented in an industrial noise analysis tool and results are shown for a number of industrial test cases. We demonstrate that a significant reduction in the number of noise failures can be obtained from considering the logic implications as proposed in this article, underscoring the need for false-noise analysis.

With operating frequencies entering the multi-gigahertz range, inductance has become an important consideration in the design and analysis of on-chip interconnects. In this paper, we present an accurate and efficient inductance modeling and analysis methodology for high-performance interconnect. We determine the critical elements for a PEEC based model by analyzing the current flow in the power grid and signal interconnect. The proposed model includes distributed interconnect resistance, inductance and capacitance, device decoupling capacitances, quiescent switching currents in the grid, pad connections, and pad/package inductance. We propose an efficient methodology for extracting these elements, using statistical models for on-chip decoupling capacitance and switching currents. Simulation results show the importance of various elements for accurate inductance analysis. We also demonstrate the accuracy of the proposed model compared to the traditional loop-based inductance approach. Since the proposed model can consist of hundreds of thousands of RLC elements, and a fully dense mutual inductance matrix, we propose a number of acceleration techniques that enable efficient analysis of large interconnect structures.

Presents a new approach for the estimation and optimization of standby power dissipation in large MOS digital circuits. We introduce a new approach for accurate and efficient calculation of the average standby or leakage current in large digital circuits by introducing the concepts of "dominant leakage states" and the use of state probabilities. Combined with graph reduction techniques and simplified nonlinear simulation, the method achieves speedups of three to four orders of magnitude over exhaustive SPICE simulations while maintaining very good accuracy. The leakage current calculation is then utilized in a new leakage and performance optimization algorithm for circuits using dual V/sub t/ processes. The approach is the first to consider the assignment of both the V/sub t/ and the width of a transistor, simultaneously. The optimization approach uses incremental calculation of leakage and performance sensitivities and can take into account a partially defined circuit state constraint for the standby mode of the device.

With operating frequencies approaching the gigahertz range, inductance is becoming an increasingly important consideration in the design and analysis of on-chip interconnect. In this paper, we give a tutorial overview of the analysis and design issues related to on-chip inductance effects.We explain the complexity of the current flow in VLSI circuits. We discuss the applicability of the PEEC approach in a detailed circuit model of the signal and power grid interconnect, switching devices, power pads and the package. Fur-ther, we explain techniques that can be used to speed-up simulation of the large PEEC model. We then discuss a simplified model that uses the so-called loop inductance approach, and compare it with the detailed model.We present experimental results, obtained from simulations of industrial circuits, for both the PEEC and loop models. We also cover design techniques that can help tackle the on-chip inductance issues.

In this paper, we present a new approach to model the impact of cross-coupling noise on interconnect delay. We introduce a new linear driver model that accurately models the noise pulse induced on a switching signal net due to cross coupling capacitance. The proposed model effectively captures the non-linear behavior of the victim driver gate during the transition and has an average error below 8% whereas the traditional approach using a Thevenin model incurs an average error of 48%. We also discuss the worst case alignment of the aggressor net transitions with respect to the victim net transition, emphasizing the need to maximize not merely the delay of the interconnect alone but the combined delay of the interconnect and receiver gate. We show that the worst case alignment of an aggressor net transition is a function of the receiver gate output loading, victim transition edge rate, and the noise pulse width and height and hence propose a pre-characterization approach to efficiently predict the worst-case alignment. The proposed methods were implemented in an industrial noise analysis tool called ClariNet. Results on industrial designs are presented to demonstrate the effectiveness of our approach.

Cross-coupled noise analysis has become a critical concern in VLSI design. Typically, noise analysis makes the assumption that all aggressing nets can simultaneously switch in the same direction. This creates a worst-case noise pulse on the victim net that often leads to false noise violations. In this paper, we present a new approach that uses logic implications to identify the maximum set of aggressor nets that can inject noise simultaneously under the logic constraints of the circuit. We propose an approach to efficiently generate logic implications from a transistor-level description and propagate them in the circuit using ROBDD representations of the DC-connected components and a newly proposed lateral propagation method. We then show that the problem of finding the worst case logically feasible noise can be represented as a maximum weighted independent set problem and show how to efficiently solve it. Initially, we restrict our discussion to zero-delay implications, which are valid for glitch-free circuits and then extend our approach to timed implications. The proposed approaches were implemented in an industrial noise analysis tool and results are shown for a number of industrial test cases. We demonstrate that a significant reduction in the number of noise failures can be obtained from considering the logic implications as proposed in this paper, underscoring the need for false-noise analysis.

Article Share on On the interaction of power distribution network with substrate Authors: Rajendran Panda Motorola, Inc., Austin, TX Motorola, Inc., Austin, TXView Profile , Savithri Sundareswaran Motorola, Inc., Austin, TX Motorola, Inc., Austin, TXView Profile , David Blaauw Motorola, Inc., Austin, TX Motorola, Inc., Austin, TXView Profile Authors Info & Claims ISLPED '01: Proceedings of the 2001 international symposium on Low power electronics and designAugust 2001Pages 388–393https://doi.org/10.1145/383082.383188Published:06 August 2001Publication History 12citation288DownloadsMetricsTotal Citations12Total Downloads288Last 12 Months1Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

We present a new method for removing user-specified false subgraphs from timing analysis and circuit optimization. Given a timing graph and a list of specified false paths, false subpaths, or false subgraphs, we generate a new timing graph in which all specified false paths are removed using a process of node splitting and edge removal. We present the necessary and sufficient condition for splitting a node, and show that the number of nodes that must be added to the timing graph is linear with the size of the false path specification. We also present an algorithm for finding the minimum set of nodes that must be split. Since this algorithm requires exponential run time for false subpaths and false subgraphs, we present a heuristic splitting approach which has linear worst-case run time, and where the number of added nodes is linear with the size of the false path specification. The heuristic approach was implemented and results are given for large industrial circuits.

We present a novel approach to compress current signatures for IR-drop analysis of large power grids. Our approach divides the original current signature into error bounded compression sets. Each compression set has a representative timepoint. The compressed current signature consists of the representative timepoints. The compression technique exploits the pattern of change of individual currents, time locality and periodicity to achieve very high quality results. We provide error guarantees for the compression. Our results are superior in compression and accuracy in comparison to other compression methods such as the single cycle compression.

With operating frequencies approaching the gigahertz range, inductance is becoming an increasingly important consideration in the design and analysis of on-chip interconnect. We present an accurate technique for modeling and analyzing the effects of parasitic inductance on power grid noise, signal delay and crosstalk. We propose a detailed circuit model composed of interconnect resistance, inductance and distributed capacitance, device decoupling capacitances, quiescent activity in the grid, pad locations, and pad/package inductance which accurately determines current distribution and, hence, on-chip inductive effects, and proves superior to the traditional simplified loop inductance approach. The model uses partial inductances, computed using an analytical formula for a pair of parallel rectangular conductors spaced in any relative position. We present experimental results, obtained from simulations of industrial circuits, that show the importance of various model components while analyzing on-chip inductance. We also propose a simple sparsification technique to handle large, dense partial inductance matrices.

Coupled noise analysis has become a critical issue for deep-submicron, high performance design. In this paper, we present, ClariNet, an industrial noise analysis tool, which was developed to efficiently analyze large, high performance processor designs. We present the overall approach and tool flow of ClariNet and discuss three critical large-processor design issues which have received limited discussion in the past. First, we present how the driver gates of a coupled interconnect network are represented with accurate linear models. Second, we show how to speed the analysis of large designs by using noise filters based on reduced interconnect representations and then pruning the nets coupled to a signal net. Third, we show how to incorporate logic and timing correlations into noise analysis to reduce its pessimism. We present the results from several industrial circuits, including a large high performance microprocessor design and a DSP design.

Careful design and verification of the power distribution network of a chip are of critical importance to ensure its reliable performance. With the increasing number of transistors on a chip, the size of the power network has grown so large as to make the verification task very challenging. The available computational power and memory resources impose limitations on the size of networks that can be analyzed using currently known techniques. Many of today's designs have power networks that are too large to be analyzed in the traditional way as flat networks. In this paper, we propose a hierarchical analysis technique to overcome the aforesaid capacity limitation. We present a new technique for analyzing a power grid using macromodels that are created for a set of partitions of the grid. Efficient numerical techniques for the computation and sparsification of the port admittance matrices of the macromodels are presented. A novel sparsification technique using a 0-1 integer linear programming formulation is proposed to achieve superior sparsification for a specified error. The run-time and memory efficiency of the proposed method are illustrated through the analysis of case studies of several multi-million node power grids, extracted from real microprocessor and DSP designs.

With operating frequencies approaching the gigahertz range, inductance is becoming an increasingly important consideration in the design and analysis of on-chip interconnect. We present an accurate technique for modeling and analyzing the effects of parasitic inductance on power grid noise, signal delay and crosstalk. We propose a detailed circuit model composed of interconnect resistance, inductance and distributed capacitance, device decoupling capacitances, quiescent activity in the grid, pad locations, and pad/package inductance which accurately determines current distribution and, hence, on-chip inductive effects, and proves superior to the traditional simplified loop inductance approach. The model uses partial inductances, computed using an analytical formula for a pair of parallel rectangular conductors spaced in any relative position. We present experimental results, obtained from simulations of industrial circuits, that show the importance of various model components while analyzing on-chip inductance.

Static timing analysis has traditionally used the PERT method for identifying the critical path of a digital circuit. Due to the influence of the slope of a signal at a particular node on the subsequent path delay, an earlier signal with a signal slope greater than the slope of the later signal may result in a greater delay. Therefore, the traditional method for timing analysis may identify the incorrect critical path and report an optimistic delay for the circuit. We show that the circuit delay calculated using the traditional method is a discontinuous function with respect to transistor and gate sizes, posing a severe problem for circuit optimization methods. We propose a new timing analysis algorithm which resolves both these issues. The proposed algorithm selectively propagates multiple signals through each timing edge in cases where there exists ambiguity regarding which arriving signal represents the critical path. The algorithm for propagating the corresponding required times is also presented. We prove that the proposed algorithm identifies a circuit's true critical path, where the traditional timing analysis method may not. We also show that under this method circuit delay and node slack are continuous functions with respect to a circuit's transistor and gate sizes. In addition, we present a heuristic method which reduces the number of signals to be propagated at the expense of a slight loss in accuracy. Finally, we show how the proposed algorithm was efficiently implemented in an industrial static timing analysis and optimization tool, and present results for a number of industrial circuits. Our results show that the traditional timing analysis method underestimates the circuit delay by as much as 38%, while that the proposed method efficiently finds the correct circuit delay with only a slight increase in run time.

We present new modeling and simulation techniques to improve the accuracy and efficiency of transient analysis of large power dis?tribution grids. These include an accurate model for the inherent decoupling capacitance of non-switching devices, as well as a sta?tistical switching current model for the switching devices. More?over, three new simulation techniques are presented for problem size-reduction and speed-up. Results of application of these tech?niques on three PowerPCtmmicroprocessors are also presented.

We present a new approach for estimation and optimization of the average stand-by power dissipation in large MOS digital circuits.To overcome the complexity of state dependence in average leakage estimation, we introduce the concept of "dominant leakage states" and use state probabilities.Our method achieves speed-ups of 3 to 4 orders of magnitude over exhaustive SPICE simulations while maintaining accuracies within 9% of SPICE.This accurate estimation is used in a new sensitivity-based leakage and performance optimization approach for circuits using dual V t processes.In tests on a variety of industrial circuits, this approach was able to obtain 81-100% of the performance achievable with all low V t transistors, but with 1/3 to 1/6 the stand-by current.

We present a methodology for the design and analysis of power grids in the PowerPC™ microprocessors. The methodology covers the need for power grid analysis across all stages of the design process. A case study showing the application of this methodology to the PowerPC™ 750 microprocessor is presented.

A novel technique to explore the performance vs design effort trade-off is proposed. Starting from an optimally synthesized design, performance-critical cells are incrementally and optimally selected and custom-sized to generate this trade-off. Efficient algorithms for the optimal selection, and for improving the reuse of custom-sized cells in the design are given. Significant performance gains are shown in several real circuits through the addition of very few customized cells.

Power management is an increasing concern for processor design. In this paper, we presented an overview of traditional power simulation tools and discussed two emerging power management design technologies: power distribution integrity analysis and standby current measurement and optimization. We present methods for accurate peak current simulation, which is needed for power grid integrity analysis, and discuss the generation and compression of the simulation vectors. Also, static approaches for calculating an upper-bound on the maximum peak current are presented. Standby leakage current is state dependent and we present methods for calculating both the average and maximum leakage current. Finally, optimization methods for minimizing the leakage current by either assigning a standby state to the circuit or by using a dual-Vt process are discussed.

Traditional synthesis techniques optimize CMOS circuits in two phases: i) logic minimization and ii) library mapping phase. Typically, the structures and the sizes of the gates in the library are chosen to yield good synthesis results over many blocks or even for an entire chip. Consequently this approach precludes an optimal design of individual blocks which may need custom structures. The authors present a new transistor level technique that optimizes CMOS circuits both structurally and size-wise. The technique is independent of a library and hence can explore a design space much larger than that possible due to gate level optimization. Results demonstrate a significant improvement in circuit performance of the resynthesized circuits.

This paper describes a tool called Focus that is currently being used for the timing verification and siting of domino CMOS circuits in a Power PC/sup TM/ microprocessor. Domino CMOS circuits introduce more complex timing and sizing requirements compared to conventional static circuits. This paper shows how these requirements are addressed in Focus. Some case studies involving the application of Focus on production circuits are also described.

Switch-level simulation has become a common means of validating the behavior of MOS circuits. In this paper, we present a new cache-based simulation method that significantly reduces the cost of subnetwork evaluation during switch-level simulation. The method speeds up simulation by as much as a factor of two. While caching may require additional memory, the structural hierarchy can be exploited to quickly identify subnetworks computing identical functions, merge their cache tables, and significantly reduce the memory requirements.>

A novel switch-level simulator, called SNEL, is presented. The SNEL simulator preprocesses the circuit description to abstract its functionality prior to simulation. Functional abstraction is concisely defined in terms of the functional domain and the functional application of circuit constructs. SNEL uses four algorithms that operate on levels ranging from single circuit elements to multiple DC-connected components. Since the functional abstraction preserves the complete functionality of the circuit, the accuracy of the simulation is maintained. However, SNEL models the circuit at a higher and more abstract level, which increases its simulation speed. The presented algorithms were implemented and tested on commercial designs. Without functional abstraction, the simulation speed of SNEL is competitive with current simulators. When functional abstraction was used, the simulation speed increased by more than an order of magnitude.>

In switch-level simulation, nodes carry a charge on their parasitic capacitance from one evaluation to the next, which gives them a memory quality. A node is classified as temporary if its memory aspect is lost and cannot affect the circuit operation, whereas a node is classified as a memory node if the memory of the node is maintained and can affect the circuit operation. Accurate classification of nodes into temporary and memory nodes increases the performance of compiled simulators and high-level model generators. An approach for reliable automatic classification of nodes in a switch-level description is introduced. Both an exhaustive, exponential-time algorithm and a polynomial-time heuristic are presented. The heuristic was implemented and tested for several large circuits, including a commercial microprocessor. For this processor, the proposed heuristics identified an average of 92% of all nodes as temporary nodes. The heuristic was applied in a high-level model generator and significantly increased its performance.>

Cost-effective and accurate fault simulation of very large digital designs on engineering workstations is proposed. The hierarchical approach reduces memory requirements drastically by storing the structure of common repeated subcircuits only once. The approach allows flexible multilevel simulation. The simulation algorithms are at the switch-level so that general MOS digital designs with bidirectional signal flow can be handled, and both stuck-at and transistor faults are treated accurately. The fault simulation algorithms have been implemented as a prototype that was used to determine the fault grade of a model of the Motorola 68000 microprocessor on SUN Microsystems workstations.>

This paper discusses the automatic generation of high-level software models from switch-level circuit descriptions. The proposed algorithms operate directly on the hierarchical description, and incorporate information about the design such as the structure, regularity, functionality, and control signals in the generation process. New algorithms are proposed and have been implemented for combinational modules and bus structures. A significant speedup has been obtained for these modules of a commercially available chip.

The design and implementation of a hierarchical switch-level simulator for complex digital circuits is discussed. The hierarchy is exploited to reduce the memory requirements of the simulation, thus allowing the simulation of circuits that are too large to simulate at the flat level. The algorithm used in the simulator operates directly on the hierarchical circuit description. Speedup is obtained through the use of high-level models. The simulator has been implemented on a SUN workstation and used to simulate a switch-level description of the Motorola 68000 microprocessor.>