Large Language Models (LLMs) with Mixture-of-Expert (MoE) architectures achieve superior model performance with reduced computation costs, but at the cost of high memory capacity and bandwidth requirements. Near-Memory Processing (NMP) accelerators that stack memory directly on the compute through hybrid bonding have demonstrated high bandwidth with high energy efficiency, becoming a promising architecture for MoE models. However, as NMP accelerators comprise distributed memory and computation, how to map the MoE computation directly determines the LLM inference efficiency. Existing parallel mapping strategies, including Tensor Parallelism (TP) and Expert Parallelism (EP), suffer from either high communication costs or unbalanced computation utilization, leading to inferior efficiency. The dynamic routing mechanism of MoE LLMs further aggravates the efficiency challenges. Therefore, in this paper, we propose HD-MoE to automatically optimize the MoE parallel computation across an NMP accelerator. HD-MoE features an offline automatic hybrid parallel mapping algorithm and an online dynamic scheduling strategy to reduce the communication costs while maximizing the computation utilization. With extensive experimental results, we demonstrate that HD-MoE achieves a speedup ranging from 1.1× to 1.8× over TP, 1.1× to 1.5× over EP, and 1.0× to 1.4× over the baseline Hybrid TP-EP with Compute-Balanced parallelism strategies.

Bioluminescence tomography (BLT) can provide non-invasive quantitative three-dimensional tumor information which has been widely applied in pre-clinical studies. Meanwhile, in recent years, deep learning methods have significantly improved the reconstruction resolution and speed by establishing a non-linear mapping relationship between surface-measured bioluminescence and light source distribution. However, this mapping relationship only works for specific biological tissues and light transmission processes under fixed wavelengths, resulting in poor stability and generalizability. To meet the requirements of diverse practical scenarios and inspired by more effective sparse regularization and graph representation theory, we propose a novel Graph Attention Iterative Contraction Network (GAICN) to conduct a finite element mesh spatial representation study. In the GAICN framework, two learnable spatial topological transforms based on the graph attention mechanism and an iterative contraction activation function were devised to achieve non-local feature aggregation and dynamic adjustment of weights between first-order neighboring nodes in the mesh. As a deep unrolling method, GAICN naturally inherits the coherence of surface bioluminescence with the light source in Forward-Backward Splitting (FBS), thus enhancing the generalizability, stability and interpretability of the network. Both simulation and in-vivo experiments further indicated that GAICN achieved superior reconstruction performance in terms of spatial location, dual light source resolution, stability, generalizability, as well as in-vivo practicability.

Training Graph Neural Networks on large graphs is challenging due to the need to store graph data and move them along the memory hierarchy. In this work, we tackle this by effectively compressing graph embedding matrix such that the model training can be fully enabled with on-chip compute and memory resources. Specifically, we leverage the graph homophily property and consider using Tensor-train to represent the graph embedding. This allows nodes with similar neighborhoods to partially share the feature representation.

Embedding retrieval is a crucial task for recommendation systems. Graph-based approximate nearest neighbor search (GANNS) is the most commonly used method for retrieval, and achieves the best performance on billion-scale datasets. Unfortunately, the existing CPU- and GPU-based GANNS systems are difficult to optimize the throughput under the latency constraints on billion-scale datasets, due to the underutilized local memory bandwidth (5-45%) and the expensive remote data access overhead (∼ 85% of the total latency). In this paper, we first introduce a practically ideal GANNS architecture for billion-scale datasets, which facilitates a detailed analysis of the challenges and characteristics of distributed GANNS systems. Then, at the architecture level, we propose DF-GAS, a Distributed FPGA-as-a-Service (FPaaS) architecture for accelerating billion-scale Graph-based Approximate nearest neighbor Search. DF-GAS uses a feature-packing memory access engine and a data prefetching and delayed processing scheme to increase local memory bandwidth by 36-42% and reduce remote data access overhead by 76.2%, respectively. At the system level, we exploit the “full-graph + sub-graph” hybrid parallel search scheme on distributed FPaaS system. It achieves million-level query-per-second with sub-millisecond latency on billion-scale GANNS for the first time. Extensive evaluations on million-scale and billion-scale datasets show that DF-GAS achieves an average of 55.4 ×, 32.2 ×, 5.4 ×, and 4.4 × better latency-bounded throughput than CPUs, GPUs, and two state-of-the-art ANNS architectures, i.e., ANNA [23] and Vstore [27], respectively.

Graph neural network (GNN) is a promising emerging application for link prediction, recommendation, etc. Existing hardware innovation is limited to single-machine GNN (SM-GNN), however, the enterprises usually adopt huge graph with large-scale distributed GNN (LSD-GNN) that has to be carried out with distributed in-memory storage. The LSD-GNN is very different from SM-GNN in terms of system architecture demand, workflow and operators, and hence characterizations.

We propose a novel architecture to efficiently perform sparse tensor decomposition/completion. As the generalization of vectors and matrices, tensors are widely used to process high-dimensional data. Sparse tensor decomposition (SpTD) is not only an emerging tensor analysis technique but also an effective tool to reduce the storage and computation costs of tensors. However, conventional general-purpose processors are inefficient to perform SpTD, mainly due to: i) variable sparsity degree and flexible buffer size requirement; ii) difficulties of fusing multiple execution kernels to pursue better performance. For domain-specific accelerator designers on the other hand, the diversity of decomposition algorithms is also an important problem that must be considered. To solve these challenges, we propose a unified abstraction for SpTD algorithms and design a specialized accelerator. First, we formulate two types of core kernels (SpLrMM and LrSampling) that serve as a standard form to fit a broad range of SpTD algorithms. Second, we design a sparse tensor engine (STE) to efficiently perform SpTD. STE uses a processing element (PE)-interactive architecture where PEs can be flexibly grouped together via Network-on-Chip (NoC) to share the buffer capacity, bandwidth, and compute resources. We evaluate our accelerator with extensive experiments, and it can achieve an average speedup of 45× over CPU and 29× over GPU.

Graph processing participates a vital role in mining relational data. However, the intensive but inefficient memory accesses make graph processing applications severely bottlenecked by the conventional memory hierarchy. In this work, we focus on inefficiencies that exist on both on-chip cache and off-chip memory. First, graph processing is known dominated by expensive random accesses, which are difficult to be captured by conventional cache and prefetcher architectures, leading to low cache hits and exhausting main memory visits. Second, the off-chip bandwidth is further underutilized by the small data granularity. Because each vertex/edge data in the graph only needs 4-8B, which is much smaller than the memory access granularity of 64B. Thus, lots of bandwidth is wasted fetching unnecessary data. Therefore, we present G-MEM, a customized memory hierarchy design for graph processing applications. First, we propose a coherence-free scratchpad as the on-chip memory, which leverages the power-law characteristic of graphs and only stores those hot data that are frequent-accessed. We equip the scratchpad memory with a degree-aware mapping strategy to better manage it for various applications. On the other hand, we design an elastic-granularity DRAM (EG-DRAM) to facilitate the main memory access. The EG-DRAM is based on near-data processing architecture, which processes and coalesces multiple fine-grained memory accesses together to maximize bandwidth efficiency. Putting them together, the G-MEM demonstrates a 2.48 × overall speedup over a vanilla CPU, with 1.44 × and 1.79 × speedup against the state-of-the-art cache architecture and memory subsystem, respectively.

The frequent data movement between the processor and the memory has become a severe performance bottleneck for deep neural network (DNN) training workloads in data centers. To solve this off-chip memory access challenge, the 3-D stacking processing-in-memory (3D-PIM) architecture provides a viable solution. However, existing 3D-PIM designs for DNN training suffer from the limited memory bandwidth in the base logic die. To overcome this obstacle, integrating the DNN related logic near each memory bank becomes a promising yet challenging solution, since naively implementing the floating-point (FP) unit and the cache in the memory die incurs a large area overhead. To address these problems, we propose DLUX, a high performance and energy-efficient 3D-PIM accelerator for DNN training using the near-bank architecture. From the hardware perspective, to support the FP multiplier with low area overhead, an in-DRAM lookup table (LUT) mechanism is invented. Then, we propose to use a small scratchpad buffer together with a lightweight transformation engine to exploit the locality and enable flexible data layout without the expensive cache. From the software aspect, we split the mapping/scheduling tasks during DNN training into intralayer and interlayer phases. During the intralayer phase, to maximize data reuse in the LUT buffer and the scratchpad buffer, achieve high concurrency, and reduce data movement among banks, a 3D-PIM customized loop tiling technique is adopted. During the interlayer phase, efficient techniques are invented to ensure the input-output data layout consistency and realize the forward-backward layout transposition. Experiment results show that DLUX can reduce FP32 multiplier area overhead by 60% against the direct implementation. Compared with a Tesla V100 GPU, end-to-end evaluations show that DLUX can provide on average 6.3× speedup and 42× energy efficiency improvement.

As deep neural networks (DNNs) continue their reach into a wide range of application domains, the neural network architecture of DNN models becomes an increasingly sensitive subject, due to either intellectual property protection or risks of adversarial attacks. Previous studies explore to leverage architecture-level events disposed in hardware platforms to extract the model architecture information. They pose the following limitations: requiring a priori knowledge of victim models, lacking in robustness and generality, or obtaining incomplete information of the victim model architecture.

In-situ approaches process data very close to the memory cells, in the row buffer of each subarray. This minimizes data movement costs and affords parallelism across subarrays. However, current in-situ approaches are limited to only row-wide bitwise (or few-bit) operations applied uniformly across the row buffer. They impose a significant overhead of multiple row activations for emulating 32-bit addition and multiplications using bitwise operations and cannot support operations with data dependencies or based on predicates. Moreover, with current peripheral logic, communication among subarrays is inefficient, and with typical data layouts, bits in a word are not physically adjacent. The key insight of this work is that in-situ, single-word ALUs outperform in-situ, parallel, row-wide, bitwise ALUs by reducing the number of row activations and enabling new operations and optimizations. Our proposed lightweight access and control mechanism, Fulcrum, sequentially feeds data into the single-word ALU and enables operations with data dependencies and operations based on a predicate. For algorithms that require communication among subarrays, we augment the peripheral logic with broadcasting capabilities and a previously-proposed method for low-cost inter-subarray data movement. The sequential processor also enables overlapping of broadcasting and computation, and reuniting bits that are physically adjacent. In order to realize true subarray-level parallelism, we introduce a lightweight column-selection mechanism through shifting one-hot encoded values. This technique enables independent column selection in each subarray. We integrate Fulcrum with Compress Express Link (CXL), a new interconnect standard. Fulcrum with one memory stack delivers on average (up to) 23.4 (76) speedup over a server-class GPU, NVIDIA P100, with three stacks of HBM2 memory, (ii) 70 (228) times speedup per memory stack over the GPU, and (iii) 19 (178.9) times speedup per memory stack over an ideal model of the GPU, which only accounts for the overhead of data movement.

Graph convolutional neural networks (GNN) have achieved state-of-the-art performance on tasks like node classification. It has become a new workload family member in data-centers. GNN works on irregular graph-structured data with three distinct phases: Combination, Graph Processing, and Aggregation. While Combination phase has been well supported by sgemm kernels in cuBLAS, the other two phases are still inefficient on GPGPU due to the lack of optimized CUDA kernels. In particular, Aggregation phase introduces large volume of DRAM storage footprint and data movement, and both Aggregation and Graph Processing phases suffer from high kernel launching time. These inefficiencies not only decrease training throughput but also limit users from training GNNs on larger graphs on GPGPU. Although these problems have been partially alleviated by recent studies, their optimizations are still not sufficient. In this paper, we propose fuseGNN, an extension of PyTorch that provides highly optimized APIs and CUDA kernels for GNN. First, two different programming abstractions for Aggregation phase are utilized to handle graphs with different average degrees. Second, dedicated GPGPU kernels are developed for Aggregation and Graph Processing in both forward and backward passes, in which kernel-fusion along with other optimization strategies are applied to reduce kernel launching time and latency as well as exploit data reuse opportunities. Evaluation on multiple benchmarks shows that fuseGNN achieves up to 5.3× end-to-end speedup over state-of-the-art frameworks, and the DRAM storage footprint is reduced by several orders of magnitude on large datasets.

With the ability to help wildlife conservation, precise medical care, and disease understanding, genomics analysis is becoming more and moe important. Recently, with the development and wide adoption of the Next-Generation Sequencing (NGS) technology, bio-data grows exponentially, putting forward great challenges for k-mer counting - a widely used application in genomics analysis.

Using deep neural networks (DNNs) in machine learning tasks is promising in delivering high-quality results but challenging to meet stringent latency requirements and energy constraints because of the memory-bound and the compute-bound execution pattern of DNNs. We propose a big-little dual-module inference to dynamically skip unnecessary memory accesses and computations to accelerate DNN inference. Leveraging the noise-resilient feature of nonlinear activation functions, we propose to use a lightweight little module that approximates the original DNN layer, termed as the big module, to compute activations of the insensitive region that are more noise-resilient. Hence, the expensive memory accesses and computations of the big module can be reduced as the results are only calculated in the sensitive region. For memory-bound models such as recurrent neural networks (RNNs), our method can reduce the overall memory accesses by 40% on average and achieve 1.54x to 1.75x speedup on a commodity CPU-based server platform with a negligible impact on model quality. In addition, our method can reduce the operations of the compute-bound models such as convolutional neural networks (CNNs) by 3.02x, with only a 0.5% accuracy drop.

Deep Neural Networks (DNNs) have been driving the mainstream of Machine Learning applications. However, deploying DNNs on modern hardware with stringent latency requirements and energy constraints is challenging because of the compute-intensive and memory-intensive execution patterns of various DNN models. We propose an algorithm-architecture co-design to boost DNN execution efficiency. Leveraging the noise resilience of nonlinear activation functions in DNNs, we propose dual-module processing that uses approximate modules learned from original DNN layers to compute insensitive activations. Therefore, we can save expensive computations and data accesses of unnecessary sensitive activations. We then design an Executor-Speculator dual-module architecture with support for balance execution and memory access reduction. With acceptable model inference quality degradation, our accelerator design can achieve 2.24x speedup and 1.97x energy efficiency improvement for compute-bound Convolutional Neural Networks (CNNs) and memory-bound Recurrent Neural Networks (RNNs).

The tremendous impact of deep learning algorithms over a wide range of application domains has encouraged a surge of neural network (NN) accelerator research. Facilitating the NN accelerator design calls for guidance from an evolving benchmark suite that incorporates emerging NN models. Nevertheless, existing NN benchmarks are not suitable for guiding NN accelerator designs. These benchmarks are either selected for general-purpose processors without considering unique characteristics of NN accelerators or lack quantitative analysis to guarantee their completeness during the benchmark construction, update, and customization. In light of the shortcomings of prior benchmarks, we propose a novel benchmarking methodology for NN accelerators with a quantitative analysis of application performance features and a comprehensive awareness of software-hardware co-design. Specifically, we decouple the benchmarking process into three stages: First, we characterize the NN workloads with quantitative metrics and select the representative applications for the benchmark suite to ensure diversity and completeness. Second, we refine the selected applications according to the customized model compression techniques provided by specific software-hardware co-design. Finally, we evaluate a variety of accelerator designs on the generated benchmark suite. To demonstrate the effectiveness of our benchmarking methodology, we conduct a case study of composing an NN benchmark from the TensorFlow Model Zoo and compress these selected models with various model compression techniques. Finally, we evaluate compressed models on various architectures, including GPU, Neurocube, DianNao, and Cambricon-X.

ReRAM-based processing-in-memory (PIM) architecture is a promising solution for deep neural networks (NN), due to its high energy efficiency and small footprint. However, traditional PIM architecture has to use a separate crossbar array to store either positive or negative (P/N) weights, which limits both energy efficiency and area efficiency. Even worse, imbalance running time of different layers and idle ADCs/DACs even lower down the whole system efficiency. This paper proposes AERIS, an Area/Energy-efficient 1T2R ReRAM based processing-In-memory NN System-on-a-chip to enhance both energy and area efficiency. We propose an area-efficient 1T2R ReRAM structure to represent both P/N weights in a single array, and a reference current cancelling scheme (RCS) is also presented for better accuracy. Moreover, a layer-balance scheduling strategy, as well as the power gating technique for interface circuits, such as ADCs/DACs, is adopted for higher energy efficiency. Experiment results show that compared with state-of-the-art ReRAM-based architectures, AERIS achieves 8.5x/1.3x peak energy/area efficiency improvements in total, due to layer-balance scheduling for different layers, power gating of interface circuits, and 1T2R ReRAM circuits. Furthermore, we demonstrate that the proposed RCS compensates the non-ideal factors of ReRAM and improves NN accuracy by 5.2% in the XNOR net on CIFAR-10 dataset.

Neural Network (NN) accelerators with emerging ReRAM (resistive random access memory) technologies have been investigated as one of the promising solutions to address the memory wall challenge, due to the unique capability of processing-in-memory within ReRAM-crossbar-based processing elements (PEs). However, the high efficiency and high density advantages of ReRAM have not been fully utilized due to the huge communication demands among PEs and the overhead of peripheral circuits. In this paper, we propose a full system stack solution, composed of a reconfigurable architecture design, Field Programmable Synapse Array (FPSA) and its software system including neural synthesizer, temporal-to-spatial mapper, and placement & routing. We highly leverage the software system to make the hardware design compact and efficient. To satisfy the high-performance communication demand, we optimize it with a reconfigurable routing architecture and the placement & routing tool. To improve the computational density, we greatly simplify the PE circuit with the spiking schema and then adopt neural synthesizer to enable the high density computation-resources to support different kinds of NN operations. In addition, we provide spiking memory blocks (SMBs) and configurable logic blocks (CLBs) in hardware and leverage the temporal-to-spatial mapper to utilize them to balance the storage and computation requirements of NN. Owing to the end-to-end software system, we can efficiently deploy existing deep neural networks to FPSA. Evaluations show that, compared to one of state-of-the-art ReRAM-based NN accelerators, PRIME, the computational density of FPSA improves by 31x; for representative NNs, its inference performance can achieve up to 1000x speedup.

Blockchain applications have shown huge potential in various domains. Proof of Work (PoW) is the key procedure in blockchain applications, which exhibits the memory-bound characteristic and hinders the performance improvement of blockchain accelerators. In order to mitigate the "memory wall" and improve the performance of memory-hard PoW accelerators, using Ethash as an example, we optimize the memory architecture from two perspectives: 1) Hiding memory latency. We propose specialized context switch design to overcome the uncertain cycles of repetitive memory requests. 2) Increasing memory bandwidth utilization. We introduce on-chip memory that stores a portion of the Ethash directed acyclic graph (DAG) for larger effective memory bandwidth, and further propose adopting embedded NOR flash to fulfill the role. Then, we conduct extensive experiments to explore the design space of our optimized memory architecture for Ethash, including number of hash cores, on-chip/off-chip memory technologies and specifications. Based on the design space exploration, we finally provide the guidance for designing the memory-bound PoW accelerator. The experiment results show that our optimized designs achieve 8.7% -- 55% higher hash rate and 17% -- 120% higher hash rate per Joule compared with the baseline design in different configurations.

Homomorphic encryption is a promising technology for enabling various privacy-preserving applications such as secure biomarker search. However, current implementations are not practical due to large performance overheads. A homomorphic encryption scheme has recently been proposed that allows bitwise comparison without the computationally-intensive multiplication and bootstrapping operations. Even so, this scheme still suffers from memory-bound performance bottleneck due to large ciphertext expansion. In this work, we propose HEGA, a near-data processing architecture that leverages this scheme with 3D-stacked memory to accelerate privacy-preserving biomarker search. We observe that homomorphic encryption-based search, like other emerging applications, can greatly benefit from the large throughput, capacity, and energy savings of 3D-stacked memory-based near-data processing architectures. Our near-data acceleration solution can speed up biomarker search by 6.3 × with 5.7× energy savings compared to an 8-core Intel Xeon processor.

Neural network accelerators are widely deployed in application systems for computer vision, speech recognition, and machine translation. Due to ubiquitous deployment of these systems, a strong incentive rises for adversaries to attack such artificial intelligence (AI) systems. Trojan is one of the most important attack models in hardware security domain. Hardware Trojans are malicious modifications to original ICs inserted by adversaries, which lead the system to malfunction after being triggered. The globalization of the semiconductor gives a chance for the adversary to conduct the hardware Trojan attacks.Previous works design Neural Network (NN) Trojans with access to the model, toolchain, and hardware platform. However, the threat model is impractical which hinders their real adoption. In this work, we propose a memory Trojan methodology without the help of toolchain manipulation and model parameter information. We first leverage the memory access patterns to identify the input image data. Then we propose a Trojan triggering method based on the dedicated input image other than the circuit events, which has better controllability. The triggering mechanism works well even with environment noise and preprocessing towards the original images. In the end, we implement and verify the effectiveness of accuracy degradation attack.

Resistive random access memory (ReRAM) demonstrates the great potential of in-memory processing for neural network (NN) acceleration. However, since the convolutional neural network (CNN) is widely known as compute-bound, current ReRAM-based accelerators are not able to support CNN efficiently. In this paper, we for the first time propose the CNN accelerator with Winograd's convolution on ReRAM (CNNWire), which minimizes the multiplications to enable fast and efficient CNN inference. We realize the convolution with Winograd Processing Element (WPE) based on convolutional tiles. Interconnections between WPEs are designed aiming to improve the data reuse. Finally, we introduce the full mapping flow to implement the Winograd convolution The results show that CNMWire gains 3.85x energy efficiency boosting and 3.24x speedup on average among different CNN benchmarks, compared with traditional GEMM based mapping.

Graph processing is an important analysis technique for a wide range of big data applications. The ability to explicitly represent relationships between entities gives graph analytics a significant performance advantage over traditional relational databases. However, at the microarchitecture level, performance is bounded by the inefficiencies in the memory subsystem for single-machine in-memory graph analytics. This paper consists of two contributions in which we analyze and optimize the memory hierarchy for graph processing workloads. First, we perform an in-depth data-type-aware characterization of graph processing workloads on a simulated multi-core architecture. We analyze 1) the memory-level parallelism in an out-of-order core and 2) the request reuse distance in the cache hierarchy. We find that the load-load dependency chains involving different application data types form the primary bottleneck in achieving a high memory-level parallelism. We also observe that different graph data types exhibit heterogeneous reuse distances. As a result, the private L2 cache has negligible contribution to performance, whereas the shared L3 cache shows higher performance sensitivity.

Domain-specific accelerators for graph analytics leverage a large on-chip memory in order to tackle the intensive random memory accesses, offering higher performance and energy efficiency than conventional architectures. However, limited by the inefficient usage of on-chip memory, current accelerators suffer from energy and performance bottlenecks due to the large amount of off-chip memory accesses. In this work, we introduce an online preprocessing step for the vertex-centric programming model based on our observation of imbalanced memory bandwidth utilization between two execution phases. Our scheme improves energy efficiency and performance by significantly reducing off-chip accesses in two ways. First, we sequence random off-chip memory accesses to balance memory bandwidth demands and improve the utilization of on-chip memory. Second, we prune active leaf vertices to avoid redundant memory accesses. We evaluate our method on a state-of-the-art graph analytics accelerator and achieve 1.6× speedup while reducing energy consumption by 42% on average.

Computational genomics has proven its great potential to support precise and customized health care. However, with the wide adoption of the Next Generation Sequencing (NGS) technology, 'DNA Alignment', as the crucial step in computational genomics, is becoming more and more challenging due to the booming bio-data. Consequently, various hardware approaches have been explored to accelerate DNA seeding - the core and most time consuming step in DNA alignment.

Graph analytics is an emerging application which extracts insights by processing large volumes of highly connected data, namely graphs. The parallel processing of graphs has been exploited at the algorithm level, which in turn incurs three irregularities onto computing and memory patterns that significantly hinder an efficient architecture design. Certain irregularities can be partially tackled by the prior domain-specific accelerator designs with well-designed scheduling of data access, while others remain unsolved.

Recent advances in deep learning (DL) have stimulated increasing interests in neural networks (NN). From the perspective of operation type and network architecture, deep neural networks can be categorized into full convolution-based neural network (ConvNet), recurrent neural network (RNN), and fully-connected neural network (FCNet). Different types of neural networks are usually cascaded and combined as a hybrid neural network (Hybrid-NN) to complete real-life cognitive tasks. Such hybrid-NN implementation is memory-intensive with large number of memory accesses, hence the performance of hybrid-NN is often limited by the insufficient memory bandwidth. A “3D + 2.5D” integration system, which integrates a high-bandwidth 3D stacked DRAM side-by-side with a highly-parallel neural processing unit (NPU) on a silicon interposer, overcomes the bandwidth bottleneck in hybrid-NN acceleration. However, intensive concurrent 3D DRAM accesses produced by the NPU lead to a serious thermal problem in 3D DRAM. In this paper, we propose a neural processor calledParanafor hybrid-NN acceleration in consideration of thermal problem of 3D DRAM. Parana solves the thermal problem of 3D memory by optimizing both the total number of memory accesses and memory accessing behaviors. For memory accessing behaviors, Parana balances the memory bandwidth by spatial division mapping hybrid-NN onto computing resources, which efficiently avoids that masses of memory accesses are issued in a short time period. To reduce the total number of memory accesses, we design a new NPU architecture and propose a memory-oriented tiling and scheduling mechanism to exploit the maximum utilization of on-chip buffer. Experimental results show that Parana reduces the peak temperature by up to 54.72$^\circ$C and the steady temperature by up to 32.27$^\circ$C over state-of-the-art accelerators with 3D memory without performance degradation.

Next Generation Sequencing (NGS) technology has become an indispensable tool for studying genomics, resulting in an exponentially growth of biological data. Booming data volume demands significant computational resources and creates challenges for 'Sequence Alignment', which is the most fundamental application in bioinformatics. Consequently, many researchers exploit both software and hardware methods to accelerate the most widely used sequence alignment algorithm - Basic Local Alignment Search Tool (BLAST). However, prior work suffers from moving huge DNA databases from the storage to computational units. Such data movement is both time and energy consuming.

Non-volatile main memory-based systems pose an opportunity for an attacker to readily access sensitive information on the memory because of its long retention time. While real-time memory encryption with dedicated AES engine can address this vulnerability, it incurs extra performance and energy overheads. As an alternative, we propose an AES in-memory implementation, AIM, to encrypt the whole/part of the memory only when it is necessary. We leverage the benefits offered by the inmemory computing architecture to address the challenges of the bandwidth intensive encryption application. We take advantage of NVM's intrinsic logic operation capability to implement the AES task. Embracing the massive parallelism inside the memory, AIM outperforms existing mechanisms with higher throughput yet lower energy consumption. Compared with state-of-the-art AES engine running at 2.1GHz, AIM can speed up the encryption process by 80 χ for a 1GB NVM.

Emerging non-volatile memories (NVM), such as phase change memory (PCM) and Resistive RAM (ReRAM), incorporate the features of fast byte-addressability and data persistence, which are beneficial for data services such as file systems and databases. To support data persistence, a persistent memory system requires ordering for write requests. The datapath of a persistent request consists of three segments: through the cache hierarchy to the memory controller, through the bus from the memory controller to memory devices, and through the network from a remote node to a local node. Previous work contributes significantly to improve the persistence parallelism in the first segment of the data path. However, we observe that the memory bus and the Remote Direct Memory Access (RDMA) network remain severely under-utilized because the persistence parallelism in these two segments is not fully leveraged during ordering. In this paper, we propose a novel architecture to further improve the persistence parallelism in the memory bus and the RDMA network. First, we utilize inter-thread persistence parallelism for barrier epoch management with better bank-level parallelism (BLP). Second, we enable intra-thread persistence parallelism for remote requests through RDMA network with buffered strict persistence. With these features, the architecture efficiently supports persistence through all three segments of the write datapath. Experimental results show that for local applications, the proposed mechanism can achieve 1.3× performance improvement, compared to the original buffered persistence work. In addition, it can achieve 1.93× performance improvement for remote applications serviced through the RDMA network.

Memory-centric architecture, which bridges the gap between compute and memory, is considered as a promising solution to tackle the memory wall and the power wall. Such architecture integrates the computing logic and the memory resources close to each other, in order to embrace large internal memory bandwidth and reduce the data movement overhead. The closer the compute and memory resources are located, the greater these benefits become. DRAM-based in-situ accelerators [1] tightly couple processing units to every memory bitline, achieving the maximum benefits among various memory-centric architectures. However, the processing units in such architectures are typically limited to simple functions like AND/OR due to strict area and power overhead constraints in DRAMs, making it difficult to accomplish complex tasks while providing high performance. In this paper, we address the challenge by applying stochastic computing arithmetic to the DRAM-based in-situ accelerator, targeting at the acceleration of error-tolerant applications such as deep learning. In stochastic computing, binary numbers are converted into stochastic bitstreams, which turns integer multiplications into simple bitwise AND operations, but at the expense of larger memory capacity/bandwidth demands. Stochastic computing is a perfect match for the DRAM-based in-situ accelerators because it addresses the in-situ accelerator's low performance problem by simplifying the operations, while leveraging the in-situ accelerator's advantage of large memory capacity/bandwidth. To further boost the performance and compensate for the numerical precision loss, we propose a novel Hierarchical and Hybrid Deterministic (H2D) stochastic computing arithmetic. Finally, we consider quantized deep neural network inference and training applications as a case study. The proposed architecture provides 2.3× improvement in performance per unit area compared with the binary arithmetic baseline, and 3.8× improvement over GPU. The proposed H2D arithmetic contributes 11× performance boost and 60% numerical precision improvement.

Spin-transfer torque magnetic random access memory (STT-RAM) technology has emerged as a potential replacement of SRAM in cache design, especially for building large-scale and energy-efficient last level caches. Compared with single-level cell (SLC), multi-level cell (MLC) STT-RAM is expected to double cache capacity and increase system performance. However, the two-step read/write access schemes incur considerable energy consumption and performance degradation. In this paper, we propose two techniques using data compression to optimize MLC STT-RAM cache design. The first technique tries to compress a cache line and fit it into only the soft-bit region of the cells, so that reading or writing this cache line takes only one step which is fast and energy-efficient. We introduce a second technique to increase the cache capacity by enabling the left hard-bit region to store another compressed cache line, which can improve the system performance for memory intensive workloads. The experimental results show that, compared with a conventional MLC STT-RAM last level cache design, our overhead minimized technique reduces the dynamic energy consumption by 38.2% on average with the same system performance, and our capacity augmented technique boosts the system performance by 6.1% with 19.2% dynamic energy saving on average, across the evaluated multi-programmed benchmarks.

Data movement between the processing units and the memory in traditional von Neumann architecture is creating the "memory wall" problem. To bridge the gap, two approaches, the memory-rich processor (more on-chip memory) and the compute-capable memory (processing-in-memory) have been studied. However, the first one has strong computing capability but limited memory capacity/bandwidth, whereas the second one is the exact the opposite.

The emerging spin-transfer torque magnetic random-access memory (STT-RAM) has attracted a lot of interest from both academia and industry in recent years. It has been considered as a promising replacement of SRAM and DRAM in the cache and memory system design thanks to many advantages, including non-volatility, low leakage power, SRAM comparable read performance and read energy consumption, higher density than SRAM, better scalability than conventional CMOS technologies, and good CMOS compatibility. However, the disadvantages of STT-RAM, such as higher write energy and longer write latency than SRAM, also bring design challenges. This paper introduces state-of-the-art architectural approaches to adopt STT-RAM in the cache and memory system design by taking advantage of the opportunities brought by STT-RAM as well as overcoming the challenges.

Processing-in-memory (PIM) provides high bandwidth, massive parallelism, and high energy efficiency by implementing computations in main memory, therefore eliminating the overhead of data movement between CPU and memory. While most of the recent work focused on PIM in DRAM memory with 3D die-stacking technology, we propose to leverage the unique features of emerging non-volatile memory (NVM), such as resistance-based storage and current sensing, to enable efficient PIM design in NVM. We propose Pinatubo1, a Processing In Non-volatile memory ArchiTecture for bUlk Bitwise Operations. Instead of integrating complex logic inside the cost-sensitive memory, Pinatubo redesigns the read circuitry so that it can compute the bitwise logic of two or more memory rows very efficiently, and support one-step multi-row operations. The experimental results on data intensive graph processing and database applications show that Pinatubo achieves a ~500× speedup, ~28000× energy saving on bitwise operations, and 1.12× overall speedup, 1.11× overall energy saving over the conventional processor.

3D die stacking and 2.5D interposer design are promising technologies to improve integration density, performance and cost. Current approaches face serious issues in dealing with emerging security challenges such as side channel attacks, hardware trojans, secure IC manufacturing and IP piracy. By utilizing intrinsic characteristics of 2.5D and 3D technologies, we propose novel opportunities in designing secure systems. We present: (i) a 3D architecture for shielding side-channel information; (ii) split fabrication using active interposers; (iii) circuit camouflage on monolithic 3D IC, and (iv) 3D IC-based security processing-in-memory (PIM). Advantages and challenges of these designs are discussed, showing that the new designs can improve existing countermeasures against security threats and further provide new security features.

Ternary Content-Addressable Memory (TCAM) is widely used in networking routers, fully associative caches, search engines, etc. While the conventional SRAM-based TCAM suffers from the poor scalability, the emerging nonvolatile memories (NVM, i.e., MRAM, PCM, and ReRAM) bring evolution for the TCAM design. It effectively reduces the cell size, and makes significant energy reduction and scalability improvement. New applications such as associative processors/accelerators are facilitated by the emergence of the nonvolatile TCAM (nvTCAM). However, nvTCAM design is challenging. In addition to the emerging device's uncertainty, the nvTCAM cell structure is so diverse that it results in a design space too large to explore manually. To tackle these challenges, we propose a circuit-level model and develop a simulation tool, NVSim-CAM, which helps researchers to make early design decisions, and to evaluate device/circuit innovations. The tool is validated by HSPICE simulations and data from fabricated chips. We also present a case study to illustrate how NVSim-CAM benefits the nvTCAM design. In the case study, we propose a novel 3D vertical ReRAM based TCAM cell, the 3DvTCAM. We project the advantages/disadvantages and explore the design space for the proposed cell with NVSim-CAM.

Processing-in-memory (PIM) is a promising solution to address the "memory wall" challenges for future computer systems. Prior proposed PIM architectures put additional computation logic in or near memory. The emerging metal-oxide resistive random access memory (ReRAM) has showed its potential to be used for main memory. Moreover, with its crossbar array structure, ReRAM can perform matrix-vector multiplication efficiently, and has been widely studied to accelerate neural network (NN) applications. In this work, we propose a novel PIM architecture, called PRIME, to accelerate NN applications in ReRAM based main memory. In PRIME, a portion of ReRAM crossbar arrays can be configured as accelerators for NN applications or as normal memory for a larger memory space. We provide microarchitecture and circuit designs to enable the morphable functions with an insignificant area overhead. We also design a software/hardware interface for software developers to implement various NNs on PRIME. Benefiting from both the PIM architecture and the efficiency of using ReRAM for NN computation, PRIME distinguishes itself from prior work on NN acceleration, with significant performance improvement and energy saving. Our experimental results show that, compared with a state-of-the-art neural processing unit design, PRIME improves the performance by ~2360x and the energy consumption by ~895x, across the evaluated machine learning benchmarks.

With the recent reincarnations of neuromorphic computing comes the promise of a new computing paradigm, with a focus on the design and fabrication of neuromorphic chips. A key challenge in design, however, is that programming such chips is difficult. This paper proposes a systematic methodology with a set of tools to address this challenge. The proposed toolset is called NEUTRAMS (Neural network Transformation, Mapping and Simulation), and includes three key components: a neural network (NN) transformation algorithm, a configurable clock-driven simulator of neuromorphic chips and an optimized runtime tool that maps NNs onto the target hardware for better resource utilization. To address the challenges of hardware constraints on implementing NN models (such as the maximum fan-in/fan-out of a single neuron, limited precision, and various neuron models), the transformation algorithm divides an existing NN into a set of simple network units and retrains each unit iteratively, to transform the original one into its counterpart under such constraints. It can support both spiking neural networks (SNNs) and traditional artificial neural networks (ANNs), including convolutional neural networks (CNNs) and multilayer perceptrons (MLPs) and recurrent neural networks (RNNs). With the combination of these tools, we have explored the hardware/software co-design space of the correlation between network error-rates and hardware constraints and consumptions. Doing so provides insights which can support the design of future neuromorphic architectures. The usefulness of such a toolset has been demonstrated with two different designs: a real Complementary Metal-Oxide-Semiconductor (CMOS) neuromorphic chip for both SNNs and ANNs and a processing-in-memory architecture design for ANNs.

Developing circuits for streaming applications written in C (or its variants) can benefit greatly from C-to-RTL (C2RTL) synthesis. Yet, most existing C2RTL tools lack system-level options to trade off various design constraints, such as delay and area. This article introduces a systematic way to accomplish C2RTL synthesis for streaming applications containing thousands of lines of C (or its variants) codes. Synthesizing circuits for such large applications presents serious challenges for existing C2RTL tools. Specifically, the proposed approach determines simultaneously the number of pipeline stages and the number of times that each functional block is duplicated in each pipeline stage. A mixed integer linear programming-based solution is formulated for obtaining the optimal solution. Furthermore, a heuristic algorithm is developed for large-scale problems. To accommodate the differences of the data rates between the adjacent hardware modules, first-in-first-out (FIFO) buffers are indispensable, but their overheads are nonnegligible. A parallelism-aware FIFO sizing method is also introduced to determine the optimal sizes of FIFOs. Experimental results on seven real-world applications demonstrate that the algorithms in the synthesis flow can make effective design trade-offs and find superior solutions in a short time compared with existing approaches. Furthermore, the algorithms achieve optimal results in most cases with subsecond running time.

The emerging nonvolatile memory (NVM) technology can potentially change the landscape of future IC designs with numerous benefits, such as high performance, low leakage power, and data retention. These advantages motivate designers to exploit utilizing NVM in in ASIC and FPGA. However, unique challenges such as large write energy and asymmetric read/write operations, lead to extra design knobs. This paper focuses on the NVM allocation and hierarchy optimization in high-level synthesis. A hierarchical hybrid memory architecture is presented. The proposed framework optimizes the memory hierarchy, type (NVM or SRAM) and capacity. Both an mixed-integer linear programming (MILP) and a branch-and-bound heuristic are developed. Experimental results demonstrate up to 69.3% power reduction compared with designs without NVM.

Energy harvesting is gaining more and more attentions due to its characteristics of ultra-long operation time without maintenance. However, frequent unpredictable power failures from energy harvesters bring performance and reliability challenges to traditional processors. Nonvolatile processors are promising to solve such a problem due to their advantage of zero leakage and efficient backup and restore operations. To optimize the nonvolatile processor design, this paper proposes new metrics of nonvolatile processors to consider energy harvesting factors for the first time. Furthermore, we explore the nonvolatile processor design from circuit to system level. A prototype of energy harvesting nonvolatile processor is set up and experimental results show that the proposed performance metric meets the measured results by less than 6.27% average errors. Finally, the energy consumption of nonvolatile processor is analyzed under different benchmarks.

Energy harvesting has been widely investigated as a promising method of providing power for ultra-low-power applications. Such energy sources include solar energy, radio-frequency (RF) radiation, piezoelectricity, thermal gradients, etc. However, the power supplied by these sources is highly unreliable and dependent upon ambient environment factors. Hence, it is necessary to develop specialized systems that are tolerant to this power variation, and also capable of making forward progress on the computation tasks. The simulation platform in this paper is calibrated using measured results from a fabricated nonvolatile processor and used to explore the design space for a nonvolatile processor with different architectures, different input power sources, and policies for maximizing forward progress.

To mitigate the “Power Wall” challenges for both mobile devices and data centers, accelerator-rich architecture with normally-off mode has been intensively studied recently. Power/energy optimization in high-level synthesis for accelerator design is critical for such accelerator-rich architecture. The emerging nonvolatile memory (NVM), offers many benefits such as ultra-low leakage power, high density, and instant power-on/off, and therefore is a promising alternative for the hardware accelerator design to achieve further power reduction. However, such NVM suffers from large write energy and latency, which brings new challenges for the buffer allocation in the custom accelerator design. This paper presents the first framework that optimizes NVM allocation in high-level synthesis for custom accelerator design, considering loop transformations. It solves the loop transformation, buffer allocation, and buffer type selection to minimize the memory power consumption, while under area, bandwidth, and performance constraints. This paper formulates the optimization problem, and solves it with a problem-specific designed stimulated annealing solution. Experiments demonstrate 32% extra power reduction compared with the previous method without optimizing loop transformations.

Solar-powered sensor nodes without specific energy maintenance have shown great promise in many applications, but they suffer from large energy storage and power converter loss. The storage-less and converter-less architecture with nonvolatile processing units has been proposed to reduce the energy loss. However, the architecture is sensitive to solar variations, since there is no energy buffering. Traditional inter-task scheduling methods may not work well due to large variations of task execution time. To tackle the challenge, we develop an algorithm for intra-task scheduling to achieve better quality of service. The experimental results show that the intra-task scheduling algorithm reduces deadline miss rate by as much as 35% and improves energy utilization by close to 20%.

Nonvolatile (NV) processors have attracted much attention in recent years due to their zero standby power, resilience to power failures, and instant-on feature. One design challenge of NV processors is the excess area needed by NV registers. This paper introduces a parallel compare and compress (PaCC) architecture to reduce such excess area. A key component of the PaCC architecture is a new codec which effectively balances area and performance. In addition, the PaCC architecture includes a configurable state table to support reference vector selection for different applications. With the proposed vector selection algorithm, the PaCC architecture can outperform other vector selection approaches by over 59% in terms of reduction in the number of NV registers. The proposed architecture has been fully realized at the circuit level and synthesized for the Rohm's 0.13-μm ferroelectric-CMOS hybrid process. Results demonstrate that the design can reduce the number of NV registers by 70%-80% with less than 1% overflow possibility, which leads to up to 30% processor area saving. The overall approach is applicable to any NV processor design regardless of the NV material used.

Developing FPGA solutions for streaming applications written in C (or its variants) can benefit greatly from automatic C-to-RTL (C2RTL) synthesis. Yet, the complexity and stringent throughput/cost constraints of such applications are rather challenging for existing C2RTL synthesis tools. This paper considers automatic partition and block-level parallelization to address these challenges. An MILP-based approach is introduced for finding an optimal partition of a given program into blocks while allowing block-level parallelization. In order to handle extremely large problem instances, a heuristic algorithm is also discussed. Experimental results based on seven well known multimedia applications demonstrate the effectiveness of both solutions.

Automatic C-to-RTL (C2RTL) synthesis can greatly benefit hardware design for streaming applications. However, stringent through-put/area constraints, especially the demand for power optimization at the system level is rather challenging for existing C2RTL synthesis tools. This paper considers a power-aware C2RTL framework using voltage-frequency islands (VFIs) to address these challenges. Given the throughput, area, and power constraints, an MILP-based approach is introduced to synthesize C-code into an RTL design by simultaneously considering three design knobs, i.e., partition, parallelization, and VFI assignment to get the global optimal solution. A heuristic solution is also discussed to deal with the scalability challenge facing the MILP formulation. Experimental results based on four well known multimedia applications demonstrate the effectiveness of both solutions.

In modern embedded systems, the C2RTL (high-level synthesis) technology helps the designer to greatly reduce time-to-market, while satisfying the performance and cost constraints. To attack the performance challenges in complex designs, we propose a FIFO-connected hierarchical approach to replace the traditional flatten one in stream applications. Furthermore, we develop an analytical algorithm to find the optimal FIFO capacity to connect multiple modules efficiently. Finally, we prove the advantages of the proposed method and the feasibility of our algorithm in seven real applications. Experimental results show that the hierarchical approach can have an up to 10.43 times speedup compared to the flatten design, while our analytical FIFO sizing algorithm shrinks design time from hours to seconds with the same accuracy compared to the simulation based approach.

Nonvolatile processor has become an emerging topic in recent years due to its zero standby power, resilience to power failures and instant on feature. This paper first demonstrated a fabricated nonvolatile 8051-compatible processor design, which indicates the ferroelectric nonvolatile version leads to over 90% area overhead compared with the volatile design. Therefore, we proposed a compare and compress recovery architecture, consisting of a parallel run-length codec (PRLC) and a state table logic, to reduce the area of nonvolatile registers. Experimental results demonstrate that it can reduce the number of nonvolatile registers by 4 times with less than 1% overflow possibility, which leads to 43% overall processor area savings. Furthermore, we implemented the novel PRLC and defined the method to optimize the optimal parallel degree to accelerate the compressions. Finally, we proposed a reconfigurable state table architecture, which supports the reference vector selecting for different applications. With our heuristic vector selecting algorithm, the optimal vector can provide over 42% better register number reduction than other vector selecting approaches. Our method is also applicable to designs with other nonvolatile materials based registers.