The Computing-In-Memory (CIM) paradigm offers a promising solution to the memory-wall bottleneck that limits conventional Von Neumann architectures. By performing data processing at the same physical location where the data are stored, CIM-based architectures minimize costly data movement and drastically improve energy efficiency. When implemented with Ferroelectric Field Effect Transistors (FeFETs), additional advantages from the non-volatility, fast switching, and low operating voltage of FeFETs are added. However, the widespread adoption of FeFETs is limited by their poor endurance, which is overcome by a Back End of the Line (BEoL) integration of FeFET-2, where a ferroelectric capacitor (FeCAP) is wired to the gate of a CMOS transistor providing high endurance compatible with low-power edge applications. These properties enable dense, low-power, and high-speed matrix operations essential for AI workloads. As a result, FeFET-2-based CIM accelerators offer a promising solution for energy-efficient, high-performance AI at the edge. The Ferro4EdgeAI project aims to develop an ultra low-power, scalable edge accelerator for AI, targeting a significant gain in energy efficiency with respect to state-of-the-art AI hardware accelerators. To attain this, our project focuses on innovation all along the value chain from materials, physic concepts, device architecture, integration technologies, and accelerators in a holistic design space exploration approach.

Deep neural networks (DNNs) have high computational and memory demands because they rely on multiplyaccumulate (MAC) operations and the amount of data that needs to be moved from the main memory to the compute unit. Since 2016, Systolic Arrays have emerged as a popular architecture for accelerating DNNs inference, although their performance is affected by multiple parameters, including the amount of used Processing Elements (PEs) and the dataflow. In this study, we perform a systematic evaluation of Systolic Array-based DNN accelerators using the ScaleSIM simulation framework to analyze different array sizes and dataflow configurations across different DNN models, measuring the effects of memory transfers, and execution latency across all DNN workloads, founding that dataflow option for routing data through an array is dependent on the specific workload; thereby, using moderate array sizes can achieve better performance than larger arrays, offering useful insights for Systolic Array-based DNN accelerators design.

The increasing adoption of AI models has driven applications toward the use of hardware accelerators to meet high computational demands and strict performance requirements. Beyond consideration of performance and energy efficiency, explainability and reliability have emerged as pivotal requirements, particularly for critical applications such as automotive, medical, and aerospace systems. Among the various AI models, Decision Tree Ensembles (DTEs) are particularly notable for their high accuracy and explainability. Moreover, they are particularly well-suited for hardware implementations, enabling high-performance and improved energy efficiency. However, a frequently overlooked aspect of DTEs is their reliability in the presence of hardware malfunctions. While DTEs are generally regarded as robust by design, due to their redundancy and voting mechanisms, hardware faults can still have catastrophic consequences. To address this gap, we present an in-depth reliability analysis of two types of DTE hardware accelerators: classical and approximate implementations. Specifically, we conduct a comprehensive fault injection campaign, varying the number of trees involved in the classification task, the approximation technique used, and the tolerated accuracy loss, while evaluating several benchmark datasets. The results of this study demonstrate that approximation techniques have to be carefully designed, as they can significantly impact resilience. However, techniques that target the representation of features and thresholds appear to be better suited for fault tolerance.

• A modular redundancy-based approximation is proposed for decision tree ensembles. • Modular redundancy is used to select only a subset of trees for classifying each class label. • This strategy allows aggressive approximation while preserving accuracy. • The effectiveness of the solution is demonstrated and shown. The deployment of machine learning models at the edge is crucial for enabling low-latency decision-making, optimizing resource utilization, and enhancing data confidentiality. Random Forest classifiers have proven to be highly accurate while offering computationally efficient inference, making them well-suited for resource-constrained edge devices. However, as the volume of training data grows, the complexity and size of these models also increase, limiting their deployment in edge computing scenarios. In order to address this challenge, we propose a novel approximation strategy for Random Forest classifiers leveraging on the concept of modular redundancy. In particular, our approach imposes that each target class is determined by only a subset of trees in a modular redundant fashion. This allows to prune from each tree the leaves related to no-longer relevant classes, significantly reducing the size of the model. To achieve an optimal balance between accuracy and resource savings with minimal computational time, we introduce an heuristic algorithm that determine the best subset of trees for each class. We evaluate our approach on multiple UCI machine learning datasets using a hardware accelerator for tree ensembles, demonstrating its effectiveness. The result shows that, on average, a 2.5% reduction in accuracy leads to save up to 50% in hardware overhead and energy consumption.

The reliability assessment of systems powered by artificial intelligence (AI) is becoming a crucial step prior to their deployment in safety and mission-critical systems. Recently, many efforts have been made to develop sophisticated techniques to evaluate and improve the resilience of AI models against the occurrence of random hardware faults. However, due to the intrinsic nature of such models, the comparison of the results obtained in state-of-the-art works is crucial, as reference models are missing. Moreover, their resilience is strongly influenced by the training process, the adopted framework and data representation, and so on. To enable a common ground for future research targeting CNN resilience analysis/hardening, this work proposes a first benchmark suite of DL models commonly adopted in this context, providing the models, the training/test data, and the resilience-related information (fault list, coverage, etc.) that can be used as a baseline for fair comparison. To this end, this research identifies a set of axes that have an impact on the resilience and classifies some popular CNN models, in both PyTorch and TensorFlow. Some final considerations are drawn, showing the relevance of a benchmark suite tailored for the resilience context.

Approximate Computing (AxC) is systematically applied across various abstraction levels to reduce overheads and enhance the performance of applications such as image processing and machine learning. However, AxC does not typically consider the specific workload (i.e., data input) of a given application. For instance, in signal processing applications like filters, some inputs are constants (filter coefficients), which allows for an additional level of approximation by considering the specific input distribution. This method is known as "Input-Aware Approximation" (IAA) and has shown potential advantages in previous studies. Unfortunately, existing input-aware design methodologies lack scalability as they mostly depend on ad-hoc, non-automatic design approaches, limiting their applicability. In this paper, we investigate how the input-aware approximate design approach can be integrated into a systematic, generic, and automatic design flow. We employ state-of-the-art approximation and multi-objective optimization techniques to achieve inputawareness. Our experimental results, focusing on classical signal processing applications like FIR filters, demonstrate that the input-aware approach can provide significant savings in both area and power consumption.

Convolutional Neural Networks (CNNs) are widely employed in various domains, including safety-critical applications such as autonomous driving. In these scenarios, the reliability of CNNs can be compromised by hardware faults occurring during inference, potentially leading to severe consequences. Evaluating the resilience of CNNs to hardware faults is primarily conducted through Fault Injection (FI) campaigns. However, a significant challenge lies in selecting an appropriate workload. Typically, the entire test set is applied for every injected fault, making the process highly time-consuming and posing difficulties for timely assessments. This paper investigates image selection strategies to rank inputs from the test dataset based on their difficulty in being classified by the CNN. The objective is to identify a minimal subset of test data that enables reliable CNN assessment while reducing computational overhead. By prioritizing challenging samples, the proposed method focuses on inputs that are more likely to reveal network vulnerabilities under fault conditions, enhancing the efficiency of the reliability evaluation process. Experimental results demonstrate that using such a subset of the test data suffices to estimate the number of critical faults for at least one image. This approach not only accelerates the reliability evaluation but also provides novel insights into CNN reliability, offering a practical framework for continuous assessments.

The IEEE European Test Symposium (ETS) has been facilitating progress in electronic systems testing since its launch in 1996. On the occasion of its 30th anniversary, this collaborative paper gathers sections by 21 ETS teams to outline their influential ideas and milestones. Each team’s section highlights historical perspective, current research, frameworks and projects as well as forward-looking research agendas in the area of electronic-based circuits and systems testing, reliability, safety, security and validation. This anniversary summary documents how research of various ETS teams, exemplifying the test community, has been evolving and transitioning from concepts to practical standards and Electronic Design Automation (EDA) tools and flows. This legacy is a strong base to drive the next generation of advances in electronic systems testing.

Assessing AI systems reliability is essential before deploying them in safety-critical applications. While recent efforts have focused on improving model resilience to random hardware faults, meaningful comparison remains difficult due to the lack of standardized reference models. Different authors use different implementations, which makes comparisons unfair and biased: resilience is influenced by the training processes, the software framework, and data representations. To address these issues, this work introduces a benchmark suite of CNN models to test the resilience of DNNs. The benchmark is structured on different axes: software framework, hardware platform, data representation, task and dataset. It is aimed at providing a shared foundation for fair and reproducible resilience evaluation.

Ferroelectric memory devices have emerged as a promising class of non-volatile memory technologies, offering a unique combination of high-speed operation, low power consumption, and good endurance compared to conventional flash memory. These devices leverage the bistable polarization states of ferroelectric materials to store data, enabling nonvolatile retention while maintaining fast read/write capabilities. The discovery of hafnium-based ferroelectric materials that are fully CMOS compatible and exhibit robust ferroelectricity at nanoscale dimensions has further enhanced their integration and scalability potential. For IoT devices, which require non-volatile state retention under constrained power budgets and frequent interruptions, we propose a novel FeAND memory cell designed to serve as a non-volatile backup for volatile memory. Unlike conventional ferroelectric memories that rely on current sensing, our design directly outputs a voltage signal, eliminating the need for sensing circuits. The cell exhibits a logical AND-like behavior, enabled by an innovative read scheme based on a CMOS inverter. The cell can function as both a non-volatile memory element and a logic gate where one input is permanently stored as a polarization state. This dual functionality enables novel Computing-in-Memory architectures by embedding logic operations directly within the memory array. We validate our design using Cadence Spectre simulations with the GlobalFoundries 28SLP technology.

Nowadays, AI applications are becoming extremely popular in our everyday life as well as for the industry. Recent incidents involving hyperscalers have revealed that even cloud-based datacenter hardware can experience failures leading to Silent Data Corruptions (SDCs), also called Silent Data Errors (SDEs). This Special Session delves into the implications of such failures on AI workloads, both during training and inference, and explores methodologies for efficiently detecting SDCs or SDEs through dedicated monitoring phases.

Evaluating the reliability of deep neural network (DNN) applications is essential to assess efficient error resilience techniques and deploy artificial intelligence (AI) on safety-critical embedded systems. This work presents a novel methodology for assessing the reliability of DNN models deployed on microcontrollers using software-level fault injection. A new tool was developed, on top of the TensorFlow Lite library for C/C++, to inject faults into memory and CPU registers during DNN inference. Providing a portable, fast, and configurable approach for injecting multiple fault types. The proposed framework was validated with two case studies, using a variety of literature-endorsed models and datasets, the results provide a comparative analysis of DNN reliability versus register and memory bit flips, induced by transient faults.

Deep Learning (DL) [1] is currently one of the most intensively and widely used predictive models in the field of machine learning. DL has proven to give very good results for many complex tasks and applications, such as object recognition in images/videos, natural language processing, robotics, aerospace, smart healthc are, and autonomous driving. Nowa-days, there is intense activity in designing custom Artificial Intelligence (AI) hardware accelerators to support the energy-hungry data movement, speed of computation, and memory resources that DL requires to realize its full potential [2]. Furthermore, there is an incentive to migrate AI from cloud to edge devices, i.e., Internet-of- Things devices, to address data confidentiality issues and bandwidth limitations, and also to alleviate the communication latency, especially for real-time safety-critical decisions, e.g., in autonomous driving.

This multi-partner-project contribution introduces the midway results of the Horizon 2020 FVLLMONTI project. In this project we develop a new and ultra-efficient class of ANN accelerators, the neural network compute cube$(N^{2}C^{2})$, which is specifically designed to execute complex machine learning tasks in a 3D technology, in order to provide the high computing power and ultra-high efficiency needed for future edgeAI applications. We showcase its effectiveness by targeting the challenging class of Transformer ANNs, tailored for Automatic Speech Recognition and Machine Translation, the two fundamental components of speech-to-speech translation. To gain the full benefit of the accelerator design, we develop disruptive vertical transistor technologies and execute design-technology-co-optimization (DTCO) loops from single device, to cell and compute cube level. Further, a hardware-software-co-optimization is executed, e.g. by compressing the executed speech recognition and translation models for energy efficient executing without substantial loss in precision.

State-of-the-Art (SotA) hardware implementations of Deep Neural Networks (DNNs) incur high latencies and costs. Binary Neural Networks (BNNs) are potential alternative solutions to realize faster implementations without losing accuracy. In this paper, we first present a new data mapping, called TacitMap, suited for BNNs implemented based on a Computation-In-Memory (CIM) architecture. TacitMap maximizes the use of available parallelism, while CIM architecture eliminates the data movement overhead. We then propose a hardware accelerator based on optical phase change memory (oPCM) called EinsteinBarrier. Ein-steinBarrier incorporates TacitMap and adds an extra dimension for parallelism through wavelength division multiplexing, leading to extra latency reduction. The simulation results show that, compared to the SotA CIM baseline, TacitMap and EinsteinBarrier significantly improve execution time by up to$\sim 154\times$and$\sim 3113\times$, respectively, while also maintaining the energy consumption within 60% of that in the CIM baseline.

Systolic array has emerged as a prominent archi-tecture for Deep Neural Network (DNN) hardware accelerators, providing high-throughput and low-latency performance essen-tial for deploying DNNs across diverse applications. However, when used in safety-critical applications, reliability assessment is mandatory to guarantee the correct behavior of DNN accelerators. While fault injection stands out as a well-established practical and robust method for reliability assessment, it is still a very time-consuming process. This paper addresses the time efficiency issue by introducing a novel hierarchical software-based hardware-aware fault injection strategy tailored for systolic array-based DNN accelerators. The uniform Recurrent Equations system is used for software modeling of the systolic-array core of the DNN accelerators. The approach demonstrates a reduction of the fault injection time up to 3 × compared to the state-of-the-art hybrid (software/hardware) hardware-aware fault injection frameworks and more than 2000 × compared to RT-level fault injection frameworks - without compromising accuracy. Additionally, we propose and evaluate a new reliability metric through experimental assessment. The performance of the framework is studied on state-of-the-art DNN benchmarks.

This paper aims to comprehensively explore challenges and opportunities to design highly efficient Neural Network (NN) systems through Approximate Computing (AxC) techniques while ensuring fault tolerance properties. By highlighting the intrinsic conflicting goals of AxC and fault tolerance principles, the study aims to stimulate and contribute to a deeper understanding of how important it is to consider fault tolerance requirements while designing approximate-computing-based systems. This is key to developing highly efficient fault-tolerant architectures for Neural Networks.

In today’s digital landscape, cryptography plays a vital role in ensuring communication security through encryption and authentication algorithms. While traditional cryptographic methods rely on hard mathematical problems for security, the rise of quantum computing threatens their effectiveness. Post-Quantum Cryptography (PQC) algorithms, like CRYSTALSKyber, aim to withstand quantum attacks. Recently standardized, CRYSTALS-Kyber is a lattice-based algorithm designed to resist quantum attacks. However, its implementation faces computational challenges, particularly with Keccak-based functions, which are crucial for security and upon which the FIPS 202 standard is based. Our paper addresses this technological challenge by designing a FIPS 202 hardware accelerator to enhance CRYSTALS-Kyber efficiency and security. We chose to implement the entire FIPS 202 standard in hardware in order to widen the applicability of the accelerator to all possible algorithms that rely on such hash functions, taking care to provide realistic assumptions on system-level integration inside a System-on-Chip (SoC). We provide results in terms of area, frequency, and clock cycles for both ASIC and FPGA targets. An area reduction of up to $22.3 \%$ is achieved with respect to state-ofthe-art solutions. In addition, we integrated the accelerator inside a 32-bit RISC-V based security-oriented SoC, where we show a strong performance gain on CRYSTALS-Kyber execution. The design presented in this paper performs better in all Kyber1024 primitives, with an improvement up to $3.21 \times$ in Kyber-KeyGen.

Convolutional Neural Networks (CNNs) are vulnerable to undetectable manipulated inputs that reduce model accuracy. There are several methods to counter these Adversarial Attacks, however, resource-constrained systems require simpler solutions due to memory and processing limitations. This work explores the application of single-layer redundancy to implement a dynamic model in TensorFlow CNNs and its impact on mitigating Adversarial Attacks.

Since Deep Neural Networks (DNNs) gracefully withstands approximation due to its inherent redundancy, Approximate Computing (AxC) can be applied to reduce power consumption and execution time. In the literature, several works adopted the AxC paradigm to DNNs in the form of quantization, precision reduction, pruning, and functional approximation. Despite the promising results demonstrated so far, most of the existing works have applied homogeneous AxC techniques, meaning that the same degree of approximation has been applied to the entire DNN. However, different DNN components (i.e., channels, filters, layers, neurons) have different resiliency levels. This paper presents a framework for applying heterogeneous AxC to DNN hardware accelerators. The framework is based on the identification of channel resilience and applying a tailored degree of approximation per channel. Preliminary results carried out on the LeNet-5 model show that by using the proposed framework it is possible to decrease resource utilization by 65.2% and power consumption by 53.4% at the cost of a marginal drop of accuracy from 98.87% to 98.03%.

The vertical nanowire field effect transistor (VN-WFET) is an emerging technology that promises to improve the sustainability of future transistor scaling beyond the limitations of conventional lateral devices. With its 3D gate-all-around (GAA) architecture, such a technology enables designs with improved energy-efficiency as well as reduced footprint and thus interconnect capacitance. In this work, and based on the compact model of a real VNWFET device, we present the design flow for the generation of a standard cell library starting from the circuit and physical design of logic cells to logic synthesis based on the VNWFET technology. The results on the synthesized benchmark cells, as compared against 45nm and 65nm CMOS libraries, demonstrate a significant decrease in the average dynamic power consumption and delay values up to 71X and 34X respectively, with anaveragearea gain of up to 5X. However, an increase in leakage power consumption (up to 2X on average) was also observed.

Reliability assessment is mandatory to guarantee the correct behavior of Deep Neural Network (DNN) hardware accelerators in safety-critical applications. While fault injection stands out as a well-established, practical and robust method for reliability assessment, it is still a very time-consuming process. This paper contributes with three recipes for optimizing the efficiency of the reliability assessment: a) hybrid analytical and hierarchical FI-based reliability assessment for systolic-array-based DNN accelerators; b) mixing techniques for the reliability assessment of in-chip AI accelerators in GPUs; c) reliability assessment of DNN hardware accelerators through physical fault injection. The experimental results demonstrate the efficiency of the proposed methods applied to their target DNN HW accelerator platforms.

Approximate Computing (AxC) paradigm was introduced to achieve higher power efficiency, lower area and better performances w.r.t. a “classical” computing system at the cost of a degraded, but still acceptable, output accuracy [1]. AxC can be applied at several abstraction levels of a given computing system: from circuit to algorithm [1], leading to a wide design exploration space that quickly became the bottleneck for successfully deploying AxC. Indeed, the literature proposes many works to automatically trade-off between output accuracy and performances [2]. However, most of them lack the capability to identify resilient elements (e.g, HW component, HDL statements, etc.) of the design to be approximated. Consequently, exploring the design for AxC generally results in a long and tedious procedure. Existing approaches generate approximate variants of the Design Under Exploration (DUE). Every variant is then executed/simulated in order to determine the accuracy degradation [3], which depends on the application and requires a specific metric to be computed (e.g., similarity index, hamming distance, etc.).

This paper investigates the potential of a compute-in-memory core based on optical Phase Change Materials (oPCMs) to speed up and reduce the energy consumption of the Matrix-Matrix-Multiplication operation. The paper also proposes a new data mapping for Binary Neural Networks (BNNs) tailored for our oPCM core. The preliminary results show a significant latency improvement irrespective of the evaluated network structure and size. The improvement varies from network to network and goes up to ~1053x.

Nowadays, Deep Neural Networks (DNNs) are one of the most computationally-intensive algorithms because of the (i) huge amount of data to be transferred from/to the memory, and (ii) the huge amount of matrix multiplications to compute. These issues motivate the design of custom DNN hardware accelerators. These accelerators are widely used for low-latency safety-critical applications such as object detection in autonomous cars. Safety-critical applications have to be resilient with respect to hardware faults and Deep Learning (DL) accelerators are subjected to hardware faults that can cause functional failures, potentially leading to catastrophic consequences. Although DNNs possess a certain level of intrinsic resilience, it varies depending on the hardware on which they are run. The intent of the paper is to assess the resilience of a systolic-array-based DNN accelerator in the presence of hardware faults, in order to identify the architectural parameters that may mainly impact the DNN resilience.

Data-centric applications such as Artificial Intelligence and IoT are putting stringent performance and energy efficiency constraints on hardware implementations of computing architectures. Computing in Memory paradigm appears as a viable approach to to address such constraints and ferroelectric FETs (FeFETs) push this paradigm at a finer grain by enabling the design of true non-volatile logic gates, by implementing tight combination of memory and logic called Logic-in-Memory (LiM). From the basic non-volatile logic gate design up to the application-level evaluation, several challenges have to be addressed.In this paper, we present a methodology to design complex operations such as cryptographic operations using FeFET, integrate them into a complete computing architecture and evaluate its benefits. Current challenges related to logic synthesis and tools for synthesis of these LiM structures will also be discussed.

Deep Learning, and in particular, Deep Neural Network (DNN) is nowadays widely used in many scenarios, including safety-critical applications such as autonomous driving. In this context, besides energy efficiency and performance, reliability plays a crucial role since a system failure can jeopardize human life. As with any other device, the reliability of hardware architectures running DNNs has to be evaluated, usually through costly fault injection campaigns. This paper explores approximation and fault resiliency of DNN accelerators. We propose to use approximate (AxC) arithmetic circuits to agilely emulate errors in hardware without performing fault injection on the DNN. To allow fast evaluation of AxC DNN, we developed an efficient GPU-based simulation framework. Further, we propose a fine-grain analysis of fault resiliency by examining fault propagation and masking in networks.

The field of neuromorphic computing has been rapidly evolving in recent years, with an increasing focus on hardware design and reliability. This special session paper provides an overview of the recent developments in neuromorphic computing, focusing on hardware design and reliability. We first review the traditional CMOS-based approaches to neuromorphic hardware design and identify the challenges related to scalability, latency, and power consumption. We then investigate alternative approaches based on emerging technologies, specifically integrated photonics approaches within the NEUROPULS project. Finally, we examine the impact of device variability and aging on the reliability of neuromorphic hardware and present techniques for mitigating these effects. This review is intended to serve as a valuable resource for researchers and practitioners in neuromorphic computing.

International audience

The computational workload involved in Convolutional Neural Networks (CNNs) is typically out of reach for low-power embedded devices. The scientific literature provides a large number of approximation techniques to address this problem. Among them, the Weight-Sharing (WS) technique gives promising results, but it requires carefully determining the shared values for each layer of a given CNN. As the number of possible solutions grows exponentially with the number of layers, the WS Design Space Exploration (DSE) time can easily explode for state-of-the-art CNNs. In this paper, we propose a new heuristic approach to drastically reduce the exploration time without sacrificing the quality of the output. The results carried out on recent CNNs (GoogleNet [1], ResNet50V2 [2], MobileNetV2 [3], InceptionV3 [4], and EfficientNet [5]), trained with the ImageNet [6] dataset, show over 5× memory compression at an acceptable accuracy loss (complying with the MLPerf [7] quality target) without any retraining step and in less than 10 hours. Our code is publicly available on GitHub [8].

In the literature, there are few studies describing how to implement Boolean logic functions as a memristor-based crossbar architecture and some solutions have been actually proposed targeting back-end synthesis. However, there is a lack of methodologies and tools for the synthesis automation. The main goal of this paper is to perform a Design Space Exploration (DSE) in order to analyze and compare the impact of the most used optimization algorithms on a memristor-based crossbar architecture. The results carried out on 102 circuits lead us to identify the best optimization approach, in terms of area/energy/delay. The presented results can also be considered as a reference (benchmarking) for comparing future work.

Today we observe amazing performance achieved by Machine Learning (ML); for specific tasks it even surpasses human capabilities. Unfortunately, nothing comes for free: the hidden cost behind ML performance stems from its high complexity in terms of operations to be computed and the involved amount of data. For this reasons, custom Artificial Intelligence hardware accelerators based on alternative computing paradigms are attracting large interest. Such dedicated devices support the energy-hungry data movement, speed of computation, and memory resources that MLs require to realize their full potential. However, when ML is deployed on safety-/mission-critical applications, dependability becomes a concern. This paper presents the state of the art of custom Artificial Intelligence hardware architectures for ML, here Spiking and Convolutional Neural Networks, and shows the best practices to evaluate their dependability.

In the literature, it is argued that Deep Neural Networks (DNNs) possess a certain degree of robustness mainly for two reasons: their distributed and parallel architecture, and their redundancy introduced due to over provisioning. Indeed, they are made, as a matter of fact, of more neurons with respect to the minimal number required to perform the computations. It means that they could withstand errors in a bounded number of neurons and continue to function properly. However, it is also known that different neurons in DNNs have divergent fault tolerance capabilities. Neurons that contribute the least to the final prediction accuracy are less sensitive to errors. Conversely, the neurons that contribute most are considered critical because errors within them could seriously compromise the correct functionality of the DNN. This paper presents a software methodology based on a Triple Modular Redundancy technique, which aims at improving the overall reliability of the DNN, by selectively protecting a reduced set of critical neurons. Our findings indicate that the robustness of the DNNs can be enhanced, clearly, at the cost of a larger memory footprint and a small increase in the total execution time. The trade-offs as well as the improvements are discussed in the work by exploiting two DNN architectures: ResNet and DenseNet trained and tested on CIFAR-10.

As modern applications demand an unprecedented level of computational resources, traditional computing system design paradigms are no longer adequate to guarantee significant performance enhancement at an affordable cost. Approximate Computing (AxC) has been introduced as a potential candidate to achieve better computational performances by relaxing non-critical functional system specifications. In this article, we propose a systematic and high-abstraction-level approach allowing the automatic generation of near Pareto-optimal approximate configurations for a Discrete Cosine Transform (DCT) hardware accelerator. We obtain the approximate variants by using approximate operations, having configurable approximation degree, rather than full-precise ones. We use a genetic searching algorithm to find the appropriate tuning of the approximation degree, leading to optimal tradeoffs between accuracy and gains. Finally, to evaluate the actual HW gains, we synthesize non-dominated approximate DCT variants for two different target technologies, namely, Field Programmable Gate Arrays (FPGAs) and Application Specific Integrated Circuits (ASICs). Experimental results show that the proposed approach allows performing a meaningful exploration of the design space to find the best tradeoffs in a reasonable time. Indeed, compared to the state-of-the-art work on approximate DCT, the proposed approach allows an 18% average energy improvement while providing at the same time image quality improvement.

International audience

In computer engineering, logic synthesis is a process by which an abstract specification of desired circuit behavior is turned into a design implementation in terms of logic gates. Historically, logic synthesis was tightly related to the physical implementation of the logic gates. Nowadays, pushed by the forecasted end of Moore's law, several emerging technologies (e.g., nanodevices, optical computing, quantum computing) are candidates to either replace or co-exist with the de facto standard CMOS technology. The main consequence of the rising of those emerging technologies is that the logic synthesis has to face new issues and, at the same time, exploits new opportunities. The goal of this paper is thus to present three emerging technologies (Vertical Nanowire Field Effect Transistors, Ferroelectric Transistors, and Memristors), how to use them to implement logic gates, and the main challenges and issues for the logic synthesis.

Solutions based on artificial intelligence and brain-inspired computations like Artificial Neural Networks (ANNs) are suited to deal with the growing computational complexity required by state-of-the-art electronic devices. Many applications that are being deployed using these computational models are considered safety-critical (e.g., self-driving cars), producing a pressing need to evaluate their reliability. Besides, state-of-theart ANNs require significant memory resources to store their parameters (e.g., weights, activation values), which goes outside the possibility of many resource-constrained embedded systems. In this light, Approximate Computing (AxC) has become a significant field of research to improve memory footprint, speed, and energy consumption in embedded and high-performance systems. The use of AxC can significantly reduce the cost of ANN implementations, but it may also reduce the inherent resiliency of this kind of application. On this scope, reliability assessments are carried out by performing fault injection test campaigns. The intent of the paper is to propose a framework that, relying on the results of radiation tests in Commercial-Off-The-Shelf (COTS) devices, is able to assess the reliability of a given application. To this end, a set of different radiation-induced errors in COTS memories is presented. Upon these, specific fault models are extracted to drive emulation-based fault injections.

Approximate Computing (AxC) trades off between the accuracy required by the user and the precision provided by the computing system to achieve several optimizations such as performance improvement, energy, and area reduction. Several AxC techniques have been proposed so far in the literature. They work at different abstraction levels and propose both hardware and software implementations. The standard issue of all existing approaches is the lack of a methodology to estimate the impact of a given AxC technique on the application-level accuracy. This paper proposes a probabilistic approach based on Bayesian networks to quickly estimate the impact of a given approximation technique on application-level accuracy. Moreover, we have also shown how Bayesian networks allow a backtrack analysis that automatically identifies the most sensitive components. That influence analysis dramatically reduces the space exploration for approximation techniques. Preliminary results on a simple artificial neural network shown the efficiency of the proposed approach.

The design and implementation of Convolutional Neural Networks (CNNs) for deep learning (DL) is currently receiving a lot of attention from both industrials and academics. However, the computational workload involved with CNNs is often out of reach for low power embedded devices and is still very costly when running on datacenters. By relaxing the need for fully precise operations, approximate computing substantially improves performance and energy efficiency. Deep learning is very relevant in this context, since playing with the accuracy to reach adequate computations will significantly enhance performance, while keeping quality of results in a user-constrained range. AdequateDL is a project aiming to explore how approximations can improve performance and energy efficiency of hardware accelerators in DL applications. This paper presents the main concepts and techniques related to approximation of CNNs and preliminary results obtained in the AdequateDL framework.

Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. To date, there are very few publications covering all the steps (from the system-level to the transistor-level) necessary to design, model, verify, implement, integrate, and test a silicon system. Our tutorial targets this empty space and intents to bridge the gap between system and circuit designers, technologists, and physicist. It is extremely important nowadays (and will be more important in the future) for system and circuit designers to understand the physical implications of system and circuit solutions based on hardware/software codesign as well as for technologists and physicists to cope with the system and circuit requirements in terms of energy, speed, and data throughput. The tutorial addresses all the steps of design, modeling, verification, implementation, integration, and test of advance silicon systems for wireless local-area networks (WLAN).

The paper addresses some of the opportunities and challenges related to test and reliability of three major emerging computing paradigms; i.e., Quantum Computing, Computing engines based on Deep Neural Networks for AI, and Approximate Computing (AxC). We present a quantum accelerator showing that it can be done even without the presence of very good qubits. Then, we present Dependability for Artificial Intelligence (AI) oriented Hardware. Indeed, AI applications shown relevant resilience properties to faults, meaning that the testing strongly depends on the application behavior rather than on the hardware structure. We will cover AI hardware design issues due to manufacturing defects, aging faults, and soft errors. Finally, We present the use of AxC to reduce the cost of hardening a digital circuit without impacting its reliability. In other words how to go beyond usual modular redundancy scheme.

In the last decade, Approximate Computing (AxC) has been studied as a possible alternative computing paradigm. It has been used to reduce the overhead cost of conventional fault tolerant schemes, such as the Triple Modular Redundancy (TMR). One of the most recent propositions is the concept of Quadruple Approximate Modular Redundancy (QAMR). QAMR reduces the overhead cost w.r.t. conventional TMR structures, while guaranteeing the same fault-tolerance capability. In this paper, we propose a new approximation technique to realize the QAMR and we perform a Design Space Exploration (DSE) to find QAMR Pareto-optimal implementations. Moreover, we provide the design of a new majority voter for the proposed architecture. Experimental results show that it is possible to find QAMR variants achieving area and/or delay gains compared to the TMR counterpart, for 85.4% and 97% of the examined circuits for FPGA and ASIC technologies respectively.

The computing continuum's actual trend is facing a growth in terms of devices with any degree of computational capability. Those devices may or may not include a full-stack, including the Operating System layer and the Application layer, or just facing pure bare-metal solutions. In either case, the reliability of the full system stack has to be guaranteed. It is crucial to provide data regarding the impact of faults at all system stack levels and potential hardening solutions to design highly resilient systems. While most of the work usually concentrates on the application reliability, the special session aims to provide a deep comprehension of the impact on the reliability of an embedded system when faults in the hardware substrate of the system stack surface at the Operating System layer. For this reason, we will cover a comparison from an application perspective when hardware faults happen in bare metal vs. real-time OS vs. general-purpose OS. Then we will go deeper within a FreeRTOS to evaluate the contribution of all parts of the OS. Eventually, the Special Session will propose some hardening techniques at the Operating System level by exploiting the scheduling capabilities.

In this paper, we present anytime instructions for floating-point additions and multiplications. Specific to such instructions is their ability to compute an arithmetic operation at a programmable accuracy of a most significant bits where a is encoded in the instruction itself. Contrary to reduced-precision architectures, the word length is maintained throughout the execution. Two approaches are presented for the efficient implementation of anytime additions and multiplications, one based on on-line arithmetic and the other on bitmasking. We propose implementations of anytime functional units for both approaches and evaluate them in terms of error, latency, area, as well as energy savings. As a result, 15% of energy can be saved on average while computing a floating-point addition with an error of less than 0.1%. Moreover, large latency and energy savings are reported for iterative algorithms such as a Jacobi algorithm with savings of up to 39% in energy.

Deep neural networks demonstrate impressive levels of performance, particularly in computer vision and speech recognition. However, the computational workload and associated storage inhibit their potential in resource-limited embedded systems. The approximate computing paradigm has been widely explored in the literature. It improves performance and energy-efficiency by relaxing the need for fully accurate operations. There are a large number of implementation options with very different approximation strategies (such as pruning, quantization, low-rank factorization, knowledge distillation, etc.). To the best of our knowledge, no automated approach exists to explore, select and generate the best approximate versions of a given convolutional neural network (CNN) according to the design objectives. The goal of this work in progress is to demonstrate that the design space exploration phase can enable significant network compression without noticeable accuracy loss. We demonstrate this via an example based on weight sharing and show that our method can obtain a 4x compression rate in an int-16 version of LeNet-5 (5-layer 1,720-kbit CNNs) without re-training and without any accuracy loss.

Approximate Integrated Circuits (AxICs) have emerged in the last decade as an outcome of Approximate Computing (AxC) paradigm. AxC focuses on efficiency of computing systems by sacrificing some computation quality. As AxICs spread, consequent challenges to test them arose. On the other hand, the opportunity to increase the production yield emerged in the AxIC context. Indeed, some particular defects in the manufactured AxIC might not catastrophically impact the final circuit quality. Therefore, some defective AxICs might still be acceptable. Efforts to detect favorable conditions to consider defective AxICs as acceptable - with the goal to increase the production yield - have been done in last years. Unfortunately, the final achieved yield gain is often not as high as expected. In this work, we propose a methodology to actually achieve a yield gain as close as possible to expectations, by proposing a technique to suitably apply tests to AxICs. Experiments carried out on state-of-the-art AxICs show yield gain results very close to the expected ones (i.e., between 98% and 100% of the expectations).

The following topics are dealt with: embedded systems; logic design; CMOS integrated circuits; neural nets; integrated circuit design; field programmable gate arrays; low-power electronics; formal verification; multiprocessing systems; Internet of Things.

Safety-critical applications are frequently based on deep learning algorithms. In particular, Convolutional Neural Networks (CNNs) are commonly deployed in autonomous driving applications to fulfil complex tasks such as object recognition and image classification. Ensuring the reliability of CNNs is thus becoming an urgent requirement since they constantly behave in human environments. A common and recent trend is to replace the full-precision CNNs to make way for more optimized models exploiting approximation paradigms such as reduced bit-width data type. If from one hand this is poised to become a sound solution for reducing the memory footprint as well as the computing requirements, it may negatively affect the CNNs resilience. The intent of this work is to assess the reliability of a CNN-based system when reduced bit-widths are used for the network parameters (i.e., synaptic weights). The approach evaluates the impact of permanent faults in CNNs by adopting several bit-width schemes and data types, i.e., floating-point and fixed-point. This determines the trade-off between the CNN accuracy and the bits required to represent network weights. The characterization is performed through a fault injection environment built on the darknet open source framework. Experimental results show the effects of permanent fault injections on the weights of LeNet-5 CNN.

Today, the concept of approximation in computing is becoming more and more a “hot topic” to investigate how computing systems can be more energy efficient, faster, and less complex. Intuitively, instead of performing exact computations and, consequently, requiring a high amount of resources, Approximate Computing aims at selectively relaxing the specifications, trading accuracy off for efficiency. While Approximate Computing gives several promises when looking at systems' performance, energy efficiency and complexity, it poses significant challenges regarding the design, the verification, the test and the in-field reliability of Approximate Computing systems. This tutorial paper covers these aspects leveraging the experience of the authors in the field to present state-of-the-art solutions to apply during the different development phases of an Approximate Computing system.

In this paper, we propose a new framework for cell-aware defect diagnosis of customer returns based on supervised learning. The proposed method comprehensively deals with static and dynamic defects that may occur in real circuits. A Naive Bayes classifier is used to precisely identify defect candidates. Results obtained on benchmark circuits, and comparison with a commercial cell-aware diagnosis tool, demonstrate the efficiency of the proposed approach in terms of accuracy and resolution.

Diagnosis is crucial in order to establish the root cause of observed failures in Systems-on-Chip (SoC). In this paper, we present a new framework based on supervised learning for cell-aware defect diagnosis of customer returns. By using a Naive Bayes classifier to accurately identify defect candidates, the proposed flow indistinctly deals with static and dynamic defects that may occur in actual circuits. Results achieved on benchmark circuits, as well as comparison with a commercial cell-aware diagnosis tool, show the effectiveness of the proposed framework in terms of accuracy and resolution. Moreover, the proposed flow has been experimented and validated on industrial circuits (two test chips and one customer return from STMicroelectronics), thus corroborating the results achieved on benchmark circuits.

Approximate computing (AxC) is increasingly emerging as a new design paradigm to produce more efficient computation systems by judiciously reducing the computation quality. In particular, AxC has been successfully applied to integrated circuits (ICs), in the last years. Hence, concerning the test of such new class of ICs, namely approximate ICs (AxICs), new challenges-as well as new opportunities-have emerged. In this survey, we provide a thorough analysis of issues related to test procedures for AxICs and review the state-of-the-art techniques to deal with them. We resort to an illustrative example having the twofold aim of: 1) guiding the reader through the AxIC testing challenges and 2) illustrating the existing solutions to correctly overcome them, while suitably taking advantage of opportunities coming from approximation. We analyze experimentally the most recent testing techniques for AxICs and highlight their mature aspects, as well as their shortcomings. Experimental outcomes show that the testing process for AxIC is not completely mature. Indeed, only under specific conditions existing testing procedures achieve good results.

Functional Safety standards like ISO 26262 require a detailed analysis of the dependability of components subjected to perturbations. Radiation testing or even much more abstract RTL fault injection campaigns are costly and complex to set up especially for SoCs and Cyber Physical Systems (CPSs) comprising intertwined hardware and software. Moreover, some approaches are only applicable at the very end of the development cycle, making potential iterations difficult when market pressure and cost reduction are paramount. In this tutorial, we present a summary of classical state-of-the-art approaches, then alternative approaches for the dependability analysis that can give an early yet accurate estimation of the safety or security characteristics of HW-SW systems. Designers can rely on these tools to identify issues in their design to be addressed by protection mechanisms, ensuring that system dependability constraints are met with limited risk when subjected later to usual fault injections and to e.g., radiation testing or laser attacks for certification.

In critical (e.g. automotive) applications, Systems-on-Chip (SoC) failures that occurred during mission mode (in the field) are the most critical since they may lead to catastrophic effects. In this context, diagnosis is crucial in order to establish the root cause of observed failures with the best accuracy. With the advent of very deep submicron technologies (i.e. 7 nm), achieving such level of accuracy will become more and more difficult with today's intra-cell diagnosis tools based on effect-cause or cause-effect paradigms. This will compromise the success of subsequent Physical Failure Analysis (PFA) done on defective SoCs. Machine Learning (ML) is now used in numerous classification problems where the knowledge on some data can be used to classify a new instance of such data. In particular, several ML-based solutions exist to address volume diagnosis for yield improvement. These learning-guided diagnosis approaches start from an existing set of defect candidates and try to minimize this set (eliminate bad candidates) owing to the use of ML tools and numerous data collected during production test (e.g. thousands of failed chips with candidates correctly labeled). Although efficient in volume diagnosis, these approaches cannot be used to identify the root cause of failures in customer returns, since only one failed chip is investigated in this case, with no information about the defective behavior of some other similar chips used in the same conditions (environment, workload, etc.). In this paper, we propose a new learning-guided approach for diagnosis of mission mode failures in customer returns. The proposed approach directly produces a minimum set of good candidates derived from the application of the learning-guided intra-cell diagnosis flow. Results obtained on a set of benchmark circuits, and comparison with a commercial intra-cell diagnosis tool, show the feasibility, effectiveness and accuracy of the proposed approach.

The paper is a contribution to the 50th anniversary celebration of the International Test Conference (ITC) and its Global Test Forum (GTF), which honors the geographic breadth of the test community and highlights the global reach of ITC during the past 50 years. It covers the past, present, and future of the International Symposium on Design and Diagnostics of Electronic Circuits and Systems (DDECS), a symposium which belongs to prominent test technology related events initiated and supported by the ITC.

Approximate Computing (AxC) is today one of the hottest topics related to circuit design and optimization. Thanks to this new computing paradigm, designers are able to reduce area, power consumption, and even production costs in the case the target application can accept a given degree of inaccuracy in the final computations. This paper investigates the impact of accepting a faulty circuit as an approximate one in order to increase the system reliability. In particular, it considers two AxC techniques: functional approximation and precision reduction. The main goal of the paper is to determine the trade-off between the degree of the approximation and the reliability to give the system an additional chance to continue working, even if it does not work exactly as originally designed. The experimental results were gathered resorting to a set of 8-bit adders that includes some precise ones used in commercial applications, as well as some approximate ones. Additionally, the actual effects of the faulty circuits were evaluated at the application level: a video encoding system called NOVA was used, assessing the validity of the proposal.

Cross-layer reliability is becoming the preferred solution when reliability is a concern in the design of a microprocessor-based system. Nevertheless, deciding how to distribute the error management across the different layers of the system is a very complex task that requires the support of dedicated frameworks for cross-layer reliability analysis. This paper proposes SyRA, a system-level cross-layer early reliability analysis framework for radiation induced soft errors in memory arrays of microprocessor-based systems. The framework exploits a multi-level hybrid Bayesian model to describe the target system and takes advantage of Bayesian inference to estimate different reliability metrics. SyRA implements several mechanisms and features to deal with the complexity of realistic models and implements a complete tool-chain that scales efficiently with the complexity of the system. The simulation time is significantly lower than micro-architecture level or RTL fault-injection experiments with an accuracy high enough to take effective design decisions. To demonstrate the capability of SyRA, we analyzed the reliability of a set of microprocessor-based systems characterized by different microprocessor architectures (i.e., Intel x86, ARM Cortex-A15, ARM Cortex-A9) running both the Linux operating system or bare metal in the presence of single bit upsets caused by radiation induced soft errors. Each system under analysis executes different software workloads both from benchmark suites and from real applications.

Memristor device represents one of the most relevant technologies to deal with CMOS technological issues. In the scientific literature, a relevant amount of works have discussed the memristor device, with a particular emphasis on memristor-based crossbar architectures. However, while the synthesis of combinational logic circuits is widely discussed, the same cannot be said for sequential logic circuits. In this work, we propose a new approach for synthesizing sequential circuits based on memristor crossbar, by enhancing an existing architecture. This approach only exploits memristors within the crossbar for implementing the state feedback mechanism, with the aim of advancing the integration process of memristor-based circuits. Moreover, to provide an automated synthesis process of memristor-based sequential circuits, we extend a pre-existing automated synthesis framework so it can be integrated with widely used tools and formats as register-transfer level (RTL) or Berkeley Logic Interchange Format (BLIF) files. We performed several experiments on publicly available benchmarks in order to compare the proposed architecture against its predecessor in terms of circuit integration and efficiency. Obtained results highlight acceptable overheads (up to a maximum of 24%) compared with the opportunity of integration offered by the proposed architecture.

Approximate Computing (AxC) emerges more and more as a new paradigm for the design of energy-efficient Integrated Circuits (ICs) at the cost of accuracy reduction. The latter has to be modeled and quantified by means of Error Metrics. From the testing point of view, AxC Integrated Circuits offer an opportunity. Instead of testing for all manufacturing defects, the goal is to test only for those that will lead to an error considered as not acceptable by the adopted Error Metrics. The main advantages are the test cost reduction, since the number of required test vectors will be reduced, and the yield improvement. We developed three approaches for generating test vectors targeting AxC Integrated Circuits. This paper aims at comparing these approaches on a public benchmark suite.

Approximate Computing (AxC) paradigm aims at designing energy-efficient systems, saving computational resources, and presenting better execution times. AxC aims to selectively violate the specifications, trading accuracy off for efficiency. It has been demonstrated in the literature the effectiveness of imprecise computation for both software and hardware components implementing inexact algorithms, showing an inherent resiliency to errors. On the other hand, the hidden cost of AxC is the reduction on the inherent resiliency to errors of an application. This paper aims at analyzing the impact of AxC on the reliability.

Approximate Computing (AxC) trades off between the level of accuracy required by the user and the actual precision provided by the computing system to achieve several optimizations such as performance improvement, energy and area reduction etc. Several AxCtechniques have been proposed so far in the literature. They work at different abstraction levels and propose both hardware and software implementations. The common issue of all existing approaches is the lack of a methodology to estimate the impact of a given AxC technique on the application-level accuracy. In this paper we propose a probabilistic approach to predict the relation between component-level functional approximation and application-level accuracy. Experimental results on a set of benchmark applications show that the proposed approach is able to estimate the approximation error with good accuracy and very low computation time.

In today's electronic designs, more and more memories are embedded in a single chip. The latest technologies make them denser and thus defects due to the manufacturing process are more prone to occur not only in the array but also in the periphery of the memory. A fast and accurate localization of such defects has become much more difficult with traditional diagnosis approaches that do not allow a fast-enough yield learning and improvement. This paper describes a new and automated intra-cell diagnosis flow for SRAMs to precisely determine the root cause of observed failures during test. Based on the electrical and topological fault signatures obtained through traditional methods, each potential fault on the identified active nets is automatically simulated to retrieve the best defect candidates to precisely guide the Failure Analysis phase. The proposed intra-cell diagnosis flow has been experimented on simulated test cases as well as silicon (industrial) test cases. The results obtained demonstrate the effectiveness of the diagnosis flow in terms of low number of defect candidates. Moreover, for the silicon test cases, these diagnosis results match with the ongoing Failure Analysis reports.

Two AxC techniques have been successfully applied to hardware components. The first one is the functional approximation [1]that modifies the circuit structure replacing the original function F with the function G. G implementation leads to area/energy reduction at the cost of reduced accuracy, meaning that some errors can be observed at the outputs of G. The observed errors are a variation between the output values of F (precise) and G (approximate). The variation is the accuracy loss measured by means of quality metric(s) [1]. The second AxC technique is the over-scaling based approximation. Basically, the HW component is forced to work outside its specified operating conditions [1]. The classical example is the reduction of the supply voltage under the minimum value.

The unceasing shrinking process of CMOS technology is leading to its physical limits, impacting several aspects, such as performances, power consumption and many others. Alternative solutions are under investigation in order to overcome CMOS limitations. Among them, the memristor is one of promising technologies. Several works have been proposed so far, describing how to synthesize boolean logic functions on memristors-based crossbar architecture. However, depending on the synthesis parameters, different architectures can be obtained. Design Space Exploration (DSE) is therefore mandatory to help and guide the designer in order to select the best crossbar configuration. In this paper, we present a formal DSE approach. The main advantage is that it does not require any simulation and thus it avoids any runtime overheads. Preliminary results show the huge gain in runtime compared to simulation-based DSE.

In the recent years, Approximate Computing (AC) has emerged as a new paradigm for energy efficient design of Integrated Circuits (ICs). AC is based on the intuitive observation that, while performing exact computation requires a high amount of resources, allowing a selective approximation or an occasional relaxation of the specification can provide significant gains in energy efficiency. This work starts from the consideration that AC-based systems can intrinsically accept the presence of faulty hardware (i.e., hardware that can produce errors). In other words, an AC-based system does not need to be built using defect-free ICs. Under this assumption, we can relax test and reliability constraints of the manufactured ICs. One of the ways to achieve this goal is to test only for a subset of faults instead of targeting all possible faults. In this way, we can reduce the manufacturing cost since we reduce the number test patterns and thus the test time. We call this approach Approximate Test (AT). The main advantage is the fact that we do not need a prior knowledge of the application. Therefore, the proposed approach can be applied to any kind of IC, reducing the test time and increasing the yield. In this work, we aim at validating the proposed AT by comparing it with a functional approach. We present preliminary results on some simple case studies. The main goal is to show that by letting some faults undetected we can save test time without having a huge impact on the application quality.

Reliability evaluation is a critical task in computing systems. From one side, the results must be accurate enough not to under-or over-estimate the overall system reliability (thus either resulting in a non-reliable system, or a system for which too expensive solutions have been adopted). On the other side, the time required for the analysis should be kept at the minimum. This paper presents some new advances in the reliability assessment of computing systems, by showing techniques targeting both hardware and software levels, and the combination of both.

Developing new methods to evaluate the software reliability in an early design stage of the system can save the design costs and efforts, and will positively impact the product time-to-market. In this paper, we propose a novel fault injection technique to evaluate the reliability of a computing system running a software at early design stage where the hardware architecture is not completely defined yet. The proposed approach efficiently operates on the original source code of the software in order to inject transient faults in the data or the instructions. To be accurate and to achieve a better characterization of the system, we simulate faults occurring in the system memory units such as the data cache and the RAM by developing a system emulator. To validate our approach, we compare the simulation results to those obtained with an FPGA-based fault injector. The similarity of the results proves the accuracy of our approach to evaluate system reliability with a gain in the execution time and without requiring a fully defined hardware system.

Power management techniques are applied at high abstraction levels to reduce chip power consumption. Accurate and efficient power models are needed as early as possible in the design flow to ensure that correct saving decisions are taken. However, accuracy at those levels cannot be ensured, as there is not exact knowledge of the circuit structure. Then, power models based on estimation techniques at lower abstraction levels are desired. In this work, we propose a hybrid power modeling approach based on an effective library characterization methodology and an efficient power estimation flow to accurately assess gate-level power consumption. The main idea is to enhance the high-level power models by providing realistic information of the physical design. We perform experiments on ISCAS'85 benchmark circuits synthesized with a 28nm FDSOI technology. To prove the validity of our approach, we compare our results with SPECTRE simulations and show that we can achieve a 144X speedup on the runtime with a transistor-like accuracy.

Functional test guarantees that the circuit is tested under normal conditions, thus avoiding any over-as well as under-test. This work is based on the use of Software-Based-Self-Test that allows a special application of functional test to the processor-based systems. This strategy applies the so-called functional test programs that are executed by the processor to guarantee a given fault coverage. The main goal of this paper is to investigate the static test compaction of a given set of functional test programs. The investigation aims at understanding and determining how to select the best functional test program candidates to obtain the smallest set having the best fault coverage. Results carried out on two different microprocessors show that a 49% reduction in test length and a 28.7% reduction in test application time can be achieved.

Selecting the ideal trade-off between reliability and cost associated with a fault tolerant architecture generally involves an extensive design space exploration. Employing state-of-the-art susceptibility estimation methods makes it unscalable with design complexity. In this paper we introduce a low-cost susceptibility analysis methodology that helps identifying the most vulnerable circuit elements for hardening with less computational effort and orders of magnitude faster. Our experimental results show that the methodology offers huge gain in terms of computational effort (2,500× faster) in comparison with a fault-injection based method and produces results within acceptable degree of accuracy.

Reliability evaluation is a high costly process that is mainly carried out through fault injection or by means of analytical techniques. While the analytical techniques are fast but inaccurate, the fault injection is more accurate but extremely time consuming. This paper presents an hybrid approach combining analytical and fault injection techniques in order to evaluate the reliability of a computing system, by considering errors that affect both the data and the instruction cache. Compared to existing techniques, instead of targeting the hardware model of the cache (e.g., VHDL description), we only consider the running application (i.e., the software layer). The proposed approach is based on the Low-Level Virtual Machine (LLVM) framework coupled with a cache emulator. As input, the tool requires the application source code, the cache size and policy, and the target microprocessor instruction set. The main advantage of the proposed approach is the achieved speed up quantified in magnitude orders compared to existing fault injection techniques. For the validation, we compare the simulation results to those obtained with an FPGA-based fault injector. The similarity of the results proves the accuracy of the approach.

System reliability estimation during early design phases facilitates informed decisions for the integration of effective protection mechanisms against different classes of hardware faults. When not all system abstraction layers (technology, circuit, microarchitecture, software) are factored in such an estimation model, the delivered reliability reports must be excessively pessimistic and thus lead to unacceptably expensive, over-designed systems. We propose a scalable, cross-layer methodology and supporting suite of tools for accurate but fast estimations of computing systems reliability. The backbone of the methodology is a component-based Bayesian model, which effectively calculates system reliability based on the masking probabilities of individual hardware and software components considering their complex interactions. Our detailed experimental evaluation for different technologies, microarchitectures, and benchmarks demonstrates that the proposed model delivers very accurate reliability estimations (FIT rates) compared to statistically significant but slow fault injection campaigns at the microarchitecture level.

Burn-In (BI) test is usually applied in manufacturing process to screen out chip early life failures, especially for safety critical applications. Unfortunately, this test method has elevated costs for companies. In recent days, Faster-than-at-Speed-Test (FAST) has become a useful technique to discover small delay defects. At the same time, overclocking methods to enhance system performances have been studied, which focus on temperature management to preserve system functionalities. In this paper, a FAST technique is approached with the aim of intentionally provoking a thermal overheating in the microprocessor by mean of the execution of functional test programs, partly regardless of system behavior preservation. The goal is to introduce an internal stress stronger than current procedures used during BI in order to speed up early detection of latent faults. The method illustrates how to avoid blocking configurations due to timing constraints violation and leads to a significant increase of the switching activity. Experimental results on a MIPS architecture show that, by using the described technique, the processor is not falling into an unpredictable state even at frequencies up to about 20 times higher than the nominal one and the switching activity is increasing up to 300% per nanoseconds.

Nowadays, power consumption is the one key factor that hinders System-on-Chip (SoC) performance. In order to reduce the power consumption, accurate and efficient power models have to be introduced early in the design flow, when most of the optimization potential is obtained. However, early accuracy cannot be ensured because of the lack of precise knowledge of the circuit structure. Current SoC design paradigm relies on Intellectual Property (IP) reuse, and low-level information about circuit components and structure is usually available. Thus, if we use this information and develop an estimation methodology that fits IP power modeling needs, the estimation accuracy at system level will be improved. This paper presents a Hybrid Power Estimation Technique (HPET). It is based on an effective library characterization methodology and an efficient hybrid power modeling approach to accurately and quickly assess gate-level power consumption. The aim is to give valuable and accurate physical information to design teams so they can ensure that the correct optimization techniques are implemented. Our approach can be used to compute both realistic instantaneous power and average power on a single simulation time. We performed experiments on different benchmark circuits synthesized using the 28nm FDSOI technology. To validate the proposed technique, we correlated our results with SPECTRE and PrimeTime-PX simulations. Our results showed that we can achieve up to 144× speedup on the simulation runtime with a mean error of about 6% and 13% for the instantaneous and average power components respectively.

The shrinking process of CMOS technology is reaching its physical limits, thus impacting on several aspects, such as performances, power consumption and many others. Alternative solutions are under investigation in order to overcome CMOS limitations. Among them, the memristor is one of the promising technologies. Several works have been proposed so far, describing how to implement boolean logic functions employing memristors in a crossbar architecture. In this paper, we propose a tool able to automatically map any boolean function to a memristor based crossbar implementation. The proposed tool helps to perform a design space exploration to identify the best implementation w.r.t. performances and area overhead.

Thermal phenomena occurring along test execution at the final stages of the manufacturing flow are considered as a significant issue for several reasons, including dramatic effects like circuit damage that is leading to yield loss. This paper tries to redeem those bad guys in order to exploit them to improve the test quality, reducing the overall test cost without affecting the yield.

Developing new methods to evaluate the software reliability in an early design stage of the system can save the design costs and efforts, and will positively impact product time-to-market. This paper introduces a new approach to evaluate, at early design phase, the reliability of a computing system running a software. The approach can be used when the hardware architecture is not completely defined yet. In order to be independent of the hardware architecture and at the same time accurate, we propose to use the Low-Level Virtual Machine (LLVM) framework. In addition, to reduce the reliability evaluation time, our approach consists in analyzing the variable lifetimes to compute the probability of masked faults. Finally, to achieve a better characterization we propose to consider also the presence of caches and register files. For this purpose, a cache emulator as well as a register file emulator are developed. Simulations run with our approach produce very similar results to those run with a hardware-based fault injector. This proves the accuracy of our approach to evaluate system reliability with a gain in the simulation time and without requiring a hardware platform.

Power-gating techniques have been adopted so far to reduce the static power consumption of an Integrated Circuit (IC). Power-gating is usually implemented by means of several power switches. Manufacturing defects affecting power switches can lead to increase the actual static power consumption and, in the worst case, they can completely isolate a functional block in the IC. Thus, efficient test and diagnosis solutions are needed. In this paper we propose a Design-for-Diagnosis architecture for Power Switches. The proposed approach has been validated through SPICE simulations on ITC'99 benchmark circuits as well as on industrial test case.

Increasing vulnerability of transistors and interconnects due to CMOS technology scaling is continuously challenging the reliability of future electronic circuits and systems. Lifetime reliability is gaining attention over performance as a design factor even for lower-end commodity applications. In this paper we propose an effective hybrid fault-tolerant architecture able to deal with permanent and transient faults in combinational parts of pipelined cores. The principle consists in triplicating the combinational logic parts but, unlike TMR, only two copies run in parallel while the third one remains in standby until an error is detected. We have implemented this approach on a MIPS microprocessor as case study. Experiments show that our approach is comparable to TMR in terms of area with a notable power saving and offers a full protection against transient and permanent faults.

Fault-tolerant architectures have been widely used in industry to prevent circuit reliability from becoming a bottleneck for the development of robust high-performance and low-power systems. One such solution is a Hybrid Fault-Tolerant Architecture that offers benefits such as low power and lifetime reliability improvement. However, it has been identified that there is room of improvement in efficiency. Thus, in this paper we present design space exploration and optimization of the Hybrid Fault-Tolerant Architecture. The study involves application of four design variants to some ITC benchmark circuits as case study. Experimental results compare the initial and optimized designs and show that the proposed optimizations offer around 65% reduction in terms of area, about 55% power saving and 87% less performance overhead as compared to the initial design without any penalty of the fault tolerance capability.

In order to improve performance and reduce cost, multi-processor system on chip (MPSoC) is increasingly becoming attractive. At the same time, 3D integration emerges as a promising technology for high density integration. 3D homogeneous MPSoCs combine the benefits of both. However, high current demand and large on-chip switching activity variations introduce severe power supply noises (PSN) for 3D MPSoCs, which can increase critical path delay, and degrade chip performance and reliability. Meanwhile, thermal gradient should also be considered for 3D MPSoCs to avoid hot spots. In the paper, we investigate the PSN effects of different workloads and propose an effective PSN estimation method. Then, a heuristic workload assignment algorithm is proposed to suppress PSN under the given thermal constraint. The experimental results show that PSNs can be reduced significantly compared with thermal-balanced workload assignment scheme, and the system performance can be improved as well.

In this paper, we investigate the generation of diagnostic test vectors targeting the intra-cell defects. Experimental results carried out on an industrial circuit show that we actually increase the diagnosis resolution by adding few more diagnostic test patterns.

Physical Design (PD) issues are becoming a major challenge with technology scaling in integrated circuits. Multi-aggressor crosstalk, power supply noise and ground bounce are some of the PD issues that cause considerable path delay variations. Therefore, these PD issues need to be considered during path delay testing to ensure better delay defect coverage. In this paper, we first show that the path delay Automatic Test Pattern Generation (ATPG) test methods are incapable of generating an input pattern that can capture worst-case path delay in circuits. We, then present our Physical Design Aware Pattern Generation (PDAPG) method to generate an input test pattern that can capture worst-case path delay in the presence of PD issues. We propose a backtrace X-filling approach to identify the relevant X-bits causing worst-case path delay. Simulations performed on ITC'99 benchmark circuits show that our PDAPG method is capable of providing high quality input test patterns in comparison with conventional path delay ATPG test methods.

With the continuous scaling down of the transistor size, the so-called intra-cell defects are more and more frequent. In this paper we propose a defect grading tool able to evaluate the efficiency of the applied test set. The test set efficiency is quantified w.r.t. the intra-cell defect coverage and the intra-cell diagnosis resolution.

Through-Silicon-Vias (TSVs) are the key enablers of 3D integration technology. Therefore, the reliability of 3D-ICs rely on the quality of TSV testing. TSVs are prone to defects that may introduce small delay variations that can cause quality and reliability issues. Moreover, physical and electrical conditions, such as TSV dimensions, coupling and IR-drop, may affect path delay variations and consequently affect the detectability of small delay faults (SDF) induced by defective TSVs. In this work, we study the test quality and pattern effectiveness for SDF induced by TSVs. We quantity test quality using statistical delay quality level (SDQL) metric and test patterns are generated with commercial ATPG tools.

Power-gating techniques have been adopted so far to reduce the static power consumption of an Integrated Circuit (IC). Power gating is usually implemented by means of several power switches. Manufacturing defects affecting power switches can lead to increase the actual static power consumption and, in the worst case they can completely isolate a functional block of the IC. In this paper we present a novel Design for Test & Diagnosis to increase the test quality and diagnosis accuracy of power switches. The proposed approach has been validated through SPICE simulations on ITC'99 benchmark circuits.

CMOS technology trends at one side open up some opportunities like making small and power efficient devices available, which in turn allow to put more functionality into a single chip. However, on the other side it poses some challenges like making devices vulnerable to hard and soft errors. In this paper we propose an efficient fault-tolerant architecture able to deal with permanent and transient faults in combinational parts of pipeline structures. The principle consists in triplicating the combinational logic parts but, unlike TMR, only two copies are running in parallel while the third one remains in standby until an error is detected. We implement this approach on a MIPS microprocessor as case study to make it resilient against transient and permanent faults.

The use of Triple Modular Redundancy (TMR) was historically introduced long time ago for improving reliability of computer systems [1]. Recently, the advances in miniaturizing of CMOS devices made digital circuits more and more unreliable. The current trend goes towards the Internet of Things and the cloud computing, where small devices have high requirements in terms of reduced power consumption and increased reliability [2]. Classical TMR solutions allow for high reliability but they cannot satisfy low-power require-ments, since they consume about three times more than the equivalent single device. However, the type of applications that are implemented in the new cloud scenario do not require high reliability all the time, but it can be assumed that some computations are more important, and thus require to be executed by a reliable hardware, while other computations are less important, and thus they can tolerate failures [3].

In this paper we propose a novel Power Supply Noise (PSN) sensor. It is based on timing uncertainty measure. Compared to state of the art it allows to measure the PSN events in more accurate way. The proposed sensor is actually under validation and patent reviewing process.

3D-IC test becomes a challenge with the increasing number of TSVs and demands for effective 3D aware test techniques. In this work, we propose a timing aware model to capture delay variations on a path due to resistive open TSVs. The key idea is to analytically model delay and apply our correlation-based resistive open TSV detection method to attain path delay fault coverage. We propose two methods to investigate timing variation introduced by resistive open TSVs in a critical path delay with multiple TSVs. Method I computes the correlation of multiple TSVs in a path to overall path delay to determine if TSVs are the source of the introduced delay. Method II pinpoints which TSV is faulty by computing the delay fault coverage of each TSV in a path with multiple TSVs. Our results indicate the accuracy of our proposed method and promotes early identification of resistive open defects TSVs.

Magnetic random access memory (MRAM) is an emerging technology with potential to become the universal on-chip memory. Among existing MRAM technologies, thermally assisted switching (TAS)-MRAM technology offers several advantages compared with other technologies: selectivity, single magnetic field, and high-integration density. In this paper, we analyze the impact of resistive-open defects on TAS-MRAM behavior. Electrical simulations were performed on a hypothetical 16 word TAS-MRAM architecture enabling any combination of read and write operations. Results show that read and write sequences may be affected by resistive-open defects that may induce single and double-cell faulty behaviors. As a next step, we will exploit the analyses results to guide the test phase by providing effective test algorithms targeting faults related to actual defects affecting TAS-MRAM architectures.

Design of power delivery network (PDN) is a constrained optimization problem. An ideal PDN must limit voltage drop that results from switching circuits' transients, satisfy current density constraints that arise from electromigration limits, yet use only minimal metal resources so that design density targets can be met. It should also provide an efficient thermal conduit to address heat flux. Furthermore, an ideal PDN should be a regular structure to facilitate design productivity and manufacturability, yet be resilient to address varying power demands across its distribution area. In 3-D ICs, these problems are further constrained by the need to minimize through-silicon via (TSV) area and bridge power lines of different dimensions across tiers, while addressing varying power demands in lateral and vertical directions. In this paper, we propose an unconventional power grid optimization solution that allows us to resize each tier individually by applying tier-specific constraints and yet be optimal in a multitier network, where each tier is locally resized while globally constrained. Tier-specific constraints are derived from electrical and thermal targets of 3-D PDNs. Two resizing algorithms are presented that optimize 3-D PDNs standalone or 3-D PDNs together with TSVs. We demonstrate these solutions on a three-tier setup where significant area savings can be achieved.

SRAM testing is becoming more and more challenging due to issues caused by continuous device scaling. Fabricated SRAMs are submitted to random and systematic process variability, which strongly affect the cell's behavior and also the ability of test algorithms to detect faults. Traditionally, bias conditions have been used to improve the behavior of the SRAM under process variations by applying body bias to compensate for the effect of variability. Based on the same principle, bias conditions also affect the cell's behavior when resistive-opens are present, hence affecting test's defect coverage capability. Both body- and source-bias conditions are analyzed in this paper to find the way to improve defect detect ability in the SRAM cell. Source-biasing has been proven to be the more effective of the two, leading to more than 3X improvement of the defect detected value. Also, by adapting the source-bias conditions to process parameter values, over- and under-testing of the SRAM can be avoided.

Low-power SRAMs embed mechanisms for reducing static power consumption. When the SRAM is not accessed during a long period, it switches into an intermediate low-power mode. In this mode, a voltage regulator is used to reduce the voltage supplied to the core-cells as low as possible without data loss. Thus, faulty-free behavior of the voltage regulator is crucial for ensuring data retention in core-cells when the SRAM is in low-power mode. This paper investigates the root cause of data retention faults due to voltage regulator malfunctions. This analysis is done under realistic conditions (i.e., industrial core-cells affected by process variations). Based on this analysis, we propose an efficient test flow for detecting data retention faults in low-power SRAMs.

Three-dimensional stacking technology promises to solve the interconnect bottleneck problem by using Through-Silicon-Vias (TSVs) to vertically connect circuit layers. However, manufacturing steps may lead to partly broken or incompletely filled TSVs that may degrade the performance and reduce the useful lifetime of a 3D IC. Due to combinations of physical factors such as switching activity, supply noise and crosstalk, path delays can experience speed-up or slow-down that could let the effect of resistive open TSV go undetected by conventional test methods. In this work, we present a metric based on probabilistic analysis to detect delay defects induced by resistive opens that occur on signal line TSVs. Our experimental result will show the accuracy of the proposed metric.

Functional operations of a Static Random Access Memory (SRAM) are strongly affected by random variability in core-cell transistors and by the variability-induced threshold voltage mismatch between the transistors of the Input-Output (IO) circuitry (especially Sense Amplifiers). This variability also affects the faulty behavior of the SRAM array. This paper is focused on the analysis of static and dynamic faults due to resistive-open defects in the SRAM core-cell, taking into account the effects of random process variability in core-cells and IO circuitry. Statistical analyses have been performed to evaluate the SRAM failure probabilities accounting for defects at each possible location. The results show that random process variability in the SRAM core-cell and IO circuitry have an important effect on the behavior of an SRAM array and also on the defect coverage of various commonly-used test sequences. It is shown that under variability, the minimum defect size detected with maximum probability is more than 2X larger than the minimum size detected in nominal conditions, thus leaving a large range of defects undetected. Several stress conditions during test have been evaluated to assess their capability to increase the defect coverage under random process variability.

In current technologies, the robustness of Static Random Access Memories (SRAM) has to be investigated under any possible source of disturbance. In this paper, we evaluate the reliability of an SRAM cell exposed to atmospheric neutron radiation, affected by random threshold voltage variation and under different operation conditions (supply voltage, process corner and temperature). The SRAM cell's Soft Error Rate (SER) at simulation level is estimated using accurate models of atmospheric neutron induced currents. The study shows that in extreme operation conditions and under random process variability, the SER of an SRAM can reach values up to 3X larger than the nominal value, or down to 2X smaller than the nominal value. This large SER range confirms the importance of our study and justifies the need for further evaluation of circuits under radiation at the simulation level.

Read and write assist techniques are widely adopted to allow voltage scaling in low-power SRAMs. In particular, this paper analyzes two assist techniques: word line level reduction and negative bit line boost. The analyzed assist techniques improve read stability and write margin of core-cells when the SRAM operates at a lowered supply voltage. In this work, we investigate the impact of such assist techniques on the faulty behavior of low-power SRAMs. This analysis is based on extensive injection of resistive-open and resistive-bridging defects in core-cells of a commercial low-power SRAM. Our study determines the most stressful configuration of assist circuits to detect each faulty behavior induced by injected defects. We show that, by applying most stressful configurations of assist circuits during test phase, defect coverage can be increased up to 89% w.r.t. test solutions that do not exploit assist circuits. Based on this analysis, we present an efficient test solution that exploits the configuration of assist circuits as a parameter to maximize the detection of studied defects, while reducing time complexity up to 73% w.r.t. test flows using state-of-the-art test algorithms.

Voltage regulation systems offer an efficient mechanism for reducing static power consumption of SRAMs. When the SRAM is not accessed for a long period, it switches into an intermediate low-power mode. In this mode, a voltage regulator is used to reduce the voltage supplied to the core-cell array as low as possible without data loss. Therefore, reliable operation of such device must be ensured by using adequate test techniques. In this work, we propose low area overhead built-in self-test (BIST) and built-in self-repair (BISR) schemes that can be embedded on the SRAM to automatically test and repair the voltage regulator. Simulation results prove the effectiveness of the proposed technique for detecting, diagnosing and repairing voltage regulators of low-power SRAMs.

Power supply noise and ground bounce can cause considerable path delay variations. Capturing the worst case power supply noise at a gate level is not a sufficient indicator for measuring the worst case path delay. Furthermore, path delay variations depend on multiple parameters such as input stimuli, cell placement, switching frequency, and available decoupling capacitors. All these variables obscure the rapport between supply noise and path delay and make the selection of stimuli for worst case path delay a difficult task during test pattern generation. In this paper, we utilize power supply noise and ground bounce distribution along with physical design data to generate test patterns for capturing worst case path delay. We propose accurate close-form mathematical models for capturing the effect of power supply noise and ground bounce on path delay. These models are based on modified nodal analysis formulation of power and ground networks, where current waveforms are obtained from levelized simulation and cell library characterization. The proposed test pattern generation flow is a simulated-annealing-based iterative process, which utilizes mathematical models for capturing the impact of supply noise on path delay for a given input pattern. We perform experiments on ITC'99 benchmarks and show that path delay variation can be considerable if test patterns are not properly selected.

3-D integration presents a path to higher performance, greater density, increased functionality and heterogeneous technology implementation. However, 3-D integration introduces many challenges for power and thermal integrity due to large switching currents, longer power delivery paths, and increased parasitics compared to 2-D integration. In this work, we provide an in-depth study of power and thermal issues while incorporating the physical design characteristics unique to 3-D integration. We provide a qualitative perspective of the power and thermal dissipation issues in 3-D and study the impact of Through Silicon Vias (TSVs) size for their mitigation. We investigate and discuss the design implications of power and thermal issues in the presence of decoupling capacitors, TSV/on-die/package parasitics, various resonance effects and power gating. Our study is based on a ten-tier system utilizing existing 3-D technology specifications. Based on detailed power distribution and heat dissipation models, we present a comprehensive analysis of TSV tapering for alleviating power and thermal integrity issues in 3-D ICs.

Magnetic Random Access Memory (MRAM) is an emerging memory technology. Among existing MRAM technologies, the Thermally Assisted Switching (TAS) MRAM technology offers several advantages such as selectivity, single magnetic field and high integration density. In this paper, we analyze resistive-bridge defects that may affect the TAS-MRAM architecture. Electrical simulations were performed on a hypothetical 16-words TAS-MRAM architecture enabling any sequences of read/write operations. Results show that both read and write operations may be affected by these defects. Especially, we demonstrate that resistive-bridge defects may have a local (single cell) or global (multiple cells) impact on the TAS-MRAM functioning. As these analysis results will be further used to develop effective test algorithms targeting faults related to actual resistive bridge-defects that may affect TAS-MRAM architecture.

High peak power consumption during test may lead to yield loss. On the other hand, reducing too much test power may lead to test escape. In order to overcome this problem, test power has to mimic the power consumed during functional mode, being as high as possible but not crossing the frontier of over-consumption. Measuring power consumption is a very time consuming activity, therefore many works in the literature focused on the indirect ways to provide power consumption estimation in a fast manner. In this paper we concentrate on a similar issue, concentrating our effort on devising a fast method for the identification and estimation of the peak power produced by test patterns. In particular we provide a detailed discussion on case studies related to peak power estimation of CPU cores when executing functional patterns, the proposed method uses the gate-level description of the CPU to identify a subset of time points over the entire test pattern that are showing the most significant peak power values. The proposed methodology has been validated on two case studies synthesized in a 65nm industrial technology.

Managing the power consumption of circuits and systems is challenging not only during functional operations but also during manufacturing test. In this paper, we first explain why it is important to control power consumption during test application. We will introduce the basic concepts and discuss issues arising from excessive power dissipation during test. Then, we explain how it is possible to control power consumption during test. We will provide an overview of existing structural and algorithmic solutions for power-aware testing, and we will show how low power circuits can be tested safely without affecting yield and reliability.

In this work, we present a new power supply noise sensor based on timing uncertainty measurements. The proposed sensor can detect power supply noise events in a more accurate way compared to the state of the art solutions. Experimental results validated the efficiency of the proposed approach.

Magnetic Random Access Memory (MRAM) is an emerging technology with the potential to become the universal on-chip memory. Among the existing MRAM technologies, the Thermally Assisted Switching (TAS) MRAM technology offers several advantages compared to the others technologies: selectivity, single magnetic field and integration density. As any other types of memory, TAS-MRAMs are prone to defects, so TAS-MRAM testing needs definitely to be investigated since only few papers can be found in the literature. In this paper we analyze the impact resistive-open defects on the heat current of a TAS-MRAM architecture. Electrical simulations were performed on a hypothetical 4×4 TAS-MRAM architecture enabling any read/write operations. Results show that W0 and/or W1 operations may be affected by the resistive-open defects. This study provides insights into the various types of TAS-MRAM defects and their behavior. As future work, we plan to utilize these analyses results to guide the test phase by providing effective test algorithm targeting fault related to actual defects that may affect TAS-MRAM architecture.

Thermally Assisted Switching Magnetic Random Access Memory (TAS-MRAM) is an emerging technology that offers several advantages compared to existing non-volatile memory technologies. In this paper we show how coupling faults induced by resistive-open defects impact the TAS-MRAM architecture. Results shows that read and write operations may be affected these defects and may induce single and double cell faulty behaviors.

Three-dimensional (3D) integration is a fast emerging technology that offers integration of high density, fast performance and heterogeneous circuits in a small footprint. Through-Silicon-Vias (TSVs) enable 3D integration by providing fast performance and short interconnects among tiers. However, they are also susceptible to defects that occur during manufacturing steps and cause crucial reliability issues. In this paper, we perform an analysis of resistive-open defects (ROD) on TSVs considering coupling effects (i.e. inductive and capacitive) and a wide frequency spectrum. Our experiments show that both substrate coupling and switching frequency can have a significant impact on weak open TSV behavior.

Summary form only given. Low-power SRAMs embed power gating mechanisms for reducing static power consumption. Power gating is applied in SRAMs using power switches for controlling the supply voltage applied to the various memory blocks (array, decoders, I/O logic, etc.). This paper provides a detailed analysis based on electrical simulations to describe the impacts of resistive-open defects on the power mode control logic, which generates control signals of power switches.

Electronic system reliability over soft errors is very critical as the transistor size shrinks. Many recent works have defined the device error rate under radiation for SRAMs in hold mode (static) and during operation (dynamic). This paper evaluates the impact of running test algorithms on SRAMs exposed to neutron radiation in order to define their stressing factor. The results that we show are based on experiments performed at the TSL facility in Uppsala, Sweden using a Quasi-Monoenergetic neutron beam. The evaluation of the test algorithms is based on the calculated device SEU cross section.

Low-power SRAMs embed power gating mechanisms for reducing static power consumption. Power gating is implemented through power switches for controlling the supply voltage applied to the various memory blocks (array, decoders, I/O logic, etc.). This way, one or more memory blocks can be disconnected from the power supply during a long period of inactivity, thus reducing static power consumption. This paper focuses on low-power SRAMs, and in particular, the power gating mechanisms of core-cells and peripheral circuitry. We provide a detailed analysis based on electrical simulations to characterize the impact of resistive-open defects on the power mode control logic. Based on this analysis, we introduce appropriate fault models that represent the observed faulty behaviors. Finally, we propose an efficient test solution targeting the set of identified fault models.

Memory design and test represent very important issues. Memories are designed to exploit the technology limits to reach the highest storage density and high-speed access. The main consequence is that memory devices are statistically more likely to be affected by manufacturing defects. The challenge of testing SRAM memories consists in providing realistic fault models and test solutions with minimal application time. Due to the complexity of the memory device, fault modeling is not trivial. Classical memory test solutions cover the so-called `static faults' (such as stuck-at, transition, and coupling faults) but are not sufficient to cover faults that have emerged in latest VDSM technologies and which are referred to as `dynamic faults'. This tutorial aims at introduce and guide to new test approaches developed so far for dealing with dynamic faults in the latest generation of SRAM memories.

Although CMOS technology scaling offers many advantages, it suffers from robustness problem caused by hard, soft and timing errors. The robustness of future CMOS technology nodes must be improved and the use of fault tolerant architectures is probably the most viable solution. In this context, Duplication/Comparison scheme is widely used for error detection. Traditionally, this scheme uses a static comparator structure that detects hard error. However, it is not effective for soft and timing errors detection due to the possible masking of glitches by the comparator itself. To solve this problem, we propose a pseudo-dynamic comparator architecture that combines a dynamic CMOS transition detector and a static comparator. Experimental results show that the proposed comparator detects not only hard errors but also small glitches related to soft and timing errors. Moreover, its dynamic characteristics allow reducing the power consumption while keeping an equivalent silicon area compared to a static comparator. This study is the first step towards a full fault tolerant approach targeting robustness improvement of CMOS logic circuits.

What is the probability that the execution state of a given microprocessor running a given application is correct, in a certain working environment with a given soft-error rate? Trying to answer this question using fault injection can be very expensive and time consuming. This paper proposes the baseline for a new methodology, based on microprocessor error probability profiling, that aims at estimating fault injection results without the need of a typical fault injection setup. The proposed methodology is based on two main ideas: a one-time fault-injection analysis of the microprocessor architecture to characterize the probability of successful execution of each of its instructions in presence of a soft-error, and a static and very fast analysis of the control and data flow of the target software application to compute its probability of success. The presented work goes beyond the dependability evaluation problem; it also has the potential to become the backbone for new tools able to help engineers to choose the best hardware and software architecture to structurally maximize the probability of a correct execution of the target software.

Launch-off-Capture (LOC) and Launch-off-Shift (LOS) are the two main test schemes for at-speed scan delay testing. In the literature, it has been shown that LOS has higher performance than LOC in terms of fault coverage and test length, but higher peak power consumption during the launch-to-capture cycle. Power reduction seems to be the key to really exploit LOS test scheme. However, it has been proven that reducing too much test power can lead to test escape due to under-test. In this context, this study proposes a smart X-filling framework able to adapt peak power consumption during the launch-to-capture cycle according to the functional power, i.e. the power consumption of the circuit in functional mode. Here, the main goal is to obtain a final test set with peak power consumption as close as possible to the functional power. Experimental results, carried out on the well-known ITC'99 benchmarks, prove the feasibility of the proposed approach.

It has become necessary to reduce power during LSI testing. Particularly, during at-speed testing, excessive power consumed during the Launch-To-Capture (LTC) cycle causes serious issues that may lead to the overkill of defect-free logic ICs. Many successful test generation approaches to reduce IR-drop and/or power supply noise during LTC for the launch-off capture (LOC) scheme have previously been proposed, and several of X-filling techniques have proven especially effective. With X-filling in the launch-off shift (LOS) scheme, however, adjacent-fill (which was originally proposed for shift-in power reduction) is used frequently. In this work, we propose a novel X-filling technique for the LOS scheme, called Adjacent-Probability-based X-Filling (AP-fill), which can reduce more LTC power than adjacent-fill. We incorporate AP-fill into a post-ATPG test modification flow consisting of test relaxation and X-filling in order to avoid the fault coverage loss and the test vector count inflation. Experimental results for larger ITC'99 circuits show that the proposed AP-fill technique can achieve a higher power reduction ratio than 0-fill, 1-fill, and adjacent-fill.

In this paper, a novel hybrid fault tolerant architecture for digital circuits is proposed in order to enable the use of future CMOS technology nodes. This architecture targets robustness, power consumption and yield at the same time, at area costs comparable to standard fault tolerance schemes. The architecture increases circuit robustness by tolerating both transient and permanent online faults. It consumes less power than the classical Triple Modular Redundancy (TMR) approach while utilizing comparable silicon area. It overcomes many permanent faults occurring throughout manufacturing while still tolerating soft errors introduced by particle strikes. These can be done by using scalable redundancy resources, while keeping the hardened combinational logic circuits intact. The technique combines different types of redundancy: information redundancy for error detection, temporal redundancy for soft error correction and hardware redundancy for hard error tolerance. Results on largest ISCAS and ITC benchmark circuits show that our approach has an area cost negligible of about 2% to 3% with a power consumption saving of about 30% compared to TMR. Finally, it deals with aging phenomenon and thus, increases the expected lifetime of logic circuits.

Low-power SRAMs embed power gating facilities for reducing power consumption. Power gating is applied using power switches for controlling the supply voltage applied to the memory cells i.e. one or more memory blocks can be disconnected from the power supply during a long time of inactivity, thus reducing the power consumption. In this paper, we provide a detailed analysis on the impact that defective power switches impose on the behavior of SRAM core-cells. Furthermore, we propose efficient test solutions to detect such faulty behaviors.

The Flash technology is the most popular non-volatile memory technology. In this paper, we present the ability of a SPICE-like model of the ATMEL TSTAC™ eFlash technology to guide the design and test phases. This model is composed of two layers: a functional layer representing the Floating Gate (FG) and a programming layer able to determine the channel voltage level controlling the Fowler-Nordheim tunneling effect. It is able to guide the test phase since it allows analyzing and modeling defects that may affect the eFlash array. This analysis highlights the interest of the proposed model to identify a realistic set of fault models that has to be tested, thus enhancing existing solutions for TSTAC™ eFlash testing. The proposed model is also helpful to guide the design phase. Data presented in the paper demonstrate its accuracy compared to silicon measurements, usefulness to predict the technology shrinking and usefulness to guide the pulse programming method.

As technology scales down, the effects of power supply noise and ground bounce are becoming significantly important. In the existing literature, it has been shown that excessive power supply noise can affect the path delay, while ground bounce is either neglected or assumed similar to power supply noise. In this work, we present a detailed study of combined and uncorrelated power supply noise and ground bounce and their impact on the path delay. Our analyses show that different combination of power supply noise and ground bounce can lead to either delay speed-up or slow-down. Furthermore, our study shows the degrading influence of supply noise resonance on the path delay. We perform HSPICE simulations for path delay analysis on various technology nodes i.e. 130nm, 90nm, 65nm and 45nm.

A comprehensive SRAM test must guarantee the correct functioning of each cell of the memory (ability to store and to maintain data), and the corresponding addressing, write and read operations. SRAM testing is mainly based on the concept of fault model used to mimic faulty behaviors. Traditionally, the effects of bit line coupling capacitances have not been considered during the fault analysis. However, recent works show the increasing impact of bit line coupling capacitances on the SRAM behavior. This paper reviews and discusses preview works addressing the issues coming from bit line parasitic capacitances and data contents on SRAM testing, pointing out the impacts of these effects on the existing test solutions. Then, we introduce two optimizations of the state-of-the-art test solution able to take into account the influence of bit line coupling capacitances while reducing the test length of about 60% and 80%, respectively.

High power consumption during test may lead to yield loss and premature aging. In particular, excessive peak power during at-speed delay fault testing represents an important issue. In the literature, several techniques have been proposed to reduce peak power consumption during at-speed LOC or LOS delay testing. On the other hand, some experiments have proved that too much test power reduction might lead to test escape and reliability problems. So, in order to avoid any yield loss and test escape due to power issues during test, test power has to map the power consumed during functional mode. In literature, some techniques have been proposed to apply test vectors that mimic functional operation from the switching activity point of view. The process consists of shifting-in a test vector (at low speed) and then applying several successive at-speed clock cycles before capturing the test response. In this paper, we propose a novel flow to determine the functional power to be used as test power (upper and lower) limits during at-speed delay testing. This flow is also used for comparison purpose between the above-mentioned test scheme and power consumption during the functional operation mode of a given circuit. The proposed methodology has been validated on an Intel MC8051 micro controller synthesized in a 65 nm industrial technology.

This paper proposes an innovative approach to cope with defects in SRAM bit-cell array. Traditional approaches use spare parts (rows, columns or blocks) to replace defective bit-cells. Instead of replacing defective bit-cells, we propose to operate the SRAM with reduced storage capacity whenever defective bit-cells are present. We implement this feature through a programmable combinational logic, called Scrambling Module (SM), which scrambles the memory addresses. The scrambling changes the addresses of the defective bit-cells, grouping them in an idle address zone located at the end of the memory address plan. The SM is described by using a mathematical formulation based on linear algebra. The proposed technique can be used in combination with traditional redundancy approaches using spare rows and/or columns. The effectiveness of three different SM is demonstrated, considering a 1MBit SRAM. For a given level of defect tolerance, it is shown that our technique can reduce the amount of spare area by several orders of magnitude. Moreover, as the SM is implemented as an external block, it does not affect the maximum operation frequency of the SRAM. Instead, it affects the memory access delay.

This paper addresses the problem of logic diagnosis of System-on-Chip (SoC). We propose a diagnosis approach based on a matching algorithm between a set of predicted failures and the set of failures observed during the test phase. The result of the diagnosis is a ranked list of suspected nets able to explain the observed failures. Experimental results show the diagnosis accuracy of the proposed approach in terms of absolute number of suspects. Moreover, a comparison with an industrial reference tool highlights the reliability of our approach.

This paper introduces a simulator that allows analyzing the radiation induced errors on memory devices. The simulator takes all the radiation effects on SRAM into account and can be easily tuned on the base of data gained during radiation experiments and/or presented in literature. We also present a case study application in which the proposed simulator is used to validate a low-cost hardware platform for soft error detection in avionic environment by the mean of atmospheric balloons in the context of HAMLET project.

Reliability analysis of SRAM core-cells requires statistical methods with very high accuracy to cope with very low failure probabilities. Although new statistical methods have been recently proposed, to the best of our knowledge, there is no method able to evaluate the joint failure probability (the probability that at least one failure mechanism occurs) of an SRAM core-cell with enough accuracy in a reasonable time. We propose a statistical simulation method based on the analytical integration of the multivariate Gaussian distribution function.

At-speed scan testing has become mandatory due to the extreme CMOS technology scaling. The two main at-speed scan testing schemes are namely Launch-Off-Shift (LOS) and Launch-Off-Capture (LOC). As it can be easily implemented, LOC has been widely investigated in the literature in the last few years, especially regarding test power consumption. Conversely, LOS has received much less attention. In this paper, we propose a comparison between the two testing schemes in terms of transition fault coverage and power consumption, in order to quantify the pros and cons of LOS with respect to LOC. This study shows that LOS not only exhibits higher performance in coverage but also does not require as much extra power as predicted, especially in terms of peak power. These facts may represent convincing arguments for its wider use and development.

In this paper, we present a comparative study on the effects of resistive-bridging defects in the SRAM core-cells, considering different technology nodes. In particular, we analyze industrial designs of SRAM core-cell at the following technology nodes: 90nm, 65nm and 40nm. We have performed an extensive number of simulations, varying the resistive value of defects, the power supply voltage, the memory size and the temperature. Experimental results show malfunctions not only within the defective core-cell, but also in other core-cells (defect-free) of the memory array.

In the context of SRAM testing, we propose a methodology to define proper conditions under which SRAMs should be tested to improve their reliability. This methodology is especially suitable to deal with the impact of threshold voltage variability affecting SRAM core-cell transistors. By establishing an objective manner of comparing different test conditions, the proposed study shows how it is possible to detect SRAM core-cells with poor quality by applying a reduced set of test runs. The proposed methodology also allows determining the most appropriate DfT (Design-for-Test) technique for each peculiar SRAM design and technology.

Flash memories are based on the floating gate technology allowing the write and erase data electronically. Such a technology can be prone to complex defects leading to faulty behaviors. In this paper, we introduce an electrical model of the ATMEL TSTACTM eFlash memory technology. The model is composed of two layers: a functional layer representing the floating gate and a programming layer able to determine the channel voltage level controlling the Fowler-Nordheim tunneling effect. The proposed model has been validated by means of simulations and comparisons with ATMEL silicon data. We apply this model for the analysis of defect-induced failures. As a case study, a resistive defect injection is considered.

In this paper we investigate the possibility of using commercial synthesis tools to build parity predictors for nano-electronic gates designs. They will be used as redundant resources for robustness improvement for future CMOS technology nodes.

This study investigates the reasons why test power reduction through X-filling techniques works well for cycle-average power reduction but is not so efficient concerning instantaneous peak power reduction.

Negative Bias Temperature Instability (NBTI) is a degradation phenomenon that occurs in PMOS transistors during circuit lifetime. Recent works have proposed transistor level and circuit level models that allow designers to deal with such phenomenon. Based on these models and taking into account Random Dopant Fluctuation (RDF), we study the possibility of detecting SRAM core-cells that are prone to NBTI failures during post-production test. For this purpose, we introduce a statistical simulation method that allows estimating the amount of NBTI affected core-cells that pass or fail under given test conditions. Supply voltage, temperature, word line pulse width, word line pulse voltage and bit line voltage are the parameters considered as test conditions. An industrial core-cell design with a 65 nm technology is used as case study.

This paper presents a comprehensive framework for logic diagnosis consisting of two main phases. In the first phase, a set of suspected faulty sites is obtained by applying an approach based on an Effect-Cause analysis. Then, in the second phase, a set of realistic fault models is associated with each suspected faulty site by analyzing specific information, called fault evidences, collected during the first phase. The main advantage of the proposed methodology is its capability to deal with several fault models at the same time. Another advantage is that it is able to handle both single and multiple fault occurrences. Experiments on ISCAS85, ISCAS89, and ITC99 benchmark circuits show the efficiency of the proposed method both in terms of diagnosis resolution and accuracy of the predicted fault models.

The importance of delay faults proportionally increases when entering in the nano-technology era, and logic diagnosis must localize delay faults as precisely as possible to speed-up yield ramp-up. This paper presents a logic diagnosis approach targeting delay faults. The proposed approach is based on the single-location-at-a-time (SLAT) paradigm used to determine a set of suspects. It addresses the case of sequential circuits tested at-speed. The main advantages of this approach are that it can manage a comprehensive set of delay faults, and that it is independent on the size of the delay (induced by the fault). Experimental results show the effectiveness of the proposed approach in terms of absolute number of suspects.

This paper presents a logic diagnosis approach targeting bridging faults. The proposed approach is performed in two phases, (i) a fault localization phase based on the single-location-at-a-time (SLAT) paradigm determining a set of suspects, and (ii) a fault model allocation phase associating a set of fault models to each suspect identified during the first phase. The main advantages of this approach are that the fault localization phase is fault model independent, and that the fault model allocation phase is able to deal at the same time with several bridging fault models leading to either static or dynamic faulty behaviors. Experimental results on full scan circuits show the diagnosis accuracy of the proposed approach in terms of absolute number of suspects. Moreover, a comparison with an industrial reference tool highlights the reliability of our approach.

This paper proposes an efficient technique for transition delay fault coverage measurement in synchronous sequential circuits. The proposed strategy is based on a combination of multi-valued algebra simulation, critical path tracing and deductive fault simulation. The main advantages of the proposed approach are that it is highly computationally efficient with respect to state-of-the-art fault simulation techniques, and that it encompasses different delay sizes in one simulation pass without resorting to an improved transition fault model. Preliminary results on ITC99 benchmarks show that the gain in terms of CPU time is up to one order of magnitude compared to previous existing techniques.

Embedded flash memories are dominated by the NOR architecture but NAND is becoming more and more adopted due to its high storage capacity. This paper presents a preliminary study on actual defects in NAND array.

This paper presents LIFTING (LIRMM fault simulator), an open-source simulator able to perform both logic and fault simulations for single/multiple stuck-at faults and single event upset (SEU) on digital circuits described in Verilog. Compared to existing tools, LIFTING provides several features for the analysis of the fault simulation results, meaningful for research purposes. Moreover, as an open-source tool, it can be customized to meet any user requirements. Experimental results show how LIFTING has been exploited on research fields. Eventually, execution time for large circuit simulations is comparable to the one of commercial tools.

An efficient on-chip infrastructure for memory test and repair is crucial to enhance yield and availability of SoCs. Most of the existing built-in self-repair solutions reuse IP-Cores for BIST without modifications. However, this prevents an optimized test and repair interaction. In this paper, the concept of modular BIST for memories is introduced, which supports a more efficient interleaving of test and repair and can be achieved with only small modifications in the BIST control.

With the technology entering the nano dimension, manufacturing processes are less and less reliable, thus drastically impacting the yield. A possible solution to alleviate this problem in the future could consist in using fault tolerant architectures to tolerate manufacturing defects. In this paper, we analyze the conditions that make the use of a classical triple modular redundancy (TMR) architecture interesting for a yield improvement purpose.

The usual techniques for memory diagnosis are mainly based on signature analysis. They consist in creating a fault dictionary that is used to determine the correspondence between the signature and the fault models affecting the memory. The effectiveness of such diagnosis methods is therefore strictly related to the fault dictionary accuracy. To the best of our knowledge, most of existing signature-based diagnosis approaches targets static faults only. In this paper, we present a new diagnosis approach that represents an alternative to signature-based approaches. This new diagnosis technique, named history-based diagnosis, makes use of the effect-cause paradigm already developed for logic design diagnosis. It consists in creating a database containing the history of operations (read and write) performed on a faulty memory core-cell. This information is crucial to track the root cause of the observed faulty behavior and it can be used to generate the set of possible fault primitives representing the set of suspected fault models. This new diagnosis method is able to identify static as well as dynamic faults. Although applied to SRAMs in this paper, it can be effective also for other memory types such as DRAMs. Experimental results are provided to prove the efficiency of the proposed methodology in generating a list of suspected faults as well as the location of the faulty components in the memory.

Manufacturing processes in the nanoscale era are less and less reliable thus leading to lower and lower yield. In this paper we investigate the usage of TMR architectures for logic cores to increase SoC yield.

Memory testing commonly faces two issues: the characterization of detailed and realistic fault models, and the definition of time-efficient test algorithms to detect them. March tests have proven to be a fast, simple and regularly structured class of memory test algorithms. This paper proposes a new polynomial algorithm to automatically generate march tests. The formal model adopted to represent memory faults allows the definition of a general methodology to deal with both static, dynamic and linked faults.

In this paper, we present a diagnosis methodology targeting the whole set of bridging faults leading to either static or dynamic faulty behavior. The adopted diagnosis algorithm resorts only to logic information provided by the tester without requiring a detailed description of the fault models. It is based on an Effect-Cause analysis providing a ranked list of suspects always including the root cause of the observed error. Experimental results on benchmarks ISCAS'89 and ITC '99 show the efficiency of the proposed solution in terms of diagnosis resolution and required computational time.

This paper presents DERRIC (Diagnosis of logic ERRors in VLSI Integrated Circuits), a diagnostic tool targeting most of the fault models used in practice today. This tool is intended to be used to diagnose faulty behaviors in nanometric circuits for which the classical stuck-at fault model is far to cover all the realistic failures. The underlying method of DERRIC is based on the Effect-Cause approach which relies on the two following main operations. The first one is based on critical path tracing (CPT) that consists in identifying critical lines in the Circuit Under Test (CUT) which can be the source of observed errors. The second one consists in allocating a set of possible fault models to each critical line, so that root causes of failures can be finally determined. The main advantage of this method is that it does not need to explicitly consider each fault model during the diagnosis process. Experiments on ISCAS'85 and ITC'99 benchmarks show the efficiency of the proposed tool in terms of diagnosis resolution.

Static linked faults are considered an interesting class of memory faults. Their capability of influencing the behavior of other faults causes the hiding of the fault effect and makes test algorithm design and validation a very complex task. This paper presents a memory fault simulator architecture targeting the full set of linked faults

Yield and reliability are two key factors affecting costs and profits in the semiconductor industry. Stress testing is a technique based on the application of higher than usual levels of stress to speed up the deterioration of electronic devices and increase yield and reliability. One of the standard industrial approaches for stress testing is high temperature burn-in. This work proposes a full-scan circuit ATPG for dynamic burn-in. The goal of the proposed ATPG approach is to generate test patterns able to force transitions into each node of a full scan circuit to guarantee a uniform distribution of the stress during the dynamic burn-in test

Static linked faults are considered an interesting class of memory faults. Their capability of influencing the behavior of other faults causes the hiding of the fault effect and makes test algorithm design a very complex task. A large number of March tests with different fault coverage have been published and some methodologies have been presented to automatically generate March tests. In this paper we present an approach to automatically generate March tests for static linked faults. The proposed approach generates better test algorithms then previous, by reducing the test length

Linked faults are considered an interesting class of memory faults. Their capability of influencing the behavior of other faults causes the hiding of the fault effect and makes test algorithm design a very complex task. Although several March tests have been developed for the wide memory faults spread, a few of them are able to detect linked faults. In the present paper March AB, a March test targeting the set of realistic memory linked fault is presented. Comparison results show that the proposed March test provides the same fault coverage of already published algorithms but, it reduces the test complexity and therefore the test time. Moreover, a complete taxonomy of linked faults will be presented

New memory production modern technologies introduce new classes of faults usually referred to as dynamic memory faults. Although some hand-made March tests to deal with these new faults have been published, the problem of automatically generate March tests for dynamic faults has still to be addressed, in this paper we propose a new approach to automatically generate March tests with minimal length for both static and dynamic faults. The proposed approach resorts to a formal model to represent faulty behaviors in a memory and to simplify the generation of the corresponding tests.

Among the different types of algorithms proposed to test static random access memories (SRAMs), March tests have proven to be faster, simpler and regularly structured. New memory production technologies introduce new classes of faults usually referred to as dynamic memory faults. A few March tests for dynamic fault, with different fault coverage, have been published. In this paper, we propose new March tests targeting unlinked dynamic faults with lower complexity than published ones. Comparison results show that the proposed March tests provide the same fault coverage of the known ones, but they reduce the test complexity, and therefore the test time.