Virtual platform frameworks have been extended to allow earlier soft error analysis of more realistic multicore systems (i.e., real software stacks and state-of-the-art ISAs). The high observability and simulation performance of underlying frameworks enable to generate and collect more error/failure-related data, considering complex software stack configurations, in a reasonable time. When dealing with sizeable failure-related data sets obtained from multiple fault campaigns, it is essential to filter out parameters (i.e., features) without a direct relationship with the system soft error analysis. In this regard, this paper proposes the use of supervised and unsupervised machine learning techniques, aiming to eliminate non-relevant information as well as identify the correlation between fault injection results and application and platform characteristics. This novel approach provides engineers with appropriate means to investigate new and more efficient fault mitigation techniques. The underlying approach is validated with an extensive data set gathered from more than 1.2 million fault injections, comprising several benchmarks, a Linux OS and parallelization libraries (e.g., MPI and OpenMP), as well as through a realistic automotive case study.
26 July 2021 in IEEE Transactions on Circuits and Systems I: Regular Papers
Deep neural networks (DNNs) are being incorporated in resource-constrained IoT devices, which typically rely on reduced memory footprint and low-performance processors. While DNNs' precision and performance can vary and are essential, it is also vital to deploy trained models that provide high reliability at low cost. To achieve an unyielding reliability and safety level, it is imperative to provide electronic computing systems with appropriate mechanisms to tackle soft errors. This paper, therefore, investigates the relationship between soft errors and model accuracy. In this regard, an extensive soft error assessment of the MobileNet model is conducted considering precision bitwidth variations (2, 4, and 8 bits) running on an Arm Cortex-M processor. In addition, this work promotes the use of a register allocation technique (RAT) that allocates the critical DNN function/layer to a pool of specific general-purpose processor registers. Results obtained from more than 4.5 million fault injections show that RAT gives the best relative performance, memory utilization, and soft error reliability trade-offs w.r.t. a more traditional replication-based approach. Results also show that the MobileNet soft error reliability varies depending on the precision bitwidth of its convolutional layers.
7 January 2022 in IEEE Transactions on Circuits and Systems II: Express Briefs
Driven by the success of machine learning algorithms for recognizing and identifying objects, there are significant efforts to exploit convolutional neural networks (CNNs) in edge devices. The growing adoption of CNNs in safety-critical embedded systems (e.g., autonomous vehicles) increases the demand for safe and reliable models. In this sense, this work investigates the soft error reliability of two CNN inference models considering single event upsets (SEUs) occurring in register files, RAM, and Flash memory sections. The results show that the incidence of SEUs in flash memory sections tend to lead to more critical faults than those resulting from the occurrence of bit-flips in RAM sections and register files.
27 March 2023 in IEEE Transactions on Nuclear Science
Deep neural network (DNN) models are being deployed in safety-critical embedded devices for object identification, recognition, and even trajectory prediction. Optimised versions of such models, in particular the convolutional ones, are becoming increasingly common in resource-constrained edge-computing devices (e.g., sensors, drones), which typically rely on reduced memory footprint, low power budget and low-performance microprocessors. DNN models are prone to radiation-induced soft errors, and tackling their occurrence in resource-constrained devices is a mandatory and substantial challenge. While traditional replication-based soft error mitigation techniques will likely account for a reasonable performance penalty, hardware solutions are even more costly. To undertake this almost contradictory challenge, this work evaluates the efficiency of a lightweight software-based mitigation technique, called Register Allocation Technique (RAT), when applied to a convolutional neural network (CNN) model running on two commercial Arm microprocessors (i.e., Cortex-M4 and M7) under the effects of neutron radiation. Gathered results obtained from two neutron radiation campaigns suggest that RAT can reduce the number of critical faults in the CNN model running on both Arm Cortex-M microprocessors. Results also suggest that the SDC FIT rate of the RAT-hardened CNN model can be reduced in up to 83% with a runtime overhead of 32%.
23 February 2021 in IET Computers & Digital Techniques
Soft error resilience has become an essential design metric in electronic computing systems as advanced technology nodes have become less robust to high‐charged particle effects. Designers, therefore, should be able to assess this metric considering several software stack components running on top of commercial processors, early in the design phase. With this in mind, researchers are using virtual platform (VP) frameworks to assess this metric due to their flexibility and high simulation performance. In this regard, herein, this goal is achieved by analysing the soft error consistency of a just‐in‐time fault injection simulator (OVPsim‐FIM) against fault injection campaigns conducted with event‐driven simulators (i.e. more realistic and accurate platforms) considering single and multicore processor architectures. Reference single‐core fault injection campaigns are performed on RTL descriptions of Arm Cortex‐M0 and M3 processors, while gem5 simulator is used to multicore Arm Cortex‐A9 scenarios. Campaigns consider different open‐source and commercial compilers as well as real software stacks including FreeRTOS/Linux kernels and 52 applications. Results show that OVPsim‐FIM is more than 1000× faster than cycle‐accurate simulators and up to 312× faster than event‐driven simulators, while preserving the soft error analysis accuracy (i.e. mismatch below to 10%) for single and multicore processors.
6 March 2023 in Foundations and Trends® in Electronic Design Automation
Over the past decade, a massive proliferation of machine learning algorithms has emerged, from applications for surveillance to self-driving cars. The turning point occurred with the arrival of Convolutional Neural Network (CNN) models and the incredible accuracy brought by Deep Neural Networks (DNNs) at the cost of high computational complexity. In this growing environment, graphic processing units (GPUs) have become the de facto reference platform for the training and inference phases of CNNs and DNNs due to their high processing parallelism and memory bandwidth. However, GPUs are power-hungry architectures. To enable the deployment of CNN and DNN applications on energy-constrained devices (e.g., IoT devices), industry and academic research have moved towards hardware accelerators. Following the evolution of neural networks (from CNNs to DNNs), this survey sheds light on the impact of this architectural shift and discusses hardware accelerator trends in terms of design, exploration, simulation, and frameworks developed in both academia and industry.