0000000000116327
AUTHOR
Alberto Bosio
Dynamic Test Methods for COTS SRAMs
International audience; In previous works, we have demonstrated the importance of dynamic mode testing of SRAM components under ionizing radiation. Several types of failures are difficult to expose when the device is tested under static (retention) mode. With the purpose of exploring and defining the most complete testing procedures and reveal the potential hazardous behaviors of SRAM devices, we present novel methods for the dynamic mode radiation testing of SRAMs. The proposed methods are based on different word address accessing schemes and data background: Fast Row, Fast Column, Pseudorandom, Adjacent (Gray) and Inverse Adjacent (Gray). These methods are evaluated by heavy ion and atmos…
Emulating the Effects of Radiation-Induced Soft-Errors for the Reliability Assessment of Neural Networks
International audience; Convolutional Neural Networks (CNNs) are currently one of the most widely used predictive models in machine learning. Recent studies have demonstrated that hardware faults induced by radiation fields, including cosmic rays, may significantly impact the CNN inference leading to wrong predictions. Therefore, ensuring the reliability of CNNs is crucial, especially for safety-critical systems. In the literature, several works propose reliability assessments of CNNs mainly based on statistically injected faults. This work presents a software emulator capable of injecting real faults retrieved from radiation tests. Specifically, from the device characterisation of a DRAM m…
Effects of Thermal Neutron Irradiation on a Self-Refresh DRAM
International audience; In this study, static and dynamic test methods were used to define the response of a self-refresh DRAM under thermal neutron irradiation. The neutron-induced failures were investigated and characterized by event cross-sections, soft-error rate and bitmaps evaluations, leading to an identification of permanent and temporarily stuck cells, block errors, and single-bit upsets.
Neutron-Induced Effects on a Self-Refresh DRAM
International audience; The field of radiation effects in electronics research includes unknowns for every new device, node size, and technical development. In this study, static and dynamic test methods were used to define the response of a self-refresh DRAM under neutron irradiation. The neutron-induced effects were investigated and characterised by event cross sections, soft-error rate, and bitmaps evaluations, leading to an identification of permanent and temporary stuck cells, single-bit upsets, and block errors. Block errors were identified in different patterns with dependency in the addressing order, leading to up to two thousand faulty words per event, representing a real threat fr…
Investigating the Impact of Radiation-Induced Soft Errors on the Reliability of Approximate Computing Systems
International audience; Approximate Computing (AxC) is a well-known paradigm able to reduce the computational and power overheads of a multitude of applications, at the cost of a decreased accuracy. Convolutional Neural Networks (CNNs) have proven to be particularly suited for AxC because of their inherent resilience to errors. However, the implementation of AxC techniques may affect the intrinsic resilience of the application to errors induced by Single Events in a harsh environment. This work introduces an experimental study of the impact of neutron irradiation on approximate computing techniques applied on the data representation of a CNN.