Single Event Upsets (SEUs) are most likely to cause bit flips within the trained parameters of a convolutional neural network (CNN). Therefore, it is crucial to analyze and implement hardening techniques to enhance their reliability under radiation. In this paper, random fault injections into the weights of LeNet-5 were carried out in order to evaluate and propose strategies to improve the reliability of a CNN. According to the results of an SEU fault injection, the accuracy of the CNN can be classified into the following three categories: benign conditions, poor conditions, and critical conditions. Two efficient methods for mitigating weight-related SEUs are proposed, as follows: weight limiting and Triple Modular Redundancy (TMR) for the critical bit of the critical layer. The hardening results show that when the number of SEU faults is small, the weight limiting almost completely eliminates the critical and poor conditions of LeNet-5’s accuracy. Additionally, even when the number of SEU faults is large enough, combining the weight limiting and TMR methods for the critical bit of the critical layer can retain the occurrence rate of benign conditions at 98%, saving 99.3% of the hardware resources compared to the Full-TMR hardening method.
13 March 2023 in IEEE Transactions on Nuclear Science
The radiation environment in the inner radiation belt is severe due to the large flux of high energy protons. Very few satellites are deployed in inner radiation belt today, and radiation shielding in this region did not get much attention. An imaging satellite developed by the Chinese Academy of Sciences will operate in the inner radiation belt in 2023. In order to shield the high energy protons, a new composite material was developed, and the shielding ability of this material was compared with that of Al, Ta, and sandwich materials by the Geant4 simulation tool. The results show that the new material can reduce the total ionizing dose by 40%, and the displacement damage dose by 50% compared with the Al shielding in same shielding mass. The number of transmitted protons under the shielding of Al and new materials was tested by using a 30 MeV~60 MeV proton beam on the ground, and the experimental results agree well with the simulation results.
1 August 2020 in IEEE Transactions on Nuclear Science
As the feature size of integrated circuit decreases, the critical charge of single-event effect decreases as well, making nano-scale devices more susceptible to the high-energy charged particles during their application in space. Here, we study the electron-induced single-event effect in 28 nm static random-access memory (SRAM)-based field programmable gate array (FPGA) utilizing high-energy electrons with energy of 1 MeV~5 MeV. The experimental results demonstrate that the 3 MeV electrons can cause single-event functional interrupts (SEFIs) in FPGA, while the electrons with other energies cannot. To further explore the mechanism of electron-induced SEFIs in this nanoscale FPGA, we combined Monte Carlo, Technology Computer-Aided Design (TCAD), and Simulation Program with Integrated Circuit Emphasis (SPICE) simulations. It is revealed that the SEFI was mainly caused by the direct ionization effect of high-energy electrons, and the SEFI was related to the interactions between multiple sensitive nodes.
Static random-access memory (SRAM), a pivotal component in integrated circuits, finds extensive applications and remains a focal point in the global research on single-event effects (SEEs). Prolonged exposure to irradiation, particularly the displacement damage effect (DD) induced by high-energy protons, poses a substantial threat to the performance of electronic devices. Additionally, the impact of proton displacement damage effects on the performance of a six-transistor SRAM with an asymmetric structure is not well understood. In this paper, we conducted an analysis of the impact and regularities of DD on the upset cross-sections of SRAM and simulated the single-event upset (SEU) characteristics of SRAM using the Monte Carlo method. The research findings reveal an overall increasing trend in upset cross-sections with the augmentation of proton energy. Notably, the effect of proton irradiation on the SEU cross-section is related to the storage state of SRAM. Due to the asymmetry in the distribution of sensitive regions during the storage of “0” and “1”, the impact of DD in the two initial states is not uniform. These findings can be used to identify the causes of SEU in memory devices.
This work has studied the co-addition of Sc and Zr elements into the Al-1.75wt%Fe-1.25wt%Ni eutectic alloy. The changes in the microstructure, electrical conductivity, and Vickers hardness of the Al-1.75wt%Fe-1.25wt%Ni-0.2wt%Sc-0.2wt%Zr alloy during heat treatment were studied. The results showed that two-step aging can effectively improve the aging response of the alloy over the single-step aging method. This was ascribed to the minimization of the diffusion difference between Sc and Zr elements. Furthermore, the homogenization treatment can also improve the aging response of the alloy by alleviating the uneven distribution of Sc and Zr. Nevertheless, the micro-alloyed elements exceeded the solid solubility limit in the Al-1.75wt%Fe-1.25wt%Ni-0.2wt%Sc-0.2wt%Zr alloy, and their strengthening effect has ever achieved the best prospect. Finally, both Sc and Zr contents were reduced simultaneously, and the aging response of the Al-1.75wt%Fe-1.25wt%Ni-0.15wt%Sc-0.1wt%Zr alloy was improved by optimized heat treatment. The underlying mechanisms for this alloy design and the corresponding microstructure–mechanical property relationship were analytically discussed.