机器学习辅助的SRAM软错误率表征：机遇与挑战.pdf

《机器学习辅助的SRAM软错误率表征：机遇与挑战.pdf》由会员分享，可在线阅读，更多相关《机器学习辅助的SRAM软错误率表征：机遇与挑战.pdf（30页珍藏版）》请在三个皮匠报告上搜索。

1、ML-assisted SRAM Soft Error Rate Characterization:Opportunities and ChallengesMasanori Hashimoto,Ryuichi Yasuda,Kazusa Takami,Yuibi Gomi Dept.Informatics,Kyoto UniversityKozo TakeuchiJAXAhashimotoi.kyoto-u.ac.jp1Cosmic ray-inducedneutrons and muons are falling into VLSI chipsExample of nuclear react

2、ion3Example of reaction in VLSI chip41 S.Abe,et.al,”Multi-scale Monte Carlo simulation of soft errors using PHITS-HyENEXSS code system,”IEEE Trans.Nuclear Science,2012Injected charge may result in bit flip called soft error.3.76 MeV 1.37 MeV 3.43 MeVn 100 MeVExample 65nm20m20mMemory cell(2.0 x 0.5 m

3、2)Multi-physics multi-layer phenomena with diverse temporal and spatial scales610-14m10-6mMulti-physics multi-layer phenomena with diverse temporal and spatial scales710-14m10-6mSRAM Soft Error Rate Characterization(Focus of this talk)Multi-physics multi-layer phenomena with diverse temporal and spa

4、tial scales810-14m10-6mControl flow monitoring(6D-5,Wed.)Agenda Background:soft error Conventional soft error rate(SER)simulation and its challenges Proposed method and experiments Future directions and conclusions9Conventional SER simulation10(3)SRAM Cell behavior(2)Charge deposition(1)Nuclear phys

5、icsSimulators:PHITS2,Geant4 3,etc10-14m10-6mMonte Carlo simulation aiming to reproduce nuclear physics,charge deposition and SRAM cell behavior.A number of event data are generated.(particle type,energy,location,direction,resultant charge deposition)2 T.Sato,et al.,“Recent improvements of the partic

6、le and heavy iontransport code system PHITS version 3.33,”J.Nuclear Sci.&Tech.,2024.3 S.Agostinelli,et al.,“Geant4-a simulation toolkit，”Nuclear Instruments and Methods in Physics Res.Sec.A，2003Conventional SER simulation11(3)SRAM Cell behavior(2)Charge deposition(1)Nuclear physicsSimulators:PHITS,G