*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Reliability and Security of Compute-In-Memory Based Deep Neural Network Accelerators
Committee:
Dr. Shimeng Yu, ECE, Chair, Advisor
Dr. Callie Hao, ECE
Dr. Saibal Mukhopadhyay, ECE
Dr. Tushar Krishna, ECE
Dr. Ada Gavrilovska, CoC
Abstract: Compute-In-Memory (CIM) is a promising solution for accelerating DNNs at edge devices, utilizing mixed-signal computations. However, it requires more cross-layer designs from algorithm levels to hardware implementations as it behaves differently from the pure digital system. On one side, the mixed-signal computations of CIM face unignorable variations, which could hamper the software performance. On the other side, there are potential software/hardware security vulnerabilities with CIM accelerators. This research aims to solve the reliability and security issues in CIM design for accelerating Deep Neural Network (DNN) algorithms as they prevent the real-life use of the CIM-based accelerators. Some non-ideal effects in CIM accelerators are explored, which could cause reliability issues, and solved by the software-hardware co-design methods. In addition, different security vulnerabilities for SRAM-based CIM and eNVM-based CIM inference engines are defined, and corresponding countermeasures are proposed.
