Techniques for Robust ReRAM-based Neural Network Accelerators

Release Time：2021-11-04Number of visits：227

Speaker:    Mr. Yu Ma

Time:         15:00-16:00 , Nov.05
Location:   SIST 1A 200
Host:          Prof. Pingqiang Zhou
Abstract:
Resistive Random Access Memory (ReRAM) is an emerging memory featuring high speed, low power and large capacity. The crossbar-structured ReRAM can not only store the weights of the neural network but also greatly speed up the matrix-vector multiplications. However, the reliability issues of ReRAM lead to significant accuracy degradation of neural networks. In this talk, I will present several techniques to mitigate the reliability issues (including stuck-at faults, conductance variation and drift) in ReRAM-based Neural Network Accelerators.

Bio:
Yu Ma received his B.D. from Qingdao University and is a PhD candidate at ShanghaiTech University. He worked as a visiting scholar at University of Minnesota and intern at AMD. His major research area is the reliability of ReRAM based neural networks. His research interests include neuromorphic computing based on emerging devices, the development and analysis of algorithms for computer graphics, optimization of neural network accelerators. His research results have turned into peer-reviewed papers in ASP-DAC, GLSVLSI, APCCAS, etc.