Position: PhD Candidate
Current Institution: UCSD
Abstract: Neuro-inspired Computing using Resistive Memory Devices
Deep neural networks (DNN) have revolutionized artificial intelligence and led to remarkable advances across diverse applications. However training the network takes substantial computing power and time due to iterative updates of a massive number of weights using conventional platforms such as CPUs/GPUs which are based on the traditional Von Neumann architecture. To overcome this my current research focuses on using algorithm and hardware co-design approach to perform in-memory computing on various machine learning tasks with emerging non-volatile memory (eNVM) devices. More specifically I have been working on 1) developing neuroinspired unsupervised learning and pruning with subquantum CBRAM arrays. 2) developing adaptive Quantization technique to Improve the performance of In-memory unsupervised learning with PCM memory. 3) benchmarking the potential use of STT-MRAM for in-memory computing against SRAM at deeply scaled technology nodes. 4) Interfacing eNVM based crossbar with neural electrodes to build a real-time high-energy efficient in-memory spike sorting system.
Yuhan Shi is a fifth year Ph.D. student at UC San Diego working with Professor Duygu Kuzum in neuroelectronics lab. Prior coming to UCSD Yuhan receievd her bachelor degree in electrical engineering at Rensselaer Polytechnic Institute. Her research interet lies in the areas of emerging non-volatile memory (eNVM) devices and machine learning. She received the Qualcomm Fellowship in 2018-2019 and UCSD Department Fellowship in 2016-2017.