Prof. Jianshi Tang, School of Integrated Circuits, Tsinghua University, China
Memristor-based Neuromorphic Computing for Accelerating AI and Signal Processing
The rapid development of artificial intelligence, such as large language models, calls for more energy-efficient computing hardware, where fundamental breakthroughs from materials and devices to architectures are needed. To overcome the von Neumann bottleneck, neuromorphic computing with emerging devices, such as memristors, emerges as an attractive computing paradigm by mimicking human brain’s working mechanism for energy saving. This lecture intends to present a comprehensive review on the fundamental principles and applications of memristor-based neuromorphic computing. The selection of memristor materials and design of neuromorphic devices are presented. Prof. Tang provides an overview of the recent progress of large-scale integration of memristors with advanced Si CMOS as well as monolithic 3D integration with other emerging devices including IGZO and carbon nanotubes. In particular, a variety of interesting demonstrations of energy-efficient computing-in-memory (CIM) on memristor crossbar arrays and prototype chips will be presented. Two seminal applications of memristor-based CIM are presented: the acceleration of artificial neural networks and the implementation of signal processing algorithms. For each application, Prof. Tang dissects the key challenges and the latest breakthroughs. Finally, he concludes the presentation with a future perspective in this field and highlight several noteworthy directions for future research including memristor-based reservoir computing and analog computing.
