Prof. An-Yeu (Andy) Wu, Distinguished Professor
Department of Electrical Engineering, National Taiwan University
Abstract
AI technology has been widely deployed on edge devices in recent days. The strong demand for AI computing calls for energy-efficient AI chips for emerging applications. In this talk, I will present some ongoing group projects at NTU, such as the Meta' SoC project and the Cross-layer Energy-Efficient Neural Network Design project. Along with the goal of edge computing, several edge AI-related chip designs have been developed. For example, the accelerating chips support mobile augmented reality, super-resolution imaging, and on-device training. Next, an embedded computing platform that links with the CIM (compute-in-memory) macro chip can perform system evaluation of SW/HW co-design for different edge applications. Lastly, how to interplay with emerging memories for storage and computation in edge AI devices becomes a new design issue. We will list some design techniques that can overcome and alleviate the shortcomings of non-volatile memories during NN training and/or inference. The research results will also be briefly addressed in this talk..
Contact information
http://access.ee.ntu.edu.tw/ [1]
Email: andywu
ntu [dot] edu [dot] tw