Abstract
Artificial intelligence (AI) has become one of the most transformative technologies of our time, with broad potential applications across nearly every sector of society. This talk will address the key challenges in advancing the functionality and performance of AI chips, with a particular focus on low power, unmanned autonomous systems that must operate in complex, dynamic environments.
We will begin with an overview of current AI technologies, particularly from the scaled CMOS point of view, followed by a discussion of emerging memory and computing paradigms — especially voltage-controlled spintronics. This includes mechanisms such as voltage-controlled magnetic anisotropy via exchange interaction and voltage-controlled Dzyaloshinskii-Moriya interaction for manipulating spin textures.
The integration challenges of these novel technologies with scaled CMOS will be examined through an illustrative example. I will also present several potential applications, with an emphasis on how voltage-controlled spintronics could enable the next generation of generative AI. In addition, we will explore the role of spintronic-based convolutional neural networks and Bayesian networks, as implemented with voltage-controlled magnetic anisotropy, the latter offering promising pathways to enhance safety, reliability, and fault tolerance, which are critical for edge applications. Finally, I will outline several plausible research directions and conceptual frameworks aimed at stimulating discussion to overcome scaling and performance bottlenecks in next-generation AI chip technologies
Time:
13:10-15:00 , April 18, 2025
Venue:
UMC room, B1F, Dept. of EE, Tzu-Chiang Campus, NCKU