Keynote Talk

Reconfigurable Wireless Multi-chiplet Networking: A "Transformer" to Exascale AI Chips

Danella Zhao

Department of Computer Science
Old Dominion University, Norfolk, VA, USA


Abstract: As we enter the "More than Moore" era of semiconductor integration, chiplets have emerged as a vital technology to enable heterogeneous system integration for high-performance AI training and inferencing at the edge. In this talk, we will discuss the key challenges faced in the chiplet paradigm and explore the core designs of a reconfigurable hybrid wireless inter-chiplet network architecture, a so-called chiplet “transformer” to scale AI-focused parallel architecture. Our ultimate goal is to exploit viable solutions to enable chiplet architecture as an AI infrastructure for human-centric computing.