Paper Title: Chiplets Enable New Heterogenous Compute Architectures

Paper Abstract: Traditional compute architectures assume one chip per package, limiting the ability to explore novel approaches. With the advent of massive AI workloads, old architectures cannot keep up with bandwidth demands. Heterogeneous compute architectures, with a mix of domain-specific chiplets, offer higher performance and energy efficiency. Examples show how they can serve several application domains, including autonomous vehicles and datacenters.

Paper Author: Philip Lewer, Sr Product Director, Untether AI

Author Bio: Philip currently works as the Senior Director of Product at Untether AI, where he is responsible for outbound marketing, software product management, and business operations. Prior to joining Untether AI, Philip worked at Arm in different positions. Philip served as the Director of Ecosystem and Developer Relations, Machine Learning, where he managed the ecosystem and developer relations team and led the Arm AI Partner Program. He completed his Master of Business Administration (MBA) with a specialization in Management and Entrepreneurial Studies.