Public Lecture #2

"Sustainable, Pervasive Gen.AI - Nowhere to go, but UP!"

Prof. Luca BENINI (ETH Zürich)

Abstract

Modern computing  is dominated by generative AI, from training, to data generation, to (embodied) inference. The fast pace of imposed by Gen.AI scaling laws requires sustained energy efficiency improvements, which cannot be matched simply by classical (Moore's Law) technology evolution.  We must move toward full three dimensional scaling: UP (in stacking and in scale) is the only way to to go.  To tackle the challenge, we need  to aggressively optimize computing platforms leveraging specialization across all levels of the design hierarchy.  In this talk I will give concrete examples of deep domain specialization, moving beyond single chips and toward three dimensional integration, while emphasizing  the strategic importance of  end-to-end (models,software, instruction set architecture, digital IPs, EDA tools) open innovation.