Rethinking Silicon - The New Architecture for Low-Power AI

Explore the fundamental challenges of implementing AI on edge devices in this comprehensive talk that examines why traditional computing architectures are inadequate for low-power AI applications. Discover how the constant movement of data between memory and arithmetic logic units (ALUs) creates unsustainable power consumption that exceeds the capabilities of edge devices. Learn about the revolutionary silicon paradigms needed to overcome these limitations, including innovative approaches centered on data locality, in-memory computing operations, and hybrid analog architectures. Understand how these architectural innovations can enable ubiquitous AI deployment by dramatically reducing power requirements while maintaining computational performance. Gain insights into the future of edge computing hardware design and the engineering principles that will make intelligent devices feasible across diverse applications and environments.