Advances in Digital vs. Analog AI Accelerators Slides

Abstract: For state-of-the-art AI accelerators, there have been large advances in both all-digital and analog/mixed-signal circuitbased designs. This tutorial presents a practical overview and comparison of recent digital and analog AI accelerators. It will first introduce recent AI algorithms for computer vision and speech applications, which have been targeted for many AI hardware designs. Next, it will present a survey of (i) all-digital AI accelerators, including designs with new dataflow, low precision, and sparsity, and (ii) analog/mixed-signal AI accelerators featuring switch-capacitor circuits and in-memory computing with analog-to-digital converters. The tutorial discusses the key trade-offs of both design approaches including circuit/architecture design, algorithm-mapping flexibility, hardware accuracy and energy efficiency