This course examines why modern artificial intelligence requires specialized hardware and provides a structured overview of the artificial intelligence hardware landscape. It analyzes the computational characteristics of prominent neural network models and the principal strategies used to make them amenable to efficient hardware execution. Building on this foundation, the course surveys the accelerator architectures in widespread use today and introduces emerging paradigms, including near-memory and in-memory computing. Lectures are complemented by student presentations of recent research papers, keeping the course aligned with a rapidly evolving field, and a project provides practical experience in evaluating state-of-the-art models and optimization techniques on real or simulated hardware.

To prepare students 1) to understand why modern AI workloads require specialized hardware and to reason about their computational and memory demands, 2) to understand the accelerator architectures, efficiency techniques, and emerging memory-centric approaches used to meet these demands, and 3) to analyze the performance of AI models on hardware platforms and simulators.