Artificial Intelligence (AI) is rapidly moving beyond niche use cases to become embedded in systems and services we rely on every day, from recommendation engines and intelligent search to robotics and autonomous vehicles. This ubiquity is driving the need for more powerful and efficient chips to run AI applications. While the underlying algorithms can run on general-purpose processors, they can be accelerated using dedicated chips that implement key functions directly in the hardware. Given the complexity of the problems they are solving, it is not surprising that designing AI chips is challenging. They tend to be some of the largest devices when used for AI training, pushing the limits of architecture, silicon, and packaging. For AI inference, challenges like low power and latency are critical. Every stage of the development process requires specialized knowledge, tools, and methodology.