Proof assistants at the hardware-software interface (invited talk)

Abstract

Some of the earliest applications of proof assistants were to correctness of digital hardware designs, but the subject doesn’t come up too frequently today at venues like CPP. I will try to make the case that proof assistants are a crucial tool for resolving both classical problems and new ones at the hardware-software interface. That is, it is important to understand exactly what guarantees a processor exports to software, it is important to verify that hardware exports those guarantees correctly, and it is important to prove end-to-end theorems covering both hardware and software. A few social developments make this an exciting time to tackle these problems: open instruction sets and open-source hardware designs are growing in real-world relevance, and surprising new classes of security vulnerabilities have gotten more practitioners thinking about precise hardware-software contracts. I will sketch the state of the research area and go into detail on a few of my own related projects. An ancillary goal is to convey that programming or proving digital hardware is a lot like programming or proving software, with a few fun distinctions, so more of the CPP crowd might want to give it a try!