Zero-knowlege Virtual Machines

Abstract

This thesis explores the concept of zero-knowledge proofs and their application in zkVMs - virtual machines capable of generating proofs of correct computation without revealing private input. We begin by introducing fundamental cryptographic tools such as zk-SNARKs and zk-STARKs, along with supporting techniques such as lookup tables. We then analyze the architecture of two concrete zkVM implementations: RISC Zero and SP1. We describe their different approaches to proof systems, recursion mechanisms, and optimization strategies. In the final part, we present comparative benchmarks on various examples, highlighting their performance differences. The results show that RISC Zero produces smaller proofs and has faster verification, while SP1 is faster in proof generation and requires fewer computation cycles.