2025-06-13 –, Side Stage - Cinema 7
This talk presents a new system for verifiable computing using the Binius proof library. Binius implements a hash-based SNARK built on towers of binary fields, which unlocks huge gains in computational efficiency compared with traditional SNARKs, particularly when using specialized hardware. Programs are written in a simple, statically-typed functional language similar to Standard ML, which compiles directly to a custom virtual machine optimized for cryptographic proofs. By co-designing the VM’s instruction set and compiler with the Binius cryptographic library, we enable practical use cases like compressing Merkle tree proofs, aggregating hash-based signatures, and recursive SNARK proofs.
In the talk, we'll discuss the design of our Binius-powered computing stack and share performance results. We will highlight key decisions made in designing the virtual machine, and explain why they lead to improved cryptographic proving efficiency. For example, we'll discuss how the write-once memory model is cheaper than read-write memory for and how the immutability inherent in functional programming supports this design choice. We will also compare our VM's performance to alternatives like RISC-V-based systems.
Jim is the co-founder and CTO of Irreducible. There he leads the cryptography research and development teams, and is an author of the cryptography research papers behind the Binius proof system. Previously, Jim started his career as an early engineer at Coinbase, where he was the Tech Lead on the Payments team, responsible for blockchain and banking integrations. He also established the Protocol team at Coinbase, and contributed to open-source projects serving Bitcoin, Ethereum, and the Lightning Network.