Microsoft Research unveils Magnetar, a system accelerating zero-knowledge proof generation for general-purpose programs

Zero-knowledge proofs allow a program to produce, alongside its output, a compact proof that the computation was executed correctly. Magnetar advances this paradigm by supporting programs written in Rust, Go, or C, reducing the burden on developers to rewrite code into specialized cryptographic formats.

The system reports average speedups of 10 to 39 times over prior fastest comparable systems, a leap attributed to architectural optimizations that exploit repetition in processor execution, such as those explored in earlier Microsoft Research systems like Phalanx and Nova-family proof systems.

Security is grounded in standard hash functions, avoiding reliance on elliptic curves vulnerable to quantum attacks or newer cryptographic assumptions lacking decades of scrutiny. Microsoft’s post-quantum security roadmap targets 2029 for full migration to post-quantum cryptography across its products and services.

Magnetar builds on over a decade of Microsoft Research work, including Spartan (2019), which generalized zero-knowledge proofs to general-purpose programs without trusted setups, and subsequent systems like Phalanx, Nova, HyperNova, NeutronNova, Neo, and SuperNeo that incrementally improved speed and generality.

The team notes that prior systems often required manual rewriting of programs into fixed arithmetic circuits, a process that could take months and introduce security bugs, or relied on elliptic-curve cryptography with quantum vulnerabilities or trusted setups.