SnarkFold: Efficient SNARK Proof Aggregation from Split Incrementally Verifiable Computation

Xun Liu, Shang Gao, Tianyu Zheng, Bin Xiao

Abstract: The succinct non-interactive argument of knowledge (SNARK) technique has been extensively utilized in blockchain systems to replace the costly on-chain computation with the verification of a succinct proof. However, most existing applications verify each proof independently, resulting in a heavy load on nodes and high transaction fees for users. Currently, the mainstream proof aggregation schemes are based on a generalized inner product argument, which has a logarithmic proof size and verification cost. To improve the efficiency of verifying multiple proofs, we introduce SnarkFold, a novel SNARK-proof aggregation scheme with constant verification time and proof size. SnarkFold is derived from incrementally verifiable computation (IVC) and is optimized further through the folding scheme. By folding multiple instance-proof pairs, SnarkFold defers the expensive SNARK verification (e.g., elliptic curve pairing) to the final step. Additionally, we propose a generic technique to enhance the verifier's efficiency by delegating instance aggregation tasks to the prover. The verifier only needs a simple preprocessing to check the validity of the delegation. We further introduce folding schemes for Groth16 and Plonk proofs. Experimental results demonstrate that SnarkFold offers significant advantages, with an aggregated Plonk proof size of just $0.5$ KB and the verification time of only $4.5$ ms for aggregating 4096 Plonk proofs.