The escalating volume of transactions on blockchains has made signature verification a critical performance bottleneck. Existing approaches of packing signatures inside SNARK often incur significant computational overhead, demanding substantial time and memory resources for proof generation. This not only hinders the participation of resource-constrained devices but also limits the scalability and accessibility of blockchain systems.
To solve the problems, a research team led by Puwen WEI published their new research on 15 April 2026 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The team presents a general approach to batch verification of arbitrary signatures on blockchain. By leveraging the memory-friendliness of incremental verifiable computation (IVC) and optimizing for blockchain environments, the proposed scheme can enhance scalability, reduce memory consumption, and ensure compatibility with common devices while supporting an arbitrary number of signature verifications. This approach allows for the concurrent generation of IVC proofs while receiving signatures from other nodes, making it particularly well-suited for low-latency blockchain applications.
As a concrete instantiation, they introduce for batch verification of the ECDSA signature. is implemented on an Intel i7-11800H CPU with 2.30 GHz, 8 cores, and 16 GB of RAM. Compare to , speeds up the prover by 3 ∼ 17 times and the verifier by 48 ∼ 240 times when handling up to ECDSA signatures, the maximum batch size supported by . For larger batches exceeding , outperforms the baseline approach, which verifies ECDSA signatures one by one without any proof. The verifier in achieved a speedup of 21 ∼ 174 times compared to the baseline as the batch size grows to . Furthermore, BEATS exhibits a remarkably low memory footprint, with peak memory usage remaining below 1 GB.
Future research directions include optimizing finite field selection for specific signature schemes and IVC implementations to minimize the computational overhead of non-native field operations, and exploring efficient applications of the resulting scheme in image authenticity verification, a critical concern in the age of AI-generated content.
DOI:10.1007/s11704-025-41269-5