High-Performance Confidentiality-Preserving Blockchain via GPU-Accelerated Fully Homomorphic Encryption

Data confidentiality is essential for safety-critical blockchain applications such as digital payment. A promising approach for preserving confidentiality is to encrypt transaction data using homomorphic encryption (HE) and prove the correctness of transaction execution through non-interactive zero-knowledge proofs (NIZKPs). However, prior work on this approach suffers from poor performance caused by the costly HE computation, hindering their adoption for real-world applications. In addition, prior work is restricted by the use of HE schemes that only support either addition or multiplication, making it challenging to implement business logic involving both types of arithmetic operations. We present Gafe, a high-performance confidentiality-preserving blockchain that carries a GPU-accelerated transaction execution workflow. Gafe encrypts transaction data with FHE, allowing both addition and multiplication on ciphertexts. For high performance, Gafe leverages parallel execution on GPUs to accelerate FHE computations. For result correctness, Gafe generates lightweight NIZKPs that incur low overhead. Evaluations show that Gafe is highly performant, achieving a 3.1 \(\times \) increase in throughput (258 transactions per second) and a 37% reduction in latency (1.61 s), surpassing the baseline without GPU acceleration.(Gafe stands for GPU-Accelerated Fully Homomorphic Encryption Blockchain.)