Leveraging FPGA Runtime Reconfigurability to Implement Multi-Hash-Chain Proof-of-Work

In the cryptocurrency mining field, algorithms have been developed to frustrate the development of ASICs that greatly out-compete general purpose hardware in both performance and power efficiency. A class of algorithms that claims to be ASIC-resistant is randomized multi-hash-chain proof-of-work algorithms, such as X16R. For these algorithms, the result of one iteration depends on the chained application of several randomly selected hash functions, which disadvantages fixed-function ASICs due to their inflexibility. FPGAs lie between GPUs and ASICs in terms of raw performance and flexibility. We investigate the usage of FPGAs for this type of proof-of-work, in particular, by leveraging the ability of modern FPGAs to quickly reconfigure at runtime. We implemented a proof-of-concept design that runs the X16R algorithm by partially reconfiguring the FPGA for every hash function in the chain and processing the data in batches. We show that our system achieves better performance when compared to GPUs from the same semiconductor process technology node, while being nine times more power efficient.