An Energy-Efficient CPU–FPGA Heterogeneous Acceleration System for Third-Generation Genomic Sequencing Based on Minimap2

Minimap2 has become a widely used software for third-generation long-read genomic sequencing. Due to the increasing complexity of data processing with long-read sequences, the analysis is computationally intensive and energy-consuming. This brief presents an end-to-end CPU-FPGA heterogeneous acceleration system for Minimap2 focusing on chaining operation, in which multi-threaded software is on the CPU, and a multi-kernel accelerator for chaining operation is on the FPGA. This brief can hold the high thread number of modern CPUs to maximize performance and energy efficiency. Hardware-efficient kernel design, software-hardware co-optimization, and memory access fusion techniques have been applied to achieve higher computational performance with less power consumption. This brief achieves at most $2.01\times $ acceleration against software and $1.65\times $ against the baseline, and EDP reduction of 72% against software and 59% against the baseline, outperforming state-of-the-art designs. The code of our acceleration system is available on GitHub, together with FPGA bitstream.