Pip-SW: Pipeline Architectures for Accelerating Smith-Waterman Algorithm on FPGA Platforms

The Smith-Waterman algorithm, which is founded on a dynamic programming approach, serves as a precise tool for aligning biological sequences. Despite its utility, the algorithm grapples with computational complexity and resource demands. Various implementations across multi-core, GPU, and FPGA platforms have sought to expedite the algorithm, yet frequently encounter issues such as suboptimal speedup, heightened reliance on external memory resources, and an exclusive focus on the forward step of the algorithm. To tackle these challenges, this study introduces an architecture aimed at accelerating the Smith-Waterman algorithm on FPGA platforms. Our architecture capitalizes on a pipeline structure that integrates optimized circuitry for parallel computations and employs memory allocation techniques, thus delivering an efficient, low power and cost-effective implementation for biological sequence alignment. Our assessments, coupled with comparisons against alternative FPGA implementations supporting protein sequence alignment, reveal a 17% increase in operating frequency and a 17% enhancement in Giga cell updates per second. Moreover, our approach competes with GPU-based solutions, showcasing comparable performance metrics alongside superior energy efficiency, with a 35% improvement. We substantiate the utility and performance of our pipeline architecture on FPGA platforms using four benchmark datasets. The validation results demonstrate a speedup ranging from 10 to 45 times for alignment score computation compared to the CPU platform.