Genome analysis is crucial for applications such as personalized medicine and disease tracking. Sequencing technologies generate single- and paired-end reads, which are aligned to a reference genome through a process known as read mapping. This process is computationally expensive and remains a significant performance bottleneck in genomic analysis. Prior works, including hardware accelerators and software filters, aim to improve read mapping performance. However, challenges persist due to 1) memory bottlenecks arising from the shift in compute-intensive to memory-intensive operations introduced by the combination of filters and hardware accelerators and 2) the poor performance of existing filters with paired-end reads.

We propose GenPairX, a hardware-algorithm co-designed accelerator to enhance paired-end read mapping performance. GenPairX introduces a memory-efficient, map-based filtering approach and a hardware-friendly lightweight alignment algorithm to reduce computational complexity. Our evaluations demonstrate that GenPairX achieves substantial performance gains over state-of-the-art solutions, delivering 1549$\times$ higher throughput per power unit compared to leading CPU-based and 1.41$\times$ compared to hardware-based read mappers, all while maintaining or exceeding accuracy levels.