Genome analysis benefits precise medical care, wildlife conservation, pandemic treatment, e.g., COVID-19, and so on. Unfortunately, the speed of data processing in genome analysis lags far behind the speed of data generation and hardware acceleration turns out to be necessary. As many key applications in genome analysis are memory-bound, the computation-centric accelerators, e.g., CPU, GPU, and FPGA, face the challenges of limited memory bandwidth and frequent data movement, which leads to sub-optimal performance and energy efficiency.

To address these challenges, this dissertation focuses on exploring efficient memory-centric accelerators for genome analysis, including designs and optimizations in both hardware and software. This dissertation proposes four memory-centric accelerators for genome analysis, covering both the emerging memory technology, i.e., ReRAM, and the conventional memory technology, i.e., DRAM. By performing in-situ computation inside the emerging memory array to leverage massive parallelism and eliminate data movement, the proposed emerging memory technology based design provides ultra-high performance and energy efficiency. As a comparison, by integrating the processing elements near the conventional memory array to utilize extra memory bandwidth and reduce data movement, the proposed conventional memory technology based designs highlight practicality and cost-effectiveness without making any modifications to the cost-sensitive DRAM dies. Multiple key applications in genome analysis are covered in this dissertation, including k-mer counting, DNA seeding, and DNA pre-alignment.