System and Circuit of Compute-in-Memory Design for Genome Alignment

Next-generation sequencing (NGS) technologies have revolutionized genomics research because they enable faster, cheaper and accurate decoding of nucleotide sequences. This has resulted in breakthroughs in diagnostics, disease risk assessment, personalized medicines, screening and more. However, despite all the NGS-related advancements that have been made, the current bottleneck lies in the memory- and compute-intensive nature of large-scale genome processing. Existing algorithms take several hours to days to process the genomic data, even when using advanced computing architectures.

Researchers at Arizona State University have developed a resistive random access memory (RRAM)-based Compute-in-Memory CIM macro design which is tailored for genome processing. This design supports the state-of-the-art Burrows-Wheeler Transform (BWT) based DNA short read alignment. This CIM design is optimized to support the major functions essential to the algorithm and is the most energy-efficient solution for genome processing. This innovative CIM macro design not only leverages the high parallelism of CIM architecture, but also keeps the low complexity of the circuit making complex genome processing tasks more efficient and less resource-intensive.

This technology represents the first RRAM CIM chip that is able to accelerate genome sequencing alignment, and shows orders of magnitude improvement in energy efficiency and throughput.

Potential Applications

RRAM-based COM for genome processing
- Diagnostics, disease risk assessment, personalized medicine, screening, etc.

Benefits and Advantages

This CIM RRAM macro is able to execute all essential functions required for alignment – NXOR based match, MEM, count and addition
Orders of magnitude improvement in energy efficiency with superior throughput over CPUs/GPUs and prior non-CIM designs
Could work independently as a parallel ‘alignment core’ that could process local correlated reference genomic data to improve system parallelism and throughput
Flexible enough to support both 1- and 2-bit per cell encoding of nucleotides
The memory and compute aren’t treated as isolated elements but are interleaved for higher throughput and reduced data traffic
Leverages the high parallelism of CIM architecture but keeps the low complexity of the circuit to make complex genome processing tasks more efficient and less resource-intensive

For more information about this opportunity, please see

Zhang et al - IEEE ESSCIRC Conference - 2023

For more information about the inventor(s) and their research, please see

Dr. Fan's departmental webpage