Acceleration of Bioinformatics Applications Using Memristive-based PIM Architecture
Genomics, the study of genes, is making it possible to predict, diagnose, and treat diseases more precisely and personally than ever. Genome projects typically involve three main phases: DNA sequencing, assembly of DNA to represent the original chromosome, and analysis of the representation.
The recent advances in technology that allow high throughput genomic sequencing to be undertaken quickly and relatively cheaply have propelled the work of genome assembly and genome analysis forward.
Both DNA sequence assembly and analysis are data-intensive applications.
Thus, the regular von Neumann architecture, in which the memory and the computation units are separated, limits the performance of such applications as it demands massive data traffic between the memory and the CPU.
To reduce data movement and eliminate this limitation, processing-in-memory (PIM) platforms have been considered.
This talk presents two memristive PIM accelerators for both:
(1) DNA sequence assembly; specifically, for the read alignment task. the first PIM-based accelerator for the base-count filter is presented.
(2) DNA sequence analysis; specifically, the sequence classification task. We introduce the first scalable PIM-based accelerator for edit-distance-tolerant classification, that is also the first taxonomic classification solution based on memristive memory.
Both accelerators utilize the reduction in data transfer together with the parallel computation the PIM platform offers to accelerate the algorithms.
* M.Sc. student under the supervision of Professor Shahar Kvatinsky.