BRAWL: A Spintronics-Based Portable Basecalling-in-Memory Architecture for Nanopore Genome Sequencing

Qian Lou, Lei Jiang

Abstract: Nanopore sequencing is one of the most promising genome sequencing technologies because of its ability to produce ultra long reads and provide portability. Basecalling, the most time-consuming step in the whole flow of Nanopore genome sequencing, translates analog signals to digital DNA symbols. The state-of-the-art basecaller relies on a complex neural network consisting of convolutional, long short-term memory and fully-connected layers, and a CTC decoder. Existing neural network portable accelerators achieve low basecalling throughput per Watt when processing such neural network inferences. In this paper, we propose BRAWL, a portable <b>B</b> asecalling-in-memory architecture, to translate <i>RAW</i> electrical signa <b>L</b> s to digital DNA symbols in SOT-MRAMs for Nanopore portable sequencers. Compared to state-of-the-art accelerators, BRAWL improves basecalling throughput per Watt by <inline-formula><tex-math notation="LaTeX">$3.88\times$</tex-math></inline-formula> .