MARS: Processing-In-Memory Acceleration of Raw Signal Genome Analysis Inside the Storage Subsystem

The preprint operates within the established paradigm of genome analysis, which traditionally involves basecalling followed by read mapping. However, it addresses the limitations of this paradigm by proposing in-storage processing of raw signals, representing a departure from the conventional approach of relying on CPU/GPU-based acceleration of individual steps within the analysis pipeline.

The preprint presents MARS, a system for accelerating Raw Signal Genome Analysis (RSGA) by leveraging in-storage processing capabilities of SSDs. It addresses the growing I/O bottleneck in genome analysis pipelines caused by increasing sequencing throughput. While RSGA itself is an emerging area representing a shift away from traditional basecalling, the preprint focuses on optimizing RSGA within the context of read mapping using novel hardware/software co-design. MARS accelerates established RSGA algorithms by integrating them with in-storage computational units, representing a model drift within the broader RSGA paradigm. While the preprint doesn't propose fundamentally new algorithms, it introduces architectural and system-level optimizations to improve performance and energy efficiency, making it a model drift. The two closest classifications are normal science, since MARS aims at solving a well-defined problem (I/O bottleneck in RSGA) within the existing RSGA paradigm, and model drift, since MARS integrates RSGA processing into the storage system to tackle the identified I/O bottleneck, representing a notable optimization but not a paradigm shift.