Darwin: A Genomics Co-processor Provides up to 15,000X Acceleration on Long Read Assembly

on YouTube ↗19m

" summary

Describes Darwin, a genomics co-processor that accelerates long-read sequence assembly by combining hardware-accelerated implementations of two algorithms: D-SOFT (a memory-efficient filtering algorithm) and GACT (an alignment algorithm that produces near-optimal alignments of arbitrarily long sequences using constant compute memory). Explains a hardware/algorithm co-design that shifts work toward faster alignment and lighter-weight filtering, plus memory-system optimizations, to achieve up to 15,000× speedup for reference-guided assembly of third-generation sequencing reads while allowing tunable speed–sensitivity tradeoffs.

AI-written summary · about

" chapters (18)

1. 0:00Start
2. 1:52Domain-Specific Acceleration in Machine Learning
3. 2:28Landscape of computing
4. 3:29DNA Sequencing, Errors and Alignment
5. 5:39Long Reads: Holy Grail of Sequencing
6. 6:10Darwin: Overview
7. 6:45Sources of acceleration
8. 8:09Evolution of Darwin (ASIC) from Graphmap
9. 9:192. Specialized Operations
10. 10:15Reduced Overhead
11. 10:48Inner-loop parallelism
12. 11:34Co-design can enable 4. Local memory
13. 12:203. Outer-loop parallelism
14. 12:59D-SOFT Algorithm: Key Points
15. 14:43Read assembly on Darwin
16. 15:25Darwin: Assembly results
17. 15:38Darwin: Paper Awards
18. 17:41Conclusion