Tajana Rosing

UC San Diego PI leading CRISP-funded processing-in-memory accelerator research. Senior author on both RAPID (ISLPED 2019) and RAPIDx (IEEE TCAD 2023, extending RAPID), the two peer-reviewed PIM DNA-alignment papers this vault used to check whether CRISP's press-claimed "20 hours to less than a second" speedup has a primary source. It does not: her papers' speedups against real state-of-the-art baselines run 2×–321× (far larger ratios — up to 9.6×10⁶× — appear only against a single-threaded CPU on synthetic sequences, not the real-chromosome workload), and RAPID's measured chromosome-wide alignment time of 18 minutes is itself roughly three orders of magnitude short of the press figure's implied magnitude.

References

• claim-rapid-rapidx-speedups-fall-short-of-crisp-press-figure
• Related hubs: entity-crisp-center
• Capture: 2026-07-30-does-a-primary-source-confirm-crisps-20-hours

Updates

• 2026-09-15 — propagation-repair. The lede's speedup range was rescoped. Was: "her papers report real 2×–321× speedups and an 18-minute measured chromosome-wide alignment time, roughly three orders of magnitude short of the press figure's implied magnitude" — an unqualified upper bound of 321×. Now: the 2×–321× range is scoped to comparisons against real state-of-the-art baselines, with the far larger ratios RAPID also reports (up to 9.6×10⁶× vs a single-threaded CPU on a synthetic 10-million-base sequence; 1585× at length 1000) named as synthetic-baseline results, and the three-orders-of-magnitude gap re-anchored on the baseline-independent measured wall-clock (1081 s ≈ 18 min for real chromosome-1 alignment). Propagates the AUDIT 2026-07-31 cross-model correction (auditor claude-opus-4-8) on claim-rapid-rapidx-speedups-fall-short-of-crisp-press-figure. Unchanged: her senior authorship on both papers, the absence of any "hours to a second" framing in either, and the conclusion that no primary of hers supports the press figure's magnitude.