Ara 2 0 64 bit RISC V Vector Processor in 22nm FD SOI

on YouTube ↗22m

" summary

Ara 2.0 is a 64-bit RISC‑V crate‑style vector processor taped out in 22nm FD‑SOI and scaled from 4 to 16 lanes; a 4‑lane chip hits ~1.125 GHz and a 16‑lane instance reaches ~1 GHz delivering ~32 GFLOPS. Matrix‑multiply kernels reach within 2–3% of peak for large N but small matrices hit an unexpected bottleneck: the scalar core single‑issue dispatch and v0.5 vector ISA cause instruction‑issue starvation, not memory. Physical design shows lanes and the vector register file dominate area; energy efficiency measures around ~41 GFLOPS/W at the 16‑lane point.

AI-written summary · about

" chapters (8)

1. 0:00Introduction to Ara vector processor
2. 3:10RISC-V vector extension overview
3. 5:41Microarchitecture and lanes design
4. 9:35Matrix multiplication benchmark
5. 11:22Performance bottlenecks analysis
6. 15:19Physical implementation results
7. 17:30ISA limitations and future work
8. 20:46Q&A discussion