Regular Session III

on YouTube ↗42m

" summary

An analytical model uses network calculus to compute worst‑case reorder buffer size for arbitrary subflow counts and a new traffic‑splitting configuration (Abby Olitz) that saves buffer space: up to 53.2% vs uniform split, up to 20.9% vs prior work, average ~19.1% improvement while preserving worst‑case delay bounds. Separately, a domain‑specific many‑ported shared‑memory 32 MB prototype (4x4 clusters, 16 masters, 256‑bit AXI trunks, NoC @1 GHz, SRAM @500 MHz) demonstrates near‑100% throughput under 16 concurrent masters, consistent latency, <30 mm2 area and ~40% logic utilization.

AI-written summary · about

" chapters (5)

1. 0:00Start
2. 11:31Experiment and Result
3. 21:40The Background
4. 29:51The Micro-architecture - SRAM Array Organization
5. 32:19The Set-up and Parameters of the Prototype