Three years building CPUs and spiking nets: one engineer's path to AI accelerators
k7agar · x · 2026-11-11
Engineer Shri publishes a three-year retrospective, "I just wanted to build accelerators," tracing her path from the GPT-2 era to building CPUs, memory, spiking networks and several small accelerators from scratch.
She lays out four long-term research directions:
- Design & verification: can an agent propose an accelerator, test it, understand failures, and improve the design?
- Iteration speed: making EDA tools fast enough to support that loop
- Access & manufacturing: easier PDKs and faster tape-outs
- New devices: what toolchains must understand if memristors become practical building blocks
A personal narrative blended with a serious research agenda on agent-driven chip design.
More from Infra
- Starlink urban US users grow 73%, double rural rate, as upload speeds jump 42% — XFreeze · 2026-11-11
- Cloudflare picks QuePaxa, an unproven consensus algorithm, for Meerkat service — sull · 2026-11-11
- Using AI to find every abandoned mall in America and turn them into data centers — LinusEkenstam · 2026-11-11
- Anthropic's $26B IPO private placement: seven suppliers buying in, AMD deal tied to GPU delivery — demian_ai · 2026-11-11
- SpaceXSI secures Mississippi air permit for 1.2 GW gas plant powering Colossus II — XFreeze · 2026-11-11
- Anthropic says AI cut an AMD chip switch from months to 3 days, eroding Nvidia's CUDA moat — The Decoder · 2026-11-11