Interesting semiconductor story here. TetraMem says it successfully taped out and validated early silicon for a 22nm analog in-memory computing SoC using multi-level RRAM on a commercial TSMC process. The bigger idea is reducing AI power consumption and memory bottlenecks by performing compute operations directly inside memory arrays instead of constantly shuffling data between RAM and processors. We have heard memristor and analog computing promises for years, so skepticism is warranted, but this appears to be actual silicon rather than another research paper or simulation. Could be worth watching as GPU power and cooling demands keep spiraling upward.