LLMs for Chip Design

Custom chips can run specialized tasks far faster and with less energy than CPUs, but are costly to design. YC wants LLMs that cut the cost of designing ASICs and FPGAs.

YC argues that each AI breakthrough raises demand for more powerful chips, and that domestic chip design and manufacturing is now about survival, not just economics. It cites OpenAI's o1 as showing that LLMs with reasoning can power breakthroughs in science and engineering.

YC is most interested in ASICs and FPGAs. Custom digital systems like these have been costly because of the custom design, development and testing they need. Most computing uses general-purpose CPUs because they are easy to reprogram. But for specialized tasks such as cryptocurrency mining, data compression or special-purpose encryption, YC says optimized hardware can run 5x to 100x faster and use 10x to 100x less energy.

The ask: startups using LLMs to improve chip design and bring those costs down, so more specialized kinds of computation become practical.

Read the full request on YC →