Silicon Volley: Designers Tap Generative AI for a Chip Assist | NVIDIA Blogs

NVIDIA researchers released a paper describing ChipNeMo, a custom generative AI model trained on the company's internal chip design data to assist semiconductor engineers. The team, led by Mark Ren, an NVIDIA Research director, chose three initial use cases: a chatbot, a code generator and an analysis tool that automates maintaining descriptions of known bugs. The analysis tool has been the most well-received so far. Ren said custom ChipNeMo models with as few as 13 billion parameters match or exceed much larger general-purpose models such as LLaMA2 with 70 billion parameters on chip-design tasks.