How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip

Description and use

OpenAI’s LLMs accelerated Jalapeño chip design by generating RTL and auto‑fixing errors, cutting development time.

Problem

Designing a custom chip (Jalapeño) required rapid, accurate RTL generation and verification, traditionally a time‑consuming, error‑prone process.

Outcome

The LLM‑driven workflow reduced RTL development time by ~70%, lowered bug count, and accelerated the Jalapeño chip’s time‑to‑market.

Workflow

• 1. Define high‑level functional spec in natural language.
• 2. Prompt OpenAI’s LLM to generate RTL code.
• 3. Run automated linting and design‑rule checks; LLM refines code.
• 4. Generate testbenches and run simulation; LLM adjusts RTL as needed.
• 5. Iterate until functional coverage and performance targets met.

Evidence

OpenAI’s LLMs generated RTL from high‑level specifications and auto‑fixed design‑rule violations.

Source