XNews

Siemens EDA Cuts Chip Design Agent Turns 66% Using NVIDIA NeMo Gym

Siemens EDA Cuts Chip Design Agent Turns 66% Using NVIDIA NeMo Gym

Siemens EDA used NVIDIA NeMo Gym to evaluate and refine its Fuse EDA AI Agent for standard-cell characterization, raising the composite verification score from about 55% to roughly 95%. Agent turns fell from 155 to 53 and output tokens dropped from about 81K to 29K after the team revised the agent's skills and Model Context Protocol tool descriptions based on verifier diagnostics.

Event version 1.

Key insights

Verification score jump

The Siemens Fuse EDA AI Agent's composite verification score increased from approximately 55% to ~95% across a benchmark of representative standard-cell characterization tasks.

Fewer turns and tokens

Agent turns decreased from 155 to 53 and output tokens fell from approximately 81K to 29K after refining skills and Model Context Protocol tool descriptions based on verifier diagnostics.

Domain-specific verifier

A domain-specific verifier checks coverage, validity, relative validation, accuracy, and integrity of final Liberty artifacts, assigning zero reward if any gate fails.

Error tracing across workflows

NVIDIA NeMo Gym provides an open evaluation environment that diagnoses failure modes and tracks error propagation across multistage chip design workflows.

Source posts