Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulationSiemens Fuse EDA AI agent10x reduction in turnaround time; 10x reduction in token costs
The agent autonomously orchestrated the silicon characterization workflow, invoking tools via skill files and MCP servers to generate liberty files, reducing manual babysitting of jobs and handling failures automatically.
“we're able to reduce the turnaround time by 10 X and reduce the token cost by 10 X. So dramatic improvement and just one example, right? So across the entire EDA workflow … you can imagine that same kind of workflow … applying applying across the board.”
Engineering & simulationNVIDIA CUDA X libraries / GPU accelerated solvers5x to 10x faster SPICE simulation
GPU acceleration was applied to the underlying simulation engines (specifically SPICE simulators) to speed up the computational loops within the engineering workflow, allowing the agent to perform high-performance engineering tasks faster.
“So, so now what the, what that flow looks like is the engineer is using reinforced learning accelerated tools, uh, within the engines themselves, the simulators are GP you accelerate it. So you're getting the advantage of five, seven, 10 X faster spice simulation through GPU acceleration.”
Engineering & simulationNVIDIA NeMo Gym
NeMo Gym provided a controlled environment to measure agent efficiency and reliability, converting these scores into reward signals for reinforcement learning to improve the agent's performance over time without altering live tapeout processes.
“And then the Nemo gym environment, what happens is the agent takes actions and the engineering workflow produces measurable outcomes … Those results can be scored by a verifier. during evaluation, the scores show us what needs to improve. … converting them into repeatable evaluation and training cases, validating the improvement, and then promoting it into production through a controlled process.”