Cadence and Nvidia release Maestro AI models for chip design

Cadence makes software for chip design. This week, it told of a partnership with Nvidia and released Maestro, a family of AI models for chip design. Cadence fine-tuned the models from Nvidia Nemotron models, to work with Cadence tools and read their results. Paul Cunningham of Cadence tells that AI agents help with 20 to 30 percent of the chips that are in chip design at this time.

How it works

Maestro works with the AI Super Agents. These are control software that use a language model to operate the chip design and test tools of Cadence. The first agents used general models. Those models had a small quantity of data about chip design code. Customers get Maestro through an API (a procedure for software to connect), in a private cloud, or on servers of customers.

The numbers

Cadence tells that first tests showed a maximum of 30 percent more output quality. It also tells of a maximum of 5x token efficiency against general models. A token is a unit of data that a model reads or writes. These numbers are from Cadence. The source gives no test data and no test from a different source.

What is not known

The source gives no cost data. Customers can adapt Maestro with data of customers. Cunningham tells that Cadence gives this more as a managed service than as a tool for customers. Cunningham tells that general models are the first step and that customers can adapt them.

Insight: Nvidia open models become the first layer in vendor AI

Cadence made Maestro from Nvidia Nemotron models. Trend AI put Nemotron in its security platform for data centers of customers. On 21 October, Nvidia released open Nemotron 3.5 models for other vendors to use. This is a possible move to make Nemotron the first layer in vendor AI, which gives Nvidia more control than its chips give. Find out which model is in the AI of each vendor.