Cadence fine-tunes Maestro models on Nvidia’s Nemotron

Cadence fine-tuned its Maestro models on Nvidia’s Nemotron models and designed them to work with and interpret results from Cadence’s own tools.[1] Cadence reported those figures from early testing.[1] Cadence’s early testing found up to 30% higher output quality for Maestro, as well as up to five times greater token efficiency compared with general-purpose foundation models.[1] The evidence does not specify how those results were measured or which tasks were tested.[1]

Cadence says customers can access Maestro models through an API, run them in a secure virtual private cloud or run them on-premises if they have the necessary hardware.[1] The company’s platform can be customized using customer data, and Tom’s Hardware reported that customers can post-train the models on their own designs and data.[1][2] Cadence executive Paul Cunningham said customization would be offered more as a managed service than as a do-it-yourself process.[1] For teams assessing a deployment, the available evidence therefore points to multiple hosting options but a service-led customization model.[1]

Cadence’s AI agents are designed to automate design and testing flows by coordinating the company’s existing electronic-design-automation, test and verification tools.[1] Its lineup includes AuraStack for advanced packaging and PCB layout, ChipStack for system-on-chip design and verification, InnoStack for PHY routing, and ViraStack for analog circuit design.[1] Cadence senior vice president Paul Cunningham estimated that AI agents are helping design 20% to 30% of chips currently under development.[1] That estimate describes current use of agents, not a measured result for Maestro specifically.[1]

Cadence’s earlier agents were largely powered by off-the-shelf large language models, which limited their exposure to register-transfer-level design.[1] The effectiveness of AI design agents’ tests depends heavily on how closely their training data resembles the inputs and outputs they receive.[1] Maestro’s focus on Cadence tools and the option to customize models with customer data make those details relevant to teams evaluating the system.[1] The evidence does not establish that customization will improve results for any particular customer’s designs or workflows.[1]