Synopsys said on 30 September that it and OpenAI had signed a multi-year agreement to develop GPT-Synopsys, a model for semiconductor design, under a revenue-sharing arrangement.
Key points
- OpenAI will license Synopsys electronic design automation tools to develop the model.
- The companies plan to sell a service bundling computing capacity, the model and tool licences.
- GPT-Synopsys will run on OpenAI-hosted infrastructure and connect to Synopsys.ai and Synopsys Autopilot.
- Synopsys says early technology engagements with semiconductor customers are under way.
OpenAI will license Synopsys design tools
The companies agreed to work as preferred partners on GPT-Synopsys, according to Synopsys. OpenAI will license Synopsys electronic design automation, or EDA, tools to develop the model. The agreement also covers joint research and development and a shared effort to bring the service to market.
Synopsys says the planned customer offering will combine computing capacity, the model and licences for its tools. The revenue-sharing arrangement gives both companies a commercial role beyond developing the technology: OpenAI is providing the hosted model, while Synopsys is supplying the design tools the model is intended to operate.
GPT-Synopsys is intended for semiconductor design workflows rather than general-purpose assistance. Synopsys says it will combine OpenAI models with its tools and chip-design expertise so that the model can work on design and verification tasks directly.
Synopsys has separately expanded its collaboration with TSMC on A14 EDA flows, agentic AI and 2nm intellectual property, AI Affairs reported. The OpenAI agreement centres on developing a specialised model to use Synopsys tools.
GPT-Synopsys targets design and verification
Synopsys says existing agentic systems connect general-purpose models to EDA tools to carry out chip-design workflows. With GPT-Synopsys, the companies plan to develop a model trained to use the tools more directly, interpret their outputs and revise designs through repeated runs.
The intended work covers power, performance and area optimisation, as well as timing and verification closure, according to Synopsys. Its proposed agents would run tools, assess the results, make changes and continue towards outcomes that engineers could review. Those capabilities are plans for the model, rather than results from customer deployments.
GPT-Synopsys will run on infrastructure hosted by OpenAI, Synopsys says. It is also being designed to work with customers’ own agent harness systems and to integrate with Synopsys.ai and Synopsys Autopilot, the company’s agentic AI platform.
Sassine Ghazi, Synopsys’ president and CEO, said the agreement would give more companies access to its design capabilities and engineering tools. Greg Brockman, OpenAI’s president and co-founder, said the companies aimed to help engineers examine more designs and reach a working chip faster.
Synopsys sets out customer data controls
Synopsys says early technology engagements with semiconductor customers are under way. The company describes those engagements separately from the planned joint service, which would bring together hosted computing capacity, GPT-Synopsys and licences for its EDA tools.
Customer-specific designs are part of the proposed workflow. Synopsys says customer data will not be used to train GPT-Synopsys and will be encrypted both while stored and while being transferred. Customers will be able to manage retention, audits and permissions through configurable controls, according to the company.
The company also says the service will include access controls and governance features. Those provisions matter for a product designed to operate within engineering workflows: the model is intended to handle design objectives, read tool outputs and propose changes for engineers to review.
Synopsys classifies statements about GPT-Synopsys’s development, capabilities, timing, availability and anticipated benefits as forward-looking. It says risks and other factors could cause actual results or time frames to differ materially from those described in its announcement.
