OpenAI has entered into a multi-year agreement with software developer Synopsys to co-create a specialized model, GPT-Synopsys. The new model will be able to independently control professional electronic design automation (EDA) tools, interpret results, and improve semiconductor architectures.
OpenAI has entered into a multi-year agreement with software developer Synopsys to co-create a specialized model, GPT-Synopsys. The new model will be able to independently control professional electronic design automation (EDA) tools, interpret results, and improve semiconductor architectures.
The main goal of the partnership is to accelerate the development process of new chips from the initial task to production readiness, while maintaining the highest requirements for accuracy, reports Interesting Engineering.
Unlike conventional language models, GPT-Synopsys is created as a full-fledged "expert user" of chip design software.
Engineers will set the system's target parameters for energy efficiency, performance, area (PPA), timings, or verification. AI agents will independently run EDA tools, analyze the data received, make changes to the design, and repeat the cycle until the desired result is achieved. The final architecture options will be approved by design engineers.
The model will run on OpenAI’s cloud infrastructure and integrate with the Synopsys.ai platform and the new Synopsys Autopilot agent service. According to Synopsys, initial tests of such agent technologies have accelerated the chip verification process by up to 50 times and increased developer productivity by 30%.
The agreement involves joint development and commercialization of the product for customers around the world. Synopsys emphasizes that customers' confidential data about their chips will not be used to train the model: the information will be fully encrypted, and companies will receive flexible access control and auditing tools.
This collaboration complements OpenAI’s own ambitions in semiconductors. The company previously unveiled Jalapeño, its first custom AI inference processor, developed in collaboration with Broadcom. With partial use of OpenAI models, the company went from concept to silicon prototype in just nine months.
