Artificial intelligence continues to evolve, with specialised models pushing the boundaries of industry-specific innovation. Just as foundational models like GPT, ADA, DALL·E, DAVINCI and Llama have redefined general-purpose AI, tailored solutions are now addressing niche sectors with precision. Aitomatic, in collaboration with the AI Alliance, has introduced SemiKong-a first-of-its-kind open-source AI model specifically designed to transform the semiconductor industry.
Built on Meta’s Llama 3.1 70B, SemiKong has been carefully fine-tuned using semiconductor-focused data, including research papers, anonymised manufacturing insights, and design documentation. This targeted training enables the model to excel in understanding and generating content for applications like chip design, manufacturing optimisation, and testing. Available in multiple variants SemiKong-8B, SemiKong-70B, SemiKong-8B Instruct, and SemiKong-70B Instruct, SemiKong is openly licensed under Apache-2.0, ensuring accessibility for both individual developers and commercial enterprises. It is available on GitHub and Hugging Face for you to try. By capturing domain-specific expertise, SemiKong not only bridges a growing knowledge gap as industry veterans retire but also accelerates innovation, potentially reducing chip design time by up to 30% and improving manufacturing success rates.
At the heart of this initiative is Aitomatic’s Domain-Expert Agent (DXA) framework, which leverages SemiKong and lightweight Llama models to automate and enhance complex processes. These agents combine structured human expertise with synthetic AI knowledge to tackle challenges such as manufacturing yield monitoring and predictive maintenance. As Aitomatic's Christopher Nguyen highlights, the open-source approach fosters collaboration and makes AI-driven advancements in semiconductor technologies, empowering organisations to innovate and scale without compromising control over intellectual property.
