Earlier this month, Aitomatic and its partners within the Foundation Models workgroup of the AI Alliance introduced a groundbreaking artificial intelligence (AI) model tailored for the semiconductor industry. Named SemiKong, this large language model (LLM) leverages a domain-expert agents (DXA) architecture and is trained on high-quality data relevant to semiconductor development. The innovative model aims to assist professionals throughout the entire spectrum of semiconductor device and process development.
SemiKong Can Reduce Chipset Design Time by Up to 30 Percent
A recent blog post by Meta detailed the functionalities of the new AI model. Developed by the Palo Alto-based Aitomatic in collaboration with various partners from the AI Alliance, SemiKong is focused solely on the semiconductor sector, incorporating data related to design, manufacturing, fabrication, and development processes.
This open-source AI model is available on GitHub and Hugging Face, distributed under the Apache-2.0 license for both individual and commercial applications. SemiKong features a bilingual language model trained on three trillion multilingual tokens and is offered in four variants: SemiKong-8B, SemiKong-70B, SemiKong-8B Instruct, and SemiKong-70B Instruct.
Christopher Nguyen, CEO of Aitomatic, stated that this AI model is designed to address the existing knowledge gap within the semiconductor industry, especially as many experienced professionals have retired in recent years without equivalent knowledge transfer. Nguyen emphasized that SemiKong could foster innovation and collaboration, enabling the industry to integrate AI in streamlining essential manufacturing and operational processes.
SemiKong utilizes a neurosymbolic agentic AI architecture known as Domain-Aware Neurosymbolic Agents (DANA). This architecture employs a three-pronged strategy: structuring expert knowledge, enhancing human expertise with synthetic information to train DXAs, and linking these trained DXAs to the manufacturing execution systems of semiconductor companies to facilitate automated technical analyses and decision-making.
According to Aitomatic’s internal evaluations, SemiKong has the potential to reduce time-to-market for new chip designs by 20 to 30 percent. The model is also reported to improve the “first-time-right” rates in chip manufacturing by up to 25 percent, in addition to expediting the onboarding process for young professionals by 40 to 50 percent.
