Semicon West 2024 was held on July 9-11 in San Francisco, and Tokyo Electron Limited (TEL), the Japanese semiconductor equipment supplier, unveiled the world's first open-source generative AI Small Specialist Agents (SSA) platform that is specifically for the semiconductor industry during a keynote.
The semiconductor industry SSA is called "SemiKong", an AI agent platform based on an industrial large language model (LLM). The speaker said its prototype will be freely available worldwide, and is expected to revolutionize the US$500 billion semiconductor manufacturing market in five years. This is also an important step taken by the AI Alliance, the main promoter of open-source, responsible generative AI platforms.
The enthusiasm generated by OpenAI's ChatGPT gave birth to the concept of AI Agents, which are generative AI models that not only can search for information or answer the questions of human beings but also can take further steps to think, plan, and propose solutions.
However, AI Agents are still susceptible to the weakness of generative AI. For example, it is prone to bigotry or errors due to insufficient or biased material input for learning. SSA aims to enhance the learning process in specific knowledge domains and industries based on a valid and verified LLM to lower the probability of errors to meet corporate customers' demands.
Nikan Kogyo Shimbun reported that open-source software can further enhance the strengths of SSA. For example, semiconductor companies including Axcelis Technologies, NXP Semiconductors, and MaxLinear participated in the development process of SemiKong, so the SSA can acquire large amounts of learning materials very quickly.
Compared to closed AI from leading GenAI operators, open-source SSA combines the advantages of broad support and protection of corporate confidentiality. With the increase in the number of partner companies and timely platform upgrades, the number of specialized knowledge domains in which the open-source SSA platform can be applied will continue to expand.
