AI by definition is any technique that enables computers to mimic human intelligence, using logic, if-then rules, decision trees, machine learning, deep learning, and generative AI techniques. AI applications have been successfully deployed in semiconductor manufacturing over the last 25+ years for process control and manufacturing variability reduction (SPC, APC, R2R, OCAPS, etc.). This paper describes those applications, new capabilities, and how they are being incorporated in Smart Manufacturing and Enterprise level control systems using modern AI and digital twins to maximize ROI (Return on Investment) and enhance process control at all levels in semiconductor manufacturing.
The challenge is understanding how to incorporate current solutions in an intelligent AI driven factory/enterprise platform that leverages existing capabilities within a framework to enable deployment from the digitized world to the physical realm of manufacturing to achieve improvements in manufacturing efficiency, cost reduction, variability reduction, quality, yield, and work force enablement: 1. Common connectivity platform leveraging current and emerging SEMI standard as well as streaming and custom containerized data for direct AI/ML applications. 2. Automated system that can effectively connect manufacturing and supply chain data together in a digital twin with minimal human involvement to enable faster decisions and better control systems at all levels of automation (equipment, factory, and enterprise). 3. Ability to automatically create and deploy AI and Machine Learning models with high accuracy and deploy to an edge control platform to enable leveraging factory data for Al/ML applications 4. The ability to link the many types of AI/ML models in a unified framework across all levels of manufacturing (equipment oems, supply chain, metrology, fab, osats, enterprise) using LLM’s and GenAI to provide a common user interface while protecting OEM and customer IP 5. Architecture that enables high performance processing to execute all of the above at the required levels of automation while maintaining Root of Trust security and Traceability for all materials and components in the supply chain.
Semiconductor manufacturers and equipment OEM’s have built a wide variety of AI and Machine Learning solutions that have been effective at reducing variability, costs, and improving yield. However, there are many AI tools specifically designed for point solutions or only available on specific equipment manufacturers tools. The innovation in the presented approach is the utilization of GenAI LLM’s to link the outputs of each AI model through LLM and interactive Agentic AI agents. This solution solves the IP protection and security concerns and enables information/results sharing as well as reinforcement learning and collaborative feedback across these AI siloes for continuous learning.
This automated system can effectively connect manufacturing and other supply chain data together in a digital twin leveraging subject matter experts to ‘teach’ the model to operate with minimal human involvement to enable faster decisions and better control systems. Results show >50% improvement in engineering efficiency.
