Siemens AG organized a conference today on AI in the semiconductor industry — optimizing your production processes with industrial AI.
The speaker was Siggi Duell, Director Artificial Intelligence, Siemens AG. Johannes Roeck, Business Developer Electronics Industry, Siemens AG, was the moderator.
Johannes Roeck, Siemens, said that digital industries focus on customer needs with software, automation, and services. Siemens offers solutions across software, factory automation, motion control, process automation, customer services, etc. There are process, hybrid, and discrete industries that are served by Siemens.
The industry value chain starts from chip manufacturing for semiconductors, followed by surface-mount technology, and assembly and test component and devices, for electronics. We have a global set up to service partners.
Addressing challenges of industrial AI for semicon
Siggi Duell, Siemens, noted there is huge potential for AI in industries. We can get optimized quantitative and qualitative target for epitaxy. There is need to use AI on the shop floor. There are ML codes that are the heart and soul of AI apps.
There are more aspects needed for productive use of AI. These can be around data verification, machine resource management, connectivity and edge management, analysis tools, process management tools, model certification and verification, feature extraction, configuration, etc.
You can get and run AI models on the shop floor directly connected to automation. There are centralized tools for commissioning AI solutions. These are integrated to the existing automation infrastructure and tools. There is collaborative environment between automation and AI engineers. You can convert and optimize the model code to an executable AI. You can access live process data with standard shop floor. Hardware and software can run AI app. We need dedicated softwarte runtimes and AI hardware.
What does it take to get AI models to operate reliably on the shop floor? Dedicated AI models are integrated and designed for use case leveraging available domain knowhow and data. Data collection, preparation, and AI deployment is directly to the edge. There are managed and secure environments integrated to the automation assets. Siemens showed some use cases for AI optimized control for monocrystalline silicon production. It has AI-based mono-Si production control software.
When industry and AI meet, there is dynamic IT/OT environment, training data does not match productive data. There are mission-critical production environments. Direct transfer of models is difficult. Also, the data does not remain static over time.
With an industrial AI approach, you can get automatically notified when node performance drops. You can check data inputs and outputs continuously for detecting the data drift. Online assessment of AI reliability requires dedicated system to monitor the AI system. False calls on test equipment may create bottlenecks though, and increases NCC and production efforts.
We can utilize ML to reduce manual efforts significantly. The online assessment of AI reliabillity requires a dedicated system to monitor your AI system. Monitoring is essential to achieve desired results data in production.
In summary, we can reduce false calls in automated optical inspection, and have improvement of first pass yield. We can also ensure process control. We can have fully automated heating power control without human intervention. We can further have process optimization, especially in semiconductor fabs. We need to get AI-industry-grade operational on the shop floor.