Dynamic Simulation Enabled Prescriptive Analytics for the AI-Driven Autonomous Factory of the Future

In the AI-Driven Autonomous Factory of the Future as propagated by SEMI, both AI techniques and Digital Twins play an important role for capacity planning and WIP flow optimization. A Digital Twin of a wafer fab in particular, in order to be able to make predictions about the fab’s future evolution, should “represent” the fab and react to key drivers in the same way as the actual fab.
As an essential empowering technology of Digital Twins, stochastic simulation is used to portray the dynamic behavior of a wafer fab and to enable essential use cases where the interdependencies between capacity and cycle time are an important consideration such as
• Prediction of wafer moves to make sure that fab performance enhancement commitments can be made with higher confidence,
• Projection of fab cycle time performance with respect to various fab utilization profile to make sure that due date commitments can be kept,
• Projection of WIP profiles and prediction of bottlenecks in order to be able to invest in bottlenecks that give best cycle time,
• Determination of the impact of adding priority lots on delivery date to increase margins and enhance due date performance,
• Analysis of time link constraints and their impact on capacity and cycle time to maximize capacity without compromising yield.
To achieve this, such Digital Twins also need to “connect” and “synchronize” with the fab operations to get a real-time view of its state and detect changes in its underlying behavioral patterns to enhance the enabling simulation model (i.e., the quality of “represent”).
How this has been successfully realized will be showcased through the benefits that have been achieved by GlobalFoundries with the deployment and application of the D-SIMCON Dynamic Capacity and Material Flow Performance (DCMF) Planner. enabling these use cases.
The presentation will also elaborate how AI techniques will empower the critical step from currently predictive to prescriptive analytics of wafer fab operations by increasing the degree of automation when an AI optimization agent has to navigate through the mind-blowing complexity of the search space associated with wafer fab operations.
The simulation testbed enabled by the D-SIMCON DCMF planner can also be used for pro-actively training AI models such as Reinforcement Learning based dispatch agents which is currently being piloted in another wafer fab.