Airborne Molecular Contamination (AMC) is a critical issue for numerous state of the art manufacturing processes. These contaminants degrade features and effect yield in the semiconductor manufacturing process, where feature sizes continues to shrink.

AMC chemical filters play a pivotal role in removing AMC contaminants and provide purified air to SEMI cleanrooms, which is a key requirement in generating quality products. Long filter lifetime is desired, and design optimization plays a central role in filter development. Filter removal efficiency (RE) and life cycle are critical design parameters. However, it is challenging to obtain removal efficiency experimentally, particularly at very low gas concentrations, due to long test time, high experimentation cost and logistical limitations. Although computational modeling is an efficient means to generate predictive models, long computational lead time and inability to capture a live Failure Mode Engineering Analysis (FMEA) analysis are shortcomings.

The digital twin (DT) is a machine learning model which is a virtual representation of real word entities and processes, synchronized at a specified frequency and fidelity. It can track the past, provide insights into the present and predict and influence future system behavior and can offer live, system-level simulated prediction with real-time data inputs. Therefore, DTs offer a unique opportunity to study virtual and physical systems, either separately or together.

This study was a pioneering effort to successfully develop and deploy Digital Twin technology in the AMC filter optimization process. A static and dynamic Digital Twin were developed from initial validated Computational Fluid Dynamics (CFD) model of a specific AMC chemical filter. The newly developed Twin model was able to predict the filter performances instantly along with live FMEA by its developed virtual sensors which critically can help filter optimization process by reducing cost of ownership significantly...