Virtual Metrology Predictive Solutions for Semiconductor Manufacturing
Industry
Semiconductor manufacturing
Asset
Semiconductor manufacturing equipment
Goal
Develop a virtual metrology predictive solution able to learn the relationship between sensor variables and metrology variables and estimate metrology variables, improving product quality and diminishing the need for costly metrology measurements
Impact
Delivered a virtual metrology solution that ensures product quality and reduces the need to perform costly measurements, saving significant time and money
Overview
In semiconductor manufacturing, costly metrology measurements must be conducted throughout the process to verify the properties of the wafers produced. Predictronics worked with a semiconductor manufacturing equipment company to develop a solution capable of learning the relationship between the sensor and metrology data. This presented many challenges, as semiconductor equipment is complex with hundreds of measured signals.
Solution
Predictronics delivered a data-driven machine learning model adept at understanding the relationship between sensor variables and metrology variables, as well as estimating metrology variables with a high level of accuracy. Predictronics established this solution for virtual metrology through the configuration of domain-driven and statistical feature extraction, machine learning models for feature selection, and regression models. With our template-driven approach, this solution can be customized and deployed with limited effort for future customers in the semiconductor industry.
Value
Our virtual metrology solution reduces the need to perform costly metrology measurements, saving both time and money. Predictronics evaluated several best-in-class machine learning models throughout the development of this solution and our findings helped us to create and refine a virtual metrology template for our PDX software.