To ensure stable processing and high-yield production, high-tech factories (e.g., semiconductor, TFT-LCD) demand product quality total inspection. Generally speaking, sampling inspection only measures a few samples and comes with metrology delay, thus it usually cannot achieve the goal of real-time and online total inspection. Automatic Virtual Metrology (AVM) was developed to tackle such problem. It can collect the data from the process tools to conjecture the virtual metrology (VM) values in the prediction model for realizing the goal of online and real-time total inspection. With the advancement of technology, the processes become more and more precise, and better accuracy of VM value prediction is demanded. The CNN-based AVM (denoted as AVM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CNN</sub> ) scheme can not only enhance the accuracy of the original AVM prediction, but also perform better on the extreme values. Nevertheless, two advanced capabilities need to be addressed for its practical applications: 1) effective initial-model-creation approach with insufficient metrology data; and 2) intelligent self-learning capability for on-line model refreshing. To possess these two advanced capabilities, the Advanced AVM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CNN</sub> System based on convolutional autoencoder (CAE) and transfer learning (TL) is proposed in this work. It is verified that the Advanced AVM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CNN</sub> System is more feasible for the onsite applications of the actual production lines.
Read full abstract