Semiconductor Process Image Generator Training for Rotation Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning methods for semiconductor process modeling struggle with recognizing rotation and symmetry transformations of input images, leading to prolonged training times due to the large number of images required.
Innovation Solution
A method of training a semiconductor process image generator by incorporating groups of mask images and their transformed versions, with controlled iterations and reduced numbers of transformed images, to accelerate convergence of the loss function.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning technology is used to model semiconductor process with input images and their transformed versions, then the model can recognize rotation and symmetry transformations, but the training time becomes excessively long
Solution Approach 1:
The training process is segmented into multiple stages: first training with original mask images only, then progressively introducing transformed images from different groups. This staged approach divides the large training task into manageable segments, reducing overall training time while maintaining transformation recognition capability
Solution Approach 2:
The method performs preliminary action by first training the neural network with original mask images to establish baseline performance, before subsequently introducing transformed images. This preliminary training phase prepares the model to better handle transformed inputs without requiring extensive retraining, thus reducing total training time
Data Source
AI summary
A method of training a semiconductor process image generator includes training the semiconductor process image generator with a plurality of mask images including a first group and a second group, training the semiconductor process image generator with the second group and a first transformed group obtained by applying a transformation to the first group, and training the semiconductor process image generator with the first group and a second transformed group obtained by applying a transformation to the second group.