Symmetric Neural Network for Order-Invariant Multi-Die Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural networks for semiconductor defect detection are sensitive to the input order of images, leading to inconsistent and unpredictable output predictions, which complicates defect analysis and debugging.
Innovation Solution
A neural network architecture is designed to be symmetric with respect to all input channels, ensuring invariant output predictions regardless of changes in input order, using a Convolutional Neural Network (CNN) with specific filter configurations to stabilize defect map generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a conventional neural network is used for defect detection, then defect detection capability is provided, but output predictions become sensitive to input order changes leading to inconsistency
Solution Approach 1:
The patent applies asymmetry in reverse by creating a symmetric neural network architecture where the processing graph remains invariant under permutations of input channels. This symmetric design ensures that regardless of the order in which images are input, the defect detection output remains consistent, resolving the contradiction between reliability and adaptability to input ordering
Solution Approach 2:
The neural network is designed with universal processing capability where any input channel can serve as the target channel for defect detection. The symmetric architecture allows the same network to handle different input orderings and channel configurations uniformly, making the system adaptable to various inspection scenarios while maintaining consistent results
Data Source
AI summary
There is provided a system and method of runtime defect detection in a semiconductor specimen. The method includes obtaining a plurality of runtime images acquired for a plurality of dies on the specimen, feeding the plurality of runtime images to a plurality of input channels of a neural network (NN) in an input order, wherein the NN is previously trained in a training phase, and processing, by the NN, the plurality of runtime images simultaneously, to obtain a plurality of defect maps, each corresponding to a respective runtime image and indicating probabilities of defect candidate presence thereof. Each given runtime image is processed as a target image using remaining images in the plurality of runtime images as reference images of the target image, and the defect map of the target image remains invariant, irrespective of changes to the input order.