Cross Autoencoder Model for Semiconductor Wafer Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wafer defect detection techniques face challenges such as sensitivity to noise, alignment accuracy issues, and high computational resource consumption, particularly in die-to-die and simulation-based inspection methods, which affect the accuracy and efficiency of defect detection in semiconductor manufacturing.
Innovation Solution
The use of a trained machine learning model incorporating cross autoencoders that compare inspection images with corresponding design layout data, allowing for feature extraction and comparison in a single step, reducing noise and computational requirements, and enabling more accurate and efficient defect detection without relying on simulation images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If die-to-die inspection or simulation-based inspection is used for defect detection, then defect detection capability is improved, but sensitivity to noise increases and computational resource consumption increases
Solution Approach 1:
The patent uses a trained machine learning model that has learned the relationship between inspection images and design layout data during training phase. During actual defect detection, the model copies this learned knowledge to directly compare inspection images with their corresponding design layout data, avoiding the need for simulation images and reducing noise sensitivity while maintaining defect detection capability
Solution Approach 2:
The patent introduces a trained machine learning model as an intermediary between inspection images and design layout data. The model automatically aligns and compares the two data types, eliminating the need for manual alignment procedures that are sensitive to noise and require high computational resources
2Measurement precision
If die-to-die inspection or simulation-based inspection is used for defect detection, then defect detection capability is improved, but computational resource consumption increases
Solution Approach 1:
The patent performs preliminary training of the machine learning model using paired inspection images and design layout data before actual defect detection. This preliminary action enables the model to automatically understand the relationship between the two data types, so during actual use, no complex simulation or alignment computations are needed, significantly reducing computational resource consumption
Solution Approach 2:
The trained model copies the learned mapping relationship from training data to production data, enabling direct comparison between inspection images and design layout data without requiring simulation-based approaches that consume high computational resources
3Device complexity
If traditional defect detection methods are used, then process is simple, but alignment accuracy between inspection images and design layout data deteriorates
Solution Approach 1:
The trained machine learning model acts as an intermediary that automatically performs alignment between inspection images and design layout data. During training, the model learns the alignment relationship from paired data, and during inference, it automatically applies this alignment knowledge, achieving high alignment accuracy without increasing apparent process complexity
Solution Approach 2:
The patent replaces traditional mechanical alignment procedures with a machine learning-based automatic alignment system. The model learns alignment transformations during training and automatically applies them during defect detection, achieving superior alignment accuracy without manual intervention
Data Source
AI summary
Systems and methods for training a machine learning model for defect detection include obtaining training data including an inspection image of a fabricated integrated circuit (IC) and design layout data of the IC, and training a machine learning model using the training data. The machine learning model includes a first autoencoder and a second autoencoder. The first autoencoder includes a first encoder and a first decoder. The second autoencoder includes a second encoder and a second decoder. The second decoder is configured to obtain a first code outputted by the first encoder. The first decoder is configured to obtain a second code outputted by the second encoder.