Multi-Scale Autoencoders for Semiconductor Feature Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in efficiently detecting and characterizing crystalline material features in semiconductor workpieces at multiple length scales, which is crucial for accurate inspection and quality control in semiconductor manufacturing.
Innovation Solution
The use of machine-learned encoding models and autoencoders, specifically deep convolutional multiscale variational autoencoders (MS-VAE), to process workpiece images and generate residual images at multiple length scales, enabling the detection and characterization of high-resolution features in semiconductor workpieces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional inspection methods are used to detect crystalline material features, then the inspection process is simple, but the detection precision and characterization accuracy at multiple length scales are insufficient
Solution Approach 1:
The patent segments the workpiece image into multiple scale representations using a multi-scale encoder architecture. The encoder processes images at different resolutions (e.g., 1x, 2x, 4x downsampling levels) to capture features at various length scales simultaneously. This segmentation approach enables precise detection of both large-scale workpiece features and fine-scale crystalline defects without requiring a single complex high-resolution system
Solution Approach 2:
The patent adds the dimension of scale resolution by processing images at multiple downsampling levels. Instead of relying on a single inspection dimension, the system creates a multi-dimensional feature space that combines information from different resolution levels, enabling comprehensive characterization of crystalline features across multiple length scales
2Measurement precision
If high-resolution imaging is used to capture fine features, then feature detection accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent applies partial action by processing images at multiple downsampling levels rather than always using full resolution. The multi-scale encoder processes a pyramid of resolutions (e.g., original, 2x downsampled, 4x downsampled), allowing the system to capture fine features when necessary while using coarser representations for broader contextual analysis, thus reducing overall processing time while maintaining detection accuracy
Solution Approach 2:
The patent segments the processing task across multiple resolution levels, where each scale captures different feature sizes. This segmentation allows parallel processing of different resolution levels and enables the system to focus computational resources on detecting features at appropriate scales rather than processing all data at maximum resolution
3Adaptability or versatility
If multiple length scales are analyzed simultaneously, then comprehensive feature detection is achieved, but the system complexity and data processing requirements increase
Solution Approach 1:
The patent merges multiple scale representations into a unified latent space through the multi-scale encoder architecture. Different resolution levels are processed through separate encoder branches and then combined in the latent space, allowing comprehensive multi-scale feature detection while maintaining a cohesive model structure that manages complexity through systematic integration
Data Source
AI summary
An example method includes obtaining a workpiece image of a semiconductor workpiece. The example method includes providing the workpiece image as input to a machine-learned encoding model. The example method includes obtaining an output from the machine-learned encoding model, the output includes an encoding corresponding to the semiconductor workpiece. The example method includes determining one or more characteristics of the semiconductor workpiece based at least in part on the encoding or modifying a semiconductor manufacturing process based at least in part on the encoding.