Unsupervised Clustering for Semiconductor Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods are inadequate for automatically detecting anomalies in large collections of SEM images generated during semiconductor manufacturing, as they struggle with processing petabytes of data, identifying new anomalies, and are prone to errors in labeling previously seen anomalies.
Innovation Solution
An unsupervised cluster method that automatically synthesizes ideal features, extracts features from images, removes distortion, and aligns them with multi-resolution alignment methods, using AI models for optimal image processing techniques like multi-image composition and feature extraction, to accurately detect and classify anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing anomaly detection methods are used on SEM images, then processing can be performed with simple algorithms, but the methods cannot process petabytes of images and make errors in labeling anomalies
Solution Approach 1:
The patent segments the anomaly detection task into multiple components: (1) extracting features from individual SEM images, (2) clustering features across multiple images to identify patterns, (3) synthesizing ideal feature representations, and (4) comparing actual features against synthesized ideals. This segmentation enables processing of petabytes of images while maintaining accuracy through specialized algorithms for each subtask.
Solution Approach 2:
The patent introduces synthesized ideal features as an intermediary representation between raw SEM image features and anomaly detection results. These synthesized features act as a mediator that captures the essential characteristics of normal manufacturing processes, enabling reliable anomaly identification by comparing actual features against this ideal reference frame.
2Measurement precision
If manual inspection methods are used, then detection accuracy can be maintained, but processing speed and scalability to petabytes of data are insufficient
Solution Approach 1:
The patent replaces manual inspection mechanisms with an automated system that uses machine learning algorithms for feature extraction and clustering. The system substitutes human visual inspection with computational processes that can analyze petabytes of SEM images at high speed while maintaining precision through sophisticated pattern recognition and synthesis of ideal feature representations.
3Adaptability or versatility
If supervised learning methods are used for anomaly detection, then previously seen anomalies can be labeled, but the methods cannot detect new, previously unseen anomalies
Solution Approach 1:
The patent inverts the traditional supervised learning approach by not requiring labeled anomaly data. Instead of training on known anomalies and hoping to detect similar ones, the system synthesizes ideal features representing normal processes and detects any deviation from this ideal, thereby enabling detection of completely new and unseen anomaly types without losing information about them.
Solution Approach 2:
The system performs self-service by automatically synthesizing ideal feature representations from the data itself without requiring external labeling or supervision. The clustering algorithm autonomously identifies patterns and creates the reference framework for anomaly detection, enabling the system to adapt to new anomaly types independently.
Data Source
AI summary
Images are accessed representing a status in a fabrication of a semiconductor chip corresponding to a particular stage in the fabrication. Distortion is removed from the images and actual features of the semiconductor chip are extracted from the images. Synthesized ideal features of the semiconductor chip associated with completion of the particular stage in the fabrication are determined from the one or more images. The actual features are compared to the ideal features to determine whether anomalies associated with the particular stage exist in the semiconductor chip.