DefectGPT Transformer for Semiconductor Defect Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect synthesis methods for semiconductor applications are limited in generating realistic defect images, particularly in multiple optical mode contexts, and are computationally intensive, making them impractical for large-scale defect synthesis.
Innovation Solution
The use of a pre-trained Defect Generative Pre-Trained Transformer (DefectGPT) encoder to determine information for a specimen, enabling the synthesis of realistic defect images aligned with process conditions and defect distributions, and facilitating defect detection in inspection and metrology applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional defect synthesis methods are used, then defect images can be generated, but the computational intensity is high and the synthesis process is computationally intensive making it impractical for large-scale defect synthesis
Solution Approach 1:
The patent applies preliminary action by pre-training the transformer model on comprehensive defect data and process condition relationships before actual defect synthesis. The model is pre-trained to learn the complex mappings between process parameters and defect characteristics, enabling rapid defect synthesis during production without intensive computational resources at runtime. This pre-computation approach resolves the contradiction by shifting computational burden to an offline training phase.
Solution Approach 2:
The patent replaces traditional mechanical/computational defect synthesis methods with an AI-based transformer model. Instead of using computationally intensive simulation or physical modeling to generate defect images, the system uses a trained neural network that can rapidly synthesize realistic defect images by processing process condition inputs. This substitution dramatically reduces computational energy consumption while maintaining defect synthesis quality.
2Manufacturing precision
If traditional defect synthesis methods are used, then defect images can be generated, but the generated defect images may not be sufficiently realistic or aligned with actual defect distributions in multiple optical mode contexts
Solution Approach 1:
The patent applies parameter changes by using the transformer model to dynamically adjust and synthesize defect images based on multiple process condition parameters. The model takes into account various parameters such as process temperature, pressure, material composition, and equipment variations to generate defect images that accurately reflect real-world defect distributions. This multi-parameter approach enables realistic defect synthesis across different optical modes without requiring overly complex manual modeling for each scenario.
Solution Approach 2:
The patent applies universality by designing a single transformer-based defect synthesis system that can generate realistic defect images across multiple optical modes and various semiconductor manufacturing processes. Instead of requiring separate complex synthesis tools for each optical mode (e.g., brightfield, darkfield, oblique illumination), the unified model learns general defect patterns and can adaptively generate mode-specific defects. This multi-functional approach achieves high manufacturing precision while avoiding the complexity of maintaining multiple specialized synthesis systems.
3Measurement precision
If more computational resources are allocated to defect synthesis, then more realistic defect images can be generated, but the computational intensity increases making it impractical for large-scale applications
Solution Approach 1:
The patent resolves this contradiction by performing preliminary training of the transformer model on large datasets of real defect images and process conditions. During this offline pre-training phase, the model learns to accurately map process parameters to defect characteristics, achieving high measurement precision for defect detection. Once trained, the model can rapidly synthesize defect images during production with high throughput, as the heavy computational learning has already been completed. This separates the precision-building phase from the throughput-critical phase.
Data Source
AI summary
Methods and systems for determining information for a specimen are provided. One system includes a computer subsystem and one or more components executed by the computer subsystem. The one or more components include a pre-trained defect generative pre-trained transformer (DefectGPT) encoder configured for determining information for a specimen based on one or more inputs specific to the specimen. The computer subsystem is configured for inputting the one or more inputs into the pre-trained DefectGPT encoder.