Digital Twin AI Synthesis for Semiconductor Data Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI methods for semiconductor manufacturing are inadequate for addressing data insufficiency and system-level problems, as they are primarily designed for specific standalone issues and lack the capability to handle multiple interacting processes.
Innovation Solution
A method utilizing a digital twin that receives physics constraints and historical samples to perform AI synthesis, generating synthesized samples that can be combined with real-world samples to address data insufficiency and system-level challenges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If individual AI methods (generative methods, data augmentation) are used for specific standalone issues, then data insufficiency for specific scenarios can be addressed, but the methods cannot handle system-level problems with multiple interacting processes
Solution Approach 1:
The patent combines multiple individual AI methods into a unified digital twin framework that can handle system-level problems. The digital twin integrates generative models, data augmentation techniques, and physics constraints into a single system that generates synthesized data across multiple interacting processes simultaneously, resolving the limitation of standalone methods
Solution Approach 2:
The digital twin framework is designed as a universal system that can address data insufficiency across multiple different semiconductor manufacturing processes. It performs multiple functions including generating synthesized process parameters, equipment states, and defect data while adhering to physics constraints, making it applicable to system-level problems rather than single-specific issues
2Reliability
If extensive real-world testing is performed to generate sufficient training data, then data quality and system-level understanding improve, but time and costs increase significantly
Solution Approach 1:
The digital twin performs preliminary data generation by synthesizing training data in advance based on physics constraints and historical data, before actual real-world testing or deployment. This allows the system to prepare comprehensive training datasets without requiring extensive physical experimentation, thereby reducing time and costs while maintaining data quality
Solution Approach 2:
The patent creates virtual copies of real-world semiconductor manufacturing processes through the digital twin. These synthesized copies replicate the behavior and characteristics of actual processes, providing high-quality training data that mirrors real-world conditions without requiring extensive physical testing, thus reducing time and resource requirements
3Quantity of substance
If more real-world samples are collected to address data insufficiency, then training data availability improves, but the complexity and resources required for data collection and processing increase
Solution Approach 1:
Instead of collecting more physical real-world samples which requires complex data collection infrastructure, the digital twin generates virtual copies of process data through AI synthesis. This approach provides sufficient training samples without needing additional physical sensors, equipment, or data collection systems, thereby avoiding increased infrastructure complexity
Data Source
AI summary
Embodiments herein use constraints from a target process (e.g., testing a semiconductor wafer or designing a new integrated circuit) and historical real-world samples (e.g., probe test sample, fault detection, or measured signals) to generate synthesized samples using AI synthesis embedded in a digital twin. These test samples can be combined with real-world test samples and then evaluated to determine next actions. In another example, a digital twin can use an new IC design and a design from an older, but related, IC to synthesize new IC design samples.