Semiconductor Prediction Model Transfer Across Fabrication Sites
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face challenges in accurately transferring patterns onto substrates due to limitations in existing lithographic projection technologies, particularly in reproducing features smaller than the classical resolution limit, which hampers the production of advanced integrated circuits and other devices.
Innovation Solution
A method is introduced to configure semiconductor manufacturing processes by providing an initial prediction model with model parameters to remote locations, where it is trained with local data to update and aggregate parameters, allowing for the adjustment of the prediction model to better reflect local conditions, thereby enhancing pattern transfer accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If an initial prediction model is provided to multiple remote locations for local training, then the model can be adjusted to reflect local conditions and improve prediction accuracy, but the complexity of managing and coordinating model updates across multiple locations increases
Solution Approach 1:
The patent segments the model training process into independent local training operations at each remote location. Each location trains the prediction model using local data independently, updating model parameters locally without requiring centralized coordination of the entire training process. This segmentation allows parallel training across multiple locations while maintaining prediction accuracy through local adaptation.
Solution Approach 2:
The patent implements a feedback mechanism where updated model parameters from each remote location are collected and used to adjust the initial prediction model. The system receives updated parameters from multiple locations, aggregates them, and uses this feedback to refine the model iteratively. This feedback loop enables continuous improvement of prediction accuracy while managing complexity through systematic parameter aggregation.
2Adaptability or versatility
If local data is used to train the prediction model at remote locations, then the model adapts to local manufacturing conditions, but the risk of data dissemination and intellectual property exposure increases
Solution Approach 1:
The patent extracts only the essential model parameters from the training process rather than sharing or transmitting the complete local datasets. Each remote location trains the model locally and extracts updated parameters, which are then sent to the service provider. This extraction approach enables local adaptation while minimizing data dissemination risk by sharing only necessary model updates rather than raw data.
Solution Approach 2:
The patent introduces an intermediary mechanism where the service provider receives and aggregates updated model parameters from multiple remote locations without accessing the underlying local data. This intermediary process allows the system to coordinate model improvements across locations while maintaining data isolation and protecting intellectual property, as the service provider only handles parameter exchanges rather than raw data.
3Measurement precision
If the prediction model is trained with local data at each remote location, then prediction accuracy for local conditions improves, but the time and computational resources required for training increase
Solution Approach 1:
The patent performs preliminary action by providing an initially trained prediction model to remote locations before local training begins. This pre-trained model serves as a starting point that already contains general knowledge from centralized training, reducing the amount of additional training needed at each location. The preliminary model enables faster local adaptation while maintaining accuracy, as local training only needs to fine-tune parameters for specific local conditions rather than training from scratch.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Described herein is a method for configuring a semiconductor manufacturing process, the method comprising: providing an initial prediction model comprising a plurality of model parameters to one or more remote locations; training the initial prediction model with local data at the one or more remote locations such that at least one model parameter is updated; receiving the at least one updated model parameter from the one or more remote locations; determining aggregated model parameters based on the at least one updated model parameter received from the one or more remote locations; and adjusting the initial prediction model based on the aggregated model parameters, the adjusted prediction model being operable to configure the semiconductor manufacturing process.