Federated Lithography Prediction Models for Cross-Facility Precision
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 achieving precise feature sizes below the classical resolution limit, which hampers the production of advanced integrated circuits and other devices.
Innovation Solution
A method is introduced to configure and enhance semiconductor manufacturing processes by adjusting prediction models used in lithography. This involves providing an initial prediction model with model parameters to remote locations, training it with local data, updating parameters, and aggregating them to improve the lithography process. The method includes using neural networks to recognize reticle contamination and determine collector contamination levels, allowing for iterative refinement until convergence is achieved.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If lithographic projection apparatus is used to transfer patterns onto substrates, then pattern transfer capability is achieved, but manufacturing precision deteriorates due to inability to achieve precise feature sizes below classical resolution limit
Solution Approach 1:
The patent changes the parameters of the prediction model (weights, biases, hyperparameters) based on local data from different facility locations. This allows the model to adapt to local variations in the lithography process, thereby achieving precise feature sizes below the classical resolution limit while maintaining pattern transfer accuracy
Solution Approach 2:
The patent implements a feedback mechanism where prediction models are trained on local data from each facility location and the updated model parameters are aggregated back to improve the global model. This continuous feedback loop enables the system to learn from local variations and improve manufacturing precision without compromising reliability
2Measurement precision
If prediction models are trained with local data at remote locations, then model accuracy is improved, but data sharing requirements increase which may compromise customer data security
Solution Approach 1:
The patent segments the training process by keeping customer data localized at each facility location. Only aggregated model parameters (not raw data) are shared across locations, which maintains prediction model accuracy while protecting customer data security by preventing direct access to sensitive information
Solution Approach 2:
The patent uses an intermediary aggregation process where local model parameters are combined to create improved global models without requiring direct sharing of customer data. This intermediary step acts as a mediator that enables model improvement while maintaining data security boundaries
Data Source
AI summary
A method for configuring a semiconductor manufacturing process, the method including: providing an initial prediction model including a plurality of model parameters to one or more remote locations; receiving at least one updated model parameter from the one or more remote locations, the at least one model parameter is updated by training the initial prediction model with local data at 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.