Solid-State Detector Characterization Using Physics-Based Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for characterizing solid-state detectors are limited in providing detailed material properties on a voxel-by-voxel basis, requiring extensive resource and time, and existing machine learning models struggle with spatial limitations and training difficulties due to weak signals.
Innovation Solution
A physics-based neural network is trained using electrode signals or free charges as ground truth, with weighted loss functions and regularization to enhance the characterization of solid-state detectors, modeling charge and voltage distribution at a microscopic level.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional experimental measurements and simulations are used to characterize solid-state detectors, then material properties can be determined, but the characterization is limited to bulk properties and requires extensive resources and time
Solution Approach 1:
The patent replaces conventional experimental measurement systems with a machine learning-based physical model. The system uses trained neural networks to predict charge and voltage distributions based on input material properties, eliminating the need for extensive physical measurements and simulations while providing detailed voxel-by-voxel characterization.
Solution Approach 2:
The patent creates a virtual copy of the solid-state detector through a trained physical model that replicates the behavior of actual detectors. This digital twin can be queried for detailed material properties without requiring physical measurements, providing bulk and voxel-level characteristics instantly.
2Productivity
If machine learning models are used to predict charge and voltage distribution, then characterization speed is improved, but spatial limitations occur in predicting weaker signals
Solution Approach 1:
The patent modifies the loss function parameters during model training to apply differential weighting. Weaker signals (such as hole signals) are assigned higher weights in the loss calculation, forcing the model to pay more attention to these difficult-to-predict quantities during training, thereby improving their prediction accuracy while maintaining fast characterization speed.
3Measurement precision
If detailed voxel-by-voxel material properties are obtained, then spatial resolution is improved, but resource requirements become unjustifiable
Solution Approach 1:
The patent replaces resource-intensive experimental setups and simulations with a computational machine learning model. Once the physical model is trained using initial measurements, it can generate detailed voxel-by-voxel material properties instantly without requiring additional physical resources, providing high spatial resolution at minimal cost.
4Measurement precision
If training data includes multiple ground truth measurements, then model accuracy is improved, but the complexity of obtaining and processing data increases
Solution Approach 1:
The patent extracts and uses only the essential ground truth data needed for training - specifically electrode signals and basic material properties. By focusing on these key measurements rather than comprehensive datasets, the system achieves effective model training with reduced data acquisition and processing complexity.
Data Source
AI summary
A physics-based network model is trained to learn weights such as trapping, detrapping, and/or transport of holes and/or electrons, as well as voltage distribution on a voxel-by-voxel basis throughout a solid-state detector model. The physics-based network may be used to estimate material property variation throughout the voxels. To reduce the number of experimental setups and information needed to train the models, the models may be trained using more easily acquired ground truth. Just the electrode signals or just the free charge data is used to train the model to characterize the solid-state detector. With this reduced data, the detector may be characterized using equivalency, such as combining multiple trapping centers to an equivalent trapping center. Regularization may be used in the loss calculation, such as where just the electrode signals are used, to deal with the reduced data available as ground truth.