Semiconductor Process Modeling with Modular ML Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for modeling semiconductor processes are costly and resource-intensive, requiring significant time and computing power to accurately predict and interpret the results of complex semiconductor processes.
Innovation Solution
A method and system for modeling semiconductor processes using machine learning, which involves obtaining measurement values from input data defining sub-process steps and measurement steps, grouping these steps into modules, and training a machine learning model to predict semiconductor device characteristics. The model includes a first sub-model that outputs feature values based on the modules and a second sub-model that outputs estimated values representing the characteristics of the semiconductor device.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional technology-based computer aided design (TCAD) methods are used to predict semiconductor process results, then prediction capability is provided, but time and computing resources are significantly consumed
Solution Approach 1:
The patent transforms the semiconductor process modeling approach by changing the fundamental parameters from traditional physics-based simulation to machine learning-based prediction. The system divides the semiconductor process into multiple sub-process steps and uses separate machine learning models for each step, changing the computational parameters from continuous physics calculations to discrete data-driven predictions, thereby reducing computation time while maintaining accuracy
Solution Approach 2:
The patent segments the complex semiconductor manufacturing process into multiple sub-process steps (e.g., deposition, etching, doping). Each sub-process is modeled by a separate machine learning model, allowing independent optimization and parallel computation. This segmentation reduces the overall computational burden compared to traditional monolithic TCAD simulations
2Reliability
If traditional TCAD methods are used to estimate semiconductor process results, then prediction capability is achieved, but computing resources are excessively required
Solution Approach 1:
The patent creates simplified copies of the complex physics-based TCAD models using machine learning. Instead of running resource-intensive physics simulations, the system trains machine learning models on historical simulation data to create lightweight predictive copies that replicate the behavior of full TCAD simulations with minimal computational resources
Solution Approach 2:
The system changes the computational parameters from continuous physics equations requiring significant computing power to discrete machine learning inference that can be executed efficiently. By pre-computing and storing relationships in the trained models, the system eliminates the need for repeated heavy calculations during actual predictions
3Measurement precision
If comprehensive process conditions are considered in semiconductor modeling, then prediction accuracy is improved, but model complexity increases
Solution Approach 1:
The patent divides the complex modeling task into multiple simpler sub-tasks, with each machine learning model handling a specific sub-process step. This segmentation allows each model to focus on specific process conditions and parameters, reducing the complexity individual models need to manage while collectively capturing comprehensive process effects
Solution Approach 2:
The patent introduces intermediate feature extraction layers between input process parameters and final predictions. These intermediate representations serve as mediators that organize and process complex process conditions in a structured manner, making the overall system more manageable while maintaining comprehensive analysis capability
Data Source
AI summary
Provided is a method of modeling a semiconductor process including obtaining a measurement value based on input data defining sub process steps and measurement steps; based on the measurement steps, grouping the sub process steps to respectively correspond to a plurality of modules; and based on the grouped sub process steps, training a machine learning model to predict at least one characteristic of a semiconductor device. The machine learning model includes a first sub model configured to output a feature value based on the plurality of modules, and based on the feature value, a second sub model configured to output an output value representing an estimated value corresponding to each of the plurality of modules and the at least one characteristic of the semiconductor device.