Semiconductor Process Simulation Using ML for Electrical Parameter Tuning

Solution Overview

Problem

The semiconductor fabrication process is time-consuming and costly when trying to find optimal process and design parameters to achieve desired electrical characteristics of semiconductor devices using physical model-based simulations.

Innovation Solution

A simulation method using a machine learning model to generate predictive data for electrical characteristics of semiconductor devices based on input process and design parameters, with adjustments made until the difference between predictive and target data is within a predetermined threshold, allowing for efficient optimization of these parameters.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If physical model-based simulations are used to find optimal process and design parameters, then desired electrical characteristics of semiconductor devices can be achieved, but the process becomes time-consuming and expensive

Engineering Contradiction:

Improveelectrical characteristicsVSAvoidsimulation time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-training a machine learning model using comprehensive physical model-based simulations to establish predictive relationships between process parameters, design parameters, and electrical characteristics. This pre-computed knowledge base enables rapid parameter optimization without requiring time-consuming simulations during actual device fabrication, thus achieving desired electrical characteristics while significantly reducing time loss.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent employs copying by creating a virtual replica of the physical simulation process through a machine learning model. This model copies the essential input-output relationships from extensive physical simulations, allowing rapid prediction of electrical characteristics based on process and design parameters without executing the full physical models again, thereby reducing simulation time while maintaining prediction accuracy.

Inventive Principle:

Principle #26Copying

2Manufacturing precision

If physical model-based simulations are used to find optimal process and design parameters, then desired electrical characteristics of semiconductor devices can be achieved, but the process becomes expensive

Engineering Contradiction:

Improveelectrical characteristicsVSAvoidcomputational cost

Core Design Contradiction:

Manufacturing precisionVSLoss of energy

Solution Approach 1:

The patent applies preliminary action by pre-training a machine learning model using comprehensive physical model-based simulations to establish predictive relationships between process parameters, design parameters, and electrical characteristics. This pre-computed knowledge base enables rapid parameter optimization without requiring time-consuming simulations during actual device fabrication, thus achieving desired electrical characteristics while significantly reducing time loss.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent employs copying by creating a virtual replica of the physical simulation process through a machine learning model. This model copies the essential input-output relationships from extensive physical simulations, allowing rapid prediction of electrical characteristics based on process and design parameters without executing the full physical models again, thereby reducing simulation time while maintaining prediction accuracy.

Inventive Principle:

Principle #26Copying

3Productivity

If machine learning models are used to predict electrical characteristics, then simulation time is reduced, but model training requires initial computational resources

Engineering Contradiction:

Improvesimulation speedVSAvoidtraining cost

Core Design Contradiction:

ProductivityVSLoss of energy

Solution Approach 1:

The patent applies preliminary action by pre-training a machine learning model using comprehensive physical model-based simulations to establish predictive relationships between process parameters, design parameters, and electrical characteristics. This pre-computed knowledge base enables rapid parameter optimization without requiring time-consuming simulations during actual device fabrication, thus achieving desired electrical characteristics while significantly reducing time loss.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11982980B2Simulation method for semiconductor fabrication process and method for manufacturing semiconductor device

Publication Date: 2024.05.14 SAMSUNG ELECTRONICS CO LTD

AI summary

According to an aspect of the present inventive concept, a simulation method for a semiconductor fabrication process includes obtaining, as input data, process parameters for controlling a semiconductor process of manufacturing semiconductor devices, or design parameters representing a structure of the semiconductor devices, or both the process parameters and the design parameters; generating predictive data for electrical characteristics of the semiconductor devices using a machine learning model based on the input data; generating reference data for the electrical characteristics of the semiconductor devices using a simulation tool based on the input data; and training the machine learning model using the predictive data and the reference data.