Semiconductor Design Automation Correcting Simulation Sampling Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity and miniaturization of semiconductor devices lead to interactions between design and manufacturing steps, resulting in unintended electrical characteristics, necessitating improved reliability and speed in generating semiconductor process models and predictions, which existing technologies fail to adequately address.
Innovation Solution
A semiconductor design automation system that employs machine learning based on a database loaded with both simulation and real data, incorporating a simulator, recovery module, preprocessing module, and graphic user interfaces to generate and visualize process models and predictions, correcting sampling errors and preprocessing data for accurate predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If simulation data is used to generate semiconductor process models, then experimental costs are reduced, but sampling errors decrease reliability
Solution Approach 1:
A recovery module is introduced as an intermediary component that receives simulation data and applies correction algorithms to compensate for sampling errors. This mediator transforms the raw simulation data into corrected simulation data with improved reliability, addressing the contradiction between using simulation data and maintaining accuracy.
Solution Approach 2:
The system changes the parameter quality of simulation data by applying correction algorithms that adjust statistical parameters to match real data distributions. This transforms the data characteristics from simulated to more realistic, improving reliability without requiring additional physical experiments.
2Productivity
If machine learning is executed with manually prepared data, then model accuracy can be maintained, but processing time and complexity increase
Solution Approach 1:
The system performs preliminary actions by automatically loading, preprocessing, and correcting simulation data before machine learning execution. The recovery module pre-processes data to correct sampling errors, and the preprocessing module prepares data in optimal formats, reducing the complexity burden during the actual machine learning phase.
Solution Approach 2:
The system implements self-service through automated data loading and preprocessing mechanisms that eliminate manual data preparation steps. The automatic simulation generator and preprocessing module autonomously handle data preparation tasks, improving productivity while managing complexity through automation rather than manual intervention.
3Measurement precision
If real experimental data is collected for model generation, then data accuracy is improved, but experimental costs and time consumption increase
Solution Approach 1:
The system creates corrected copies of simulation data that replicate the statistical characteristics of real experimental data without requiring actual physical experiments. The recovery module generates corrected simulation data that copies the essential features and distributions of real data, providing accurate training data instantly without experimental time delays.
Solution Approach 2:
The data processing system serves multiple functions: it processes both simulation data and real data through the same recovery and preprocessing modules, enabling the system to improve accuracy without always requiring new experimental data collection. The universal processing pipeline handles different data sources efficiently.
Data Source
AI summary
A semiconductor design automation system comprises a simulator configured to generate simulation data, a recovery module configured to correct a sampling error of the simulation data to generate recovery simulation data, a hardware data module configured to generate real data, a preprocessing module configured to preprocess the real data to generate preprocessed real data, a database configured to store the recovery simulation data and the preprocessed real data, a first graphic user interface including an automatic simulation generator configured to generate a machine learning model of the recovery simulation data and the preprocessed real data and generate predicted real data therefrom, and a second graphic user interface including a visualization unit configured to generate a visualized virtualization process result from the machine learning model.