Feature Vector Learning for Semiconductor Process Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for simulating semiconductor manufacturing processes are inefficient, as they require extensive parameter changes and lack effective data processing to accurately model and predict outcomes.
Innovation Solution
A learning device and simulation system that uses machine learning to process simulation data into feature vectors, associating XY coordinates with simulation data from semiconductor manufacturing processes, allowing for efficient simulation and prediction of process results using a learned model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional simulation methods are used for semiconductor manufacturing processes, then comprehensive process modeling is achieved, but simulation efficiency is low and time-consuming
Solution Approach 1:
The patent pre-calculates and stores feature vectors from simulation data before actual inference is needed. By preparing the data in advance and organizing it into structured feature vectors that capture essential process characteristics, the system enables rapid querying and comparison during actual simulations, significantly reducing the time required for process optimization iterations.
Solution Approach 2:
The patent creates simplified representations (feature vectors) that copy the essential characteristics of complex simulation data. Instead of repeatedly running full simulations, the system uses these copied feature representations to quickly assess process outcomes and compare different parameter configurations, maintaining accuracy while dramatically improving efficiency.
2Adaptability or versatility
If extensive parameter changes are performed for process optimization, then comprehensive process exploration is achieved, but data processing complexity increases
Solution Approach 1:
The patent extracts only the most relevant features from complex simulation data to create compact feature vectors. By identifying and extracting key characteristics that truly matter for process optimization while discarding redundant information, the system maintains comprehensive process exploration capability but with significantly reduced data processing complexity.
Solution Approach 2:
The patent transforms complex simulation data into standardized feature vectors with consistent dimensions and formats. This parameter transformation approach allows extensive exploration of different process configurations while maintaining uniform data structures that simplify processing and comparison across all scenarios.
3Measurement precision
If detailed simulation data is collected for accurate modeling, then model precision is improved, but data processing time increases
Solution Approach 1:
The patent creates compressed feature vector representations that capture the essential information from detailed simulation data. These copied representations maintain the precision needed for accurate modeling while containing far less data than the original simulations, enabling rapid processing without sacrificing model accuracy.
Solution Approach 2:
The patent pre-processes detailed simulation data into feature vectors in advance, extracting and storing only the critical information needed for accurate modeling. This preliminary extraction of essential features allows the system to maintain high model accuracy while avoiding the time cost of processing complete detailed simulations during iterative optimization.
Data Source
AI summary
A learning device for performing a machine learning based on a learning model using data input to an input layer, includes: a calculation part configured to calculate a predetermined number of features, in which simulation data as a result of simulating semiconductor manufacturing processes by setting environmental information inside a process vessel in which the semiconductor manufacturing processes are performed and using a predetermined component provided in the process vessel as a variable, and XY coordinates parallel to a plane of a wafer are associated with each other; and an input part configured to input the calculated predetermined number of features to the input layer.