OPC Model Fitting with Parallel Gradient Composition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing OPC model simulation methods are inefficient in fitting complex photolithography technologies with high precision requirements due to increased data volume and computational demands.
Innovation Solution
A method utilizing distributed computing nodes for parallel computing and random data sampling to perform gradient composition and iteration, optimizing fitting parameters through precision judgment functions, avoiding local overfitting and improving convergence rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of fitting parameters is increased to improve fitting precision, then the fitting precision is improved, but the computing time and complexity increase greatly
Solution Approach 1:
The patent divides the large dataset into multiple subsets and distributes them across multiple computing nodes for parallel processing. Each computing node independently processes a portion of the data, which reduces the computational burden on individual nodes and enables efficient handling of increased data volume while maintaining fitting precision
Solution Approach 2:
The patent transitions from sequential computing to parallel computing by adding the dimension of multiple computing nodes. This dimensional change allows simultaneous processing of multiple data subsets, dramatically reducing computing time while maintaining the ability to handle increased numbers of fitting parameters
2Measurement precision
If the number of fitting parameters is increased to improve fitting precision, then the fitting precision is improved, but the device complexity increases
Solution Approach 1:
The computational task is segmented across multiple computing nodes, with each node handling a specific subset of data and parameters. This segmentation distributes the computational complexity across the system rather than concentrating it in a single computational unit, making the overall system capable of handling increased precision requirements
Solution Approach 2:
The computing cluster is designed as a universal parallel computing system that can handle various photolithography OPC modeling tasks. Each computing node performs similar functions (parameter fitting on local data), creating a scalable and flexible system that can adapt to different complexity requirements
3Productivity
If traditional sequential computing is used for fitting, then the system complexity is low, but the productivity and fitting efficiency are insufficient
Solution Approach 1:
The dataset is divided into multiple subsets that can be processed in parallel across different computing nodes. This segmentation enables simultaneous processing of multiple data portions, dramatically increasing fitting efficiency while the modular architecture keeps individual node complexity manageable
Solution Approach 2:
Multiple computing nodes are merged into a coordinated computing cluster that functions as a unified system. The nodes work in parallel on different data subsets and combine their results to achieve overall fitting objectives, achieving high productivity through the synergistic combination of distributed computational resources
Data Source
AI summary
The present application discloses an OPC model simulation method. The method includes the following steps: step 1, establishing a precision judgment function which is formed by multiplying each square of the difference between a simulation point of an OPC model and an actual point on a wafer, by weight, and then superposing all the squares; step 2, performing random data sampling, comprising forming distributed computing nodes; randomly distributing data to each computing node, and meanwhile distributing a current state value of fitting parameter space composed of all fitting parameters to each computing node; computing a local precision judgment function of each computing node; step 3, performing parallel computing to obtain the gradient of each local precision judgment function, and computing a first derivative and a first order approximate value of the gradient of each local precision judgment function; step 4, performing gradient composition and iteration.