Inverse Lithography Mask Synthesis Using Synthetic Pattern Training

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The existing inverse lithography technology for mask synthesis in integrated circuit fabrication is computationally expensive and time-consuming, limiting its industrial applicability due to the need for iterative simulation and mathematical computation to achieve high accuracy, which is not efficiently scalable for complex designs.

Innovation Solution

A computing system implementing a mask synthesis system that generates synthetic image clips and corresponding mask data using machine learning to determine pixelated output masks, reducing the computational burden by dividing layout designs into image cells and utilizing machine learning-based inverse lithography technology for optical proximity correction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If inverse lithography technology is used to accurately reproduce small features in photolithographic process, then manufacturing precision is improved, but computational cost and processing time increase significantly

Engineering Contradiction:

Improvepattern fidelityVSAvoidprocessing time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent pre-calculates and stores optimal mask patterns for various design shapes and feature sizes in a lookup table during an offline training phase. During actual mask synthesis, the system simply queries this pre-computed database rather than performing complex iterative calculations, thereby achieving high precision rapidly without real-time computational burden

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates a simplified computational model (neural network) that copies the behavior of the complex inverse lithography technology. This model is trained offline to replicate the high-precision mask generation capabilities of traditional ILT, enabling fast inference during production while maintaining pattern fidelity

Inventive Principle:

Principle #26Copying

2Manufacturing precision

If inverse lithography technology is used to accurately reproduce small features in photolithographic process, then manufacturing precision is improved, but computational complexity increases

Engineering Contradiction:

Improvepattern fidelityVSAvoidcomputational complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent replaces the complex iterative mathematical optimization system of traditional inverse lithography technology with a machine learning-based neural network system. The neural network, once trained, provides direct mask pattern generation without requiring complex iterative calculations, thus reducing computational complexity while maintaining precision

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

A simplified neural network model is created to copy the essential functionality of the complex ILT algorithm. This model captures the relationship between design shapes and optimal mask patterns through learning, enabling complex mask synthesis tasks to be performed with simpler, faster computations during inference

Inventive Principle:

Principle #26Copying

3Productivity

If traditional mask synthesis methods are used, then computational resources are consumed, but productivity remains low due to time-consuming iterations

Engineering Contradiction:

Improvemask synthesis efficiencyVSAvoidcomputational cost

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The system performs comprehensive mask pattern calculations and stores optimal solutions in advance during an offline training phase. This pre-computation eliminates the need for repeated iterative calculations during production, significantly improving productivity while reducing online computational energy consumption

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

A neural network model is trained to copy the mask generation capabilities of traditional methods. Once trained, this model provides rapid mask synthesis through simple forward propagation, dramatically improving productivity and reducing computational energy requirements compared to traditional iterative approaches

Inventive Principle:

Principle #26Copying

Data Source

PatentUS20230064987A1Training of machine learning-based inverse lithography technology for mask synthesis with synthetic pattern generation

Publication Date: 2023.03.02 SIEMENS INDUSTRY SOFTWARE INC

AI summary

This application discloses a computing system implementing a mask synthesis system to generate synthetic image clips of design shapes and corresponding mask data for the synthetic image clips. The mask data can describe lithographic masks capable of being used to fabricate the design shapes on an integrated circuit. The mask synthesis system can utilize the synthetic image clips of the design shapes and the corresponding mask data to train a machine-learning system to determine pixelated output masks from portions of the layout design. The mask synthesis system can identify one or more pixelated output masks for portions of a layout design describing an electronic system using the trained machine-learning. The mask synthesis system can synthesize a mask layout design for the electronic system based, at least in part, on the layout design describing the electronic system and the one or more pixelated output masks for the layout design.