Abstract:Optical lithography is the main enabler to semiconductor manufacturing. It requires extensive processing to perform the Resolution Enhancement Techniques (RETs) required to transfer the design data to a working Integrated Circuits (ICs). The processing power and computational runtime for RETs tasks is ever increasing due to the continuous reduction of the feature size and the expansion of the chip area. State-of-the-art research sought Machine Learning (ML) technologies to reduce runtime and computational power, however they are still not used in production yet. In this study, we analyze the reasons holding back ML computational lithography from being production ready and present a novel highly scalable end-to-end flow that enables production ready ML-RET correction.

What problem does this paper attempt to address?

The paper attempts to address the issue of how to improve the efficiency and manufacturability of resolution enhancement techniques (RET) in nano-lithography using machine learning (ML) methods. Specifically: 1. **Limitations of Traditional RET Techniques**: As feature sizes continue to shrink, traditional lithography techniques require increasingly more computational resources to perform RET, leading to significantly increased processing time and computational costs. 2. **Problems with Existing ML-RET Methods**: - **Information Loss**: Using binarized mask data during training leads to information loss in intermediate operations, making it difficult for the model to infer physical relationships in the process. - **Full-Chip Scale Issue**: Existing studies mainly focus on the accuracy and runtime of local areas, lacking effective full-chip scale processing methods. - **Consistency Issue**: Performing lithography corrections in different window slices may lead to inconsistent results. - **Hardware Requirements**: Traditional RET requires a large amount of CPU resources, while ML models are more suited to GPU environments, increasing hardware migration costs. - **Pattern Interaction Distance**: Patterns at different locations may have different interaction distances, leading to consistency issues. - **Input Resolution**: The choice of resolution during image conversion significantly affects the correction results. - **Unseen Pattern Integration**: The model needs to be retrained to adapt to newly emerging pattern types. - **RET Process Segmentation**: Segmented RET processing increases complexity. 3. **Proposed New Method TPM-RET**: The paper proposes a new framework called "True Pixel-based Machine-learning RET (TPM-RET)" aimed at overcoming the above issues to achieve an efficient, consistent, and full-chip scale ML-RET solution. The design choices of TPM-RET include: - **Pixel-Level Correction**: Avoids boundary conflict issues brought by window slicing. - **Minimal Model Footprint**: Uses convolutional neural networks (CNN) to reduce computational resource requirements. - **Inverse Intensity Profile**: Introduces the Inverse Intensity Profile (IIP) algorithm to recover process information under binarized masks. - **Non-Uniform Image Compression**: Optimizes input image size while maintaining directional asymmetry. - **End-to-End Correction Process**: Simplifies dataset engineering, reduces training and validation time, lowers hardware requirements, and fully leverages the ML model's ability to handle complex patterns. Through these improvements, TPM-RET has demonstrated good performance and scalability in practical applications.

Novel End-to-End Production-Ready Machine Learning Flow for Nanolithography Modeling and Correction

Computational Lithography Using Machine Learning Models

Machine Learning Applied to Electron Beam Lithography to Accelerate Process Optimization of a Contact Hole Layer

Litho-aware redundant local-loop insertion framework with convolutional neural network

Generic Lithography Modeling with Dual-band Optics-Inspired Neural Networks

Fast Level-Set-Based Inverse Lithography Algorithm for Process Robustness Improvement and Its Application

Improving Fabrication Fidelity of Integrated Nanophotonic Devices Using Deep Learning

Towards smart scanning probe lithography: a framework accelerating nano-fabrication process with in-situ characterization via machine learning

A Survey on Machine Learning in Lithography

Model-driven Convolution Neural Network for Inverse Lithography

Data Efficient Lithography Modeling with Residual Neural Networks and Transfer Learning.

Applying Machine Learning Models on Metrology Data for Predicting Device Electrical Performance

Patterning, mask life, throughput and overlay improvements for high volume semiconductor manufacturing using nanoimprint lithography

Neural Lithography: Close the Design-to-Manufacturing Gap in Computational Optics with a 'Real2Sim' Learned Photolithography Simulator

Open-Source Differentiable Lithography Imaging Framework

Physics-Informed Optical Kernel Regression Using Complex-valued Neural Fields

Optimal feature vector design for computational lithography

Lithography tool improvement at productivity and performance with data analysis and machine learning

LithoGAN

From Computational Lithography to Computational Inspection: Inverse Lithography Technology (ILT) and Inverse Inspection Technology (IIT)

LithoBench: Benchmarking AI Computational Lithography for Semiconductor Manufacturing