Abstract:Lithography simulation is one of the key steps in physical verification, enabled by the substantial optical and resist models. A resist model bridges the aerial image simulation to printed patterns. While the effectiveness of learning-based solutions for resist modeling has been demonstrated, they are considerably data-demanding. Meanwhile, a set of manufactured data for a specific lithography configuration is only valid for the training of one single model, indicating low data efficiency. Due to the complexity of the manufacturing process, obtaining enough data for acceptable accuracy becomes very expensive in terms of both time and cost, especially during the evolution of technology generations when the design space is intensively explored. In this work, we propose a new resist modeling framework for contact layers, utilizing existing data from old technology nodes and active selection of data in a target technology node, to reduce the amount of data required from the target lithography configuration. Our framework based on transfer learning and active learning techniques is effective within a competitive range of accuracy, i.e., 3-10X reduction on the amount of training data with comparable accuracy to the state-of-the-art learning approach.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the low data efficiency in lithography simulation. Specifically, although the existing learning - based resist modeling methods can achieve high precision, they require a large amount of training data, and this data can only be used for specific lithography configurations, resulting in low data utilization efficiency. As the technology nodes continue to evolve (for example, from 10 nanometers to 7 nanometers), obtaining sufficient training data becomes very expensive and time - consuming. To solve these problems, the author proposes a new resist modeling framework, aiming to reduce the amount of training data required for the target lithography configuration in the following two ways: 1. **Transfer Learning**: Utilize the data and knowledge from old technology nodes and transfer them to the models of new technology nodes. 2. **Active Learning**: Actively select the most representative data samples in the target technology node for labeling. ### Main Contributions - Propose a high - performance resist modeling technique based on Residual Neural Network (ResNet). - Design a transfer learning scheme to reduce the amount of data required to reach the target precision by using data from other configurations. - Propose an active learning scheme based on the K - Medoids algorithm, which is suitable for CNN and ResNet in combination with theoretical analysis. - The experimental results show that compared with the state - of - the - art learning methods, this method can reduce the amount of training data by 3 to 10 times while maintaining similar precision. ### Problem Definition The paper defines two main problems: 1. **Learning - based Resist Modeling Problem**: Given a data set containing aerial images and their central position thresholds, train a resist model to maximize the precision of predicting thresholds. 2. **Data - efficient Resist Modeling Problem**: Given a labeled N10 data set containing aerial images and thresholds, and an unlabeled N7 data set containing only aerial images, train a resist model for the N7 data set so that it queries the labels of N7 data samples as little as possible while achieving high precision. ### Definition of Data Efficiency Data efficiency refers to the amount of target - domain data required to learn a model under a given precision. If a model can obtain higher precision than another model at a specific amount of data, it means that this model has higher data efficiency. ### Training Process Figure 5 shows the overall training process: 1. Use the labeled source - domain data to train the source - domain model. 2. Before training the target - domain model, apply active learning to select data samples for label query. 3. Finally, use the selected data samples and the knowledge transferred from the source - domain model to train the target - domain model. In this way, this method not only improves the precision of the model, but also significantly reduces the amount of required training data, thereby enhancing data efficiency.

Data Efficient Lithography Modeling with Transfer Learning and Active Data Selection

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

Lithography Hotspot Detection Based on Heterogeneous Federated Learning with Local Adaptation and Feature Selection

Lithography Hotspot Detection Via Heterogeneous Federated Learning with Local Adaptation

LithoGAN

Computational Lithography Using Machine Learning Models

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

An Adversarial Active Sampling-based Data Augmentation Framework for Manufacturable Chip Design

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

Automated Lithography Resolution Enhancement with Deep Learning Enabled Layout Modification During Physical Design Stage

Efficient Creation of Jettability Diagrams Using Active Machine Learning

Deep Learning for Mask Synthesis and Verification: A Survey.

Efficient ILT via Multi-level Lithography Simulation.

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

Optimal feature vector design for computational lithography

Subresolution Assist Feature Generation with Supervised Data Learning

DIRECT: An Efficient Optimization Scheme for Mask Generation Using Inverse Lithography

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

LithoBench: Benchmarking AI Computational Lithography for Semiconductor Manufacturing

Hybrid physics-guided data-driven modeling for generalizable geometric accuracy prediction and improvement in two-photon lithography

Model-driven Convolution Neural Network for Inverse Lithography