Data Efficient Lithography Modeling With Transfer Learning and Active Data Selection

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.

Re-examining VLSI manufacturing and yield through the lens of deep learning

“…Machine learning-based techniques have been proposed as a substitute for compact models for better simulation quality [2], [10], [11], [44]. [44] proposed an artificial neural network (ANN) for resist height prediction.…”

“…[10] proposed a convolutional neural network (CNN) model that predicts the slicing thresholds in aerial images accurately. Recently, [11] proposed a transfer learning model to cope with the deficiency in the manufacturing data at advanced technology nodes.…”

“…However, the steady decrease of the feature sizes along with the growing complexity and variation of the manufacturing process have tremendously increased the lithography modeling complexity and prolonged the already-slow simulation procedure. Considering the fact that machine learning based approaches have demonstrated superior efficacy in a particular stage during lithography modeling [10], [11], we have recently proposed LithoGAN [12] as a novel end-to-end lithography modeling framework based on conditional generative adversarial network (CGAN) that has demonstrated tremendous success in computer vision over the past few years [13]- [17].…”

Accurate prediction of EUV lithographic images and 3D mask effects using generative networks

“…The data requisites of machine learning can largely impede the efficiency when the training data involves rigorously simulated 3D mask model aerial images. Lin et al 11 proposed an approach that tackles the data inefficiency by implementing active data selection in combination with transfer learning, where data from a simpler technology node is utilized to reduce the amount of training data the network required from the more advanced technology node.…”

Neural-ILT

“…In the past decade, both academia and industry have been actively working on facilitating the conventional lithography-related processes as well as maintaining competitive QoR. Significant efforts have been made, including but not limited to (1) migrating highperformance computational lithography to GPU acceleration [9]; (2) introducing fast modeling approaches for rigorous/compact lithosimulations [10]; (3) considering multiple patterning [11,12] and (4) applying the SOTA machine learning techniques on lithographyrelated applications such as lithography system modeling [13,14], hotspot detection [15ś17] and OPC [13, 18ś20]. Among them, Yang et al [19] (GAN-OPC) for the first time applied conditional generative adversarial networks (CGAN) to mimic the process of typical ILT OPC.…”