Making digital twins using the Deep Learning Kit (DLK)

In lithography, optical proximity and process bias/effects need to be corrected to achieve the best wafer print. Efforts to correct for these effects started with a simple bias, adding a hammer head in line-ends to prevent line-end shortening. This first-generation correction was called rule-based optical proximity correction (OPC). Then, as chip feature sizes continued to shrink, OPC became more complicated and evolved to a model-based approach. Some extra patterns were added to masks, to improve the wafer process window, a measure of resilience to manufacturing variation. Around this time, the concept of inverse lithography technology (ILT), a mathematically rigorous inverse approach that determines the mask shapes that will produce the desired on-wafer results, was introduced. ILT has been explored and developed over the last three decades as the next generation of OPC, promising a solution to several challenges of advanced-node lithography, whether optical or extreme ultraviolet (EUV). Today, both OPC and ILT are part of an arsenal of lithography technologies called resolution enhancement technologies. Since OPC and ILT both involve computation, they are also considered as part of computational lithography. We explore the background and history of ILT and detail the significant milestones that have taken full-chip ILT from an academic concept to a practical production reality.

Section: Applying Deep Learning To Iltmentioning

confidence: 99%

Inverse lithography technology: 30 years from concept to practical, full-chip reality

Pang

2021

“…In consequence of the continuous development of AI-techniques and specifically deep learning-techniques, another powerful approach is increasingly applied: the representation of digital twins with generative adversarial networks. 28 In the case of SEM images, the idea of this approach is to generate output data in the style of an SEM image from arbitrary input data. To do so, the network has to learn how SEM images look.…”

Section: Artificial Training Datamentioning

confidence: 99%

“…This opens the field for another kind of deep learning application: the creation of digital twins for the efficient generation of such data. 28,29 One example is the generation of SEM images with defects from SEM images without defects. 30 In this work, the digital twin approach is used to generate simplified SEM images with defects from mask pattern data.…”

Section: Introductionmentioning

confidence: 99%

Mask defect detection with hybrid deep learning network

Evanschitzky

Auth

Heil

et al. 2021

Background: Deep learning is a very fast-growing field in the area of artificial intelligence with remarkable results in recent years. Many works in the lithography and photomask field have shown progress of technology in this application area and the potential to improve lithography by the automation of processes. Despite this progress, the use of machine learning techniques in the field of mask repair still seems to be at the beginning.Aim: We show that deep learning-based methods can successfully be applied to mask repair applications.Approach: The presented system is a hybrid and modular approach based on a combination of several deep learning networks and analytical methods, enabling the detection of mask pattern defects and the determination of the exact defect shapes from SEM images. In the current version, the system is trained for line/space patterns and contact patterns with typical defect types. The modularity allows for the extensibility to new use cases. The issue of an insufficient amount of training data is addressed using purely computer-generated simplified SEM data in combination with a specific network architecture. Results:The very good functionality and defect detection accuracy of the system are demonstrated with a set of real SEM images with line/space patterns and contact patterns with numerous defects. In particular, a 100% true defect detection rate could be obtained. Conclusions:The presented machine learning approach demonstrates the successful defect identification, location, and shape determination from real mask SEM images.

“…Several regression-based OPC methods were conducted to learn edge displacement from fragmented target layout with linear algorithm, nonlinear algorithms, or hierarchical Bayes model 1 – 3 Another mainstream was mask pattern generation by utilizing U-Net architecture or generative adversarial networks (GANs) 4 – 6 Generally, the latter approach produces curvilinear mask patterns that best mimic the ILT technique, but critical dimension (CD) accuracy in practical production are less investigated.…”

Section: Introductionmentioning

confidence: 99%

“…[1][2][3] Another mainstream was mask pattern generation by utilizing U-Net architecture or generative adversarial networks (GANs). [4][5][6] Generally, the latter approach produces curvilinear mask patterns that best mimic the ILT technique, but critical dimension (CD) accuracy in practical production are less investigated.…”

Section: Introductionmentioning

confidence: 99%

Machine learning optical proximity correction with generative adversarial networks

Ciou

Tsai

et al. 2022

Background: Algorithmic breakthroughs in machine learning (ML) have allowed increasingly more applications developed for computational lithography, gradually shifting focus from hotspot detection to inverse lithography and optical proximity correction (OPC). We proposed a pixelated mask synthesis method utilizing deep-learning techniques, to generate afterdevelopment-inspection (ADI) contour and mask feature generation.Aim: Conventional OPC correction consists of two parts, the simulation model which predicts the expected contour signal, and the correction script that modifies the actual layout. With practicality in mind, we collected modeling wafer data from scratch, then implemented ML models to reproduce conventional OPC actions, mask to contour prediction, and design to mask correction.Approach: Two generative adversarial networks (GANs) were constructed, a pix2pix model was first trained to learn the correspondences between mask image and paired ADI contour image collected on wafer. The second model is embedded into machine learning mask correction (ML-OPC) framework, output mask is optimized through minimizing pixel difference between design target and simulated contour.Results: Two different magnification SEM image datasets were collected and studied, with the higher magnification showing better simulator pixel accuracy. Supervised training of the correction model provided a quick prototype mask synthesis generator, and combination of unsupervised training allowed mask pattern synthetization from any given design layout. Conclusions:The experimental results demonstrated that our ML-OPC framework was able to mimic conventional OPC model in producing exquisite mask patterns and contours. This ML-OPC framework could be implemented across full chip layout.