“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

Machine learning is a powerful computer science technique that can derive knowledge from big data and make predictions/decisions. Since nanometer integrated circuits (IC) and manufacturing have extremely high complexity and gigantic data, there is great opportunity to apply and adapt various machine learning/deep learning techniques in IC design and technology co-optimization (DTCO), including physical verification and physical design. This talk will first give an overview on machine learning/deep learning, and then discuss several case studies in various DTCO applications, including lithography modeling and hotspot detection, optical proximity correction (OPC) and sub-resolution assist feature (SRAF) insertion, and physical design. It will further discuss some challenges and research directions.

Machine learning for IC design and technology co-optimization in extreme scaling

Optical proximity correction (OPC) is a critical step in semiconductor manufacturing due to its high complexity and significant influence on the subsequent process steps. Conventional OPC using the Maxwell equation can become more and more challenging as a fully vectorized three-dimensional simulation is required for advanced technology nodes. Machine learning (ML) has been a promising alternative recently. This work proposes machine-learning-assisted human decision, which can be more in line with the clean room engineer's practice and can potentially surpass pure human decision and the pure machine learning approach. Using 10-step optimization in the photolithographic mask, the averaged mean square error (MSE) at the optimized cases are 6360 and 2101 for two-bar patterns and 7132 and 5931 for tri-line attackers when comparing pure human decision and ML-assisted human. The average MSEs at the first 3 steps are 26019 and 6023 for the two-bar pattern and 79979 and 7738 for the tri-line attacker when comparing pure ML and ML-assisted human. It is suggested that the strength of the ML-assisted human decision lies in the early-stage superiority over pure ML, flexibility, incorporating past experience, and a human sense that cannot be formulated concretely by statistical models.

Optical Proximity Correction Using Machine Learning Assisted Human Decision

Optical proximity correction (OPC) is a critical step in semiconductor manufacturing due to its high complexity and significant influence on the subsequent process steps. Conventional OPC using the Maxwell equation can become more and more challenging as a fully vectorized three-dimensional simulation is required for advanced technology nodes. Machine learning (ML) has been a promising alternative recently. This work proposes machine-learning-assisted human decision, which can be more in line with the clean room engineer's practice and can potentially surpass pure human decision and the pure machine learning approach. Using 10-step optimization in the photolithographic mask, the averaged mean square error (MSE) at the optimized cases are 6360 and 2101 for two-bar patterns and 7132 and 5931 for tri-line attackers when comparing pure human decision and ML-assisted human. The average MSEs at the first 3 steps are 26019 and 6023 for the two-bar pattern and 79979 and 7738 for the tri-line attacker when comparing pure ML and ML-assisted human. It is suggested that the strength of the ML-assisted human decision lies in the early-stage superiority over pure ML, flexibility, incorporating past experience, and a human sense that cannot be formulated concretely by statistical models. 

/pdf/optical-proximity-correction-using-machine-learning-assisted-1i4ge1lg.pdf

With the continuous improvement and development of wireless sensor network, it have been enriched to a great extent. Monitoring, processing and transmitting all kinds of sensing data is its main function, so their coverage issues have received widespread attention. Among them, the WSN coverage based on DV-Hop node positioning technology has low cost and power consumption with high scalability, and is extremely widely used. However, the current error control and WSN coverage of DV-Hop node positioning are not enough for practical applications. This research innovatively adopts an improved sparrow search pattern to optimise the DV-Hop localisation algorithm. The study introduces a deviation correction factor to adjust the minimum number of node hops and uses the minimum mean squared error criterion to correct the calculation error to reduce error. In addition, the study improved the sparrow search algorithm by means of a GPS optimisation population initialisation. In the algorithm performance comparison, GSSA showed the best convergence efficiency compared to other algorithms. The average error of the GSSADV-Hop localisation constructed in the study is 0.72 m, which is 77.71% less than the traditional DV-Hop error. The study provides a reference idea for the application of DV-Hop in WSN coverage, and offers a novel solution for the optimization of DV-Hop.

Improved Sparrow Search Algorithm Optimized DV-Hop for Wireless Sensor Network Coverage

Background: In datasets for hotspot detection in physical verification, data are predominantly composed of non-hotspot samples with only a small percentage of hotspot ones; this leads to the class imbalance problem, which usually hinders the performance of classifiers.
Aim: We aim to enrich datasets by applying a data augmentation technique.
Approach: We propose a data augmentation flow-based generative adversarial network (GAN) to generate high-resolution hotspot samples.
Results: We evaluated our flow with the current state-of-the-art convolutional neural network hotspot classifier by comparison with conventional data augmentation techniques. Experimental results demonstrate that the accuracy improvement of our work can reach 3% at the same false alarm rate and the false alarm rate reduction can reach 5% at the same accuracy.
Conclusions: Our study demonstrates that rational hotspot classification can improve the efficiency of data. It also highlights the potential of GAN to generate complicated layout patterns.

Data augmentation in hotspot detection based on generative adversarial network

Mass production can be planned by utilizing the multiple patterning technology of 193 nm immersion scanners at the 7 nm technology node. In deep ultraviolet lithography, imaging performance is significantly affected by distortions of projection optics. For 7 nm immersion lithography layer patterns, distortions of the projection optics must be tightly controlled. This paper proposes an optimization method to determine the distribution of Zernike aberration coefficients. First, we build aberration prediction models using the backpropagation (BP) neural network. Then, we propose an aberration optimization method based on the sparrow search algorithm (SSA), using the common indicators of the lithography process window, depth of focus, mask error enhancement factor, and image log slope as the objective function. Some sets of optimized aberration distributions are obtained using the SSA optimization method. Finally, we compare the results of the SSA optimization algorithm with those obtained by rigorous computational simulations. The aberration combination distribution optimized by the SSA method is much more significant than the value under the zero aberration (ideal conditions), a nonoptimal distribution in deep ultraviolet lithography image simulation. Furthermore, the results indicate that the aberration optimization method has a high prediction accuracy.

Aberration analysis and compensate method of a BP neural network and sparrow search algorithm in deep ultraviolet lithography.

The miniaturization of nodes poses new challenges in semiconductor manufacturing. Optical proximity correction (OPC) is typically performed to satisfy technical requirements through iterative optimization. However, this method is expensive and slow. This study proposes a framework based on patch loss and a generative adversarial network through unsupervised learning to address these problems. The target pattern is used as the input of the model to avoid dependence on OPC tools. Thus, a fast approach is proposed for realizing OPC swiftly.

Peng Xu

Papers

Data augmentation in hotspot detection based on generative adversarial network

Aberration analysis and compensate method of a BP neural network and sparrow search algorithm in deep ultraviolet lithography.

Optical proximity correction by using unsupervised learning and the patch loss function.

Deep-learning-based image preprocessing for particle image velocimetry

Infrared target recognition with deep learning algorithms