A Review of DNN and GPU in Optical Proximity Correction
Presenter: Xiangyu Jiang, Xidian University
Abstract: Optical Proximity Correction (OPC) is a resolution enhancement technique. It compensates for imaging distortions by modifying the mask patterns. In advanced nodes, inverse lithography technology (ILT) is used to produce more complex and finer mask shapes. However, ILT increases the computational complexity and runtime. In recent years, researchers have attempted to accelerate the process using GPU and improved the accuracy using deep neural networks. In the last ten years, the rapid advancement of deep neural networks has given rise to numerous OPC algorithms based on DNN. The high performance of GPU has provided a foundation for them. Finally, we summarize the key OPC technologies during this period and made projections for it in the future.
Model-based OPC Extension in OpenILT
Presenter: Li Xie, GWX Technology
Abstract: Optical proximity correction (OPC) is a technique to improve the accuracy of pattern transfer from the mask to the wafer in optical lithography. Model-based OPC (MB-OPC)uses mathematical models to simulate the image formation process and adjust the mask layout accordingly. In this paper,we extend the open-source computational lithography library OpenILT to support MB-OPC. The extension provides a flexible and modular framework for implementing OPC algorithms for large-scale layouts. It also supports GPU acceleration to speed up the OPC process. We demonstrate the performance and scalability of the library on different mask patterns. The experimental results show that our method can achieve more than 5 times speedup over the CPU-based MB-OPC method,while maintaining the same correction accuracy and quality. Our MB-OPC extension can provide a powerful baseline for future research on OPC.
End-to-end Lithography Modeling Based on Process Parameters and Deep Learning
Presenter: Zebang Lin, Zhejiang University
Abstract: Lithography is one of the most important processes in integrated circuit manufacturing, and with the continuous advancement of technology nodes, the cost of computational lithography is also increasing. Previous studies have mostly viewed lithography system as a black box mapping from image to image, lacking guidance on process parameters and supporting experimental data when comparing experimental results. This article models the lithography system based on process parameters and deep learning, and verifies it using experimental data. The optical system combines process parameters and uses sum of coherent systems (SOCS) to reduce the computational complexity of optical simulation. The resist system takes aerial images as input and outputs binary masks. Our proposed method has a 9.36% accuracy improvement in mean intersection over union (mIOU) compared to traditional compact resist models.
Inverse Lithography with Structured Sub-Resolution Assist Features
Abstract: With the continuous advancement of semiconductor manufacturing technology, integrated circuits now incorporate billions of mask pattern data. In the event of non-compliance with mask design rules before manufacturing, the entire mask needs to be re-optimized to address mask rule violations, incurring significant resource and time costs. In this paper, we address mask rule violations of width, space, minimum areas during the optimization process by constructing structured sub-resolution assist features (SRAFs) and incorporating them into the optimization process of inverse lithography technology (ILT) using the level-set method. An initial mask is calculated by combining the structured SRAFS and the optical proximity correction (OPC) region of the main features which excludes the SRAFs outside of the spacing spec. Level-set based ILT is further implemented incorporating mask rule checks (MRC) and corrections during the optimization process: features with spacing violations are merged while features with width violations are expanded to a structured formation; the contour of the detected non-compliant locations are fixated, while allowing the contour of the MRC-compliant features to evolve. This ensures that features already corrected will not experience new or previously encountered violations during the subsequent optimization iterations. Upon completion of the level set optimization, structured SRAFs with fixed widths will be defined. Simulation results indicate that the inclusion of mask rule check and correction dose compromise pattern fidelity with comparable performance, it can effectively solve the problem of mask rule violation. Furthermore, by constructing structured SRAFs, mask complexity is significantly reduced.
A Fast Imaging Model of Plasmonic Lithography for Line/space Patterns based on Parameter Sweep
Presenter: Huwen Ding, Institute of Microelectronics of the Chinese Academy of Sciences
Abstract: As a new and alternative lithography technology, plasmonic lithography can break through the diffraction limit of traditional lithography by exciting the surface plasmon polaritons (SPPs) to make the evanescent wave at the mask participate in imaging. The photoresist aerial image distribution of different mask patterns can be calculated by establishing an imaging model, which is the basis for understanding and further optimizing imaging. Based on the idea of machine learning and parameter sweep, a fast imaging model for plasmonic lithography is established, including periodic line/space patterns. Compared with the rigorous numerical method, the fast imaging model can greatly improve the calculation speed with high accuracy, which creates conditions for the development of computational lithography technology.
Budget analysis of multiple parameters in EUV lithography system based on support vector machine
Presenter: Jiashuo Wang, Institute of Microelectronics of the Chinese Academy of Sciences
Abstract: As one of the most critical pattern transfer technologies in semiconductor manufacturing, lithography directly affects the performance of devices or circuits. With the development of technology node, the critical dimensions of patterns are continually shrinking, which places increasingly high demands on lithography. In order to explore the influence of various parameters of scanners on the lithography results, and to provide quantitative and precise direction for the development of extreme ultraviolet (EUV) scanners, this paper proposes a methodology of budget analysis of multiple parameters based on support vector machine (SVM). First, establish a SVM classification model between parameters in lithography system and lithography results. Then, identify the parameter combinations that meet the lithography requirements according to the SVM model. Finally, calculate basic budget range for each parameter according to the statistics on the general distribution of each parameter, as well as the strict budget range according to the joint distribution between parameters. For patterns that commonly used in EUV single exposure at the 5nm technology node, we obtained the budget ranges for 8 parameters simultaneously, where the parameters come from the light source, illumination system, and projection system. Compared to the basic range of each parameter, the strict budget range of flare and dose has been reduced by more than 50%, which provides clearer goals for the development of each subsystem in EUV scanners. Since the proposed method considers the coupling effects of multiple parameters simultaneously , the results are more reliable compared to those considering the effects of every single parameter only. Furthermore, since the operations of SVM modeling and budget analysis of this method are simple and easily extendable to more parameters, this method can serve as an effective alternative for budget analysis of parameters in future research and development of EUV scanners.
A Review of DNN and GPU in Optical Proximity Correction
Presenter: Xiangyu Jiang, Xidian University