Open-Source Differentiable Lithography Imaging Framework (2409.15306v1)

Abstract: The rapid evolution of the electronics industry, driven by Moore's law and the proliferation of integrated circuits, has led to significant advancements in modern society, including the Internet, wireless communication, and AI. Central to this progress is optical lithography, a critical technology in semiconductor manufacturing that accounts for approximately 30\% to 40\% of production costs. As semiconductor nodes shrink and transistor numbers increase, optical lithography becomes increasingly vital in current integrated circuit (IC) fabrication technology. This paper introduces an open-source differentiable lithography imaging framework that leverages the principles of differentiable programming and the computational power of GPUs to enhance the precision of lithography modeling and simplify the optimization of resolution enhancement techniques (RETs). The framework models the core components of lithography as differentiable segments, allowing for the implementation of standard scalar imaging models, including the Abbe and Hopkins models, as well as their approximation models. The paper introduces a computational lithography framework that optimizes semiconductor manufacturing processes using advanced computational techniques and differentiable programming. It compares imaging models and provides tools for enhancing resolution, demonstrating improved semiconductor patterning performance. The open-sourced framework represents a significant advancement in lithography technology, facilitating collaboration in the field. The source code is available at https://github.com/TorchOPC/TorchLitho

• The paper introduces an open-source differentiable lithography imaging framework that integrates GPU acceleration and adjoint back-propagation for efficient optimization.
• It achieves state-of-the-art results with an average mPA of 99.45% and mIOU of 99.23%, outperforming traditional benchmarks.
• The framework promotes collaborative research and paves the way for advanced lithography simulations in semiconductor manufacturing.

Open-Source Differentiable Lithography Imaging Framework: An Expert Overview

The paper, "Open-Source Differentiable Lithography Imaging Framework," introduces an innovative framework in the domain of semiconductor manufacturing, central to the electronics industry's advancement underpinned by Moore's Law. The proposed framework leverages differentiable programming and GPU computational power to enhance the precision and efficiency of optical lithography, a crucial technology accounting for a significant portion of production costs in integrated circuit (IC) manufacturing.

Framework Overview

This work presents a framework that incorporates differentiable programming within computational lithography—a field that amalgamates optical and chemical processes with advanced computational methods to optimize semiconductor patterning. The framework supports both Abbe and Hopkins imaging models, known for their utility in partial coherence illumination scenarios. Notably, the framework is open-source, facilitating wider collaboration and development within the field, with its code accessible via GitHub.

Numerical Results and Performance

The paper demonstrates the framework's efficacy through rigorous optical modeling, showing its capability to handle both via and metal layers simultaneously, with configurable optical parameters and GPU acceleration. It advances beyond traditional methods by integrating neural network technologies, showcasing substantial improvements in Mean Pixel Accuracy (mPA) and Mean Intersection Over Union (mIOU) across multiple benchmarks. Specifically, the framework achieves an average mPA of 99.45% and mIOU of 99.23%, outperforming existing models like DAMO, TEMPO, and DOINN in accuracy and precision metrics.

Technical and Theoretical Contributions

Technically, the differentiable lithography engine introduces adjoint back-propagation for efficient memory usage in large-scale optimization problems, enabling the model's parameters to be fine-tuned via gradient descent. This method is pivotal for optimizing source shape, lens design, mask parameters, and neural network modules jointly, thereby enhancing the lithographic simulation's fidelity.

Theoretically, the framework contributes to the refinement of image simulation techniques and resolution enhancement strategies (RETs), such as Optical Proximity Correction (OPC) and Source Mask Optimization (SMO). The integration of forward and backward automatic differentiation within the framework allows for optimized computational paths, reducing inefficiencies and improving the precision of the lithographic process.

Implications and Future Prospects

The implications of this research are manifold. Practically, the open-source nature of the framework is poised to accelerate advancements in computational lithography by promoting collaborative enhancements and adaptability across varying semiconductor manufacturing processes. Theoretically, the framework sets a foundation for further exploration into differentiable programming's capabilities in resolving complex lithographic challenges.

Looking forward, the framework's integration with machine learning tools such as GANs and neural network-driven models highlights a potential trajectory for expanding lithography simulations and enhancing future IC design processes. The adaptability and precision offered by the framework pave the way for innovation in semiconductor manufacturing technologies, aligning with the industry's drive towards miniaturization and efficiency under Moore's Law.

By sharing this advanced tool with the broader research community, the authors contribute significantly to the trajectory of machine learning and computational imaging in semiconductor technology, setting the stage for future developments in this critical field.