DEEP LEARNING-BASED LITHOGRAPHIC HOTSPOT DETECTION FOR ENHANCED SEMICONDUCTOR DESIGN

The semiconductor manufacturing industry is vital to modern technology, powering devices from smartphones to supercomputers. A critical challenge within this industry is the detection of lithographic hotspots—areas in integrated circuit designs that are prone to manufacturing defects. Traditional methods for hotspot detection, primarily rule-based and statistical approaches, often struggle to address the complexities of contemporary IC designs, leading to potential yield losses and compromised device performance. This paper proposes a deep learning-based framework for lithographic hotspot detection, leveraging convolutional neural networks to analyze design data more effectively than conventional methods. By integrating both simulated and real-world datasets, the proposed model significantly enhances detection accuracy and generalization capabilities across various design scenarios. Furthermore, this research explores multi-task learning, allowing the model to not only identify hotspots but also predict design rule violations, thereby streamlining the design process. The findings indicate that deep learning techniques can revolutionize hotspot detection, providing a robust solution that meets the increasing demands for smaller and more efficient semiconductor devices. This work contributes to the field by offering a comprehensive framework that enhances the efficiency and effectiveness of semiconductor design and manufacturing processes, paving the way for future advancements in the industry.