A massively parallel GPU rasterizer for next-generation computational lithography

Computational lithography, critical for advanced semiconductor manufacturing, demands high-performance rasterization to meet nanometer-scale precision. Traditional CPU-based rasterizers struggle with the increasing complexity and data volumes of modern designs. This paper presents a massively parallel GPU rasterizer designed to accelerate high-resolution mask synthesis, lithography simulation, and optical proximity correction (OPC). Our innovative GPU-accelerated approach leverages a GPU-friendly algorithm that ensures high precision, fractional pixel coverage, and connectivity preservation for sub-pixel geometries. Benchmarking on NVIDIA H100 GPUs demonstrates significant speedups—up to 290x for Manhattan shapes and 45x for curvilinear shapes—compared to highly optimized CPU algorithms, with less than 1 percent absolute error. This methodology provides a robust solution for the demanding computational requirements of next-generation lithography, enabling faster time to market and improved design quality.

What you'll learn:

• How to implement a massively parallel GPU rasterizer for computational lithography tasks.
• The benefits of using GPU acceleration for mask synthesis, lithography simulation, and OPC.
• Benchmark results showing significant speedups compared to CPU algorithms.

Who should read this:

• Researchers and engineers interested in high-performance computing.
• Professionals in electronic design automation (EDA) and semiconductor manufacturing.
• Developers working on algorithms for computational lithography.