Beyond the EPE Budget: Building the Next Generation of Computational Lithography
As semiconductor manufacturing pushes toward the 2nm node and beyond, the edge placement error (EPE) budget has become a central constraint. Traditional computational lithography methods are straining to keep pace with the tightening tolerances demanded by advanced patterning. In 2026, the industry is rethinking the approach—moving beyond incremental EPE improvements to fundamentally new architectures for computational lithography.
The Shrinking EPE Budget
The EPE budget—the allowable deviation between designed and printed features—has shrunk dramatically with each node. At 2nm and below, stochastic effects, mask 3D effects, and process variations consume nearly all available margin. Conventional optical proximity correction (OPC) and inverse lithography technology (ILT) alone can no longer guarantee sufficient process windows.
Why Traditional Methods Fall Short
Traditional computational lithography relies on deterministic models and rule-based corrections. While these have served the industry well, they struggle with:
- Stochastic failures at EUV limits, where random photon and resist effects cause defects.
- Complex 2D and 3D geometries that exceed the capability of rule-based OPC.
- Multi-patterning interactions that introduce new sources of overlay and CD error.
The Next Generation: Data-Driven and Holistic Approaches
The next generation of computational lithography is characterized by:
1. Machine Learning and AI-Driven Correction
ML models trained on vast process data can predict and compensate for stochastic and systematic errors more effectively than traditional models. In 2026, AI-driven OPC and ILT are becoming production-ready, enabling faster convergence and better EPE control.
2. Co-Optimization Across the Flow
Rather than treating lithography as an isolated step, next-gen computational lithography co-optimizes design, mask, and process. This holistic approach—sometimes called design-technology co-optimization (DTCO) extended to manufacturing—allows trade-offs that minimize EPE across the entire flow.
3. Stochastic-Aware Modeling
New models explicitly account for stochastic effects, using probabilistic methods to predict defectivity and guide correction. This is essential for EUV high-NA and beyond.
4. High-NA EUV Readiness
With high-NA EUV entering early production in 2026, computational lithography must handle the changed imaging conditions, including tighter pitches and new mask architectures.
Conclusion
The EPE budget is no longer just a number to be met—it is a driver for innovation. The next generation of computational lithography will be defined by AI, holistic co-optimization, and stochastic-aware methods. As the industry moves beyond 2nm, these advances will be critical to sustaining Moore's Law.