MorphOPC: Neural Morphological Learning for Mask Optimization
A recent study published on arXiv (2605.12528v1) presents MorphOPC, a hierarchical model that operates on multiple scales and employs neural morphological modules to learn geometric transformations essential for optical proximity correction (OPC) in semiconductor production. With feature sizes decreasing to the nanometer range, precise pattern transfer from photomasks to wafers is vital. Current generative encoder-decoder models often struggle to accurately capture these transformations, resulting in inferior mask quality. MorphOPC approaches mask generation through a series of morphological operations applied to local layout features. Tests on benchmarks for edge-based OPC and inverse lithography technology (ILT) across metal and via layers demonstrate its consistent superiority, tackling a significant challenge in advanced lithography for chip manufacturing.
Key facts
- MorphOPC is a multi-scale hierarchical model with neural morphological modules.
- It learns geometric transformations from target layouts to mask patterns.
- The model outperforms existing generative encoder-decoder approaches.
- Experiments were conducted on edge-based OPC and ILT benchmarks.
- Benchmarks covered metal and via layers.
- The paper is published on arXiv with identifier 2605.12528v1.
- Optical proximity correction ensures pattern fidelity in semiconductor manufacturing.
- Feature sizes are shrinking to nanometer scale.
Entities
Institutions
- arXiv