Resist / Lithography Capability Matrix

This document maps every resist and lithography forward-model implementation across the OpenLithoHub ecosystem, clarifying fidelity levels, parameter surfaces, and intended use cases so that contributors can choose the right backend and avoid accidental duplication.

Implementation	Location	Fidelity	Key Parameters	Use Case
apply_differentiable_resist	openlithohub/_utils/resist_model.py	Light	threshold, steepness, resist_diffusion_nm, quencher	Built-in default for ILT/OPC optimization loops. Dispatches to sigmoid threshold or soft CAR model depending on whether diffusion/quencher are set.
HopkinsSimulator	openlithohub/simulators/hopkins_sim.py	Medium	Hopkins SOCS kernels, partial coherence (sigma, sigma_inner), illumination shape, dose, threshold, resist_diffusion_nm, quencher, resist_backend	Core lithography forward model. Bundled reference backend with full SOCS decomposition. Supports opt-in acid diffusion and DiffNano resist backend.
CalibreSimulator	openlithohub/simulators/calibre.py	External	config.extra["calibre_home"], config.extra["runset"], config.extra["mock_mode"], dose, threshold	Siemens EDA Calibre nmOPC adapter. Requires commercial toolchain + license; mock mode for testing.
TachyonSimulator	openlithohub/simulators/tachyon.py	External	config.extra["tachyon_home"], config.extra["recipe"], config.extra["mock_mode"], dose, threshold	ASML Brion Tachyon adapter. Requires commercial toolchain + license; mock mode for testing.
DiffCFDLithoSimulator (Dill + Mack)	openlithohub/plugins/diffcfd_process.py	High	dill_A, dill_B, dill_C, r_max, r_min, mack_n, mack_a, gamma_solvent, thickness_m, residual_solvent, dev_time_s	High-fidelity exposure + development chain via diffcfd plugin. Requires [diffcfd] extra.
DiffNanoResistAdapter (CAR/PEB)	openlithohub/plugins/diffnano_resist.py	High	acid_diffusion_length_nm, development_contrast, threshold_dose, peb_diffusion_nm, pixel_size_nm	High-fidelity chemically-amplified resist with acid diffusion + PEB via diffnano plugin. Requires [diffnano] extra.
HopkinsLithoModel	diffnano/solvers/litho.py	Medium	wavelength_nm, na, sigma_source, n_kernels, pixel_size_nm, resist_threshold, resist_beta	Hopkins PSF model in DiffNano. Functionally equivalent to OpenLithoHub's HopkinsSimulator for the PSF convolution + sigmoid resist path. See dedup note below.
LithoSolver (Dill + Mack)	diffcfd/solvers/litho.py	High	dill_A, dill_B, dill_C, r_max, r_min, mack_n, mack_a, gamma_solvent	Dill exposure + Mack development solver in DiffCFD. Wrapped by DiffCFDLithoSimulator in OpenLithoHub.
DifferentiableResistModel	diffnano/solvers/resist.py	High	grid_shape, dl, acid_diffusion_length_nm, development_contrast, threshold_dose, peb_diffusion_nm	Full CAR/PEB analytical model in DiffNano. Takes (H, W) grid shape and grid spacing dl (nm). Wrapped by DiffNanoResistAdapter in OpenLithoHub, which adds pixel_size_nm for nm-to-pixel conversion.

Which Model Should I Use?

Scenario	Recommended Model	Rationale
ILT gradient descent on 64x64 synthetic tiles	apply_differentiable_resist (Light)	Fast, differentiable, no external dependencies. The scored leaderboard path uses this.
Mask optimization with realistic aerial images	HopkinsSimulator (Medium)	Full SOCS decomposition with partial coherence. Still uses simple threshold resist.
Validation against commercial tool	CalibreSimulator or TachyonSimulator (External)	Uses vendor forward simulator for ground-truth comparison. Requires commercial toolchain + license; mock_mode for CI testing.
Process-window analysis, CD prediction	DiffCFDLithoSimulator via [diffcfd] (High)	Physics-based Dill exposure + Mack development. Models PAC bleaching, dissolution rate, residual solvent. Requires calibrated Dill/Mack parameters from the target process.
Chemically-amplified resist modeling	DiffNanoResistAdapter via [diffnano] (High)	CAR acid diffusion + post-exposure bake + development contrast. Suitable for EUV CAR resists where acid diffusion dominates print fidelity. Parameters calibratable to SEM CD data.
Joint spin-coating + lithography co-optimization	DiffCFDLithoSimulator + DiffCFD spin-coating solver	The only path that couples film thickness variation (from spin coating) into the exposure/development chain.
DiffNano-internal metalens with litho DFM	HopkinsLithoModel in DiffNano	Uses simplified Gaussian PSF rather than full SOCS. Retained for DiffNano standalone workflows; prefer OpenLithoHub's HopkinsSimulator for ILT/OPC work.

Important: Plugin backends (DiffCFD, DiffNano) produce non-comparable metric values. The scored leaderboard default remains HopkinsSimulator + CTR (threshold 0.225) without diffusion.

Fidelity Levels

• Light -- Sigmoid threshold with optional Gaussian acid diffusion. No physics-based exposure or dissolution model. Suitable for ILT gradient descent where speed matters more than absolute CD accuracy.
• Medium -- Hopkins/SOCS aerial image with physically correct partial coherence and dose scaling. Resist model is still a simple threshold or sigmoid. Good for mask-optimization loops that need realistic aerial images.
• High -- Full physics-based models (Dill exposure, Mack development, CAR acid diffusion + PEB). Suitable for process-window analysis, CD prediction, and joint process optimization.
• External -- Vendor simulator (Calibre nmOPC, ASML Tachyon). Highest fidelity but requires commercial toolchain + license. Mock mode available for API testing without the vendor binary.

Deduplication Notes

Hopkins / PSF Convolution

HopkinsLithoModel in DiffNano (diffnano/solvers/litho.py) implements the same PSF convolution + sigmoid resist pipeline as OpenLithoHub's HopkinsSimulator + apply_differentiable_resist. The DiffNano version uses a simplified Gaussian PSF approximation rather than full SOCS decomposition. For new ILT/OPC work, prefer OpenLithoHub's HopkinsSimulator which supports multiple illumination types and proper SOCS kernel caching.

Dill / Mack Solvers

LithoSolver in DiffCFD is the canonical implementation. OpenLithoHub wraps it via DiffCFDLithoSimulator (the adapter) and should not reimplement the physics.

CAR / PEB Resist Models

DifferentiableResistModel in DiffNano is the canonical implementation. OpenLithoHub wraps it via DiffNanoResistAdapter and should not reimplement the physics.