RFC 0003 — Standard MRC Rule-Deck Schema (v1)

Status

Accepted (May 2026). Implemented in openlithohub.benchmark.compliance.rule_deck.

Motivation

Different fabs and process nodes ship Mask Rule Check (MRC) decks in incompatible formats — Mentor Calibre SVRF, Synopsys ICV runset, vendor-specific YAML, ad-hoc Excel. None of them are easy to read or exchange across teams. When OpenLithoHub scores a mask, every contributor needs to know exactly which rules were applied and at which thresholds; re-translating SVRF for every benchmark run is brittle and audit-hostile.

This RFC defines a single JSON / TOML rule-deck format that:

1. Captures every parameter the OpenLithoHub MRC checkers consume (min_width_nm, min_spacing_nm, min_curvature_radius_nm, min_feature_area_nm2).
2. Carries enough metadata (process node, layer, mask polarity, pixel size, provenance) to be reproducible and auditable.
3. Validates against a published JSON Schema (RULE_DECK_SCHEMA) before being trusted.

It is intentionally not a full SVRF replacement. It covers what the optical / OPC side needs to score a mask. DRC-on-the-design side (transistor sizing, well rules, etc.) stays in the existing process-node configs.

Schema (v1)

{
  "schema_version": "1",
  "process_node": "freepdk45",
  "layer": "metal1",
  "polarity": "clear",
  "pixel_size_nm": 1.0,
  "rules": {
    "min_width_nm": 65.0,
    "min_spacing_nm": 65.0,
    "min_curvature_radius_nm": 25.0,
    "min_feature_area_nm2": 4225.0
  },
  "source": {
    "vendor": "NCSU FreePDK45",
    "translated_from": "manual",
    "reference": "https://eda.ncsu.edu/freepdk/freepdk45/"
  },
  "notes": "Optional free-text."
}

Field	Required	Notes
schema_version	yes	Currently "1". Bumped on breaking changes.
process_node	yes	Tech-node tag (asap7, freepdk45, n5, …).
layer	yes	Mask layer tag (metal1, poly, …).
polarity	no ("clear")	"clear" = features are 1; "dark" = features are 0.
pixel_size_nm	no	Raster pixel size; required if any rule uses physical units.
rules.min_width_nm	no	Minimum allowed feature width.
rules.min_spacing_nm	no	Minimum allowed gap between features.
rules.min_curvature_radius_nm	no	Min local radius of curvature (curvilinear).
rules.min_feature_area_nm2	no	Minimum connected-component area (curvilinear).
source	no	Provenance — vendor, original tool, citation URL.
notes	no	Free-text.

The validator rejects unknown top-level keys and unknown rule keys, so typos surface immediately rather than silently dropping rules.

SVRF → JSON translation

The Calibre SVRF lines below are the most commonly-encountered MRC constructs. Mappings to v1 fields:

SVRF	v1 JSON
EXTERNAL "metal1" < 65 nm	"rules": { "min_spacing_nm": 65.0 }
INTERNAL "metal1" < 65 nm	"rules": { "min_width_nm": 65.0 }
WIDTH "metal1" < 65 nm	"rules": { "min_width_nm": 65.0 }
SPACE "metal1" < 65 nm	"rules": { "min_spacing_nm": 65.0 }
AREA "metal1" < 4225 nm2	"rules": { "min_feature_area_nm2": 4225.0 }
RADIUS "metal1" < 25 nm CONVEX EDGE	"rules": { "min_curvature_radius_nm": 25.0 }
Layer derivation LAYER metal1 ...	"layer": "metal1"
Tech-node header comment	"process_node": "..."

Rules SVRF expresses but v1 does not yet capture (use notes for now, add to v2 if needed): enclosure, extension, end-of-line spacing, notch, jog, off-grid alignment, density windows, antenna ratios.

Synopsys ICV → JSON

ICV uses width(...), space(...), area(...), and radius(...) function-style operations on a layer expression; the parameters map directly onto the same v1 fields. The density family does not map to v1.

Loading

from openlithohub.benchmark.compliance import load_rule_deck, check_mrc

deck = load_rule_deck("rules/freepdk45_metal1.json")
result = check_mrc(mask, **deck.kwargs_manhattan())

load_rule_deck validates the file against the schema before returning a RuleDeck dataclass. kwargs_manhattan() and kwargs_curvilinear() produce kwargs for the existing check_mrc / check_curvilinear_mrc functions — only rules actually present in the deck are forwarded, so partial decks are safe.

Versioning

Backwards-incompatible changes (renamed keys, narrowed enums) bump schema_version to "2"; the loader rejects unknown major versions explicitly. Backwards-compatible additions (new optional rules) do not change schema_version; older decks continue to load.

Open questions

• Should we accept multiple layers per file (an array of decks)? Currently no — one file = one (process_node, layer) pair, which keeps diffing cleaner. Multi-layer bundles can be a follow-up if demand emerges.
• Should pixel_size_nm be required when any rule is in nm? Currently optional — defaults to whatever the checker is called with.