I am working on a local Auto-ADK evaluation that generates 3D-IC interface collateral and then validates it through OpenROAD-flow-scripts. I need to avoid treating self-consistent OpenRCX output as calibrated RC evidence.
Current local evidence:
- Source OpenRCX rules are available and executable for public ORFS platforms. In my current run I see source-RCX SPEF/report evidence for
nangate45, sky130hd, and gf180.
- The local RC calibration boundary scan sees 50 public
gf180 source OpenRCX rule candidates, plus nangate45/rcx_patterns.rules and sky130hd/rcx_patterns.rules in the broader extraction evidence.
- The same scan finds 0 foundry-qualified signoff extraction deck candidates and 0 golden/reference SPEF or golden extractor replay candidates for calibrating a generated 3D interface stack.
Request:
- Is any public OpenROAD-flow-scripts artifact considered an owner-approved reference for RC calibration of generated 3D interface collateral, beyond ordinary source OpenRCX rule execution?
- If yes, which deck/file set, corner mapping, and comparison thresholds should be used for net capacitance, resistance, and coupling capacitance error metrics?
- If no, is there a recommended public reference SPEF, extractor replay, or documented owner policy that downstream generated-collateral experiments should use to distinguish calibration from self-consistent extraction?
The specific blockers I am trying to resolve are:
- foundry-qualified or owner-approved extraction deck
- golden/reference SPEF or golden extractor replay
- corner mapping from public OpenRCX corners to accepted reference corners
- comparison threshold policy for RC metrics
- owner acceptance of generated interface layer remapping when source platform RCX rules are reused
Non-claim: I am not asking ORFS to sign off this generated 3D collateral. I am trying to keep a local experiment honest by not calling public OpenRCX SPEF generation "calibrated RC" unless there is an appropriate reference artifact or owner guidance.
I am working on a local Auto-ADK evaluation that generates 3D-IC interface collateral and then validates it through OpenROAD-flow-scripts. I need to avoid treating self-consistent OpenRCX output as calibrated RC evidence.
Current local evidence:
nangate45,sky130hd, andgf180.gf180source OpenRCX rule candidates, plusnangate45/rcx_patterns.rulesandsky130hd/rcx_patterns.rulesin the broader extraction evidence.Request:
The specific blockers I am trying to resolve are:
Non-claim: I am not asking ORFS to sign off this generated 3D collateral. I am trying to keep a local experiment honest by not calling public OpenRCX SPEF generation "calibrated RC" unless there is an appropriate reference artifact or owner guidance.