analog-simulation-metrics

Analog Simulation Metrics

A Codex skill for token-efficient analog IC simulation loops.

Instead of reading entire Spectre logs, ADE outputs, netlists, or waveform dumps into the chat, this skill pushes the workflow toward small metric samplers: scripts or calculator expressions that return only the values needed for design decisions.

Why

Analog design iteration often produces huge artifacts:

• simulator logs
• PSF waveform databases
• ADE Explorer/Assembler runs
• Monte Carlo tables
• transient waveform exports

Most sizing decisions only need a few numbers: DC gain, UGF, phase margin, power, common-mode level, settling time, offset, PSRR, noise, or a pass/fail diagnosis. This skill helps Codex ask for those numbers first, then inspect detailed data only when something fails or looks inconsistent.

What It Contains

• SKILL.md: the Codex skill instructions.
• scripts/metric_sampler.py: a small Python helper for extracting compact JSON metrics from Spectre logs or CSV AC data.
• agents/openai.yaml: optional UI metadata for Codex.

Examples

Successful AC Metric Extraction

Parse exported AC data:

python scripts/metric_sampler.py ac-csv ac.csv --freq freq --mag gain_db --phase phase_deg

Example CSV:

freq,gain_db,phase_deg
1,72.8,-0.1
1000000,40.2,-72.4
100000000,0.0,-118.5
1000000000,-22.6,-171.0

Example output:

{
  "analysis": "ac-csv",
  "status": "ok",
  "metrics": {
    "dc_gain_db": 72.8,
    "phase_at_ugf_deg": -118.5,
    "unity_gain_frequency_hz": 100000000.0
  }
}

STB Failure Diagnosis

Parse a Spectre STB log:

python scripts/metric_sampler.py spectre-log path\to\spectre.out

Example output:

{
  "analysis": "stb",
  "status": "fail",
  "metrics": {
    "gain_margin_db": null,
    "phase_margin_deg": null,
    "phase_zero_crossing_hz": 26302700.0
  },
  "diagnosis": "Spectre reported positive-feedback instability: phase crosses 0 deg while |loopGain| > 1."
}

Installation

Copy this folder into your Codex skills directory:

C:\Users\<you>\.codex\skills\analog-simulation-metrics

Then ask Codex to use $analog-simulation-metrics when iterating analog simulations.

Design Philosophy

• Read summaries before raw data.
• Read warning lines before full logs.
• Export crossing points before full waveforms.
• Preserve simulator diagnoses when metrics are undefined.
• Keep a small history table of design variables and resulting metrics.

The goal is not to replace deep circuit judgment. The goal is to spend context on decisions, not on bulky simulation artifacts.