behavioral-modeling
/behavioral-modeling
Author, compile, and validate analog behavioral models — Verilog-A, Verilog-AMS, VHDL-AMS, and SystemVerilog real-number models — plus the connect modules that bridge analog and digital domains. This is the analog-HDL core of the marketplace. Use when building a fast, accurate model of an analog block for AMS verification or system simulation.
- Author
- hdl-tools ↗
- Repository
- analog-chip-design-agents ↗
- Licence
- MIT
- Last repository change
- 09.06.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 09.06.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Author, compile, and validate analog behavioral models — Verilog-A, Verilog-AMS, VHDL-AMS, and SystemVerilog real-number models — plus the connect modules that bridge analog and digital domains. This is the analog-HDL core of the marketplace. Use when building a fast, accurate model of an analog block for AMS verification or system simulation.
Original author description
Author, compile, and validate analog behavioral models — Verilog-A, Verilog-AMS, VHDL-AMS, and SystemVerilog real-number models — plus the connect modules that bridge analog and digital domains. This is the analog-HDL core of the marketplace. Use when building a fast, accurate model of an analog block for AMS verification or system simulation.
How it works
- Invocation
- Pre-run Context
- Purpose
- Supported EDA Tools
- Stage: modelplanning
- Stage: vaauthoring
- Stage: modelcompilation
- Stage: connectrulesetup
- Stage: modelvalidation
- Stage: modelsignoff
SKILL.md preview
--- name: behavioral-modeling description: > Author, compile, and validate analog behavioral models — Verilog-A, Verilog-AMS, VHDL-AMS, and SystemVerilog real-number models — plus the connect modules that bridge analog and digital domains. This is the analog-HDL core of the marketplace. Use when building a fast, accurate model of an analog block for AMS verification or system simulation. version: 1.0.0 author: chuanseng-ng license: MIT allowed-tools: Read, Write, Bash --- # Skill: Behavioral / AMS Modeling ## Invocation - **If invoked by a user** presenting a modeling task: immediately spawn the `analog-chip-design-agents:behavioral-modeling-orchestrator` agent and pass the full user request and any available context. Do not execute stages directly. - **If invoked by the `behavioral-modeling-orchestrator` mid-flow** (including fix-request servicing): do not spawn a new agent. Treat this file as read-only — return the requested stage rules, sign-off criteria, or loop-back guidance. Spawning the orchestrator from within an active orchestrator run causes recursive delegation and must never happen. ## Pre-run Context Before executing or advising on **any** stage, read the following if they exist: 1. `memory/modeling/knowledge.md` — known Verilog-A idioms, OpenVAF/ADMS quirks, convergence recipes, RNM pitfalls. Incorporate its guidance into every authoring decision. 2. `memory/modeling/run_state.md` — current run identity (`run_id`, `design_name`, `pdk`, `last_stage`) for resume-after-interruption. ## Purpose Author, compile, and validate analog behavioral models and the connect modules that bridge analog and digital domains — the "analog HDL" capability. Six stages with explicit QoR gates and loop-back criteria enforced by the behavioral-modeling orchestrator. A model is only signed off when it is both **accurate** (vs a SPICE reference) and **faster** than SPICE. --- ## Supported EDA Tools ### Open-Source - **OpenVAF** (`openvaf`) — Verilog-A → OSDI compiler for ngspice / Xyce; the primary path - **ADMS** (`admsXml`) — legacy Verilog-A → C model generation - **ngspice** (`ngspice`) / **Xyce** (`Xyce`) — OSDI loading + model-vs-SPICE co-sim - **SystemVerilog RNM** (`nettype`/`wreal`) via **Verilator** (`verilator`) / **Icarus** (`iverilog`) - **Hdl21 + VLSIR** (`python -m hdl21`) — programmatic (Python) analog HDL - **GHDL-AMS** (`ghdl`) — limited VHDL-AMS support ### Proprietary (detect-only — never installed) - **Cadence AMS Designer / Xcelium AMS** (`xrun`) — Verilog-AMS / Verilog-A reference - **Spectre Verilog-A** (`spectre`) — Verilog-A in Spectre - **Synopsys VCS-AMS / CustomSim** (`vcs`) — AMS co-sim - **Siemens Symphony / Symphony Pro** — AMS simulation --- ## Stage: model_planning ### Domain Rules 1. Choose the modeling language per use: **Verilog-A** (→ OSDI via OpenVAF for ngspice/Xyce) for continuous analog behaviour with conservation laws; **SystemVerilog RNM** (`nettype`/`wreal`) for fast event-driven signal-flow; **VHDL-AMS** (GHDL-AMS, limited) only when mandated. 2. Define the model's port natures (electrical vs real/wreal vs logic) and the abstraction level (functional / behavioural-with-parasitics) up front. 3. Record the target model-vs-SPICE error tolerance and the parameters that map to `design_state.constraints.specs` (gain, BW, noise, offset) so the model is spec-traceable. 4. Identify the conserved quantities and independent sources the model must expose, and which analog↔digital boundaries (if any) will need connect modules. ### QoR Metrics to Evaluate - Language choice justified for the use case - Spec-bearing outputs identified and traced to `constraints.specs`/`rf_specs` - Error tolerance and abstraction level fixed before authoring ### Common Issues & Fixes | Issue | Fix | |-------|-----| | RNM chosen for a conservation-law block | Switch to Verilog-A (RNM cannot carry KCL/KVL) | | No SPICE reference identified | Pick the circuit-design netlist (`design_state.circuit.netlist`) or a reference cell | | Tolerance left implicit | Fix the error tolerance from the modelled spec before authoring | ### Output Required - Model plan (language, port natures, abstraction, error tolerance) - Spec-traceability map (model output → constraint key) --- ## Stage: va_authoring ### Domain Rules 1. Write Verilog-A per the LRM: declare `electrical` disciplines, use `branch`/contribution operators (`<+`), and gate parameter computation with `@(initial_step)` — never recompute constants every evaluation. 2. Ensure continuity and differentiability of all branch contributions; use `$limexp`, `tanh`-smoothed clamps, and `ddx`-friendly forms to keep the Jacobian well-conditioned (avoid hard `if` discontinuities on solution variables). 3. Bound parameters with `from [...]` ranges and provide physical defaults; emit `$strobe`/`$error` only outside the inner solve loop to avoid convergence noise. 4. Model noise explicitly where the block's noise spec matters (`white_noise`, `flicker_noise`) so model-vs-SPICE noise comparison is meaningful. 5. For RNM, drive `wreal`/`nettype` resolution functions and quantize/threshold at digital boundaries; never let `wreal` `'z`/`'x` propagate into arithmetic. ### QoR Metrics to Evaluate - All branch contributions continuous/differentiable (no hard discontinuity on solve vars) - Parameters bounded with physical defaults - Noise modelled where a noise spec applies ### Common Issues & Fixes | Issue | Fix | |-------|-----| | Hard `if` on a solution variable | Replace with `$limexp`/`tanh`-smoothed transition | | Constants recomputed every step | Hoist to `@(initial_step)` | | RNM `x`/`z` propagating into math | Add resolution/threshold at the boundary; default `wreal` to a finite value | ### Output Required - Verilog-A / Verilog-AMS or SystemVerilog-RNM source --- ## Stage: model_compilation ### Domain Rules 1. Compile Verilog-A to OSDI with OpenVAF; treat all OpenVAF warnings (unused params, implicit conversions, non-differentiable constructs) as actionable, not noise. 2. Load the OSDI into ngspice/Xyce and confirm the module instantiates and runs a trivial DC/tran without error before validation. 3. On compile failure, read the OpenVAF diagnostic, fix the offending construct in va_authoring, and recompile (loop, cap 3×). Classify a hard tool crash as `tool_error`; a convergence-shaping construct failure as `convergence`. 4. Record the exact OpenVAF/compiler version and keep generated OSDI artifacts for reproducibility. ### QoR Metrics to Evaluate - OSDI compiles with 0 errors; warnings triaged - Module instantiates and runs a trivial DC/tran in ngspice/Xyce ### Common Issues & Fixes | Issue | Fix | |-------|-----| | OpenVAF compile error (unsupported construct) | Replace with an OSDI-supported form; remove non-differentiable `if` on solve vars | | OSDI loads but won't instantiate | Check discipline/nature mismatch and port order vs the netlist | | Non-convergence when OSDI loaded | Smooth discontinuities (`$limexp`/`tanh`); add small conductance to floating nodes | ### Output Required - Compiled OSDI artifact (+ compiler version) - Compile/load log (clean) --- ## Stage: connect_rule_setup ### Domain Rules 1. Define connect modules / connect rules (E2L, L2E, E2R, R2E bidirectional) for every analog↔digital boundary the model touches; specify supply-relative thresholds (VIH/VIL referenced to `constraints.supply.vdd_v`) and drive strength / impedance. 2. Ensure connect-module coverage is complete: every crossing net has an explicit rule and no implicit/default insertion is relied upon (flag implicit insertions). 3. Parameterise thresholds and rise/fall so the same connect modules work across the corner supply range. ### QoR Metrics to Evaluate - 100% of analog↔digital crossings have an explicit connect rule - Thresholds parameterised vs `supply.vdd_v` (corner-robust) - No reliance on implicit insertion ### Common Issues & Fixes | Issue | Fix | |-------|-----| | Connect module missing at a boundary | Add an explicit connect rule; disable reliance on implicit insertion | | Fixed thresholds break at corner Vmin | Parameterise VIH/VIL vs `supply.vdd_v` | ### Output Required - Connect-module sources + connect-rule map (boundary → rule) --- ## Stage: model_validation ### Domain Rules 1. Co-simulate the model against the SPICE reference (the circuit-design netlist in `design_state.circuit.netlist` if present, else a reference cell) on matched stimuli; compute model-vs-SPICE error % on the spec-bearing outputs (gain, BW, offset, noise). 2. Measure the simulation speed-up factor (SPICE wall-time / model wall-time) — the model must be both accurate AND faster to justify itself. 3. For RNM, measure toggle/branch coverage of the model's behavioural arcs; an under-exercised RNM is not validated. 4. If error > tolerance, loop back to va_authoring (cap 3×) targeting the failing output; classify error > tol as `functional` (behaviour wrong) or `spec_violation` (quantitatively off the spec it models). ### QoR Metrics to Evaluate - Model-vs-SPICE error ≤ tolerance (default 5%) on each spec-bearing output - Sim speed-up ≥ target (default ≥ 10×) vs the SPICE reference - RNM toggle/branch coverage ≥ target (default ≥ 90%) for RNM models ### Common Issues & Fixes | Issue | Fix | |-------|-----| | Model-vs-SPICE error > tol | Add the missing parasitic/noise term; refine parameter extraction from the reference | | Low sim speed-up | Hoist constant computation to `@(initial_step)`; reduce internal nodes | | RNM coverage low | Add directed stimulus exercising the unhit behavioural arcs | ### Output Required - Model-vs-SPICE validation report (error %, speed-up, RNM coverage) --- ## Stage: model_signoff ### Sign-off Pass Criteria (all must pass) | Check | Criterion | |-------|-----------| | Compile | OSDI compiles clean (0 errors), loads in ngspice/Xyce | | Accuracy | model-vs-SPICE error ≤ tolerance on all spec-bearing outputs | | Speed-up | ≥ target speed-up factor vs the SPICE reference | | RNM coverage | ≥ target toggle/branch coverage (RNM models) | | Connect modules | 100% boundary coverage, explicit rules | ### Domain Rules 1. Review the model against the model plan and the spec-traceability map. 2. Confirm accuracy, speed-up, and (for RNM) coverage all pass with margin. 3. Hand off the compiled OSDI / connect modules to ams-verification; record paths in `design_state.modeling`. ### Failure Escalation - Accuracy miss the author cannot resolve → va_authoring (×3) → escalate (`functional`/`spec_violation`) - Compile crash → `tool_error` (regenerate); model-shaping non-convergence → `convergence` - Missing connect rule → `connectivity` ### Output Required - Sign-off record (compile, accuracy, speed-up, coverage) - Compiled OSDI / connect modules + validation report for ams-verification --- ## Constraint Validation See [`plugins/meta/skills/pipeline-orchestration/SKILL.md`](../../../meta/skills/pipeline-orchestration/SKILL.md) §Constraints Schema for the authoritative schema and stage-entry validation rule. **Required at entry (`model_planning`) — hard-fail if missing:** - at least one spec to model — a non-null `constraints.specs`/`rf_specs` entry, or an explicit per-block spec passed in (e.g. from `architecture.blocks[]`) **Required at `connect_rule_setup` (only if the model has analog↔digital boundaries):** - `constraints.supply.vdd_v` — for supply-relative connect-module thresholds **Optional:** - `constraints.pdk` — modeling is technology-light; absence is not a failure Skip constraint validation entirely when invoked in fix-request-servicing mode (a `fix_request.id` was passed in the prompt). --- ## Memory ### Run state (write before first stage, update after each stage) Write `memory/modeling/run_state.md` as the **first action** before launching any tool: ```markdown run_id: modeling_<YYYYMMDD>_<HHMMSS> design_name: <design> pdk: <pdk or unknown> tool: <primary tool> start_time: <ISO-8601> last_stage: null ``` Update `last_stage` to the completed stage name only after each stage finishes successfully. ### Write on stage completion After each stage completes, upsert one JSON record in `memory/modeling/experiences.jsonl` keyed by `run_id` (do not append a second line for the same run). `key_metrics` fields: `model_error_pct`, `sim_speedup_x`, `rnm_coverage_pct`. Set `signoff_achieved: false` until model_signoff passes; then `true`. Create the file and parent directories if they do not exist. ### Optional: claude-mem index If `mcp__plugin_ecc_memory__add_observations` is available, emit each applied fix as an observation to entity `analog-design-modeling-fixes` after writing to `experiences.jsonl`. Skip silently if absent — the JSONL file is the canonical record.
Versions
The database has no other skills named behavioral-modeling.
Database snapshot: 13.11.2026 · package version 48f2f2a24dde (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100ams-verification
Build mixed-signal testbenches, define connect rules, run analog-digital co-simulation and real-number-model regressions, and close functional coverage. Use when verifying an analog/mixed-signal block against its digital control across corners, or re-validating after a… - 100rtl-reading-web
Generate or improve a Chinese RTL code-reading website from Verilog, SystemVerilog, or VHDL sources, for quickly understanding AI-generated code or unfamiliar code written by others. Use for guided module, transaction, signal, source-code, and optional waveform explanations with… - 100mixed-signal-cosim
Run mixed-signal co-simulation — digital gate-level Verilog + analog behavioral models together. Use when the user says "mixed-signal sim", "co-simulation", "digital-analog integration test", or at Step A9 of the analog track. - 100hdl-module-design
Use when writing, refactoring, or deciding how to test an HDL module, component, or IP block (ROHD, Chisel, SpinalHDL, Verilog, VHDL) and you need it parameterized, validated, and covered by exhaustive tests rather than a one-off - 100ams-integration
Qualify analog IP, assemble the mixed-signal top, define boundary/connect rules, run chip-level AMS simulation, and check power intent through integration sign-off. Use when integrating signed-off analog/RF blocks into a mixed-signal top, or re-validating after an integration… - 100verilog-fpga-digital-design
Design, synthesize, and verify digital hardware circuits using Verilog and SystemVerilog for FPGAs (Xilinx Artix/Zynq, Intel Cyclone, Lattice iCE40). Implement synthesizable finite state machines (FSM), synchronous reset logic, clock domain crossing (CDC) synchronizers, and…