Add MosfetModel class with lambda (channel-length modulation)

[esaruoho]

Summary

• Adds MosfetModel class following the DiodeModel/TransistorModel pattern — named, reusable MOSFET parameter sets stored in a static HashMap
• Model parameters: threshold voltage (Vt), beta, lambda (channel-length modulation, 1/V), gate-source capacitance (Cgs), gate-drain capacitance (Cgd)
• Lambda is applied in both saturation and linear (triode) regions, matching SPICE Level 1:
  • Saturation: Ids = 0.5*β*(Vgs-Vt)²*(1+λ*Vds), with proper gm and Gds derivatives
  • Linear: Ids = β*((Vgs-Vt)*Vds - Vds²/2)*(1+λ*Vds), with full product-rule Gds derivative
• MosfetElm edit dialog shows model selector with Create/Edit buttons; old circuits without FLAG_MODEL load inline vt/beta as before (full backward compatibility)
• JfetElm is unaffected — overrides needsModel() to return false (1-line change)
• Gate capacitance fields (Cgs/Cgd) are stored in the model but companion-model simulation is deferred to a follow-up PR

Changes

File	Change
MosfetModel.java	New — model class with HashMap registry, text dump (type 36), XML serialization (tag "mm"), Editable interface
EditMosfetModelDialog.java	New — model editor dialog (follows EditTransistorModelDialog pattern)
MosfetElm.java	Model reference (modelName/model), FLAG_MODEL (128) for backward compat, setup(), dump()/dumpXml()/undumpXml() with model support, lambda in calculate() (both linear and saturation regions), model selector in edit dialog
JfetElm.java	needsModel() returns false (1 line)
CircuitLoader.java	Register type code 36 → MosfetModel.undumpModel()
XMLDeserializer.java	Register tag "mm" → MosfetModel.undumpModelXml()
CirSim.java	Add mosfetModelEditDialog static field, MosfetModel.clearDumpedFlags()
CommandManager.java	MosfetModel.clearDumpedFlags() in copyOfSelectedElms()
SimulationManager.java	MosfetModel.clearDumpedFlags() in getCircuitAsComposite()

Context

This addresses Paul's feedback on PR #266 and #240: "Between this and the lambda change, I think we need mosfet models." The approach follows the established model pattern exactly — DiodeModel for diodes, TransistorModel for BJTs, now MosfetModel for MOSFETs.

The linear-region lambda improvement was informed by independent work from Jonathan Klamroth (@jonnykl), who implemented channel-length modulation in their fork. Their approach applied (1+λ*Vds) in the linear region as well, matching SPICE Level 1. We adopted this improvement with corrected linear-region gm (including the (1+λ*Vds) factor, which their version omitted) and the Gds convergence floor preserved.

Based on v3-dev branch.

Test plan

• Create new n-MOSFET and p-MOSFET — should use "default" model
• Open edit dialog — model selector with "default" selected, Create/Edit buttons
• Create new model with non-zero lambda — verify saturation Ids increases with Vds
• Verify linear-region behavior with lambda — Ids should also show Vds dependence before saturation
• Load old circuit files (no FLAG_MODEL) — should read inline vt/beta, no model
• Save and reload circuit with model — model persists via FLAG_MODEL
• Create JFET — should show old-style Threshold/Beta fields, no model selector
• Verify XML save/load round-trip preserves model reference
• With lambda=0, verify behavior is identical to original (no regression)

🤖 Generated with Claude Code

[pfalstad]

If you load an existing circuit with a MOSFET (example: Circuits->MOSFETs->n-MOSFET) and right click the MOSFET you get an exception.

If you check and then uncheck "Simulate Body Diode" you get the edit model dialog...?

I assume you have another PR coming soon with Cgs/Cgd support so I'd probably wait for that... Otherwise remove Cgs/Cgd from the model details.

[esaruoho]

Fixed both bugs and added Cgs/Cgd companion model simulation in the latest push:

Bug fixes:

• Old circuits without FLAG_MODEL now get "default" model assigned (matching TransistorElm pattern) — this prevents the NPE on right-click and fixes the body diode toggle issue (both caused by null model)

Cgs/Cgd support:

• Trapezoidal companion model (same approach as BJT junction caps)
• startIteration() computes geq/ceq from model's capGS/capGD
• Stamps go in calculate() alongside the FET equations
• stepFinished() saves cap voltage/current state
• Gate node marked nonlinear when caps are present
• getConnection() returns true when caps bridge gate to S/D

The saturable inductor (dL/dt) is already up as PR #308.