analog M4b-q: the dynamic comparator with the devices charge and the body effect at 27 C, so its decision-time row is computed, not 98 times too fast #11312

Part of the analog epic (see #10486). The one FAIL left in the acceptance evaluator (specs/analog/accept_sar.t27, 6 PASS, 1 FAIL, 6 NOT MODELLED of 13 after #11288) is the dynamic comparator's decision time: the PSP103-table comparator (specs/analog/sar_comparator_psp.t27) is a NO-CHARGE model, the devices' own capacitance is not in it, so it decides about 98 times too fast against ngspice (5.35 ps against 523 ps at +2 mV, 50 C; 5.12 ps on the 27 C tables) and fails the report's 10 ps lower limit.

Everything needed for a 27 C slice (the report's temperature) is now on master: the tt log current tables (#11146), the drain and gate charge tables of tt (M3c, #10893, #11123; the gate charge read by the gate-node circuit of M3c-3), and the source-to-bulk axis of tt (M3e, #11288), which the input pair needs (its sources sit on the tail node, not on the bulk).

Only .t27 (skill only-t27); decks under specs/analog/decks/.

Requirements

• FR-001: The comparator's multi-node Newton step (sar_comparator.t27 _np forms, reused, not copied) MUST take the devices' charge from the tt charge tables, the same backward Euler charge step the 500 fF inverter uses (M3c), at 27 C.
• FR-002: The devices whose source is not at their bulk MUST read the current through the Vsb axis of analog M3e: a source-to-bulk axis on the PSP103 log tables, so the tgate Ron with its bulks on the rails is within 0.09 percent of ngspice #11288.
• FR-003: The decision time at +2 mV and -2 mV (and the report's other comparator rows the model can then compute) MUST be checked against our own ngspice-47 run of the same comparator at 27 C on the reference host (container jobs, at most 6 of its 12 cores, removed afterwards); the gap is printed in percent.
• FR-004: accept_sar.t27's dynamic_comparator rows are recomputed from this model; a row turns PASS only within the report's limit AND within 10 % of ngspice; a negative control (charge removed) must still FAIL.

User Scenarios

• Given +2 mV at the inputs, when the comparator with charge runs on the 27 C tables, then its decision time is printed next to ngspice's with the gap.
• Given the charge removed, when the same run is judged, then the decision-time row FAILs as today.

Boundary

• specs/analog/sar_comparator.t27
• specs/analog/sar_comparator_psp.t27
• specs/analog/accept_sar.t27
• specs/analog/decks/
• docs/analog/

Success Criteria

• t27c test-report on every changed spec prints FAIL 0 and vacuous passes 0
• the decision time at +2 mV is within 10 % of ngspice at 27 C, or the gap is written as measured and the row stays FAIL
• the dynamic_comparator block is PASS, or the reason it is not is written in accept_sar.t27's header

🤖 Generated with Claude Code