analog M3e: a source-to-bulk axis in the PSP103 tables (tt), so the tgate Ron sees the body effect #11283

Part of the analog epic (see #10486). After M5d (#11263, PR #11279) the external team's per-block checks give 5 PASS, 1 FAIL, 7 NOT MODELLED of 13. Two of the open rows fail for one reason: the PSP103 tables are Id(Vgs, Vds) with each bulk tied to its source, so they have no body effect.

• The transmission gate's Ron (rows NOT MODELLED): ours 1718.0 ohm at 0.75 V and 2109.3 ohm at the worst of 27 points; ngspice's real cell, bulks on the rails, 2459.6 ohm and 2776.1 ohm. With each bulk tied to its source ngspice gives 1717.2 and 2108.4 ohm, so the whole gap is the body effect (specs/analog/mos_psp103_gates.t27, header).
• The comparator on PSP103 tables decides 6.5 % early for the same reason (0.65 % with bulks tied to sources; M4b, PR analog M4b-psp: the StrongARM comparator on PSP103 tables at 50 C against ngspice -- sign, times, flip point, imbalance sensitivity (Closes #10856) #10884).

Only .t27 (skill only-t27): decks are ngspice netlists under specs/analog/decks/, every number lands in a .t27 data file, the read and the checks are .t27 tests.

Requirements

• FR-001: A tt deck MUST print log-domain current tables of sg13_lv_nmos (1/0.45 um) and sg13_lv_pmos (3/0.45 um) at several source-to-bulk voltages from 0 to 1.5 V on the grid of the existing tt log tables (TT_LG_N, TT_LG_P), plus held-out currents at cell centres between the Vsb slices.
• FR-002: The table read MUST take the source-to-bulk voltage as a third axis (extend the existing read in specs/analog/mos_table.t27; no copied body) and MUST give back the existing 2-D result exactly when Vsb = 0.
• FR-003: The tgate Ron at 0.75 V and its worst Ron over 0.1 to 1.4 V MUST be computed with the bulks on the rails and compared with ngspice's real cell (2459.6 ohm, 2776.1 ohm); the accept_sar.t27 Ron rows become PASS only if within the 10 % bar of mos_psp103_gates.t27 and inside the report's limit.
• FR-004: Every reference number MUST come from our own ngspice-47 run on the reference host (container jobs, at most 6 of its 12 cores, files removed afterwards) and its deck MUST be committed.
• FR-005: A negative control MUST show the 2-D read (Vsb forced to 0) refused by the same bar on the real cell.

User Scenarios

• Given the tgate at 0.75 V with bulks on the rails, when Ron is read through the Vsb axis, then it is printed next to ngspice's 2459.6 ohm with the relative gap.
• Given Vsb = 0, when the 3-D read runs, then every value equals the 2-D read.
• Given the Vsb axis forced to 0 on the real cell, when the tgate check runs, then the bar refuses it.

Boundary

• specs/analog/mos_table.t27
• specs/analog/mos_psp103_data.t27 (or a new data file next to it for the Vsb tables)
• specs/analog/mos_psp103_gates.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 tgate Ron at 0.75 V is within 10 % of 2459.6 ohm, or the gap is reported as measured and the row stays NOT MODELLED
• the NOT MODELLED count of accept_sar.t27 drops from 7, or the reason it cannot is written in the spec header

🤖 Generated with Claude Code