by John Lauer
An independently designed 12 V to 5 V, 1 A synchronous buck molecule (TI TPS54202), with calculations, ngspice and PSpice simulation, KiCad 10 schematic and board, and the board laid out and analysed with adom-aiflow.
61f1679
Probing before ordering
When the boards come back from the fab, they go into the Adom probing workcell: a gantry with a flying probe, a programmable supply, an electronic load and an oscilloscope. This guide is the plan for that bench, written before the board is ordered, so the board can still change if the bench needs it to. It did change: the plan found a test the board could not support, and the board went back to layout for one more pad.
This is the probing stage of Adom AI Flow, which is still in development. Everything below is a plan and a simulation; the bench numbers come when the boards arrive.
Every bench test is a SPICE run
Each test on the workcell reproduces one of the SPICE runs from the simulation guide: the same input voltage, the same load, the same 1 A to 0.5 A step at 7 ms, the same trigger. The scope trace should lie on top of the SPICE trace, and the same measurement code scores both, so the comparison is like for like.
Until the boards exist, each capture is simulated: the ngspice waveform with every part value drawn inside its tolerance, seen through the measurement chain (the probe and its ground, the 20 MHz bandwidth limit, scope noise, 12-bit sampling, averaging). That proves the plan, not the board.
| Test | Probe | Measured on the simulated bench | ngspice | TI PSpice model | Pass band |
|---|---|---|---|---|---|
| Soft start, 10 to 90 % rise | TP3 | 4.01 ms | 4.00 ms | 4.10 ms | 3 to 5 ms |
| Switching frequency | TP1 | 530 kHz | 500 kHz | 390 to 630 kHz (datasheet) | |
| Ripple at 12 V, 1 A | TP3 | 3.6 mV | 3.1 mV | 4.0 mV | 1 to 8 mV |
| Ripple at 9 V, 1 A | TP3 | 3.0 mV | 2.3 mV | 0.8 to 6 mV | |
| Ripple at 16 V, 1 A | TP3 | 4.5 mV | 3.9 mV | 1.2 to 9 mV | |
| Load step, 1 A to 0.5 A | TP3 | 91 mV | 79 mV | 90 mV | 55 to 125 mV |
| Load step, 0.5 A to 1 A | TP3 | 90 mV | 78 mV | 89 mV | 55 to 125 mV |
| Settling after the step up | TP3 | 208 us | 194 us | 178 us | under 300 us |
The load step is the question the real board answers: the fitted ngspice loop and TI's model disagree by about 15 %, and the bench says which one to trust for the next layout.
The plan also covers DC levels at TP3 and TP2, the VMON monitor on the Control Panel's ADC, the undervoltage lockout thresholds (a hand calculation, since the switching model has no UVLO), efficiency at 9, 12 and 16 V, and a thermal soak.


More scope plots: soft start, switch node, ripple at 9 V, ripple at 16 V.
The loop: probing sends the board back to layout
The probe carries its ground pin beside the tip, about 2 mm away. TP1 (the switch node) already had TP4 (GND) 2.1 mm away, but TP3, where the ripple is measured, had no ground pad within reach. Measured through a long ground lead instead, the simulated bench reads 12 mV of ripple where the board has about 3.5 mV: the lead picks up the switching edges and measures itself.
So the AI went back to the layout step and added TP5, a 1.0 mm GND pad 2.16 mm from TP3, joined to J5 with a short trace. The video shows the loop: the probing result, the return recorded in the run, the board before, the change made live in KiCad 10, the design-rule check, and the probing plan passing again.
Download the video (54 s, narrated)
The space beside TP3 is tight (L1, the J5 and J6 contacts and their silkscreen frames), so TP5 uses a 0.75 mm courtyard instead of the library's 1.0 mm, has no silkscreen circle, and the contact frames were trimmed back off the pad. KiCad 10's check of the saved board: 0 errors, 0 unconnected. TP5 is a bare pad: nothing changes in the BOM or the pick-and-place file, only the copper, mask and silkscreen.
# Probing before ordering
When the boards come back from the fab, they go into the Adom probing workcell: a gantry with a flying probe, a programmable supply, an electronic load and an oscilloscope. This guide is the plan for that bench, written **before** the board is ordered, so the board can still change if the bench needs it to. It did change: the plan found a test the board could not support, and the board went back to layout for one more pad.
This is the probing stage of [Adom AI Flow](https://wiki.adom.inc/adom/adom-aiflow), which is still in development. Everything below is a plan and a simulation; the bench numbers come when the boards arrive.
## Every bench test is a SPICE run
Each test on the workcell reproduces one of the SPICE runs from the [simulation guide](simulation.md): the same input voltage, the same load, the same 1 A to 0.5 A step at 7 ms, the same trigger. The scope trace should lie on top of the SPICE trace, and the same measurement code scores both, so the comparison is like for like.
Until the boards exist, each capture is simulated: the ngspice waveform with every part value drawn inside its tolerance, seen through the measurement chain (the probe and its ground, the 20 MHz bandwidth limit, scope noise, 12-bit sampling, averaging). That proves the plan, not the board.
| Test | Probe | Measured on the simulated bench | ngspice | TI PSpice model | Pass band |
|---|---|---|---|---|---|
| Soft start, 10 to 90 % rise | TP3 | 4.01 ms | 4.00 ms | 4.10 ms | 3 to 5 ms |
| Switching frequency | TP1 | 530 kHz | 500 kHz | | 390 to 630 kHz (datasheet) |
| Ripple at 12 V, 1 A | TP3 | 3.6 mV | 3.1 mV | 4.0 mV | 1 to 8 mV |
| Ripple at 9 V, 1 A | TP3 | 3.0 mV | 2.3 mV | | 0.8 to 6 mV |
| Ripple at 16 V, 1 A | TP3 | 4.5 mV | 3.9 mV | | 1.2 to 9 mV |
| Load step, 1 A to 0.5 A | TP3 | 91 mV | 79 mV | 90 mV | 55 to 125 mV |
| Load step, 0.5 A to 1 A | TP3 | 90 mV | 78 mV | 89 mV | 55 to 125 mV |
| Settling after the step up | TP3 | 208 us | 194 us | 178 us | under 300 us |
The load step is the question the real board answers: the fitted ngspice loop and TI's model disagree by about 15 %, and the bench says which one to trust for the next layout.
The plan also covers DC levels at TP3 and TP2, the VMON monitor on the Control Panel's ADC, the undervoltage lockout thresholds (a hand calculation, since the switching model has no UVLO), efficiency at 9, 12 and 16 V, and a thermal soak.


More scope plots: [soft start](docs/probing/t2-soft-start.png), [switch node](docs/probing/t3-sw-node.png), [ripple at 9 V](docs/probing/t5-ripple-9v.png), [ripple at 16 V](docs/probing/t6-ripple-16v.png).
## The loop: probing sends the board back to layout
The probe carries its ground pin beside the tip, about 2 mm away. TP1 (the switch node) already had TP4 (GND) 2.1 mm away, but TP3, where the ripple is measured, had no ground pad within reach. Measured through a long ground lead instead, the simulated bench reads 12 mV of ripple where the board has about 3.5 mV: the lead picks up the switching edges and measures itself.
So the AI went back to the layout step and added **TP5**, a 1.0 mm GND pad 2.16 mm from TP3, joined to J5 with a short trace. The video shows the loop: the probing result, the return recorded in the run, the board before, the change made live in KiCad 10, the design-rule check, and the probing plan passing again.
<video controls preload="metadata" width="1080" poster="docs/probing/tp5-loop-poster.jpg" src="docs/probing/tp5-loop.mp4"></video>
[Download the video](docs/probing/tp5-loop.mp4) (54 s, narrated)
The space beside TP3 is tight (L1, the J5 and J6 contacts and their silkscreen frames), so TP5 uses a 0.75 mm courtyard instead of the library's 1.0 mm, has no silkscreen circle, and the contact frames were trimmed back off the pad. KiCad 10's check of the saved board: 0 errors, 0 unconnected. TP5 is a bare pad: nothing changes in the BOM or the pick-and-place file, only the copper, mask and silkscreen.