Skip to the content
4 layers150 × 100 mmPower profiler carrier, the printed circuit board in 3D

Power profiler carrier

Guiimartinho/open-power-profiler

Open-source PPK2-class power profiler built on the Raspberry Pi Pico 2: 100 kSPS, 100 nA to 1 A auto-ranging, source meter and ampere meter modes, USB stream.

Size
150 × 100 mm
Copper layers
4
Thickness
1.6 mm
Parts
425
Nets
258
Pads
1,347
Tracks, vias
3,743, 694
Pours
45
Routed
Yes
KiCad
KiCad 10.0 (file 20260206)
Created
8 Oct 2026
Updated
14 Oct 2026

Overview

Power profiler carrier is an open-source printed circuit board by Guiimartinho, from the Guiimartinho/open-power-profiler repository on GitHub, published under the MIT licence. It is a 4-layer board of 150 × 100 mm with 425 parts on 258 nets, routed with 3,743 tracks and 694 vias.

It is built around the OPA197IDBV, OPA365AIDBV and TC4427EOA713, with the TPD4E1U06DBVR, LP5907MFX-3.3/NOPB(MS) and SN74LV165ADRE4 among its other parts. Its parts include 114 resistors, 108 capacitors, 21 ICs, 19 diodes and 16 transistors. It belongs with the sensors and measurement boards and the RP2040 and RP2350 boards of this library.

Main components

12

README

README.md

Open Power Profiler

Checks: run on demand Status: early development Software license: MIT Hardware license: CERN-OHL-P v2

Open-source power profiler in the class of the Nordic PPK2, built around the Raspberry Pi Pico 2. It is designed to sample current at 100 kSPS from 100 nA to 1 A with automatic range switching, to work either as a source meter that powers the device under test or as an ampere meter in series with an external supply, and to stream the samples to a PC over USB.

The instrument is two boards: a Raspberry Pi Pico 2, bought ready-made, plugged into a carrier board designed in this project that holds the measurement electronics.

Rendered view of the carrier board, draft A2

The picture is a rendering from the KiCad files of draft A2, with the Pico 2 on its sockets. It is not a photograph: no board has been built.

[!NOTE] The project is in early development. No hardware has been built and nothing has been measured. The performance figures in this repository are design targets, the circuits of the board are simulated from its netlist, and the figures of the board are calculated from the drawn copper. Project Status says what exists, what it rests on and what comes next.

Table of Contents

Project Status

As of 2026-10-13. Draft A2 of the carrier board is the present state of the hardware: the whole schematic, a specification written for it, and a board with every part placed and every one of its 984 connections drawn. The board was first routed by an autorouter, which left 28 connections open. Its layout has since been reviewed: the open connections are closed and the copper that an autorouter does not draw is drawn, the pours of the 1 A path, the guard, the sense pairs and the copper of the converters. The review was done with scripts and checked by independent calculation. The seven points where the reviewed board departs from the first wording of section 10 of the specification are recorded decisions, D-87 to D-93, and the specification follows them.

On 2026-10-13 a person looked at the drawings for the first time: the project owner opened the schematic and the board in the KiCad editor and found both coherent. It was a look, not a review item by item, and no edit was saved. The same day the footprints and the placement were reviewed on a copy of the project, with scripts on the board file, 3D renderings of every region of the board and the datasheets. Five findings of that review were worked into the board at single points, with no new routing, and the silkscreen got the frames and names of the functional blocks (Review of Footprints and Placement). What is open is the review of the board item by item in the KiCad editor, a list of items that need parts moved, what the silkscreen and the test points still lack, and the thermal check of D-93. Every figure of the board is calculated from the drawn copper, none is measured; the package in tools/board/ reproduces them from the board file.

Later on 2026-10-13 the project owner asked for the files that are needed to send the board to fabrication. The fabrication outputs of draft A2 are filed in hardware/fabrication/draft-a2/ for an order of proof-of-concept boards: eleven Gerber files of the four-layer board, two drill files and two files for machine assembly. That is decision D-105 of the specification. The items that the specification puts before an order are still open, boards made from these files are draft A2 and not revision A, and those items stay open. No order is recorded: no board has been ordered or made as far as the repository shows.

On 2026-10-10 the circuits of the schematic were simulated as well, block by block, from its netlist: 134 benches in ngspice, of which 93 run circuits of the board and 41 put the models into the test circuits of their datasheets. The central design figures of the specification come out again. The simulations raised four points for a decision before boards are ordered, and the project owner decided them the same day: they are decisions D-95 to D-98 of the specification. Three of them are drawn since that day, as changes of values, part numbers and texts that move no copper: 10 nF at three capacitors of the rail monitor (D-95), the two parts of the damper on the module input fitted (D-98), and the text of the detector position at the boost converter, which stays without a part (D-97). The schematic, the board, the bill of materials and the netlist snapshot of the simulations hold them, and the electrical rules check and the design rules check report nothing on these files (run again on 2026-10-13). The fourth is not drawn: a position for a capacitor at the buffer of the driver rail (D-96), which needs a footprint and copper. The simulations were not run again, so their results are still those of the circuit before the three changes.

In fourteen places the simulations differed from a figure or a rule of the specification or showed something that it did not state. The owner asked for the text to follow the simulations: twelve places are corrected in the specification, three of them by decisions D-99 to D-101, two wait for the run of the benches on the drawn decisions, and seven questions that the corrections leave are open, each with one recommendation. A simulation is not a measurement: every active part in it is a model written from a datasheet, a typical part at room temperature. Circuit Simulations shows the state.

The software is written ahead of the hardware and is tested without it. Its state is that of 2026-10-12. On that day the firmware got a design of its own, one level below the specification; the host protocol went to version 2, which holds what draft A2 reports (D-103); and the firmware directory became a project of the Pico SDK that builds an image for the Pico 2. That image starts a console and the device state machine and nothing more, and it has not been loaded onto a board: nothing has run on a controller.

AreaWhat existsWhat is open
SpecificationThe baseline of draft A2: requirements R-01 to R-17, analog and firmware design, pin map, host protocol, 105 recorded decisions and 36 firmware rules that guard hardware. D-95 to D-98 are the decisions of the project owner on the four points that the simulations raised, D-99 to D-101 are the three corrections of the text that change a rule, a test or advice to the user, and D-102 settles the license of the simulation folder. D-103 and D-104 are the decisions of 2026-10-12 on the software: version 2 of the host protocol, and the over-current trip as a program of its own in PIO 2. D-105 is the decision of 2026-10-13 on the first order: the fabrication outputs of draft A2 are filed for proof-of-concept boards while the items that precede an order are openEvery figure is a datasheet value, a calculation, a simulation or an estimate; the tests of its section 11 have to confirm them. The items of its section 16 under "Before the board is ordered" stay open after D-105 and remain conditions of revision A. Of the four decisions D-95 to D-98, three are drawn since 2026-10-10; the fourth, a position for a capacitor at the buffer of the driver rail (D-96), is stated here and not drawn. Of the fourteen other places where the simulations of 2026-10-10 differed from its text, twelve are corrected; two wait for the new run of the benches, and seven questions that the corrections leave are open in its section 16
Schematic428 parts on 15 pages. Three decisions that follow the simulations are drawn in it since 2026-10-10: 10 nF in place of 1 nF at three capacitors of the rail monitor (D-95), the two parts of the damper on the module input fitted (D-98), and the text of the detector position at the boost converter, which stays without a part (D-97). It passes the electrical rules check, run again on 2026-10-13. Before these three changes the netlist of the draft was checked independently against the datasheets, and its circuits were simulated from it. On 2026-10-13 its symbols were compared with the libraries and the datasheets: 55 of the 59 symbols of the parts are unchanged symbols of the KiCad library, and the four symbols of the project agree pin by pin with their datasheets. One field changed that day and no connection: the pre-regulator names a new project footprintThe component checks: none is closed, so every part is a candidate; the comparison of 2026-10-13 closes none. One decision that follows the simulations is not drawn: a position for a capacitor at the buffer of the driver rail (D-96). The circuits that the three drawn changes touch have not been simulated again
Board150 mm × 100 mm, four layers, 425 footprints, 984 of 984 connections drawn, no violation in the design rules check, run again on 2026-10-13. The layout was reviewed on 2026-10-10 with scripts and checked by independent calculation. The three decisions drawn that day changed fields of six footprints (values, part numbers, fitted flags and one text) and no copper. On 2026-10-13 the project owner opened the project in the KiCad editor and found the schematic and the board coherent, and the footprints and the placement were reviewed with scripts on the board file, 3D renderings and the datasheets: every footprint equals its library footprint, no courtyard overlaps another, and nothing stands under the module. Five findings of that review are worked into the board: the pre-regulator stands on the land pattern of its own package, nine vias were moved away from pads and two from under the wall of the shield can, the reference texts stand at their own parts, and the console header has its pin names. The silkscreen carries the frames and names of the sixteen functional blocks. Calculated from the drawn copper, again after that round and unchanged: the 1 A path is 19.5 squares in source meter mode and 23.3 squares in ampere meter mode against a limit of 30, and the surface leakage into the measured node is 5.1 nA against a budget of 10 nA. The seven points where the layout departs from the first wording of section 10 of the specification are recorded as decisions D-87 to D-93No person has reviewed the board item by item in the KiCad editor: the owner's look of 2026-10-13 was not that review, and the review of footprints and placement was made outside the editor. Open items need parts moved, the capacitor loops of the converters among them. On the silkscreen 95 of the 425 reference texts are hidden, and the net names of the test points, the functions of the two jumpers and a frame for the serial number are missing; probe grounds at the test points are missing too. Two vias still stand at a pad, the lower edge of the frame of the shield can is not confirmed in a drawing of its maker, and the finger lands of the pre-regulator are not drawn. The temperature rise of the linear regulator is an open check, read on the first board (D-93). The position for a capacitor of D-96 is not on the board yet: it adds one footprint
Bill of materials151 lines for 370 parts, each line with a part number and a maker, exported from the schematic on 2026-10-10 with the drawn decisions: the three capacitors of the rail monitor at 10 nF (D-95) and the two parts of the damper on the module input (D-98). Exported again on 2026-10-13: one line changed, the footprint of the pre-regulatorThree lines had no stock on 2026-10-09, among them the capacitor that the damper uses. One position stays without a part and is not in the file: the detector position at the boost converter (D-97)
Fabrication outputsFiled on 2026-10-13 for an order of proof-of-concept boards of draft A2, by decision D-105 of the project owner: eleven Gerber files of the four-layer board and two drill files, also as one archive, and two files for machine assembly with the 360 surface-mount placements of the top side in 142 lines. They were exported from the board file with kicad-cli. Two things were checked on them: the drill file holds 787 plated holes, which are the 694 vias and the 93 holes of pads of the board, and every layer, the drill file and the placement file lie in one coordinate frameNo order is recorded: no board has been ordered or made as far as the repository shows. No Gerber viewer has read the files back, and no manufacturer has reviewed them. The assembly files carry manufacturer part numbers only, and their rotations are those of KiCad. The order is the project owner's; it still needs the surface finish ENIG in the order form, a word on the mark of the order number, the rotations checked in the preview of the assembler, and the parts matched by their manufacturer part numbers. Boards made from these files are draft A2, not revision A
Circuit simulationsThe circuits of draft A2 simulated in ngspice on 2026-10-10 from the netlist of the schematic as it stood that day, before the three decisions were drawn, in a package of their own (D-94): 134 benches, 93 on the circuits of the board in ten blocks and 41 that put the models written for the project into the test circuits of their datasheets. Of 3162 figures 1834 pass, 122 fail and 1206 carry no limit. The central design figures of the specification come out again block by block: shunt ladder, signal chain, range logic with requirement R-07, source meter, output stage, path switches, rails, power input and digital lines. The package has 777 tests; lines and branches are covered in full, the one module that loads the simulator excepted; its gates passed again on 2026-10-12The four points that the simulations raised are decided (D-95 to D-98): the capacitors of the rail monitor, the buffer of the driver rail, the detector position at the boost converter and the damper on the module input. Three of them are drawn since 2026-10-10 and are in the netlist snapshot of the package (D-95, D-97, D-98); the snapshot still matches the schematic on 2026-10-13. The results are still those of the run of 2026-10-10, on the circuit before these changes: the benches were not run again, and those of the power input still leave the two parts of the damper out. Open work: the benches brought in line with the snapshot, and the blocks that the changes touch run again, the rail monitor with the power-up and the power-off and the input from the module. In fourteen places the simulation and the text of the specification differed: twelve are corrected in the specification, two wait for the new run, and seven questions are open. The limits of the benches are still those of the earlier text. What stays for the bench: leakage at the nanoampere level, the multiplexer at +12 V and −4 V, the programs of the controller, coupling through the board, temperature. Every active part is a model written from its datasheet, and nothing is measured. The simulations ran on one machine; the Simulation workflow has not been started yet
FirmwareA design of the whole firmware: architecture, state machines, the three PIO programs as sketches, the USB link, the place and the test of every firmware rule. A project of the Pico SDK that builds an image for the Pico 2 with warnings as errors: 33 kB that start the console and the device state machine and drive no other pin. A core in C that needs no hardware, with 22 unit tests on the PC and every line and branch covered: wire protocol at version 2, block queue, raw capture word, settling window, step-down and under-range rules, device state machine, fault causes, the sequences of DUT power on and off, pin table, DAC frame, monitor scales and rails testThe image has not been loaded onto a board: nothing has run on a controller. No PIO program assembled or run, no adapter for any hardware, no USB link, no command handler, no storage of the calibration record. Of the 36 firmware rules none is implemented end to end: the tables, scales and sequence orders behind F-4 to F-6, F-11, F-12, F-23 to F-26 and F-28 to F-31, the under-range rule of F-22 and two of the three marks of F-35 have tests, and nothing calls them; the other rules have nothing. Six points of the design wait for the project owner. The build job of the Firmware workflow, rewritten for the Pico SDK, has not been started yet
Host softwarePython package with the protocol at version 2, transports, a device client, capture helpers and a device simulator, tested against that simulator: 500 tests, every line and branch covered. The client reads the status and the identity that version 2 carries, takes the four zero offsets from a zero calibration, and reads and stores the calibration record as a blockIts nominal calibration does not follow draft A2 yet, and flagged samples are left out of the statistics where the specification asks to fill them in. No code builds or reads the content of the calibration record; no command that captures from an instrument, no viewer and no export
ProtocolOne definition, at version 2 since 2026-10-12 (D-103). It holds what draft A2 reports: marks for over-range and under-range samples and for the output switch, a fault flag for each of the fifteen causes, wider records of status and identity, the supply in use with its budget as a current, notices, and the calibration record as one block. The constants of firmware and host, the layouts of the records and the shared test vectors are generated from it, and both codecs reproduce the vectorsNo firmware fills these fields yet, and no frame has passed between a controller and a PC. The layouts stay provisional until the protocol is frozen in phase 6. Three points of sections 6 to 8 of the specification are not decided; its section 16 lists them
Toolstools/board/: a Python package that calculates the layout figures from a dump of the board file: copper resistance in squares and milliohms, lengths of the Kelvin pairs, surface leakage into the measured node. 260 tests on synthetic boards, lines and branches fully coveredCalibration and production test: two empty folders. The Tools workflow has not been started yet. The scripts that drew the layout are not in the repository
MeasurementsNone: no board has been builtThe phases of the roadmap

Next steps, in this order:

  1. Finish the decisions that follow the simulations. The project owner decided the four points on 2026-10-10: decisions D-95 to D-98 of the specification. Three are drawn since that day, as changes of values, part numbers and texts on parts and pads that the board already had: 10 nF in place of 1 nF at three capacitors of the rail monitor, the comparators that tell the controller that the analog supplies stand (D-95); the two parts of a damper, a resistor with a capacitor, fitted on the module input, the supply that comes through the USB connector of the Pico 2 (D-98); and the text of the position for a voltage detector at the boost converter, which makes the supply of the analog rails: that position stays without a part (D-97). The electrical rules check and the design rules check report nothing on the changed files, and the bill of materials, the pictures and the netlist snapshot of the simulations were made again from them. Two things are open. The fourth decision is not drawn: a position for a capacitor at the buffer of the driver rail, the supply of the amplifier that drives the converter, which needs a footprint and so a change of the board (D-96). And the benches were not run again: those of the blocks that the changes touch, the rail monitor with the power-up and the power-off and the input from the module, have to be brought in line with the new snapshot and run. Until then the results of the simulations describe the circuit before the three changes.
  2. Decisions that are still open, and the first order. The open decisions are the project owner's. Seven questions that the corrections of the specification leave, because their answer needs a design value that no simulation gives: the criterion of one check, two limits of firmware rules against the leakage of the monitor converter, the upper limit of the 5 V rail, the threshold for a step of that rail, the capacitor at the VIN terminals with heavy leads, a withstand voltage of the logic inputs and the gaps of one test. Section 16 of the specification gives each with its options and one recommendation. One decision of this step is taken: on 2026-10-13 the project owner decided that proof-of-concept boards of draft A2 may be ordered on this state, with the items that the specification puts before an order still open (D-105). The fabrication outputs are filed, and the order is the owner's. It still needs four things from the owner, which the README of the outputs explains: the surface finish ENIG in the order form, a word on the mark of the order number, the rotations of the parts checked in the preview of the assembler, and the parts matched by their manufacturer part numbers. No order is recorded: no board has been ordered or made as far as the repository shows.
  3. Board. Four things stand between the reviewed layout and the fabrication of revision A. The proof-of-concept boards of draft A2 do not wait for them, and the decision to order such boards closes none of the four. First, a person reviews the board in the KiCad editor item by item. On 2026-10-13 the project owner opened the project there and found the schematic and the board coherent: that was a look, with nothing recorded item by item. The same day the footprints and the placement were reviewed with scripts, 3D renderings and datasheets, and the five findings of that review are worked into the board; that review was not made in the editor either. Second, the open items that need parts moved are a change of placement with a local redraw: the capacitor loops of the converters, the inductor of the pre-regulator and the others. Third, what the silkscreen and the test points still lack: net names at the test points, the functions of the two jumpers, probe grounds, a frame for the serial number, and the 95 reference texts that are hidden. The frames and names of the functional blocks and the pin names of the console header are printed since 2026-10-13. Fourth, the thermal check of D-93: the temperature rise of the linear regulator at full dissipation is read on the first board. Beside these four the round of 2026-10-13 left three small points: two vias that still stand at a pad, the lower edge of the frame of the shield can, which no drawing of its maker confirms, and the finger lands of the pre-regulator, which are not drawn. The hardware guide lists the open items one by one. The simulations judge the circuits, not the copper: of their decisions only the position of D-96 touches it.
  4. Software. The project owner took this step up on 2026-10-12, ahead of the three above, and three parts of it are done. What draft A2 reports is in version 2 of the protocol, in the definition and in the codecs of both sides (D-103). The fault causes are in the firmware core. And the firmware builds for the Pico 2 with the Pico SDK. What follows, in the order of the firmware design: the six points of that design that wait for the owner; the rest of the firmware core on the PC, from the command handlers to the calibration record; the three PIO programs in an emulator; the adapters for the hardware; and the firmware core tested against the host package. In the host package the nominal calibration of draft A2 and the statistics that section 9 of the specification asks for are still to do. The firmware guide and the host guide list what exists and what is missing.

Until 2026-10-10 this list had a step to file the simulations. It is done in another form. The files of the earlier simulations, from which the specification took its figures marked "simulated", are not filed; the circuits were simulated again from the netlist, and the package, the benches and their results are in simulation/, where a reader can run them again.

The component checks follow, of which none is closed, and the phases of the roadmap, starting with the risk prototypes of phase 1. Three of those prototypes settle what the simulations leave open: the reaction time of the range sequencer, which is an input of every simulation of the range logic, the loop of the pre-regulator at its low end, and the start of the boost converter from a supply limited to 0.7 A.

Target Specifications

ItemTargetRequirement
Sampling rate100 kSPS, timed by hardwareR-01
Data integrityNo lost sample in a capture of 10 minutes; any loss is reportedR-02
Current range100 nA to 1 A in four ranges, switched automaticallyR-03
Resolution100 nA or better in the lowest range; RMS noise there at most 5 nA in ampere meter mode from a quiet supply and 40 nA in source meter modeR-04
Accuracy±1 % of reading ±0.1 % of range after calibrationR-05
Burden voltage100 mV maximum across the shunt; 200 mV from input to output at 1 AR-06
Range changeOn a step from 1 µA to 500 mA the instrument adds at most 0.5 V of drop with 1 µF at its output and 0.25 V with 10 µFR-07
Source meter modeProgrammable output from 0.8 V to 5.0 V in 1.3 mV steps; 1.0 A up to 2.0 V, falling to 0.6 A at 5.0 VR-08
Ampere meter modeExternal supply from 0.8 V to 5.0 V through the shunts; the input withstands −20 V to +20 V while its switch is openR-09
Digital inputs8 logic channels sampled together with the current, logic levels from 1.65 V to 5.5 VR-10
Host linkUSB Full-Speed, CDC ACM, binary protocolR-11
CalibrationGain and offset of each range, stored on the deviceR-12
ProtectionOver-current trip, reverse polarity, over-voltage, thermalR-13
Power inputUSB-C, 5 V, on the carrier board, used whenever it is present, or the USB cable of the Pico 2 alone; the input current is limited to what the source offers: 0.45 A, 1.4 A or 1.7 AR-14
Host softwareCapture, live view, statistics and export on Windows, Linux and macOSR-15
OpennessFirmware and host software under MIT, hardware under CERN-OHL-P v2R-16
Software qualityAutomated tests with coverage gates, static analysis and continuous integrationR-17
ControllerRaspberry Pi Pico 2 (RP2350) on pin socketsD-39
DUT connectionsPin header and lever terminal block, in the pin order of the PPK2D-44

These are design targets, not measurements. Draft A2 set seven of them (R-04, R-06, R-07, R-08, R-09, R-10 and R-14) to what the architecture delivers, with the PPK2 as the yardstick. Section 2 of the specification gives the full wording of each requirement and the evidence behind each figure: a datasheet value, a calculation, a simulation or an estimate.

Where the specification compares a target with the PPK2, it quotes the PPK2 User Guide v1.0.1:

ItemThis design, targetPPK2, as its user guide states
Resolution in the lowest range100 nA or better (R-04)0.2 µA, with measurement down to about 200 nA
Current in source meter mode1.0 A up to 2.0 V, falling to 0.6 A at 5.0 V (R-08)600 mA, with a rated power of 5 W
External supply in ampere meter mode0.8 V to 5.0 V (R-09)The same range
Logic levels of the digital inputs1.65 V to 5.5 V (R-10)1.65 V to 5.5 V
Burden voltage100 mV across the shunt, 200 mV in the whole path at 1 A (R-06)Not stated
Drop at a range change0.5 V with 1 µF at the output, 0.25 V with 10 µF (R-07)Not stated
Voltage that the supply input withstands−20 V to +20 V with its switch open (R-09)Not stated

The middle column holds design targets, none of them measured; the right column is taken from section 2 of the specification.

How It Works

text
USB-C 5 V ────► limiter ─┐
                         ├► multiplexer ─► 5 V rail ─► supplies of the carrier
Pico 2 USB ───► limiter ─┘

5 V rail ───► pre-regulator ──► LDO (source mode) ◄── DAC ◄── SPI
                                    │
VIN (ampere mode) ──► protection ──►├── mode switch
                                    │
                         shunt ladder + range FETs ◄── range sequencer
                                    │         │ Kelvin sense   ▲
                         output switch     in-amp ──┬──► comparators
                                    │               └──► ADC ──► Pico 2 ──► USB
                                  VOUT ──► DUT          D0..D7 ──► level shift
  • Two boards. Everything in the diagram except the controller is on the carrier board. The Raspberry Pi Pico 2 brings the microcontroller and the USB port, and uses every one of its 26 GPIO pins.
  • One programmable part. The range sequencer and the sampling clock are programs for the PIO blocks of the RP2350: small state machines that run from the system clock whatever the processors do. Firmware is loaded through USB; no programmer and no vendor tool is needed.
  • A supply that spares its source. Each USB input has its own current limiter, and a multiplexer feeds the 5 V rail from USB-C whenever it is present. A supervisor switches the supplies of the carrier off when that rail collapses, and four comparators report through one line, PWR_GOOD, that the analog rails and the reference are present.
  • Four shunt ranges. Shunts of 1 kΩ, 33 Ω, 1 Ω and 0.1 Ω, each limited to 100 mV of burden voltage. Three comparators drive the range sequencer, which switches to a higher range with no firmware involved: on a large step the 1 A range conducts 0.35 µs after its threshold (simulated, with a sequencer that reacts within 100 ns, which phase 1 has to measure). A current step therefore costs the device under test only a short dip; requirement R-07 of the specification gives its size. The firmware decides when to step back down.
  • Hardware-timed sampling. A 16-bit SAR ADC is clocked by a PIO state machine, so the sampling instant does not depend on firmware latency. The same state machine reads the range and the logic inputs with every conversion result.
  • Self-describing samples. Every sample is a 32-bit word that carries the ADC code, the active range, the eight logic inputs and five marks: the sample lies in a settling window, the over-current trip is latched, the reading is over-range, it is under-range, the output switch is requested on.
  • Two cores, two jobs. One core acquires data in DMA blocks. The other frames the blocks and streams them over USB.
  • Programmable supply. In source meter mode a DAC sets a low-noise linear regulator, fed by a pre-regulator that tracks the output voltage.
  • Protection that does not wait for firmware. The over-current trip is a PIO state machine. An over-voltage on the external supply input holds its switch open through one transistor, a second transistor keeps the two mode switches from closing together, and with its switch open that input withstands −20 V to +20 V (simulated).
  • A quiet front end. The shunts of the three lower ranges, the amplifier and the converter are placed under a shield can, and the switching converters stand at least 30 mm away from it.

The reasons behind each choice, with the decision that records it, are in sections 3 and 4 of the specification.

Connections

The back edge of the carrier has the USB connector of the Pico 2 (data, and power for small loads) and a USB-C connector for power. Either one powers the whole instrument. The front edge has the connections to the device under test, in the pin order of the Nordic PPK2:

ConnectorTypePins, left to right
DUT1×4 pin header, 2.54 mmGND, VIN, VOUT, GND
DUTLever terminal block, 3.5 mmGND, VIN, VOUT, GND
Logic port1×10 pin header, 2.54 mmVCC, GND, D7 down to D0
  • Source meter mode: the device under test connects to VOUT and GND.
  • Ampere meter mode: the external supply connects to VIN and GND, the device under test to VOUT and GND.
  • Logic port: as built, the logic inputs follow the output voltage, and the VCC pin is not needed. A solder jumper makes the VCC pin the reference instead, for a device whose logic runs on another voltage.

Hardware Draft

The carrier board is drawn in KiCad 10 as draft A2. Its layout was reviewed on 2026-10-10. On the same day three decisions that follow the circuit simulations were drawn into it as changes of values, part numbers and texts (D-95, D-97 and D-98); they changed no copper and moved no part. On 2026-10-13 the footprints and the placement were reviewed, and a round at single points followed: the pre-regulator got the land pattern of its own package, eleven vias were moved, and the silkscreen was worked on. No part moved, nothing was routed again and the circuit did not change. The two rule checks were run again on these files on 2026-10-13. The fabrication outputs of the board in this state are filed since that day, for an order of proof-of-concept boards (D-105).

Draft A2 after the layout review, the three drawn decisions and the round of 2026-10-13
Schematic428 parts on 15 A4 pages: the root page and 14 sheets; unchanged by the layout review. Drawn on 2026-10-10: 10 nF at three capacitors of the rail monitor (D-95), the two parts of the damper on the module input fitted (D-98), and the text of the detector position at the boost converter, which stays without a part (D-97). Not drawn: the position for a capacitor at the buffer of the driver rail (D-96). On 2026-10-13 the footprint field of the pre-regulator changed, and no connection
Electrical rules check0 errors, 0 warnings
Board150 mm × 100 mm, four layers, 425 footprints, all on the top side; three of them are footprints of the project library, the others those of the library that KiCad 10 installs
Routing984 of the 984 connections: 8.40 m of track in 3,743 segments on three layers (top 3.77 m, second inner layer 2.84 m, bottom 1.79 m), 694 vias and 42 copper zones. The first inner layer is the ground plane, in one piece and with no track on it. Before the round of 2026-10-13, which moved eleven vias with short pieces of track, it was 8.38 m in 3,716 segments
Open connectionsNone. The autorouted board had 28, and 25 of them broke a function
Design rules check0 violations, 0 unconnected pads, 0 footprint errors, 0 differences between board and schematic. The rule file is stricter than the one the autorouter ran with: the exemption of the larger spacings near pins now holds on the top layer only. Decision D-87 records its spacings
SilkscreenThe frames and names of the sixteen functional blocks, the pin names of the DUT connectors, of the logic port and of the console header, and the limits of VOUT and VIN. 330 of the 425 reference texts are visible, each at its own part; 95 are hidden where no place is free beside the part, and every hidden one is on the fabrication layer
Bill of materials151 lines for 370 parts, each line with a part number and a maker: hardware/doc/bom.csv; unchanged by the layout review, exported again on 2026-10-10 with the three capacitors at 10 nF and the two parts of the damper, and on 2026-10-13 with the footprint of the pre-regulator in one line
Fabrication outputsFiled on 2026-10-13 in hardware/fabrication/draft-a2/, for an order of proof-of-concept boards (D-105): eleven Gerber files of the four-layer board (four copper layers, two solder masks, two paste layers, two silkscreens and the outline) and two drill files, also as one archive, and two files for machine assembly with 360 surface-mount placements in 142 lines. The drill file holds 787 plated holes, the 694 vias and the 93 holes of pads. No Gerber viewer has read the files back, no manufacturer has reviewed them, and no order is recorded
Circuit simulationsThe circuits of the schematic simulated from its netlist on 2026-10-10 in 93 benches: 1356 figures pass, 105 fail and 1136 carry no limit. The four points that they raised are decided (D-95 to D-98). Three of them are drawn since, and the benches were not run again: the results are those of the circuit before these changes (Circuit Simulations)

The schematic was checked independently against the datasheets of its parts. That closes none of the component checks: the parts are candidates. The simulations close none either: they run on models written from those datasheets.

The board came about in three steps. A script placed every footprint inside the area of its functional block. An autorouter drew a first set of tracks: 956 of the 984 connections, with 28 left open, no pour and no guard. Then the layout was reviewed with scripts on the board file, block by block: the open connections were closed and the pours of the 1 A path, the guard, the Kelvin pairs, the copper of the converters, the 5 V rail and the ground fills were drawn. Each block was drawn against the rules of section 10 of the specification and then checked by an independent check that calculated every rule again. A final review of the whole board by rule group followed, with 75 findings, then a repair round for the leakage paths into the measured node that it found. No part was added or removed; 35 of the 425 footprints moved or turned, 33 of them by 2.75 mm or less. A fourth step followed on 2026-10-13, at single points and with no new routing: the round after the review of footprints and placement.

The result is a drawn and calculated layout, not a checked design. The project owner opened it in the KiCad editor on 2026-10-13 and found it coherent. That was a look: nobody has reviewed it there item by item, as a person would before ordering boards, and every check ran from the command line. The scripts that calculate the figures are in the repository, as the package in tools/board/; the helpers that drew the layout are not.

The project owner decided on 2026-10-13 that proof-of-concept boards may be ordered from the draft as it stands, and asked for the files (D-105). The fabrication outputs are filed in hardware/fabrication/draft-a2/. Boards made from them are draft A2, as their silkscreen says, and not revision A: everything that the sections below list as open stays open for them. The repository records no order.

What the Layout Review Reached

Every figure in the column "Now" is calculated from the drawn copper. Nothing is built and nothing is measured. The resistances count 35 µm copper at 40 °C and leave out the transistors, the shunt and the fuse. The round of 2026-10-13 moved eleven vias and gave the pre-regulator other lands. The 1 A path, the leakage and the count of nets without a via were calculated again on the board after it and are unchanged; the other rows are those of the layout review.

Rule, with its section of the specificationLimitAutorouted boardNow
Connections drawn984956984
1 A path in source meter mode, regulator output to the VOUT terminal (10.4)30 squares397 squares, about 211 mΩ19.5 squares, 10.4 mΩ
1 A path in ampere meter mode, VIN terminal to the VOUT terminal (10.4)30 squares217 squares, about 116 mΩ23.3 squares, 12.3 mΩ
Surface leakage into the measured node (10.3)10 nANo figure5.1 nA; it was about 20 nA at the final review, before the repair round
Guard around the measured node (10.3)A closed ring on the top layerNoneOne piece of copper: three arcs on the top layer, open where the measured node leaves the shield can, joined through the other layers
Sense and guarded nets on the top layer without a via (10.3)All 11No figure10 of 11
5 V rail from the bulk capacitor to the input capacitors of the converters (10.1)10 mΩ to the pre-regulator and 25 mΩ to the boost converter: targets of the layout review, the specification gives no figureTracks of 0.4 mmA pour on the second inner layer: 2.7 mΩ and 3.0 mΩ to the pre-regulator, 3.6 mΩ to the boost converter
Copper from the output capacitors of the pre-regulator to the bead (10.5)15 mΩNo figure7.1 mΩ

The figures of the 1 A path, of the leakage, of the pairs and of the vias on the sense nets are reproduced from the board file by the report command of the package in tools/board/; its guide says how to install it, dump the board and run it. The resistances are calculated on a grid of 0.1 mm; on a grid of 0.05 mm the two sums of the 1 A path read about 2 % higher, 19.9 squares and 23.8 squares, and a via is counted in a way that errs to the high side.

The leakage figure rests on the assumptions that section 10.3 of the specification itself uses: a clean surface of 10¹¹ Ω per square and the voltages of the neighboring nets. Solder mask, cleanliness and humidity are not modeled. Almost all of what is left comes from the pitch of pads: the gate beside the source in a transistor package, the poles of the two DUT connectors, neighboring pins of the multiplexer.

Decisions of the Layout Review

In seven points the reviewed board departs from what section 10 of the specification first asked. The owner accepted all seven on 2026-10-10. They are recorded in the decision log of the specification (section 15), and section 10 now describes the board as it is drawn:

  • D-87. The rule file of the board states every spacing that the layout needs.
  • D-88. The high-side sense line of the 0.1 Ω shunt keeps two vias and 1.9 mm on the second inner layer.
  • D-89. The Kelvin pairs: one runs side by side, two are equal in length within 1 mm, and the taps of the two lowest ranges run through one via each.
  • D-90. The guard is one piece of copper, open on the top layer at the three exits of the measured node and joined on other layers.
  • D-91. The supply, address and enable pins of the multiplexer and their parts stand inside the guard.
  • D-92. The 1 A path is judged by its resistance, 30 squares or less in either mode; two of its pieces run on the bottom layer.
  • D-93. The island of output copper under the linear regulator is 291 mm² on the bottom layer and 102 mm² on the top layer, joined by 22 vias; its temperature rise is an open check, read on the first board.

A decision accepts a drawn layout, not a measured one: where it rests on a resistance or a leakage, the figure is calculated from the drawn copper. The hardware guide quotes, for each point, what section 10 said before, what the board has and why.

Review of Footprints and Placement

On 2026-10-13 the project owner opened the project in the KiCad editor, the schematic and the board, and found both coherent. That was a look, not a review item by item, and no edit was saved.

The same day the footprints and the placement were reviewed on a copy of the project, from three sources: measurements by scripts from the board file, 3D renderings of every region of the board, read one by one, and the datasheets and drawings of the makers. It is not a review in the KiCad editor, nothing was measured, and it closes no component check. What it found in order:

  • All 425 footprints of the board equaled the footprint of the same name in their library, pad by pad. Of the 59 symbols of the parts, 55 are unchanged symbols of the KiCad library, and the four symbols of the project agree pin by pin with their datasheets.
  • The terminal block, the USB-C receptacle, the socket strips of the module and the 0.1 Ω shunt agree with the drawings of their makers.
  • Every part is on the top side, no courtyard overlaps another, no part stands under the module, and 58 parts stand inside the shield can with none across its wall.

It led to five findings. The project owner asked for all five to be worked into the board, at those points only and with no new routing, and that was done the same day:

FindingWhat was done
The pre-regulator stood on a library footprint for another package, with a thermal land of 1.70 mm × 3.30 mmIt has the land pattern of its own package, the DSJ package of Texas Instruments, as a project footprint: a thermal land of 1.58 mm × 2.85 mm
123 reference texts of the silkscreen were hidden, and visible ones stood at another part or up to 7 mm from their own61 texts were moved to their own part and 31 hidden ones are shown again: 330 of the 425 are visible, and 95 are hidden
The console header had no pin names, and the text "USB DATA/POWER" stood beside itIts pins read TX, RX and GND, and the text reads "PICO USB" over "DATA/POWER"
Nine via holes reached into a pad or stood within 0.05 mm of one, where solder can drain into the holeThe nine vias were moved by 0.2 mm to 1.6 mm, to 0.25 mm or more from the pad; two more were left where no free place exists
Two tented vias stood under the wall line of the shield canBoth were moved inside the wall line

The silkscreen also got the frames and names of the sixteen functional blocks, which were on a documentation layer of the board until then. No part was added or moved, and the circuit did not change. Nothing was routed again: the 3,716 track segments that the board had are unchanged, 27 short pieces with 15.9 mm in all were added at the eleven moved vias, and the other 683 vias stand where they stood. After the round the board has 8.40 m of track in 3,743 segments and 694 vias, the figures of the 1 A path and of the leakage are unchanged, and both rule checks report nothing. The six pictures of the board below were plotted again from the final board file. The hardware guide gives every comparison and every finding with its figures.

What Is Still Open on the Board

Four things stand before the fabrication of revision A, and the component checks beside them. The proof-of-concept boards of draft A2 may be ordered with all of them open (D-105), and that decision closes none of them. Two more came after the layout review. One is what is left of the decisions that follow the circuit simulations: three of them are drawn since 2026-10-10 and changed no copper, and the fourth is not drawn, the position for a capacitor at the buffer of the driver rail, which adds a footprint to this board (D-96; Project Status). The other is what the review of 2026-10-13 left: two vias that still stand at a pad, and two points that are named below.

No person has reviewed the board item by item in the KiCad editor. The project owner looked at it there on 2026-10-13 and found it coherent, and the footprints and the placement were reviewed that day outside the editor; neither is the review that precedes revision A.

Eleven items need parts moved or another footprint and were left open by the layout review. The capacitor loops of the pre-regulator and of the boost converter close over 6.8 mm to 17.1 mm of top copper where the limit is 5 mm, and the inductor of the pre-regulator stands 4.5 mm from its switch pins where the limit is 3 mm; nine more are smaller. One of the nine is done in part since 2026-10-13: of the vias at pads that get solder paste nine were moved, and two still stand at a pad.

The silkscreen is not finished. Since 2026-10-13 it carries the frames and names of the functional blocks and the pin names of the console header, and its reference texts stand at their own parts. 95 reference texts are hidden, and the net names of the test points and the functions of the two jumpers are not printed; there is no frame for the serial number. 32 of the 40 signal test points have no probe ground within 5 mm.

Beside those two vias, two points are left from the review of 2026-10-13: no drawing of its maker confirms the lower edge of the frame of the shield can, and the finger lands that the drawing of the manufacturer shows at the two short ends of the pre-regulator package are not drawn.

The temperature rise of the linear regulator at full dissipation is an open check, to be read on the first board (D-93).

The component checks are open too: none is closed. The hardware guide lists the open items one by one, with their figures and with what would close each.

Pictures

All pictures are renderings and plots of the board file, not photographs. The six pictures of the board were made again on 2026-10-13, from the board file after the round of that day: the two renderings and the plots of the copper layers, of the front end, of the 1 A path and of the placement.

The top side, rendered from the board file. The gold line around the front end is the guard under open solder mask, and the thin white frames with a name in each are the sixteen functional blocks as the silkscreen prints them:

Rendered top view of the carrier board

The four copper layers side by side. Top left, the top layer: the parts, the pours of the 1 A path, the sense pairs and the guard. Top right, the first inner layer: the ground plane, in one piece. Bottom left, the second inner layer: the 5 V rail, the rail trunks and a ground fill. Bottom right, the bottom layer: the crossings, the second layer of the 1 A path and a ground fill:

The four copper layers of the board

The measured node under the shield can, top layer in red, openings of the solder mask in pink. Look for the pink line that runs around the large red area: it is the guard, bare of solder mask, around the copper of the measured node and the guard pour. From the multiplexer in the middle, thin lines run in pairs to the shunts: the Kelvin lines. At the lower right the pair of the 1 A range leaves the can in a lane with a guard track on each side:

The measured node and its guard on the top layer

The 1 A path on the top layer. Follow the wide copper from the linear regulator at the bottom left, along the supply band to the right, up to the shunt branch and the output switch, and out to the terminal block at the right edge. The dashed rectangle is made of the lands of the shield can:

The pours of the 1 A path on the top layer

The areas of the sixteen functional blocks and the parts inside them. The plot shows the silkscreen with the frames and names of the blocks, the fabrication layer with the outlines and the references of the parts, and the courtyards:

Placement of the functional blocks

The root page of the schematic, with one block for each of the 14 sheets:

Block level of the schematic

Every sheet and the board are shown in hardware/doc/, and the whole schematic is in hardware/doc/schematic.pdf. The status, the checks and the open work are in the hardware guide.

Circuit Simulations

The circuits of the carrier board are simulated from the netlist of the schematic. A bench names parts of the schematic by their reference designators, the package in simulation/ writes their elements from a snapshot of that netlist, and ngspice runs the circuit: no part of the board is typed by hand. What stands around the parts is typed by the bench and named in its notes: sources, loads, cables, and in some benches a stand-in for a neighboring block of the board. The simulator is the ngspice library that KiCad 10 ships. Every active part is a model written for this project from its datasheet, a typical part at room temperature. 41 benches put these models into the test circuits of their datasheets; four models have no such bench, the analog multiplexer among them. The models of the manufacturers may not be copied into the repository; where a copy is present they give a second value.

Everything in this section is simulated. Nothing is built and nothing is measured: a figure that passes says that the simulated circuit keeps its limit with these models, not that a board will. Elsewhere on this page a figure marked "simulated" is one of the specification, taken from the earlier simulations of the draft.

The results are those of the run of 2026-10-10, on the circuit as it was drawn that day. Three decisions were drawn afterwards (D-95, D-97 and D-98), and the netlist snapshot of the package holds them since: 10 nF at the three capacitors of the rail monitor, the damper on the module input fitted, the new text of the detector position. The benches were not run again. Every figure below therefore comes from the circuit before these changes, with 1 nF at those capacitors and without the damper. Two cases that a bench set on purpose are the exception: one crossing of a rail with 10 nF, and the data cable plugged again with the damper. The benches of the power input still leave the two parts of the damper out. Bringing the benches in line with the snapshot and running the blocks that the changes touch, the rail monitor with the power-up and the power-off and the input from the module, is open work.

BlockBenchesPassFailNo limit
Shunt ladder22006
Signal chain111138111
Range control logic121436122
Source meter121763247
Output stage61516142
Path switching1528819235
Power input and logic supplies1319027113
Analog rails and rail monitor1010525108
Digital lines and monitors101451149
Whole measuring path22503
Circuits of the board, ten blocks9313561051136
Models against their datasheets414781770
All13418341221206

A bench is one circuit with its stimulus. It takes figures from the waveforms and judges each one against its limit: a figure of the specification where it states one; else a rating or a figure of a datasheet, a value that the bench calculates, or a limit that the bench sets and names. For a model the limit is the figure of its datasheet with the tolerance that this project asks of a model. 1206 figures carry no limit: they are values that nothing limits. A figure that fails is kept as it is. Most of the 122 are of five kinds: a point that is decided since, whose figures are still those of the circuit before the decision, a figure of the specification that the circuit does not give, a rating that a fault case passes, an option that is not in use, and a model that misses its datasheet. The guide sorts them by block.

How close the simulated values lie: for 595 figures of the board blocks the pages print the distance from a stated value, in most cases the one that the specification states. 368 lie less than 5 % from it, 434 less than 10 % and 489 less than 25 %. The index of the results has every bench, and the page of each block has the values, the graphs and the decks, the circuit files that the simulator ran. The whole suite ran on one Windows machine and takes 29 minutes there.

What the Simulations Say

Reproduced. The central design figures of the specification come out of circuits taken from the netlist. That is a second calculation with models, not a confirmation by measurement. Among them:

  • Requirement R-07: on a step from 1 µA to 500 mA with 1 µF at the output the drop is 301.9 mV with nominal delays, and 388.2 mV with every delay at its limit and the capacitor 10 % low, against the limit of 500 mV. With 10 µF it is 168.3 mV and 179.2 mV, against 250 mV. An amplifier with half the bandwidth of its model, which is an assumption, takes the two worst cases to 400.3 mV and 180.8 mV.
  • The jump to the 1 A range takes 360.8 ns from its threshold with nominal delays and 480.3 ns with the worst ones, against a target of 550 ns. The 100 ns of the sequencer are an input of that run.
  • Gain 19.93 and zero code 1313; one code is 1.914 nA in the lowest range and 19.14 µA in the highest. After a jump to the highest range the chain settles to 0.1 % of the range in 45.37 µs, and in 49.2 µs at the most over the cases that were run. The noise of one sample in the lowest range is 2.098 nA from a quiet supply.
  • The source meter gives 799.5 mV and 5 V at its two nominal codes in steps of 1.282 mV, and its regulator keeps 449.2 mV to 631.9 mV of head room on the curve of R-08 with nominal parts.
  • The in-rush into 2200 µF at the output is 856.1 mA, below the trip level, and the closed path drops 155.7 mV at 1 A in ampere meter mode, with the 20 mΩ that the specification allows for copper and contacts (limit 200 mV).
  • With its switch open the external supply input takes 184.6 µA at +20 V and 597.9 µA at −20 V, against a limit of 1 mA.

Four points, decided. The simulations raised them for a decision before boards are ordered, and the project owner decided them on 2026-10-10. Three are drawn since that day and one is not. The benches were not run again, so the results above are still those of the circuit before them:

  1. The capacitors of the rail monitor are 10 nF (D-95), drawn since 2026-10-10. With 1 nF, the value of the simulated circuit, and a comparator at its least hysteresis, an edge of PWR_GOOD moves its own threshold and gives a burst of edges on a slow rail. With 10 nF the threshold moves by 525 µV, below that hysteresis, where 1 nF gives 3.149 mV to 3.606 mV; the one run made with 10 nF, at a slower slope, shows one edge. The capacitance between the pins that couples the edge is an assumption. The edges of PWR_GOOD at power-up and at power-off have not been simulated with 10 nF.
  2. The buffer of the driver rail of the converter gets a position for a capacitor at its input, whose value is chosen on the bench (D-96). Its loop has 41.92° of phase margin as drawn, not the 54° that the specification stated, and it is stable. The model of the manufacturer gives 68.99°, with input capacitances that are not those of its datasheet. No run with the capacitor exists, and the position is not drawn: it needs a footprint.
  3. The position for a voltage detector at the boost converter stays without a part (D-97). As drawn the converter starts in one go from an input limited to 0.67 A to 0.85 A. A detector with a time-out of 0.24 s in that position would stop that start and, with 1 mA of load, repeat it four times in the 1.1 s of the run without completing it. The result rests on a model of the converter below 2.7 V, where its datasheet says nothing, on a source with a flat current limit and on the 1 mA, so the start on a prototype stays a test of phase 1. Since 2026-10-10 the drawing no longer names the 0.24 s part at that position.
  4. The damper on the module input is fitted (D-98), drawn since 2026-10-10. When the data cable is plugged again on a port at 5.5 V behind a short cable while USB-C supplies the instrument, the input of the power multiplexer reaches 5.941 V with a limiter of typical reaction, and 6.002 V with one assumed to react in 15 µs, against a rating of 6.0 V. With the damper fitted the peak is 5.502 V in the runs made; the case that is worst with the damper, a longer gap of the contact, was not run, and neither was the recharge pulse at a change of input.

Fourteen places where the simulation and the text differed. They were figures and rules of the specification, and things on which it was silent: how long the over-current comparator stays high while a large capacitor recharges, the ladder clamp in a short circuit, the advice on a capacitor at the VIN terminals, the times at power-off, the opening time of the output switch at 0.8 V, the current in the output suppressor at a trip, three figures of the mode switches, a check of section 16 that the simulation does not meet, the list of terms of the noise budget, the timing estimates of the converter lines, the leakage of the monitor converter, figures of the source meter, the in-rush on a computer port and the range that a logic input withstands. Twelve are corrected in the specification since 2026-10-10, each with the simulated figure. Three of the corrections are decisions: the over-current trip counts its time in the 1 A range only (D-99), the advice on a capacitor at the VIN terminals is limited to what was simulated (D-100), and the range of a logic input is stated as 0 V to 5.5 V (D-101). The times at power-off and the figures of the power input wait for the run of the benches on the drawn decisions. Seven questions stay open, each with one recommendation in section 16 of the specification. The fourteen are not every difference: the guide names the further figures that fail against a rating of a datasheet.

What stays for the bench. The simulations cannot settle the linear range of the amplifier near the negative rail, the loop of the pre-regulator at its low end, the linear regulator at light load, a supply that leaves its range while the path is closed, leakage at the nanoampere level, the programs of the controller, coupling through the board and temperature.

The simulation guide gives each of these with its figures, with the bench that shows it and with the limits of the models behind it.

Two of the Results

Both pictures are plots of simulation runs, not measurements.

Simulated: a step from 1 µA to 500 mA with 1 µF at the terminals

Requirement R-07 in a simulation. The load steps from 1 µA to 500 mA in the lowest range, with 1 µF at the terminals and 5 V at the output. From the top: the drop between the supply node and the output terminal, the voltage at the terminal, the current in the branch of the 0.1 Ω shunt, and the selected range. The step-up comparator takes the next range, then the jump comparator takes the path to the 1 A range. The drop peaks at 301.9 mV with nominal delays and at 388.2 mV with every delay at its limit and the capacitor 10 % low, against the limit of 500 mV. The requirement also allows the drop above 0.2 V for 1 µs at the most, which is the dashed line; the two curves are above it for 302 ns and 390 ns. The third curve gives the sequencer 300 ns, three times what the specification allows it, and reaches 445.1 mV. The reaction time of the sequencer is an input of this run: the program that has to keep it is not written.

Simulated: load current to code and to the reading, every range held

The whole measuring path in a simulation: ladder, multiplexer and amplifier chain as one circuit from the netlist, each range held while the load current rises to 120 % of its full scale. Above, the code of an ideal 16-bit converter at the converter input: the four ranges cover 100 nA to 1.2 A, with zero at code 1313. Below, the error of the reading with the nominal calibration against the load current; at small currents it is the rounding to one code. The amplifiers have no offset, no bias current and no noise in this run, and the converter is arithmetic.

Repository Structure

PathContentLicense
firmware/Firmware of the controller: a project of the Pico SDK for the Pico 2, with a hardware-independent core and its unit tests; no adapter for the hardware and no PIO program yetMIT; the helper cmake/pico_sdk_import.cmake is a copy from the Pico SDK and keeps the license in its header
hardware/KiCad project of the carrier board, its pictures and bill of materials, and the fabrication outputs of draft A2: Gerber and drill files and two assembly files, filed for an order of proof-of-concept boards (D-105)CERN-OHL-P v2
simulation/Circuit simulations of the carrier board: the package circuit-sim with its tests, a snapshot of the netlist of the schematic, the models written for the project, the benches and their resultsMIT; the netlist snapshot in netlist/ and the results in results/: CERN-OHL-P v2
host/Python package: protocol, transports, device client, capture helpers, simulatorMIT
protocol/Protocol definition, generator and shared test vectorsMIT
tools/tools/board/: the package that calculates the layout figures from the board file, with its tests. The folders for the calibration and production-test scripts are emptyMIT
docs/Specification, firmware design, component checks (all open), test reports (none yet)MIT

Roadmap

The work is split into phases. Each phase is closed by a test report with recorded measurements. The exit criteria are in section 13 of the specification.

PhaseGoalStatus
1Risk prototypes. On a Pico 2 with an ADC evaluation module: firmware on the Pico SDK, hardware-timed ADC capture, reaction time of the range sequencer, USB throughput. On evaluation modules: the pre-regulator with its tracking amplifier, and the start of the boost converter from a supply limited to 0.7 ANot started: nothing has run on a Pico 2. Written ahead of it and tested without hardware: the firmware core, the protocol at version 2 and the host package. The firmware builds for the Pico 2 with the Pico SDK since 2026-10-12; the image has not been loaded onto a board, and what the phase measures is not written: the capture, the PIO programs and the USB link. The bench work is not started
2Analog front end with one fixed range on a test boardNot started
3Shunt ladder and range logicNot started
4Source mode and powerNot started
5Carrier board, revision A, with the Pico 2 plugged in. Entry: the open checks that precede fabrication are closed and the layout is reviewedNot started; draft A2 of the schematic and a board with every connection drawn exist, drawn ahead of the phases. The layout was reviewed with scripts and by calculation, and its seven deviations from section 10 are recorded decisions (D-87 to D-93). Of the four decisions that follow the simulations, three are drawn since 2026-10-10 (D-95, D-97 and D-98). On 2026-10-13 the project owner looked at the project in the KiCad editor and found it coherent, and the footprints and the placement were reviewed with scripts, 3D renderings and datasheets; the five findings of that review are worked into the board. Still to do: the fourth decision drawn (D-96), the benches that the drawn decisions touch run again, the review of the board item by item by a person in the KiCad editor, and the open items. The fabrication outputs of draft A2 are filed since 2026-10-13 for an order of proof-of-concept boards (D-105). Such boards are draft A2 and not revision A, and no order is recorded
6Calibration, protocol freeze and host softwareNot started
7Revision B and releaseNot started

Draft A2 is the hypothesis that the phases test: the component checks and the results of phases 1 to 4 change it before revision A is fabricated. Proof-of-concept boards of the draft may be made before that: the project owner decided so on 2026-10-13, and the fabrication outputs are filed (D-105). Such boards are draft A2, not revision A. The circuit simulations test the same hypothesis ahead of the bench, with models in place of parts. They are not a phase and close none: a phase is closed by measurements. The work that comes first is listed under Project Status.

Getting Started

sh
git clone https://github.com/Guiimartinho/open-power-profiler.git
cd open-power-profiler

Start with the specification. Each part of the repository documents its own setup and commands:

PartNeedsGuide
FirmwareCMake, Ninja and gcc for the unit tests; for the image of the Pico 2 also the Pico SDK 2.3.1 and the Arm GNU Toolchainfirmware/README.md
Host softwarePython 3.10 or laterhost/README.md
ProtocolPython 3.11 or laterprotocol/README.md
HardwareKiCad 10hardware/README.md
Board figuresPython 3.10 or later; KiCad 10 to dump the board filetools/board/README.md
Circuit simulationsPython 3.10 or later; the ngspice shared library, which KiCad 10 shipssimulation/README.md
DocumentationNode.js, for npx --yes markdownlint-cli2@0.23.3CONTRIBUTING.md

No instrument is needed to work on the software: the firmware logic runs in unit tests on the PC and its image builds without a board, the host package talks to a simulator, the board figures package is tested on small synthetic boards, and the circuit simulations need only the ngspice library that KiCad installs.

Engineering Practices

  • Logic that does not need hardware is kept apart from the code that does, in firmware and in host software, and is covered by unit tests.
  • The wire protocol has one definition. The constants of both sides are generated from it, and both codecs must reproduce the same test vectors.
  • Seven continuous integration workflows, one per area (Firmware, Host, Protocol, Tools, Simulation, Hardware, Docs), build the firmware, run the tests with coverage floors, run the formatters and static analyzers and check the hardware project. The firmware build is the one for the Pico 2 since 2026-10-12: its job in the Firmware workflow was rewritten for the Pico SDK and has not been started yet. For now the workflows are started by hand, not on every push (D-22); the same checks run locally before a change is recorded.
  • The gates of the firmware and of the host package ran locally on Windows on 2026-10-12. Firmware: 22 unit tests, every line and branch of the core covered against floors of 90 % and 80 %, clang-format, clang-tidy and cppcheck without a finding, and the image for the Pico 2 built with warnings as errors. Host: 500 tests, every line and branch covered against a floor of 90 %, ruff without a finding, mypy in strict mode clean. The generated protocol files are up to date. The sanitizer builds of the firmware and the host tests on Linux and macOS cannot run on that machine and did not run.
  • The board figures package in tools/board/ has the gates of the host software: tests with a coverage floor of 90 % of lines and branches, ruff, mypy in strict mode, and the wheel built and installed. They ran locally on Windows; the Tools workflow, which repeats them on Linux and macOS too, has not been started yet.
  • The simulation package in simulation/ has those gates as well: 777 tests, of which 11 need the ngspice library, ruff, and mypy in strict mode over the package, its tests and the benches. Lines and branches are covered in full against a floor of 90 %; the one module that loads the simulator runs in a child process and is outside that measurement. The gates ran locally on Windows with Python 3.11, last on 2026-10-12; the Simulation workflow has not been started yet. They cover the package, not the board: a figure that fails in a simulation breaks no gate. It is a finding, and it is kept.
  • Timing, throughput and analog behavior are not claimed from tests. They are measured on the bench and recorded in a report. A simulation is filed as a simulation: it closes no check and no phase.

The rules are in section 18 of the specification and in the contributing guide.

Documentation

  • System specification: requirements, analog and firmware architecture, pin map, host protocol, calibration, verification plan, risks and decision log.
  • Hardware guide and the draft in pictures: the carrier board, sheet by sheet.
  • Fabrication files of draft A2: the Gerber, drill and assembly files for an order of proof-of-concept boards, the order data of the bare board, and what was checked on the files and what was not.
  • Firmware design: the firmware one level below the specification, with its architecture, its state machines, the three PIO programs, the USB link, the place and the test of every firmware rule, the decisions that are taken and the points that are open. It is a design: of what it describes, a first part of the core and the build exist.
  • Firmware guide, host guide and protocol guide: architecture, commands and tests of each part, with what exists and what is missing.
  • Simulation guide and simulation results: what the circuit simulations reproduce, the four points that are decided, where they differ from the text of the specification, the limits of the models, and every figure with its graph and its circuit file.
  • Open checks: what has to be confirmed, part by part and on the bench, before the schematic is frozen, and the records that accept a candidate part against its datasheet; none is filed yet.
  • Test reports: the measurements that close each phase; no report is filed yet.
  • Calibration and protocol reference: what the specification defines for each and what is not written.
  • Tools: the board figures, which exist, and what the calibration tool and the production test will have to do; neither is started.
  • Board figures guide: how the figures of the layout are calculated from the board file, with the method, the assumptions and the limits.
  • Documentation index: every document of the repository and how they relate.
  • Changelog: notable changes by release.

Contributing

Contributions are welcome. Read the contributing guide first: it covers the workflow, the commit message format, the quality gates and how changes to the specification are recorded.

License

  • Hardware design files under hardware/: CERN-OHL-P v2.
  • What derives from those files in simulation/: the snapshot of the netlist in simulation/netlist/ and the decks, records, graphs and pages in simulation/results/, under the same CERN-OHL-P v2, with a copy of the license in each of the two folders.
  • Everything else, including firmware, host software, tools, the simulation package with its benches and models, and documentation: MIT.

Acknowledgments

  • Inspired by Gedankenn/power_profiller, an ESP32 and INA226 power profiler with a web dashboard. This project started on the ESP32-S3 too, and was called esp32-s3-power-profiler until its controller changed.
  • The feature set and the pin order of the connectors follow the Nordic Semiconductor Power Profiler Kit II (PPK2). This project is independent and is not affiliated with or endorsed by Nordic Semiconductor.
  • The controller module is a Raspberry Pi Pico 2. This project is independent and is not affiliated with or endorsed by Raspberry Pi Ltd.

Parts

425
QtyValueFootprintReferences
51100n 50V X7RCapacitor_SMD:C_0603_1608MetricC1, C13, C25, C100, C101, C102, C103, C104, C105, C106, C107, C108, C109, C110, C111, C113 +35
201k 1%Resistor_SMD:R_0603_1608MetricR4, R26, R53, R60, R62, R70, R72, R75, R88, R89, R93, R94, R111, R112, R118, R146 +4
201u 25V X7RCapacitor_SMD:C_0603_1608MetricC6, C7, C14, C15, C16, C17, C19, C24, C41, C47, C56, C62, C64, C65, C78, C112 +4
17100k 1%Resistor_SMD:R_0603_1608MetricR17, R24, R27, R42, R50, R78, R80, R85, R86, R103, R106, R109, R115, R148, R149, R151 +1
1310k 1%Resistor_SMD:R_0603_1608MetricR16, R19, R23, R25, R41, R54, R113, R114, R126, R144, R145, R150, R152
10BAV199Package_TO_SOT_SMD:SOT-23D10, D13, D15, D16, D17, D18, D19, D20, D22, D23
74.7u 25V X7RCapacitor_SMD:C_0805_2012MetricC2, C18, C20, C22, C28, C30, C63
7CSD17577Q3APackage_SON:VSON-8_3.3x3.3mm_P0.65mm_NexFETQ4, Q5, Q8, Q9, Q14, Q15, Q16
510R 1%Resistor_SMD:R_0603_1608MetricR31, R61, R108, R120, R129
51N5819HWDiode_SMD:D_SOD-123D1, D3, D4, D11, D12
522R 1%Resistor_SMD:R_0603_1608MetricR77, R79, R92, R100, R130
522u 25V X5RCapacitor_SMD:C_1206_3216MetricC42, C43, C45, C50, C53
5GNDTestPoint:TestPoint_Pad_2.0x2.0mmTP15, TP20, TP25, TP28, TP29
410n 50V C0GCapacitor_SMD:C_0603_1608MetricC32, C33, C34, C74
410u 25V X5RCapacitor_SMD:C_0805_2012MetricC5, C39, C40, C46
45.1k 1%Resistor_SMD:R_0603_1608MetricR7, R11, R91, R95
4B0530W 30V 0.5ADiode_SMD:D_SOD-123D6, D7, D8, D9
4IRLML0030Package_TO_SOT_SMD:SOT-23Q10, Q11, Q12, Q13
4M3MountingHole:MountingHole_3.2mm_M3_PadH1, H2, H3, H4
4TPD4E1U06DBVRPackage_TO_SOT_SMD:SOT-23-6U2, U35, U36, U37
30.33R 1%Resistor_SMD:R_0805_2012MetricR14, R20, R59
3100k 0.1% 25ppmResistor_SMD:R_0603_1608MetricR36, R67, R74
310u 25V X7RCapacitor_SMD:C_1210_3225MetricC27, C29, C31
32.2M 1%Resistor_SMD:R_0603_1608MetricR83, R84, R117
322u 10V X7RCapacitor_SMD:C_1206_3216MetricC9, C10, C21
347R 1%Resistor_SMD:R_0603_1608MetricR102, R105, R119
349.9k 1%Resistor_SMD:R_0603_1608MetricR44, R46, R47
36.8k 1%Resistor_SMD:R_0603_1608MetricR52, R81, R82
3BSS138Package_TO_SOT_SMD:SOT-23Q1, Q2, Q3
3FiducialFiducial:Fiducial_1mm_Mask2mmFID1, FID2, FID3
3GNDTestPoint:TestPoint_Pad_D1.5mmTP7, TP8, TP39
3OPA197IDBVPackage_TO_SOT_SMD:SOT-23-5U17, U25, U26
3OPA365AIDBVPackage_TO_SOT_SMD:SOT-23-5U19, U28, U29
3TC4427EOAPackage_SO:SOIC-8_3.9x4.9mm_P1.27mmU20, U22, U23
21.5k 1%Resistor_SMD:R_0603_1608MetricR3, R51
2100kResistor_SMD:R_Array_Convex_4x0603RN6, RN7
210k 0.1% 25ppmResistor_SMD:R_0603_1608MetricR22, R57
211k 0.1% 25ppmResistor_SMD:R_0603_1608MetricR58, R124
2120k 1%Resistor_SMD:R_0603_1608MetricR140, R142
21k 0.1% 25ppmResistor_SMD:R_0603_1608MetricR127, R137
21n 50V C0GCapacitor_SMD:C_0603_1608MetricC12, C58
22.00k 0.1% 25ppmResistor_SMD:R_0603_1608MetricR55, R56
2220p 50V C0GCapacitor_SMD:C_0603_1608MetricC23, C77
222p 50V C0GCapacitor_SMD:C_0603_1608MetricC75, C76
23.01k 1%Resistor_SMD:R_0603_1608MetricR6, R10
23.9k 1%Resistor_SMD:R_0603_1608MetricR125, R128
2330RResistor_SMD:R_Array_Convex_4x0603RN8, RN9
24.53k 1%Resistor_SMD:R_0603_1608MetricR5, R9
24.7kResistor_SMD:R_Array_Convex_4x0603RN2, RN3
24.7k 1%Resistor_SMD:R_0603_1608MetricR1, R71
2402k 1%Resistor_SMD:R_0603_1608MetricR8, R12
2470kResistor_SMD:R_Array_Convex_4x0603RN10, RN11
2BC847BPackage_TO_SOT_SMD:SOT-23Q6, Q7
2LP5907MFX-3.3Package_TO_SOT_SMD:SOT-23-5U7, U8
2MCP6561T-E/OTPackage_TO_SOT_SMD:SOT-23-5U21, U32
2SN74LV165ADRPackage_SO:SOIC-16_3.9x9.9mm_P1.27mmU33, U34
2TPS259621DDARPackage_SO:TI_SO-PowerPAD-8_ThermalViasU3, U4
1+12V_ATestPoint:TestPoint_Pad_D1.5mmTP16
1+13V5TestPoint:TestPoint_Pad_D1.5mmTP13
1+3V3_ATestPoint:TestPoint_Pad_D1.5mmTP10
1+3V3_CTestPoint:TestPoint_Pad_D1.5mmTP11
1+5VTestPoint:TestPoint_Pad_D1.5mmTP6
1-4V_ATestPoint:TestPoint_Pad_D1.5mmTP12
10.1R 0.25% 50ppmResistor_SMD:R_Shunt_Ohmite_LVK12R110
10.22R 1%Resistor_SMD:R_0603_1608MetricR131
10.47R 1% 0.25WResistor_SMD:R_1206_3216MetricR37
10.47R 1% 0.5WResistor_SMD:R_1206_3216MetricR87
10RResistor_SMD:R_0603_1608MetricR2
11.02k 0.1% 25ppmResistor_SMD:R_0603_1608MetricR122
11.21k 1%Resistor_SMD:R_0603_1608MetricR13
11.3k 1%Resistor_SMD:R_0805_2012MetricR69
11.5n 50V C0GCapacitor_SMD:C_0603_1608MetricC85
11.5u 4.6AInductor_SMD:L_Coilcraft_XxL4020L2
1100R 1%Resistor_SMD:R_0603_1608MetricR139
1100n 50V C0GCapacitor_SMD:C_1206_3216MetricC71
110n 25V C0GCapacitor_SMD:C_0603_1608MetricC90
110p 50V C0GCapacitor_SMD:C_0603_1608MetricC52
110u 16V X7RCapacitor_SMD:C_0805_2012MetricC26
110u 3AInductor_SMD:L_Wuerth_MAPI-4030L1
1115k 0.1%Resistor_SMD:R_0603_1608MetricR73

The 309 most common of 425 parts.

Similar boards

  • 2 layers79 × 83 mmInterface module, the printed circuit board in 3D

    Interface module

    psas/mints-scada-hardware

    Minimal Test Stand Supervisory

    113 parts
  • 4 layers120 × 107 mmSL23-modem tester, the printed circuit board in 3D

    SL23-modem tester

    JitterCompany/debug-tools

    Collection of useful utility PCBs for development purposes

    229 partsMIT
  • 2 layers50 × 35 mmSoil pH AFE, the printed circuit board in 3D

    Soil pH AFE

    xmyoot/guarded-ph-acquisition-module

    Guarded pH analog front end: KiCad design, fabrication files, ESP32-S3 bring-up firmware and engineering review. Unbuilt prototype.

    44 parts
  • 4 layers202.3 × 90 mmMythra psu, the printed circuit board in 3D

    Mythra psu

    bad-alloc-heavy-industries/Mythra

    45 parts7BSD-3-Clause
  • 8 layers280 × 195 mmMaxwell main, the printed circuit board in 3D

    Maxwell main

    azonenberg/starshipraider

    Open hardware test equipment

    997 parts193BSD-3-Clause
  • 4 layers54.6 × 61 mmSapphineMinimal, the printed circuit board in 3D

    SapphineMinimal

    ChrisHayward-dev/SapphireKicad

    See the LastProduction branch for the most current revision

    159 parts

Questions

How big is the Power profiler carrier PCB?

It measures 150 × 100 mm and has 4 copper layers, 1.6 mm thick.

What parts does the Power profiler carrier use?

425 parts, built around the OPA197IDBV, OPA365AIDBV, TC4427EOA713 and TPD4E1U06DBVR. The full list of its parts is on this page, with links to each part's symbol and footprint where PCBKit's library has them.

Where can I download the Power profiler carrier KiCad files?

From the Guiimartinho/open-power-profiler repository on GitHub, at the commit this page was made from: the board is hardware/kicad/power-profiler-carrier.kicad_pcb and its schematic hardware/kicad/analog_rails.kicad_sch. PCBKit opens the board in the browser, with nothing to install.

Can I use the Power profiler carrier design in my own project?

Its repository publishes it under the MIT licence: you may reuse it on that licence's terms (keep the author's notice, and for a copyleft licence share your changes alike).