← AXS-042

Circuit review & bench-test guide

AXS-042 — SCD40 true-NDIR CO2 sensor PMOD module

Document purpose#

This document explains the axs-042-scd40 prototype at component level and turns the design evidence into a practical manual-review and bench-test plan. It is based on the implemented schematic (generate_design.py, axs-042-scd40.kicad_sch), the module README, and the Sensirion SCD4x datasheet (version 1.5, July 2023).

The board is schematic-only. ERC is clean (0/0) and the netlist has been reviewed, but no PCB layout exists, no board has been fabricated, and there is no recorded bench result. Every numeric value below — currents, command timings, CRC bytes, expected CO2 readings — is a verification target taken from the schematic and the datasheet, not a measured result.

1. What the board does#

The board is an I2C PMOD module around the Sensirion SCD40-D-R2, a true photoacoustic-NDIR CO2 sensor (not a metal-oxide eCO2 estimator) with an integrated SHT4x-class temperature/humidity sensor for on-chip compensation. It reports CO2 in ppm (specified range 400–2000 ppm for the SCD40; output range 0–40000 ppm), temperature, and relative humidity over I2C at fixed address 0x62.

The interface is Sensirion's command protocol: 16-bit commands, 16-bit data words each followed by a CRC-8 byte, and millisecond-to-second command latencies. The IR-emitter pulses draw large supply-current peaks (up to 205 mA maximum at 3.3 V), which is why the schematic fits two 10 µF bulk reservoirs plus a 100 nF decoupler at the sensor.

Functional block diagram#

 PMOD host J1
 3V3 (6,12) ────┬────────┬────────┬──────────┐
 GND (5,11) ────┼────────┼────────┼──────┐   │
                │        │        │      │   │
              C1 100n  C2 10µ   C3 10µ   │   │  emitter-pulse reservoirs
                │        │        │      │   │
 SCL (3,9) ──┬──┴────────┴────────┴──> U1 SCD40 (addr 0x62)
 SDA (4,10) ─┼─┬                       VDD+VDDH = 3V3, GND
             │ │                       photoacoustic NDIR CO2 + T/RH
  R2 4.7k ───┘ │  (via JP2, SCL)
  R1 4.7k ─────┘  (via JP1, SDA)

2. Safety and scope boundaries#

• Electrically this is a benign 3.3 V board, but the peak supply current is up to 205 mA at 3.3 V during emitter pulses. A bench current limit set below ≈250 mA will brown out the sensor mid-measurement and produce confusing resets — set the limit accordingly and note it in the record.
• Sensirion specifies the unloaded supply ripple below 30 mV peak-to-peak; a long, thin bench lead defeats the on-board reservoirs. Keep leads short.
• The white protection membrane on the sensor top must not be removed or touched; the sensor opening must stay unobstructed. Handling damage is invisible and permanent.
• Do not blow breath directly at the sensor as a "CO2 test" and then trust the absolute number; exhaled air saturates it (>>5000 ppm). Use it only as a gross respond/no-respond check.
• The automatic self-calibration (ASC) algorithm, enabled by default, assumes weekly exposure to ~400 ppm fresh air. Long sealed-box storage of a running sensor will corrupt its baseline; that is an algorithm property, not a board fault.
• MSL 3 package: after the vacuum bag is opened, floor life and any reflow must follow Sensirion's soldering instructions. Hand-rework of the LGA-21 with a hot-air pencil is high-risk to the membrane.
• This prototype has no layout, no fab, and no bench history. All numbers below are targets to verify, not achieved results.

3. Power and control sequence#

1. 3V3 and GND come directly from the PMOD host (J1 pins 6/12 and 5/11); the board has no regulator. VDD and VDDH of the SCD40 are the same 3V3 net, as the datasheet requires (they must be kept at the same voltage).
2. Power-up: the sensor is ready for commands at most 30 ms after VDD passes the power-up threshold (idle state). It powers up in idle: no measurement runs and supply current stays low until commanded.
3. start_periodic_measurement (0x21B1) begins the 5-second measurement cycle. From then on, average current is ≈15 mA (18 mA max) at 3.3 V, with pulse peaks up to 205 mA drawn from C2/C3 and the supply.
4. First data become available ≈5 s after start (one signal-update interval). Data readiness is polled with get_data_ready_status (0xE4B8).
5. stop_periodic_measurement (0x3F86) returns to idle; the sensor accepts no other command for 500 ms afterwards. Most configuration commands are legal only in idle mode.
6. There is no reset pin; recovery paths are the reinit command (0x3646, 30 ms, only in idle) or a power cycle.

4. Interfaces#

PMOD pin map (Type 6 extended, J1)#

Pins 9/10 repeat the bus for module chaining.

Jumpers#

There are no address jumpers: the SCD40 address is fixed at 0x62. Two SCD40 modules cannot share one bus.

Test points#

None are defined in the schematic. Probe at J1 pins and jumper pads. Adding TPs for 3V3 (at the sensor), SDA, and SCL is a recommended layout task, especially for observing supply droop during emitter pulses.

5. Component-by-component review#

Every reference designator in generate_design.py is covered.

5.1 Active device and connector#

5.2 Pull-ups, decoupling, and jumpers#

5.3 ERC-only artifacts#

6. Datasheet summary and design interpretation#

Key commands used in this guide (datasheet Table 9):

Official references: Sensirion SCD4x datasheet v1.5, Sensirion SCD40 product page, Yageo RC thick-film resistors, Murata capacitor search. Check the latest datasheet revision before procurement.

7. Expected values before bench testing#

All values are calculated/datasheet targets, not measurements.

8. Manual schematic and assembly review checklist#

• Verify U1 pin-1 orientation via the notched corner of the protection membrane; LGA-21 is not visually reversible after reflow without X-ray.
• Confirm VDD (pin 7) and VDDH (pin 19) are both on 3V3 and GND pads (6, 20, center pads) are grounded; DNC pads soldered to floating pads only.
• Confirm no solder paste on the exposed thermal-pad opening per the schematic note and Sensirion handling guidance.
• Membrane clean, no flux residue, no paste splatter; opening unobstructed by neighboring tall parts (layout review item).
• JP1/JP2 bridged (default) or deliberately cut per chain plan.
• C1 placed nearest the sensor supply pads, C2/C3 adjacent (layout task).
• Buzz J1 pin 3 → U1 SCL, pin 4 → U1 SDA, and the 9/10 duplicates.
• 3V3-to-GND: no hard short (ohmmeter rises as 20 µF charges).
• Confirm the CRC-8 implementation in the host/HDL passes 0xBEEF → 0x92 before first attach.

9. Ordered bench-test procedure#

Stop at the first abnormal result. Assumed bench: PMOD host (3.3 V) or current-limited 3.3 V bench supply, limit ≥250 mA, short leads.

A. Unpowered inspection and resistance tests#

1. Complete the checklist in section 8 under magnification.
2. Ohmmeter: 3V3-to-GND no hard short; SDA-to-3V3 ≈4.7 kΩ; SCL-to-3V3 ≈4.7 kΩ (JP1/JP2 fitted verification — cut jumpers must read open).
3. SDA-to-SCL: open (no bridge across the jumper bank).

B. First power-up, bus idle#

1. Apply 3.3 V, current limit 250 mA. Wait ≥30 ms (power-up time).
2. Supply current: expect a low idle value well under the 15 mA periodic average; record it. Sustained tens of mA before any command suggests a sensor stuck mid-measurement from a previous run or a fault.
3. DMM: SDA and SCL both ≈3.3 V (bus idle high through the pull-ups).

C. Address ACK and first transaction: serial number with CRC check#

1. I2C scan at 100 kHz: expect ACK at 0x62 only.
2. Defensive start: because the sensor may still be in a periodic mode from a previous session, first send stop_periodic_measurement: S 0x62+W ACK 0x3F ACK 0x86 ACK P, then wait 500 ms.
3. Serial-number read (bring-up transaction):

• Write command 0x3682: S 0x62+W 0x36 0x82 P
• Wait ≥1 ms (command execution time).
• Read 9 bytes: S 0x62+R b0 b1 crc0 b2 b3 crc1 b4 b5 crc2 NACK P
• Verify each CRC-8 (poly 0x31, init 0xFF) over its preceding word pair. Datasheet example: a serial of 273'325'796'834'238 returns 0xF8 0x96 0x31 0x9F 0x07 0xC2 0x3B 0xBE 0x89 — your serial differs, but all three CRCs must verify and the 48-bit serial must be nonzero.

1. A NACK on the read header here means the 1 ms wait was skipped or the sensor is still busy (e.g. inside the 500 ms stop window).

D. First CO2 measurement#

1. Send start_periodic_measurement: S 0x62+W 0x21 0xB1 P.
2. Poll get_data_ready_status no faster than ~every 500 ms: write 0xE4B8, wait ≥1 ms, read 3 bytes (word + CRC). Data not ready while the least-significant 11 bits are all 0 (e.g. response 0x80 0x00 0xA2); ready otherwise. Expect readiness after ≈5 s.
3. Send read_measurement 0xEC05, wait ≥1 ms, read 9 bytes:

• word[0] = CO2 in ppm (direct); word[1] = raw temperature, T = −45 + 175 × raw / (2^16 − 1); word[2] = raw RH, RH = 100 × raw / (2^16 − 1); each word followed by its CRC.
• Datasheet worked example: 0x01 0xF4 0x7B 0x66 0x67 0xA2 0x5E 0xB9 0x3C = 500 ppm, 25 °C, 37 % RH.

1. Plausibility: indoor CO2 400–2000 ppm (a just-ventilated room sits near 400–600 ppm; an occupied closed room drifts up). Temperature within a few °C of ambient (the default 4 °C offset is tuned for enclosed products; on an open board expect a warm-reading bias to verify and tune later). RH within ±9 % of a reference.
2. Gross response check: exhale near (not onto) the sensor; CO2 must rise by hundreds-to-thousands of ppm within a few update intervals and decay afterwards. Treat direction of change, not the absolute number, as the result.
3. Observe supply behavior during measurement: pulses on the 3V3 rail with droop < 0.1 V at the sensor. If the bench limit trips, raise it — peaks up to 205 mA are specified behavior.

E. Stop discipline and idle-mode commands#

1. Send stop_periodic_measurement 0x3F86 and wait 500 ms. Confirm the sensor NACKs/ignores idle-only commands sent inside that window (expected protocol behavior, useful driver test).
2. In idle, run perform_self_test 0x3639: wait 10 s, read 3 bytes; expect word 0x0000 + valid CRC (no malfunction detected).
3. Optional: reinit 0x3646, wait 30 ms, then repeat the serial-number read to confirm recovery paths work.

F. Chaining and endurance (deferred)#

1. Chain test with another AXS module: cut JP1/JP2 on one board, verify ohms (cut = open, fitted = 4.7 kΩ), confirm both devices ACK.
2. Multi-hour soak logging CO2/T/RH; watch ASC behavior across days and verify against a reference meter before trusting absolute calibration.
3. Do not run FRC (forced recalibration) or persist_settings casually: they write sensor EEPROM/calibration state.

10. Troubleshooting map#

11. Bench record template#

12. Review conclusion#

The schematic follows the Sensirion application circuit: commoned VDD/VDDH, pull-ups behind cut jumpers, and a deliberate 20 µF + 100 nF local reservoir sized for the documented 205 mA emitter peaks — the single most likely bench failure source if ignored. The protocol burden (16-bit commands, per-word CRC-8, execution-time waits, NACK-not-stretch semantics, the 500 ms stop window) lives in the host/HDL, and this guide pins the exact bytes to send and expect. Principal open risks: no layout (sensor-opening keep-out, paste keep-out on the exposed pad, and reservoir placement are unproven), no fabricated first article, no bench data, and no reference CO2 instrument yet identified for absolute-accuracy verification. Nothing above should be read as evidence the board works — it is the plan for finding out.