Board bring-up and debugging

A board that powers on has not passed anything. Bring-up is a staged plan in which each stage gives you evidence before the next one risks the hardware, and in which every surprise gets written down. The goal is to find faults while each is still a single fault.

The bring-up order

1. Inspect. Under a loupe or microscope check polarity, pin 1, bridges and missing parts. The stripe on a moulded tantalum marks the positive end, the opposite of an aluminium electrolytic, and on a diode it marks the cathode.
2. Power off, measure. Resistance from every rail to ground (hundreds of ohms to kilohms is normal, near zero is a short), rail to rail, and continuity of ground.
3. Current-limited power-up. Set the bench supply to the battery voltage with the limit just above the expected idle: 100 to 150 mA for the motor board, motors unplugged. Raise the voltage from zero and watch the current.
4. Rails. Check each rail against its tolerance (3.3 V within 3 % is 3.20 to 3.40 V), then look at ripple on a scope.
5. Clocks and reset. Crystal running, reset line high.
6. Programming. Read the chip ID with esptool before writing anything.
7. Peripheral by peripheral. LED, buttons, UART, then the motor driver with a motor off the ground at 20 % PWM, then encoders and sensors. Add one thing at a time.

What the supply tells you

A bench supply is the cheapest diagnostic you own. In constant-current mode with the voltage collapsed, the ratio of the two is the apparent load resistance: 1.9 V at 0.8 A is 2.4 Ω, which is no working board. A healthy idle shows the full voltage and a small, steady current that matches your estimate. If the current creeps up over several seconds, something is heating and its resistance is falling. Remember the power, too: a supply limited to 1 V and 0.5 A can put at most 0.5 W into a shorted rail, and a real short drags the voltage down (0.5 A through 0.45 Ω is only 0.2 V), so the failed part itself dissipates about 0.1 W. That is enough to warm it for a thermal camera but far too little to damage the board.

Instruments and what each is for

A multimeter (10 MΩ input) is for DC voltage, resistance with the power off, continuity and the diode test, where a good silicon junction reads 0.5 to 0.7 V forward and open in reverse. Its current ranges add burden voltage in series, so fuse them and use them sparingly.

An oscilloscope shows shape. Pick a bandwidth of at least five times the highest clock; its rise time is 0.35 divided by its bandwidth, so a 100 MHz scope cannot show an edge faster than 3.5 ns. A 10× passive probe is 10 MΩ with about 10 pF, and its ground lead is the part most people get wrong. Set the 20 MHz bandwidth limit when you want ripple rather than noise, AC-couple a 3.3 V rail to read millivolts, and trigger single-shot on a falling edge just below the rail to catch a dip.

A logic analyzer records many digital channels at 24 to 500 MS/s and decodes I2C, SPI and UART. It tells you what bits were sent, not whether the edges were clean, so sample at least four times the bus clock and set the threshold for your logic level. A thermal camera turns a short into a picture: power the board from a current-limited supply and look for the hot part, remembering that bare copper and shiny solder have low emissivity (about 0.1) and read cold, while mould compound and solder mask (about 0.9) read true. Without a camera, a drop of isopropyl alcohol evaporates first on the hottest part, and freeze spray on the culprit makes the supply current change. A clamp-style current probe, or a sense resistor that is already in the circuit, shows pulsed currents without breaking it; a new series shunt has to be put in the path.

The usual suspects

Most first-article faults are boring: solder bridges on fine-pitch pins, opens from cold joints or tombstoned parts, rotated or flipped parts (diodes, tantalums and ICs), wrong values or footprints, brown-outs when a motor or the radio starts, and ESD damage that shows up as a leaky pin days later. A brown-out is a rail that dips below the MCU's threshold for a few microseconds, and only a single-shot trigger will catch it. Handle boards on a grounded mat, because ESD damage often appears days later as a leaky pin or an input that sticks. Probing artefacts account for the rest, which is why the ground lead matters.

A systematic method

Observe, hypothesise, change one thing, record. Split the system in half and ask which half the fault lives in. Compare a bad board with a known-good one at the same node. Never change two variables at once. Keep a bring-up log with the board serial and revision, firmware hash, supply settings, every reading with its units, scope screenshots with the probe point marked, and what you changed. Next week's you will need it.

Rework basics

Use a temperature-controlled iron (about 350 °C for lead-free) with fresh flux, spend no more than three to four seconds on a pad so it does not lift, and let it cool. Bridges come off by drag soldering with flux or wick. Small SMD parts lift by heating both ends. QFNs and SOICs need hot air at 350 to 380 °C with the board preheated. Clean flux residue with isopropyl alcohol and inspect again.

First walk the decision tree for a symptom and see which instrument each step uses. Then take the find-the-short bench: power the board from the battery input and from the 3V3 rail, and notice which method gives the culprit away.