A little ARM board that had been running happily in the cupboard for two years stopped powering on. No lights, no fan twitch, nothing. The honest, sensible thing would have been to spend eight pounds on a replacement. I did not do the sensible thing, because there's a particular satisfaction in bringing a dead board back, and because if I always replaced things I'd never learn why they die. So out came the iron.
start with the boring questions
Before touching anything, I ruled out the obvious. Was it the power supply? I had a known-good one of the right voltage and current, swapped it in, no change. Was it the cable? Swapped that too. It is genuinely embarrassing how often a "dead" board is a dead barrel jack on a four-quid PSU, and checking takes thirty seconds. This time it wasn't the supply, which meant the fault was on the board.
The first tool out after that is always the eyes, then the nose. Burnt components have a smell you don't forget, and a blown capacitor often bulges or leaks visibly. I went over the board under a desk lamp with a loupe, looking for scorch marks, bulged cans, or lifted components. Nothing dramatic. But near the power input there was a capacitor whose top looked very slightly domed, the kind of thing you'd miss if you weren't suspicious.
the multimeter does the actual work
Eyes find the obvious; the meter finds the rest. With the board unpowered I went to continuity mode and checked for the unforgivable fault first: a dead short across the power rail. If the input rail is shorted to ground, powering it up just feeds a fire. There was no short, which was a relief and also meant the board probably wasn't catastrophically dead.
Then I powered it carefully through a bench supply with the current limit turned right down, so that if something was shorted it would simply refuse to draw rather than cook. It drew a tiny trickle and stalled. That pattern (draws a little, doesn't boot) often points at a power problem upstream of the processor rather than a dead processor, which is good news because power components are big, cheap and easy to get at compared to a BGA chip you can't touch at home.
Probing voltages along the input, I found the 5V rail present at the jack but not making it past that suspicious capacitor cleanly. Measuring across the cap, the reading wobbled. Combined with the slight doming, that was enough for a working theory: the cap had degraded, and possibly its solder joint had cracked with two years of thermal cycling in a warm cupboard.
the repair
Reflowing came first because it's free and reversible. I touched the iron to each leg of the suspect capacitor and a couple of nearby joints that looked dull rather than shiny, added a whisker of fresh solder, and let surface tension pull them into clean fillets. A cracked joint will often look fine to the eye and only show itself as a grey, slightly grainy surface rather than a bright cone. Reflowing alone sometimes fixes the whole thing.
It didn't, quite. The board now powered up and showed a light, but it was unstable, rebooting under load. So the capacitor itself was tired, not just its joints. I desoldered it (braid to wick the old solder, then gentle heat on both legs together with a bit of rocking) and fitted a replacement of the same capacitance and voltage rating from the parts drawer. The rules here are simple and not negotiable: match or exceed the voltage rating, match the capacitance, and get the polarity right on an electrolytic or it'll vent. The negative stripe goes to the marked side of the footprint. Backwards is a small bang and a worse smell.
With the new cap in, the input rail measured clean and steady, and the board booted and stayed booted under load. It's been running for two days since without a wobble.
what I'd tell myself
The repair was easy. The diagnosis was the whole job, and the order matters: rule out the supply and cable first, then look and smell, then meter for shorts, then power up current-limited, then probe the rails to find where the voltage stops behaving. Each step is cheap and narrows the search before you ever heat the iron.
A few things that earn their place on the bench. A cheap temperature-controlled iron beats an expensive fixed one, because for through-hole work you mostly want it hot enough and stable. Solder braid is worth more than you'd think for getting old joints clean. And a bench supply with an adjustable current limit turns "power it up and hope" into a controlled experiment, which is the difference between diagnosing a fault and creating a new one.
Was it worth it against eight pounds for a new board? Not on the maths. But I now know exactly why this one died, I've got the muscle memory for next time, and there's one fewer dead board in landfill. That's a good enough trade for a quiet Thursday evening.