💨 The Smoke Test

The smoke test is the first time the board pushes back, and the name is a grim joke from the days before current limiting, when “power it and see if it smokes” was literally the test. The modern version never lets the smoke out. It is a tight loop of two halves: a power-on half that catches the fault the instant it draws too much, and a localization half that walks the heat down to a single part. The ordering rule of this whole module shows up here as the rule of the loop: never raise the limit to make a fault go away, lower the voltage and chase the heat instead.

By the end, you can

1. Set a bench supply's current limit from a board's expected idle and run a current-limited power-on
2. Decide what to do the instant a board goes over-limit, and explain why raising the limit is the wrong move
3. Localize a low-impedance fault using thermal/IR imaging, freeze spray, and DMM rail readings
4. Bound a fault by isolating a section (lift a pin / cut a rail) and name the common root causes

Intuition first

Picture a city’s water main with a pressure-relief valve on it. Open the main and water should trickle out to the taps. If a pipe has burst underground, the flow does not trickle, it gushes, and the relief valve clamps the pressure down so the burst pipe floods a basement instead of bursting the whole street. You stand at the valve and watch the pressure gauge sag: you cannot see the leak, but you know it is there, and you know roughly how big it is.

A current-limited supply is that relief valve for electricity. Voltage is pressure; current is flow. A healthy board sips a known, small flow at idle. A short is a burst pipe: it wants to gush. Set the limit just above the healthy idle and the supply refuses to gush. It trades voltage for a fixed maximum current, the output droops, and you are left standing at the gauge watching a fault you cannot yet see, with all the time in the world to find it. Nothing has exploded. The evidence is intact. The only thing that has happened is that the board told you, loudly and safely, that something is wrong.

The localization half is then a heat hunt. Every watt the fault swallows turns into heat in one small place, because that place is where the resistance is. Find the warmest square millimeter on the board and you have found, or nearly found, the fault.

Setting the limit and ramping

The limit is the whole game, so set it deliberately. You need two numbers: the rail voltage the board expects, and its expected idle current. The voltage comes from the schematic; the idle current comes from the datasheets (sum the quiescent draws of the regulators, the MCU in its reset or sleep state, and any always-on parts) or from a known-good sibling board if you have one.

Set the current limit just above that expected idle, with enough margin that a healthy board does not trip it but not so much that a fault can do real damage. For a board that should idle around 70 mA, a limit of 90 to 100 mA is sensible: a good board lives comfortably below it, and a short slams into it long before any trace overheats.

Then ramp. Start the voltage low and bring it up while watching the current the whole way. Two things can happen:

• Healthy: the current rises briefly as the decoupling capacitors charge, then settles to about the expected idle as the voltage reaches its target. The supply stays in constant-voltage mode, delivering the set voltage at a current well under the limit.
• Faulted: the current climbs and pins at the limit while the voltage stops rising and sags. The supply has flipped into constant-current mode. This is the over-limit branch, and it is your stop sign.

The instant you are over-limit, the move is the same every time: kill power. Not raise the limit. Raising the limit feeds the fault more current and converts your safe, diagnosable event back into the destructive one you set up the leash to prevent. Kill it, breathe, and switch into localization.

The localization loop

You are over-limit, you have killed power, and you have a board that hides a low-impedance path you cannot see. Re-power it, current-limited at a deliberately low limit so the fault dissipates only a little power, just enough to warm the guilty part without cooking it. Now find the heat.

Thermal / IR camera. A thermal imager is the fastest tool here. The fault is the lowest-resistance path on the rail, so it carries the most current, so it dissipates the most power as heat, so it lights up first and brightest on the camera. Pan across the board and the hot spot is usually obvious within seconds, a single bright part or joint against a cool background.

Freeze spray. If you do not have a camera, or want to confirm what the camera found, freeze spray (a can of fast-evaporating coolant) is the cheap counterpart. Chill a suspect part and watch the supply current: if cooling the part changes the current, or if the part re-warms noticeably faster than its neighbors after a spritz, you have found the one doing the work. Spraying the camera’s suspect and seeing the current twitch is a clean confirmation.

A careful finger. With no tools at all, a light touch finds gross heat: brush a fingertip across the parts and the hot one announces itself. Do this only on a current-limited, low-voltage board, and only on parts that cannot be hot enough to burn. Never finger-test a mains-side circuit or a high-voltage rail.

The DMM. Once heat points at a region, the multimeter pins it down. Measure the collapsed rail to ground (the one whose voltage sagged) and confirm it reads a low impedance, the same single-digit-ohms signature the pre-power short check looks for. The DMM tells you which rail; the heat tells you which part on it.

The diagram’s right-hand branch is exactly this loop. The current-limited power-on (box 2) asks one question: over limit? The “no” path continues down the bring-up ladder to verify the rails. The “yes” path peels off into the kill-and-localize box: kill power, scan with thermal or freeze-spray, DMM the shorted rail, isolate a section, and arrive at a root cause. You only rejoin the main ladder once the fault is found and reworked.

Bounding and naming the fault

Heat usually lands you on the part, but a stubborn fault (a short under a connector, a bridge buried in a fine-pitch footprint) may not radiate cleanly. Then you bound it. The idea is to cut the suspect region off from the rest of the rail and re-measure: if the short vanishes, it was in the part you isolated; if it remains, it is elsewhere.

You bound by isolating a section: lift a single pin of a suspect IC off its pad, or cut a rail at a deliberate break point (a jumper, a zero-ohm link, or a scored trace you can re-bridge later). Lift the motor-driver’s supply pin and re-check the rail-to-ground resistance: if the dead short is gone, the driver (or its decoupling) is the culprit; if it persists, the fault is upstream and you isolate the next section. Each cut halves the search space, the same divide-and-conquer you would use to bisect a bad commit.

Once bounded, name the root cause. The usual suspects on a freshly assembled board are a short list:

• A solder bridge between two adjacent pins or a pin and a ground pour, dragged across by too much paste or a reflow that wicked wrong.
• A reversed electrolytic capacitor, soldered in backwards, whose oxide dielectric breaks down the instant the rail comes up and presents a near short.
• A shorted decoupling capacitor, a ceramic cracked by board flex or a manufacturing defect, sitting directly across the rail it is meant to clean.
• A misoriented part, an IC or diode rotated 180 degrees, tying the rail to ground through an internal path that was never meant to conduct.

Then rework: reflow the bridge, replace the reversed or cracked cap, reorient the part. Re-run the pre-power short check to confirm the rail reads high again, and only then return to the current-limited power-on and resume the ladder.

See it: the over-limit branch

The bring-up flow above is one figure, but its shape is the whole lesson, so read it as a decision tree rather than a checklist. Trace the spine from box 1 down: inspect, then power-on current-limited. At box 2 the diamond asks over limit? and everything in this lesson hangs off the “yes” arrow.

Follow that arrow to the kill-and-localize box on the right and read its lines in order: kill power, scan with thermal or freeze-spray, DMM the shorted rail, isolate a section, arrive at the root cause. That is the localization loop drawn as a single block. Notice what the diagram does not let you do: there is no arrow that loops back and raises the limit. The only way off the “yes” branch is through the fault and back to the start of the ramp. When you are standing at a pinned supply, picture this branch and walk its lines one at a time.

?During a current-limited ramp, the supply pins at its 90 mA limit and the output voltage sags from 12 V to 1.6 V and stays there. What is the correct first action?

• A Raise the current limit until the voltage recovers to 12 V
• B Kill power, then re-power at a low limit and localize the heat
• C Leave it ramped and load firmware to see if the board still boots
• D Lower the voltage setpoint and call the board healthy

?Thermal imaging points at one warm IC, but you want to prove it is the fault and bound the short. Which step bounds it?

• A Touch every part on the board to compare temperatures
• B Lift the IC's supply pin and re-measure the rail-to-ground resistance
• C Increase the supply voltage to make the hot part hotter
• D Replace every decoupling capacitor on the board at once

Lab: bring a board up current-limited and log it

Take any board you can power from a bench supply and bring it up the disciplined way. Before switching on, write down the expected idle current per rail from the datasheets, and a current-limit value just above it. Power on with the limit set, and ramp the voltage while watching current. For each rail, log three numbers against your budget: the set voltage, the measured voltage at the rail, and the measured current, then compare each to the expected idle. A healthy board fills the table with currents below the limit and voltages at target. If any rail pins the supply and sags, stop, note it, and run the localization loop: re-power at a low limit, find the heat with thermal or freeze spray, DMM the collapsed rail, and isolate the section. The deliverable is a per-rail table, voltage and current versus budget, with a one-line verdict per rail.

↘ Why a reversed electrolytic becomes a near-short the instant the rail rises

The reason a reversed electrolytic capacitor shows up as a low-impedance fault, rather than just a bad capacitor, is in how its dielectric is built. An electrolytic capacitor is a polarized device: its anode is a metal foil (aluminium, tantalum, or niobium) on which a wafer-thin insulating oxide layer has been grown electrochemically by anodization, and that oxide is the dielectric. A liquid, gel, or solid electrolyte makes contact with the oxide and serves as the cathode. The oxide is extraordinarily thin, on the order of nanometers per volt of rating, which is exactly why these parts pack so much capacitance into so little volume. But that thinness is only safe while the correct polarity holds the oxide in place.

Operate the part with the anode more positive than the cathode and the oxide stays intact. Reverse the polarity, or exceed the rated voltage by as little as a volt or so, and the oxide breaks down. The formerly insulating layer starts conducting, leakage current heats the electrolyte, and the part can pressurize and vent or, in the worst case, explode. So a backwards electrolytic does not present a clean open or a normal capacitance: as the rail rises and the reverse voltage breaks the oxide down, it presents a collapsing impedance, which on your supply reads as a current that pins at the limit while the voltage sags. This is also why polarity is a pre-power inspection item in the last lesson: once the rail is up on a reversed electrolytic, the breakdown is already under way.

The same physics explains why the short check works at all. A healthy decoupling network looks like an open at DC once its capacitors finish their brief initial charge, because an ideal capacitor passes no steady current. Resistance from rail to ground should therefore be high. A reversed or cracked capacitor, a solder bridge, or a part rotated so an internal junction ties the rail to ground all defeat that, presenting a DC path of a few ohms. Ohm’s law then sets the stakes: a rail meant to sit at $V$ volts that reads $R$ ohms to ground would, at full power, try to pass

$$I = \frac{V}{R}$$

through the defect, and dissipate

$$P = I \cdot V = \frac{V^2}{R}$$

right at the fault. A 3.3 V rail shorted through 5 Ω would attempt $I = 3.3 / 5 = 0.66 \ \text{A}$ and dump $P = 3.3^2 / 5 \approx 2.2 \ \text{W}$ into one tiny spot, which is why the unleashed version smokes and the current-limited version merely warms. The thermal camera reads that $P$ as a bright pixel, freeze spray perturbs it, and the DMM measures the $R$ behind it.

Grounded in Wikipedia: “Electrolytic capacitor”, “Printed circuit board” (CC BY-SA).

Key takeaways

• ▸The smoke test never lets the smoke out: a current-limited supply turns a short into a safe, watchable event.
• ▸Set the limit just above expected idle, ramp the voltage, and watch current; pinned current with a sagging voltage is the over-limit signature.
• ▸Over-limit means kill power, never raise the limit. Raising it feeds the fault and destroys the evidence.
• ▸Localize by heat: thermal/IR camera, freeze spray, or a careful finger on a low-voltage board, then DMM the collapsed rail.
• ▸Bound the fault by isolating a section (lift a pin / cut a rail), name the root cause (bridge, reversed electrolytic, shorted cap, misoriented part), rework, and re-check.
• ▸Inrush and long-ground-lead ringing are false alarms, not faults. Settle the first, shorten the ground spring for the second.

Practice 1 warm-up

A board’s datasheets sum to an expected idle of about 45 mA. You want a current limit that lets a healthy board run but slams into a fault early. What limit would you dial in, and what would you expect the supply to do as you ramp the voltage on a good board?

↘ Show worked solution

Set the limit a little above idle, around 60 mA (roughly 1.3 times the 45 mA idle gives headroom without giving a fault much room to do damage). On a good board, as you ramp the voltage you would see a brief current bump while the decoupling capacitors charge, then the current settles to about 45 mA as the voltage reaches its target. The supply stays in constant-voltage mode, delivering the set voltage at a current comfortably below the 60 mA limit. If instead the current climbs and pins at 60 mA while the voltage sags, the board is faulted and you switch to localization.

Practice 2 core

On the over-limit finger-driver board you re-power at a low current limit. A thermal camera shows one decoupling capacitor near the motor driver glowing hot. Describe, in order, how you would confirm it is the fault, identify the shorted rail, and bound the problem to that one part.

↘ Show worked solution

First, confirm with freeze spray: chill the hot capacitor and watch the supply current. If the current changes when you cool it (or the part re-warms much faster than its neighbors after a spritz), the camera’s suspect is doing the work. Next, identify the rail with the DMM: measure the rail whose voltage collapsed during the ramp to ground and confirm it reads a low impedance (single-digit ohms), the short-check signature. That tells you the motor rail is the shorted one. Finally, bound it by isolating the section: lift the capacitor (or the motor-driver supply pin it sits on) off the rail and re-measure rail-to-ground resistance. If the short clears and the rail reads high, the fault was that capacitor (most likely a cracked ceramic shorting the rail); if the short persists, the cap is innocent and the fault is elsewhere, so you isolate the next section. Then rework: replace the cracked cap, re-run the pre-power short check, and resume the current-limited ramp.

Practice 3 stretch

Two engineers hit over-limit on identical boards. Engineer A raises the supply’s current limit “just to see if it pushes through,” and the voltage recovers to target. Engineer B kills power and localizes. Explain what physically happened on A’s board when the limit was raised, why the recovered voltage is a trap, and why B’s slower path is the correct one.

↘ Show worked solution

When Engineer A raised the limit, the supply was freed to deliver more current into the low-impedance fault. The voltage “recovered” only because the supply could now drive enough current to hold the rail up while feeding the short, so the rail looks healthy even though a few-ohm path is still dumping power into one spot. With $P = V^2/R$, a 5 V rail shorted through, say, 4 Ω now dissipates roughly $25/4 \approx 6$ W into a single part. That is the trap: the gauge looks fine while the board cooks itself, which is exactly the destructive “power it and see if it smokes” failure the current limit was meant to prevent. The evidence (and likely the part) is being destroyed as you watch. Engineer B’s path is slower because it refuses to push current into an unknown fault: kill power, re-power at a low limit so the fault only warms, find the heat, DMM the rail, isolate the section, and fix the root cause. B ends with a working board and a known cause; A ends, at best, with a board that “works” but hides a damaged part, and at worst with a fresh crater where the fault used to be.