How Do You Use Thermal Imaging to Debug a PCB?

Detailed Explanation

A thermal (infrared) camera turns power dissipation into a picture. Every component and joint on a running board is dissipating some power as heat, normally an amount that matches its expected function and stays within its rating. A thermal camera makes that dissipation visible without probing anything: a regulator running hotter than its datasheet suggests, a resistor pulling more current than the design intends, or a joint with more resistance than a good connection should have, all show up as a localised hot spot relative to their surroundings.

That makes it a genuinely different debugging tool from an oscilloscope or logic analyser (see oscilloscope basics and logic analyser vs oscilloscope) rather than a competitor to them. A scope tells you what a signal is doing at one point over time; a thermal camera tells you where power is actually going across the whole board at once. It's a coarse, non-contact, whole-board instrument, best used to point a more precise investigation (multimeter, scope, or a re-flow of a specific joint) at the right location rather than as the final diagnostic step on its own.

Finding Hot Components During Bring-Up

During PCB bring-up, thermal imaging answers a simple question fast: is anything on this board dissipating power it shouldn't be? A component that's warmer than its function explains (an MCU running hot at idle, a logic-level component showing a clear thermal signature when it should barely dissipate anything, a passive component that's warm when it should be nearly invisible to the camera) is a strong, early signal of a fault: a wrong component value pulling excess current, a reversed polarised part, a solder bridge shorting two nets, or a supply rail sitting at the wrong voltage and overdriving something downstream.

The practical bring-up sequence is: power the board with a current-limited bench supply, let it settle for the first 10-30 seconds, then scan the board with the thermal camera before touching anything else. A component that's noticeably hot within seconds of power-on, before firmware has even loaded, is dissipating power in its quiescent state: that's a hardware fault, not a firmware one, and worth stopping to investigate before applying full power or loading application code. This is the same principle behind the "touch components lightly" step in a manual bring-up checklist; a thermal camera does the same check faster, without physical contact, and quantifies the temperature rather than relying on a fingertip's rough sense of "too hot."

For components with a known thermal budget from the design stage (see thermal design and heatsink selection), thermal imaging is also a direct way to validate the calculation against reality: measure the actual case or PCB-adjacent surface temperature under representative load, and compare it against the junction temperature the θJA-based calculation predicted. A significant mismatch usually means the real copper area, airflow, or enclosure ambient differs from what the calculation assumed.

Identifying Bad or High-Resistance Solder Joints

A high-resistance joint (mechanically connected but with a smaller-than-intended contact area, a cold joint with poor wetting, or a hairline crack from thermal or mechanical stress) dissipates power as I²R at that specific connection point. Under load, that shows up as a localised hot spot exactly at the joint, distinct from the normal thermal signature of the component or trace on either side of it.

This is a functional test, not a static inspection: with no current flowing, a marginal joint and a perfect joint read the same temperature, because neither is dissipating anything. Run the board at a representative load, ideally the actual worst-case current for the node in question, and give a marginal joint time to heat measurably above the surrounding board temperature; this can take anywhere from tens of seconds to a few minutes depending on the joint's thermal mass and how much current it's actually carrying. A joint carrying a high-current path (a power connector pin, a large MOSFET's drain connection, a high-current inductor pad) will show a resistive fault faster and more visibly than a joint on a low-current signal line, where the same resistance produces far less dissipated power and a much smaller thermal signature.

Because thermal imaging only catches defects that are carrying current at the moment you look, it complements rather than replaces structural inspection methods. AOI and X-ray inspection catch a defect (a void, a bridge, an open) whether or not the board is powered, but can't tell you whether a structurally marginal joint is actually causing an electrical problem under real operating conditions. Thermal imaging under load answers exactly that question, and is particularly useful for chasing intermittent field failures that AOI passed cleanly at assembly.

Spotting Thermal Runaway

Thermal runaway is a positive-feedback failure: rising temperature increases current (or leakage) through a device, which increases power dissipation, which raises temperature further, in a loop that accelerates rather than settles. It's a particular risk with parallel devices that don't share current evenly: paralleled MOSFETs, paralleled BJTs, or LED strings, where devices with a negative temperature coefficient (or simply slightly lower resistance/forward voltage from part-to-part variation) pull disproportionately more current as they warm, at the expense of their neighbours.

A single thermal snapshot can miss a runaway condition that hasn't fully developed yet. The signal to look for is divergence over time between devices that should be running identically: image the same set of parallel components repeatedly over a period of sustained load (most thermal cameras support recording a short video or a time-lapse sequence rather than only a single still), and watch whether one device's temperature is climbing away from its neighbours rather than settling to a stable, matched temperature alongside them. A device settling to a stable temperature above its neighbours but not still climbing may simply have slightly different characteristics within normal part tolerance; a device whose temperature keeps climbing relative to its neighbours after the others have stabilised is the signature worth stopping the test over.

This component-level runaway (a single device or a small group of paralleled devices on a board) is a distinct failure mode from cell-to-cell thermal runaway propagation in a multi-cell battery pack, which involves different mechanisms (electrochemical, not just resistive/junction heating) and different mitigations. See preventing thermal runaway propagation in a battery pack for that separate problem.

For production troubleshooting, the same divergence principle applies across units rather than across parallel devices on one board: build a thermal reference image of a known-good unit at a defined load point (a "golden board" signature), then compare failing or suspect units from a production batch against it. A unit that shows the same overall pattern but with one region running measurably hotter than the reference points directly at the fault location, often faster than working through a full electrical test sequence from scratch.

Camera Resolution and Thermal Sensitivity: What Actually Matters

Two specifications determine whether a thermal camera is actually useful for board-level work, and they aren't the same thing.

Spatial resolution (the detector's pixel count, e.g. 160×120, 320×240, 640×480) determines how much physical detail the camera can resolve at a given working distance: whether it can distinguish one pin of a fine-pitch IC from its neighbour, or whether an entire QFN package blurs into a single thermal blob. What matters in practice isn't the pixel count alone but the resulting spatial resolution at your actual working distance, sometimes described by a camera's distance-to-spot ratio: the same detector resolves far more detail on a large power inductor from 10 cm away than it does trying to distinguish the pins of a 0.4 mm-pitch QFN from 30 cm away. For dense boards with small SMD packages, a camera in the 320×240 class or higher, with a close-focus or macro lens option if available, resolves meaningfully more useful detail than a phone-clip-on accessory at typical working distances.

Thermal sensitivity, specified as NETD (Noise Equivalent Temperature Difference, in milliKelvin), determines the smallest temperature difference the camera can distinguish from its own noise floor. A lower NETD figure means better sensitivity. A camera with poor sensitivity (a higher mK figure) can still spot a component that's grossly overheating, because that's a large, easy-to-see temperature delta, but it will miss a subtle few-degree difference between a marginal joint and its surroundings, which is exactly the kind of fault most worth catching before it becomes a field failure. Instrument-grade thermal cameras aimed at electronics and predictive-maintenance work typically specify NETD in the tens of milliKelvin; consumer and phone-accessory-class devices are usually a good deal higher, which is the main practical reason they're better suited to gross fault-finding than to fine solder-joint or component-matching work.

Resolution and sensitivity both matter, and a high pixel count with poor sensitivity is not a substitute for genuine thermal sensitivity, nor is excellent sensitivity a substitute for enough spatial resolution to actually locate the fault on a densely populated board.

Emissivity: The Biggest Source of Measurement Error on PCBs

A thermal camera doesn't measure temperature directly. It measures infrared radiation and converts that to a temperature using an assumed emissivity value for the surface being imaged. Emissivity describes how efficiently a surface radiates compared to a perfect blackbody at the same temperature; a low-emissivity surface reflects a meaningful portion of the ambient infrared around it (including the camera itself, or a warm hand nearby) rather than emitting radiation proportional to its own true temperature, and the camera has no way to tell the difference unless it's told what emissivity to assume.

This matters enormously on a PCB, because the surfaces on a typical board span a wide emissivity range within centimetres of each other:

• Solder mask, conformal coating, and most component plastics: typically emissivity around 0.85-0.95, and read close to their true temperature directly with a default high-emissivity setting.
• Bare or plated metal (component leads, exposed thermal pads, connector shells, unmasked copper): typically emissivity in the range of roughly 0.1-0.2 for shiny/polished surfaces, and reads significantly cooler than its true temperature unless corrected for, because most of what the camera sees is reflected ambient IR rather than the metal's own emission.

The practical consequences: a shiny lead sitting right next to a solder-masked pad at the same physical temperature can show a large, entirely false temperature difference on the raw thermal image; and a genuinely hot exposed metal surface (a connector pin, a bare copper pour, an exposed thermal pad under a QFN) can be under-read badly enough to miss a real fault. Camera manufacturers document typical emissivity ranges for common electronics surfaces, and this is one of the few numeric parameters in thermal imaging genuinely worth checking against the camera manufacturer's published guidance, rather than assuming a single default value works across an entire board.

Two practical ways to work around low-emissivity metal surfaces: if the camera or its software supports multiple regions of interest, set a region-specific emissivity value, calibrated against a contact thermometer or thermocouple reading on a representative point; or apply a thin layer of a known high-emissivity material (a small dot of flat black paint, or specialised high-emissivity tape) to the specific surface of interest before testing. The second option is a common technique when a genuinely reliable absolute temperature reading, not just a relative comparison, is needed on a low-emissivity component.

For most board-level fault-finding, absolute accuracy matters less than relative comparison: whether one component or joint is meaningfully hotter than its identical neighbours, or hotter than a known-good reference board, is usually the actionable signal, and that comparison holds even when the absolute temperature reading on a low-emissivity surface is understated, as long as you're comparing like-for-like surfaces (metal lead against metal lead, not metal lead against solder mask).

Design Considerations

• Design in accessible thermal targets on high-power components. Where practical, avoid covering critical power components entirely in solder mask cutouts or under mechanical shielding that blocks a clear line of sight for a thermal camera during bring-up and troubleshooting. A component that's fully enclosed or obscured by a shield can only be imaged after partial disassembly, which is far slower during debug.
• Keep a known-good thermal reference image for production-volume boards. A thermal image of a working unit at a defined load point, taken once and kept on file, turns "does this failing unit look different from normal" into a fast visual comparison rather than a from-scratch investigation for every field return or line reject.
• Treat thermal imaging as a localisation tool, not a final verdict. A hot spot tells you where to look with a multimeter, oscilloscope, or physical inspection under magnification. It rarely tells you definitively why a joint or component is hot without that follow-up step.

Zeus Design's product development process includes thermal validation as part of hardware bring-up and production test strategy, catching marginal solder joints and thermal design issues before they reach the field.

Common Mistakes

• Imaging a board with no load applied and concluding solder joints are fine. A resistive joint with no current flowing dissipates nothing and reads identically to a good joint. Thermal imaging for joint problems requires the board under representative load, not a static power-on check.
• Comparing a shiny lead's reading directly against an adjacent solder-masked surface. The two materials have very different emissivity, so a raw temperature comparison between them is measuring an emissivity artefact as much as a real temperature difference. Compare like-for-like surfaces, or correct for emissivity first.
• Relying on a low-resolution or low-sensitivity camera for fine-pitch or subtle-fault work. A phone-clip-on thermal accessory is genuinely useful for finding a grossly overheating component, but its resolution and NETD are usually not enough to resolve individual pins on a fine-pitch package or catch a marginal, few-degree solder joint fault.
• Taking a single snapshot to rule out thermal runaway. A runaway condition that hasn't fully developed can look like a stable, minor temperature difference in one still image. Watching temperature trend over a sustained load period is what actually distinguishes normal part-to-part variation from a genuinely diverging runaway condition.
• Assuming the displayed temperature is accurate without checking the camera's emissivity setting against the surface being measured. The single biggest source of measurement error on a mixed-material PCB is an emissivity assumption that doesn't match the actual surface in view, not detector noise or calibration drift.