Action cameras are specified in a way that hides their real engineering constraints. The headline numbers are resolution, frame rate and battery life, and all three are measured under conditions that the product will rarely see: room temperature, a fresh card, a battery at nominal voltage, and a recording that stops before the body heats up.

The failures that generate returns are different. A camera that stops recording after twelve minutes in the sun, a card that corrupts its last file when the battery dies, a body that becomes too hot to hold. All three trace back to storage and power decisions made at the schematic stage.

The workload is sustained, not peak

A 4K60 HEVC stream at 100 to 150 Mbps writes 12 to 18 MB per second, continuously, for as long as the recording runs. That is unremarkable for a modern card on paper. What matters is that the rate never drops, because a single stall produces a dropped frame or a broken file.

Two things push the real requirement above the average. Bitrate is variable: high-detail scenes — foliage, water, confetti — push instantaneous demand well above the nominal figure. And the camera does not only write. It reads back thumbnails, updates the file system, and occasionally writes metadata and a second low-resolution proxy.

So the number to design against is not "supports 4K", it is "sustains the peak bitrate for thirty minutes at 60 °C case temperature without a single stall". Very few card datasheets answer that question directly, which is why a write test on the actual card and the actual firmware is not optional.

Card choice is a system decision

A microSD card is a small computer with its own controller, its own garbage collection and its own idea of when to stall. Three consequences follow for a camera design.

Sustained write differs from burst write. Cards quote sequential write in ideal conditions. As the card fills, its controller runs out of pre-erased blocks and has to erase before writing, and the write rate drops. A card that benchmarks at 90 MB/s when empty can fall to 30 MB/s at 80 percent full. Cameras that record long takes need the worst case, not the fresh-card case.

Wear is real and uneven. A camera deletes and rewrites the same directory constantly. Unlike a phone, it does not spread writes across the card. The card's wear levelling is internal, but the file system's allocation pattern still concentrates on the same regions. A card used daily for two years in a dashcam-style workload will develop errors before its endurance rating suggests.

Power loss corrupts the directory, not just the file. If the camera loses power mid-write, the file allocation table can be left inconsistent. The next boot may see a card that needs repair, or worse, an empty card. This is a firmware and file-system problem more than a card problem, and it is solved by writing to pre-allocated files and updating the directory at controlled points rather than by choosing a better card.

On-board storage as a buffer

Many action cameras add eMMC or UFS on the main board, either as a second recording target or as a buffer. The buffer approach is underrated: write to on-board storage continuously, flush to the card in the background at a rate the card can sustain, and use the buffer to absorb the stalls.

The cost is capacity and BOM. The benefit is that the card's worst-case behaviour stops being on the critical path. For a camera that must not drop frames, that trade is usually worth making, and it also gives the firmware a place to keep the file system consistent if the card is removed mid-record.

Power: the sag matters more than the capacity

A 4K camera's current draw is spiky. The encoder, the image sensor, the card write and the WiFi radio do not peak together by design, but they overlap in practice, and the peaks land on a battery whose internal resistance rises as it discharges and as it cools.

The failure mode is not "battery empty". It is the camera shutting down at 15 percent charge because a write burst pulled the rail below the brownout threshold. Users experience that as a bad battery, not as a power design problem.

Three mitigations, in order of cost. Bulk capacitance close to the load, sized for the worst-case burst duration rather than the average. A brownout detector that gives the firmware a warning window, so it can close the file before the rail collapses. And a current limit or a sequencing policy that prevents the radio and the encoder from being enabled at the same instant.

The second one is the one that protects the recording. Without a warning signal, the firmware has no way to distinguish a deliberate power-off from a battery sag, and the file system takes the hit.

Thermal: the specification nobody writes down

A sealed camera body with no fan has to dissipate the encoder's heat through its surface. At 4K60 the SoC may run at two to three watts, and the body is a few hundred grams of aluminium and plastic. In still air at 35 °C ambient, the case temperature will rise until the SoC throttles.

That throttling is not a bug; it is the thermal design working. What matters is whether it is documented and whether the storage survives it. A card rated for 85 °C is fine. A card rated for 70 °C, sitting next to the SoC, may not be.

The design decisions that help: keep the card slot away from the SoC and the battery, use the body as a heat spreader rather than a sealed box, and specify the throttling curve so the frame rate drops predictably instead of the recording stopping.

A validation set that predicts returns

Sustained write at temperature. Thirty minutes at the highest bitrate, in a chamber at the specified maximum ambient, with the card at 80 percent full. Watch for dropped frames and for the write rate falling below the bitrate.

Power-cut during write. Hundreds of cycles, cutting power at random points, then checking that the card mounts and the previous files are intact. This is the test that prevents the "my card is corrupted" returns.

Battery sag under peak load. Discharge the battery while running the heaviest workload, and log the rail voltage. If the camera shuts down above 10 percent charge, the capacitance or the sequencing is wrong.

Drop and vibration. Not for the electronics, but for the card retention and the connector. A card that works loose during a bike mount is a support ticket.

If you are designing a camera or a similar battery-powered recorder, we can review the storage, power and thermal architecture against your bitrate and runtime targets. We supply storage and power components for these designs and have supported products through this kind of validation, so the conversation tends to be about the workload rather than the datasheet.