Why MicroSD Cards Slow Down Under Sustained Write (Thermal Throttling)

TL;DR

A microSD card that writes at high speed for a short burst is not the same product as one that must write continuously for hours. Under sustained write load, the controller and NAND die heat up, the card’s internal firmware reduces write throughput to stay inside safe operating limits, and the effective speed drops to a lower plateau. That plateau — not the peak number on the datasheet — is what determines whether a dashcam drops frames, a logger loses samples, or a device’s buffer overflows.

For B2B buyers the practical consequences are:

• Specify sustained write rate, not burst rate. Ask for a throughput-over-time curve, not a single benchmark figure.
• Thermal design is a host problem as much as a card problem. Cards sit in sealed enclosures with no airflow; the host’s thermal path matters more than any card-level heat spreader.
• Heat and endurance interact. Higher die temperature accelerates NAND wear and shifts data-retention behaviour, so throttling and lifetime are linked, not separate topics.
• Test in the real enclosure. A card that holds speed on an open bench can throttle inside a product housing.

This article explains the mechanism, how to measure it, what to specify, and where the limits of card-level mitigation lie. WINTEL’s confirmed capabilities are stated only where they appear in our company data; everything else below is general industry and engineering knowledge.

What actually generates heat in a microSD card

A microSD card is a small package containing a flash controller, one or more NAND die, and usually DRAM or SRAM for mapping tables. Heat comes from three places:

1. The controller. It runs firmware, manages wear levelling, garbage collection and error correction. During heavy writes it is the busiest and often the hottest element.
2. The NAND die. Programming flash cells involves driving charge through the cell structure; program and erase operations dissipate energy. Sustained programming keeps the die warm.
3. The host interface. At high bus speeds the interface itself contributes, though usually less than the controller and NAND.

The card has almost no thermal mass and very little surface area. In a phone, camera or dashcam it is enclosed by the host, often with a plastic or metal shell around it and no direct airflow. The thermal path out of the card is through the card’s contacts, the socket, and the surrounding PCB — which is why the host design dominates the outcome.

General engineering note: the SD Association’s physical layer specification defines the card’s operating conditions and bus modes; it does not mandate a thermal throttling curve. Throttling behaviour is implemented by each controller vendor’s firmware and varies by product. See the SD Association specifications index for the normative definitions.

How throttling works in practice

Most card firmware implements a thermal management loop:

• A temperature sensor (on-die or in the controller) reports temperature.
• Below a threshold, the card runs at full speed.
• As temperature rises, the firmware reduces write throughput — by inserting idle time, lowering the effective program rate, or deferring garbage collection.
• When the card cools, speed recovers.

The result on a throughput-over-time plot is characteristic: a fast start, a gradual or stepwise decline, then a stable lower plateau. That plateau is the number that matters for continuous-write products.

Two things follow from this:

Throttling is protective, not a fault. A card that throttles is doing what it was designed to do. A card that does not throttle under sustained load is either very efficient, very slow to begin with, or not adequately protecting itself.

Peak benchmarks mislead. Short benchmarks finish before the card reaches steady state. A 30-second test can report a rate the card cannot hold for ten minutes, let alone an hour.

A worked illustration of why the plateau matters

(Illustrative only — not WINTEL data) Suppose a card bursts at 90 MB/s for the first 20 seconds, then settles at 35 MB/s once warm. A device that must record continuously at 50 MB/s will buffer for the first 20 seconds and then overflow. The datasheet’s 90 MB/s is technically true and practically useless for that application. The specification that would have prevented the problem is “sustained write ≥ 50 MB/s after 10 minutes of continuous write in the target enclosure.”

Heat, endurance and data retention are one topic

It is tempting to treat thermal throttling as a performance issue and endurance as a separate reliability issue. In flash memory they are coupled.

• Higher temperature accelerates wear. Program/erase cycling at elevated die temperature stresses the oxide and increases the rate at which cells degrade. NAND vendors characterise endurance at specified temperatures for this reason.
• Retention depends on temperature history. Data retention is specified as a function of storage temperature; a card that ran hot in service and then sat unpowered is not in the same condition as one that ran cool. JEDEC’s standards documents define the endurance and retention terminology the industry uses — see the JEDEC standards index.
• Throttling reduces the thermal load. By slowing writes, the firmware keeps the die inside its qualified envelope. A card that never throttled and ran hot would wear faster, not slower.

For products that write continuously — dashcams, body cameras, telemetry loggers, industrial recorders — the practical design goal is not “avoid throttling” but “keep the sustained duty cycle inside what the card and host can dissipate.”

Measuring throttling properly

A useful test answers three questions: what is the peak rate, what is the steady-state rate, and how long does it take to get there.

Test design checklist:

• Write sequentially for long enough to reach steady state — minutes, not seconds.
• Log throughput over time, not just an average.
• Test in the actual host enclosure, or a thermally equivalent rig.
• Test at the ambient temperature the product will see in the field, including the hot end of the range.
• Repeat with the file sizes and access pattern the product actually uses.
• Record recovery behaviour: how fast does speed return after the load stops?

What to look for in the results:

Pattern	Interpretation
Flat line from start	Card is not throttling, or is limited by something else (host, bus mode, file system)
Fast start, step down, stable plateau	Normal thermal throttling; plateau is your real number
Continuous decline with no plateau	Thermal design is inadequate for the duty cycle; expect reliability problems
Sawtooth	Firmware cycling between throttle states; often indicates marginal thermal margin

Host-side factors confound these measurements. Caching, file system flush behaviour and buffer sizes all shape the write pattern the card actually sees. Microsoft’s file system documentation (see Sources) is a reasonable starting point for understanding write-back and flush semantics on Windows hosts; Linux hosts behave differently again, and the kernel’s MMC subsystem documentation covers how the host negotiates modes and handles errors.

What card-level mitigation can and cannot do

Buyers often ask whether a heat sink, thermal pad or metallised label on the card solves the problem. The honest answer is that card-level mitigation has very little room to work:

• Surface area is tiny. A microSD card is roughly the size of a fingernail. There is not much to spread heat into.
• The card is usually enclosed. In most products the card sits inside a housing with the socket around it; a label on the outside of the card is not in the thermal path that matters.
• The contacts are the main path. Heat leaves through the card’s electrical contacts into the socket and PCB. Host-side thermal design — socket choice, PCB copper, airflow, duty cycle — therefore dominates.

Where card-level measures can help is at the margins: better thermal contact between card and socket, avoiding unnecessary re-writes, and choosing a card whose firmware throttles gracefully rather than abruptly.

Specifying for sustained-write applications

If your product writes continuously, the procurement conversation should be about the plateau, not the peak. A specification that works looks like this:

1. State the required sustained write rate for your worst-case duty cycle, in your enclosure, at your field ambient temperature.
2. Ask for a throughput-over-time curve, not a single benchmark number.
3. Ask how the card behaves at the hot end of the ambient range — not just at room temperature.
4. Agree the test method so that your acceptance test and the supplier’s test measure the same thing.
5. Define what happens on failure to meet the plateau — retest, re-sample, or reject.

For related reading on how cards are specified and where the numbers come from, see our guides on A1/A2 application performance classes and capacity drift across temperature. If your product logs continuously and may lose power, power-loss corruption and what protects data covers the failure mode that heat discussions often get confused with.

For dashcam-specific selection criteria, how to choose a TF card for a dashcam walks through the duty-cycle and endurance questions in more detail.

Where WINTEL fits

WINTEL (Shenzhen Wintel Digital Technology Limited) manufactures custom TF/microSD cards, USB flash drives, UDP modules and semi-finished boards for OEM/ODM buyers. Our confirmed company profile states a 5,000 m² facility, 200+ employees, monthly capacity of 500,000+ units and certifications including SGS, RoHS, CE and FCC.

We do not publish speed classes, endurance ratings, temperature ranges, MOQ or lead times as fixed figures here, because those are product- and project-specific and should be agreed against your application rather than assumed from a web page. If you need a sustained-write specification for a continuous-recording product, the right next step is a scoped conversation about your duty cycle, enclosure and field temperature, followed by sample testing against an agreed method.

• Browse TF/microSD cards and semi-finished boards.
• Estimate capacity and duty-cycle requirements with the TF card capacity estimator.
• Plan acceptance testing with the sample acceptance checklist.
• Send your sustained-write requirement through Request a quote.

FAQ

Does a microSD card really get hot enough to throttle? Yes. Continuous sequential writes at high bus speed keep the controller and NAND die busy, and a card inside a sealed device with no airflow is effectively thermally insulated. Most cards manage this internally by reducing write speed rather than failing outright, but the result is a sustained-write rate well below the burst rating on the label.

Is thermal throttling the same as a slow card? No. A slow card is slow from the first second. A throttling card starts fast and then falls to a lower plateau after seconds or minutes of load, then recovers when it cools. If you plot write throughput over time, throttling shows as a step down, not a flat low line.

Does a heat sink on a microSD card help? Marginally, and usually not in the way buyers expect. There is very little surface area to work with, and the card is normally enclosed by the host. Improving airflow or thermal contact in the host device, or reducing sustained write duty cycle, generally matters more than any sticker or pad on the card itself.

Will throttling damage my data? Throttling itself is a protective behaviour, not a fault. The real data risk in sustained-write products is power loss during a write, which is a separate failure mode from heat. Both should be designed for in dashcam, bodycam and logging applications.

How do I test for throttling before committing to a production order? Run a long sequential write — long enough that the card reaches steady state, not a 30-second benchmark — while logging throughput over time in the actual host or a thermally similar enclosure. Compare the plateau rate, not the peak, against what the product must sustain.

Sources

• SD Association — SD Specifications (Physical Layer, Speed Class definitions): https://www.sdcard.org/downloads/pls/ (accessed 2026-10-24)
• SD Association — Application Performance Class (A1/A2): https://www.sdcard.org/developers/sd-standard-overview/application-performance-class/ (accessed 2026-10-24)
• JEDEC — Standards and documents index (NAND reliability, endurance and retention terminology): https://www.jedec.org/standards-documents (accessed 2026-10-24)
• Linux kernel documentation — MMC/SD subsystem: https://docs.kernel.org/driver-api/mmc/index.html (accessed 2026-10-24)
• Microsoft Learn — File systems and write/flush semantics: https://learn.microsoft.com/en-us/windows/win32/fileio/file-systems (accessed 2026-10-24)