Your power board passes every test on the bench, sitting in open air with a desk fan nearby. Then it goes into its sealed enclosure, runs at full load, and the MOSFET hits 120 °C and starts derating. The silicon was never the problem — the thermal path was, and a heatsink is one of the most under-specified, badly sourced parts in a power product. A cheap one quietly costs you efficiency, lifespan, and field returns.
Heatsink manufacturing: extrusion vs skived vs forged
The process determines what fin geometry you can get and what it costs.
- Extruded aluminum — the default. Cheap, widely available, fine for moderate power. The limit is fin aspect ratio: extrusion can't make very tall, thin fins, which caps surface area in a given footprint. Good to roughly 20-30 W in natural convection depending on size.
- Skived — fins are shaved from a solid block, allowing very tall, dense, thin fins from one piece of metal (no joint resistance). Higher surface area per volume, better for tight, high-power spaces. Costs more.
- Forged — pin-fin geometry pressed into shape, good for omnidirectional airflow and impingement cooling, where a fan blows straight down onto the sink.
Picking the process before you know your airflow direction is backwards. Decide convection strategy first, then geometry, then the process that can make it.
Natural convection vs forced air
This is the first fork, and it changes everything downstream:
- Natural convection — no fan, fins must be vertical for the chimney effect, spacing wider (6-12 mm) so air can rise between them. Quiet, reliable, no moving parts to fail. The penalty: you need a lot more surface area, so the sink is bigger.
- Forced air — a fan lets you pack fins tighter (2-4 mm spacing) and shrink the sink dramatically. The cost is acoustic noise and a fan as a wear item with a finite MTBF.
Rule of thumb: if you can hit your junction temperature with natural convection inside the real enclosure, do it — a fan is a part that fails. If the math doesn't close, size the fan for worst-case ambient (a 40 °C warehouse, not your 22 °C office).
Thermal interface materials matter more than people think
The gap between the device and the heatsink is filled by a TIM, and a bad one wastes the whole design. Options range from thermal grease (best performance, messy, pumps out over thermal cycles) to thermal pads (easy assembly, higher thermal resistance) to gap fillers.
The number to watch is thermal resistance (°C/W) across the interface. A poorly applied pad can add 2-5 °C/W — enough to push a marginal design over the edge. Specify the exact TIM part number and thickness; "thermal pad" on a BOM invites substitution.
Sourcing custom heatsinks without getting burned
Stock heatsinks are easy. Custom extrusions are where sourcing goes wrong:
- Tooling cost: a custom extrusion die runs roughly $500-1,500, amortized over your volume. Factor it into per-unit cost honestly at low volumes.
- Alloy substitution: 6061 and 6063 are common; a factory may quote a cheaper, lower-conductivity alloy. Specify the alloy.
- Surface treatment: black anodizing improves radiative cooling (worth a few degrees in natural convection) and prevents corrosion. State it.
- Flatness: the mating surface flatness directly affects TIM performance. Call out a flatness tolerance.
A concrete case from a LoRa gateway with a power-hungry RF amplifier: the supplier's first article used a lower-grade alloy and an unspecified TIM pad, and the amplifier ran 14 °C hotter than the thermal model predicted. Pinning the alloy to 6063-T5, specifying the TIM, and adding black anodize brought it back into spec — and direct factory sourcing of the corrected part cut about 22% off the cost the team had been paying through a distributor.
This is hard to verify from a quote. If you don't have an engineer who can validate the alloy and measure thermal performance on first articles, an engineering-led agent like China Sourcing Agents will check the material and run a loaded thermal test during sampling — before you commit to a tool.
Thermal spec items to lock before tooling
- [ ] Convection strategy decided (natural vs forced) for worst-case ambient
- [ ] Target junction/case temperature with margin (aim ≥15 °C below derating)
- [ ] Alloy specified (e.g. 6063-T5), not just "aluminum"
- [ ] TIM part number, thickness, and thermal resistance called out
- [ ] Surface treatment (anodize) and mating-face flatness tolerance
- [ ] Loaded thermal test on first articles inside the real enclosure
The bench number lies. The only thermal result that counts is measured at full load, in the actual box, at the highest ambient your product will ever see.
Top comments (0)