Heat sink, heat spreader, cold plate: the three words get used interchangeably in casual conversation, and the confusion costs real money - because each does a different job, and a stack that uses one where it needs another ends up overbuilt in the wrong place. Here is what each layer does and how they combine.
Three layers, three jobs
| Layer | Job | Metric that matters | Typical material |
|---|---|---|---|
| Thermal interface (pad) | Fill the microscopic gap between two solid surfaces | Impedance at the real pressure (°C·cm²/W) | Elastomer pads, grease, phase-change films |
| Heat spreader | Take a concentrated hot spot and distribute it over a larger area | Conductivity and weight; lateral spreading | Copper, aluminium, graphene plate |
| Heat sink / cold plate | Move heat into air or coolant - the final exit | Fin area, airflow or coolant flow | Extruded aluminium, liquid cold plates |
The ordering is fixed by physics: the die produces heat at a point, the interface crosses the joint, the spreader flattens the flux, the sink exhausts it. Each layer is only as good as what feeds it.
Why the spreader exists
A heatsink fin cannot help the part of it that never gets warm. Put a 700 W die against a cold plate directly and the coolant channels under the die absorb a violent local flux while the plate edges stay cool - the junction temperature is set by the hot spot, not the average. A high-conductivity spreader between them (our graphene plate: 1200 W/m·K through-thickness, 1.5-4 g/cm³) carries flux laterally so the sink is loaded uniformly. On the same footprint, the hot-spot temperature drop shrinks even though the sink has not changed.
Why the interface exists at all
Even polished surfaces touch at a few percent of their nominal area - the rest is air, and air is an insulator at 0.026 W/m·K. The interface material replaces the air with something conductive while remaining compliant enough to close across tolerance, warpage and thermal movement. It is usually the thinnest and cheapest layer in the stack, and frequently the most misunderstood: conductivity alone does not specify it - impedance at the real clamp pressure does.
Worked comparison: 700 W accelerator
| Stack | Interface term | Spreading term | Relative junction rise |
|---|---|---|---|
| Die - thin pad - cold plate | Low (thin bond line) | High - hot spot under die | Baseline |
| Die - 180 W/m·K pad - cold plate | Low and stable (no pump-out) | High - hot spot remains | Moderately better |
| Die - 180 W/m·K pad - 1200 W/m·K plate - cold plate | Low and stable | Low - flux spread uniformly | Best; also evens coolant load |
The numbers behind each row depend on die size, pressure and plate thickness - the point is structural: the pad fixes the joint, the plate fixes the distribution, the cold plate fixes the exit. Solving a hot spot with a better cold plate usually means paying for more coolant flow to fix a problem one thin layer upstream already solved.
Choosing what your stack actually needs
- Small die, modest power, good clamp: a quality pad alone is usually enough.
- Multi-die module, local hot spot, or weight-sensitive platform: pad + graphene spreader - that is the stack's whole reason to exist.
- Any sealed, long-life system: solid interfaces throughout (no grease), because the aged joint is the specification.
Send us the stack - power, footprint, gap, pressure - and we will tell you honestly whether the plate earns its place in your design. Sometimes the answer is no, and that answer is part of the service.