Every thermal pad datasheet quotes resistance at one pressure. Assembly drawings rarely say what pressure the mount actually delivers. Between those two facts sits most of the thermal problems we get asked about: joints that measure beautifully in the lab and run hot in the product, because nobody checked what the screws were doing.
What clamping pressure actually does
A pad does not conduct heat through its surface - it conducts through the real contact area, which at low pressure is a small fraction of the nominal footprint. Pressure does one job: it converts nominal area into real contact area, squeezing out the air gaps left by surface roughness and flatness error. Air is a 0.026 W/m·K material; every unclosed pocket is a thermal open circuit.
This is why measured resistance falls steeply at first and then flattens. The steep part is contact being established. The flat part is the material's own bulk resistance, which no amount of torque will remove.
Where the knee is - measured, not estimated
From our three-thickness dataset (26 × 26 mm samples, 80 °C, 5-50 psi, ten points per thickness), roughly 90% of the total resistance reduction is already banked by 20 psi. Going from 5 to 20 psi delivers about 91% of what 50 psi achieves on the 0.20 mm pad; the remaining 30 psi buys single-digit percentages.
| Pressure (psi) | R 0.15 mm | R 0.20 mm | R 0.30 mm | Note |
|---|---|---|---|---|
| 5 | 0.168 | 0.255 | 0.158 | Contact barely established |
| 10 | 0.088 | 0.109 | 0.079 | Steep region |
| 20 | 0.054 | 0.058 | 0.054 | Knee - ~90% of gain banked |
| 30 | 0.045 | 0.045 | 0.049 | Diminishing returns |
| 50 | 0.038 | 0.039 | 0.042 | Bulk-dominated floor |
Resistance in °C·cm²/W. Practical reading: design for 15-30 psi. Below 10 psi you are still on the cliff; above 35 psi you are mainly pre-loading the hardware, not improving the interface.
What your screws actually deliver
This is the step most drawings skip. Bolt preload converts to pressure through the contact area, and the numbers get large fast:
F = T / (K · d), where T is torque, K the nut factor (about 0.2 for dry steel), d the nominal diameter.
| Fastener (dry, K=0.2) | Torque | Preload F | Pressure over 20x20 mm |
|---|---|---|---|
| M2.5 | 0.35 N·m | ~700 N | ~250 psi |
| M3 | 0.50 N·m | ~830 N | ~300 psi |
| M4 | 1.20 N·m | ~1.5 kN | ~540 psi |
In other words: a single lubricated M3 tightened by hand can put more than ten times the pad's optimum pressure into a 20 × 20 mm foot. Three consequences follow, and all three show up in real failures:
- Over-compression damages the pad and the board. Above roughly 40-50% compression, filled elastomers take permanent set; the pad thins, the joint opens when the load is released, and the interface that was fine at assembly is starved at end of life.
- Point loading bows the hardware. A screw boss directly over a thin PCB or a ceramic substrate bends it. The pad under the screw is crushed while the rest of the footprint is barely touched - the average pressure on the drawing is a fiction.
- Torque is a poor proxy. Friction, thread condition and washer stack change preload by a factor of two or more. If pressure matters to the thermal budget, specify it as pressure, not as torque.
How to control it instead
- Hard stops and shoulder screws. Let a machined step set the compressed thickness rather than the torque. This is the single most effective fix, and it decouples thermal performance from operator technique.
- Compression limiters. Specify pad thickness so that the worst-case gap plus a defined compression (say 20-30%) lands on the stop.
- Spring elements. Belleville washers or spring clips flatten the pressure-torque curve; if you must use clips, expect 5-15 psi and choose thickness from the low-pressure end of the curve, not the datasheet headline.
- Verify once with pressure film. A single pressure-indicating film map at first article tells you more than a page of torque calculations, and it shows the pressure distribution your FEA assumed but never checked.
Low-pressure mounts: the counter-intuitive case
If the mount cannot deliver more than about 10 psi - spring clips, plastic snap-fits, hand-tightened brackets - do not automatically reach for the thinnest pad. In our measurements the 0.20 mm pad had the highest resistance at 5 psi (0.255) while the 0.30 mm pad was lowest (0.158), and the curves cross near 26.5 psi. A thin, stiff pad needs pressure to conform; a thicker one grabs the surface finish early. Thickness selection has to be read off the curve at your pressure, not at the datasheet's.
Pressure is not free - the other side of the ledger
Clamping load also means stress: on solder joints under a die, on a ceramic substrate, on the optical surface of a crystal, on the flatness of a cold plate. Above the knee, additional pressure buys very little thermally while adding mechanical risk. The engineering target is not "as much as possible" - it is enough, uniformly distributed, and repeatable.