Thickness is the variable that undoes most thermal designs. Conductivity gets the attention in the data sheet; the bond line you actually achieve in assembly is usually what determines the result.
The only equation you need
What bond line will you really get?
| Assembly | Nominal gap | Real bond line once tolerances stack |
|---|---|---|
| GPU lid to cold plate | 0.2 mm | 0.5–1.0 mm |
| IGBT baseplate to heatsink | 0.1 mm | 0.3–0.8 mm (bow dominates) |
| Prismatic cell to cold plate | 0.5 mm | 0.8–1.5 mm (module stack-up) |
| LED board to chassis | 0.3 mm | 0.3–0.5 mm (if surfaces are controlled) |
Contributors: component and heatsink flatness, machining tolerance, board warpage, mounting torque distribution, and under load, thermal bow. None of these appear on a TIM data sheet, and all of them end up in your bond line.
What that costs you
| Bond line | 3 W/m·K pad | 6 W/m·K pad | Lenecold 180 W/m·K |
|---|---|---|---|
| 0.3 mm | 0.100 | 0.050 | 0.0017 |
| 0.5 mm | 0.167 | 0.083 | 0.0028 |
| 1.0 mm | 0.333 | 0.167 | 0.0056 |
| 1.5 mm | 0.500 | 0.250 | 0.0083 |
How to choose
Measure the assembled gap, not the drawn gap
Use feeler gauge, pressure film, or measure the compressed thickness of a sample in the real stack. Do this on production-representative parts, not on a prototype built by hand.
Add tolerance, not hope
Specify the pad thickness at the worst-case gap, not the nominal. A pad that is slightly too thick compresses; one that is too thin leaves a void.
Check the compression range
Most pads are designed to work in a compression window. Confirm your assembly lands inside it — ask for the compression-deflection curve.
Verify at end of life
Compression set over time effectively thins the pad. Our material holds compressibility drift to within 10% after 1,500 h at 150 °C.