Power & Thermal

Conduction Cooling

Q: How do I calculate the cold-plate temperature needed for a conduction-cooled GaN amplifier?

Work backward from the maximum allowed junction temperature through the conduction stack. For a GaN device with Tj,max of 225 C dissipating 40 W, with a junction-to-case resistance of 0.6 C/W, a case-to-baseplate resistance of 0.2 C/W through the solder and interface material, and a baseplate-to-cold-plate resistance of 0.15 C/W, the total RthJ-coldplate is about 0.95 C/W. The temperature rise is 40 W times 0.95 C/W, or 38 C. To keep Tj at a reliability-safe 175 C, the cold plate must be held at 175 minus 38, which is 137 C maximum. In practice designers target a cold-plate temperature of 70 to 85 C to leave reliability margin, since GaN median time to failure roughly halves for every 10 C rise in channel temperature.

Q: Why does conduction cooling use a thermal interface material instead of bolting metal directly to metal?

Two nominally flat machined surfaces touch only at their high spots, so real metal-to-metal contact covers only 1 to 3 percent of the apparent area. The remaining gaps are filled with air, which has a thermal conductivity near 0.025 W/m-K and adds a large contact resistance. A thermal interface material such as a gap pad, grease, or a soft indium foil displaces that air and conforms to the surface roughness, dropping the interface resistance from roughly 1 C-cm2/W for a dry joint to 0.1 to 0.3 C-cm2/W. The trade-off is that most interface materials are far less conductive than the base metals, so the layer must be kept thin (25 to 100 micrometers) and the bolts torqued to a specified pressure of 100 to 300 kPa.

Q: When is conduction cooling preferred over forced-air or liquid cooling for RF systems?

Conduction cooling is chosen when reliability, sealing, or environment rules out moving parts. Sealed airborne and space avionics use it because there is no fan to fail and no coolant to leak, and the chassis can be hermetically sealed against dust, humidity, and altitude. Conduction-cooled chassis standards such as VPX and VME64x route heat from the card edge into wedge locks and then into the cold wall. The limit is heat flux: conduction cooling comfortably handles board-level dissipation up to roughly 50 to 100 W per slot, but kilowatt transmitters and dense phased-array tiles usually need a pumped liquid loop or microchannel cold plate to keep junction temperatures in range.

/kuhn-DUK-shuhn KOO-ling/

A thermal management approach in which heat generated by an RF power device travels through an unbroken chain of solid materials, the die, its baseplate, a thermal interface layer, and the mounting hardware, into a cold plate or chassis wall where it is finally rejected. There is no fan, blower, or circulating coolant in the device-level path; heat flow obeys Fourier's law and is governed entirely by the cross-sectional area, thickness, and conductivity of each layer in series. Conduction cooling is the standard for sealed avionics, space hardware, and ruggedized transceivers because it has no moving parts to fail, but its low heat-flux ceiling means high-power transmitters must keep junction temperature in check by minimizing every interface resistance between the device and the heat sink.

Category: Power & Thermal

Board power limit: 50 to 100 W/slot

Cold-plate target: 70 to 85 °C

How Heat Travels Through a Conduction Path

In a conduction-cooled assembly, every watt of dissipated power must cross a series of solid layers before it reaches a surface cool enough to reject it. Each layer behaves like a thermal resistor, and because the layers are stacked in series their resistances add directly. The total resistance from the transistor channel to the cold plate sets the temperature rise for a given dissipation, so the design problem reduces to making each resistor as small as practical: short conduction paths, large contact areas, high-conductivity materials, and the thinnest possible interface layers. A GaN power amplifier die only a few square millimeters in area can dissipate 30 to 50 W, producing local heat fluxes above 1 kW/cm² at the channel, which is why the first millimeter of the path dominates the thermal budget.

The dominant obstacle in most real designs is not the bulk metal but the joints between parts. Two machined surfaces meet only at microscopic high spots, so the true contact area is a small fraction of the apparent footprint and the gaps fill with poorly conducting air. This contact resistance is managed with thermal interface materials and with controlled bolt pressure. Indium foil, silver-filled epoxy, or graphite pads displace the air and conform to surface waviness, cutting the joint resistance by an order of magnitude. For the highest-power devices, a soldered or sintered attach replaces the mechanical joint entirely, eliminating the interface layer at the cost of repairability.

Because there are no moving parts, conduction-cooled hardware is inherently quiet, sealed, and reliable, which is why it dominates space payloads, missile electronics, and ruggedized field radios. The penalty is a hard ceiling on how much heat the path can carry before the cold plate or chassis itself runs too hot. Once board dissipation climbs past roughly 50 to 100 W per slot, designers add heat pipes, embedded vapor chambers, or transition to a pumped active cooling loop.

Governing Equations

Fourier conduction through a layer:
R_θ = t / (k × A) °C/W

Series thermal stack (junction to cold plate):
R_θ,total = R_θJC + R_θ,TIM + R_θCS + R_θ,coldplate

Junction temperature:
T_j = T_cp + P × R_θ,total

Interface (contact) resistance:
R_contact = R″ / A where R″ ≈ 0.1 to 1 °C·cm²/W

Where t = layer thickness, k = thermal conductivity (W/m·K), A = conduction area, P = dissipated power, T_cp = cold-plate temperature, R″ = area-specific interface resistance. Example: P = 40 W, R_θ,total ≈ 0.95 °C/W, T_cp = 85 °C → T_j ≈ 123 °C.

Conductive Path Materials and Cooling Methods

Material / method	k (W/m·K)	Role in path	Notes
CVD diamond	1000 to 2000	Die heat spreader	GaN-on-diamond, costly
Copper (C110)	400	Baseplate / cold plate	Heavy, best metal
CuMoCu (CPC)	200 to 250	CTE-matched carrier	Matches GaAs/GaN
Aluminum 6061	167	Chassis / wedge lock	Light, low cost
Indium foil (TIM)	82	Interface layer	Soft, conformal
Thermal grease (TIM)	3 to 8	Interface layer	Thin bondline only
Trapped air gap	0.025	Parasitic (avoid)	Dominates dry joints

Common Questions

Frequently Asked Questions

How do I calculate the cold-plate temperature needed for a conduction-cooled GaN amplifier?

Work backward from T_j,max through the stack. A 40 W GaN device with R_θJC = 0.6, case-to-baseplate 0.2, and baseplate-to-cold-plate 0.15 °C/W gives R_θ,total ≈ 0.95 °C/W, so the rise is 40 × 0.95 = 38 °C. To hold a reliability-safe T_j of 175 °C the cold plate may reach 137 °C, but most designs target 70 to 85 °C for margin since GaN lifetime roughly halves per 10 °C rise.

Why does conduction cooling use a thermal interface material instead of bolting metal directly to metal?

Two flat machined surfaces touch at only 1 to 3% of their apparent area; the rest is air, which conducts near 0.025 W/m·K and adds large contact resistance. A gap pad, grease, or indium foil displaces that air and conforms to the roughness, dropping the joint from roughly 1 °C·cm²/W dry to 0.1 to 0.3 °C·cm²/W. Keep the bondline thin (25 to 100 µm) and torque the bolts to 100 to 300 kPa.

When is conduction cooling preferred over forced-air or liquid cooling for RF systems?

It wins where reliability, sealing, or environment rules out moving parts: sealed airborne and space avionics have no fan to fail and no coolant to leak. Standards such as VPX and VME64x route heat from the card edge through wedge locks into a cold wall. The limit is heat flux, comfortably 50 to 100 W per slot; kilowatt transmitters and dense phased-array tiles need a pumped liquid loop instead.

Related Terms

← Conducted Susceptibility Conductive Adhesive →