Best materials for heat sinks: copper, aluminium or AlN?

Heat-sink candidates at a glance

Which material conducts heat best for a heat sink?

Silver (Ag) conducts heat best among the metals compared here, at 429 W/(m·K), with Copper (Cu) close behind at 391 W/(m·K) and Gold (Au) at 320 W/(m·K). The silver and copper figures come from different suppliers and grades and sit within 1.1 times of each other, so treat them as about equal; the silver profile makes no ranking. Several non-metal entries are higher on paper: suspended graphene flakes, a single-crystal silicon carbide table value, and two carbon profiles in the database, pyrolytic graphite and carbon nanotubes. The last two are below our full-grade bar, so we do not chart or quote them. Graphene and silicon carbide are treated below.

Silver is a poor default for a bulk sink: our data lists sulphide tarnish in air and a black oxide that flakes in low Earth orbit. Gold (Au) does not tarnish but is soft and very dense, and gold dissolving into tin-lead solder can make a joint brittle. Our records use gold for contacts and mirrors, not bulk sinks.

Aluminium is next. The commercially pure 1100-O grade gives 220 W/(m·K), the extrusion alloy Aluminium alloy 6063 gives 200 W/(m·K) and lists heat sinks among its uses, and the structural alloy Aluminium alloy 6061 gives 167 W/(m·K) in the T6 temper. The 6063 figures come from one article that calls its data indicative. Other aluminium alloys in the database are not charted; their conductivities are lower than or equal to 6063's. Commercially pure aluminium conducts 1.3 times as much as 6061, but its yield strength is 29 MPa against 276 MPa. That is the trade: a little conductivity for a part you can machine and bolt down.

Among ceramics, Aluminium nitride (AlN) is listed at 140 to 180 W/(m·K), and Silicon carbide (SiC) at 140 to 490 W/(m·K). Read the silicon carbide range with care. A manufacturer datasheet and a reference page give 140 to 150 W/(m·K) for direct-sintered and reaction-bonded grades; the top of the range is a 4H single-crystal figure from a semiconductor table, and another source in the same profile gives a lower value than that table without explaining the gap.

The device materials sit lower still: Silicon at 150 W/(m·K) and Gallium nitride (GaN) at 130 W/(m·K). Heat made in a transistor has to cross these first, which is why the base under them matters.

General engineering knowledge, not in our data, with no numbers. Contact and interface resistance and the sink geometry often matter more than the conductivity gap between candidates. Finned sinks are usually extruded, so Aluminium alloy 6063 is the alloy to start from and Aluminium alloy 6061 suits parts machined from bar. Heat leaves the fins by airflow, and in vacuum on a spacecraft by radiation, so the radiating surface matters. Joining copper to aluminium risks galvanic corrosion at the joint, and anodising an aluminium sink changes its radiating surface. Cost is not in our data.

How much weight do you pay for better conductivity?

Copper pays for its conductivity in mass. Its density is 8.9 to 8.96 g/cm³, which is 3.3 times that of Aluminium alloy 6061, for 2.3 times the conductivity (larger end of each range). Against commercially pure Aluminium the ratios are 3.3 times and 1.8 times. Either way the density ratio is the larger, so for a solid conduction path aluminium moves more heat per kilogram, and copper more per unit of volume.

Silver (Ag) at 10.5 g/cm³ and Gold (Au) at 19.3 g/cm³ are heavier still, and Tungsten (W) is as dense as gold at 19.3 g/cm³, with 174 W/(m·K). In a mass-limited design these three are hard to justify as sinks.

The ceramics change the picture. Copper (Cu) is 2.7 times as dense as Aluminium nitride (AlN) but only 2.2 times as conductive at the larger end of the nitride's range. Silicon carbide (SiC) has a density of 3 to 3.17 g/cm³, similar to the nitride. Against aluminium, the nitride is 1.2 times as dense, a modest weight cost for an insulating base.

Which heat-sink material has an expansion that matches a chip?

The ceramics expand far less than the metals, but our data holds no silicon expansion value, so we cannot give you the mismatch. Aluminium nitride (AlN) has 4.5 ppm/K (Accuratus, temperature not stated) and 5.5 to 5.6 ppm/K (CoorsTek grades, 25 to 1000 °C), and Silicon carbide (SiC) has 4.4 ppm/K (25 to 1000 °C). The metals are far higher: Aluminium alloy 6061 23.6 ppm/K (20 to 100 °C), Copper (Cu) 17.6 ppm/K, Free-cutting brass C36000 20.5 ppm/K. The temperature intervals differ, so these comparisons are indicative. The aluminium nitride profile quotes its supplier as saying the expansion is close to silicon's; that is a supplier statement we have not checked. General engineering knowledge, not in our data: silicon is a low-expansion material. Check the figure on your die datasheet, for its own temperature interval, before relying on any value here.

Also general engineering knowledge: when a hot chip is bonded to a part that expands much more, each temperature swing loads the solder or adhesive layer between them. A ceramic base between the chip and a metal sink is a common way to keep that stress small.

Low expansion is not the same as a good sink. Invar 36 (FeNi36) is the low-expansion reference at 1.18 to 4.92 ppm/K, but it carries only 10.5 W/(m·K). Tungsten (W) has 4.5 ppm/K and 174 W/(m·K), but it is dense and brittle and oxidises in air when hot. Gallium nitride (GaN) sits at 5.59 ppm/K, Alumina (Al2O3) at 8.2 ppm/K and Silicon nitride (Si3N4) at 1.4 ppm/K near room temperature and 3.2 ppm/K to 1000 °C.

Can one material give high conductivity, low expansion and electrical insulation?

In our data, Aluminium nitride (AlN) comes closest. It combines 140 to 180 W/(m·K), the low expansion above, and an AC dielectric strength of 16 to 20 kV/mm across the grades listed. The profile describes it as a heat-spreading ceramic that stays electrically insulating, and lists substrates, IC packages, power transistor bases and heat sinks among its uses. The cost is that it is brittle, with a flexural strength of 300 to 375 MPa, and it hydrolyses slowly in water, with no moisture limit printed. Its supplier sheets print volume resistivity as >10^14 ohm·cm (Accuratus) and >10^12 to >10^14 ohm·cm (CoorsTek); this is not entered as a property in our data. General engineering knowledge, not in our data: ceramic bases usually need metallisation or a direct-bonded copper layer before they can be soldered or brazed.

Alumina (Al2O3) is the plain insulating alternative. In the AD-995 grade it has a volume resistivity above 10¹⁴ ohm-cm at room temperature, an expansion of 8.2 ppm/K and a conductivity of only 30 W/(m·K); density spans AD-96 and AD-995. The top of the typical nitride range (180 W/(m·K)) is 6 times that figure; at the 140 end the gap is smaller. Alumina's resistivity also falls steeply as it gets hot, so check it at your operating temperature.

Silicon nitride (Si3N4) is the strong ceramic: a flexural strength of 670 to 900 MPa in our record, well above the nitride and oxide above, with 24 to 29 W/(m·K). Our data holds no electrical resistivity for it, so we do not claim insulation here.

Silicon carbide (SiC) looks attractive on conductivity (sintered grades only) and on expansion, but it is both a hard ceramic and a wide-bandgap semiconductor, and our profile gives no resistivity for the ceramic grades. Do not assume it insulates. Tungsten (W) and Invar 36 (FeNi36) are metals and conduct current, so if the part must block current only the ceramics are candidates.

Is graphene a better heat spreader than copper?

On paper, yes, and the gap is not closed by a substrate alone. Suspended graphene flakes are quoted up to 5,000 W/(m·K), many times Copper (Cu) at 391 W/(m·K). The review cites about 600 W/(m·K) for graphene on a silicon dioxide on silicon substrate, still above copper in-plane, and about 160 W/(m·K) for a single layer encased in oxide, which is below it. The substrate and the encasement take away much of the advantage.

The real limit is thickness. General engineering knowledge, not in our data: a film is one to a few atoms thick, so the total heat it can carry is small even when its conductivity is high. It works as a spreading coat on a hot spot, not as a bulk sink. The uses in our record are research: few-layer graphene films on transistors on silicon carbide as a proof of concept, CVD graphene on thermal evaluation chips, and simulated spreaders in silicon-on-insulator circuits. No commercial product use was read.

The database also holds pyrolytic graphite, whose profile names heat spreading as its role, and carbon nanotubes. Both are below our full-grade bar, so they are not charted here. Graphene has no density or expansion value in our data, so it is absent from the density and expansion charts.

Which common materials make poor heat sinks?

Structural metals chosen for strength or corrosion resistance. Ti-6Al-4V has 6.7 W/(m·K), Stainless steel 316 has 13 to 17 W/(m·K) and Invar 36 (FeNi36) has 10.5 W/(m·K). Copper (Cu) is 58.4 times as conductive as the titanium alloy. Use these where strength, corrosion resistance or a fixed dimension is the job, and give the heat a separate path in copper or aluminium.

For the grades we hold, the ceramics Alumina (Al2O3) and Silicon nitride (Si3N4) are in the same range (30 W/(m·K) and 24 to 29 W/(m·K)). General engineering knowledge, not in our data: a thin layer passes heat across its thickness even at modest conductivity, so these grades can serve as thin insulating layers, but they are no match for aluminium nitride in a thick base.

Free-cutting brass C36000 is a free-cutting alloy with 2.5 to 3.0 % lead that dezincifies in water, so avoid it for a sink that sees water.

Terms used

Thermal conductivity

how fast heat flows through a material for a given temperature difference. Higher means the material carries heat better. Our data gives it in W/(m·K).

Heat sink and heat spreader

a spreader moves heat sideways away from a hot spot. A sink hands that heat to air, liquid or a radiating surface through a large area. Many parts do both.

Substrate

the base that a chip or circuit is built on or mounted on. A ceramic substrate is often the mechanical base and the heat path at the same time.

CTE

coefficient of thermal expansion: how much a material grows per degree of temperature rise, here in ppm/K (millionths per kelvin). A mismatch between bonded parts loads the joint when the temperature changes.

Dielectric strength

the electric field an insulator survives before it breaks down, in kV/mm. Higher is better for an insulating substrate.

Suspended graphene

a graphene flake held across a gap with nothing underneath. This is the condition that gives the highest conductivity values; on a surface the value falls.

What to check before you order

1. Ask for the grade and temper behind every conductivity figure. In our data copper is the soft oxygen-free grade and aluminium is the commercially pure, annealed 1100-O grade.
2. For aluminium nitride, ask for the measured thermal conductivity of your lot. Our data holds a typical supplier range (140 to 180 W/(m·K)), not a guaranteed minimum.
3. For ceramics, ask for volume resistivity and dielectric strength at your operating temperature, not only at room temperature.
4. Ask for the expansion coefficient over your own temperature range, and take the silicon expansion from your die datasheet.
5. Ask how aluminium nitride is stored and sealed. Our sources say it hydrolyses slowly in water and print no moisture limit.
6. Ask how the part will be joined to the chip or sink (solder, braze or adhesive), what the interface layer is, and whether copper meets aluminium at a joint. Our data holds none of this.

What do we not know yet?

• Silicon has no coefficient of thermal expansion in our data, so no mismatch to a silicon die can be computed; use your die datasheet. Graphene has no density or expansion value either.
• Aluminium nitride conductivity is a typical supplier range. The CoorsTek grade sheets print the conductivity unit incomplete, so those grades are not entered, and no aluminium nitride volume resistivity is entered as a property.
• Silicon carbide has no resistivity for its ceramic grades, and the single-crystal 4H conductivity cannot be taken as a sintered-ceramic value. Silicon nitride has no electrical resistivity either, and high-conductivity silicon nitride substrate grades are not among those we hold.
• Not charted or not in our data: pyrolytic graphite and carbon nanotubes (profiles below our full-grade bar, so not quoted), diamond, beryllia, copper-tungsten and copper-molybdenum composites, thermal interface materials and any joining data. The graphite-anode profile is a battery record with no thermal conductivity. Molybdenum disulfide is a lubricant with no thermal conductivity in our data.
• Almost all values are for room temperature or an unstated temperature. No candidate has a conductivity curve over an operating range, and no sink-level thermal resistance or fin performance data is held. The 6063 values come from one article that calls its data indicative.
• Cost and supply are not compared, and the space-environment record for these materials as heat-path parts is thin: it covers silver and copper oxidation in orbit, while the ceramics have only ground tests.

Questions people ask