What Has Shipped, And What Is Still A Sample

“Diamond cooling” is used as if it were one product. It is at least five, and they are at very different stages. Mixing them up is how buyers end up specifying powder for a supply chain that will not exist for three years, or missing the one that is buying today.

The numbers attached to the first route are the loudest: single-crystal diamond conducts heat at 2,000 W/m·K and above against roughly 400 for copper, which is where “five times” comes from. That is a material property, not a finished-part performance figure, and the wafer that made the news still has to be polished before anyone can bond it to a die.

Forecasts quoted in the same coverage put AI-chip diamond heat spreaders near RMB 8.7 billion in 2026 and RMB 59 billion by 2030. Treat that as a direction of travel. The narrower fact is that the composite routes are the only ones buying diamond powder in tonne quantities this year.

Why The Loudest Route Buys The Least Powder

A single-crystal wafer is grown from a gas mixture on a seed crystal. There is no powder anywhere in that process, and there never will be. The same holds for polycrystalline film: methane and hydrogen in a plasma, then a laser plane cut and a release step.

So the headline reads “diamond replaces copper”, while from a powder supplier’s side it is a story about consumables. Every one of those wafers has to be turned into a flat, sub-nanometre surface first, and the only thing that grinds diamond is diamond. The faster the CVD routes scale, the more fine grit they consume — not less.

The Stage Nobody Prices: Finishing

Growth produces a disc that is bowed, rough and unusable. Published process descriptions for 3 inch heat spreader production put the target at nanometre-scale surface roughness with total warp held under 10 µm. Warp is the harder of the two, because it is a stress problem spread across the whole wafer, and it gets harder as diameter grows.

The sequence is conventional and unforgiving: wire or laser separation, then double-side lapping, then mechanical polishing, then a chemical-mechanical step for the final surface. Grit does the work in the first two stages.

• Lapping, 3–6 µm — removes the damage layer left by separation. Narrow distribution matters more than average size, because a single coarse particle sets the depth of the scratches you then have to polish out.
• Polishing, 0.5–1 µm — brings Ra down and starts to reveal subsurface damage.
• Final, below 0.25 µm — where agglomerates show up as directional sleeks rather than as roughness numbers.

None of this is new to anyone who has finished 4H-SiC, with one twist: the workpiece is now the same material as the abrasive, so there is no hardness differential to exploit. Removal relies on mechanical action and carrier chemistry, not on the grit being harder than the work.

Practical consequence for specification: the finishing grades are bought on consistency, not on cutting power. A supplier who can hold D50 within a tight window across a hundred batches is worth more here than one with a sharper particle. The bands used run from 3 µm down to sub-micron: HDG micron powder covers the lapping stages, and HDY the tighter windows for final polish.

What The Wafer-Size Race Changes

The 200 mm announcement matters less as a record and more as a statement about where the roadmap points: 300 mm before 2030, matching the size leading-edge fabs already run.

Every step up in diameter moves the powder requirement the same way. Polished area scales with the square of diameter, and uniformity across a larger field is harder to hold. A part that is flat in the centre and 15 µm off at the edge is scrap.

That pushes the buying decision away from “what is your finest grade” and toward “what is your batch-to-batch variation”. For a 3 inch heat spreader, a 5 % swing in D50 is absorbable. For a 300 mm wafer it is not, because the same swing now appears as a thickness variation across a much wider field.

Where It Goes Wrong

Four failures we see repeatedly in thermal work, and what each one actually points to.

What To Put On The Order

Two questions decide everything else, and they are usually skipped.

• Is the powder a filler or an abrasive in this application? A filler is bought on packing, coating and interface; an abrasive is bought on shape, toughness and size control. The same grade number means different things on the two sides.
• Which route is it feeding? Composite, TIM or finishing. A specification written for one is not transferable to the others.

Then the details that survive an audit:

• PSD per fraction, with the instrument named — laser diffraction, sedimentation and sieving do not agree
• Maximum particle size as a single value, plus coarse-particle count
• Coating type and weight gain as a percentage, not as a grade name, plus the continuity criterion
• Ash ceiling and the analytical method — the residue ends up exactly at the interface you are trying to engineer
• A consistency commitment, stated as the range of D50 across the last five batches

Next Step

If you are building a composite, the interface is the whole problem and the diamond composite thermal management page covers coating and packing in detail. If you are filling a grease or a gap pad, start with thermal interface materials, where loading rather than particle quality sets the result. If you are finishing wafers or spreaders, the grit logic is the same one used on SiC and sapphire — wafer polishing sets it out.

For the material property side of the same question, functional diamond powder for thermal management compares the coated and uncoated routes, and quality control across batches is where the consistency argument is made in numbers.

The two coatings this work depends on are both stocked: titanium-coated diamond for metal-matrix composites, where the coating has to become the bond, and nickel-coated diamond where the matrix is a polymer or a softer alloy. Each page lists coating weight as a percentage and the size bands available, so you can compare against what your process actually calls for.

Send the route and the target finish and we will come back with a grade, a PSD window and a sample quantity that matches the stage you are actually at.