AI News

Diamond Foundry Creates 200mm Single-Crystal Diamond—Can It Cool GPUs?

Semiconductor Industry ObserverUpdated Nov 16 21:10

AI summary

On November 5, 2026, Diamond Foundry announced it had produced single-crystal diamond 200 mm (8 inches) in diameter, pushing previous 100 mm-class single crystals to the wafer scale. High-quality single-crystal diamond has a thermal conductivity of about 2000 to 2200 W/(m·K), roughly 5 times that of copper, and can be used for near-junction heat-spreading layers or substrates. However, the sample is an unpolished as-grown crystal; dislocation density, cost, and yield were not disclosed, and single-crystal substrates are priced in the tens of thousands of yuan per wafer, so in the short term it is more likely to be used first in RF and power devices.

Body · AI translation

English text is being prepared.

The next major breakthrough in semiconductors may not come from smaller transistors, more ingenious chip architectures, or more complex packaging technologies. It could also come from a material humanity has known for thousands of years—diamond.

On November 5, 2026, California-based company Diamond Foundry announced that it had successfully produced single-crystal diamond (SCD) with a diameter of 200 millimeters (8 inches). According to the company, this is a single crystal grown continuously as one piece, not assembled from small crystal pieces, nor is it polycrystalline diamond. The company's previous milestone was 100-mm-class single crystal; this time it has pushed the size to wafer scale. It should be noted that public information shows the sample displayed this time is an as-grown, unpolished raw crystal, and the company did not disclose quantitative data such as dislocation density, measured thermal conductivity, cost, and yield. In other words, there is still a distance between "meeting the size target" and "being usable at device level," and this distinction determines how this news should be viewed.

The significance of this achievement goes far beyond producing an oversized synthetic diamond. It means that a material long constrained by crystal size has, for the first time, crossed the "wafer-level" threshold. Previously, diamond could only appear in the form of heat spreaders, coatings, or small-size substrates, and it was difficult for it to enter semiconductor equipment and process systems that operate with a full wafer as the unit. A clarification of scope is also needed: advanced logic processes currently use 300-mm wafers, while 200 mm corresponds to mature nodes and the production-line specifications for power and RF devices. Therefore, a more accurate statement is that diamond has begun to align with the size specifications of some mainstream silicon production lines, rather than directly entering the most advanced processes.

Thermal conductivity about 5 times that of copper: why semiconductors need diamond

Diamond is not just one of the hardest materials in nature. It also possesses a set of properties—extremely high hardness, excellent wear resistance, broad-spectrum optical transparency, good insulation, and extremely high thermal conductivity—that are highly attractive for advanced electronic devices.

The core value of diamond in semiconductors is efficient heat conduction while maintaining electrical insulation. This combination allows heat to be redistributed without introducing extra conductive paths. The thermal conductivity of high-quality single-crystal diamond is about 2000 to 2200 W/(m·K), about 5 times that of copper (about 400 W/(m·K)) and more than ten times that of silicon.

Traditional chip cooling relies on copper vapor chambers, thermal interface materials (TIM), vapor chambers, and increasingly complex liquid cooling systems. But as the power density of AI processors continues to climb, the underlying contradiction in cooling has changed. The challenge is no longer just how to expel heat from the server rack, but how, once heat is generated, to spread it over the shortest distance and prevent local hot spots from limiting chip performance first. This is especially critical for advanced packaging architectures featuring 3D integrated circuits, high-bandwidth memory (HBM), and high-density heterogeneous integration. Local hot spots create thermal bottlenecks, forcing chips to reduce frequency, accelerating device aging, and reducing long-term reliability.

Diamond offers a potential solution at the material level. Its value is not only in making existing cooling solutions more efficient, but also in potentially changing the conduction path of heat within the chip structure at the root. Looking at the composition of thermal resistance from junction to case, the interfaces between the chip, TIM, and lid often account for a large proportion; using diamond as a near-junction heat-spreading layer or substrate can significantly reduce thermal resistance in this segment. At the same time, its scope of action must be defined: diamond addresses heat diffusion inside the chip and at epitaxial interfaces, while heat removal at the rack and data center level still relies on systems such as liquid cooling and air cooling; the two are complementary, not substitutes.

From material breakthrough to industrial implementation: landscape, costs, and variables

This track is not being pursued by only one company. In June 2026, Element Six (part of the De Beers Group), a leading UK synthetic diamond company, and Japanese precision materials company Orbray jointly announced that they had established a repeatable production process for 3-inch (about 75-mm) wafer-level single-crystal diamond, with 4-inch substrates entering the development stage and 2-inch thermally bonded substrates entering the final stage of mass-production preparation; Orbray plans to start 2-inch mass production in 2027. The two parties had previously also announced 50-mm single-crystal substrates with thermal conductivity exceeding 2200 W/(m·K). In horizontal comparison, 200 mm is a clear leap in size, but competitors have moved earlier in the engineering links of "repeatable and deliverable." The real watershed in the industry is often not who first produces a large size, but who first brings dislocation density, thickness uniformity, surface processing quality, and batch consistency to device-level standards—these metrics directly determine whether downstream customers can run full-wafer processes.

Cost is another threshold. The cost of diamond-copper composite materials is about 3 to 5 times that of pure copper, polycrystalline diamond heat sink sheets are more expensive, and single-crystal substrates are priced at tens of thousands of yuan per piece, with limited supply. From this it can be inferred that the more realistic short-term application order is roughly: RF and power devices (GaN-on-diamond, where improved heat dissipation can directly translate into higher power and efficiency), then high-power lasers, optical modules, and other scenarios, and only gradually AI accelerator packaging. For single-crystal diamond to enter GPU packaging, supporting bonding processes, interface thermal resistance control, cutting and inspection capabilities, and stress management due to thermal expansion coefficient mismatch are also needed—diamond's thermal expansion coefficient is about 1 ppm/K, significantly different from silicon and GaN.

From the perspective of the industrial chain, once diamond enters packaging, the value chain will be redistributed: material suppliers move forward in position, the packaging and testing segment needs to add bonding and processing capabilities, equipment and metrology segments need to adapt to non-silicon materials, and the design side needs to re-evaluate thermal design models and reliability criteria. Geographically in terms of supply, large-size single-crystal growth capabilities are currently concentrated in a small number of companies in the United States, Japan, and the United Kingdom. China already has a relatively mature industrialization foundation in polycrystalline diamond and diamond-copper composite materials, but still needs to make progress in large-size single-crystal substrates. This means both the risk of increased supply chain concentration and the possibility that bargaining power may shift partly from traditional thermal management material manufacturers to a small number of companies that master crystal growth technology.

A judgment that needs to be clarified: diamond will not replace silicon logic. Its role in this path is as a thermal management material, while as a wide-bandgap semiconductor material it also has long-term possibilities—p-type doping is relatively feasible, while n-type remains an industry challenge. The emergence of 200-mm single crystal proves that the long-standing bottleneck of "size" is loosening, but industrialization still must undergo long-cycle testing of yield, cost, standards, and customer validation. For current AI chips, it provides a new cooling path, not the endpoint.

This article is reproduced from media reports or online platforms and represents the author's personal stance or views. Our republication is solely for sharing and does not mean that we endorse or agree with it. If the source attribution is incorrect or your lawful rights and interests are infringed, please contact customer service promptly, and as a neutral platform service provider we will promptly correct, delete, or handle it in accordance with the law. Comments: No user comments yet. Latest News More Related Recommendations

Dekryptek launches millimeter-wave ultra-high-speed interconnect solution, targeting connections within 10 meters in AI data centers 2026-11-16 16:17

Why is an ST chip being made by Samsung in the United States? 2026-11-16 16:06

Southeast Asia smartphone shipments fell 36% in Q3, with only Apple achieving positive growth 2026-11-16 15:55

Source

Semiconductor Industry Observer