Carbice CEO Baratunde Cola once dreamed of the NFL, rather than carbon nanotubes, but fate had other ideas.

A all-star footballer at high school in his hometown of Pensacola, Florida, Cola headed to Vanderbilt University in Tennessee as an undergraduate, determined to earn a place in the NFL draft by playing a starring role for the college team, the Vanderbilt Commodores.

“I went to college on a partial engineering scholarship with the idea of earning a football scholarship,” he recalls. “I was a pretty good player, but the day before I was due to discuss a scholarship with the coach, I tore my anterior cruciate ligament playing basketball. I had to go into his office on crutches, and he took one look at me and said, ‘I can’t give you anything’.”

In recounting this tale to DCD, Cola is being modest. Despite suffering a second serious injury to the same knee while on the comeback trail, he managed to regain fitness and fight his way into the team, playing for the Commodores in the Southeastern Conference as starting full-back, and eventually earning that precious football scholarship.

But by then, he had decided on a different path. During his time on the treatment table, he had channeled his efforts into mechanical engineering. Today, he leads Carbice, a company marketing a carbon-based ‘connecter’ that can be slotted into data center cooling systems between the cold plate and the server hardware. By utilizing the high thermal connectivity of the material, Carbice believes it can help data center operators make big savings through greater cooling efficiency.

Big ideas for small particles

As the name suggests, carbon nanotubes are very small tubes of carbon molecules, with a single tube usually measuring 0.5-2 nanometers (a nanometer is one billionth of a meter). When arranged into sheets, they exhibit properties including great strength and, importantly for Carbice’s purposes, high thermal conductivity.

Much of the work on their development occurred in the 1990s and is credited to Japanese physicist Sumio Iijima, though the tubes themselves were discovered much earlier. Cola first became aware of their existence while rehabbing his initial knee injury. “I got introduced to nanotubes by a professor at Vanderbilt and figured they were going to be the next big thing,” he says. “That professor moved to Purdue University in 1999, and I said to him, ‘I want to come and do my PhD with you and then start a business’.”

By 2011, PhD in hand, Cola had moved to Georgia Tech as a professor, and founded Carbice after receiving a $20 million grant from DARPA. The US Department of Defense lab, famous for funding early work on foundational technologies such as the Internet, GPS, and autonomous vehicles, had set up a programme investigating possible uses for carbon nanotubes, and Carbice was one of the beneficiaries. But what’s all this got to do with data centers?

“Carbon nanotubes are probably one of the most researched materials in human history,” Cola says. “I think I read like 100,000 research papers when we were starting Carbice. The frustration has been finding a useful application. It turns out the best use case is an interface in cooling systems.”

By interface, Cola is referring to the point where the hottest thing in the data center, the chip, comes into contact with the coldest, the cooling technology. In most systems, it’s normal to rely on a thermal paste to make the connection between the processor and the cold plate that chills it, but Cola says that this is fraught with issues. “You can’t control the paste well during production - in a lot of factories it’s applied by a guy manually squeezing it onto the cold plate - and when it’s shipped it vibrates and moves around,” he says. “So when you get the data center, you can see a 15-degree variance in temperature across the interface, depending on where the paste is, and that’s in a system where everything else is controlled with precision.”

This has negative implications, both for the performance of the components and for the amount of power needed to chill them. That wasn’t a big deal, “back in 2005 when I was doing my PhD, because the chip power was low,” Cola says. “You couldn’t see these issues as much. But as chip power went up, the problems became visible, and people were still trying to get by using old solutions, some of which weren’t even originally designed for this application.”

Carbice’s solution, known as the Ice Pad, solves this problem both on a thermal and a mechanical level, Cola says. By laying a sheet of nanotubes across the chip, where the paste would be, he says it is much easier to achieve predictable cooling performance, leading to a more efficient and cost-effective system.

“The way the nanotubes are made solves the mechanical issue, because the fibers are vertical and swirly like little springs, meaning they can stretch out and compress,” he says. “So if you have an AI chip that’s moving like a trampoline - they can go from 100 to 300 microns of curvature during a normal operation cycle - you have a material that can move with it.”

Though it uses some very advanced technology, the Carabice product is built on something you might find in your kitchen drawer alongside the cling film and baking paper. “The base is aluminium foil,” Cola explains. “Then we grow the carbon fibers on both sides of the foil sticking up. Growing the fibers out of the aluminium makes them very robust, and because of that, it can handle any kind of dynamic flexing of the chip.”

US manufacturing power

Working in materials science is a notoriously slow process. One only has to look at the lack of progress made in finding uses for another carbon-based compound, graphene, which has long been touted as a wonder material, to see how difficult it is to bring these kinds of products to market.

In comparison, Carabice appears to have taken significant steps over the last 15 years. It raised a $1.5 million seed funding round in 2017, and started shipping product to customers, including data center operators, the following year. The company also sees space as a large potential market, and Carbice products have been to the International Space Station as part of the MISSE-10 mission, which tested how various materials perform in space.

More recently, it has entered consumer technology through a partnership with CyberPowerPC, which enables the Ice Pad to be installed as a premium option in gaming PCs.

In 2020, it raised $20 million in Series A funding, helping the company build out a 69,445 sq ft (6,450 sqm) manufacturing facility in Georgia. This enables it to produce 30 million interfaces a year, each of which is big enough to cover a single GPU.

Cola says this manufacturing power is key to the Carabice proposition. “You can find hundreds of labs around the world saying they can manufacture carbon nanotubes,” he says. “But nobody has ever scaled the vertical orientation of carbon nanotubes like we have.”

The facility also allows the firm to do characterization of chip interfaces to find the optimal cooling solution for a particular server, Cola says. “Customers pay us to do their characterization for them,” he says. “We have Nvidia and AMD chips in our lab, we have a H200 server, we have Bitcoin miners. We run the qualification, and that’s a key part of our IP.”

This, presumably, is also useful for Carbice in proving how its technology is superior to tried and tested thermal management solutions. Cola admits that there is an element of “behavioral change” required for clients looking to adopt carbon nanotubes, but says operators - and other vendors - are becoming more open to new solutions.

“Five years ago, I was always telling the team, ‘we’ve got to teach the customer’,” Cola says. “Now we try to show them what we can do. We’re partnering with big system integrators, and once you can show customers that conservative, low-margin businesses like that are interested in our technology, it makes things easier.

“We have direct relationships with the hyperscalers, but the system integrators are going to be an important channel for us, because if you want a big order from a hyperscaler, it’s going to come via them telling their SI to use our technology.”

Indeed, signing up customers is now critical for Carbice, and Cola is optimistic that his long journey with carbon nanotubes is ready to bear fruit.

“Nobody likes working with thermal paste,” he says. “What we’re offering is not a competitor, it’s a replacement technology. We have that conviction, and the key is to sign up a few big customers which we hope to be able to announce in the coming months. We’re already at a point where our costs are profitable, and now it’s just about scale.”

Most materials companies, Cola concludes, “fail when they go to market,” but he says: “Cooling is a great go-to market for us because it’s a problem everybody struggles with. I won’t live forever, but Carbice is a 100-year company, it’s being built for that purpose.”

This feature first appeared in the DCD Cooling Supplement. Register here to read the entire supplement free of charge.