Dr. Baratunde Cola, CEO and founder at Carbice Corporation, is probably one of the longest continuous nanotube researchers.  

Since 1999, whilst a student at Vanderbilt University, carbon nanotubes (CNTs) and their potential for energy technology and thermal management have been Cola’s life’s work. Today, Carbice’s global production facility in Atlanta, GA, is a 23,000 square foot unit that can manufacture 30 million square inches of thermal interface materials (TIMs) a year.

Watch the full interview with Dr. Baratunde Cola below, or continue reading for key insights.

Carbon nanotubes possess exceptional thermal conductivity and are among the most thermally conductive structures known; Cola says carbon nanomaterials are ‘the most invested-in research material in the history of humans.’ Carbice aligns billions of nanotubes to, as the company puts it, ‘form a highway for heat to move from the hot to the cold side.’ 

The applications and industries are myriad, from space and avionics to sensors and cameras, to desktop computers and data centers. Satellites were the first success story for the company. The first Carbice-cooled space mission was launched in 2018, and once cost-effectiveness and manufacturing at scale were worked out, the sky was the limit.  

“When we did that, everybody in the space industry reached out to us, because they had already tried nanotubes, they already knew they worked,” says Cola. “We were the fastest material ever to get qualified and flown in space.” 

“We’ve gotten better over time, because roadmaps increase, but that’s the thing that catches people by surprise about Carbice and about carbon nanotubes in particular,” Cola adds. “They work. The science is binary. If you could figure out how to make what works in a lab work across millions of units, now you have an ability to get the industry to adopt.” 

While Carbice has different product designs, Cola stresses that the key problem being solved across the industry is universal. 

“Whenever new materials come out, it’s not really about the material [or] the properties… it’s about what the material and the properties enable,” says Cola. “What Carbice sells is a new design capability. Before Carbice, you could not design assembly joints accurately in real-use conditions, because the materials would have non-linear properties.”  

“The material set, and the way we make it, has collapsed the assembly joint problem into a predictable elastic problem set. With that, it unlocks design capabilities.” 

Data centers, however, are a particularly interesting application for Carbice’s technology. For AI data centers, the cooling methods that have sustained traditional enterprise computing will no longer cut it. As the company notes in a recent blog post, GPUs are ‘no longer short-cycle compute assets optimised purely for peak performance’, but multi-year balance sheet investments. Existing TIMs were not designed for AI, and thermal degradation in GPUs can cause a major sustainability headache for operators and tenants alike. 

Cola notes that, while Carbice’s solution set is similar for every industry, for AI data centers the cost impact is the highest. “People talk about chips and the GPUs being large and curved, but they don’t talk about dynamic curvature, if the curvature changes when it powers up,” says Cola. “So, the trampolining that happens needs a material that can both expand and contract elastically to keep up with that dynamic nature, because you’re going to lose contact otherwise.” 

“That’s where we play. We enable the winners by ensuring that their products stay attached to the chip.” 

Getting a foothold in this potentially rocket-fuelled market is key. In February, Carbice announced a partnership with DarkNX, a company building next-generation AI data centers, supporting its customers’ projects with ‘expertise and data collection’ through the Carbice Lab center of thought leadership in thermal interface reliability. 

“I view companies like DarkNX like the beginning of the iPhone,” Cola explains. “It’s a full tech stack integration approach that allows for security, effectiveness, and user experience. You can’t get from a disconnected ecosystem.” 

Where DarkNX and Carbice align is on the understanding that the life of the GPU, the depreciation cost and curve, matters. “The analogy we like to use  is that these issues are kind of like the cost of fraud in credit cards being built into the structure,” says Cola. “That only works when the size and use space is the size of what it is today. But imagine a 100x increase in the user base of credit cards; somebody’s going to figure out how to take the fraud in. 

“That’s what is happening in the data center,” Cola adds. “Data centers are driving the cost structure of the whole support system so high that you can’t ignore solving these problems. You can’t ignore paying attention to reliability and long-term performance.”  

Cola is speaking at Microelectronics US on April 22-23, on a panel session around transforming modern data centers with photonic technologies. He hopes the audience will grasp the idea that ‘the options are unlimited with the right design tool.’ “You don’t have to just test to make assembly interfaces,” he says. “You can design them. And the benefit of that is tremendous.” 

The other takeaway is not to be afraid of taking it on, even for large curve chips and chiplets. “I think there’s an assumption out there about what works to solve these problems and what doesn’t,” says Cola. “The industry’s view is limited to a couple of options.”