A pioneering thermal management method adapted directly from nuclear reactor systems is supercharging artificial intelligence-driven materials research aimed at powering the future of clean energy.

Through a strategic alliance between data center cooling innovator Ferveret and the Massachusetts Institute of Technology’s Materials in Extreme Environments Laboratory, researchers are utilizing graphics processing units equipped with Ferveret's adaptive cooling technology. This setup empowers the team to train massive machine learning models, accelerate iteration cycles, and execute complex simulations to evaluate material performance under severe operational stressors inherent to advanced nuclear reactors.

According to Reza Azizian, founder and chief executive officer of Ferveret, this partnership provides a vital testing ground to assess their cooling infrastructure against authentic AI workloads rather than simulated dummy loads, simultaneously delivering crucial computational bandwidth to the MIT researchers.

Azizian emphasizes that enhancing thermal efficiency offers a direct solution to the escalating power crisis facing modern data centers. The firm's proprietary Adaptive Phase Cooling methodology extracts higher usable computational output from a constrained power budget, enabling scientists and enterprises to maximize their current infrastructure.

Adaptive Phase Cooling is inspired by how nuclear reactors are cooled, a process that industry has used efficiently and reliably for about 70 years. Cooling is a big one, because in a typical data center a large share of the power goes to cooling overhead instead of compute. — Reza Azizian, CEO of Ferveret

Bridging Academic Research and Industrial Innovation

Highlighting the convergence of modern computing and scientific breakthroughs, Azizian noted that digital processing serves as a fundamental catalyst for discovery. It grants researchers the capability to model intricate physical mechanics—such as the structural degradation of novel materials inside an active nuclear reactor—at unprecedented scales.

For Ericmoore Jossou, Assistant Professor of Nuclear Science and Engineering alongside Electrical Engineering and Computer Science, and leader of the Jossou Research Group, the infusion of dedicated GPU power fundamentally transforms investigations into material resilience under intense heat, radiation, and corrosive atmospheres.

“Understanding how materials behave under extreme environments, such as high temperatures, irradiation, or corrosive conditions, is important for the design of better and longer-lasting materials,” stated Jossou.

Furthermore, Jossou's division is engineering novel artificial intelligence architectures that integrate the underlying physics of diffraction straight into machine learning frameworks, pushing for heightened precision and dependability in image reconstruction.

  • Transitioning from single-run reconstructions on shared academic servers to massive, parallel testing on dedicated GPU hardware.
  • Generating thousands of synthetic, high-fidelity datasets to train sophisticated generative AI models.
  • Automating experimental workflows to drastically shorten discovery timelines.

Jossou highlighted that moving beyond legacy bottlenecks allows his team to target real-time deployment capabilities during active laboratory experiments.

“We previously used to run these reconstructions one at a time on a shared lab server. With sustained access to Ferveret’s GPUs, we can test our AI model in parallel and at scale on state-of-the-art GPU resources. We can also generate thousands of realistic datasets to train generative AI models and automate our experiments,” Jossou explained.

By achieving real-time precision, materials scientists will gain the unprecedented ability to track structural defects as they emerge within material matrices. Within the realm of nuclear engineering, this capability threatens to drastically shorten development timelines for next-generation fission and fusion energy systems.

Jossou points out that these very same material vulnerabilities plague a broad spectrum of industrial sectors—ranging from advanced battery manufacturing and chemical catalysts to semiconductors—where microscopic flaws dictate overall device efficacy and lifespan.

Ultimately, unconstrained access to high-performance computing enables the Jossou Research Group to tackle ambitious scientific inquiries that would typically remain out of reach for a standard academic facility.

“We want image reconstruction to be accurate and fast enough to deploy in real time during experimental campaigns. This will enable materials scientists to monitor various defects causing failure within materials as they occur,” Jossou concluded.