Why Materials Discovery Suddenly Got Interesting
Behind every cool new device there is usually a breakthrough in materials. Think liquid cooled data centers that do not overheat, flexible OLED screens, or batteries that last all day. To build this future, scientists need new chemicals and materials that are more efficient, more durable and more energy friendly.
The problem is that discovering these materials the old way is slow and expensive. You mix and test in the lab, tweak a few variables and hope for the best. Now NVIDIA is using accelerated computing and AI to flip that workflow. Instead of trying a handful of ideas in the lab, researchers can simulate millions or even billions of options virtually and only bring the best candidates into real experiments.
At the SC25 conference in St. Louis, NVIDIA showed how its new data processing pipelines and AI microservices are changing the game for chemistry and materials science. The spotlight is on two key platforms:
- NVIDIA Holoscan for real time sensor and imaging data
- NVIDIA ALCHEMI with NIM microservices for chemistry and materials simulations
Three very different players are already putting this tech to work: Brookhaven National Laboratory, energy giant ENEOS and OLED pioneer Universal Display Corporation. Together they show how GPUs and AI are becoming core tools for modern materials discovery.
Seeing At The Nanoscale In Real Time
Brookhaven National Laboratory runs the National Synchrotron Light Source II, one of the world’s most advanced X ray facilities. It uses powerful X ray beamlines to probe the structure of materials like batteries, microelectronics and nanoparticle systems at extremely high resolution, down to a few nanometers.
Each experiment creates a firehose of data. Before, researchers would have to wait for full scans to finish and then run heavy processing workloads before they could see anything useful. That meant slow feedback loops and less efficient use of very expensive instruments.
By integrating NVIDIA Holoscan into their pipeline, Brookhaven’s team can now process streaming data at the edge as it comes off the sensors. The result is near instant feedback during an experiment rather than after it.
That matters for a few reasons:
- Scientists can spot interesting regions on the fly and zoom in immediately
- They can watch how properties evolve during a measurement instead of guessing after the fact
- They can run more experiments in the same amount of beam time, cutting costs and boosting throughput
The data engineering lead at the facility points out that this efficiency is not just convenient. It means they can support more users and do more science overall. They also see a path to AI assisted operation, where models help decide what to scan next and even run autonomous experiments.
In other words, Holoscan is turning a massive scientific camera into something closer to a real time, AI aware sensor rig that lets researchers iterate like software developers.
Searching Millions Of Candidates For Better Energy And Cooling
Japanese company ENEOS works on large scale energy problems. Two of their big targets are next generation immersion cooling liquids for data centers and better catalysts for processes like hydrogen fuel production.
Both problems involve exploring huge chemical search spaces. You want liquids that safely pull heat away from racks of servers and catalysts that make energy conversion reactions efficient and practical. Testing candidates one by one in the lab is basically impossible at the scale needed.
With NVIDIA ALCHEMI NIM microservices, ENEOS can run GPU accelerated conformer search and molecular dynamics simulations on the cloud. In plain language, that means they can quickly:
- Generate plausible 3D shapes of molecules
- Simulate how those molecules move and interact at the atomic level
These simulations let them prescreen huge numbers of molecules virtually. Only the most promising ones graduate to physical experiments. The result is a massive speedup in research and lower development costs.
Using ALCHEMI, ENEOS ramped up to testing around 10 million candidate liquids for immersion cooling and 100 million candidate materials for oxygen evolution reactions in just a few weeks. That is at least a ten times jump compared with what they could manage before.
The team says they had never even considered running searches at the 10 to 100 million scale. With the new tools, calculations finish fast enough that scientists spend their time analyzing and designing instead of waiting for jobs to complete.
Designing The Next Wave Of OLED Displays
Universal Display Corporation is behind many of the OLED materials used in phones, TVs, laptops, cars and extended reality headsets. Their challenge is intense: the number of possible molecules for OLED materials is estimated around 10 to the power of 100. Somewhere in that astronomical space are combinations that could give brighter, more efficient, more colorful screens.
Historically UDC used conventional CPU clusters for simulations. That limited how broadly they could explore. Researchers had to lean heavily on their own intuition to pick narrow regions of chemistry to investigate.
Now UDC is using NVIDIA ALCHEMI NIM microservices to run AI accelerated conformer searches and molecular dynamics, boosted by GPUs. This lets them:
- Evaluate billions of candidate OLED molecules much faster, in some cases up to 10,000 times quicker than older methods for the initial screening
- Run detailed dynamics simulations of the most promising compounds up to 10 times faster per run
- Distribute workloads across multiple GPUs and shrink simulation times from days to seconds
The company is applying this stack to ambitious projects such as blue phosphorescent OLEDs, which could bring big energy savings and longer battery life to consumer devices.
UDC’s leadership describes the impact as a shift in how scientists think. With capacity and throughput less of a concern, individual researchers can be more creative and exploratory. They get immediate feedback on new ideas instead of waiting on long queues, which in turn accelerates the pace of discovery.
All of this is built on top of NVIDIA’s CUDA X ecosystem, which now includes more than 150 libraries and frameworks targeting real scientific and engineering workloads. For materials discovery, NVIDIA ALCHEMI and Holoscan are quickly becoming core tools in the lab, letting researchers treat chemistry and physics problems with the same rapid iteration mindset that game and software developers use every day.
Original article and image: https://blogs.nvidia.com/blog/ai-science-materials-discovery-sc25/
