Imagine a future where researchers discover groundbreaking materials with properties exceeding conventional expectations at an unprecedented pace. Researchers at Oak Ridge National Laboratory (ORNL) are turning this vision into reality with a groundbreaking autonomous materials synthesis tool. The system marries the power of AI, automated experimentation, and real-time diagnostics to accelerate the hunt for new materials.

Navigating a sea of possibilities with Bayesian optimization

At the heart of this innovation lies Bayesian optimization (BO), the AI method for guiding the experiments and exploring the parameter space, used with pulsed laser deposition (PLD), a technique used to create thin films. The system taps BO to predict the optimal conditions for materials synthesis by analyzing how the quality of the newly created material relates to the synthesis conditions, such as temperature, pressure, and energy emitted during the PLD process. As the experiment progresses, the BO evolves, updating its surrogate function and variance projections in each 2D parameter plane. This allows it to suggest a revised set of conditions that may yield improved quality. Orchestrated by open-source Python software, the autonomous workflow seamlessly integrates popular AI and machine-learning libraries. This allows the system to efficiently navigate the vast landscape of potential parameters, which would be incredibly time-consuming for human researchers.

Ten times faster with AI

“We built computer control of all processes into the system and incorporated some hardware innovations to enable AI to drive experimentation,” the study’s leader, Sumner Harris of the Center for Nanophase Materials Sciences at ORNL, said in a press release. “The automation allows us to perform our work 10 times faster, and the AI can understand huge parameter spaces with fewer samples.

During synthesis, the system constantly gathers data through 50 ICCD images and an ion probe trace. This real-time feedback allows the AI to make informed decisions, dynamically adjusting the synthesis conditions for optimal results.

Synthesizing superior materials with AI

Demonstrating its capabilities, the autonomous system successfully synthesized WSe2 thin films using PLD. The AI, analyzing the data, determined that increasing the background pressure from 45 mTorr to 107 mTorr significantly enhanced the quality of the synthesized material, a conclusion the Raman score validated. “After the 125 active learning steps, a series of 10 samples were grown where only the temperature was varied and the other parameters were fixed,” the study in Small Methods noted. “The BO prediction for Raman score along this unexplored axis in parameter space qualitatively matches the experimental outcome.”

Toward a new era of materials discovery

This power of AI in materials discovery isn’t limited to ORNL’s initiative. For instance, Microsoft partnered with the Pacific Northwest National Laboratory (PNNL) to accelerate the development of new battery materials. The AI system, trained on vast datasets of materials and their properties, screened 32.6 million potential candidates in 80 hours — a feat that the researchers estimate would have taken an estimated 20 years using traditional methods.

The ORNL’s autonomous materials synthesis tool could represent a significant leap forward in materials science. By uniting AI, automation, and in situ diagnostics, researchers could explore vast parameter spaces with unprecedented speed and efficiency, uncovering optimal synthesis conditions without constant human oversight. This breakthrough, along with parallel advances like the Microsoft-PNNL collaboration, holds the potential to upend materials discovery, paving the way for the creation of novel materials boasting extraordinary properties and pushing the boundaries of scientific innovation.