U.S. Laboratories Plan "Materials Discovery Cloud" to Use AI to Predict Impact of Defects in Microelectronic Devices

Researchers at the U.S. Department of Energy's Argonne National Laboratory, Lawrence Berkeley National Laboratory, Oak Ridge National Laboratory, and Northwestern University are planning to develop a "Materials Discovery Cloud" platform that leverages a physics-based artificial intelligence framework to predict how tiny defects affect the performance and lifespan of microelectronic devices.

Visualization of the Materials Discovery Cloud, a physics-based AI framework that integrates experimental data, simulations, and high-performance computing to predict how tiny defects affect the performance and lifespan of microelectronic devices. (Image provided by ChatGPT.)

Microelectronic devices are widely used in smartphones, laptops, secure communications, and artificial intelligence hardware, among other fields. As device dimensions continue to shrink and operating speeds increase, tiny defects within materials and at interfaces have a more pronounced impact on device stability. These defects can cause overheating, leakage current, switching instability, and other problems, or in certain conditions, improve electrical or thermal performance. Identifying critical defects and understanding their evolution under real-world operating conditions has become a key challenge in the design of next-generation microelectronic materials.

According to the project concept, the "Materials Discovery Cloud" will integrate experimental data, advanced simulations, and high-performance computing to link material composition, structure, and operating conditions with defect evolution and functional properties such as electrostatic potential, current density, and temperature. The research team hopes that through this platform, fragmented measurement results can be transformed into predictive capabilities applicable to device design.

The platform will leverage data resources from U.S. Department of Energy Office of Science user facilities and advanced computing systems, including the Advanced Photon Source, the Center for Nanoscale Materials, and the Argonne Leadership Computing Facility at Argonne National Laboratory; the Advanced Light Source, the Molecular Foundry, and the National Energy Research Scientific Computing Center at Lawrence Berkeley National Laboratory; and the Center for Nanophase Materials Sciences at Oak Ridge National Laboratory. These facilities provide support for X-ray measurements, microscopy characterization, materials simulation, and large-scale computing.

The researchers noted that no single experimental technique can fully reveal defect behavior in microelectronic materials. Electron microscopy can observe features at extremely small scales, X-ray methods can analyze strain, internal structure, and defect motion, and other techniques are used to measure chemical composition, electrical behavior, and heat flow. Placing diverse data into a unified framework helps provide a more comprehensive understanding of the relationship between material defects and device performance.

The project will also introduce autonomous discovery methods, combining artificial intelligence, machine learning, and robotics to assist researchers in determining which measurements to conduct next and to synthesize samples and perform multiple characterizations in higher-throughput workflows. Synthetic data generated from simulations will also be used to supplement insufficient experimental data and help train AI models.

Subramanian Sankaranarayanan, an Argonne National Laboratory scientist and principal investigator of the project, stated that the advantage of the "Materials Discovery Cloud" lies in bringing experiments, simulations, and artificial intelligence into the same workflow, enabling researchers to learn from existing limited data and continuously refine the framework as new data are added.

The research team emphasized that the system is not a "black box" that merely outputs results, but will operate under the guidance of established physical laws to ensure that predictions reflect the true behavior of materials and devices. If successfully developed, the platform is expected to help scientists identify potential issues in microelectronic materials design earlier, shorten testing cycles, and support the development of more reliable and energy-efficient electronic devices.