Center for Functional Nanomaterials Seminar

Recent advances in materials theory and automated experimentation, tightly integrated with artificial intelligence, have accelerated the discovery and design of novel materials. However, most successes to date have been confined to near-equilibrium processes. Many functional solid-state materials demand precise control over defects, structural architecture, and non-equilibrium synthesis pathways—posing fundamentally different challenges for AI, both algorithmically and operationally. In this work, we explore the deployment of AI for real-time control of dynamic, non-equilibrium synthesis. Achieving this vision requires full-stack co-design—from experiment to silicon and back. We discuss several key components: automated scientific data curation at the terabyte-to-petabyte scale; architectural choices for deployment, spanning low-latency, near-edge inference on FPGAs (sub-millisecond scale) to high-throughput AI services on distributed Kubernetes clusters; and the training of ultra-compact machine learning models tailored to harsh resource constraints, where every binary operation matters. We further address issues in data pipelining, synchronization, and real-time integration with scientific instrumentation. Finally, we highlight our ongoing efforts to leverage this approach to reimagine high-performance computing as a service (HPCaaS) for deployable scientific services. Scientific use cases include real-time feedback for scanning probe spectroscopy, high-speed (>500 Hz) reflection high-energy electron diffraction (RHEED), imaging of pulsed laser deposition plume dynamics, and microsecond-latency control of tokamak plasma. Bio: Dr. Joshua C. Agar is a former Assistant Professor in the Department of Mechanical Engineering and Mechanics at Drexel University. With a foundational background in experimental materials science, Dr. Agar is renowned for pioneering contributions to AI algorithms, computing infrastructure, and the development of cyber-physical systems for materials synthesis and microscopy. His has applied these concepts to several disciplines including particle and plasma physics, materials science, and fluid dynamics, and has core expertise in synthesis and characterization of ferroelectric thin films. Now focusing full-time on a startup dedicated to advancing high-performance AI systems and ultra-low latency machine learning co-design, Dr. Agar remains an active member of the global AI community, particularly within the FastML community. His work continues to receive recognition from institutions such as the National Academy of Engineering and the National Science Foundation