University of Oklahoma researcher wins DoD funding for AI-driven material design
Mike Banad, Ph.D., an associate professor in the School of Electrical and Computer Engineering at the University of Oklahoma, has secured U.S. Department of Defense funding to develop advanced materials for energy-efficient electronics and photonics. His team is building an inverse design workflow to discover, validate and manufacture materials with specific, measurable properties.
The initial focus is on metal-insulator transition (MIT) chalcogenides-materials that can switch between conductive and insulating states without changing their crystal structure. That switch makes them attractive for computing, sensing and defense applications, yet they're hard to make consistently without costly trial and error.
PMD³: a property-driven path from idea to device
Banad's group embeds AI at every step: candidate discovery, screening, recipe prediction and fabrication. The framework-property-driven material discovery, design and deployment (PMD³)-targets performance metrics researchers care about: lower energy per switch, reliable state transitions and stability under stress.
- Learn from known MIT materials: train models on structure, composition and figures-of-merit such as switching threshold, hysteresis, endurance and thermal stability.
- Propose candidates and stress-test with physics-based simulations under heat, repeated cycles and harsh environments relevant to defense use.
- Downselect to a short list and predict synthesis "recipes" (precursors, temperatures, anneals, timing) with trained models.
- Fabricate and measure against application targets, feeding results back to improve the models.
Only candidates that survive simulation and performance checks move to the lab, which cuts wasted runs and shortens iteration time.
Why this matters for science and engineering teams
For neuromorphic computing, the aim is fast, low-energy switching that holds up in demanding conditions. PMD³ is built to produce MIT chalcogenides with more consistent on/off behavior, lower energy consumption and higher tolerance to temperature and cycling-key traits for brain-inspired hardware and compact military systems.
The same workflow can extend beyond MIT chalcogenides. Combining physics-informed models with AI optimization offers a practical route to materials engineered for specific targets across electronics, sensing and secure communications.
Practical takeaways you can apply now
- Data quality drives everything: include failed syntheses and negative results; they sharpen model boundaries.
- Optimize for multiple objectives: stability, switching energy, endurance and manufacturability often trade off.
- Close the loop: use active learning to decide the next simulation or experiment, not a fixed queue.
- Recipe prediction needs a digital thread: track conditions, instruments and post-process steps for reproducibility and model retraining.
Backed by DEPSCoR
The project is funded through the DoD's Defense Established Program to Stimulate Competitive Research (DEPSCoR), which supports research capacity in underutilized U.S. states and territories. Learn more about the program's goals and scope at the DoD Basic Research Office's DEPSCoR page: basicresearch.defense.gov/Programs/DEPSCoR.
What's next
As the dataset grows and models keep training, PMD³ should produce stronger predictions and more reproducible syntheses. Banad's team plans to pursue additional support to expand the approach and apply it to new material classes.
If your lab is building skills in AI-driven modeling, optimization and experimentation, this curated catalog of courses can help your team level up faster: Latest AI courses.
Your membership also unlocks:
