Ohio State Materials and Manufacturing Conference Program

Abstract: I would like to discuss the impact of artificial intelligence and self-driving lab on the practice of research & development, in particular, clean energy research. ["Autonomous experiments using active learning and AI," Nature Reviews Materials 8 (2023) 563­564; "A multimodal robotic platform for multi-element electrocatalyst discovery," Nature 647 (2025) 390-396]   Rapid growth in modeling, experiment and reasoning capabilities, such as universal neural network interatomic potential (UNIP), large language model based hypothesis generation, robotic high-throughput experimentation, and knowledge-based Bayesian optimization, could usher in an era of “mass production of science”, but plenty of challenges and peril lie ahead. 

Abstract: Artificial Intelligence (AI) is rapidly reshaping almost every aspect of our life, just as what lithium-ion battery did about four decades ago. As a special application of AI, AI4Science represents a fundamental shift in how research is conducted, accelerating discovery across nearly every scientific domain, from materials design to system data analysis, from image processing to manufacturing control, and eventually the fully autonomous laboratories.

Among those numerous materials discovery domains, battery perhaps represents the most challenging scenario, not only because battery is a complicated system that is subject to the influences of myriads of parameters in many dimensions, but also because all components in a battery are working at electrochemical extremities far beyond their thermodynamical limits.

Yet again, in this challenging system, electrolyte is undoubtedly the most unique and challenging component, because it must interface with every other components in the device, be it active (anode, cathode or other redox species), assisting (conductive additive, binder) or inactive (current collectors, separators and packaging materials). These interfaces often dictate whether the battery chemistry could work according to the designed electrochemical pathways, at what rate (power density), or how reversible (cycle-life).

Molecular Universe (MU) is the world’s first AI-platform that aims to address the acceleration of battery materials discoveries. Constructed upon property databases of astronomical scale (108), agentic systems operating with various large-language and machine-learning models, as well as all-inclusive literature databases on the scale of 107, MU enables us to explore the universe of all small organic molecules that has been impossible to reach just a few years ago.

This talk will cover how MU was conceived, its underneath logic and framework as well as its successes in real battery world.

Bio: Kang Xu is an MRS Fellow, ECS Fellow, ARL Fellow (emeritus), and currently the Chief Technology Officer of SES AI Corp based in Boston, MA. He has been conducting electrolytes and interphasial chemistry research for the past 38 years, published 350+ papers, wrote/edited 5 books/chapters, and obtained 20+ US Patents, with total citation of 87,000+ and an h-index of 142. He is a Clarivate’s highly-cited author since 2018, and one of the top 2% most influential researchers in the Stanford Database.  

Abstract: The speaker will introduce the newly formed AI(X) Hub at The Ohio State University, a collaborative research initiative dedicated to advancing artificial intelligence and harnessing its transformative potential to address complex challenges, accelerate discovery, and create meaningful societal impact. The AI(X) Hub brings together faculty, students, and industry to advance AI-enabled research across engineering, health, science, business, and the arts. From discovering new materials and energizing manufacturing processes to developing radical new treatments in medicine, the Hub is committed to shaping the future of the AI revolution, driving economic growth, and preparing the next generation of leaders in this rapidly evolving field.

Within this broader ecosystem, the speaker will also describe the NSF AI-EDGE Institute, which focuses on the co-design of AI and networking to enable intelligent edge systems. As sensing, experimentation, and manufacturing processes become increasingly distributed, AI-EDGE develops networked intelligence architectures that support real-time adaptation, resilience under uncertainty, efficient, privacy-preserving coordination across devices, and enables low-cost training and inference of powerful machine learning models. These capabilities are relevant to applications in smart manufacturing, materials characterization, healthcare, and robotics where low-cost inference and training is required, and where distributed sensors and autonomous agents must operate under tight communication and latency constraints.

Abstract: The speaker will share Ohio State’s approach to AI Fluency, accelerating student entrepreneurship, and advancing interdisciplinary learning in technology.

Bio: Shereen Agrawal is the associate vice president of student innovation and entrepreneurship at Ohio State University and the inaugural executive director of the Center for Software Innovation. She is a seasoned technology leader with experience ranging from early-stage startups to large, publicly traded companies. Throughout her career, she has gained experience launching new technology, leading go-to-market initiatives, forging networks across global technology communities and developing highperforming teams.

She served as vice president of business development at Columbus-based Root Insurance, where she developed innovative partnerships and products supporting the company’s growth through to IPO. At Twitter and Cloudera, she supported the growth of an ecosystem of software partnerships to develop new customer experiences and grow revenue. As part of the finance team at Intel,
Shereen focused on developing a competitive advantage in new mobile
technology.

Shereen earned her MBA from Harvard Business School and her bachelor’s
degree in Business Administration from the University of Michigan.

Abstract: Delivering a materials solution into common use has historically been an undertaking measured in terms of careers. Traditional materials maturation relies on a linear discovery-development-deployment pipeline, which, though generally robust, is painfully slow and inefficient. The advent of computation is now rapidly accelerating each aspect of this pipeline: novel alloys can be discovered and digitally tested in seconds, automated laboratories can characterize material samples with high throughput, and roboticized facilities can manufacture components at speed. Yet delivering materials solutions for real-world problems remains a slow process. In this talk, we will discuss the challenges that continue to impede accelerating materials solutions. We will showcase historical problems and successes in materials development, tie this history to current research on accelerating pieces of the materials pipeline, and finally highlight critical challenge areas that must be addressed to lead-turn the future of digital materials.

Bio: Sean P. Donegan, a member of the scientific and professional cadre of senior executives, is the Senior Scientist for Digital Advanced Materials, Air Force Research Laboratory, Wright-Patterson Air Force Base, Ohio. He serves as the principal scientific authority and independent researcher in accelerating advanced materials development, integrating computational modeling, high-performance computing, and AI/ML to transition digital-centric materials solutions from materials discovery through deployment for the Department of the Air Force. 

Abstract: We present an update on computational digital twin being built by the Institute for Model-Based Qualification & Certification of Additive Manufacturing (IMQCAM) under NASA support. Many of the component models for microstructure development in 3D printing and micro-mechanical response leading to prediction of fatigue are being both developed and calibrated. Crucially, with upwards of a dozen major components, the multi-institution team is devoting substantial effort not merely to data curation but also to data transfer which involves significant learning and effort on both sides of the exchange. Uncertainty quantification is being applied to both process and micromechanical models. The need to predict microstructure across a wide range of process parameter values motivates the use of reduced order models calibrated by physics-based models at the grain scale, e.g., Potts, which are themselves abstractions of models at the dendrite level, e.g., cellular automata.

Bio: Rollett earned an MA in Metallurgy & Materials Science in 1980 from Cambridge and a PhD in Materials Engineering from Drexel in 1987. He has been a member of the Dept. of Materials Science & Engineering at Carnegie Mellon University since 1995 which included five years as Department Head at CMU. He is a University Professor and the Co-Director of the NextManufacturing Center on additive manufacturing. Previously, he worked at the Los Alamos National Laboratory with five years as a Group Leader then Deputy Division Director.  He has been a Fellow of ASM since 1996, Fellow of the Institute of Physics (UK) since 2004 and Fellow of TMS since 2011.  He received the Cyril Stanley Smith Award from TMS in 2014, was elected as Member of Honor by the French Metallurgical Society in 2015 and then became the US Steel Professor of Metallurgical Engineering and Materials Science in 2017. He received the Cyril Stanley Smith Award from the International Conference on Recrystallization and Grain Growth in 2019 and the Hans Bunge Award from the Intl. Conf. on Textures of Materials (ICOTOM) in 2024. He was an International Francqui Professor (Belgium) in 2022 and received the ASM Gold Medal in 2024.  His research focuses on processing-microstructure-properties relationships with interests in additive manufacturing, the measurement and prediction of microstructural evolution, the relationship between microstructure and properties, especially three-dimensional effects, texture & anisotropy and the use of synchrotron x-rays. 

Abstract: When automating a process, roboticists define the actions to achieve a goal and how to execute those actions. While advances in deep learning have helped execute more complicated actions, sequencing these actions for a particular task is still done manually. Task planning is bottlenecked by this need for manual modeling, which requires dual expertise in process knowledge and formal logic languages. We leverage the current capabilities of LLMs to model planning problems using the Planning Domain Definition Language (PDDL), without relying on pre-written examples or PDDL experts-in-the-loop. We present a preliminary approach that uses evolutionary generation to remove errors in the PDDL representation. To ensure validity, we iteratively identify errors in the generations and reprompt the LLM with general, templated feedback for cross-domain application. Our approach generated models without syntax errors in multiple different domains, bypassing the need for tedious human-in-the-loop debugging or prompt engineering. This demonstrates a potential pathway for more accessible and dynamically defined task planning models for robotics applications.

Bio: Brennan Swick is a National Research Council Postdoctoral Fellow at the Air Force Research Laboratory. His current research focuses on increasing the programming flexibility of autonomous systems by making high-level planning representations more accessible through neuro-symbolic approaches.

Abstract: Physics-based simulation has long been a key tool for understanding thermomechanical processes, particularly in relation to residual stress. However, these simulations are often computationally intensive, making them unsuitable for supporting real-time process planning on the shop floor. In welding, where the input variation may require a revised plan, the physics-based simulation is not able to support shop floor updates on the fly.

Recent advancements in machine learning (ML) and artificial intelligence (AI) provide a potential solution by enabling real-time analysis of assembly variations and dynamic updates to welding plans. These technologies mimic the capabilities of computationally intensive simulations while offering faster execution at production speeds. This results in improved operational efficiency and optimized fabricated products. 

This talk explores ongoing efforts to create hybrid physics-based simulation - advanced ML frameworks. This system dynamically updates welding plans to address deviations in real-world conditions, enhancing the value of simulations during production and improving as-fabricated quality. Additionally, this approach opens new possibilities for extending these concepts to other manufacturing processes and more closely linking robotic system behavior to desired manufacturing outcomes.

Bio: At Southwest Research Institute, Matt Robinson is the Program Manager for the ROSIndustrial Consortium – Americas, an industry-driven open source program bringing advanced manufacturing solutions for the industrial robotics community. When not supporting open source for industry Matt supports and manages numerous robotics projects within the organization. Prior to his current role, Matt was team leader for Caterpillar’s Manufacturing Technology Automation Research where he led development and deployment of automation tools to improve the performance and productivity of Caterpillar manufacturing facilities around the globe. Matt also led manufacturing value stream design initiatives that led to the deployment of over 50 robotic/automated manufacturing systems around the world. Matt holds a Master’s Degree in Welding Engineering from Ohio State University.

Abstract: Data-driven Koopman-theoretic approaches have proven effective in output prediction, state estimation, and control of nonlinear dynamical systems ubiquitous in autonomous manufacturing. For control-affine systems, the Koopman generator's affine input dependence enables finite-dimensional bilinear approximations. However, a significant challenge in constructing Koopman generators for actuated systems is its reliance on appropriate basis functions or observables, with no unified framework for their selection. Real-world applications often involve noisy, partially observed states, requiring Koopman observables to capture system behavior from input-output data. The challenge of identifying Koopman observables under partial observation is well studied, e.g., time-delayed observables offering viable solutions. However, the presence of actuation reduces the efficacy of these methods. To address this limitation, a recent data-driven method leverages a learned Koopman generator-based bilinear surrogate model with linear reconstruction, demonstrating promise for actuated nonlinear system identification. Further study is needed to assess its effectiveness in complex, partially observed nonlinear systems with actuation and sensitivity to initialization. To address this, we model the dynamics of a control-affine nonlinear system as a bilinear Hidden Markov Model (HMM) defined via Koopman generators with a nonlinear observation map (decoder) modeled using a multilayer perceptron (MLP). The parameters of the HMM and decoder are learned from noisy output data using a neural expectation maximization (EM) approach. In the EM method, the E-step employs an extended Kalman filter and smoother, while the M-step utilizes a least-squares approximation of the Koopman generators, combined with a gradient-based optimization for the decoder parameters. In addition, we present a model-predictive control (MPC)-based output regulation method using the learned HMM as a predictive model. We demonstrate the performance of our method on three nonlinear systems: (1) an actuated polynomial system with a slow manifold and partial observation, (2) a forced Duffing oscillator with partial observation, and (3) an unforced Kuramoto-Sivashinsky equation with noisy observation.

Authors: Santosh Rajkumar, Sriram Narayanan , Samuel Otto, and Debdipta Goswami

Bio: Debdipta Goswami joined the Department of Mechanical and Aerospace Engineering, the Ohio State University, in 2022 as an assistant professor. He received his Ph.D. degree in Electrical and Computer Engineering from the University of Maryland in 2020 under the supervision of Prof. Derek A. Paley. Between 2020 and 2022, he has worked as a postdoctoral research associate at the Department of Mechanical and Aerospace Engineering in Princeton University where he worked with Prof. Clancy Rowley. His research interests lie at the intersection of control systems and machine learning with a focus on motion planning and agile control of aerial robots.

He has worked on data-driven discovery and control of dynamical systems using operatortheoretic methods and reservoir computers. His current research focuses on the structured learning of control systems from data with guaranteed performance and simultaneous learning and control of dynamical systems. He also works on model predictive control and motion planning for unmanned aerial vehicles.

Abstract: Yb-doped laser sources offer a robust high-average power scalable platform which have found diverse applications both in scientific and industrial settings. By employing nonlinear post-compression techniques, one can obtain a tunable laser source with durations ranging from many to few optical cycles. Such flexibility allows them to support high-order harmonic generation-based spectroscopic techniques with varying demands over the time and energy resolution. This talk will describe how state-of the-art compression techniques are applied to Yb-based laser systems to obtain tunable high-repetition rate sources for time-resolved spectroscopies, and how the NeXUS laser systems take advantage of them to support our different experimental beam-lines. 

Bio: Laser scientist at NSF-NeXUS, focus on laser source development and maintenance. PhD at the University of Central Florida, research on compression techniques for high-average power laser systems. Postdoc at University of California Berkeley, research on attosecond pump-probe spectroscopy of molecular systems

Abstract: Laser powder bed fusion (LPBF) has experienced exponential growth in interest, and machine technology has improved in like manner.  The industry is reaching ever-closer to the boundaries of what is technically possible with focused CW lasers using Gaussian beam shapes, but the technical challenges provided by key customers in the aerospace industry grow more difficult with every success.  Much of the LPBF industry is unfamiliar with fs laser technology and how it could be used to augment LPBF processes.  This talk details what work has already been done in this area, how fs lasers may be integrated into LPBF systems in the future, and assesses the improvements LPBF systems could see as short and ultrashort pulsed lasers are adopted in the future. 

Bio: Dr. Middendorf has been developing his own Laser powder bed fusion (LPBF) machines since July 2014 and currently leads the additive manufacturing division at the Center for Design and Manufacturing Excellence at The Ohio State University.  John was the technical leader behind the development of the Open Additive (OA) brand, selling the PANDA LPBF machines and AMSENSE sensor suite, and in his past has develop several other innovations such as Selective Laser Ablation & Melting (SLAM) and Graded Alloy Processing (GAP). 

Abstract: Two-dimensional (2D) transition metal dichalcogenides (TMDs) are an emerging materials platform for electronics, photonics, energy conversion, and quantum technologies. Because their electronic and optical properties are highly sensitive to thickness, lateral size, and defect landscape, scalable synthesis with control over both morphology and defects is essential.

We first develop alkali-metal-mediated CVD to grow large-area, epitaxial monolayer TMD flakes and films with improved uniformity and device-relevant performance, and implement a clean detachment-based transfer that preserves crystalline integrity for assembling twisted van der Waals heterostructures. We then synthesize monolayer/bilayer TMD nanoribbons via Ni nanoparticle–enabled VLS tip growth, where ribbon width is set by nanoparticle diameter, revealing confinement- and strain-driven functionalities such as width-dependent Coulomb blockade in MoS2 nanoribbons (<20 nm) up to 80 K and high-purity single-photon emission from WSe2nanoribbons (up to 98% at 120 K).

Beyond dimensional control, I will present defect engineering in monolayer WS2 using alkalihalide-assisted CVD to introduce dense point defects and strong defect-bound exciton emission, and use spatially resolved ultrafast pump–probe microscopy to directly track defect-bound/free exciton populations, uncovering sub-ps defect trapping and ultrafast interconversion dynamics, including efficient up-conversion into free excitons under sub-resonant pumping.

Looking forward, these synthesis and spectroscopy capabilities open pathways to wafer-scale 2D electronics and catalysis, defect-tailored excitonic energy harvesting, and on-chip quantum light sources and hybrid heterostructures for scalable quantum communications.

References
1. ACS Nano 14, 6570 (2020).
2. Nano Lett. 19, 8118 (2019).
3. ACS Catal. 11, 12159 (2021).
4. Sci. Adv. 7, eabk1892 (2021).
5. Nat. Commun. 15, 10080 (2024).
6. Nano Lett. 25, 17475 (2025).
7. Phys. Rev. B 111, 075410 (2025
8. ACS Nano 20, 2904 (2026).

Bio: Xufan Li is a principal scientist at Honda Research Institute USA, Inc. He earned his Ph.D. from the University of Georgia in 2013, and then carried out postdoctoral research at Oak Ridge National Laboratory from 2013 to 2017. He joined Honda Research Institute in 2017, where his work focuses on controllable synthesis and property engineering of low-dimensional twodimensional transition metal dichalcogenides (TMDs), including epitaxial monolayer films, widthconfined nanoribbons, and defect-engineered monolayers. His research combines materials growth with device- and spectroscopy-driven studies, such as quantum transport in nanoribbons, strainenabled quantum emitters, and ultrafast pump–probe measurements of exciton dynamics, to advance applications in electronics, energy conversion, and quantum photonics.

Abstract: Lasers are ever expanding in applications in everyday life devices (screens of your cell-phones and other handheld devices are processed with lasers, for example), materials and manufacturing industry and medical device industry. Utrafast lasers, with their ultrafast heating and cooling rates (a trillion Kelvins per second ) allow capabilities that are beyond the reach of continuous wave (CW) or micro to nanosecond lasers. Ultrafast laser irradiation The interaction is initiated by strong laser field excitation of electrons, which are initially bound in a many body system.  The excited electrons then couple energy into the lattice by multiple pathways and irreversible surface modification occurs due to (non-thermal) melting and vaporization. It is also exciting from an application point of view, as femtosecond laser pulses can produce nm-scale features in metals and non-metals due to extreme spatio-temporal localization of pulse energy preventing heat diffusion into the surrounding volume. However, to utilize their full potential, fundamental understanding of the various stages of the process is necessary. We will present a comprehensive computational model for this process, and also demonstrate some applications via experimentation.

Bio: Prof. Chowdhury is a leading expert in the field of short pulse lasers and laser damage, ultra-intense and high energy density laser matter interaction. He led the design and construction of the 400 TW SCARLET laser system at the OSU High Energy Density Physics (HEDP) Laboratory, which was completed in 2012. His research spans over twelve orders of magnitude in laser intensity (1010 – 1022 Wcm-2 and beyond). At the ‘lower’ intensities (1010 – 1015 Wcm-2), he studies femtoseconed laser matter interaction near material damage threshold, which has concentrated on how laser fundamental damage mechanisms evolve as a function of various laser and material parameters. His ongoing computational efforts concentrate on modeling ultrafast laser damage using strong field Keldysh ionization while fully incorporating plasma and particle dynamics using particle-in-cell (PIC) HPC infrastructure to show that traditional models like Two Temperature Model (TTM) developed for near IR laser solid interaction may not be adequate, and new wavelength scaled paradigm may be necessary to explain intense laser solid interaction at longer wavelengths.

Abstract: To reduce batter manufacturing costs by 30% will require the elimination of solvents and drying during processing.  This presentation will discuss the differences between solution processed and dry processed battery electrodes with an emphasis on the fundamental transformations in polymers, pore structures, and tortuosity.   

Bio:

Abstract: Photonic integrated circuits (PICs) based on GaSb with monolithically integrated active and passive optoelectronic components that operate in the short- and mid-wave infrared wavelength regime are currently of significant research interest due to a wide range of emerging applications. Realizing complex PICs often requires spatially selective control of electrical and optical properties of materials. Proton implantation on GaSb helps electrically isolate adjacent active devices on a PIC by the selective introduction of ion-induced damage. Subsequent annealing minimizes the implantation-induced optical absorption while retaining a high electrical resistivity. This talk presents the recent progress made in this research. Furthermore, passive waveguides of a PIC, responsible for light propagation from active components, must be transparent to the operating wavelength (have a bandgap larger than that of the active components) so that optical power loss due to absorption during transit is minimum. Two promising approaches to this end are quantum well intermixing induced by ion implantation and impurity-free vacancy disordering. The recent results related to induced blueshifts in the photoluminescence spectrum of quantum well materials will also be discussed in this talk.  

Bio: Dr. Shamsul Arafin is an Associate Professor in the Electrical and Computer Engineering Department at The Ohio State University. Prior to joining OSU, he worked as a Project Scientist in the Department of Electrical and Computer Engineering at the University of California at Santa Barbara (UCSB), USA. Dr. Arafin has 15+ years of experience in design and fabrication of widely tunable semiconductor lasers and photonic integrated circuits. Till now, he has authored or co-authored over 165 publications in the areas of growth of III-V materials, optoelectronic devices and quantum photonics. He is a senior member of the IEEE, IEEE Photonics Society, SPIE and OSA.  

Abstract: The monolithic, epitaxial integration of III-V compound semiconductor materials and devices with the ubiquitous Si microelectronics platform has been a central goal across the optoelectronics space for decades. In the case of infrared photodetectors, Si integration offers the promise of not only the convenience of direct ROIC integration, but also substantially larger production areas (for large-scale arrays) at significantly lower materials costs. However, accomplishing the requisite heteroepitaxial integration with sufficiently high materials quality, which direct growth is unlikely to provide, at sufficiently low total thickness to avoid warpage and cracking, below that typically enabled by conventional graded buffers, is far from trivial. To this end, recent work has suggested an alternative, hybrid approach that, which we have adapted to successfully yield MOCVD-grown GaAs-on-Si virtual substrates with threading dislocation densities of ≤ 4×10⁶ cm^-2 at a total III-V thickness under 2 µm. We are now working to push the integration endpoint out to larger lattice constants, GaSb and beyond, to support high-quality growth of antimonide-based infrared materials and devices. This talk will provide an overview of the underlying metamorphic platform development, and will cover ongoing work to adapt it for antimonide applications. 

Abstract: In this talk, I will first present the device design of a vertical GaN-on-GaN PN power diode using a double field plate structure and guard-rings for electrical field management to achieve >10 kV breakdown voltage. Experimentally, the fabricated diode with a ~74 μm thick drift layer and Nd-Na concentration of 1×1015 cm-3 demonstrates a breakdown voltage of 11.45 kV on a bulk GaN substrate. The device has an on-resistance of 10.8 mΩ·cm2 and a Baliga figure of merit of 12 GW/cm2. Compared to the theoretical breakdown voltage of this device design, the fabricated device has an overall breakdown efficiency >90%. Then I will discuss the correlation between the device performance and material properties. The critical field and electron mobility dependences on the doping density in GaN are established. With the doping dependences, specific resistance Ron versus breakdown voltage is mapped for comparison of GaN and SiC on Baliga figure of merit at different doping concentrations, which gives the design strategies and space for GaN power devices. The prediction suggests that GaN power devices have 4~5 times higher BFOM than SiC power devices eat a high blocking voltage rating. Finally, an outlook is given on the challenges of developing GaN power devices with blocking voltage of 20 kV and the needs for accurate prediction and design of ultrawide bandgap semiconductors.

Bio: Wu Lu is a professor at Department of Electrical and Computer Engineering, The Ohio State University. His current research interests focus on wide bandgap semiconductor devices, nanofabrication, chemical and biological sensors, and nanobiotechnology.