Thrust 1 Projects

SODA ULTRA: Software-Defined Architectures for Ultra-Low-Latency Reasoning

Principal Investigator: A. Tumeo

SODA ULTRA provides AT SCALE projects with the capability to generate and explore domain-specialized architectures to perform ultra-low-latency data processing at the edge in experimental scientific workflows for material synthesis. Following a codesign approach, the project will extend the SODA synthesizer to enable the automatic generation of specialized systems for artificial intelligence (AI) and machine learning (ML) models that current edge platforms and existing tools cannot manage. The resulting specific-purpose accelerators will enable real-time decisions in complex experimental materials synthesis workflows. SODA will leverage and contribute to the open-source hardware design tools community to establish an end-to-end flow for the generation of low-latency domain-specific systems for precision materials synthesis. In the longer term, SODA ULTRA will provide a unique capability enabling the generation of highly specialized hardware systems for autonomous characterization and synthesis platforms with low-latency and continual adaptation on edge.

Toward Integration of Closed-Loop Manufacturing across Scales: Sensing, Signal, and Algorithmic Synergies

Principal Investigators: M. Taheri and R. Rallo

Codesign of AI/ML, computing, and control methodologies across a range of microscopy and deposition platforms will enable precise, low-latency decisions on-the-fly. This project pursues a combination of algorithm development, translation to edge computing, and on-edge decision and control toward a generalizable ML-based platform for rapid decision-making. Using electron microscopy and precision processing in Johns Hopkin’s University’s Materials Characterization and Processing Center as a canvas, a broadly deployable decision framework will be developed on edge for variable modes and spatiotemporal scales. The developments will be translated to toolsets that will be deployed within a future autonomous testbed in Pacific Northwest National Laboratory’s Energy Sciences Center. 

Spectroscopic Probes of Vapor Deposited Two-Dimensional Materials to Enable Predictive Synthesis Control

Principal Investigator: M. Prange

This project will pioneer advances in (non)linear optical microscopy and spectroscopy of transition metal dichalcogenides (TMDs). The goal is to develop a detailed understanding of the relationships between processing parameters and the properties of final products. This will pave the way for developing real-time control over the properties of TMD devices that feature desired defect and concentration gradients. The make-measure-model paradigm will be applied as follows. (1) Make: chemical vapor transport and deposition synthesis of TMD samples reflecting different growth stages and suitable for semi in situ analysis. (2) Measure: identify optical fingerprints across scales. (3) Model: develop simulations of optical spectra that reflect the interaction of defects with the thermal bath of phonons present in growing TMD crystals.

Differentiable Programming for Low-latency Control of Material Synthesis Processes

Principal Investigator: E. King

This project seeks to develop a novel theoretical and algorithmic framework for differentiable digital twins of multiscale physics present in AT SCALE material synthesis projects. This framework will then allow for the design of fast predictive control strategies deployable on edge devices with low-latency constraints. These developments will enable low-latency closed-loop autonomous control of material synthesis processes across ongoing AT SCALE projects through active collaboration. The project will initially focus on two material synthesis process control use cases (epitaxial synthesis and laser processing) that range in data set types and the availability of information for physics-based digital twin development.

Digital Twin Enabled Accelerated Development of Topological Insulator

Principal Investigator: Z. J. Xu

This project will investigate the science behind the fabrication topological insulators into devices. We will identify appropriate materials, handling procedures, chemistry, and lithography while keeping synthesis constant to establish a fabrication workflow based on available Pacific Northwest National Laboratory capabilities. This will enable the fabrication of devices that can be characterized and used to develop a figure of merit based on the process workflow. This information will be fed into a digital twin framework involving a suite of models to identify the relationships between material composition, structure, fabrication parameters, and device performance. This requires integrated multiscale and multiphysics models, their AI surrogates, and experiment/testing data. The digital twin will enable rapid prediction of device performance for given fabrication properties and will eventually include material synthesis properties.

DECENTRALIZED DATA MESH FOR AUTONOMOUS MATERIALS SYNTHESIS 

PIs: S. Allec, N. Tallent

The lack of domain-informed composable data management concepts for materials science is a fundamental barrier to exploiting multi-modal materials data. This project will define and design the necessary data management concepts, realized as general building blocks of data services that compose into larger units, interface with a scientific data mesh, and are quickly customized to meet the varying needs of domain scientists. The work will develop helpful data management services and use them to compose customizable prototype data pipelines for each project. A pilot project will work with two other projects to build a prototype data mesh that can be expanded in future work.

Thrust 2 Projects

AI-Accelerated Materials Design: Closing the Synthesis-Characterization-Device Loop

Principal Investigator: T. Kaspar

This project is developing AI-accelerated analysis of in situ data streams to enable the targeted synthesis of novel oxide-based energy materials with desired structure, chemical stability, and functional properties. Atomically precise discovery and design of functional materials underpins many important emerging technologies, including quantum computing, advanced microelectronics, catalysis, and energy storage. We are employing ML-based methods to interpret in situ thin film deposition data in real time to enable AI-guided feedback control of key synthesis parameters. We are developing an open-source ML analysis toolsuite called RHAAPsody for on-the-fly analysis of in situ reflection high energy electron diffraction (RHEED) images collected during thin film deposition by molecular beam epitaxy. We are collaborating closely with the AT SCALE project “Differentiable Programming for Low-latency Control of Material Synthesis Processes” to implement AI-guided closed-loop autonomous control of the synthesis process. The methodologies developed in this project will be broadly applicable to a wide array of synthesis methods, in situ and ex situ probes, and desired materials.

Dynamic Control of Laser Processing for Nonequilibrium Materials

Principal Investigator: M. Olszta

The realization of predictive capabilities for the accelerated insertion of additive manufacturing parts and laser welding are limited in precision and reliability due to our limited understanding of nano-to-meso-scale determining factors. A critical missing link is at-scale, closed-loop capabilities. An “intelligent approach” that integrates simulation, ML, and related statistical analysis tools could speed up the insertion and reliability process by directly connecting process parameters using in situ observations and resulting microstructures to desired properties. In this project, we will codevelop AI, ML, and control methodologies in concert with electron microscopy and additive manufacturing to provide a unique platform to enable precision manufacturing on-the-fly and at scale.

Automated In Operando Experimental Analysis of Far from Equilibrium Phase Evolution During Additive Manufacturing of Multiple-Principal Element Alloys

Principal Investigator: A. Devaraj

This project seeks to advance cutting-edge in situ experimental techniques and data processing pipelines to enable precise detection of phase evolution in multiple-principal element metal alloys undergoing laser processing. By leveraging these enhanced capabilities, the system will facilitate rapid interpretation of observed data and offer real-time recommendations for adaptive laser process parameters, enabling the formation of optimized microstructures with superior properties. Additionally, the project emphasizes the integration of Python-based data processing methods to analyze in situ synchrotron X-ray imaging and diffraction results with low latency, employing ML models for efficient data analysis and interpretation. These insights will feed into an edge computing framework to deliver optimized process controls, advancing the field of autonomous materials processing and property prediction.

Autonomous Materials Discovery in Nonequilibrium Reaction-Diffusion Systems

Principal Investigator: E. Nakouzi

Non-equilibrium materials synthesis is the modus operandi in natural and biological systems; however, it remains beyond traditional synthesis techniques due to the immense and challenging phase space of parameters that can potentially be controlled. To solve this problem, we will develop an autonomous experimentation capability to explore, control, and optimize nanomaterials synthesis in non-equilibrium environments. Specifically, we will program reaction-diffusion precipitation systems to create multicomponent metal oxide nanomaterials with controlled chemistry, size, morphology, crystallographic phase, and spatial distribution. The vast experimental parameter space will be navigated using an iterative feedback loop informed by real-time spectroscopic data and modeling. We anticipate three major components toward achieving this autonomous experimentation capability: 1) synthesis using reaction-diffusion in liquid-gel systems, 2) real-time characterization of product evolution using spatially resolved Raman and other spectroscopies, and 3) real-time data-driven modeling to benchmark experimental products against a spectral database.

AI-DRIVEN AUTONOMOUS STOICHIOMETRY CONTROL IN PLD THIN FILM SYNTHESIS

PI: L. Wang

Precise stoichiometry control is critical for growing functional materials. This project will develop an AI-driven, autonomous synthesis approach for pulsed laser deposition (PLD) by combining real-time laser ablation–induced plasma diagnostics via optical emission spectroscopy (OES) to monitor the plume before it reaches the substrate and reflection high-energy electron diffraction (RHEED) to monitor film growth on the substrate surface. The case study will address precise lithium stoichiometry control during thin film growth, relevant to energy storage, materials science, and electronics applications. The autonomous synthesis approach will enable precise stoichiometry control and dynamic adjustment of deposition parameters, accelerating the development of advanced materials.

Auto FIM/OAP for Imaging Transition Metal Dichalcogenides (TMDs) Synthesis at the Nanoscale: Towards Enhanced Control of TMDs Fabrication

PIs: S. Lambeets, M. Ziatdinov

Transition metal dichalcogenides (TMDs), such as WSe₂ and MoS₂, are 2D materials with promising electronic and catalytic properties as well as great resistance and mechanical flexibility, which opens them to a wide range of applications. However, synthesizing TMD films with high control remains a delicate process. This project aims to explore the electric field-assisted synthesis of WSe₂ on W surfaces, using a combination of field ion microscopy (FIM) and operando atom probe (OAP). AI-driven control of the FIM/OAP system will be optimized to adjust synthesis parameters (precursor pressure, voltage, temperature) in real time based on OAP experimental measurements. The goal is to understand how electric fields can trigger and influence the TMD synthesis mechanism at the atomic scale, improving the quality and control of TMD films.