Search NASA⌕ Search

SEARCH · Search NASA

Results for “machine-learned potentials”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 73 records · Page 4
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

HydraGNN v5.0

HydraGNN v5.0 expands the code base into a more portable, scalable, and flexible framework for scientific graph learning, with particular strength in atomistic machine-learning interatomic potentials and large-scale distributed training. The release adds Fully Sharded Data Parallel (FSDP) support alongside existing DDP and DeepSpeed paths, including FSDP-aware checkpointing and optimizer integration, and introduces a configurable multi-precision training workflow supporting FP32, BF16, and FP64 across GPUs and Intel XPUs. For atomistic modeling, HydraGNN v5.0 strengthens its MLIP capabilities through dynamic graph construction at every forward pass, energy-conserving force prediction via automatic differentiation, and per-atom energy loss formulations, while extending EGNN models to properly handle periodic boundary conditions. The release also broadens model expressiveness through graph-level attribute conditioning, adds new multi-task and model-parallel extensions such as MACE support and encoder/decoder branch optimization, and expands application coverage with integrated examples for datasets including OC25, Nabla2-DFT, QCML, Open Polymers 2026, and OPF. In parallel, HydraGNN v5.0 improves production readiness through performance optimizations for large-scale runs, stratified sampling and linear-regression preprocessing utilities, and tested installation scripts for DOE supercomputers including Frontier, Aurora, Perlmutter, and Andes. Overall, the release advances HydraGNN as a robust software platform for scalable graph neural networks across materials science, chemistry, and scientific machine learning workflows

Lupo Pasini, Massimiliano [Oak Ridge National Labo↗
DOE OSTI2026

matsim-agents v1.0

matsim-agents is a multi-agent AI framework for atomistic materials simulation and discovery. It orchestrates large language models (LLMs), machine-learned interatomic potentials (MLIPs), and DFT codes into a single agentic loop running on laptops and DOE leadership-class supercomputers. MULTI-AGENT ORCHESTRATION A LangGraph state machine with three nodes: a Planner that converts a natural-language research objective into structured tasks; an Executor that dispatches atomistic tools and loops until the queue is empty; and an Analyst that summarizes results into a human-readable report. State is checkpointed after every step and human-in-the-loop gates can be inserted at any edge. HYPOTHESIS-DRIVEN DISCOVERY CHAT An interactive REPL (matsim-agents chat) that couples LLM dialogue with atomistic simulation. Chemical formulas are automatically detected in conversation turns and trigger a full crystal-phase exploration: structure generation → relaxation → stability scoring → result injection back into the conversation, creating a closed hypothesis-refinement loop. CRYSTAL PHASE ENUMERATION Given a composition, the phase explorer enumerates prototypes by stoichiometry: elemental (fcc/bcc/hcp/sc/diamond), binary 1:1 (rocksalt/CsCl/zincblende/ wurtzite/fluorite/rutile), ternary 1:1:3 (cubic perovskite), ternary 1:2:4 (perovskite + spinel), quaternary 1:1:2:6 (Fm-3m double perovskite). 2-D prototypes (graphene, h-BN, MoS2 2H/1T) and multilayer stacking are also supported via --include-2d and --num-layers. SUPERCELL GENERATION AND SITE DECORATION Auto-tiling to a minimum atom count (--min-atoms), explicit NxNxN tiling (--supercell), symmetry-distinct site decorations (--n-orderings), and isotropic lattice-scale sweeps (--lattice-scales) for volume bracketing. MLFF RELAXATION AND STABILITY SCORING HydraGNN (multi-headed GNN) drives structure relaxation via ASE with FIRE, BFGS, or BFGSLineSearch. Stability output: delta-E/atom ranking across phases and a max-residual-force dynamical-stability proxy. Other MLIPs (MACE, NequIP, Orb) can be plugged in through the same interface. DFT BACKENDS Quantum ESPRESSO pw.x and VASP 6.6 are first-class labellers. Both have validated GPU builds and SLURM/PBS launchers for three DOE platforms: Frontier (AMD MI250X, ROCm), Aurora (Intel PVC, oneAPI), Perlmutter (NVIDIA A100, CUDA). QE produces ~100 binaries (pw.x, ph.x, epw.x, ...). VASP supports scf, relax, vc-relax, and vc-relax-shape run types. ACTIVE-LEARNING LOOP matsim-agents al run CONFIG.yaml drives an iterative HydraGNN-DFT loop: MD generates candidates → ensemble/MC-dropout uncertainty selects the most informative → DFT labels them in parallel inside one allocation → dataset grows → HydraGNN retrains → repeat. DFT backend is a single YAML toggle (dft.backend: vasp | qe). LLM-generated seed structures are supported (no curated POSCAR library needed). Config uses ${VAR}, ${VAR:-default}, ${VAR:?msg} shell-style substitution for cross-user/cross-site portability. LLM BACKENDS Ollama (local, default), vLLM (HPC multi-GPU serving), OpenAI, Anthropic, HuggingFace Transformers+Accelerate. Selected at runtime via flag or env var with no code changes. HPC PORTABILITY Same Python entry points run on Frontier (ROCm 7.2), Aurora (oneAPI), and Perlmutter (CUDA 12). DFT and ML stacks are never co-loaded in the same shell; they couple through the scheduler and filesystem. Advanced multi-node launchers (serve, discovery-chat, single-relaxation, active-learning, QE warm-start) are provided for all three platforms. CODABENCH COMPETITION BUNDLE A self-contained benchmark: 159 atomistic test structures across 11 material classes, 5 tasks (formation energy, forces, ML relaxation, AI-DFT relaxation, phase stability ranking), public/private leaderboard split (30/70), and four ready-to-run baselines: MACE-MP-0, HydraGNN, UMA, AllScAIP.

Lupo Pasini, Massimiliano [Oak Ridge National Labo↗
DOE OSTI2026

User Manual - HydraGNN v5.0: Distributed Implementation of Multi-Tasking Graph Neural Networks

This document serves as the user manual for HydraGNN v5.0, a scalable graph neural network (GNN) architecture for simultaneous prediction of multiple target properties using multi-task learning (MTL). This version of HydraGNN has been developed primarily to support the development, training, and deployment of predictive graph-based deep learning (DL) models for atomistic materials modeling. HydraGNN is templated over 13 message-passing policies, including invariant models (GIN, PNA, PNAPlus, GAT, MFC, CGCNN, SAGE, SchNet, DimeNet) and equivariant models (EGNN, PNAEq, PAINN, MACE), and supports distributed training via distributed data parallelism (DDP), DeepSpeed, and Fully Sharded Data Parallelism (FSDP) on leadership-class supercomputers. Although HydraGNN can be applied to problems beyond atomistic materials modeling, its current use is confined to homogeneous graphs. Additional capabilities include machine-learned interatomic potentials with energy-conserving forces, General, Powerful, and Scalable Graph Transformer (GraphGPS) global attention, periodic boundary conditions, hyperparameter optimization, mixed-precision training, and uncertainty quantification.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Actinium–DOTA coordination in water from hybrid ML/MM: Structure, free energies, and water-exchange pathways

Quantitative simulation of trivalent ƒ-block chelates in water remains challenging because bonded and non-bonded force-field models make different approximations for coordination structure, exchange dynamics, and ion–ligand interactions in highly charged systems. Here, we develop a hybrid machine-learning/molecular-mechanics (ML/MM) framework for Ac 3+ –DOTA in explicit solvent by training an E(3)-equivariant neural network potential (MACELES) on mechanically embedded QM/MM data for Ac aquo and Ac–DOTA species and coupling it to NAMD 2.14 with particle-mesh Ewald electrostatics. Nanosecond ML/MM trajectories remain numerically stable and preserve chelate integrity, yielding a compact DOTA inner shell with an inner-sphere water coordination number of CN Ac,O w ≈ 1.7 arising from a dynamic equilibrium between one- and two-water states (37.5% and 59.9% of frames; three waters 2.5%). A 5 ns potential of mean force shows two low-lying basins at CN Ac,O w ≈ 1 and CN Ac,O w ≈ 2. DFT end-state free energies are consistent with the ML/MM profile, and DFT minimum-energy paths provide a qualitative electronic-structure reference for the observed basin connectivity. State-resolved kinetics reveal picosecond water-exchange pathways that couple hydration changes to transient DOTA arm fluctuations, and training-set comparisons show that temperature-matched Ac–DOTA data optimize energy/force accuracy while more diverse solvated data improve charge prediction. Overall, the present hybrid ML/MM model provides a practical description of Ac 3+ –DOTA hydration thermodynamics and short-time exchange behavior in explicit water at MD-like cost.

Actinium↗
DOE OSTI2025

A foundation model for atomistic materials chemistry

Atomistic simulations of matter, especially those that leverage first-principles (ab initio) electronic structure theory, provide a microscopic view of the world, underpinning much of our understanding of chemistry and materials science. Over the last decade or so, machine-learned force fields have transformed atomistic modeling by enabling simulations of ab initio quality over unprecedented time and length scales. However, early machine-learning (ML) force fields have largely been limited by (i) the substantial computational and human effort required to develop and validate potentials for each particular system of interest and (ii) a general lack of transferability from one chemical system to the next. Here, we show that it is possible to create a general-purpose atomistic ML model, trained on a public dataset of moderate size, that is capable of running stable molecular dynamics for a wide range of molecules and materials. We demonstrate the power of the MACE-MP-0 model-and its qualitative and at times quantitative accuracy-on a diverse set of problems in the physical sciences, including properties of solids, liquids, gases, chemical reactions, interfaces, and even the dynamics of a small protein. The model can be applied out of the box as a starting or "foundation" model for any atomistic system of interest and, when desired, can be fine-tuned on just a handful of application-specific data points to reach ab initio accuracy. Establishing that a stable force-field model can cover almost all materials changes atomistic modeling in a fundamental way: experienced users obtain reliable results much faster, and beginners face a lower barrier to entry. Foundation models thus represent a step toward democratizing the revolution in atomic-scale modeling that has been brought about by ML force fields.

Batatia, Ilyes↗
DOE OSTI2025

Deep-learning atomistic semi-empirical pseudopotential model for nanomaterials

The semi-empirical pseudopotential method (SEPM) has been widely applied to provide computational insights into the electronic structure, photophysics, and charge carrier dynamics of nanoscale materials. We present “DeepPseudopot”, a machine-learned atomistic pseudopotential model that extends the SEPM framework by combining a flexible neural network representation of the local pseudopotential with parameterized non-local and spin-orbit coupling terms. Trained on bulk quasiparticle band structures and deformation potentials from GW calculations, the model captures many-body and relativistic effects with very high accuracy across diverse semiconducting materials, as illustrated for silicon and group III-V semiconductors. DeepPseudopot’s accuracy, efficiency, and transferability make it well-suited for data-driven in silico design and discovery of novel optoelectronic nanomaterials.

Lin, Kailai [University of California, Berkeley, C↗
DOE OSTI2026

Single-cell chromatin accessibility and cis -regulatory element analyses in plants using the scPlantReg platform

Understanding gene regulation is fundamental to plant improvement, but the lack of plant-specific single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) frameworks and cross-species databases has limited insights into cell-type-specific cellular regulation. Here we present ‘scPlantReg’, an integrated framework and database for plant scATAC-seq data. scPlantReg supports end-to-end analyses from raw data processing to biological interpretation and features ‘scATACtor’, a supervised machine-learning approach that outperforms existing tools for cell-type annotation. We applied scPlantReg to pearl millet to characterize cell-type-specific chromatin accessibility and identify validated activating and repressing accessible chromatin regions (ACRs), revealing WRKY transcription factors as potential regulators of xylem development. Furthermore, we reanalysed scATAC-seq datasets from 8 plant species, spanning 11 tissues and multiple developmental stages, enabling cross-species comparisons. Furthermore, these analyses uncovered conserved regulatory programmes, including AP2/EREBP-associated ACRs linked to cell wall development and cell-type-conserved TFs across grasses. Collectively, scPlantReg provides a general framework and resource for comparative regulatory analysis in plants.

Epigenomics↗
DOE OSTI2025

Treyson Ricks - Intern Showcase Poster

Quinone-based sorbents offer a tunable, energy-efficient route to electrochemical CO2 capture, but systematic guidance for molecular design is lacking. Here, we report a high-throughput computational workflow that combines density functional theory (DFT) screening with machine-learning (ML) modeling to evaluate CO2 binding thermodynamics across several quinone derivatives, spanning benzoquinones, naphthoquinones, and anthraquinones. In addition to using solvents to stabilize the quinone anion and dianion, we studied the effect of ion-pairing on the reduction potentials and the CO2 binding energy. Automated Python scripts handled geometry optimizations and adduct-formation energies on an HPC cluster, reducing manual effort significantly. This integrated platform can uncover structure–property relationships and enables rapid in silico evaluation of untested candidates. We present one example from our workflow to showcase the capability of using quinones with ion-pairing to effectively capture CO2. Our approach paves the way for the rational selection of optimal quinone sorbents and can be extended with experimental thermochemical and kinetic data, alternative redox cycles, and stability assessments to accelerate development of next-generation electrochemical CO2 capture materials.

37 - INORGANIC, ORGANIC, PHYSICAL AND ANALYTICAL C↗
DOE OSTI2025

A Computational Workflow of Elucidating Viral Impact on Mediating Microbial Response to In-situ Experimental Warming: Bridging microbial modeling to carbon and mineral modeling

Viruses are abundant in soils and shape microbial communities in ways that can potentially influence ecosystem processes, yet their contributions to carbon cycling and mineral transformations remain poorly understood. Here we present a multi-phase framework that links virus-host interactions to soil biogeochemistry by combining ecological simulations, genome- and community-scale metabolic modeling, and statistical and machine-learning analyses. We first calibrated microbial abundance profiles under explicit infection scenarios to capture how viral pressure alters community structure, then explored alternative interaction strategies, including kill-the-winner, piggyback-the-winner, and mixed lytic-lysogenic modes, through forward simulations. These ecological shifts were translated into metabolic consequences using exchange fluxes summarized into biologically meaningful categories, while integrated statistical and machine-learning screens elevated subtle but consistent signals. Application of this framework revealed that viral infections shift the balance between organic and inorganic fluxes, redirecting metabolism from diffuse organic transformations toward inorganic pools such as protons and CO 2 , directly linking viral regulation to respiration and soil carbon balance. The roll-up analysis also isolated perturbations in critical mineral ions, including magnesium, manganese, zinc, and copper, which serve as essential enzymatic cofactors. In piggyback-the-winner scenarios, uptake of these ions was strongly suppressed. Contrasting viral strategies produced distinct community structures and metabolic outcomes, from broad suppression under kill-the-winner dynamics to dramatic redistributions under high-lytic and high-gain lysogenic regimes that collapsed vulnerable microbial populations while promoting opportunists. Together, these results provide a tractable path to trace viral perturbations from host abundance shifts to metabolic flux adjustments and ecosystem-scale processes, offering a practical way to include viruses in earth system models.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Neutron-Antineutron Transition Search at DUNE

Neutron-antineutron transition is a baryon number violating process with ΔB=2, providing a unique insight into potential explanations of the baryon asymmetry in our universe, especially in the context of post-sphaleron baryogenesis. Studies have been conducted across various neutron-rich environments, including free neutron sources, neutron stars, and bound neutrons in large underground neutrino detectors, among which large neutrino experiments have historically obtained the most stringent constraints. The forthcoming Deep Underground Neutrino Experiment (DUNE) will offer especially strong prospects for this search, enabled by the high spatial resolution of its liquid argon time projection chamber. This poster presents a recent study on neutron-antineutron transition at DUNE using machine-learning based event classification.

Yu, Kaiwen [UC, Irvine (main)]↗
DOE OSTI2025

Investigation of the Effect of Framework Flexibility on CO 2 Adsorption in SIFSIX-3-Cu Using a Machine-Learned Force Field

Metal–organic frameworks (MOFs) offer promise as selective CO 2 sorbents, but successful MOF sorbent materials need high CO 2 binding affinity and selectivity for CO 2 over water. This work focuses on the use of machine-learned force fields (MLFFs) to model CO 2 adsorption in flexible MOFs, with a focus on SIFSIX-3-Cu, an anion-pillared MOF known for its high CO 2 affinity. A preliminary high-throughput screening of over 900 anion-pillared MOFs was performed using rigid UFF+DDEC6 force fields to predict zero-loading heats of adsorption for CO 2 and H 2 O. SIFSIX-3-Cu was selected for further computational study due to its predicted CO 2 heat of adsorption and experimental relevance. A DeePMD-based MLFF was trained to reproduce DFT (PBE+D3) energies and forces, with an iterative sampling scheme combining molecular dynamics, geometry optimization, random geometric insertion, and NVT Monte Carlo-based configuration generation to capture both attractive and repulsive regions of the potential energy surface. Flexibility of the MOF was explicitly included, contrasting with previous models that approximated the MOF as rigid. Hybrid Monte Carlo/molecular dynamics (MC/MD) simulations with the MLFF produced CO 2 adsorption isotherms in good agreement with experimental data at direct air capture (DAC) pressures (e.g., 40 Pa), in contrast to previous overestimations of CO 2 sorption by models with rigid structures. Bond and angle histogram analysis showed that MOF flexibility increased the variance of fluorine–fluorine diagonal distances at adsorption sites, resulting in a lower predicted sorption for flexible, asymmetric SIFSIX-3-Cu pore geometries compared to the rigid, symmetric DFT-optimized SIFSIX-3-Cu pore geometry. A detailed description of flexibility afforded by the MLFF resulted in an accurately predicted CO 2 uptake (0.88 mmol/g) at low pressure (40 Pa) compared to the experimentally measured value (1.24 mmol/g). In conclusion, these results underscore the importance of including framework flexibility when modeling adsorption phenomena in MOFs, particularly for low-pressure applications.

adsorption↗
DOE OSTI2026

Machine-learning modeling of magnetization dynamics in quasi-equilibrium and driven metallic spin systems

Here, we present a perspective on recent progress in machine-learning (ML) force-field approaches for large-scale Landau–Lifshitz–Gilbert (LLG) simulations of metallic spin systems. Building on a generalization of the Behler–Parrinello (BP) architecture originally developed for quantum molecular dynamics, we develop scalable and transferable ML models that faithfully capture the complex, environment-dependent electron-mediated exchange fields characteristic of itinerant magnets. A central ingredient of this framework is the implementation of symmetry-aware magnetic descriptors based on group-theoretical bispectrum formalisms. Leveraging these ML force fields, LLG simulations faithfully reproduce hallmark non-collinear magnetic orders—such as the 120° and tetrahedral states—on the triangular lattice, and successfully capture the complex spin textures emerging in the mixed-phase states of a square-lattice double-exchange model under thermal quench. We further discuss a generalized potential theory that extends the BP formalism to incorporate both conservative and nonconservative electronic torques, thereby enabling ML models to learn nonequilibrium exchange fields from computationally demanding microscopic approaches such as nonequilibrium Green’s-function techniques. This extension yields quantitatively accurate predictions of voltage-driven domain-wall motion and establishes a foundation for quantum-accurate, multiscale modeling of nonequilibrium spin dynamics and spintronic functionalities.

Descriptors↗
DOE OSTI2025

Geologic hydrogen: From natural occurrences to anthropogenic generation – A review of fundamentals, potential, challenges and prospects

Growing demand for hydrogen is exposing the environmental and economic limits of reforming-based and carbon-managed supply chains, while the scale-up of electrolytic capacity remains capital-constrained. Geologic hydrogen, defined as molecular H₂ generated and stored within the Earth's crust offers a complementary, potentially lower-cost resource, yet exploration is still ad hoc. This review (1) revisits a global inventory of confirmed hydrogen seeps and subsurface occurrences; (2) analyzes the controlling reactions, migration pathways, and trapping conditions governing these occurrences; (3) proposes a process-based geologic hydrogen system concept analogous to, yet distinct from, the petroleum system; and (4) evaluates potential geologic hydrogen systems within the United States as a representative case study. Here, we contrast natural systems powered by serpentinization, mantle degassing or radiolysis with anthropogenic systems that stimulate the same reactions or convert in-situ hydrocarbons. Stable hydrogen accumulations require generation rates that outpace combined physical, chemical and microbial losses; the Bourakébougou field (Mali) exemplifies a self-recharging, free-gas reservoir sustained by meteoric-water serpentinization beneath an efficient caprock. Prospective geologic hydrogen resources are likely to occur in regions where iron-rich lithologies, deep-seated faults, and low-permeability sealing formations coexist. Applying this principle, we highlight three promising hydrogen play types in U.S. geological terrains: ophiolite belts (Appalachian and Californian regions), the Midcontinent Rift and the Lake Superior banded‑iron formations. Multiphysics numerical models and positive-unlabeled machine-learning workflows help to accelerate play screening and de-risk future production; yet, reaction kinetics, stimulation strategies, and full techno-economic and life-cycle assessments remain pivotal knowledge gaps.

Anthropogenic hydrogen generation↗
DOE OSTI2025

Tuning water dissociation at oxide–electrolyte interfaces with electric fields

Understanding how electric fields influence water dissociation at heterogeneous interfaces is crucial for controlling interfacial chemical reactions and advancing next-generation energy technologies. Herein, ab initio–based machine learning simulations show that even small electric field changes can significantly alter the water dissociation fraction at planar TiO 2 –electrolyte interfaces. The resulting free energy difference between undissociated and dissociated interfacial water exhibits a linear dependence on the field change with a slope of 1.97 eÅ, which far exceeds the dissociation-induced dipole change of a water molecule. Employing a machine-learned collective variable to investigate the reaction statistics of thousands of water dissociation/recombination events, we find that small electric field changes exert minor effects on individual reaction energy barriers but significantly influence the populations of local configurations associated with initial states that are most favorable for reactions. These findings elucidate the pronounced impact of electric fields on interfacial water dissociation and reveal a mechanism for electric-field-controlled chemical reactions.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

Rapid Coal-Ash Characterization using Geophysical Methods & Machine Learning

Coal combustion products (CCP) are challenging to delineate in heterogeneous field settings. Conventional methods (test pits, coring, and laboratory analyses) are labor-intensive, slow, invasive, and provide sparse spatial coverage. This study evaluates whether rapid non-invasive geophysical screening methods—induced polarization (IP), magnetic susceptibility, and nuclear magnetic resonance (NMR) —combined with surface colorimetry (RGB_24), can discriminate CCP-soil mixtures and provide reliable estimates of CCP content. Laboratory measurements were collected on five CCP-soil mixtures (series) and modeled using (i) a linear baseline, (ii) a calibrated non-linear (power-mean) model, and (iii) a machine-learning (ML) Random Forest approach, with validation via leave-one-series-out and site-specific tests. Across the five series, individual signals—particularly IP and magnetic susceptibility—were strongly predictive of ash content but were consistently outperformed by combined models. The pooled calibrated non-linear and ML models captured the observed non-linearity and achieved high accuracy and precision, improving on linear fits. Colorimetry showed the weakest direct relationship with ash content for the tested samples but improved performance when included in multi-signal models. At pre-selected 3.5% decision threshold, calibrated and ML approaches yielded near-perfect classification (Matthews correlation coefficient ˜ 1), suggesting strong practical operability for field screening. Additionally, field-analog tests highlighted the role of endmembers—accuracy declined without access to end-member measurements but was largely recovered by collecting a minimal labeled pair for local recalibration. With end members, accuracy remained high. Globally trained models performed well on three operational unknowns; however, series-specific refits provided the most accurate predictions. Overall, these results highlight the potential of combining rapid geophysics and minimal local calibration for improved coal-ash delineation.

Peshtani, Klaudio↗
DOE OSTI2025

Exocortex Network for AI-Augmented Human-Led Scientific Expedition

AI advances in science can be viewed along two main directions with a fluid boundary: enhancing efficiency through automation and smart tools to accelerate tasks that humans can already perform; and enabling exploration into uncharted territories and potentially toward AGI. These advances manifest in the AI cognitive core through the development and explainability of foundation models; in the physical embodiment of instruments and facilities; and in the integrated agency of AI workflows exemplified by the science exocortex. To address the role of humans in this evolving landscape, in this Perspective, we suggest a third direction: the development of personalized agents that form human-centered networks, supporting both efficiency and exploration while ensuring that AI remains aligned with human vision.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Even Higher-Level Synthesis: An Exploration of AI Hardware Accelerators using HLS4ML

With the rise of artificial intelligence, the popularization of deep learning, and a constantly evolving industry, the demand for flexible and efficient tools has never been greater. As algorithms grow more complex, their runtime and energy consumption increase exponentially. Customized hardware accelerators, long used for specific mathematical operations, remain essential for managing modern applications' computational and power demands. Hardware accelerators can speed up complex computations by orders of magnitude, but their manual design and verification processes are often challenging and time-consuming. High-Level Synthesis (HLS) provides a solution by transforming high-level algorithm descriptions, typically written in C++ or SystemC, into synthesizable RTL suitable for hardware implementation. This approach reduces development time for RTL engineers while offering flexibility beyond what traditional handwritten RTL can provide. We extended this capability to the machine-learning domain with the open-source framework hls4ml, which allows neural networks trained in Python frameworks like Tensorflow or PyTorch to be synthesized into efficient hardware representations for the traditional FPGA and ASIC flows. This breakthrough addresses the growing need for reduced design turnaround and easy verification of ML hardware accelerators with low latency and power efficiency constraints. During this tutorial, we will demonstrate how Python complements HLS by simplifying the ML design process, bridging the gap between software and hardware development. Attendees will explore how we translate neural networks modeled in Python into fixed-point C++ models suitable for HLS workflows. We will dive into strategies like Value-Range Analysis and Quantization-Aware Training, which optimize these designs for deployment and evaluate their accuracy, power consumption, and energy efficiency. To exemplify these concepts, experts from Fermilab will share their experiences applying this technology to high-energy physics experiments, where real-time, low-latency processing is critical. Over the years, Fermilab engineers have demonstrated how deep neural networks, optimized for hardware using hls4ml, can meet the stringent requirements of trigger systems at the CERN Large Hadron Collider. These systems rely on rapid decision-making to process immense data volumes while retaining only the most relevant events for further analysis. The application of hls4ml has also been extended to innovative technologies like smart pixel arrays. These smart pixels integrate ML inference capabilities directly into sensor devices, enabling localized data processing at the pixel level. This approach drastically reduces the need to transmit raw data to external processing units, significantly decreasing power consumption and latency. By embedding neural networks within the pixel architecture, the smart pixels can identify and prioritize relevant data in real time, providing a highly efficient solution for edge computing in scenarios such as particle detectors and imaging systems. Fermilab's work highlights the potential of hardware-accelerated ML in scenarios where both speed and power efficiency are mission-critical. Through this tutorial, attendees will gain valuable insights into the challenges and solutions of deploying ML in hardware. Understanding how HLS and hls4ml streamline the development of neural network-based hardware accelerators is fundamental for the industry's future. Participants will learn how these technologies are shaping the future of AI and scientific computing.

Di Guglielmo, Giuseppe [Fermilab]↗
DOE OSTI2026

Crowdsourcing the Frontier: Advancing Hybrid Physics‐ML Climate Simulation via a $\$$50,000 Kaggle Competition

Subgrid machine-learning (machine learning [ML]) parameterizations have the potential to introduce a new generation of climate models that incorporate the effects of higher-resolution physics without incurring the prohibitive computational cost associated with more explicit physics-based simulations. However, important issues, ranging from online instability to inconsistent online performance, have limited their operational use for long-term climate projections. To more rapidly drive progress in solving these issues, domain scientists and ML researchers opened up the offline aspect of this problem to the broader ML and data science community with the release of ClimSim, a NeurIPS Data sets and Benchmarks publication, and an associated Kaggle competition. This paper reports on the downstream results of the Kaggle competition by coupling emulators inspired by the winning teams' architectures to an interactive climate model (including full cloud microphysics, a regime historically prone to online instability) and systematically evaluating their online performance. Our results demonstrate that online stability in the low-resolution real-geography setting is reproducible across multiple diverse architectures, which we consider a key milestone. All tested architectures exhibit strikingly similar offline and online biases, though their responses to architecture-agnostic design choices (e.g., expanding the list of input variables) can differ significantly. Multiple Kaggle-inspired architectures achieve state-of-the-art results on certain metrics such as zonal mean bias patterns and global Root Mean Squared Error, indicating that crowdsourcing the essence of the offline problem is one path to improving online performance in hybrid physics-AI climate simulation.

Environmental sciences↗