SEARCH · Engineering Papers

Results for “hardware accelerators”

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.

SourceResearch connectionFrom yearThrough year

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

At least 883 records · Page 49

All sources NASA NTRS DOE OSTI

DOE OSTI2025

Establishing model credibility for process-microstructure-property relationships in additive manufacturing using exascale computing

Additive Manufacturing (AM) of alloys holds significant promise as a disruptive technology in various industries, yet its adoption is often hindered by challenges in achieving consistent part quality. These issues are primarily due to the complex process-microstructure-property (PSP) relationships inherent to AM. Computational models can greatly aid in understanding these relationships, but their widespread impact and adoption has been limited by a lack of validated, open-source, and computationally efficient PSP modeling frameworks and hardware limitations. Here, this study leverages the ExaAM software suite and data from the AMBench-2018 series of laser powder bed fusion (LPBF) benchmark experiments to perform a comprehensive model assessment, including verification, validation, sensitivity analysis, and uncertainty quantification. The RADICAL-EnTK workflow manager was used to perform an ensemble of heat transport, solidification, and mechanical response simulations on the exascale computer Frontier, considering uncertainties in critical model inputs such as laser spot size and nucleation parameters, and consisting of 125 explicit grain structure simulations and 7875 crystal plasticity simulations. For a selected location within the Inconel 625 AMBench-2018 test artifact, sensitivity analysis and uncertainty quantification were performed using the predicted distributions of grain structure and mechanical properties. Qualitative agreement was found between the predicted grain size and texture and the observed AMBench-2018 microstructure, the mean predicted yield stress was within 5% of the experimental measurement mean, and the mean predicted engineering stress at 5% strain was within 10% of the experimental measurement mean. The insights gained from development and validation of the ExaAM PSP modeling framework will help guide future directions for enhancing the credibility and reliability of PSP models in AM, thereby accelerating the adoption of AM technologies in various industries.

Additive manufacturing↗

DOE OSTI2026

Bridging Cloud and Edge Computing at NREL Using CONNECT: Cloud Optimized Networking for Next-Gen Edge Computing Technologies [Slides]

CONNECT is an innovative on-premise hardware and software solution that integrates edge and cloud computing infrastructure at NREL. Built on the AWS Greengrass middleware and leveraging the MQTT protocol, CONNECT enables real-time data streaming from IoT devices and gateways to both cloud and local services, empowering researchers to rapidly capture, analyze, and act upon edge-generated data while leveraging cloud capabilities. The platform addresses research infrastructure challenges by providing a pre-approved platform which is already configured with the correct networking and cybersecurity baselines thus eliminating procurement delays and enabling on-demand availability. CONNECT's hybrid architecture efficiently manages burstable workloads, allowing research teams to dynamically scale computational capacity, handle peak data loads, and reduce operational bottlenecks. Advanced capabilities include built-in GPU support for executing machine learning models which enables low-latency inference at the edge from models trained in the cloud. This architecture supports real-time analytics and filtering, providing a mechanism to allow only transmitting and processing high-value data. Cloud-based configuration management permits engineers to manage on-premise systems remotely, optimizing operational efficiency. By bridging edge and cloud computing, CONNECT provides NREL researchers with a flexible, scalable platform that accelerates scientific discovery while maintaining robust security and performance standards.

97 MATHEMATICS AND COMPUTING↗

NASA NTRS2006

Methodology for Life Testing of Refractory Metal/Sodium Heat Pipes

The focus of this work was to establish an approach to generate carefully controlled data that can conclusively establish heat pipe operating life with material-fluid combinations capable of extended operation. To accomplish this goal acceleration is required to compress 10 years of operational life into 3 years of laboratory testing through a combination of increased temperature and mass fluence. Specific test series have been identi3ed, based on American Society for Testing and Materials (ASTM) specifications, to investigate long term corrosion rates. The refractory metal selected for demonstration purposes is a Molybdenum-44.5%Rhenium alloy formed by powder metallurgy. The heat pipe makes use of an annular crescent wick design formed by hot isostatic pressing of Molybdenum-Rhenium wire mesh. The heat pipes are filled using vacuum distillation and purity sampling is considered. Testing of these units is round-the-clock with 6-month destructive and non-destructive inspection intervals to identify the onset and level of corrosion. Non-contact techniques are employed for providing power to the evaporator (radio frequency induction heating at I to 5 kW per unit) and calorimetry at the condenser (static gas gap coupled water cooled calorimeter). The planned operating temperature range would extend from 1123 to 1323 K. Accomplishments prior to project cancellation included successful demonstration of the heat pipe wick fabrication technique, establishment of all engineering designs, baselined operational test requirements and procurement/assembly of supporting test hardware systems.

Martin, James J.↗

NASA NTRS2004

Lessons learned about spaceflight and cell biology experiments

Conducting cell biology experiments in microgravity can be among the most technically challenging events in a biologist's life. Conflicting events of spaceflight include waiting to get manifested, delays in manifest schedules, training astronauts to not shake your cultures and to add reagents slowly, as shaking or quick injection can activate signaling cascades and give you erroneous results. It is important to select good hardware that is reliable. Possible conflicting environments in flight include g-force and vibration of launch, exposure of cells to microgravity for extended periods until hardware is turned on, changes in cabin gases and cosmic radiation. One should have an on-board 1-g control centrifuge in order to eliminate environmental differences. Other obstacles include getting your funding in a timely manner (it is not uncommon for two to three years to pass between notification of grant approval for funding and actually getting funded). That said, it is important to note that microgravity research is worthwhile since all terrestrial life evolved in a gravity field and secrets of biological function may only be answered by removing the constant of gravity. Finally, spaceflight experiments are rewarding and worth your effort and patience.

manned↗

DOE OSTI2020

Evaluation of a streaming readout solution for Jefferson Lab experiments

Jefferson Lab (JLab) operates the CEBAF accelerator that provides a high energy (up to 12 GeV) and high current (up to 150 µA) polarized electron beam for fixed target experiments. JLab's mission is to provide substantial progress in understanding the Quantum Chromo Dynamics (QCD), the current theory of the spectrum and the dynamics of hadronic matter. Given the broad scientific program a general-purpose detector, CLAS12, has been designed and constructed. Such a detector requires a sophisticated trigger and current experiments use an on-line FPGA-based system that relies upon custom firmware and electronics both of which are difficult reconfigure from one experiment to the next. To overcome these challenges transitioning to a streaming readout has been proposed since it would allow a more flexible, easier to debug, software trigger to be developed. To facilitate this JLab is gaining experience with streaming readout systems and is actively working to design and test a streaming data acquisition architecture. Significant work has already been done to make existing front-end electronics compatible with operation in streaming mode. Software-wise, much work remains to be done but an existing framework, TriDAS, which was developed for underwater neutrino telescopes can, in principle, be used with the existing JLab hardware. A test stand is already in place in the INDRA DAQ development lab at JLab. This paper will present the development of this TriDAS based test system and it?s operation in-beam using parts of the CLAS12 detector to evaluate streaming readout for a large general purpose detector.

Battaglieri, Marco↗

NASA NTRS2003

Reliable Transport over SpaceWire for James Webb Space Telescope (JWST) Focal Plane Electronics (FPE) Network

NASA's James Webb Space Telescope (JWST) faces difficult technical and budgetary challenges to overcome before it is scheduled launch in 2010. The Integrated Science Instrument Module (ISIM), shares these challenges. The major challenge addressed in this paper is the data network used to collect, process, compresses and store Infrared data. A total of 114 Mbps of raw information must be collected from 19 sources and delivered to the two redundant data processing units across a twenty meter deployed thermally restricted interface. Further data must be transferred to the solid-state recorder and the spacecraft. The JWST detectors are kept at cryogenic temperatures to obtain the sensitivity necessary to measure faint energy sources. The Focal Plane Electronics (FPE) that sample the detector, generate packets from the samples, and transmit these packets to the processing electronics must dissipate little power in order to help keep the detectors at these cold temperatures. Separating the low powered front-end electronics from the higher-powered processing electronics, and using a simple high-speed protocol to transmit the detector data minimize the power dissipation near the detectors. Low Voltage Differential Signaling (LVDS) drivers were considered an obvious choice for physical layer because of their high speed and low power. The mechanical restriction on the number cables across the thermal interface force the Image packets to be concentrated upon two high-speed links. These links connect the many image packet sources, Focal Plane Electronics (FPE), located near the cryogenic detectors to the processing electronics on the spacecraft structure. From 12 to 10,000 seconds of raw data are processed to make up an image, various algorithms integrate the pixel data Loss of commands to configure the detectors as well as the loss of science data itself may cause inefficiency in the use of the telescope that are unacceptable given the high cost of the observatory. This combination of requirements necessitates a redundant, fault tolerant, high- speed, low mass, low power network with a low Bit error Rate(1E-9- 1E-12). The ISIM systems team performed many studies of the various network architectures that meeting these requirements. The architecture selected uses the Spacewire protocol, with the addition of a new transport and network layer added to implement end-to-end reliable transport. The network and reliable transport mechanism must be implemented in hardware because of the high average information rate and the restriction on the ability of the detectors to buffer data due to power and size restrictions. This network and transport mechanism was designed to be compatible with existing Spacewire links and routers so that existing equipment and designs may be leveraged upon. The transport layer specification is being coordinated with European Space Agency (ESA), Spacewire Working Group and the Consultative Committee for Space Data System (CCSDS) PlK Standard Onboard Interface (SOIF) panel, with the intent of developing a standard for reliable transport for Spacewire. Changes to the protocol presented are likely since negotiations are ongoing with these groups. A block of RTL VHDL that implements a multi-port Spacewire router with an external user interface will be developed and integrated with an existing Spacewire Link design. The external user interface will be the local interface that sources and sinks packets onto and off of the network (Figure 3). The external user interface implements the network and transport layer and handles acknowledgements and re-tries of packets for reliable transport over the network. Because the design is written in RTL, it may be ported to any technology but will initially be targeted to the new Actel Accelerator series (AX) part. Each link will run at 160 Mbps and the power will be about 0.165 Watt per link worst case in the Actel AX.

Rakow, Glenn↗

NASA NTRS2011

Space Shuttle Main Engine - The Relentless Pursuit of Improvement

The Space Shuttle Main Engine (SSME) is the only reusable large liquid rocket engine ever developed. The specific impulse delivered by the staged combustion cycle, substantially higher than previous rocket engines, minimized volume and weight for the integrated vehicle. The dual pre-burner configuration permitted precise mixture ratio and thrust control while the fully redundant controller and avionics provided a very high degree of system reliability and health diagnosis. The main engine controller design was the first rocket engine application to incorporate digital processing. The engine was required to operate at a high chamber pressure to minimize engine volume and weight. Power level throttling was required to minimize structural loads on the vehicle early in flight and acceleration levels on the crew late in ascent. Fatigue capability, strength, ease of assembly and disassembly, inspectability, and materials compatibility were all major considerations in achieving a fully reusable design. During the multi-decade program the design evolved substantially using a series of block upgrades. A number of materials and manufacturing challenges were encountered throughout SSME s history. Significant development was required for the final configuration of the high pressure turbopumps. Fracture control was implemented to assess life limits of critical materials and components. Survival in the hydrogen environment required assessment of hydrogen embrittlement. Instrumentation systems were a challenge due to the harsh thermal and dynamic environments within the engine. Extensive inspection procedures were developed to assess the engine components between flights. The Space Shuttle Main Engine achieved a remarkable flight performance record. All flights were successful with only one mission requiring an ascent abort condition, which still resulted in an acceptable orbit and mission. This was achieved in large part via extensive ground testing to fully characterize performance and to establish acceptable life limits. During the program over a million seconds of accumulated test and flight time was achieved. Post flight inspection and assessment was a key part of assuring proper performance of the flight hardware. By the end of the program the predicted reliability had improved by a factor of four. These unique challenges, evolution of the design, and the resulting reliability will be discussed in this paper.

VanHooser, Katherine P.↗

NASA NTRSDate not supplied

Results from Two Full-Scale Hawker 4000 Fuselage Drop Tests

In the spring of 2022, NASA Langley Research Center (LaRC) conducted two full-scale drop tests on a Hawker 4000 partial fuselage structure for the evaluation of composite material response under dynamic impact loading conditions. The specific objectives of the tests were to induce dynamic composite damage initiation and progression and in primary structure and to understand the nature of the failure modes. The tests were conducted via dividing the fuselage hardware into three smaller sections. Dynamic drop tests were conducted on the forward and aft portions of the fuselage under different dynamic loading conditions, while the middle section was reserved for materials testing. Both tests were conducted by dropping the test articles from a height of 14 feet with significant amounts of pitch, which was added to initiate damage in specific regions of the structure. The test articles were ballasted then instrumented with accelerometers at various locations and speckle-coated on both exterior sides with a black and white stochastic pattern for use with three-dimensional digital image correlation (3D-DIC). In addition, the interior belly portion of the forward section test article was also speckle-coated, which was intended to capture deformations on the belly of the test article from the interior at impact. The interior of the aft section test article was not speckle-coated. Prior to testing, bow-tie markers were applied at various locations for use in marker tracking, to measure impact conditions such as vertical impact velocity and pitch. In this report, a partial summary of the data collected from the tests is presented. The data was primarily in the form of accelerometers to measure impact acceleration and 3D-DIC to measure deformation and failure onset and propagation. Post-test inspections revealed the primary damage mechanism was fiber cracking and delamination primarily in and around the area that contacted the impact surface. Acceleration pulse shapesin the ballast locations were trapezoidal in nature and ranged between 10 g and 40 g, depending on the location and test.

full scale testing↗

NASA NTRSDate not supplied

Results from Two Full-Scale Hawker 4000 Fuselage Drop Tests

full scale testing↗

DOE OSTI2026

Science Uses Deployment Operations-Advanced Wireless: Exploring Open Radio Access Network Technologies for Energy Science

Open Radio Access Network is emerging as a solution to the increasing demand for more flexible, cost-effective, and advanced mobile network infrastructures. This evolution is driven by advancements in wireless technologies and the growing complexity of deploying and managing these networks. O-RAN represents a significant shift in wireless technology, building upon the 3rd Generation Partnership Project framework to foster openness, flexibility, and interoperability. By decoupling hardware and software components, Open Radio Access Network enables a multi-vendor ecosystem that encourages innovation and diverse solutions. Open Radio Access Network's potential extends beyond traditional wireless applications, with growing interest in its role in advancing energy systems, particularly in the context of smart grids, microgrids, and the integration of renewable energy sources. While the role of open-wireless technologies in driving energy transformation is increasingly recognized, further exploration is needed. Vendors and utilities are investigating how Open Radio Access Network technologies can optimize energy use cases and improve the performance of 5G and beyond applications. This report outlines efforts under the Science Uses Deployment Operations Advance Wireless project, a collaboration between the National Laboratory of the Rockies' Cybersecurity Research Center, Argonne National Laboratory, Lawrence Berkeley National Laboratory, and the Department of Energy's Energy Science Network research and operations staff. The focus of this project is on due diligence, through testing and evaluation, preparing for the deployment of advanced wireless infrastructure for scientific use cases, with an emphasis on Open Radio Access Network technology, its components, integrations, and its ability to support vertical stack application across the energy sector. Additionally, the report highlights the value cases for utilities, underscoring how adopting open wireless standards can accelerate the evolution of energy systems, foster innovation, and improve the integration of critical energy technologies.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2025

The critical importance of software for HEP

Particle physics has an ambitious and broad global experimental programme for the coming decades. Large investments in building new facilities are already underway or under consideration. Scaling the present processing power and data storage needs by the foreseen increase in data rates in the next decade for HL-LHC is not sustainable within the current budgets. As a result, a more efficient usage of computing resources is required in order to realise the physics potential of future experiments. Software and computing are an integral part of experimental design, trigger and data acquisition, simulation, reconstruction, and analysis, as well as related theoretical predictions. A significant investment in computing and software is therefore critical. Advances in software and computing, including artificial intelligence (AI) and machine learning (ML), will be key for solving these challenges. Making better use of new processing hardware such as graphical processing units (GPUs) or ARM chips is a growing trend. This forms part of a computing solution that makes efficient use of facilities and contributes to the reduction of the environmental footprint of HEP computing. The HEP community already provided a roadmap for software and computing for the last EPPSU, and this paper updates that, with a focus on the most resource critical parts of our data processing chain.

97 MATHEMATICS AND COMPUTING↗

NASA NTRS2004

A "Kane's Dynamics" Model for the Active Rack Isolation System Part Two: Nonlinear Model Development, Verification, and Simplification

Many microgravity space-science experiments require vibratory acceleration levels that are unachievable without active isolation. The Boeing Corporation's active rack isolation system (ARIS) employs a novel combination of magnetic actuation and mechanical linkages to address these isolation requirements on the International Space Station. Effective model-based vibration isolation requires: (1) An isolation device, (2) an adequate dynamic; i.e., mathematical, model of that isolator, and (3) a suitable, corresponding controller. This Technical Memorandum documents the validation of that high-fidelity dynamic model of ARIS. The verification of this dynamics model was achieved by utilizing two commercial off-the-shelf (COTS) software tools: Deneb's ENVISION(registered trademark), and Online Dynamics Autolev(trademark). ENVISION is a robotics software package developed for the automotive industry that employs three-dimensional computer-aided design models to facilitate both forward and inverse kinematics analyses. Autolev is a DOS-based interpreter designed, in general, to solve vector-based mathematical problems and specifically to solve dynamics problems using Kane's method. The simplification of this model was achieved using the small-angle theorem for the joint angle of the ARIS actuators. This simplification has a profound effect on the overall complexity of the closed-form solution while yielding a closed-form solution easily employed using COTS control hardware.

Beech, G. S.↗

DOE OSTI2023

Exploration of Optimizing FPGA-based Qubit Controller for Experiments on Superconducting Quantum Computing Hardware

This work explores avenues and target areas for optimizing FPGA-based control hardware for experiments conducted on superconducting quantum computing systems and serves as an introduction to some of the current research at the intersection of classical and quantum computing hardware. With the promise of building larger-scale error-corrected quantum computers based on superconducting qubit architecture, innovations to room-temperature control electronics are needed to bring these quantum realizations to fruition. The QICK (Quantum Instrumentation Control Kit) is one leading experimental FPGA-based implementations. However, its integration into other experimental quantum computing architectures, especially those using superconducting radiofrequency (SRF) cavities, is largely unexplored. We identify some key target areas for optimizing control electronics for superconducting qubit architectures and provide some preliminary results to the resolution of a control pulse waveform. With optimizations targeted at 3D superconducting qubit setups, we hope to bring to light some of the requirements in classical computational methodologies to bring out the full potential of this quantum computing architecture, and to convey the excitement of progress in this research.

43 PARTICLE ACCELERATORS↗

DOE OSTI2023

High-bandwidth image-based predictive laser stabilization via optimized Fourier filters

Controlling the delivery of kHz-class pulsed lasers is of interest in a variety of industrial and scientific applications, from next-generation laser-plasma acceleration to laser-based x-ray emission and high-precision manufacturing. The transverse position of the laser pulse train on the application target is often subject to fluctuations by external drivers (e.g., room cooling and heating systems, motorized optics stages and mounts, vacuum systems, chillers, and/or ground vibrations). For typical situations where the disturbance spectrum exhibits discrete peaks on top of a broad-bandwidth lower-frequency background, traditional PID (proportional-integral-derivative) controllers may struggle, since as a general rule PID controllers can be used to suppress vibrations up to only about 5%–10% of the sampling frequency. Here, a predictive feed-forward algorithm is presented that significantly enhances the stabilization bandwidth in such laser systems (up to the Nyquist limit at half the sampling frequency) by online identification and filtering of one or a few discrete frequencies using optimized Fourier filters. Furthermore, the system architecture demonstrated here uses off-the-shelf CMOS cameras and piezo-electric actuated mirrors connected to a standard PC to process the alignment images and implement the algorithm. To avoid high-end, high-cost components, a machine-learning-based model of the piezo mirror’s dynamics was integrated into the system, which enables high-precision positioning by compensating for hysteresis and other hardware-induced effects. A successful demonstration of the method was performed on a 1 kHz laser pulse train, where externally-induced vibrations of up to 400 Hz were attenuated by a factor of five, far exceeding what could be done with a standard PID scheme.

Natal, Joseph↗

NASA NTRS2012

Characterization of the Vectron PX-570 Crystal Oscillator for Use in Harsh Environments

Computing hardware, data-acquisition systems, communications systems, and many electronic control systems require well-controlled timing signals for proper and accurate operation. These signals are, in most cases, provided by circuits that employ crystal oscillators due to availability, cost, ease of operation, and accuracy. In some cases, the electronic systems are expected to survive and operate under harsh conditions that include exposure to extreme temperatures. These applications exist in terrestrial systems as well as in aerospace products. Well-logging, geothermal systems, and industrial process control are examples of ground-based applications, while distributed jet engine control in aircraft, space-based observatories (such as the James Webb Space Telescope), satellites, and lunar and planetary landers are typical environments where electronics are exposed to harsh operating conditions. To ensure these devices produce reliable results, the digital heartbeat from the oscillator must deliver a stable signal that is not affected by external temperature or other conditions. One such solution is a recently introduced commercial-off-the-shelf (COTS) oscillator, the PX-570 series from Vectron International. The oscillator was designed for high-temperature applications and as proof, the crystal oscillator was subjected to a wide suite of tests to determine its ruggedness for operation in harsh environments. The tests performed by Vectron included electrical characterization under wide range of temperature, accelerated life test/aging, shock and vibration, internal moisture analysis, ESD threshold, and latch-up testing. The parametric evaluation was performed on the oscillator's frequency, output signal rise and fall times, duty cycle, and supply current over the temperature range of -125 C to +230 C. The evaluations also determined the effects of thermal cycling and the oscillator's re-start capability at extreme hot and cold temperatures. These thermal cycling and restart tests were performed at the NASA Glenn Research Center. Overall, the crystal oscillator performed well and demonstrated very good frequency stability. This paper will discuss the test procedures and present details of the performance results.

Li, Jacob↗

DOE OSTI2025

Development of an FGPA-Based Cavity Simulator for Testing RF Controls

LLRF is used to precisely control the amplitude and phase of the RF field in cavities. Often times, access to test the control algorithms with RF equipment, especially in the presence of beam, is limited or beyond reach. In such cases, testing must be done through computer modeling or simulations. Computer modeling is often too slow and difficult to interface with the LLRF hardware. Analog or digital cavity simulators are preferred as they allow for interaction with the LLRF controls platform in real-time, and compared to their analog counterparts, FPGA-based digital cavity simulators allow for a more adjustable and sophisticated implementation. The newly developed FPGA-based cavity simulator includes the cavity electrical model, the cavity mechanical model including Lorentz Force Detuning and microphonics, an amplifier model which can simulate real amplifier nonlinearities, and a beam model. The simulator has been validated using measurements from BNL’s CeC 704 MHz 5-cell SRF cryomodule.

43 PARTICLE ACCELERATORS↗

DOE OSTI2024

IRIS: Exploring Performance Scaling of the Intelligent Runtime System and its Dynamic Scheduling Policies

High-Performance Computing is becoming increasingly heterogeneous, relying on a diverse mix of hardware to achieve good performance. Paradoxically, current drivers and frameworks for these devices typically require separate languages and implementations for each vendor. Furthermore, there are few tools and little support to schedule codes between these devices in a truly heterogeneous manner-partly because of this fragmentation between vendors and the languages each supports. To overcome both limitations, the Intelligent Runtime System (IRIS) was developed. It allows a common task abstraction to automatically be shared among contemporary vendors and is run from a single host-side API. At runtime, IRIS queries the host system and registers which frameworks and drivers are available, these determine which kernels can be used by the scheduler-CPUs via OpenMP, Nvidia GPUs (CUDA), AMD GPUs (HIP), and Intel and Xilinx FPGAs with OpenCL. IRIS enables tasks to be scheduled to any heterogeneous device and resolves to the appropriate kernel binary at runtimeit only uses the devices supported by the system on which it is run. IRIS supports single-task and graph-based expressions of dependencies of tasks. Additionally, IRIS features a range of dynamic scheduling policies, allowing complex chains of tasks and interactions to be executed, relieving the programmer/user from considering the system to assign tasks to devices optimally. This paper presents the peak performance attainable by IRIS over a range of systems-each with different numbers and types of accelerator devices, it highlights the flexibility of IRIS since these devices are truly heterogeneous, relying on different backends (drivers, frameworks, and languages) which historically required unique implementations to utilize them. We then use this peak performance as a baseline to compare increasingly complex chains of tasks (with increasingly complex task dependencies) and evaluate how IRIS copes. Finally, we consider the performance of different IRIS scheduling policies on this range of task graphs.

Johnston, Beau↗

DOE OSTI2025

Neural architecture codesign for fast physics applications

We develop a pipeline to streamline neural architecture codesign for physics applications to reduce the need for ML expertise when designing models for novel tasks. Our method employs neural architecture search and network compression in a two-stage approach to discover hardware efficient models. This approach consists of a global search stage that explores a wide range of architectures while considering hardware constraints, followed by a local search stage that fine-tunes and compresses the most promising candidates. We exceed performance on various tasks and show further speedup through model compression techniques such as quantization-aware-training and neural network pruning. We synthesize the optimal models to high level synthesis code for FPGA deployment with the hls4ml library. Additionally, our hierarchical search space provides greater flexibility in optimization, which can easily extend to other tasks and domains. We demonstrate this with two case studies: Bragg peak finding in materials science and jet classification in high energy physics, achieving models with improved accuracy, smaller latencies, or reduced resource utilization relative to the baseline models.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗