Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “inference 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.

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

At least 91 records · Page 5

Nuclear winds and the narrow-line emission from active galaxies

This paper discusses active galaxy narrow-line emission in the context of a supersonic wind generated by the active nucleus, a nuclear wind. Wind acceleration can naturally produce the inferred narrow-line cloud velocities as well as those of the broad-line and broad absorption line clouds with only a weak dependence on the parameters of the active nucleus. The primary obstacle to wind acceleration of emission-line clouds is the destructive effect of instabilities; however, the stability and structure of wind-confined clouds remain areas of future research. Observations, particularly of the Seyfert galaxy NGC 1068, are interpreted in the context of nuclear winds.

Smith, Steven J.↗

Towards On-Chip Learning for Low Latency Reasoning with End-to-End Synthesis

The Software Defined Architectures (SODA) Synthesizer is an open-source compiler-based tool able to automatically generate domain-specialized systems targeting Application-Specific Integrated Circuits (ASICs) or Field Programmable Gate Arrays (FPGAs) starting from high-level programming. SODA is composed of a frontend, SODA-OPT, which leverages the multilevel intermediate representation (MLIR) framework to interface with productive programming tools (e.g., machine learning frame-works), identify kernels suitable for acceleration, and perform high-level optimizations, and of a state-of-the-art high-level synthesis backend, Bambu from the PandA framework, to generate custom accelerators. One specific application of the SODA Synthesizer is the generation of accelerators to enable ultra-low latency inference and control on autonomous systems for scientific discovery (e.g., electron microscopes, sensors in particle accelerators, etc.). This paper provides an overview of the flow in the context of the generation of accelerators for edge processing to be integrated in transmission electron microscopy (TEM) devices, focusing on use cases from precision material synthesis. We show the tool in action with an example of design space exploration for inference on reconfigurable devices with a conventional deep neural network model (LeNet). Finally, we discuss the research directions and opportunities enabled by SODA in the area of autonomous control for scientific experimental workflows.

Castellana, Vito G.↗

Radiative acceleration in outflows from broad absorption line quasi-stellar objects. 2: Wind models

We investigate the dynamics of radiatively driven broad absorption-line (BAL) outflows in quasi-stellar objects (QSOs) by developing radial and time-independent numerical models. Two limits are explored. The first assumes that the absorbing matter is not forced to comove with the substrate, which provides pressure confinement. This assumption allows us to explore in detail a case in which the acceleration is entirely due to radiation pressure. Using the parameters inferred from observations, we find that under these conditions radiative acceleration (mainly due to resonance line scattering) can readily accelerate the flow to the observed velocities. An important feature of the noncoupled flow is that the line profiles tend to stay relatively flat throughout the velocity interval covered by the line. We discuss how relaxing the assumptions of radial symmetry and time independent may help to explain the structures observed in BALs. In the second class of models, the absorbing flow is assumed to be completely coupled to the substrate in which it is embedded. Aside from being more plausible physically, these models produce line profiles that trail off at higher velocities, a behavior observed in some BALs. We show that, even if the substrate is massless, we have to assume a starting radius very close to the inferred radius of the broad emission-line region (approximately 0.1 pc) in order to obtain a significant contribution from radiative acceleration, given a typical active galactic nucleus (AGN) spectrum. The reason is that the energy input needed to pressurize the substrate, allowing the flow to become supersonic and to retain a reasonable ionization equilibrium, at the same time contributes appreciably to the acceleration. A way to relax the small starting radius constraint is to use a softer ionizing spectrum.

Arav, Nahum↗

Detecting Thermally-Induced Spinodal Decomposition with Picosecond Ultrasonics in Cast Austenitic Stainless Steels

Destructive techniques to monitor nuclear reactor component health may not always be available during service, as they are time-consuming and often require pre-installed inspection coupons. Non-destructive evaluation (NDE) techniques can bridge this gap by rapidly identifying the state of mission-critical reactor components, via inference between NDE-measurable material properties and those of ultimate interest, such as ductility and toughness. Here, we demonstrate one such inference about the health of thermally aged cast austenitic stainless steels. Observations of surface acoustic wave peak (SAW) splitting correlate with spinodal decomposition-induced embrittlement as destructively measured by Charpy impact energy. Elastodynamic calculations and molecular dynamics simulations of the effects of spinodal decomposition on elastic moduli support that the new acoustic modes present are due to stiffening in the δ-ferrite domains. Finally, this discovery enables one to probe structure-property relationships in materials in a greatly accelerated manner, suggesting that similar inference methods can be used to determine material fitness-for-service, or to quickly uncover new structure-property relationships.

304-type stainless steel↗

Parametric 3D Atmospheric Reconstruction in Highly Variable Terrain with Recycled Monte Carlo Paths and an Adapted Bayesian Inference Engine

We describe a method for accelerating a 3D Monte Carlo forward radiative transfer model to the point where it can be used in a new kind of Bayesian retrieval framework. The remote sensing challenge is to detect and quantify a chemical effluent of a known absorbing gas produced by an industrial facility in a deep valley. The available data is a single low resolution noisy image of the scene in the near IR at an absorbing wavelength for the gas of interest. The detected sunlight has been multiply reflected by the variable terrain and/or scattered by an aerosol that is assumed partially known and partially unknown. We thus introduce a new class of remote sensing algorithms best described as "multi-pixel" techniques that call necessarily for a 3D radaitive transfer model (but demonstrated here in 2D); they can be added to conventional ones that exploit typically multi- or hyper-spectral data, sometimes with multi-angle capability, with or without information about polarization. The novel Bayesian inference methodology uses adaptively, with efficiency in mind, the fact that a Monte Carlo forward model has a known and controllable uncertainty depending on the number of sun-to-detector paths used.

Monte Carlo↗

Towards Automated Generation of Chiplet-Based Systems

The Software Defined Architectures (SODA) Synthesizer is an open-source compiler-based tool able to automatically generate domain-specialized systems targeting Application- Specific Integrated Circuits (ASICs) or Field Programmable Gate Arrays (FPGAs) starting from high-level programming. SODA is composed of a high-level frontend, SODA-OPT, which leverages the multilevel intermediate representation (MLIR) framework to interface with productive programming tools (e.g., machine learning frameworks), identify kernels suitable for acceleration, and perform high-level optimizations, and of a state-of-the-art high-level synthesis backend, Bambu from the PandA framework, to generate custom accelerators. One specific application of the SODA Synthesizer is the generation of accelerators to enable ultra-low latency inference and control on autonomous systems for scientific discovery (e.g., electron microscopes, sensors in particle accelerators, etc.). This talk will discuss ongoing work on the SODA synthesizer to enable no-human-in-the-loop generation and design space exploration of the chiplets for highly specialized artificial intelligence accelerators. Connecting these highly specialized chiplets to general-purpose cores or programmable accelerators will allow to quickly deploy autonomous systems for scientific discovery.

Limaye, Ankur M.↗

FPGA Acceleration of GCN in Light of the Symmetry of Graph Adjacency Matrix

Graph Convolutional Neural Networks (GCNs) are widely used to process large-scale graph data. Different from deep neural networks (DNNs), GCNs are sparse, irregular, and unstructured, posing unique challenges to hardware acceleration with regular processing elements (PEs). In particular, the adjacency matrix of a GCN is extremely sparse, leading to frequent but irregular memory access, low spatial/temporal data locality and poor data reuse. Furthermore, a realistic graph usually consists of unstructured data (e.g., unbalanced distributions), creating significantly different processing times and imbalanced workload for each node in GCN acceleration. To overcome these challenges, we propose an end-to-end hardware-software co-design to accelerate GCNs on resource-constrained FPGAs with the features including: (1) A custom dataflow that leverages symmetry along the diagonal of the adjacency matrix to accelerate feature aggregation for undirected graphs. We utilize either the upper or the lower triangular matrix of the adjacency matrix to perform aggregation in GCN to improve data reuse. (2) Unified compute cores for both aggregation and transform phases, with full support to the symmetry-based dataflow. These cores can be dynamically reconfigured to the systolic mode for transformation or as individual accumulators for aggregation in GCN processing. (3) Preprocessing of the graph in software to rearrange the edges and features to match the custom dataflow. This step improves the regularity in memory access and data reuse in the aggregation phase. Moreover, we quantize the GCN precision from FP32 to INT8 to reduce the memory footprint without losing the inference accuracy. We implement our accelerator design in Intel Stratix10 MX FPGA board with HBM2, and demonstrate 1.3x-110.5x improvement in end-to-end GCN latency as compared to the state-of the-art FPGA implementations, on the graph datasets of Cora, Pubmed, Citeseer and Reddit.

Nair, Gopikrishnan R.↗

Actinides in the Source of Cosmic Rays and the Present Interstellar Medium

The abundances of the actinide elements in the cosmic rays can provide critical constraints on the major sites of their acceleration. Using recent calculations of the r-process yields in core collapse supernovae, we have determined the actinide abundances averaged over various assumed time intervals for their supernova generation and their cosmic-ray acceleration. Using standard Galactic chemical evolution models, we have also determined the expected actinide abundances in the present interstellar medium. From these two components, we have calculated the U/Th and other actinide abundances expected in the supernova-active cores of superbubbles, as a function of their ages and mean metallicity resulting from dilution with interstellar cloud debris. Then, using observations of the fractions of Galactic supernovae that occur in superbubbles and in the rest of the interstellar medium, we calculate the expected actinide abundances in cosmic rays accelerated by Galactic supernovae. We find that the current measurements of actinide/Pt-group and preliminary estimates of the UPuCm/Th ratio in cosmic rays are all consistent with the expected values if superbubble cores have mean metallicities of around 3 times solar. Such metallicities are quite comparable to the superbubble core metallicities inferred from other cosmic-ray observations. Future, more precise measurements of these ratios with experiments such as ECCO are needed to provide a better measure of the mean source metallicity sampled by the local Galactic cosmic rays. Measurements of the cosmic- ray actinide abundances have been favorably compared with the protosolar ratio, inferred from present solar system abundances, to infer that the cosmic rays are accelerated from the general interstellar medium. We suggest, however, that such an inference is not valid because the expected actinide abundances in the present interstellar medium are very different from the protosolar values, which sampled the interstellar medium 4.5 Gyr ago and included an additional fresh ejecta component from a neighboring supernova.

Lingenfelter, R. E.↗

Electron Densities in Solar Flare Loops, Chromospheric Evaporation Upflows, and Acceleration Sites

We compare electron densities measured at three different locations in solar flares: (1) in Soft X-Ray (SXR) loops, determined from SXR emission measures and loop diameters from Yohkoh Soft X-Ray Telescope maps (n(sub e, sup SXR) = (0.2-2.5) x 10(exp 11)/ cu cm); (2) in chromospheric evaporation upflows, inferred from plasma frequency cutoffs of decimetric radio bursts detected with the 0.1-3 GHz spectrometer Phoenix of ETH Zuerich (n(sub e, sup upflow) = (0.3-11) x 10(exp 10)/cu cm; and (3) in acceleration sites, inferred from the plasma frequency at the separatrix between upward-accelerated (type III bursts) and downward-accelerated (reverse-drift bursts) electron beams [n(sub e, sup acc) = (0.6-10) x 10(exp 9)/cu cm]. The comparison of these density measurements, obtained from 44 flare episodes (during 14 different flares), demonstrates the compatibility of flare plasma density diagnostics with SXR and radio methods. The density in the upflowing plasma is found to be somewhat lower than in the filled loops, having ratios in a range n(sub e, sup upflow)/n(sub e, sup SXR) = 0.02-1.3, and a factor of 3.6 higher behind the upflow front. The acceleration sites are found to have a much lower density than the SXR-bright flare loops, i.e., n(sub e, sup acc)/n(sub e, sup SXR) = 0.005- 0.13, and thus must be physically displaced from the SXR-bright flare loops. The scaling law between electron time-of-flight distances l' and loop half-lengths s, l'/s = 1.4 +/- 0.3, recently established by Aschwanden et al. suggests that the centroid of the acceleration region is located above the SXR-bright flare loop, as envisioned in cusp geometries (e.g., in magnetic reconnection models).

Aschwanden, Markus J.↗

Analog architectures for neural network acceleration based on non-volatile memory

Analog hardware accelerators, which perform computation within a dense memory array, have the potential to overcome the major bottlenecks faced by digital hardware for data-heavy workloads such as deep learning. Exploiting the intrinsic computational advantages of memory arrays, however, has proven to be challenging principally due to the overhead imposed by the peripheral circuitry and due to the non-ideal properties of memory devices that play the role of the synapse. We review the existing implementations of these accelerators for deep supervised learning, organizing our discussion around the different levels of the accelerator design hierarchy, with an emphasis on circuits and architecture. We explore and consolidate the various approaches that have been proposed to address the critical challenges faced by analog accelerators, for both neural network inference and training, and highlight the key design trade-offs underlying these techniques.

97 MATHEMATICS AND COMPUTING↗

Aeronomy report no. 74: The Urbana meteor-radar system; design, development, and first observations

The design, development, and first observations of a high power meteor-radar system located near Urbana, Illinois are described. The roughly five-fold increase in usable echo rate compared to other facilities, along with automated digital data processing and interferometry measurement of echo arrival angles, permits unsurpassed observations of tidal structure and shorter period waves. Such observations are discussed. The technique of using echo decay rates to infer density and scale height and the method of inferring wind shear from radial acceleration are examined. An original experiment to test a theory of the Delta-region winter anomaly is presented.

Hess, G. C.↗

Solitary Waves in the Western Equatorial Pacific Ocean

During the spring tides of early January and February 1993, groups of solitary internal waves were observed propagating through the Intensive Flux Array of the TOGA COARE experiment. The waves appear to originate near the islands of Nugarba (3 deg S 30 deg S - 154 deg 30'E). They travel north-eastward at 2.5-3 m/s, closely coupled with the semi-diurnal baroclinic tide. Peak amplitudes exceed 60 m. Velocities are in excess of .8 m/s. Sea-surface vertical displacements of order.3 m can be inferred directly from the lateral acceleration of surface waters. The Equatorial Undercurrent is displaced by soliton passage but apparently is unaffected otherwise. The intrinsic shear of the solitary crests is small compared to ambient equatorial shears. The crests, while not themselves unstable, are effective at triggering instabilities on the background flow. The motions potentially contribute 10-15 Watts/sq m to the flux of heat into the mixed layer.

Pinkel, R.↗

A First-Principles Model Based on Saturation of the Electron Cyclotron Drift Instability for Electron Transport in Hydrodynamics Simulations of Hall Thruster Plasmas

We present a computational model based on the hypothesis that the higher-than-expected electron mobility across magnetic field lines observed in Hall thrusters is due to the growth and eventual saturation of the electron cyclotron drift instability. The key improvement of this model with respect to previous work is that it decouples the saturation of the instability from the correlation that exists between mobility and magnitude of the wave perturbations in linear theory, thereby enabling first-principles simulations. The model has been incorporated in the 2-D (r-z) multi-fluids code Hall2De and simulations of the H6 thruster at 300 V, 20 A are numerically stable and achieve a steady-state solution at a computational cost that is not significantly higher than previous Hall2De simulations. The computed location of the acceleration region is found to be within 10% of the length of the acceleration channel compared to that inferred by experiments. The simulations also capture well the plasma gradients along the channel centerline of this thruster. To further establish the validity of this model, we plan to conduct simulations of other thrusters and/or operating conditions.

Lopez Ortega, Alejandro↗

A Magnetized Strongly Turbulent Corona as the Source of Neutrinos from NGC 1068

Abstract The cores of active galactic nuclei are potential accelerators of 10–100 TeV cosmic rays, in turn producing high-energy neutrinos. This picture was confirmed by the compelling evidence of a TeV neutrino signal from the nearby active galaxy NGC 1068, leaving open the question of what is the site and mechanism of cosmic-ray acceleration. One candidate is the magnetized turbulence surrounding the central supermassive black hole. Recent particle-in-cell simulations of magnetized turbulence indicate that stochastic cosmic-ray acceleration is nonresonant, in contrast to the assumptions of previous studies. We show that this has important consequences on a self-consistent theory of neutrino production in the corona, leading to a more rapid cosmic-ray acceleration than previously considered. The turbulent magnetic-field fluctuations needed to explain the neutrino signal are consistent with a magnetically powered corona. We find that strong turbulence, with turbulent magnetic energy density higher than 1% of the rest-mass energy density, naturally explains the normalization of the IceCube neutrino flux, in addition to the neutrino spectral shape. Only a fraction of the protons in the corona, which can be directly inferred from the neutrino signal, are accelerated to high energies. Thus, in this framework, the neutrino signal from NGC 1068 provides a testbed for particle acceleration in magnetized turbulence.

Fiorillo, Damiano F. G. (ORCID:0000000349279850)↗

Particle Acceleration in Active Galactic Nuclei

The high efficiency of energy generation inferred from radio observations of quasars and X-ray observations of Seyfert active galactic nuclei (AGNs) is apparently achieved only by the gravitational conversion of the rest mass energy of accreting matter onto supermassive black holes. Evidence for the acceleration of particles to high energies by a central engine is also inferred from observations of apparent superluminal motion in flat spectrum, core-dominated radio sources. This phenomenon is widely attributed to the ejection of relativistic bulk plasma from the nuclei of active galaxies, and accounts for the existence of large scale radio jets and lobes at large distances from the central regions of radio galaxies. Reports of radio jets and superluminal motion from galactic black hole candidate X-ray sources indicate that similar processes are operating in these sources. Observations of luminous, rapidly variable high-energy radiation from active galactic nuclei (AGNs) with the Compton Gamma Ray Observatory show directly that particles are accelerated to high energies in a compact environment. The mechanisms which transform the gravitational potential energy of the infalling matter into nonthermal particle energy in galactic black hole candidates and AGNs are not conclusively identified, although several have been proposed. These include direct acceleration by static electric fields (resulting from, for example, magnetic reconnection), shock acceleration, and energy extraction from the rotational energy of Kerr black holes. The dominant acceleration mechanism(s) operating in the black hole environment can only be determined, of course, by a comparison of model predictions with observations. The purpose of the work proposed for this grant was to investigate stochastic particle acceleration through resonant interactions with plasma waves that populate the magnetosphere surrounding an accreting black hole. Stochastic acceleration has been successfully applied to the problem of ion and electron energization in solar flares, and is capable of accounting for a wide range of both neutral and charged particle emissions. It is also a component in diffusive shock acceleration, since pitch-angle scattering (which is necessary for multiple shock crossings) is accompanied by diffusion in momentum space, which in turn yields a net systematic energy gain; however, stochastic energization will dominate the first-order shock process only in certain parameter regimes. Although stochastic acceleration has been applied to particle energization in the lobes of radio galaxies, its application to the central regions of AGNs has only recently been considered, but not in detail. We proposed to systematically investigate the plasma processes responsible for stochastic particle acceleration in black hole magnetospheres along with the energy-loss processes which impede particle energization. To this end we calculated acceleration rates and escape time scales for protons and electrons resonating with Alfven waves, and for electrons resonating with whistlers. Assuming either a Kolmogorov or Kraichnan wave spectrum, accretion at the Eddington limit, magnetic field strengths near equipartition, and turbulence energy densities approx. 10% of the total magnetic field energy density, we find that Alfven waves accelerate protons to Lorentz factors approx, equals 10(exp 4) - 10(exp 6) before they escape from the system. Acceleration of electrons by fast mode and whistler waves can produce a nonthermal population of relativistic electrons whose maximum energy is determined by a competition with radiation losses.

Miller, James A.↗

End-to-end AI framework for interpretable prediction of molecular and crystal properties

We introduce an end-to-end computational framework that allows for hyperparameter optimization using the DeepHyper library, accelerated model training, and interpretable AI inference. The framework is based on state-of-the-art AI models including CGCNN, PhysNet, SchNet, MPNN, MPNN-transformer, and TorchMD-NET. We employ these AI models along with the benchmark QM9, hMOF, and MD17 datasets to showcase how the models can predict user-specified material properties within modern computing environments. We demonstrate transferable applications in the modeling of small molecules, inorganic crystals and nanoporous metal organic frameworks with a unified, standalone framework. We have deployed and tested this framework in the ThetaGPU supercomputer at the Argonne Leadership Computing Facility, and in the Delta supercomputer at the National Center for Supercomputing Applications to provide researchers with modern tools to conduct accelerated AI-driven discovery in leadership-class computing environments. We release these digital assets as open source scientific software in GitLab, and ready-to-use Jupyter notebooks in Google Colab.

36 MATERIALS SCIENCE↗

A differentiable simulation package for performing inference of synchrotron-radiation-based diagnostics

The direction of particle accelerator development is ever-increasing beam quality, currents and repetition rates. This poses a challenge to traditional diagnostics that directly intercept the beam due to the mutual destruction of both the beam and the diagnostic. An alternative approach is to infer beam parameters non-invasively from the synchrotron radiation emitted in bending magnets. However, inferring the beam distribution from a measured radiation pattern is a complex and computationally expensive task. To address this challenge we present SYRIPY ( SYnchrotron Radiation In PYthon ), a software package intended as a tool for performing inference of synchrotron-radiation-based diagnostics. SYRIPY has been developed using PyTorch , which makes it both differentiable and able to leverage the high performance of GPUs, two vital characteristics for performing statistical inference. The package consists of three modules: a particle tracker, Lienard–Wiechert solver and Fourier optics propagator, allowing start-to-end simulation of synchrotron radiation detection to be carried out. SYRIPY has been benchmarked against SRW , the prevalent numerical package in the field, showing good agreement and up to a 50× speed improvement. Finally, we have demonstrated how SYRIPY can be used to perform Bayesian inference of beam parameters using stochastic variational inference.

43 PARTICLE ACCELERATORS↗