Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “deep potential”

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 1,243 records · Page 69
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials

Abstract This work presents Neural Equivariant Interatomic Potentials (NequIP), an E(3)-equivariant neural network approach for learning interatomic potentials from ab-initio calculations for molecular dynamics simulations. While most contemporary symmetry-aware models use invariant convolutions and only act on scalars, NequIP employs E(3)-equivariant convolutions for interactions of geometric tensors, resulting in a more information-rich and faithful representation of atomic environments. The method achieves state-of-the-art accuracy on a challenging and diverse set of molecules and materials while exhibiting remarkable data efficiency. NequIP outperforms existing models with up to three orders of magnitude fewer training data, challenging the widely held belief that deep neural networks require massive training sets. The high data efficiency of the method allows for the construction of accurate potentials using high-order quantum chemical level of theory as reference and enables high-fidelity molecular dynamics simulations over long time scales.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
NASA NTRS2018

Auto-Gopher-2 – An Autonomous Wireline Rotary Piezo-Percussive Deep Drilling Mechanism

Drilling deep into the subsurface of planetary bodies in the Solar System for samples acquisition enables critical capabilities for future NASA exploration missions in its quest to understand the origins of the Solar System and potentially the search for life. Such planetary bodies as Mars and Europa are key targets for potential missions that would require reaching great depths. Performing drilling while using minimal mass/volume systems and with low energy consumption are the main requirements that are imposed on such technologies. A wireline deep drill, called Auto-Gopher-2, is currently being developed as a joint effort between JPL and Honeybee Robotics Ltd. The Auto-Gopher II is a wireline rotary piezo-percussive deep drilling mechanism that combines formation breaking by rotating and piezoelectric actuator hammering and cuttings removal by rotating a fluted bit. The hammering mechanism is based on the Ultrasonic/Sonic Drill/Corer (USDC) mechanism that has been developed as an adaptable tool for many drilling and coring applications. The USDC uses an intermediate ball-shape free-mass to transform high frequency vibrations of a piezoelectric transducer horn tip into sonic hammering of the drill bit. The lessons learned from the previous studies are being implemented into the development of the Auto-Gopher-II, an autonomous deep wireline drill with integrated cuttings and sample management and drive electronics. Subsystems of the wireline drill are being developed in parallel at JPL and Honeybee Robotics, Ltd. Issues related to the bit and its ability to retain the cuttings for caching and removal are currently being addressed. This paper presents the development efforts of the piezoelectric actuator, cuttings removal and retention flutes and drive electronics.

Bar-Cohen, Yoseph↗
DOE OSTI2020

Deep convolutional neural networks for multi-scale time-series classification and application to disruption prediction in fusion devices

The multi-scale, mutli-physics nature of fusion plasmas makes predicting plasma events challenging. Recent advances in deep convolutional neural network architectures (CNN) utilizing dilated convolutions enable accurate predictions on sequences which have long-range, multi-scale characteristics, such as the time-series generated by diagnostic instruments observing fusion plasmas. Here we apply this neural network architecture to the popular problem of disruption prediction in fusion tokamaks, utilizing raw data from a single diagnostic, the Electron Cyclotron Emission imaging (ECEi) diagnostic from the DIII-D tokamak. ECEi measures a fundamental plasma quantity (electron temperature) with high temporal resolution over the entire plasma discharge, making it sensitive to a number of potential pre-disruptions markers with different temporal and spatial scales. Promising, initial disruption prediction results are obtained training a deep CNN with large receptive field ({$\sim$}30k), achieving an $F_1$-score of {$\sim$}91\% on individual time-slices using only the ECEi data.

convolutional neural networks↗
NASA NTRS2018

Cryogenic Selective Surfaces

There are many challenges involved in deep-space exploration, but several of these can be mitigated, or even solved, by the development of a coating that reflects most of the Sun’s energy, yet still provides far-infrared heat emission. Such a coating would allow non-heat-generating objects in space to reach cryogenic temperatures without using an active cooling system. This would benefit deep-space sensors that require low temperatures, such as the James Webb Telescope focal plane array. It would also allow the use of superconductors in deep space, which could lead to magnetic energy storage rings, lossless power delivery, or perhaps a large-volume magnetic shield against galactic cosmic radiation. However, perhaps the most significant enablement achieved from such a coating would be the long-term, deep space storage of cryogenic liquids, such as liquid oxygen (LOX). In our Phase I NIAC study, we realized that a combination of scattering particles and a silver backing could yield a highly effective, very broadband, reflector that could potentially reflect more than 99.9% of the Sun’s irradiant power. We developed a sophisticated model of this reflector and theoretically showed that cryogenic temperatures could be achieved in deep space at one astronomical unit (1 AU) from the Sun. We showed how this new reflector could minimize heat conduction into the cryogenic tanks by coating the tank support struts. We then modelled a strawman architecture for a mission to Mars, using a coated LOX tank, coated struts, and infrared shields, to show that with our new coating it would be possible to maintain liquid oxygen passively. As a result of this work a patent application was generated and a paper published in Optics Letters. Our Phase II NIAC study had two primary goals, to develop a rigid version of the cryogenic selective surface proposed in Phase I and to test its performance in a simulated deep space environment. During the first year of the project the work concentrated on developing rigid tiles of BaF2, leading to tiles as large as 4 inches in diameter that transmitted very little visible light. In addition, during the first year a simulated deep space environment was created using a vacuum chamber and cryocooler. Using this facility, we showed that our BaF2 tiles absorbed less than ¼% of 375 nm radiation, a significant milestone for the work. During the second year of the project, we continued to develop the BaF2 tiles and we put significant effort into the construction of a deep space environment where we could project simulated solar radiation onto a sample. In the spring of 2018, we conducted our first solar simulator test with BaF2 and saw about 3.6% absorption. This is better than the state-of-the-art, but disappointing since predictions were for much lower absorption. We, erroneously, attributed this absorption to water retention by the BaF2, and decided to change materials. We considered several oxides and settled on yttrium oxide (Y2O3) for further development, because it is broadband, lightweight, has high index, and is hydrophobic. In July 2018 we conducted our first test of a rigid tile of Y2O3 in the simulated deep space environment and saw significant absorption again. We then realized that the issue was not water, but mid-wave radiation passing through the tile and being absorbed by the temperature sensor and the varnish used to hold it in place. We wrapped the sensor in silver foil, re-ran the test, and saw much lower absorption; only 1.1%. We then re-ran the BaF2 tile and saw 1.4% absorption. These values are almost adequate to maintain LOX in deep space, but we suspect that there are still issues in our test apparatus; we suspect thermocouple wires may be absorbing radiation. Further, post-NIAC, testing will better determine the performance of our new solar reflector. In order to restrict the size of this report, we will only briefly describe topics that we have previously published, allowing us to devote more time to new material. So minimal material will be devoted to modeling the material and deep space cryogenic storage, while longer sections will cover our material development, simulated deep space testing, and new applications. The Launch Service Program (LSP) requested that we explore ways to use this new coating to maintain LOX in low Earth Orbit and that work is described. In addition, the Nuclear Thermal Propulsion (NTP) Program asked us to explore ways to reduce the heat load for liquid hydrogen, resulting in the development of a spray-on version of the coating that should significantly improve in-space multi-layer insulation performance.

Robert C. Youngquist,↗
DOE OSTI2016

Potential hydrothermal resource temperatures in the Eastern Snake River Plain, Idaho

The Eastern Snake River Plain (ESRP) in southern Idaho is a region of high heat flow. Sustained volcanic activities in the wake of the passage of the Yellowstone Hotspot have turned this region into an area with great potential for geothermal resources as evidenced by numerous hot springs scattered along the margins of the plain and several hot-water producing wells and hot springs within the plain. Despite these thermal expressions, it is hypothesized that the pervasive presence of an overlying groundwater aquifer in the region effectively masks thermal signatures of deep-seated geothermal resources. The dilution of deeper thermal water and re-equilibration at lower temperature are significant challenges for the evaluation of potential resource areas in the ESRP. Over the past several years, we collected approximately 100 water samples from springs/wells for chemical analysis as well as assembled existing water chemistry data from literature. We applied several geothermometric and geochemical modeling tools to these chemical compositions of ESRP water samples. Geothermometric calculations based on principles of multicomponent equilibrium geothermometry with inverse geochemical modeling capability (e.g., Reservoir Temperature Estimator, RTEst) have been useful for the evaluation of reservoir temperatures. RTEst geothermometric calculations of ESRP thermal water samples indicated numerous potential geothermal areas with elevated reservoir temperatures. Specifically, areas around southern/southwestern side of the Bennett Hills and within the Camas Prairies in the western-northwestern regions of the ESRP and its margins suggest temperatures in the range of 140-200°C. In the northeastern portions of the ESRP, Lidy Hot Springs, Ashton, Newdale, and areas east of Idaho Falls have expected reservoir temperature =140 °C. In the southern ERSP, areas near Buhl and Twin Falls are found to have elevated temperatures as high as 160 °C. These areas are likely to host potentially economic geothermal resources; however, further detailed study is warranted to each site to evaluate hydrothermal suitability for economic use.

Eastern Snake River Plain, ESRP, RTEst, geothermal↗
DOE OSTI2024

Future North Atlantic tropical cyclone intensities in thermodynamically modified historical environments

Tropical cyclones (TCs) rank as the deadliest and most financially crippling natural disasters in the United States for the last half-century. It is imperative to assess potential shifts in TC intensity within the paradigm of an evolving climate. In this study, we have modeled the intensities of 620 historical TC events in the North Atlantic Basin using the Risk Analysis Framework for Tropical Cyclones (RAFT)'s deep learning intensity model. By applying a thermodynamic warming signal extrapolated from Global Climate Models, we rerun historical events under eight different future climate scenarios, providing a spectrum of potential TC intensity outcomes. One of the future simulations indicates a staggering 43% increase in the number of major hurricanes, underscoring the critical impact of climate change on TC intensity. Additionally, an interactive dashboard has been created to enable users to explore individual storm simulations and understand the influence of future climate signals on environmental conditions of TC development and resulting TC intensities. This dataset and the user-friendly tool offer invaluable resources for systematic exploration of the discrete effects that changes in the air-sea thermodynamic state have on the intensities of TCs.

Climate Change↗
NASA NTRS2006

Volume Ice Radiolysis in the Outer Solar System

The primary energy flux of charged particle components of the heliospheric and magnetospheric environments of the solar system is primarily carried by highly penetrating energetic particles. Although laboratory experiments on production of organics and oxidants typically only address effects on very thin surface layers, energy deposition occurs on surfaces of icy bodies of the outer solar system to meters in depth. Time scales for significant radiolytic deposition vary from thousands of years at millimeter depths on Europa to billions of years in the meters-deep regolith of Kuiper Belt Objects. Radioisotope decay (e.g., K-40) also contributes to volume radiolysis as the only energy source at much greater depths. Radiolytic oxygen is a potential resource for life within Europa and a partial source of oxygen for Saturn's magnetosphere and Titan's upper atmosphere. Interactions of very high energy cosmic rays with ices at Titan's surface may provide one of the few sources of oxidants in that highly reducing environment. The red colors of low-inclination classical Kuiper Belt Objects at 40-50 AU, and Centaur objects originating from this same population, may arise from volume radiolysis of deep ice layers below more refractory radiation crusts eroded away by surface sputtering and micrometeoroid impacts. A variety of techniques are potentially available to measure volume radiolysis products and have been proposed for study as part of the new Space Physics of Life initiative at NASA Goddard Space Flight Center. The technique of Electron Paramagnetic Resonance (EPR) has been used in medical studies to measure oxidant production in irradiated human tissue for cancer treatment. Other potential techniques include optical absorption spectroscopy and standard wet chemical analysis. These and other potential techniques are briefly reviewed for applicability to problems in solar system ice radiolysis and astrobiology.

Cooper, John F.↗
DOE OSTI2025

A Predictive Deep-Reinforcement-Learning-Based Connected Automated Vehicle Anticipatory Longitudinal Control in a Mixed Traffic Lane Change Condition

Maintaining safety and efficiency for mixed traffic consisting of connected automated vehicles (CAVs) and human-driven vehicles (HDVs) is an arduous task due to the inherent HDVs’ stochasticity. Especially for longitudinal control, which is the basic function of vehicle automation, prevailing research primarily considers CAV’s car-following control merely the acceleration and deceleration of leading vehicles. However, this approach overlooks the potential disruptions caused by surrounding vehicles executing lane changes, which can significantly impact the control vehicle’s stability and overall safety. Hence, our study introduces a predictive deep reinforcement learning (DRL) longitudinal CAV controller. This innovative approach leverages prediction from a physics-informed neural network as well as the control capability of DRL to better anticipate and mitigate issues arising from lane-changing, enhancing the safety and efficiency of CAVs in such scenarios. Finally, validated by the numerical simulations embedded with the real-world data, the results indicate that the proposed controller significantly enhances the safety and efficiency of CAVs in situations involving lane changes by other vehicles, showcasing its potential as a valuable tool in advancing CAV technology in mixed traffic.

33 ADVANCED PROPULSION SYSTEMS↗
NASA NTRS2013

Concentrators Enhance Solar Power Systems

"Right now, solar electric propulsion is being looked at very seriously," says Michael Piszczor, chief of the photovoltaic and power technologies branch at Glen Research Center. The reason, he explains, originates with a unique NASA mission from the late 1990s. In 1998, the Deep Space 1 spacecraft launched from Kennedy Space Center to test a dozen different space technologies, including SCARLET, or the Solar Concentrator Array with Refractive Linear Element Technology. As a solar array that focused sunlight on a smaller solar cell to generate electric power, SCARLET not only powered Deep Space 1 s instruments but also powered its ion engine, which propelled the spacecraft throughout its journey. Deep Space 1 was the first spacecraft powered by a refractive concentrator design like SCARLET, and also utilized multi-junction solar cells, or cells made of multiple layers of different materials. For the duration of its 38-month mission, SCARLET performed flawlessly, even as Deep Space 1 flew by Comet Borrelly and Asteroid Braille. "Everyone remembers the ion engine on Deep Space 1, but they tend to forget that the SCARLET array powered it," says Piszczor. "Not only did both technologies work as designed, but the synergy between the two, solar power and propulsion together, is really the important aspect of this technology demonstration mission. It was the first successful use of solar electric propulsion for primary propulsion." More than a decade later, NASA is keenly interested in using solar electric propulsion (SEP) for future space missions. A key issue is cost, and SEP has the potential to substantially reduce cost compared to conventional chemical propulsion technology. "SEP allows you to use spacecraft that are smaller, lighter, and less costly," says Piszczor. "Even though it might take longer to get somewhere using SEP, if you are willing to trade time for cost and smaller vehicles, it s a good trade." Potentially, SEP could be used on future science missions in orbit around the Earth or Moon, to planets or asteroids, on deep space science missions, and even on exploration missions. In fact, electric propulsion is already being used on Earth-orbiting satellites for positioning.

Source record↗
DOE OSTI2025

Evaluating the carbon capture potential of industrial waste as a feedstock for enhanced weathering

Abstract Limiting anthropogenic global climate warming since the start of the industrial period to less than 2 °C will very likely require both deep and rapid reductions in anthropogenic greenhouse gas emissions and a range of approaches toward carbon dioxide removal (CDR). One prominent CDR approach is enhanced weathering (EW), in which crushed silicate rock is applied on land or in the open ocean to accelerate natural weathering processes that absorb carbon dioxide from Earth’s ocean–atmosphere system. However, in addition to a range of potential environmental, socioeconomic, and ethical issues associated with this pathway, bottlenecks in feedstock sourcing represent a key barrier for deployment of EW at scale. Here, we evaluate the potential of silicate wastes produced from industrial processes—such as steel slag and cement waste—as feedstocks for the EW process. An empirical model that links industrial alkaline waste production to gross domestic product at purchase power parity is developed to forecast waste production in the alternative futures described by the shared socioeconomic pathway (SSP) framework. By incorporating these results into an intermediate-complexity Earth system model, we also explore the impacts of EW using industrial waste on changes to global temperature, ocean pH, and ocean aragonite saturation state, while also quantifying overall CDR efficiency through the end of the century. We estimate a maximum cumulative end-of-century capture potential of ∼400 GtCO 2 for industrial waste, which could represent a significant fraction of the projected CDR requirement of many mitigation scenarios in the SSP framework. However, feedstock-dependent environmental impacts and the technoeconomics of feedstock redistribution may ultimately limit deployment scope.

Xu, Pengxiao (ORCID:0009000633724293)↗
DOE OSTI2022

A deep learning approach to real-time HIV outbreak detection using genetic data

Pathogen genomic sequence data are increasingly made available for epidemiological monitoring. A main interest is to identify and assess the potential of infectious disease outbreaks. While popular methods to analyze sequence data often involve phylogenetic tree inference, they are vulnerable to errors from recombination and impose a high computational cost, making it difficult to obtain real-time results when the number of sequences is in or above the thousands. Here, we propose an alternative strategy to outbreak detection using genomic data based on deep learning methods developed for image classification. The key idea is to use a pairwise genetic distance matrix calculated from viral sequences as an image, and develop convolutional neutral network (CNN) models to classify areas of the images that show signatures of active outbreak, leading to identification of subsets of sequences taken from an active outbreak. We showed that our method is efficient in finding HIV-1 outbreaks with R0 ≥ 2.5, and overall a specificity exceeding 98% and sensitivity better than 92%. We validated our approach using data from HIV-1 CRF01 in Europe, containing both endemic sequences and a well-known dual outbreak in intravenous drug users. Our model accurately identified known outbreak sequences in the background of slower spreading HIV. Importantly, we detected both outbreaks early on, before they were over, implying that had this method been applied in real-time as data became available, one would have been able to intervene and possibly prevent the extent of these outbreaks. This approach is scalable to processing hundreds of thousands of sequences, making it useful for current and future real-time epidemiological investigations, including public health monitoring using large databases and especially for rapid outbreak identification.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2016

Methods for Finding Legacy Wells in Large Areas

United States. When abandoned, many wells were not adequately sealed and now provide a potential conduit for the vertical movement of liquids and gases. Today, groundwater aquifers can be contaminated by surface pollutants flowing down wells or by deep, saline water diffusing upwards. Likewise, natural gas, carbon dioxide (CO2), or radon can travel upwards via these wells to endanger structures or human health on the surface. Recently, the need to find and plug wells has become critical with the advent of carbon dioxide injection into geologic formations for enhanced oil recovery (EOR) or carbon storage. The potential for natural gas or brine leakage through existing wells has also been raised as a concern in regions where shale resources are hydraulically fractured for hydrocarbon recovery. In this study, the National Energy Technology Laboratory (NETL) updated existing, effective well finding techniques to be able to survey large areas quickly using helicopter or ground-vehicle-mounted magnetometers, combined with mobile methane detection. For this study, magnetic data were collected using airborne and ground vehicles equipped with two boom-mounted magnetometers, or on foot using a hand-held magnetometer with a single sensor. Data processing techniques were employed to accentuate well-casing-type magnetic signatures. To locate wells with no magnetic signature (wells where the steel well casing had been removed), the team monitored for anomalous concentrations of methane, which could indicate migration of volatile compounds from deeper sedimentary strata along a well or fracture pathway. Methane measurements were obtained using the ALPIS DIfferential Absorption Lidar (DIAL) sensor for helicopter surveys and the Apogee leak detection system (LDS) for ground surveys. These methods were evaluated at a 100-year-old oilfield in Wyoming, where a helicopter magnetic survey accurately located 93% of visible wells. In addition, 20% of the wells found by the survey were previously unknown or inaccurately located. This study found helicopter magnetic surveys to be an accurate, cost- and time-effective means to locate steel-cased wells in large areas, and is a first step in evaluating whether well detection techniques can be applied effectively for well location screening across broad geographic areas.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2017

Ultrawide‐Bandgap Semiconductors: Research Opportunities and Challenges

Abstract Ultrawide‐bandgap (UWBG) semiconductors, with bandgaps significantly wider than the 3.4 eV of GaN, represent an exciting and challenging new area of research in semiconductor materials, physics, devices, and applications. Because many figures‐of‐merit for device performance scale nonlinearly with bandgap, these semiconductors have long been known to have compelling potential advantages over their narrower‐bandgap cousins in high‐power and RF electronics, as well as in deep‐UV optoelectronics, quantum information, and extreme‐environment applications. Only recently, however, have the UWBG semiconductor materials, such as high Al‐content AlGaN, diamond and Ga 2 O 3 , advanced in maturity to the point where realizing some of their tantalizing advantages is a relatively near‐term possibility. In this article, the materials, physics, device and application research opportunities and challenges for advancing their state of the art are surveyed.

Tsao, J. Y.↗
DOE OSTI2023

Robust deep learning framework for constitutive relations modeling

Modeling the full-range deformation behaviors of materials under complex loading and materials conditions is a significant challenge for constitutive relations (CRs) modeling. Here, we propose a general encoder-decoder deep learning framework that can model high-dimensional stress-strain data and complex loading histories with robustness and universal capability. The framework employs an encoder to project high-dimensional input information (e.g., loading history, loading conditions, and materials information) to a lower-dimensional hidden space and a decoder to map the hidden representation to the stress of interest. We evaluated various encoder architectures, including gated recurrent unit (GRU), GRU with attention, temporal convolutional network (TCN), and the Transformer encoder, on two complex stress-strain datasets that were designed to include a wide range of complex loading histories and loading conditions. All architectures achieved excellent test results with an root-mean-square error (RMSE) below 1 MPa. Additionally, we analyzed the capability of the different architectures to make predictions on out-of-domain applications, with an uncertainty estimation based on deep ensembles. The proposed approach provides a robust alternative to empirical/semi-empirical models for CRs modeling, offering the potential for more accurate and efficient materials design and optimization.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Harnessing graph convolutional neural networks for identification of glassy states in metallic glasses

Graph Convolutional Neural Networks (GCNNs) have emerged as powerful tools for analyzing materials. In this study, we employ GCNNs to examine structural characteristics of CuZr metallic glasses (MGs) and identify their states. We use molecular dynamics to simulate the quenching process of CuZr, using cooling rates ranging from 10 9 to 10 15 K/s, to produce six unique glassy states. For each state, we create a dataset comprising 1,800 distinct samples. We evaluate the effectiveness of various GCNNs, including Graph Attention Neural Network (GANN), Graph Sample and AggreGatE (GraphSAGE), Graph Isomorphism Network (GIN), and Relational Graph Convolutional Neural Network (RGCN). GANN and GraphSAGE demonstrate comparable performance, achieving an overall accuracy of 81% in classifying the MG states. Furthermore, these results underscore the potential of GCNNs to detect subtle structural variances in disordered materials and point to broader application of deep learning in the analysis of MGs and other amorphous substances.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Modeling nuclear energy’s future role in decarbonized energy systems

Increased attention has been focused on the potential role of nuclear energy in future electricity markets and energy systems as stakeholders target rapid and deep decarbonization and reductions in fossil fuel use. This paper examines models of electric sector planning and broader energy systems optimization to understand the prospective roles of nuclear energy and other technologies. In this perspective, we survey modeling challenges in this environment, illustrate opportunities to propagate best practices, and highlight insights from the deep decarbonization literature on the range of visions for nuclear energy's role. Nuclear energy deployment is highest with combinations of stringent emissions policies, nuclear cost reductions, and constraints on the deployment of other technologies, which underscores model dimensions related to these areas. New modeling capabilities are needed to adequately address emerging issues, including representing characteristics and applications of nuclear energy in systems models, and to ensure the relevance of models for policy and planning as deeper decarbonization is explored.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2014

A model for arsenic anti-site incorporation in GaAs grown by hydride vapor phase epitaxy

GaAs growth by hydride vapor phase epitaxy (HVPE) has regained interest as a potential route to low cost, high efficiency thin film photovoltaics. In order to attain the highest efficiencies, deep level defect incorporation in these materials must be understood and controlled. The arsenic anti-site defect, As{sub Ga} or EL2, is the predominant deep level defect in HVPE-grown GaAs. In the present study, the relationships between HVPE growth conditions and incorporation of EL2 in GaAs epilayers were determined. Epitaxial n-GaAs layers were grown under a wide range of deposition temperatures (T{sub D}) and gallium chloride partial pressures (P{sub GaCl}), and the EL2 concentration, [EL2], was determined by deep level transient spectroscopy. [EL2] agreed with equilibrium thermodynamic predictions in layers grown under conditions in which the growth rate, R{sub G}, was controlled by conditions near thermodynamic equilibrium. [EL2] fell below equilibrium levels when R{sub G} was controlled by surface kinetic processes, with the disparity increasing as R{sub G} decreased. The surface chemical composition during growth was determined to have a strong influence on EL2 incorporation. Under thermodynamically limited growth conditions, e.g., high T{sub D} and/or low P{sub GaCl}, the surface vacancy concentration was high and the bulk crystal was close to equilibrium with the vapor phase. Under kinetically limited growth conditions, e.g., low T{sub D} and/or high P{sub GaCl}, the surface attained a high GaCl coverage, blocking As adsorption. This competitive adsorption process reduced the growth rate and also limited the amount of arsenic that incorporated as As{sub Ga}. A defect incorporation model which accounted for the surface concentration of arsenic as a function of the growth conditions, was developed. This model was used to identify optimal growth parameters for the growth of thin films for photovoltaics, conditions in which a high growth rate and low [EL2] could be attained.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2020

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↗