Search NASA⌕ Search

SEARCH · Search NASA

Results for “layout optimization”

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 235 records · Page 13
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

SMART SiC Power ICs: Scalable, Manufacturable, and Robust Technology for SiC Power Integrated Circuits (Final Technical Report)

This collaborative project was initiated with the goal of developing Scalable, Manufacturable, and Robust Technology for SiC Power Integrated Circuits (SMART SiC Power ICs). In pursuit of this objective, innovative designs and fabrication processes were implemented, enabling the development of large-scale (>1 cm²) SiC Complementary Metal-Oxide-Semiconductor (CMOS) integrated circuits and high-voltage (400–600 V) lateral power MOSFETs (HV-LDMOS) on 150 mm 4H-SiC substrates. The resulting SMART SiC Power ICs are tailored to support a wide range of applications requiring diverse voltage and power levels, including automotive systems, industrial equipment, electronic data processing, energy harvesting, and power conditioning. To achieve the proposed ‘SMART’ technology for SiC ICs, the team focused on 1) the Development of highly scalable CMOS (with high channel mobilities for n-type and p-type MOSFETs), LDMOS (~600V, 10A rated), and IC technologies, 2) Establishment of a manufacturable process baseline in a production-grade-, 150mm, SiC fabrication facility, and 3) Demonstration of SMART SiC ICs. The project initially comprised of fabricating 5 lots. In lot 1 monolithic integration using a single process was achieved. Here, we were able to successfully accomplish Integrated HV NMOSFET with LV CMOS on N-epi/N+ Substrate. The HV NMOS demonstrated a Breakdown Voltage (BV) more than 600V. Circuit demonstration of CMOS was also another achievement from this lot. In lot 2, priority was in place for isolation and integration. Here we addressed the isolation concerns and integrated the HV NMOS and LV CMOS using the N-epi/P-epi/N+ substrate. Similar to the lot 1, we were able to achieve a BV of 600 V for HV NMOS. Optimized gate oxide process with high channel mobilities, better gate oxide reliability, development of SPICE models, successful ohmic process development, novel wafer area saving design layouts, P+ isolation schemes with channeling implantations and high temperature operational circuits demonstrations are some of the key highlights from lot 1 and lot2. In lot 3, discrete device performances of HV NMOS with a BV ~700V and reliable LV CMOS performances were achieved. Also, novel architectural solutions were successfully implemented to suppress the electric field crowding at the gate oxide for reliable operations. In lot 4, half bridge power driver ICs with a conversion efficiency of (target 90% to 95%) in the 1-5MHz switching frequency range for output power between 25 W to 3 kW have been included in. However, due to the unfortunate events of sudden foundry shutdown (SiCamore Semi) the processing of lot 4 wafers came to a complete stop (January 2024). Arrangements have recently been made to shift the fabrication to another foundry, General Electric Aerospace. The fabrication process now on course (as of December 2024). Characterizations are delayed due to this unfortunate circumstance. The proposed trench architectural-based devices and ICs (lot 5) underwent modifications from the original project proposal. This change was necessitated by limitations in the availability of trench-based processes at commercial production-grade fabrication facilities in the US. Apart from above achievements, a Process Development Kit (PDK) was successfully developed for planar type SiC CMOS/LDMOS.

42 ENGINEERING↗
DOE OSTI2020

waveSZ: A Hardware-Algorithm Co-Design of Efficient Lossy Compression for Scientific Data

Error-bounded lossy compression is critical to the success of extreme-scale scientific research because of ever-increasing volumes of data produced by today's high-performance computing (HPC) applications. Not only can error-controlled lossy compressors significantly reduce the I/O and storage burden but they can retain high data fidelity for post analysis. Existing state-of-the-art lossy compressors, however, generally suffer from relatively low compression and decompression throughput (up to hundreds of megabytes per second on a single CPU core), which considerably restrict the adoption of lossy compression by many HPC applications especially those with a fairly high data production rate. In this paper, we propose a highly efficient lossy compression approach based on field programmable gate arrays (FPGAs) under the state-of-the-art lossy compression model SZ. Our contributions are fourfold. (1) We adopt a wavefront memory layout to alleviate the data dependency during the prediction for higher-dimensional predictors, such as the Lorenzo predictor. (2) We propose a co-design framework named WAVESZ based on the wavefront memory layout and the characteristics of SZ algorithm and carefully implement it by using high-level synthesis. (3) We propose a hardware-algorithm co-optimization method to improve the performance. (4) We evaluate our proposed WAVESZ on three real-world HPC simulation datasets from the Scientific Data Reduction Benchmarks and compare it with other state-of-the-art methods on both CPUs and FPGAs. Experiments show that our WAVESZ can improve SZ's compression throughput by 6.9x similar to 8.7x over the production version running on a state-of-the-art CPU and improve the compression ratio and throughput by 2.1x and 5.8x on average, respectively, respectively, compared with the state-of-the-art FPGA design.

Compression Ratio↗
DOE OSTI2014

SU-E-T-387: Achieving Optimal Patient Setup Imaging and Treatment Workflow Configurations in Multi-Room Proton Centers

Purpose: To simulate patient flow in proton treatment center under uncertainty and to explore the feasibility of treatment preparation rooms to improve patient throughput and cyclotron utilization. Methods: Three center layout scenarios were modeled: (S1: In-Tx room imaging) patient setup and imaging (planar/volumetric) performed in treatment room, (S2: Patient setup in preparation room) each treatment room was assigned with preparation room(s) that was equipped with lasers only for patient setup and gross patient alignment, and (S3: Patient setup and imaging in preparation room) preparation room(s) was equipped with laser and volumetric imaging for patient setup, gross and fine patient alignment. A 'snap' imaging was performed in treatment room. For each scenario, the number of treatment rooms and the number of preparation rooms serving each treatment room were varied. We examined our results (average of 100 16-hour (two shifts) working days) by evaluating patient throughput and cyclotron utilization. Results: When the number of treatment rooms increased ([from, to]) [1, 5], daily patient throughput increased [32, 161], [29, 184] and [27, 184] and cyclotron utilization increased [13%, 85%], [12%, 98%], and [11%, 98%] for scenarios S1, S2 and S3 respectively. However, both measures plateaued after 4 rooms. With the preparation rooms, the throughput and the cyclotron utilization increased by 14% and 15%, respectively. Three preparation rooms were optimal to serve 1-3 treatment rooms and two preparation rooms were optimal to serve 4 or 5 treatment rooms. Conclusion: Patient preparation rooms for patient setup may increase throughput and decrease the need for additional treatment rooms (cost effective). Optimal number of preparation rooms serving each gantry room varies as a function of treatment rooms and patient setup scenarios. A 5th treatment room may not be justified by throughput or utilization.

43 PARTICLE ACCELERATORS↗
DOE OSTI2020

Advanced Gas Turbine and sCO 2 Combined Cycle Power System (Final Report)

The team of Southwest Research Institute (SwRI), Hanwha Power Systems Americas (HPSA), and Solar Turbines Incorporated (Solar) has worked to develop a modular, highly efficient combined-cycle power system consisting of a supercritical carbon dioxide (sCO 2 )-based waste-heat recovery system (WHRS) coupled to a Solar Titan 130 gas turbine (GT). During Phase 1 of the project, the project team used advanced thermodynamic cycle and techno-economic modeling to find an optimal sCO 2 WHRS bottoming cycle and machinery configuration for a Solar Titan 130. In addition to this, a conceptual design of the machinery and overall system layout was created and technology gaps were identified.

30 DIRECT ENERGY CONVERSION↗
DOE OSTI2024

Rational Optimization of Microbial Processing for High Yield CO 2 -to-Isopropanol Conversion: Cooperative Research and Development Final Report, CRADA Number CRD-20-17114

This project focuses on the production of the fuel blendstock isopropanol using a CO 2 -fixing Clostridium by metabolic engineering and process optimizations. The project will initiate from a baseline isopropanol producer and pursue isopropanol production at high carbon-conversion efficiency. We will lead engineering work by in-depth pathway analyses including thermodynamics optimization, enzyme expense analysis, metabolic robustness analysis, and -omics analysis. The isopropanol production will be optimized via genome editing followed by fermentation optimizations. This project will deliver a novel microbial process that efficiently converts waste CO 2 to isopropanol at ~g/L titer level within 18-months. This project will layout a solid knowledge basis and technology platform for renewable CO 2 valorization to bio-blendstock that help achieve Co-Optima and Shell's goals.

09 BIOMASS FUELS↗
DOE OSTI2026

SOLPS-ITER modeling of a dedicated divertor for negative triangularity operation on DIII-D

The design of a new dedicated divertor for negative triangularity (NT) operation on DIII-D with neutral baffles and pumping is informed by SOLPS-ITER transport modeling. This dedicated NT divertor is the latest step in a progression of NT shapes with various divertor characteristics explored on DIII-D, including NT shapes at reduced triangularity and a campaign with stronger shaping that included new armored components on the outboard side. SOLPS simulations played a key role in these divertor designs. Interpretive simulations, using cross-field diffusivities constrained by experimental data in the NT Shelf shape were used to inform the design of the 2023 armor campaign components. A similar procedure used armor campaign data to predict conditions for the dedicated NT divertor. The predictive simulations were used to assess the divertor fluxes, detachment threshold, pumped flux, and neutral leakage. For the dedicated NT divertor, SOLPS simulations and two-point-modeling were used to show the relative impact of magnetic topology (mainly longer connection length) and divertor closure on the divertor conditions relative to the armor campaign. It is predicted that the dedicated NT divertor reaches detachment (measured by target ion flux rollover) at a lower upstream density (≈(1.75−−2.4)×1⁢019m−3) as compared to the armor campaign shape. For the preliminary design geometry, divertor closure reduces the neutral leakage by ≈10%. Parametric optimization indicating further ≈20%–60% improvement in the leakage flux and recycled flux crossing the pump entrance is possible for relatively minor changes to the divertor and baffle layout.

Lore, Jeremy [ORNL] (ORCID:000000029192465X)↗
DOE OSTI2021

A Reconfigurable Neural Network ASIC for Detector Front-End Data Compression at the HL-LHC

Despite advances in the programmable logic capabilities of modern trigger systems, a significant bottleneck remains in the amount of data to be transported from the detector to off-detector logic where trigger decisions are made. We demonstrate that a neural network (NN) autoencoder model can be implemented in a radiation-tolerant application-specific integrated circuit (ASIC) to perform lossy data compression alleviating the data transmission problem while preserving critical information of the detector energy profile. For our application, we consider the high-granularity calorimeter from the Compact Muon Solenoid (CMS) experiment at the CERN Large Hadron Collider. The advantage of the machine learning approach is in the flexibility and configurability of the algorithm. By changing the NN weights, a unique data compression algorithm can be deployed for each sensor in different detector regions and changing detector or collider conditions. To meet area, performance, and power constraints, we perform quantization-aware training to create an optimized NN hardware implementation. The design is achieved through the use of high-level synthesis tools and the hls4ml framework and was processed through synthesis and physical layout flows based on a low-power (LP)-CMOS 65-nm technology node. The flow anticipates 200 Mrad of ionizing radiation to select gates and reports a total area of 3.6 mm 2 and consumes 95 mW of power. The simulated energy consumption per inference is 2.4 nJ. Furthermore, this is the first radiation-tolerant on-detector ASIC implementation of an NN that has been designed for particle physics applications.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2022

Simulation of X-ray Hartmann wavefront sensing with the Synchrotron Radiation Workshop

X-ray wavefront measurement is an important beam diagnostic tool, especially for the diffraction-limited X-ray beam. These in situ diagnostics give a better understanding of beam imperfections, and they enable feedback for possible corrections and/or optical alignment improvements. Hartmann wavefront sensing is one of the promising techniques to perform in situ X-ray wavefront measurements. In this work, a simulation tool of the X-ray Hartmann Wavefront Sensor (HWS) is developed under the Synchrotron Radiation Workshop (SRW) framework. Using this new simulation capability, one can take advantage of the full SRW package to simulate Hartmann wavefront sensing with the beam traveling from the X-ray source to the sample through different X-ray optical components. This SRW HWS simulation tool can help to optimize the wavefront sensor parameters for a specific X-ray energy range. It can also simulate an in situ wavefront measurement experiment with a particular beamline optical layout and predict the expected results of the wavefront measurement under different beamline configurations.

36 MATERIALS SCIENCE↗
DOE OSTI2022

OPTICS FOR STRONG HADRON COOLING IN EIC HSR-I

Insertion Region 2 (IR2) of the Relativistic Heavy Ion Collider will be modified to accommodate a Strong Hadron Cooling facility in the Hadron Storage Ring (HSR) of the Electron-Ion Collider (EIC). This paper describes the current proof-of-principle design of HSR-IR2 - layout, optical performance, design methodology, and engineering requirements. It also describes the challenges and opportunities in the future development of the HSR-IR2 design, in order to further optimize Strong Hadron Cooling performance.

Peggs, S.↗
DOE OSTI2026

Demonstration of Algorithm for Sensor Placement Optimization using Simulation Data

This deliverable reports FY26 progress in advancing a neural-network-based Green’s-function framework for reconstructing neutron-flux distributions from ex-core measurements and for translating reconstruction requirements into a practical detector-layout strategy. Building on the FY25 formulation, the present work had two main objectives: (1) refine and re-evaluate the reconstruction methodology on an updated Purdue University Reactor Number One (PUR-1) model, and (2) develop a systematic, Green’s-function-guided procedure for boundary detector placement.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2023

Modeling, analysis, and optimization of complex nuclear processes and facilities via computational methods: The HALEU process case study

Improving and adapting industrial systems to timely meet changing programmatic and market demands is an important goal to achieve, including when operating and maintaining complex nuclear processes and facilities. However, changes to these complex systems are costly, particularly when they are already in place and bounded to stringent requirements and constraints such as when handling radioactive material and contaminated equipment. These conditions often exist when treating spent nuclear fuel remotely within shielded nuclear radiation chambers, commonly referred as hot cells, to condition nuclear material and/or fabricate products for utilization in other nuclear enterprises such as in the manufacture of advanced nuclear fuel. The illustrative case considered here is the production of high assay low enriched uranium (HALEU) products supporting the deployment of advanced nuclear reactors. For the HALEU program, resources invested were and are being systematically analyzed so that these investments are maximized in a facility that is nearly 60 years old. A methodology that has effectively enabled optimized and improvements in the Spent Fuel Treatment (SFT) program, and consequently the HALEU program, involves discrete event simulation as addressed in this article. Here, the quantification of multiple productivity metrics, including material processing rates, cycle times, bottlenecks, number of material transfers as well as equipment, workstation, and material handling utilization, has resulted in a myriad of diverse discoveries and data-informed decisions regarding process layout and constituent, labor levels and schedules, selection of new process units, storage needs, and other critical process configurations. This article describes such a computational capability being applied for decision-making, illustrates its application to an actual process and program, provides illustrative results, and argues how computational methods for the modeling, analysis, and optimization of complex processes and facilities does lead to informed decisions derived from data and not only from intuition.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2013

QCAD Simulation and Optimization of Semiconductor Double Quantum Dots

We present the Quantum Computer Aided Design (QCAD) simulator that targets modeling quantum devices, particularly silicon double quantum dots (DQDs) developed for quantum qubits. The simulator has three differentiating features: (i) its core contains nonlinear Poisson, effective mass Schrodinger, and Configuration Interaction solvers that have massively parallel capability for high simulation throughput, and can be run individually or combined self-consistently for 1D/2D/3D quantum devices; (ii) the core solvers show superior convergence even at near-zero-Kelvin temperatures, which is critical for modeling quantum computing devices; (iii) it couples with an optimization engine Dakota that enables optimization of gate voltages in DQDs for multiple desired targets. The Poisson solver includes MaxwellBoltzmann and Fermi-Dirac statistics, supports Dirichlet, Neumann, interface charge, and Robin boundary conditions, and includes the effect of dopant incomplete ionization. The solver has shown robust nonlinear convergence even in the milli-Kelvin temperature range, and has been extensively used to quickly obtain the semiclassical electrostatic potential in DQD devices. The self-consistent Schrodinger-Poisson solver has achieved robust and monotonic convergence behavior for 1D/2D/3D quantum devices at very low temperatures by using a predictor-correct iteration scheme. The QCAD simulator enables the calculation of dot-to-gate capacitances, and comparison with experiment and between solvers. It is observed that computed capacitances are in the right ballpark when compared to experiment, and quantum confinement increases capacitance when the number of electrons is fixed in a quantum dot. In addition, the coupling of QCAD with Dakota allows to rapidly identify which device layouts are more likely leading to few-electron quantum dots. Very efficient QCAD simulations on a large number of fabricated and proposed Si DQDs have made it possible to provide fast feedback for design comparison and optimization.

77 NANOSCIENCE AND NANOTECHNOLOGY↗
DOE OSTI2024

Simulation of the Multi-Wake Evolution of Two Sandia National Labs/National Rotor Testbed Turbines Operating in a Tandem Layout

The future of wind power systems deployment is in the form of wind farms comprised of scores of such large turbines, most likely at offshore locations. Individual turbines have grown in span from a few tens of meters to today’s large turbines with rotor diameters that dwarf even the largest commercial aircraft. These massive dynamical systems present unique challenges at scales unparalleled in prior applications of wind science research. Fundamental to this effort is the understanding of the wind turbine wake and its evolution. Furthermore, the optimization of the entire wind farm depends on the evolution of the wakes of different turbines and their interactions within the wind farm. In this article, we use the capabilities of the Common ODE Framework (CODEF) model for the analysis of the effects of wake–rotor and wake-to-wake interactions between two turbines situated in a tandem layout fully and partially aligned with the incoming wind. These experiments were conducted in the context of a research project supported by the National Rotor Testbed (NRT) program of Sandia National Labs (SNL). Results are presented for a layout which emulates the turbine interspace and relative turbine emplacement found at SNL’s Scaled Wind Technologies Facility (SWiFT), located in Lubbock, Texas. The evolution of the twin-wake interaction generates a very rich series of secondary transitions in the vortex structure of the combined wake. These ultimately affect the wake’s axial velocity patterns, altering the position, number, intensity, and shape of localized velocity-deficit zones in the wake’s cross-section. This complex distribution of axial velocity patterns has the capacity to substantially affect the power output, peak loads, fatigue damage, and aeroelastic stability of turbines located in subsequent rows downstream on the farm.

Baruah, Apurva (ORCID:0000000252354068)↗
DOE OSTI2022

Design concept for a second interaction region for the Electron-Ion Collider

In addition to the day-one primary Interaction Region (IR), the design of the Electron Ion Collider (EIC) must support operation of a 2nd IR potentially added later. The 2nd IR is envisioned in an existing experimental hall at RHIC IP8, compatible with the same beam energy combinations as the 1st IR over the full center of mass energy range of ~20 GeV to ~140 GeV. The 2nd IR is designed to be complementary to the 1st IR. In particular, a secondary focus is added in the forward ion direction of the 2nd IR hadron beamline to optimize its capability in detecting particles with magnetic rigidities close to those of the ion beam. We provide the current design status of the 2nd IR in terms of parameters, magnet layout and beam dynamics.

Gamage, B. R.↗
DOE OSTI2017

Optimal reblocking as a practical tool for neighborhood development

Fast urbanization is a common feature of many developing human societies. In many cases, past and present, explosive population growth in cities outstrips the rate of provision of housing and urban services and leads to the formation of informal settlements or slums. Slums are extremely varied in terms of their histories, infrastructure, and rates of change, but they share certain common features: informal land use, lack of physical accesses, and nonexistent or poor quality urban services. Currently, about 1 billion people worldwide live in slums, a number that could triple by 2050 if no practical solutions are enacted to reverse this trend. Underlying most problems of slums is the issue of lack of physical accesses to places of work and residence. This prevents residents and businesses from having an address, obtaining basic services such as water and sanitation, and being helped in times of emergency. In this paper, we show how the physical layout of any neighborhood can be classified quantitatively in terms of its access topology in a way that is independent of its geometry. Topological indices capturing levels of access to structures within a city block can then be used to define a constrained optimization problem, whose solution generates an access network that makes each structure in the settlement accessible to services with minimal disruption and cost. We discuss the general applicability of these techniques to several informal settlements in developing cities and demonstrate various technical aspects of our solutions. In conclusion, we discuss how these techniques could be used on a large scale to speed up human development processes in cities throughout the world while respecting their local identity and history.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Hydrogen Energy Storage Integrated with a Combined Cycle Plant

A project is being developed that will build upon the existing infrastructure and resources at the Intermountain Power Project (IPP) site to provide reliable, dispatchable energy and to support the transmission of renewable energy resources while transitioning to an economical green energy future. The concept study depicted in this report outlines a techno-economic optimization to fulfill the demand for 30% vol hydrogen co-firing in the IPP 840 MW advanced class combined cycle power plant. In an initial step, a site assessment concluded the site has sufficient land available to co-locate a hydrogen production and storage facility. The team evaluated and defined a scalable concept that considered technology characteristics, including input and output models to be used for optimization purposes. The concept for the hydrogen production and storage system integrates multiple technologies, to determine system size and scalable approach, for each of the technologies evaluated, the team defined component and subcomponent sizes, minimum and maximum capacity, modularity, component utility consumption (electric, water), component flexibility and servicing, layout, and technology status, as well as technology alternatives. For hydrogen generation, the project considers Siemens Energy’s Silyzer-300 (S300) technology, a 17.5 MW modular Proton Exchange Membrane (PEM) electrolyzer. For the S300 configuration, the team determined that three S300 arrays, or approximately 1,000 kg/hr, per block would yield a compact block design. This configuration results in a fairly wide and flexible arrangement that fits well into the spaces available at the site. Therefore, the overall design approach is based on multiple identical blocks of 3 arrays to minimize engineering cost and optimize constructability. In parallel, a transmission screening study was conducted to determine any potential transmission constraints from the energy sources that could feed the hydrogen production equipment. The study results show that minimum transmission constraints would be encountered to deliver 400 MW renewable generation from southern California, or south-central Wyoming. At last, the techno-economic analysis concluded that a scenario that uses solar and wind power yields the lowest levelized cost of hydrogen (LCOH 2 ) production and the lowest cost per tonne of CO 2 reduced. In this optimized scenario, the hydrogen production plant was determined as 6,201 kg/hr and the hydrogen storage (underground cavern) was determined as 4,600 tonnes. The resulting capacity factor for the hydrogen production plant was 66.33% with 8,745 operating hours in one year. This techno-economic analysis provided various options for integrating hydrogen storage at the Intermountain Power Plant site to co-fire the CCPP units. The results provide insightful data about the magnitude of capacity needed and the economics of producing hydrogen and reducing CO 2 emissions.

08 HYDROGEN↗
DOE OSTI2024

Exploring code portability solutions for HEP with a particle tracking test code

Traditionally, high energy physics (HEP) experiments have relied on x86 CPUs for the majority of their significant computing needs. As the field looks ahead to the next generation of experiments such as DUNE and the High-Luminosity LHC, the computing demands are expected to increase dramatically. To cope with this increase, it will be necessary to take advantage of all available computing resources, including GPUs from different vendors. A broad landscape of code portability tools—including compiler pragma-based approaches, abstraction libraries, and other tools—allow the same source code to run efficiently on multiple architectures. In this paper, we use a test code taken from a HEP tracking algorithm to compare the performance and experience of implementing different portability solutions. While in several cases portable implementations perform close to the reference code version, we find that the performance varies significantly depending on the details of the implementation. Achieving optimal performance is not easy, even for relatively simple applications such as the test codes considered in this work. Several factors can affect the performance, such as the choice of the memory layout, the memory pinning strategy, and the compiler used. The compilers and tools are being actively developed, so future developments may be critical for their deployment in HEP experiments.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2023

Accelerating network layouts using graph neural networks

Graph layout algorithms used in network visualization represent the first and the most widely used tool to unveil the inner structure and the behavior of complex networks. Current network visualization software relies on the force-directed layout (FDL) algorithm, whose high computational complexity makes the visualization of large real networks computationally prohibitive and traps large graphs into high energy configurations, resulting in hard-to-interpret “hairball” layouts. Here we use Graph Neural Networks (GNN) to accelerate FDL, showing that deep learning can address both limitations of FDL: it offers a 10 to 100 fold improvement in speed while also yielding layouts which are more informative. We analytically derive the speedup offered by GNN, relating it to the number of outliers in the eigenspectrum of the adjacency matrix, predicting that GNNs are particularly effective for networks with communities and local regularities. Finally, we use GNN to generate a three-dimensional layout of the Internet, and introduce additional measures to assess the layout quality and its interpretability, exploring the algorithm’s ability to separate communities and the link-length distribution. The novel use of deep neural networks can help accelerate other network-based optimization problems as well, with applications from reaction-diffusion systems to epidemics.

97 MATHEMATICS AND COMPUTING↗