SEARCH · Search NASA

Results for “High-bandwidth”

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

SourceResearch connectionFrom yearThrough year

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

At least 73 records · Page 4

All sources NASA NTRS DOE OSTI

DOE OSTI2025

Pushing the limits of NAND technology scaling with ferroelectrics

Artificial intelligence (AI) continues to drive transformative advancements across various industries. The data-intensive nature of AI training (and inferencing) has resulted in the generation of unprecedented volumes of data with machine-generated content surpassing human-generated data by more than 100-fold in 2025. Efficiently managing this data influx necessitates advanced digital storage technologies. However, traditional NAND flash memory, which is critical for supporting data flows in AI systems—alongside high-bandwidth memory, for AI training—faces fundamental scaling limitations as it approaches the 1000-layer milestone, encompassing more than 40 trillion transistors. This article delves into the potential of hafnia-based ferroelectric materials as a breakthrough solution to these challenges. Recent advancements indicate that the intrinsic limitations of ferroelectric field-effect transistors (FEFETs) can be mitigated through material and device-level engineering. These advancements enable FEFETs to meet the stringent density, reliability, and scalability requirements of future three-dimensional NAND technology. The role of ferroelectrics in addressing NAND scaling challenges and expanding storage capabilities presents a promising avenue for meeting the storage demands of the AI-driven era.

3D NAND↗

DOE OSTI2022

Longitudinal Emittance Measurements at PIP2IT

The PIP-II particle accelerator is a new upgrade to the Fermilab accelerator complex, featuring an 800-MeV H⁻ superconducting linear accelerator that will inject the beam into the present Fermilab Booster. A test accelerator known as PIP-II Injector Test (PIP2IT) has been built to validate the concept of the front-end of PIP-II. One of the paramount challenges of PIP2IT was to demonstrate a low longitudinal emittance at the end of the front end. Having a low longitudinal emittance is crucial in order to ensure the stability of the beam in the accelerator. We present a longitudinal emittance calculation at 14.3 MeV at the SSR1-8 cavity in the High Energy Transport line (HEBT). The signal is collected by a Fast Faraday Cup (FFC) at the end of HEBT and recorded by a high-bandwidth oscilloscope.

43 PARTICLE ACCELERATORS↗

DOE OSTI2021

Plasma Photonics: Manipulating Light Using Plasmas (Full Technical Final Report)

In this LDRD, we made significant advances on multiple fronts on the investigation and design of plasma-based optical structures to manipulate light at extreme intensities. We have unveiled new fundamental properties of laser-plasma interaction that were leveraged to introduce transformative applications. We first investigated new ways to generate high-amplitude plasma-based optical structures, based on the auto-resonance mechanism in order to produce refractive structures in plasmas. New insights into the interaction of such plasma structures with light waves allowed us to propose new methods for polarization manipulation and control for high-intensity lasers in the nonlinear regime, i.e. allowing the manipulation of an intense laser via its interaction with a less intense “control” laser beam in plasmas. Another application that we demonstrated experimentally is the capability to manipulate the velocity of light waves in plasmas by tailoring its refractive index; we reported the first demonstration of “slow light” and supra-luminal light propagation in plasmas. We also developed a new technological concept that would allow the probing of the type of plasma structures investigated in such applications, using high-bandwidth laser probing [Ludwig Phys. Plasmas 2019]. Finally, first steps were taken towards the demonstration of a plasma-based amplifier and compressor for the National Ignition Facility, in order to achieve powers equal of greater than those delivered by the Advanced Radiographic Capability (ARC). This technique uses a plasma-based grating acting like a compressor and amplifier, which could potentially compress and amplify an 88 ps, 56 J beam from the National Ignition Facility to <10 ps and hundreds of J.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DOE OSTI2021

Roadmap for Advanced Power System Measurements

Conditions in the electric power grid are changing rapidly. Changes are driven, in particular, by the increased penetration of renewable resources in the generation mix. Part of the solution to dealing with the changing conditions is the better use of measurement systems, taking advantage of the better understanding of measurements that was heralded by digital technology. Another part of the solution is the communication of samples of the directly observable voltages and currents by means of continuous point-on-wave values. The increased bandwidth capability of such a system allows more complicated measurement models to be built at the point of use. Additional measurements and analyses can then be performed. This Roadmap hinges on these two aspects of measuring: improved measurements supported by high-bandwidth (raw data) communications. Following an introduction to the overall subject, the report explores the impact of what can be thought of as the digital revolution in measurement. In essence, that revolution has served to clarify ideas about measurement and is continuing to lead to improved processes. An extensive survey of the relevant literature was performed . Webinars were held to learn from the wider community of practitioners. Based on this information, a number of gaps were identified whose continued existence would hamper the solution of the challenges now facing measurement and communication in the power grid. The underlying value propositions are interoperability (presently not very satisfactory in many measurements); greater (and less challenging) adoption of distributed energy resources; and improved international competitiveness. Underlying all that, of course, is progress toward a more reliable and resilient power system. The identified gaps were addressed in four major thrusts: technology deployment and demonstration, education, instrumentation testing and standards, and analytical methods. For each of the thrusts, multiple goals are defined to guide the implementation. The thrusts were designed to close gaps and unlock high-value use cases. After discussing the literature review, gap analysis, high-value use cases, and research thrusts in chapters 2-5, the report culminates in chapter 6 with a roadmap that links each aspect together to form a unified plan. Developing a research portfolio aligned with this plan will ensure that advanced measurement systems are in place to support reliable operation of the power system as it modernizes.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2022

Hybrid Hydraulic-Electric Architecture (HHEA) for Mobile Machines (Final Report)

Traditionally, off-road mobile machines such as excavators and wheel loaders are primarily powered by hydraulics, and throttling valves are used to control their work circuits. In recent years, two general trends are towards more energy efficient systems and electrification. With electrification, both efficiency and control performance can be improved by the elimination of throttling losses and the use of high-bandwidth inverter control. Electrification is generally accomplished with Electro-hydraulic actuators (EHA) but they are limited to lower powered systems due to the high cost of electric machines capable of high power or high torque. This project proposes a new system architecture for off-road vehicles - Hybrid Hydraulic Electric Architecture (HHEA) to improve efficiency and control performance without requiring large electric machines. The widely applicable architecture combines hydraulic power and electric power in such a way that the majority of power is provided hydraulically while electric drives are used to modulate this power. In particular, HHEA utilizes multiple common pressure rails to transmit the majority of power and small electric machines to modulate the power. The energy-saving potential of the the HHEA has been validated for the work circuits of a variety of mobile machines, from small 5-ton excavators to medium sized 20-ton excavators and wheel loaders, and representative duty cycles to reduce energy input by 50-80% compared to the commercial state-of-art load-sensing systems. In addition, the corner power requirements of the electrical machines can be downsized by 85% compared to the EHA approach. Various tradeoff studies have also been conducted, including sensitivities to individual components performances, controllers, accumulator sizes, and variations of the system architecture etc. A control strategy has been developed to maintain or exceed the motion control precision compared to current systems. The motion control strategy consists of a nominal controller, based on a passivity-based backstepping design, and a transition controller, based on least-norn feedforward design. The nominal controller is used in between common pressure rail switches whereas the transition controller compensates for any disturbance that common pressure rail switchings inflict on the system. The control strategy has been experimentally validated on both a medium pressure (200bar) hardware-in-the-loop (HIL) testbed and a high pressure (300+bar) HIL testbed. An efficient and power-dense integrated electric-hydraulic machine consisting of an axial flux electric machine and a radial hydrostatic piston hydraulic machine has been designed, constructed and tested. The machine has an active material power density of 6.1kW/kg, a rated speed of 12500 RPM, and a design efficiency of 85%. This is among the highest power density electric machines using conventional materials. While the integrated machine was designed for modulating the hydraulic power within the HHEA, it can also be used in other applications.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2024

Intelligent, grid-friendly, modular extreme fast charging system with solid-state DC protection

The development of electric vehicle (EV) charging infrastructure is crucial for the widespread adoption of electric transportation. However, implementing such infrastructure is a complex task that requires consideration of factors such as space limitations, adherence to industry standards, grid capacity, and other technical and policy issues. This project seeks to create a framework for the efficient design of compact medium voltage (MV) extreme fast charging (XFC) stations for EVs. The station design involves the use of a solid-state transformer (SST) that connects to the MV distribution network, delivering power to a shared DC bus. This innovative approach eliminates the need for a step-down transformer to provide low-voltage service by connecting directly to the MV distribution network. Eliminating the low-frequency transformer not only reduces the system footprint and losses but also eliminates inrush currents during grid black-start. Additionally, placing power electronics directly on the distribution system allows for high-bandwidth filtering and power factor correction. The inclusion of a shared DC bus enables multiple charging dispensers and DC storage/generation units to connect, forming a DC microgrid. This setup facilitates power sharing with minimal conversion stages. The project showcases a DC distribution network protected by intelligent solid-state (SS) DC circuit breakers (DCCB) capable of isolating the smallest section of the faulted circuit much faster than existing mechanical solutions.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2023

IceNet for FireBox - A Berkeley Warehouse-Scale Computer

Berkeley’s FireBox is a next-generation warehouse-scale computer (WSC) that utilizes the energy-efficiency and bandwidth density of integrated silicon-photonic interconnects to enable a new high-bandwidth and low-latency network fabric connecting thousands of compute nodes to petabytes of DRAM and Flash storage. The high bandwidth, low latency and high connectivity of FireBox’s WSC network fabric (IceNet) will enable dramatic improvements in the overall system energy efficiency enabling fine-grain power control on system resources (processors, links and memory/storage components). IceNet is a special 3-stage photonic Clos network architected to achieve ultra-low-latency connectivity between processor nodes and memory, drastically cutting down on the energy wasted in resource idling (processors and memory stalled due to pending network requests). This is achieved by integration of the first and last switch stages into processor/memory hub clients and by heavy over-provisioning of the high-radix middle switches (FlareSwitches). A key hardware component developed in this program is an active laser power management photonic integrated circuits called LightSpark. It interacts with the FlareSwitch and provides laser power to a subset of occupied switch ports, increasing the utilization of laser light in the photonic network by an order of magnitude. In addition to guiding the laser power where it is needed, the laser-power management module enables both wavelength and laser redundancy, significantly increasing the robustness of the system. The goal of the IceNet fabric is to enable communication between 1000s of processor nodes and PBs of memory/storage with <100ns latency, <10pJ/b wall-plug energy cost at multiple Pb/s of available connectivity bandwidth. These metrics represent two-orders of magnitude improvement with respect to the status of current data-center technology.

42 ENGINEERING↗

DOE OSTI2025

Unified Universal Control and Coordination of Inverter-Based Resources, and Validation for a PV + Battery Hybrid Plant

As renewable energy deployment grows, hybrid power plants (HPPs) combining photovoltaic (PV) and battery systems must evolve to offer both energy and grid stability services. These systems typically include a mix of grid-following (GFL) and grid-forming (GFM) inverters, presenting unique coordination and control challenges. This Department of Energy–funded project developed and validated a Unified Universal Control and Coordination (UUCC) framework for such PV + battery hybrid plants, enabling seamless and stable operation, including ultrafast black start, autonomous synchronization, and robust frequency and voltage regulation, under different grid conditions. The project significantly advanced the understanding of inverter-based resource (IBR) control by developing and validating three complementary system-level approaches for hybrid GFL/GFM operation: 1. A combined Virtual Resistance (VR)-based GFL and Virtual Oscillator Control (VOC)-based GFM method, where each inverter type is governed by a specialized control strategy. Together, these achieve stable, fast-response coordination, eliminating inrush current and enabling smooth black start and grid synchronization across a wide range of grid strengths. 2. A Deadbeat-based UUCC strategy, which uses discrete-time, switching-cycle-level control for both GFL and GFM inverters. This approach replaces traditional PI/PLL control with a control parameter-free, high-bandwidth framework that supports stable LVRT and instantaneous synchronization under all conditions. 3. A benchmark comparison with Siemens’ commercial GFM microgrid controller, which provided a fast baseline platform. The commercial approach decoupled v & f control was implemented on a commercial microgrid controller.The baseline commercial benchmark helped highlight superior transient response and black start performance offered by the deadbeat and VOC approaches. These technical contributions offer substantial improvements over conventional inverter control schemes, which often rely on slow phase-locked loop (PLL)-based synchronization, require careful control parameters tuning, and prone to unstable in weak grids with GFL inverters and in stiff grid with GFM inverters therefore challenging for hybrid GFL+GFM under all grid conditions. The deadbeat-based UUCC framework enables simpler, faster, and more robust operation of hybrid IBR systems using wide-bandgap (WBG) devices such as SiC power semiconductors. The rapid expansion of hybrid distributed energy resources (DERs), including residential and commercial PV-BESS installations such as Tesla Powerwall, PV with vehicle-to-grid (V2G) capability, and other integrated configurations, presents complex operational challenges for medium-voltage radial distribution feeders. These networks are subject to frequent disturbances such as faults, switching operations, rapid reclosing sequences, and feeder reconfigurations, all of which introduce dynamic stress on IBRs. In addition, planned feeder segmentation and deliberate islanding for resilience will require DERs that can autonomously perform blackstart, establish voltage and frequency references, and resynchronize with the main grid. The advanced deadbeat-based UUCC control and blackstart functionalities developed in this project directly address these requirements, enabling decentralized and autonomous operation of inverter-dominated DERs in distribution systems under a wide range of fault and reconfiguration scenarios. From a public benefit perspective, these innovations enable more reliable and cost-effective integration of renewable energy into distribution networks. The ability to autonomously black start and stabilize grids under varying grid conditions support accelerates recovery from outages and support decentralized resilient energy systems. By reducing system complexity and improving performance, this project lays critical groundwork for future inverter-dominated power grids that are clean, reliable, and accessible to all.

14 SOLAR ENERGY↗

DOE OSTI2025

Measurement of Local Electric Field Turbulence in Magnetically Confined Plasmas (Final Report)

This final report describes the development of high-bandwidth plasma fluctuation diagnostics, encompassing the design and implementation of Beam Emission Spectroscopy systems on several fusion experiments such as TFTR, DIII-D and NSTX, the advancement of Alpha particle and Ultrafast Charge-Exchange Recombination Spectroscopy, and the creation of advanced spectrometers for localized studies of electric and magnetic field fluctuations based on the Motional Stark Effect.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DOE OSTI2025

Deployment and Evaluation of SciStream on OLCF's Advanced Computing Ecosystem (ACE)

The growing demand for real-time analysis, experimental steering, and decision-making in scientific workflows has created a need for tightly coupled integrations between experimental facilities and high-performance computing (HPC) systems. The Department of Energy’s Integrated Research Infrastructure (IRI) initiative highlights data streaming as a key capability for enabling memory-to-memory data transfers, bypassing the limitations of traditional store-and-forward models. SciStream is a toolkit developed by researchers at Argonne National Laboratory (ANL) to support such streaming by addressing cross-domain security, delegated authentication, and application transparency. We deployed and evaluated SciStream on the Oak Ridge Leadership Computing Facility’s (OLCF) Advanced Computing Ecosystem (ACE) infrastructure, leveraging the Olivine OpenShift cluster and its high-bandwidth Data Streaming Nodes (DSNs) as gateway nodes. Our evaluation included synthetic streaming workloads derived from IRI science workflows, a streaming simulator, and integration with RabbitMQ to handle low-level messaging. This report documents the deployment process, performance evaluation, and challenges encountered, along with opportunities for future improvements.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2025

Characterization of Peripheral Neurophotonic Systems for High-Performance Human-Computer Interfaces (CRADA Final Report)

As part of the Cyclotron Road program, Morphosis Inc. sought to investigate a non-invasive neuromuscular sensing approach for use as an intuitive and secure human–computer interface. These highly miniaturized, wearable neural interfaces were completely non-invasive and maintained stable, high-bandwidth, long-term access to a user’s actions, intent, and identity, while offering an exceptionally high signal-to-noise ratio compared to contemporary neural recording technologies. The widespread adoption of neural interfaces had the potential to reshape how people interact with technology, with profound societal impacts. Millions worldwide suffered from movement and/or speech disabilities, and these tools had the potential to democratize access to technology to enhance autonomy and quality of life. More broadly, interfaces capable of accurately conveying intentions and safeguarding identities could serve as a cornerstone for privacy, trust, and personal authenticity in digital environments. The use of thought-driven control of digitally enabled devices and governance of digital identities had the potential to revolutionize relationships with technology, transforming how people learn, communicate, and interact with the world.

42 ENGINEERING↗

DOE OSTI2025

Improved Charge Sensing on a SiMOS Double Quantum Dot using a Cryogenic Skipper Readout ASIC (Quandarum)

Major outstanding questions in high-energy physics such as the nature of dark matter and the existence of interactions beyond the standard model require new measurement techniques which are extremely sensitive to minute electromagnetic fields. An array of entangled spin qubits is a promising system for building novel detectors due to its combination of sensitivity and controllability. CMOS-based electron spin qubits, which have demonstrated the operational requirements for fault-tolerant quantum computing [1], offer a particular opportunity due to their compatibility with classical electronics, which allows the leveraging of decades of development of low-noise cryogenic detectors for physics. In this work, we combine a SiMOS double-quantum dot device architecture with a state-of-the-art cryoelectronic readout circuit [2-3] aimed to demonstrate improved charge readout using a single-electron transistor (SET). We identify the design characteristics for an SET that facilitate the use of on-chip classical electronics as a low-power, high-bandwidth first amplification stage and explore opportunities for sensor-readout co-design to minimize noise. This is the first of a series of steps to demonstrate high-fidelity readout of a large array of spin qubit with enough sensitivity to probe processes of interest for the investigation of beyond-standard-model physics.

Quinn, Adam [Fermilab]↗

DOE OSTI2022

Distributed Real-time Plume Monitoring for Deep Sea Mineral Extraction

In the emerging industry of deep-sea mining for minerals and deposits (e.g. polymetallic nodules for nickel, cobalt, copper, and manganese), more data is required to understand the effects of sediment plume generation and predict the distribution of disturbed sediment. There are two main sources of plume generation, the first being at the active mining site where the “collector” directly removes the top layer of the sea floor. The other is the “midwater plume” consisting of unwanted sediment that was collected during extraction that is pumped back into the aphotic zone. The vast majority of plume generation is caused by the collector, causing detrimental and long-lasting impacts on seafloor ecosystems due to the lack of wave activity or strong currents at the sea floor. Therefore, it is crucial to invest in the infrastructure to support the study and constant monitoring over a large area of the sea floor where plume generation is present. Due to the limited number of usable channels and power requirements, current subsea wireless communications technologies are not well suited to instrumenting the large areas of the sea floor needed to monitor plume migration. The scope of this effort is to transition experimental demonstrations of high-bandwidth, full-duplex scalable underwater laser communications to the seafloor in an open ocean environment. Specifically tackling challenges associated with the dynamic nature of the subsea world, including but not limited to, deployment logistics, sustainability, and range. The goal is to enable the internet of underwater things for deep sea industries by broadening the capabilities of subsea communications. By using high-precision laser transmitters, many of the challenges current subsea optical systems face can be circumvented, such as power consumption, interference, and bandwidth limitations. This approach lends itself to wireless interlinking multi-node networks, in series or parallel, facilitating the implementation of a wide array of sensor types. This interlinking allows all the data gathered from the network to be processed through a single hardline uplink to the surface, lowering the complexity required for near real-time data processing. Additionally, the laser control systems produce metadata that can be used to help characterize the water column between the nodes. Combining data from various sensors such as turbidity, temperature, current velocity with metadata such as beam attenuation and deflection can produce a high-resolution model of sea floor conditions around an active mining zone. The resulting near real-time model can be used to optimize location and flow rate of the mining operation to minimize and quantify the environmental impact.

Mons, Ishan↗

DOE OSTI2020

Generic characterization of electrical test benches for AC- and HVDC-connected wind power plants

The electrical test and assessment of wind turbines go hand in hand with standards and network connection requirements. In this paper, the generic structure of advanced electrical test benches, including grid emulator or controllable grid interface, wind torque emulator, and device under test, is proposed to harmonize state-of-the-art test sites. On the other hand, modern wind turbines are under development towards new features, concerning grid-forming, black-start, and frequency support capabilities as well as harmonic stability and control interaction considerations, to secure the robustness and stability of renewable-energy-based power systems. Therefore, it is necessary to develop new and revised test standards and methodologies to address the new features of wind turbines. This paper proposes a generic test structure within two main groups, including open-loop and closed-loop tests. The open-loop tests include the IEC 61400-21-1 standard tests as well as the additional proposed test options for the new capabilities of wind turbines, which replicate grid connection compliance tests using open-loop references for the grid emulator. In addition, the closed-loop tests evaluate the device under test as part of a virtual wind power plant and perform real-time simulations considering the grid dynamics. The closed-loop tests concern grid connection topologies consisting of AC and HVDC, as well as different electrical characteristics, including impedance, short-circuit ratio, inertia, and background harmonics. The proposed tests can be implemented using available advanced test benches by adjusting their control systems. The characteristics of a real power system can be emulated by a grid emulator coupled with real-time digital simulator systems through a high-bandwidth power-hardware-in-the-loop interface.

17 WIND ENERGY↗

DOE OSTI2020

Sensors with Intelligent Measurement Platform and Low-cost Equipment (SIMPLE) – Performance Characterization

Design and performance of a digital signal processing platform for enhancing off-the-shelf voltage and current sensors are presented. The platform was developed for enabling advanced distribution grid applications with existing low-cost sensors. This digital sensor system leverages a variety of off-the-shelf voltage and current sensing technologies, analog-to-digital conversion, and sensor correction algorithms to yield more accurate and reliable digital measurements that support a multitude of applications and use cases. Sample performance and accuracy results are provided, including measurements at power frequency and higher harmonics. The system extends the native performance of the medium voltage sensors used in the system, for example, broadening measurement bandwidth beyond a few kHz, appropriate for measuring switching surges and harmonics on medium voltage distribution systems. The digital sensor system is installed and characterized in the field with a high-bandwidth optical calibration system to determine site-dependent calibration and performance issues.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2021

Acoustic gas volume fraction measurement in a multiphase flowing liquid

Apparatus and methods for the measurement of gas volume fraction of produced oil are described. A first method measures the response of a pipe containing the produced oil excited by a source of vibration in the form of an acoustic frequency chirp containing a linearly varying range of frequencies in the tens of kilohertz range encompassing at least one resonant mode of the pipe. As the gas volume fraction increases, the location of the peak maximum of the measured frequency spectrum responsive to the excitation increases in frequency, and the height of the peak maximum increases, thereby permitting a linear calibration curve to be obtained. A second method measures the response of a pipe containing the produced oil to excitation by a continuous source of vibration having a chosen frequency above those which excite flexural vibrations in the pipe and simultaneously excite acoustic waves in the fluid contained in the pipe, known as the coincidence frequency. Gas present in the fluid will interrupt sound propagation or reverberation, thereby generating fluctuations in the amplitude of the measured vibrations of the pipe. The amplitude fluctuation level provides a measure of the gas volume present inside the pipe. A third method measures the response of a pipe containing the produced oil to excitation by a high-bandwidth, short pulse having a chosen center frequency above the coincidence frequency. Gas present in the fluid will interrupt pulse propagation, thereby generating fluctuations in the amplitude of the measured vibrations of the pipe.

Sinha, Dipen N.↗

DOE OSTI2020

Performance Potential of Mixed Data Management Modes for Heterogeneous Memory Systems

Many high-performance systems now include different types of memory devices within the same compute platform to meet strict performance and cost constraints. Such heterogeneous memory systems often include an upper-level tier with better performance, but limited capacity, and lower-level tiers with higher capacity, but less bandwidth and longer latencies for reads and writes. To utilize the different memory layers efficiently, current systems rely on hardware-directed, memory -side caching or they provide facilities in the operating system (OS) that allow applications to make their own data-tier assignments. Since these data management options each come with their own set of trade-offs, many systems also include mixed data management configurations that allow applications to employ hardware- and software-directed management simultaneously, but for different portions of their address space. Despite the opportunity to address limitations of stand-alone data management options, such mixed management modes are under-utilized in practice, and have not been evaluated in prior studies of complex memory hardware. In this work, we develop custom program profiling, configurations, and policies to study the potential of mixed data management modes to outperform hardware- or software-based management schemes alone. Our experiments, conducted on an Intel ® Knights Landing platform with high-bandwidth memory, demonstrate that the mixed data management mode achieves the same or better performance than the best stand-alone option for five memory intensive benchmark applications (run separately and in isolation), resulting in an average speedup compared to the best stand-alone policy of over 10 %, on average.

Effler, Chad↗

DOE OSTI2023

Runtime extension for neural network training with heterogeneous memory

Systems, apparatuses, and methods for managing buffers in a neural network implementation with heterogeneous memory are disclosed. A system includes a neural network coupled to a first memory and a second memory. The first memory is a relatively low-capacity, high-bandwidth memory while the second memory is a relatively high-capacity, low-bandwidth memory. During a forward propagation pass of the neural network, a run-time manager monitors the usage of the buffers for the various layers of the neural network. During a backward propagation pass of the neural network, the run-time manager determines how to move the buffers between the first and second memories based on the monitored buffer usage during the forward propagation pass. As a result, the run-time manager is able to reduce memory access latency for the layers of the neural network during the backward propagation pass.

Mappouras, Georgios↗