Search NASA⌕ Search

SEARCH · Search NASA

Results for “Accelerator design”

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 343 records · Page 19

An MLIR-based Compiler Flow for System-Level Design and Hardware Acceleration

The generation of custom hardware accelerators for applications implemented within high-level productive programming frameworks requires considerable manual effort. To automate this process, we introduce \sodaopt, a compiler tool that extends the MLIR infrastructure. \sodaopt automatically searches, outlines, tiles, and pre-optimizes relevant code regions to generate high-quality accelerators through high-level synthesis. \sodaopt can support any high-level programming framework and domain-specific language that interface with the MLIR infrastructure. By leveraging MLIR, \sodaopt solves compiler optimization problems with specialized abstractions. Backend synthesis tools connect to \sodaopt through progressive intermediate representation lowerings. \sodaopt interfaces to a design space exploration engine to identify the combination of compiler optimization passes and options that provides high-performance generated designs for different backends and targets. We demonstrate the practical applicability of the compilation flow by exploring the automatic generation of accelerators for deep neural networks operators outlined at arbitrary granularity and by combining outlining with tiling on large convolution layers. Experimental results with kernels from the PolyBench benchmark show that \sodaopt high-level optimizations improve execution delays of synthesized accelerators up to 60x. We also show that for the selected kernels, our solution outperforms the current of state-of-the art in more than 70% of the benchmarks and provides better average speedup in 55% of them.

Bohm Agostini, Nicolas↗

A Revised Capsule Design for the Accelerated Testing of Advanced Reactor Fuels

Testing of fast reactor fuel is currently challenging because of the thermal neutron spectrum of the Advanced Test Reactor (ATR) and the fact that ATR coolant is substantially below typical temperatures found in liquid metal-cooled fast reactors. Despite these issues, since 2003 the Idaho National Laboratory (INL) has successfully tested a variety of metallic, oxide, and nitride fuels for fast reactor applications. There are, however, opportunities to improve the experiments so that the time required to reach a desired burnup can be reduced, and so the sensitivity of the experiments to fabrication tolerances can be minimized. A revised capsule design exploits the use of smaller diameter rodlets to increase power densities and correspondingly reduce the irradiation time required to reach high burnup. Thermal and nuclear analyses indicate that the time to reach 30at% burnup could be reduced from about 12 years to approximately 2-3 years if the fuel diameter is reduced by one-half and to around 1-2 years if the diameter is reduced by one-third. In addition, reducing fuel diameter improves the radial power distribution in the fuel without the use of cadmium shrouding. This observation opens up the possibility of testing advanced reactor fuel systems in Small I locations in the ATR where there is substantial unused capacity. Although not directly tied to a revised capsule design, post-irradiation furnace testing is proposed to provide time-to-failure data that will be helpful in the benchmarking of fuel performance models and in assessing the performance of new fuel designs relative to more standard fast reactor fuel systems. In this paradigm, ATR irradiation experiments will be viewed more as preparation for furnace testing than as experiments that exist in isolation. An added benefit to furnace testing is that uncertainties in temperatures will significantly be reduced relative to non-instrumented capsules in the ATR. The time and cost in developing this experimental capability is expected to be reduced considerably because of the availability of the blister anneal furnace used by the HPRR program. A key issue in the experiment approach is the potential for non-prototypical rodlet dimensions to influence the observed performance of a given fuel design. For example, a qualitative assessment indicates that the effects of fuel-cladding chemical interactions during furnace testing might be more pronounced in reduced diameter rodlets simply because thinner cladding would be more sensitive to a given depth of fission product penetration. In a traditional empirically-driven approach to fuel qualification, any change from prototypical behavior is undesirable because the resulting data cannot easily be assimilated into a statistical correlation. However, in the current approach, systematic variations in fuel diameter provide the opportunity to benchmark fuel performance models over a wider range of parameter space, which in turn is expected to improve understanding and result in better, more physically based models in the long term.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

NoC-enabled Software/Hardware Co-Design Framework for Accelerating k-mer Counting

Counting k-mers (substrings of fixed length k) in DNA and protein sequences generate non-uniform and irregular memory access patterns. Processing-in-Memory (PIM) architectures have potential to significantly reduce the overheads associated with such frequent and irregular memory accesses. However, existing k-mer counting algorithms are not designed to exploit the advantages of PIM architectures. Furthermore, owing to thermal constraints, the allowable power budget is limited in conventional PIM designs. Moreover, k-mer counting generates long-range, non-uniform traffic patterns which needs to be handled by an efficient Network-on-Chip (NoC). In this paper, we present an NoC-enabled hardware/software co-design framework to implement high-performance k-mer counting. The proposed architecture enables more computational power, efficient communication between cores/memory – all without creating a thermal bottleneck. The software component overcomes the I/O bottleneck, aids in NoC design and exposes more in-memory opportunities to exploit in PIM. Experimental results show that the proposed architecture outperforms a state-of-the-art software implementation using Hybrid Memory Cube (HMC), by up to 7.14X, while allowing significantly higher power budget.

Joardar, Biresh K.↗

Flexible silicon photonic architecture for accelerating distributed deep learning

The increasing size and complexity of deep learning (DL) models have led to the wide adoption of distributed training methods in datacenters (DCs) and high-performance computing (HPC) systems. However, communication among distributed computing units (CUs) has emerged as a major bottleneck in the training process. In this study, we propose Flex-SiPAC, a flexible silicon photonic accelerated compute cluster designed to accelerate multi-tenant distributed DL training workloads. Flex-SiPAC takes a co-design approach that combines a silicon photonic hardware platform with a tailored collective algorithm, optimized to leverage the unique physical properties of the architecture. The hardware platform integrates a novel wavelength-reconfigurable transceiver design and a micro-resonator-based wavelength-reconfigurable switch, enabling the system to achieve flexible bandwidth steering in the wavelength domain. The collective algorithm is designed to support reconfigurable topologies, enabling efficient all-reduce communications that are commonly used in DL training. The feasibility of the Flex-SiPAC architecture is demonstrated through two testbed experiments. First, an optical testbed experiment demonstrates the flexible routing of wavelengths by shuffling an array of input wavelengths using a custom-designed spatial-wavelength selective switch. Second, a four-GPU testbed running two DL workloads shows a 23% improvement in job completion time compared to a similarly sized leaf-spine topology. We further evaluate Flex-SiPAC using large-scale simulations, which show that Flex-SiPAC is able to reduce the communication time by 26% to 29% compared to state-of-the-art compute clusters under representative collective operations.

Wu, Zhenguo (ORCID:0000000322847985)↗

Collaboration for Advanced Modeling of Particle Accelerators

The pverarching purpose is to accelerate and expand the scope of discoveries from high energy physics (HEP) particle accelerators by enabling the design of accelerators that are significantly more compact and cheaper to build and run. This will be realized through (i) developing high-performance computing (HPC) accelerator and beam modeling capabilities to design the full range of systems required (ii) developing community simulation ecosystems that seamlessly integrate accelerator elements to facilitate the design and control of next-generation accelerators.

43 PARTICLE ACCELERATORS↗

Magnetohydrodynamic Augmented Propulsion Experiment

A fundamental obstacle to routine space access is the specific energy limitations associated with chemical fuels. In the case of vertical take-off, the high thrust needed for vertical liftoff and acceleration to orbit translates into power levels in the 10 GW range. Furthermore, useful payload mass fractions are possible only if the exhaust particle energy (i.e., exhaust velocity) is much greater than that available with traditional chemical propulsion. The electronic binding energy released by the best chemical reactions (e.g., LOX/LH2 for example, is less than 2 eV per product molecule (approx. 1.8 eV per H2O molecule), which translates into particle velocities less than 5 km/s. Useful payload fractions, however, will require exhaust velocities exceeding 15 km/s (i.e., particle energies greater than 20 eV). As an added challenge, the envisioned hypothetical RLV (reusable launch vehicle) should accomplish these amazing performance feats while providing relatively low acceleration levels to orbit (2-3g maximum). From such fundamental considerations, it is painfully obvious that planned and current RLV solutions based on chemical fuels alone represent only a temporary solution and can only result in minor gains, at best. What is truly needed is a revolutionary approach that will dramatically reduce the amount of fuel and size of the launch vehicle. This implies the need for new compact high-power energy sources as well as advanced accelerator technologies for increasing engine exhaust velocity. Electromagnetic acceleration techniques are of immense interest since they can be used to circumvent the thermal limits associated with conventional propulsion systems. This paper describes the Magnetohydrodynamic Augmented Propulsion Experiment (MAPX) being undertaken at NASA Marshall Space Flight Center (MSFC). In this experiment, a 1-MW arc heater is being used as a feeder for a 1-MW magnetohydrodynamic (MHD) accelerator. The purpose of the experiment is to demonstrate that an MHD accelerator can be an effective augmentation system for increasing engine exhaust velocity. More specifically, the experiment is intended to show that electromagnetic effects are effective at producing flow acceleration whereas electrothermal effects do not cause unacceptable heating of the working fluid. The MHD accelerator was designed as an externally diagonalized segmented Faraday channel, which will be inserted into an existing 2-tesla electromagnet. This allows the external power to be connected through two terminals thereby minimizing the complexity and cost associated with powering each segment independently. The design of the accelerator and other components in the flow path has been completed and fabrication activities are underway. This paper provides a full description of MAPX including performance analysis, design, and test plans, and current status.

Litchford, Ron J.↗

A Revised Capsule Design for the Accelerated Testing of Advanced Reactor Fuels

The Advanced Test Reactor (ATR) has been used successfully for the testing of fast reactor fuel for nearly two decades. These successes have been in spite of numerous challenges for testing fast reactor fuel in the ATR (a thermal spectrum reactor), but the solutions to those challenges have resulted in excessively long irradiation times (~10 years) for high-burnup targets as well as experiments that are highly sensitive to fabrication tolerances and eccentricities. This paper presents a solution to the problems of extended irradiation times and fabrication sensitivities. Thermal and neutronic analyses were performed to show that a reduced-diameter fuel pin with an equivalent linear heat generation rate can provide a prototypic thermal profile (peak centerline and inner clad temperature) along with a near-prototypic power profile within the ATR thermal spectrum. This allows the experiment to reach a high burnup in an expeditious timeframe compared to traditional ATR fast fuel irradiations. In addition, problems with fabrication sensitivities were addressed by introducing a double-encapsulated experiment that pushes the high heat flux helium gap farther away from the fuel pin. Fuel pin position eccentricities are also mitigated by using a large sodium bond between the pin and capsule fuel. The advantages and potential pitfalls of this revised design are discussed, including the effect of length scales on fuel system behavior.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Accelerated Discovery and Design of Ultralow Lattice Thermal Conductivity Materials Using Chemical Bonding Principles

Semiconductors with very low lattice thermal conductivities are highly desired for applications relevant to thermal energy conversion and management, such as thermoelectrics and thermal barrier coatings. Although the crystal structure and chemical bonding are known to play vital roles in shaping heat transfer behavior, material design approaches of lowering lattice thermal conductivity using chemical bonding principles are uncommon. In this work, an effective strategy of weakening interatomic interactions and therefore suppressing lattice thermal conductivity based on chemical bonding principles is presented and a high-efficiency approach of discovering low κ L materials by screening the local coordination environments of crystalline compounds is developed. The resulting first-principles calculations uncover 30 hitherto unexplored compounds with (ultra)low lattice thermal conductivities from 13 prototype crystal structures contained in the Inorganic Crystal Structure Database. Furthermore, an approach of rationally designing high-performance thermoelectrics is demonstrated by additionally incorporating cations with stereochemically active lone-pair electrons. Here these results not only provide atomic-level insights into the physical origin of the low lattice thermal conductivity in a large family of copper/silver-based compounds but also offer an efficient approach to discover and design materials with targeted thermal transport properties.

36 MATERIALS SCIENCE↗

A Bridge for Accelerating Materials by Design

Recent technical advances in the area of nanoscale imaging, spectroscopy, and scattering/diffraction have led to unprecedented capabilities for investigating materials structural, dynamical and functional characteristics. In addition, recent advances in computational algorithms and computer capacities that are orders of magnitude larger/faster have enabled large-scale simulations of materials properties starting with nothing but the identity of the atomic species and the basic principles of quantum- and statistical-mechanics and thermodynamics. Along with these advances, an explosion of high-resolution data has emerged. This confluence of capabilities and rise of big data offer grand opportunities for advancing materials sciences but also introduce several challenges. In this editorial we identify challenges impeding progress towards advancing materials by design (e.g., the design/discovery of materials with improved properties/performance), possible solutions, and provide examples of scientific issues that can be addressed by using a tightly integrated approach where theory and experiments are linked through big-deep data.

97 MATHEMATICS AND COMPUTING↗

Silicon buried gratings for dielectric laser electron accelerators

This paper describes design and simulations of dielectric laser electron accelerators that achieve Gigavolt-per-meter (GV/m) accelerating gradients and wide electron channels (>1 μm). The accelerator design is based on a silicon buried grating structure that enables flexible phase synchronization, large electron channel fields, and low standing-wave ratio in the material. This design increases the accelerating gradients to more than double those of reported quartz grating accelerators, thereby reducing the input laser fluence by 60% for the same accelerating gradient. With a 100 fs pulsed laser, our silicon buried gratings can achieve a maximum gradient of 1.1 GV/m, indicating that these accelerators have potential for numerous electron-accelerator applications.

36 MATERIALS SCIENCE↗

Accurate and Accelerated Neuromorphic Network Design Leveraging A Bayesian Hyperparameter Pareto Optimization Approach

Neuromorphic systems allow for extremely efficient hardware implementations for neural networks (NNs). In recent years, several algorithms have been presented to train spiking NNs (SNNs) for neuromorphic hardware. However, SNNs often provide lower accuracy than their artificial NNs (ANNs) counterparts or require computationally expensive and slow training/inference methods. To close this gap, designers typically rely on reconfiguring SNNs through adjustments in the neuron/synapse model or training algorithm itself. Nevertheless, these steps incur significant design time, while still lacking the desired improvement in terms of training/inference times (latency). Designing SNNs that can mimic the accuracy of ANNs with reasonable training times is an exigent challenge in neuromorphic computing. In this work, we present an alternative approach that looks at such designs as an optimization problem rather than algorithm or architecture redesign. We develop a versatile multiobjective hyperparameter optimization (HPO) for automatically tuning HPs of two state-of-the-art SNN training algorithms, SLAYER and HYBRID. We emphasize that, to the best of our knowledge, this is the first work trying to improve SNNs’ computational efficiency, accuracy, and training time using an efficient HPO. We demonstrate significant performance improvements for SNNs on several datasets without the need to redesign or invent new training algorithms/architectures. Our approach results in more accurate networks with lower latency and, in turn, higher energy efficiency than previous implementations. In particular, we demonstrate improvement in accuracy and more than 5× reduction in the training/inference time for the SLAYER algorithm on the DVS Gesture dataset. In the case of HYBRID, we demonstrate 30% reduction in timesteps while surpassing the accuracy of the state-of-the-art networks on CIFAR10. Further, our analysis suggests that even a seemingly minor change in HPs could change the accuracy by 5 - 6×.

Parsa, Maryam↗

Evaluation of Data Lake Design for the Accelerator Control System

In this modern world, the user expects faster processing and real-time response to operate accelerator control devices. The existing framework with its infrastructure does not have the ability to satisfy these future needs. Therefore, modernization is required to reach industry standards and develop a modular framework that can provide flexibility and dynamic scalability. The data lake architecture, comprising three layers, ingestion, processing, and data consumption, provides flexibility and scalability to meet current and future demands for the accelerator control system.

Jaikar, Amol [Fermilab]↗