Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “memory array”

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 145 records · Page 8
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Combining different 3-D global and regional seismic wave propagation solvers towards box tomography in the deep Earth

SUMMARY In previous publications, we presented a general framework, which we called ‘box tomography’, that allows the coupling of any two different numerical seismic wave propagation solvers, respectively outside and inside a target region, or ‘box’. The goal of such hybrid wavefield computations is to reduce the cost of computations in the context of full-waveform inversion for structure within the target region, when sources and/or receivers are located at large distances from the box. Previously, we had demonstrated this approach with sources and receivers outside the target region in a 2-D acoustic spherical earth model, and demonstrated and applied this methodology in the 3-D spherical elastic Earth in a continental scale inversion in which all stations were inside the target region. Here we extend the implementation of the approach to the case of a 3-D global elastic earth model in the case where both sources and stations are outside the box. We couple a global 3-D solver, SPECFEM3D_GLOBE, for the computation of the wavefield and Green’s functions in a reference 3-D model, with a regional 3-D solver, RegSEM, for the computation of the wavefield within the box, by means of time-reversal mirrors. We briefly review key theoretical aspects, showing in particular how only the displacement is needed to be stored at the boundary of the box. We provide details of the practical implementation, including the geometrical design of the mirrors, how we deal with different sizes of meshes in the two solvers, and how we address memory-saving through the use of B-spline compression of the recorded wavefield on the mirror. The proposed approach is numerically efficient but also versatile, since adapting it to other solvers is straightforward and does not require any changes in the solver codes themselves, as long as the displacement can be recovered at any point in time and space. We present benchmarks of the hybrid computations against direct computations of the wavefield between a source and an array of stations in a realistic geometry centred in the Yellowstone region, with and without a hypothetical plume within the ‘box’, and with a 1-D or a 3-D background model, down to a period of 20 s. The ultimate goal of this development is for applications in the context of imaging of remote target regions in the deep mantle, such as, for example, Ultra Low Velocity Zones.

Geochemistry & Geophysics↗
DOE OSTI2024

Ferroelectric capacitors and field-effect transistors as in-memory computing elements for machine learning workloads

Abstract This study discusses the feasibility of Ferroelectric Capacitors (FeCaps) and Ferroelectric Field-Effect Transistors (FeFETs) as In-Memory Computing (IMC) elements to accelerate machine learning (ML) workloads. We conducted an exploration of device fabrication and proposed system-algorithm co-design to boost performance. A novel FeCap device, incorporating an interfacial layer (IL) and$$\text {Hf}_{0.5}\text {Zr}_{0.5}\text {O}_2$$ Hf 0.5 Zr 0.5 O 2 (HZO), ensures a reduction in operating voltage and enhances HZO scaling while being compatible with CMOS circuits. The IL also enriches ferroelectricity and retention properties. When integrated into crossbar arrays, FeCaps and FeFETs demonstrate their effectiveness as IMC components, eliminating sneak paths and enabling selector-less operation, leading to notable improvements in energy efficiency and area utilization. However, it is worth noting that limited capacitance ratios in FeCaps introduced errors in multiply-and-accumulate (MAC) computations. The proposed co-design approach helps in mitigating these errors and achieves high accuracy in classifying the CIFAR-10 dataset, elevating it from a baseline of 10% to 81.7%. FeFETs in crossbars, with a higher on-off ratio, outperform FeCaps, and our proposed charge-based sensing scheme achieved at least an order of magnitude reduction in power consumption, compared to prevalent current-based methods.

Science & Technology - Other Topics↗
DOE OSTI2020

TuckerMPI: A Parallel C++/MPI Software Package for Large-scale Data Compression via the Tucker Tensor Decomposition

With this study, our goal is compression of massive-scale grid-structured data, such as the multi-terabyte output of a high-fidelity computational simulation. For such data sets, we have developed a new software package called TuckerMPI, a parallel C++/MPI software package for compressing distributed data. The approach is based on treating the data as a tensor, i.e., a multidimensional array, and computing its truncated Tucker decomposition, a higher-order analogue to the truncated singular value decomposition of a matrix. The result is a low-rank approximation of the original tensor-structured data. Compression efficiency is achieved by detecting latent global structure within the data, which we contrast to most compression methods that are focused on local structure. In this work, we describe TuckerMPI, our implementation of the truncated Tucker decomposition, including details of the data distribution and in-memory layouts, the parallel and serial implementations of the key kernels, and analysis of the storage, communication, and computational costs. We test the software on 4.5 and 6.7 terabyte data sets distributed across 100 s of nodes (1,000 s of MPI processes), achieving compression ratios between 100 and 200,000×, which equates to 99--99.999% compression (depending on the desired accuracy) in substantially less time than it would take to even read the same dataset from a parallel file system. Moreover, we show that our method also allows for reconstruction of partial or down-sampled data on a single node, without a parallel computer so long as the reconstructed portion is small enough to fit on a single machine, e.g., in the instance of reconstructing/visualizing a single down-sampled time step or computing summary statistics. The code is available at https://gitlab.com/tensors/TuckerMPI.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Optimizing Magnetic Force Microscopy Resolution and Sensitivity to Visualize Nanoscale Magnetic Domains

Magnetic force microscopy (MFM) enables mapping local magnetic fields across a sample surface with nanoscale resolution. To perform MFM, an atomic force microscopy (AFM) probe whose tip has been magnetized vertically (i.e., perpendicular to the probe cantilever) is oscillated at a fixed height above the sample surface. The resultant shifts in the oscillation phase or frequency, which are proportional to the magnitude and sign of the vertical magnetic force gradient at each pixel location, are then tracked and mapped. Although the spatial resolution and sensitivity of the technique increases with decreasing lift height above the surface, this seemingly straightforward path to improved MFM images is complicated by considerations such as minimizing topographical artifacts due to shorter range van der Waals forces, increasing the oscillation amplitude to further improve sensitivity, and the presence of surface contaminants (in particular water due to humidity under ambient conditions). In addition, due to the orientation of the probe's magnetic dipole moment, MFM is intrinsically more sensitive to samples with an out-of-plane magnetization vector. Here, high-resolution topographical and magnetic phase images of single and bicomponent nanomagnet artificial spin-ice (ASI) arrays obtained in an inert (argon) atmosphere glovebox with <0.1 ppm O 2 and H 2 O are reported. Further, optimization of lift height and drive amplitude for high resolution and sensitivity while simultaneously avoiding the introduction of topographical artifacts is discussed, and detection of the stray magnetic fields emanating from either end of the nanoscale bar magnets (~250 nm long and <100 nm wide) aligned in the plane of the ASI sample surface is shown. Likewise, using the example of a Ni-Mn-Ga magnetic shape memory alloy (MSMA), MFM is demonstrated in an inert atmosphere with magnetic phase sensitivity capable of resolving a series of adjacent magnetic domains each ~200 nm wide.

47 OTHER INSTRUMENTATION↗
DOE OSTI2021

Superconducting neural networks with disordered Josephson junction array synaptic networks and leaky integrate-and-fire loop neurons

Fully coupled randomly disordered recurrent superconducting networks with additional open-ended channels for inputs and outputs are considered the basis to introduce a new architecture to neuromorphic computing in this work. Various building blocks of such a network are designed around disordered array synaptic networks using superconducting devices and circuits as an example, while emphasizing that a similar architectural approach may be compatible with several other materials and devices. A multiply coupled (interconnected) disordered array of superconducting loops containing Josephson junctions [equivalent to superconducting quantum interference devices (SQUIDs)] forms the aforementioned collective synaptic network that forms a fully recurrent network together with compatible neuron-like elements and feedback loops, enabling unsupervised learning. This approach aims to take advantage of superior power efficiency, propagation speed, and synchronizability of a small world or a random network over an ordered/regular network. Additionally, it offers a significant factor of increase in scalability. Here, a compatible leaky integrate-and-fire neuron made of superconducting loops with Josephson junctions is presented, along with circuit components for feedback loops as needed to complete the recurrent network. Several of these individual disordered array neural networks can further be coupled together in a similarly disordered way to form a hierarchical architecture of recurrent neural networks that is often suggested as similar to a biological brain.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

Imaging the magnetic nanowire cross section and magnetic ordering within a suspended 3D artificial spin-ice

Artificial spin-ice systems are patterned arrays of magnetic nanoislands arranged into frustrated geometries and provide insight into the physics of ordering and emergence. The majority of these systems have been realized in two-dimensions, mainly due to the ease of fabrication, but with recent developments in advanced nanolithography, three-dimensional artificial spin ice (ASI) structures have become possible, providing a new paradigm in their study. Such artificially engineered 3D systems provide new opportunities in realizing tunable ground states, new domain wall topologies, monopole propagation, and advanced device concepts, such as magnetic racetrack memory. Direct imaging of 3DASI structures with magnetic force microscopy has thus far been key to probing the physics of these systems but is limited in both the depth of measurement and resolution, ultimately restricting measurement to the uppermost layers of the system. In this work, a method is developed to fabricate 3DASI lattices over an aperture using two-photon lithography, thermal evaporation, and oxygen plasma exposure, allowing the probe of element-specific structural and magnetic information using soft x-ray microscopy with x-ray magnetic circular dichroism (XMCD) as magnetic contrast. The suspended polymer–permalloy lattices are found to be stable under repeated soft x-ray exposure. Analysis of the x-ray absorption signal allows the complex cross section of the magnetic nanowires to be reconstructed and demonstrates a crescent-shaped geometry. Measurement of the XMCD images after the application of an in-plane field suggests a decrease in magnetic moment on the lattice surface due to oxidation, while a measurable signal is retained on sub-lattices below the surface.

36 MATERIALS SCIENCE↗
DOE OSTI2021

A participant-derived xenograft model of HIV enables long-term evaluation of autologous immunotherapies

HIV-specific CD8+ T cells partially control viral replication and delay disease progression, but they rarely provide lasting protection, largely due to immune escape. Here, we show that engrafting mice with memory CD4+ T cells from HIV+ donors uniquely allows for the in vivo evaluation of autologous T cell responses while avoiding graft-versus-host disease and the need for human fetal tissues that limit other models. Treating HIV-infected mice with clinically relevant HIV-specific T cell products resulted in substantial reductions in viremia. In vivo activity was significantly enhanced when T cells were engineered with surface-conjugated nanogels carrying an IL-15 superagonist, but it was ultimately limited by the pervasive selection of a diverse array of escape mutations, recapitulating patterns seen in humans. By applying mathematical modeling, we show that the kinetics of the CD8+ T cell response have a profound impact on the emergence and persistence of escape mutations. This “participant-derived xenograft” model of HIV provides a powerful tool for studying HIV-specific immunological responses and facilitating the development of effective cell-based therapies.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2024

Fail-Safe Logic Design Strategies Within Modern FPGA Architectures

Fail-safe computing refers to computing systems that revert to a non-operational safe state when a fault occurs. In this paper, we investigate a circuit level technique as mitigation for single event upsets (SEUs) and fault injection attacks on field programmable gate arrays (FPGAs), and analyze the effectiveness of the technique as a fail-safe monitor for an encryption algorithm. The propagation of fault effects through FPGA primitives including lookup tables (LUTs) and programmable interconnect points (PIPs) is assessed within an FPGA architecture created using an open source tool, and validated using fault injection experiments on an FPGA. The analysis reveals additional vulnerabilities exist within reconfigurable architectures over those in equivalent fail-safe application specific integrated circuit (ASIC), thus requiring a more elaborate network of redundant circuits and checking logic. The configuration memory bits (CMBs), which configure routing and designate logic functions within the LUTs of the FPGA, add complexity to fail-safe design strategies by introducing additional fault conditions and fault propagation paths. A resource-efficient fail-safe circuit design technique called DEsign for Fail-safe in reCONfigurable systems (DEFCON) is proposed. The benefits and limitations associated with DEFCON are described in the context of fault injection experiments carried out as simulations and in FPGA hardware.

Bhakta, Priya A. [Univ. of New Mexico, Albuquerque↗
DOE OSTI2020

High performance context switching for virtualized FPGA accelerators

A hardware context manager in a field-programmable gate array (FPGA) device includes configuration logic configured to program one or more programming regions in the FPGA device based on configuration data for implementing a target configuration of the one or more programming regions. Context management logic in the hardware context manager is coupled with the configuration logic and saves a first context corresponding to the target configuration by retrieving first state information from the set of one or more programming regions, where the first state information is generated based on the target configuration, and storing the retrieved first state information in a context memory. The context management logic restores the first context by transferring the first state information from the context memory to the one or more programming regions, and causing the configuration logic to program the one or more programming regions based on the configuration data.

42 ENGINEERING↗
DOE OSTI2024

A nonlinear journey from structural phase transitions to quantum annealing

Motivated by an exact mapping between equilibrium properties of a one-dimensional chain of quantum Ising spins in a transverse field (the transverse field Ising (TFI) model) and a two-dimensional classical array of particles in double-well potentials (the “$\phi$ 4 model”) with weak inter-chain coupling, we explore connections between the driven variants of the two systems. Here, we argue that coupling between the fundamental topological solitary waves in the form of kinks between neighboring chains in the classical $\phi$ 4 system is the analog of the competing effect of the transverse field on spin flips in the quantum TFI model. As an example application, we mimic simplified measurement protocols in a closed quantum model system by studying the classical $\phi$ 4 model subjected to periodic perturbations. This reveals memory/loss of memory and coherence/decoherence regimes, whose quantum analogs are essential in annealing phenomena. In particular, we examine regimes where the topological excitations control the thermal equilibration following perturbations. This paves the way for further explorations of the analogy between lower-dimensional linear quantum and higher-dimensional classical nonlinear systems.

74 ATOMIC AND MOLECULAR PHYSICS↗
DOE OSTI2026

Gaia: segmented germanium detector for high-energy X-ray fluorescence and spectroscopic imaging

We present Gaia, a monolithic array of 96 high-purity germanium pixel detectors integrated with a custom low-noise application-specific integrated circuit (ASIC) and a field-programmable gate array (FPGA)-based data acquisition system. The sensor operates at ∼100 K using a commercial closed-cycle cryocooler, with the in-vacuum electronics thermally isolated from the cold finger to ensure thermal stability. The system demonstrates an average energy resolution of 711 eV at 122 keV, measured using a 57 Co source, and 253 eV at 5.89 keV, measured with 55 Fe across all channels. The readout architecture incorporates a high-performance FPGA paired with a dual-core ARM processor, forming a complete embedded Linux-based computing platform. Communication between the processor and FPGA is handled via memory-mapped I/O, and data are streamed over high-speed gigabit Ethernet. A full-scale 384-pixel Gaia detector, based on this 96-element module, is currently under fabrication.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Erratum to: Soft- and reactive landing of ions onto surfaces: concepts and applications

Soft- and reactive landing of mass-selected ions is gaining attention as a promising approach for the precisely-controlled preparation of materials on surfaces that are not amenable to deposition using conventional methods. A broad range of ionization sources and mass filters are available that make ion soft-landing a versatile tool for surface modification using beams of hyperthermal (<100?eV) ions. The ability to select the mass-to-charge ratio of the ion, its kinetic energy and charge state, along with precise control of the size, shape, and position of the ion beam on the deposition target distinguishes ion soft landing from other surface modification techniques. Soft- and reactive landing have been used to prepare interfaces for practical applications as well as precisely-defined model surfaces for fundamental investigations in chemistry, physics, and materials science. For instance, soft- and reactive landing have been applied to study the surface chemistry of ions isolated in the gas-phase, prepare arrays of proteins for high-throughput biological screening, produce novel carbon-based and polymer materials, enrich the secondary structure of peptides and the chirality of organic molecules, immobilize electrochemically-active proteins and organometallics on electrodes, create thin films of complex molecules, and immobilize catalytically active organometallics as well as ligated metal clusters. In addition, soft landing has enabled investigation of the size-dependent behavior of bare metal clusters in the critical subnanometer size regime where chemical and physical properties do not scale predictably with size. The morphology, aggregation, and immobilization of larger bare metal nanoparticles, which are directly relevant to the design of catalysts as well as improved memory and electronic devices, have also been studied using ion soft landing. This review article begins in section 1 with a brief introduction to the existing applications of ion soft- and reactive landing. Section 2 provides an overview of the ionization sources and mass filters that have been used to date for soft landing of mass-selected ions. A discussion of the competing processes that occur during ion deposition as well as the types of ions and surfaces that have been investigated follows in section 3. Section 4 discusses the physical phenomena that occur during and after ion soft landing, including retention and reduction of ionic charge along with factors that impact the efficiency of ion deposition. The influence of soft landing on the secondary structure and biological activity of complex ions is addressed in section 5. Lastly, an overview of the structure and mobility as well as the catalytic, optical, magnetic, and redox properties of bare ionic clusters and nanoparticles deposited onto surfaces is presented in section 6. © 2015 Wiley Periodicals, Inc. Mass Spec Rev 35:439–479, 2016.

Johnson, Grant E.↗
DOE OSTI2020

Quantum electronics and optics at the interface of solid neon and superfluid helium

Here, we predict a new quantum electronic structure at the interface between two condensed phases of noble-gas elements: solid neon and superfluid helium. An excess electron injected onto this interface self-confines its wavefunction into a nanometric dome structure. Its size varies with pressure and optical transitions cover a broad mid-infrared spectrum. A collection of such electrons can form a classical Wigner crystal resembling a quantum-dot array on a triangular lattice. Under ultrafast laser illumination, this Wigner crystal can exhibit the quantum optical phenomenon of superradiance in the picosecond time scale. The ultralong spin-coherence time and micron-scale deterministic configurability allow the electrons in this system to serve as quantum information carriers. Their spin states can be controlled and readout by on-chip single-electron devices.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2020

Self‐Assembled NiO Nanocrystal Arrays as Memristive Elements

Abstract Memristive switching, a nonlinear current–voltage ( I – V ) characteristic, has seen a tremendous surge in interest as an approach to achieve implementation of synaptic functions. The memristive switching behavior of self‐assembled NiO nanocrystals is investigated via scanning probe microscopy, based on first‐order reversal curve current–voltage spectroscopy. Synaptic switching is clearly observed as a direct consequence of filament growth (i.e., gradually increased conductance) in the nanocrystals. A spatial dependency of the conduction in the nanocrystals suggests that there is a localization of the switching filament. The current understanding of this localization ignores features related to local lateral variation current, which can generate an excessive local heat and temperature such as electrical dissipation. The observation of low electrical dissipation at the edge of the nanocrystals shows that less energy is wasted as heat such that the bias applied can be utilized more efficiently to assist the nucleation of the filament and thus reduces the power consumption. Electrical power dissipation is also found to scale with nanocrystal height and has spatial dependence within the nanocrystals. The combination of synaptic switching and high density of the nanocrystals demonstrate that it is feasible to exploit them to create a basic architecture for neuromorphic memory devices.

Kurnia, Fran↗
DOE OSTI2020

A massively parallel and scalable multi-CPU material point method

Harnessing the power of modern multi-GPU architectures, we present a massively parallel simulation system based on the Material Point Method (MPM) for simulating physical behaviors of materials undergoing complex topological changes, self-collision, and large deformations. Our system makes three critical contributions. First, we introduce a new particle data structure that promotes coalesced memory access patterns on the GPU and eliminates the need for complex atomic operations on the memory hierarchy when writing particle data to the grid. Second, we propose a kernel fusion approach using a new Grid-to-Particles-to-Grid (G2P2G) scheme, which efficiently reduces GPU kernel launches, improves latency, and significantly reduces the amount of global memory needed to store particle data. Finally, we introduce optimized algorithmic designs that allow for efficient sparse grids in a shared memory context, enabling us to best utilize modern multi-GPU computational platforms for hybrid Lagrangian-Eulerian computational patterns. We demonstrate the effectiveness of our method with extensive benchmarks, evaluations, and dynamic simulations with elastoplasticity, granular media, and fluid dynamics. In comparisons against an open-source and heavily optimized CPU-based MPM codebase [Fang et al. 2019] on an elastic sphere colliding scene with particle counts ranging from 5 to 40 million, our GPU MPM achieves over 100x per-time-step speedup on a workstation with an Intel 8086K CPU and a single Quadro P6000 GPU, exposing exciting possibilities for future MPM simulations in computer graphics and computational science. Moreover, compared to the state-of-the-art GPU MPM method [Hu et al. 2019a], we not only achieve 2x acceleration on a single GPU but our kernel fusion strategy and Array-of-Structs-of-Array (AoSoA) data structure design also generalizes to multi-GPU systems. Our multi-GPU MPM exhibits near-perfect weak and strong scaling with 4 GPUs, enabling performant and large-scale simulations on a 10243 grid with close to 100 million particles with less than 4 minutes per frame on a single 4-GPU workstation and 134 million particles with less than 1 minute per frame on an 8-GPU workstation.

Wang, Xinlei↗
DOE OSTI2023

Computing the Properties of Matter with Leadership Computing Resources (Closeout Report for DE-SC0018121)

In order to add more capabilities to Halide, we have designed a new framework called Tiramisu and integrated this framework into Halide. Since Tiramisu enables Halide to target heterogeneous architectures, our development efforts have been refocused on Tiramisu. Most high-performance computer systems today are complex and increasingly heterogeneous; they may have CPUs, GPUs and FPGAs. Achieving best performance requires taking full advantage of all these different architectures. To address this issue, we have designed Tiramisu, an optimization framework that enables Halide (and other DSLs) to target heterogeneous architectures. Tiramisu is an optimization framework that takes as input a high level, architecture-independent representation of code and a set of scheduling and data mapping commands that guide code transformation. The input can either be generated by a domain-specific language (DSL) compiler such as Halide or directly written by a programmer. Tiramisu then applies the user-specified code and data-layout transformations and generates an architecture-specific, low-level intermediate representation (IR) that takes advantage of modern architectural features such as multicore parallelism, non-uniform memory (NUMA) hierarchies, clusters, and accelerators like GPUs and FPGAs. We integrated Tiramisu within Halide and implemented a representative set of benchmarks to evaluate this integration. Tiramisu is now open source and is available for public use (http://tiramisu-compiler.org/). A paper about Tiramisu was published, it shows that Tiramisu extends Halide with many new capabilities and that Tiramisu can generate efficient code for multicores, GPUs, FPGAs and distributed heterogeneous systems. The performance of code generated by the Tiramisu backends matches or exceeds hand optimized reference implementations. For example, the multicore backend matches the highly optimized Intel MKL library on many kernels and shows speedups reaching 4x over the original Halide. In addition to making Tiramisu more robust, we have used Tiramisu to implement a set of representative tensor operation for constructing baryon building blocks required for multi baryon contractions in LQCD. In order to implement this code, we needed to generalize Tiramisu in two ways: first we needed to support indirect array accesses, and second, we needed to add support for complex numbers to Tiramisu. The code generated by Tiramisu is 6x faster than the reference code. Our efforts towards an MPI based multi-node version of tiramisu have matured and the resulting code scales well on multiple nodes (tests up to 512 KNL nodes have been undertaken).

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2021

Nanosecond Gated CMOS Camera (NSGCC) ICD (Rev. 2.1)

The Ultra-Fast X-ray Imager (UXI) program is an ongoing effort at Sandia National Laboratories to create high speed, multi-frame, time-gated Read Out Integrated Circuits (ROICs), and a corresponding suite of photodetectors to image a wide variety of High Energy Density (HED) physics experiments on both Sandia’s Z-Machine and LLNL’s National Ignition Facility (NIF). Several cameras have been designed over the length of the program; one of the most recent is the Icarus, which is an improvement on past imagers (Furi and Hippogriff). A second sensor that can be connected is the Daedalus sensor. The Icarus is a 1024 × 512-pixel array with either 25 μm or 8 µm spatial resolution containing four frames of storage per pixel and has improved timing generation and distribution components while achieved 2 ns time gating. The Daedalus sensor is also a 1024 x 512-pixel array with 25 µm special resolution containing three frames of storage per pixel and has an increased set of features for a wider variety of applications from interlacing of rows in each frame to configurability of all shutters. See Section 14 for details regarding the Icarus implementation of the firmware and Section 15 for details regarding the Daedalus implementation. Due to the unique test environments UXI sensors are targeted for, full custom hardware was required to physically mount an Icarus or Daedalus sensor, manage its various functions, and read out pixel data for transfer to a host computer. Beyond experimental functionality, the hardware also needed to accommodate sensor characterization requirements. Lawrence Livermore National Laboratory’s ‘Version 4.0 Board’ was the result of these efforts. It mounts all the components required to fully utilize the Icarus and Daedalus sensors including analog to digital converters to convert pixel data and various system voltages to digital form for readout and analysis, DAC channels for remote configuration of critical bias voltages, static random-access memories to buffer pixel data, RS422 and Gigabit Ethernet communications for remote access, and an FPGA to tie these components together. This document describes the FPGA electrical interfaces in detail to allow the reader a greater understanding of the device, and to facilitate implementation of custom software to control and manage it. The Version 4.0 Board is a continuation of the Nano-second Gated CMOS hardware design that retains much of the functionality of the Version 1.0 Board while adding features including a DAC instead of digital potentiometers, as well as sensors for pressure and radiation. The Version 4.0 board is intended for applications requiring tight form-factor enclosures. It is composed of two stacking boards; one holds the FPGA and regulators to power the various components of the board while the other contains the mating connector to the sensor, image-readoff ADCs, the DAC, and other components.

42 ENGINEERING↗
DOE OSTI2024

Bridging the Gap Between LLMs and LNS with Dynamic Data Format and Architecture Codesign

Deep neural networks (DNNs) have achieved tremendous success in the past few years. However, their training and inference demand exceptional computational and memory resources. Quantization has been shown as an effective approach to mitigate the cost, with the mainstream data types reduced from FP32 to FP16/BF16 and recently FP8 in the latest NVIDIA H100 GPUs. With increasingly aggressive quantization, however, the conventional floating-point formats suffer from limited precision in representing numbers around zero. Recently, NVIDIA demonstrated the potential of using a Logarithmic Number System (LNS) for the next generation of tensor cores. While LNS mitigates the hurdles in representing small numbers, in this work we observed a mismatch between LNS and the emerging Large Language Models (LLM), where LLM exhibits significant outliers when directly adopting the LNS format. In this paper, we present a data-format/architecture codesign to bright this gap. On the format side, we propose a dynamic LNS format to flexibly represent outliers at a higher precision, by exploiting asymmetry in the LNS representation and identifying outliers through a per-vector basis. On the architecture side, for demonstration, we realize the dynamic LNS format in a systolic array, which can handle the irregularity of the outliers at runtime. We implement our approach on an Alveo U280 FPGA as a prototype. Experimental results show that our design can effectively handle the outliers and resolve the mismatch between LNS and LLM, contributing to an accuracy improvement of 15.4% and 16% over the floating-point and the original LNS baselines, using four state-of-the-art LLM models. Our observation and design lay a solid foundation for the large-scale adoption of the LNS format in the next-generation deep learning hardware.

Haghi, Pouya↗