SEARCH · Search NASA
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.
This paper describes the operation of FPGAs at very low temperatures eg -160(deg)C. Both Actel and Xilinx parts are tested It was found that low temperature operations is not a problem with the parts tested, but there is a problem with powering on an FPGA at cold temperatures.
A document discusses the replication of the functionality of the onboard space-shuttle general-purpose computers (GPCs) in field-programmable gate arrays (FPGAs) and application-specific integrated circuits (ASICs). The purpose of the replication effort is to enable utilization of proven space-shuttle flight software and software-development facilities to the extent possible during development of software for flight computers for a new generation of launch vehicles derived from the space shuttles. The replication involves specifying the instruction set of the central processing unit and the input/output processor (IOP) of the space-shuttle GPC in a hardware description language (HDL). The HDL is synthesized to form a "core" processor in an FPGA or, less preferably, in an ASIC. The core processor can be used to create a flight-control card to be inserted into a new avionics computer. The IOP of the GPC as implemented in the core processor could be designed to support data-bus protocols other than that of a multiplexer interface adapter (MIA) used in the space shuttle. Hence, a computer containing the core processor could be tailored to communicate via the space-shuttle GPC bus and/or one or more other buses.
Static random access memory (SRAM) upset rates in field programmable gate arrays (FPGAs) from the Xilinx Virtex 2 family have been tested for radiation effects on configuration memory, block RAM and the power-on-reset (POR) and SelectMAP single event functional interrupts (SEFIs). Dynamic testing has shown the effectiveness and value of Triple Module Redundancy (TMR) and partial reconfiguration when used in conjunction. Continuing dynamic testing for more complex designs and other Virtex 2 capabilities (i.e., I/O standards, digital clock managers (DCM), etc.) is scheduled.
A device denoted a board saver has been developed as a means of reducing wear and tear of a printed-circuit board onto which an antifuse field programmable gate array (FPGA) is to be eventually soldered permanently after a number of design iterations. The need for the board saver or a similar device arises because (1) antifuse-FPGA design iterations are common and (2) repeated soldering and unsoldering of FPGAs on the printed-circuit board to accommodate design iterations can wear out the printed-circuit board. The board saver is basically a solderable/unsolderable FPGA receptacle that is installed temporarily on the printed-circuit board. The board saver is, more specifically, a smaller, square-ring-shaped, printed-circuit board (see figure) that contains half via holes one for each contact pad along its periphery. As initially fabricated, the board saver is a wider ring containing full via holes, but then it is milled along its outer edges, cutting the via holes in half and laterally exposing their interiors. The board saver is positioned in registration with the designated FPGA footprint and each via hole is soldered to the outer portion of the corresponding FPGA contact pad on the first-mentioned printed-circuit board. The via-hole/contact joints can be inspected visually and can be easily unsoldered later. The square hole in the middle of the board saver is sized to accommodate the FPGA, and the thickness of the board saver is the same as that of the FPGA. Hence, when a non-final FPGA is placed in the square hole, the combination of the non-final FPGA and the board saver occupy no more area and thickness than would a final FPGA soldered directly into its designated position on the first-mentioned circuit board. The contact leads of a non-final FPGA are not bent and are soldered, at the top of the board saver, to the corresponding via holes. A non-final FPGA can readily be unsoldered from the board saver and replaced by another one. Once the final FPGA design has been determined, the board saver can be unsoldered from the contact pads on the first-mentioned printed-circuit board and replaced by the final FPGA.
In this sliide presentation the statistical impact of PMOS Negative Bias Temperature Instability (NBTI) degradation on circuit delay is reviewed. An on-board timing circuitry that was been developed to detect and measure timing delays in SRAM-based LUT commercial FPGAs is described. The initial experimental data on the on-board timing circuit are reviewed.
The invention allows a field-programmable gate array (FPGA) or similar device to be efficiently reconfigured in whole or in part to provide higher capacity, non-redundant operation. The redundant device consists of functional units such as adders or multipliers, configuration memory for the functional units, a programmable routing method, configuration memory for the routing method, and various other features such as block RAM, I/O (random access memory, input/output) capability, dedicated carry logic, etc. The redundant device has three identical sets of functional units and routing resources and majority voters that correct errors. The configuration memory may or may not be redundant, depending on need. For example, SRAM-based FPGAs will need some type of radiation-tolerant configuration memory, or they will need triple-redundant configuration memory. Flash or anti-fuse devices will generally not need redundant configuration memory. Some means of loading and verifying the configuration memory is also required. These are all components of the pre-existing redundant FPGA. This innovation modifies the voter to accept a MODE input, which specifies whether ordinary voting is to occur, or if redundancy is to be split. Generally, additional routing resources will also be required to pass data between sections of the device created by splitting the redundancy. In redundancy mode, the voters produce an output corresponding to the two inputs that agree, in the usual fashion. In the split mode, the voters select just one input and convey this to the output, ignoring the other inputs. In a dual-redundant system (as opposed to triple-redundant), instead of a voter, there is some means to latch or gate a state update only when both inputs agree. In this case, the invention would require modification of the latch or gate so that it would operate normally in redundant mode, and would separately latch or gate the inputs in non-redundant mode.
High-performance digital electronic circuits for onboard processing of return signals in an airborne precipitation- measuring radar system have been implemented in commercially available field-programmable gate arrays (FPGAs). Previously, it was standard practice to downlink the radar-return data to a ground station for postprocessing a costly practice that prevents the nearly-real-time use of the data for automated targeting. In principle, the onboard processing could be performed by a system of about 20 personal- computer-type microprocessors; relative to such a system, the present FPGA-based processor is much smaller and consumes much less power. Alternatively, the onboard processing could be performed by an application-specific integrated circuit (ASIC), but in comparison with an ASIC implementation, the present FPGA implementation offers the advantages of (1) greater flexibility for research applications like the present one and (2) lower cost in the small production volumes typical of research applications. The generation and processing of signals in the airborne precipitation measuring radar system in question involves the following especially notable steps: The system utilizes a total of four channels two carrier frequencies and two polarizations at each frequency. The system uses pulse compression: that is, the transmitted pulse is spread out in time and the received echo of the pulse is processed with a matched filter to despread it. The return signal is band-limited and digitally demodulated to a complex baseband signal that, for each pulse, comprises a large number of samples. Each complex pair of samples (denoted a range gate in radar terminology) is associated with a numerical index that corresponds to a specific time offset from the beginning of the radar pulse, so that each such pair represents the energy reflected from a specific range. This energy and the average echo power are computed. The phase of each range bin is compared to the previous echo by complex conjugate multiplication to obtain the mean Doppler shift (and hence the mean and variance of the velocity of precipitation) of the echo at that range.
We present our work on implementing Radiation Hardening by Software (RHBSW) techniques on the Xilinx Virtex5 FPGAs PowerPC 440 processors on the SpaceCube 2.0 platform. The techniques have been matured and tested through simulation modeling, fault emulation, laser fault injection and now in a flight experiment, as part of the Space Test Program- Houston 4-ISS SpaceCube Experiment 2.0 (STP-H4-ISE 2.0). This work leverages concepts such as heartbeat monitoring, control flow assertions, and checkpointing, commonly used in the High Performance Computing industry, and adapts them for use in remote sensing embedded systems. These techniques are extremely low overhead (typically <1.3%), enabling a 3.3x gain in processing performance as compared to the equivalent traditionally radiation hardened processor. The recently concluded STP-H4 flight experiment was an opportunity to upgrade the RHBSW techniques for the Virtex5 FPGA and demonstrate them on-board the ISS to achieve TRL 7. This work details the implementation of the RHBSW techniques, that were previously developed for the Virtex4-based SpaceCube 1.0 platform, on the Virtex5-based SpaceCube 2.0 flight platform. The evaluation spans the development and integration with flight software, remotely uploading the new experiment to the ISS SpaceCube 2.0 platform, and conducting the experiment continuously for 16 days before the platform was decommissioned. The experiment was conducted on two PowerPCs embedded within the Virtex5 FPGA devices and the experiment collected 19,400 checkpoints, processed 253,482 status messages, and incurred 0 faults. These results are highly encouraging and future work is looking into longer duration testing as part of the STP-H5 flight experiment.
Irradiations and subsequent failure analyses were performed to investigate single event dielectric rupture (SEDR) in Actel FPGAs as a function of ion LET, angle, bias, temperature, feature size, and device type. The small cross sections imply acceptably low risk for most spacecraft uses.
Although Transformer-based language representations achieve state-of-the-art accuracy on various natural language processing (NLP) tasks, the large model size has been challenging the resource constrained computing platforms. Weight pruning, as a popular and effective technique in reducing the number of weight parameters and accelerating the Transformer, has been investigated on GPUs. However, the Transformer acceleration using weight pruning on field-programmable gate array (FPGAs) remains unexplored. This paper investigates the column balanced block-wise pruning on Transformer and designs an FPGA acceleration engine to customize the balanced blockwise matrix multiplication. We implement the Transformer model with proper hardware scheduling, and the experiments show that the Transformer inference on FPGA achieves 10.35 ms latency with the batch size of 32, which is 10.96 × speed up comparing to CPU platform and 2.08 × speed up comparing to GPU platform.
In this paper, we show how to adapt and deploy anomaly detection algorithms based on deep autoencoders, for the unsupervised detection of new physics signatures in the extremely challenging environment of a real-time event selection system at the Large Hadron Collider (LHC). We demonstrate that new physics signatures can be enhanced by three orders of magnitude, while staying within the strict latency and resource constraints of a typical LHC event filtering system. This would allow for collecting datasets potentially enriched with high-purity contributions from new physics processes. Through per-layer, highly parallel implementations of network layers, support for autoencoder-specific losses on FPGAs and latent space based inference, we demonstrate that anomaly detection can be performed in as little as $80\,$ns using less than 3% of the logic resources in the Xilinx Virtex VU9P FPGA. Opening the way to real-life applications of this idea during the next data-taking campaign of the LHC.
We present the first sub-microsecond transformer implementation on an FPGA achieving competitive performance for state-of-the-art high-energy physics benchmarks. Transformers have shown exceptional performance on multiple tasks in modern machine learning applications, including jet tagging at the CERN Large Hadron Collider (LHC). However, their computational complexity prohibits use in real-time applications, such as the hardware trigger system of the collider experiments up until now. In this work, we demonstrate the first application of transformers for jet tagging on FPGAs, achieving $\mathcal{O}(100)$ nanosecond latency with superior performance compared to alternative baseline models. We leverage high-granularity quantization and distributed arithmetic optimization to fit the entire transformer model on a single FPGA, achieving the required throughput and latency. Furthermore, we add multi-head attention and linear attention support to hls4ml, making our work accessible to the broader fast machine learning community. This work advances the next-generation trigger systems for the High Luminosity LHC, enabling the use of transformers for real-time applications in high-energy physics and beyond.
Explore the source record for details and available documents.
Active feedback control in magnetic confinement fusion devices is desirable to mitigate plasma instabilities and enable robust operation. Optical high-speed cameras provide a powerful, non-invasive diagnostic and can be suitable for these applications. Here, in this study, we process high-speed camera data, at rates exceeding 100 kfps, on in situ field-programmable gate array (FPGA) hardware to track magnetohydrodynamic (MHD) mode evolution and generate control signals in real time. Our system utilizes a convolutional neural network (CNN) model, which predicts the n = 1 MHD mode amplitude and phase using camera images with better accuracy than other tested non-deep-learning-based methods. By implementing this model directly within the standard FPGA readout hardware of the high-speed camera diagnostic, our mode tracking system achieves a total trigger-to-output latency of 17.6 μs and a throughput of up to 120 kfps. This study at the High Beta Tokamak-Extended Pulse (HBT-EP) experiment demonstrates an FPGA-based high-speed camera data acquisition and processing system, enabling application in real-time machine-learning-based tokamak diagnostic and control as well as potential applications in other scientific domains.
We present the implementation of binary and ternary neural networks in the hls4ml library, designed to automatically convert deep neural network models to digital circuits with FPGA firmware. Starting from benchmark models trained with floating point precision, we investigate different strategies to reduce the network's resource consumption by reducing the numerical precision of the network parameters to binary or ternary. We discuss the trade-off between model accuracy and resource consumption. In addition, we show how to balance between latency and accuracy by retaining full precision on a selected subset of network components. As an example, we consider two multiclass classification tasks: handwritten digit recognition with the MNIST data set and jet identification with simulated proton-proton collisions at the CERN Large Hadron Collider. The binary and ternary implementation has similar performance to the higher precision implementation while using drastically fewer FPGA resources.
In this paper, we investigate how field programmable gate arrays can serve as hardware accelerators for real-time semantic segmentation tasks relevant for autonomous driving. Considering compressed versions of the ENet convolutional neural network architecture, we demonstrate a fully-on-chip deployment with a latency of 4.9 ms per image, using less than 30% of the available resources on a Xilinx ZCU102 evaluation board. The latency is reduced to 3 ms per image when increasing the batch size to ten, corresponding to the use case where the autonomous vehicle receives inputs from multiple cameras simultaneously. We show, through aggressive filter reduction and heterogeneous quantization-aware training, and an optimized implementation of convolutional layers, that the power consumption and resource utilization can be significantly reduced while maintaining accuracy on the Cityscapes dataset.
Currently, there is a lack of availability of low cost, low power neuromorphic hardware. In this work, we introduce the μCaspian architecture along with an associated development PCB design which provides a low cost and SWaP (size, weight, and power) optimized neuromorphic hardware platform. Further, our proposed system only uses commercial off the shelf components and an open source FPGA workflow to maximize the accessibility of μCaspian to all researchers.
Here, we develop an end-to-end workflow for the training and implementation of co-designed neural networks (NNs) for efficient field-programmable gate array (FPGA) hardware. Our approach leverages Hessian-aware quantization of NNs, the Quantized Open Neural Network Exchange intermediate representation, and the hls4ml tool flow for transpiling NNs into FPGA firmware. This makes efficient NN implementations in hardware accessible to nonexperts in a single open sourced workflow that can be deployed for real-time machine-learning applications in a wide range of scientific and industrial settings. We demonstrate the workflow in a particle physics application involving trigger decisions that must operate at the 40-MHz collision rate of the CERN Large Hadron Collider (LHC). Given the high collision rate, all data processing must be implemented on FPGA hardware within the strict area and latency requirements. Based on these constraints, we implement an optimized mixed-precision NN classifier for high-momentum particle jets in simulated LHC proton-proton collisions.