Search

Invasive cortical brain-machine interfaces (BMIs) can significantly improve the life quality of motor-impaired patients. Nonetheless, externally mounted pedestals pose an infection risk, which calls for fully implanted systems. Such systems, however, must meet strict latency and energy constraints while providing reliable decoding performance. While recurrent spiking neural networks (RSNNs) are ideally suited for ultra-low-power, low-latency processing on neuromorphic hardware, it is unclear whether they meet the above requirements. To address this question, we trained RSNNs to decode finger velocity from cortical spike trains (CSTs) of two macaque monkeys. First, we found that a large RSNN model outperformed existing feedforward spiking neural networks (SNNs) and artificial neural networks (ANNs) in terms of their decoding accuracy. We next developed a tiny RSNN with a smaller memory footprint, low firing rates, and sparse connectivity. Despite its reduced computational requirements, the resulting model performed substantially better than existing SNN and ANN decoders. Our results thus demonstrate that RSNNs offer competitive CST decoding performance under tight resource constraints and are promising candidates for fully implanted ultra-low-power BMIs with the potential to revolutionize patient care., Comment: 5 pages, 2 figures. This work has been submitted to the IEEE BioCAS 2024 conference

Training spiking neural networks to approximate universal functions is essential for studying information processing in the brain and for neuromorphic computing. Yet the binary nature of spikes poses a challenge for direct gradient-based training. Surrogate gradients have been empirically successful in circumventing this problem, but their theoretical foundation remains elusive. Here, we investigate the relation of surrogate gradients to two theoretically well-founded approaches. On the one hand, we consider smoothed probabilistic models, which, due to the lack of support for automatic differentiation, are impractical for training multi-layer spiking neural networks but provide derivatives equivalent to surrogate gradients for single neurons. On the other hand, we investigate stochastic automatic differentiation, which is compatible with discrete randomness but has not yet been used to train spiking neural networks. We find that the latter gives surrogate gradients a theoretical basis in stochastic spiking neural networks, where the surrogate derivative matches the derivative of the neuronal escape noise function. This finding supports the effectiveness of surrogate gradients in practice and suggests their suitability for stochastic spiking neural networks. However, surrogate gradients are generally not gradients of a surrogate loss despite their relation to stochastic automatic differentiation. Nevertheless, we empirically confirm the effectiveness of surrogate gradients in stochastic multi-layer spiking neural networks and discuss their relation to deterministic networks as a special case. Our work gives theoretical support to surrogate gradients and the choice of a suitable surrogate derivative in stochastic spiking neural networks., Comment: 26 pages, 7 figures + 3 supplementary figures

Spiking neural networks (SNNs) underlie low-power, fault-tolerant information processing in the brain and could constitute a power-efficient alternative to conventional deep neural networks when implemented on suitable neuromorphic hardware accelerators. However, instantiating SNNs that solve complex computational tasks in-silico remains a significant challenge. Surrogate gradient (SG) techniques have emerged as a standard solution for training SNNs end-to-end. Still, their success depends on synaptic weight initialization, similar to conventional artificial neural networks (ANNs). Yet, unlike in the case of ANNs, it remains elusive what constitutes a good initial state for an SNN. Here, we develop a general initialization strategy for SNNs inspired by the fluctuation-driven regime commonly observed in the brain. Specifically, we derive practical solutions for data-dependent weight initialization that ensure fluctuation-driven firing in the widely used leaky integrate-and-fire (LIF) neurons. We empirically show that SNNs initialized following our strategy exhibit superior learning performance when trained with SGs. These findings generalize across several datasets and SNN architectures, including fully connected, deep convolutional, recurrent, and more biologically plausible SNNs obeying Dale's law. Thus fluctuation-driven initialization provides a practical, versatile, and easy-to-implement strategy for improving SNN training performance on diverse tasks in neuromorphic engineering and computational neuroscience., Comment: 30 pages, 7 figures, plus supplementary material

The Wide-Field Infrared Survey Telescope (WFIRST) will answer fundamental questions about the evolution of dark energy over time and expand the catalog of known exoplanets into new regions of parameter space. Using a Hubble-sized mirror and 18 newly developed HgCdTe 4K x 4K photodiode arrays (H4RG-10), WFIRST will measure the positions and shapes of hundreds of millions of galaxies, the light curves of thousands of supernovae, and the microlensing signals of over a thousand exoplanets toward the bulge of the Galaxy. These measurements require unprecedented sensitivity and characterization of the Wide Field Instrument (WFI), particularly its detectors. The WFIRST project undertook an extensive detector development program to create focal plane arrays that meet these science requirements. These prototype detectors have been characterized and their performance demonstrated in a relevant space-like environment (thermal vacuum, vibration, acoustic, and radiation testing), advancing the H4RG-10's technology readiness level (TRL) to TRL-6. We present the performance characteristics of these TRL-6 demonstration devices., Comment: 47 pages, 19 figures manuscript submitted to JATIS

Muon spin spectroscopy has been used to study in detail the onset of large-moment antiferromagnetism (LMAF) in UPt_{3} as induced by Th substitution. Zero-field experiments have been carried out on a series of polycrystalline U_{1-x}Th_{x}Pt_{3} (0 <= x <= 0.05) samples in the temperature range 0.04 - 10 K. At low Th content (x <= 0.002) magnetic ordering on the time scale of the uSR experiment (10^{-8} s) is not detected. For x = 0.005 a weak magnetic signal appears below T = 2 K, while for 0.006 <= x <= 0.05, spontaneous oscillations in the uSR spectra signal the presence of the LMAF phase. The data are well described by a two-component depolarization function, combining the contribution of a polycrystalline antiferromagnet and a Kubo-Lorentzian response. However, the transition into the antiferromagnetic phase is quite broad. For x = 0.01 and 0.02, a weak magnetic signal appears below about 7 K, which is well above the mean-field transition temperatures. The broadening may be a result of the effects of disorder on the time fluctuations associated with anomalous small-moment antiferromagnetism., Comment: 29 pages, 9 figures

Transverse-field muon spin relaxation measurements have been carried out on the heavy-fermion superconductor UPt3 doped with small amounts of Pd. We find that the critical Pd concentration for the emergence of the large-moment antiferromagnetic phase is ~0.6 at.%Pd. At the same Pd content, superconductivity is completely suppressed. The existence of a magnetic quantum critical point in the phase diagram, which coincides with the critical point for superconductivity, provides evidence for ferromagnetic spin-fluctuation mediated odd-parity superconductivity, which competes with antiferromagnetic order., Comment: 4 pages (includes 3 figures); postscript file

The magnetic response of the heavy fermion superconductor UPt_3 has been investigated on a microscopic scale by muon Knight shift studies. Two distinct and isotropic Knight shifts have been found for the field in the basal plane. While the volume fractions associated with the two Knight shifts are approximately equal at low and high temperatures, they show a dramatic and opposite temperature dependence around T_N. Our results are independent on the precise muon localization site. We conclude that UPt_3 is characterized by a two component magnetic response., Comment: 5 pages, 4 figures

The angular and temperature (10-250 K) variation of the Knight shift of single-crystalline U(Pt0.95Pd0.05)3 has been measured in transverse field (B=0.6 T) mSR experiments. By analysing the temperature variation of the Knight shift with a modified Curie-Weiss expression the muon localization site in this hexagonal material is determined at (0,0,0)., Comment: 12 pages (including 4 figures); postscript file; Proc. 8th Int. Conf. on Muon Spin Rotation, Relaxation and Resonance (Aug.30-Sept.3, Les Diablerets); 2nd version with minor corrections

We report mSR experiments carried out on a series of heavy-electron pseudobinary compounds U(Pt1-xPdx)3 (x<=0.05). For x<=0.005 the zero-field muon depolarisation is described by the Kubo-Toyabe function. However the temperature variation of the Kubo-Toyabe relaxation rate does not show any sign of the small-moment antiferromagnetic phase with TN~6 K (signalled by neutron diffraction), in contrast to previous reports. The failure to detect the small ordered moment suggests it has a fluctuating (> 10 MHz) nature, which is consistent with the interpretation of NMR data. For 0.01<=x<=0.05 the muon depolarisation in the ordered state is described by two terms of equal amplitude: an exponentially damped spontaneous oscillation and a Lorentzian Kubo-Toyabe function. These terms are associated with antiferromagnetic order with substantial moments. The Knight-shift measured in a magnetic field of 0.6 T on single-crystalline U(Pt0.95Pd0.05)3 in the paramagnetic state shows two signals for B perpendicular to c, while only one signal is observed for B||c. The observation of two signals for B perpendicular to c, while there is only one muon localisation site (0,0,0), points to the presence of two spatially distinct regions of different magnetic response., Comment: 25 pages including 12 figures (PS), J. Phys.: Condens. Matter, in print

Wide Field Infrared Survey Telescope (WFIRST) is the next generation astrophysics flagship mission. The primary mission goal is to probe the expansion history of the Universe and the growth of cosmic structure. The wide field instrument has an operating temperature of 90K. We measured the thermal conductance of several candidate materials and components for thermal subsystem of the instrument at cryogenic temperatures. This paper describes these measurements and presents the results.

NASA's Wide Field Infrared Survey Telescope (WFIRST) is being designed to deliver unprecedented capability in dark energy and exoplanet science, and to host a technology demonstration coronagraph for exoplanet imaging and spectroscopy. The observatory design has matured since 2013; we present a comprehensive description of the observatory configuration as refined during the WFIRST Phase-A study. The observatory is based on an existing, repurposed 2.4 meter space telescope coupled with a 288 megapixel near-infrared (0.6 to 2 microns) HgCdTe focal plane array with multiple imaging and spectrographic modes. Together they deliver a 0.28 square degree field of view, which is approximately 100 times larger than the Hubble Space Telescope, and a sensitivity that enables rapid science surveys. In addition, the coronagraph technology demonstration will prove the feasibility of new techniques for exoplanet discovery, imaging, and spectral analysis. A composite truss structure meters both instruments to the telescope assembly, and the instruments and the spacecraft are flight serviceable. We present configuration changes since 2013 that improved interfaces, improved testability, and reduced technical risk. We provide an overview of our Integrated Modeling results, performed at an unprecedented level for a phase-A study, to illustrate performance margins with respect to static wavefront error, jitter, and thermal drift.

X-ray microcalorimeters using transition-edge sensors (TES) show great promise for use in astronomical x-ray spectroscopy. We have obtained very high energy resolution (2.8 electronvolts at 1.5 kiloelectronvolts and 3.7 electronvolts at 3.3 kiloelectronvolts) in a large, isolated TES pixel using a Mo/Au proximity-effect bilayer on a silicon nitride membrane. We will discuss the performance and our characterization of that device. In order to be truly suitable for use behind an x-ray telescope, however, such devices need to be arrayed with a pixel size and focal-plane coverage commensurate with the telescope focal length and spatial resolution. Since this requires fitting the TES and its thermal link, a critical component of each calorimeter pixel, into a far more compact geometry than has previously been investigated, we must study the fundamental scaling laws in pixel optimization. We have designed a photolithography mask that will allow us to probe the range in thermal conductance that can be obtained by perforating the nitride membrane in a narrow perimeter around the sensor. This mask will also show the effects of reducing the TES area. Though we have not yet tested devices of the compact designs, we will present our progress in several of the key processing steps and discuss the parameter space of our intended investigations.

The XRS instrument on Astro-E is a fully self-contained microcalorimeter x-ray instrument capable of acquiring optimally filtering, and characterizing events for 32 independent pixels. With the launch of the Astro-E spacecraft, a full flight spare detector system has been integrated into a laboratory cryostat for use on the electron beam ion trap (EBIT) at Lawrence Livermore National Laboratory. The detector system contains a microcalorimeter array with 32 instrumented pixels heat sunk to 60 mK using an adiabatic demagnetization refrio,erator. The instrument has a composite resolution of 8eV at 1 keV and 12eV at 6 keV with a minimum of 95% quantum efficiency. This will allow high spectral resolution, broadband observations of collisionally excited plasmas which are produced in the EBIT experiment. Unique to our instrument are exceptionally well characterized 1000 Angstrom thick aluminum on polyimide infrared blocking filters. The detailed transmission function including the edc,e fine structure of these filters has been measured in our laboratory using an erect field grating spectrometer. This will allow the instrument to perform the first broadband absolute flux measurements with the EBIT instrument. The instrument performance as well as the results of preliminary measurements will be discussed. Work performed under the auspices of the U.S. D.o.E. by Lawrence Livermore National Laboratory under contract W-7405-ENG-48 and was supported by the NASA High Energy Astrophysics Supporting Research and Technology Program.

The XRS instrument has an array of 32 micro-calorimeters at the focal plane. These calorimeters consist of ion-implanted silicon thermistors and HgTe thermalizing x-ray absorbers. These devices have demonstrated a resolution of 9 eV at 3 keV and 11 eV at 6 keV. We will discuss the basic physical parameters of this array, including the array layout, thermal conductance of the link to the heat sink, operating temperature, thermistor size, absorber choice, and means of attaching the absorber to the thermistor bearing element. We will present representative performance data, though a more detailed presentation of the results of the instrument calibration is presented elsewhere in these proceedings. A silicon ionization detector is located behind the calorimeter array and serves to reject events due to cosmic rays. We will briefly describe this anti-coincidence detector and its performance in conjunction with the array.

Zero- and longitudinal-field muon spin relaxation measurements on a heavy fermion system CeCu2.1 Si2 have revealed an onset of static magnetic ordering below T(M) approximately 0.8 K, which coexists with superconductivity below T(c) = 0.7 K. The line shapes of the observed muon spin depolarization functions suggest an ordering in either spin glass or incommensurate spin-density-wave state, with a small averaged static moment of the order of 0.1 micro-B per formula unit at T approaches 0.

Zero- and longitudinal-field muon spin relaxation measurements on a polycrystal sample of a heavy fermion superconductor CeCu2.1 Si2 (T(c) = 0.7 K) have revealed an onset of static magnetic ordering below T approximately 0.8 K. The line shapes of the observed spectra in zero field indicate a wide distribution of static random local fields at muon sites, suggesting that the ordering is either spin glass or incommensurate spin-density-wave state. The observed width of the random local field at T = 0.05 K corresponds to a small averaged static moment of the order of 0.1 micro-B per formula unit.

This simple cryocooler provides a small reservoir of liquid helium at a stable temperature of 4.2K. It uses a novel adaptation of the Simon expansion cryocooler to provide continuous cooling. Operation is in a four stage cycle: (1) A closed vessel of helium under high pressure is cooled to 12K using a conventional Gifford-McMahon closed-cycle cryocooler. (2) The pressure is released adiabatically providing cooling to 4.2K. (3) Liquid helium is collected in a second, well insulated, vessel. (4) The first vessel is repressurized. The cycle time is 15-30 minutes. In this manner, a pool of liquid helium is continuously maintained in the second vessel, with a temperature stability of 0.03 degrees. The continuous cooling power available is 3mW. This design provides simplicity and reliability through the absence of any orifices or moving parts at cryogenic temperatures except for the conventional Gifford-McMahon cryocooler.