Neuromorphic Sensor Based on Force-Sensing Resistors.

This work introduces a neuromorphic sensor (NS) based on force-sensing resistors (FSR) and spiking neurons for robotic systems. The proposed sensor integrates the FSR in the schematic of the spiking neuron in order to make the sensor generate spikes with a frequency that depends on the applied force. The performance of the proposed sensor is evaluated in the control of a SMA-actuated robotic finger by monitoring the force during a steady state when the finger pushes on a tweezer. For comparison purposes, we performed a similar evaluation when the SNN received input from a widely used compression load cell (CLC). The results show that the proposed FSR-based neuromorphic sensor has very good sensitivity to low forces and the function between the spiking rate and the applied force is continuous, with good variation range. However, when compared to the CLC, the response of the NS follows a logarithmic-like function with improved sensitivity for small forces. In addition, the power consumption of NS is 128 µW that is 270 times lower than that of the CLC which needs 3.5 mW to operate. These characteristics make the neuromorphic sensor with FSR suitable for bioinspired control of humanoid robotics, representing a low-power and low-cost alternative to the widely used sensors.

Coordinated reset stimulation of plastic neural networks with spatially dependent synaptic connections.

Background: Abnormal neuronal synchrony is associated with several neurological disorders, including Parkinson's disease (PD), essential tremor, dystonia, and epilepsy. Coordinated reset (CR) stimulation was developed computationally to counteract abnormal neuronal synchrony. During CR stimulation, phase-shifted stimuli are delivered to multiple stimulation sites. Computational studies in plastic neural networks reported that CR stimulation drove the networks into an attractor of a stable desynchronized state by down-regulating synaptic connections, which led to long-lasting desynchronization effects that outlasted stimulation. Later, corresponding long-lasting desynchronization and therapeutic effects were found in animal models of PD and PD patients. To date, it is unclear how spatially dependent synaptic connections, as typically observed in the brain, shape CR-induced synaptic downregulation and long-lasting effects. Methods: We performed numerical simulations of networks of leaky integrate-and-fire neurons with spike-timing-dependent plasticity and spatially dependent synaptic connections to study and further improve acute and long-term responses to CR stimulation. Results: The characteristic length scale of synaptic connections relative to the distance between stimulation sites plays a key role in CR parameter adjustment. In networks with short synaptic length scales, a substantial synaptic downregulation can be achieved by selecting appropriate stimulus-related parameters, such as the stimulus amplitude and shape, regardless of the employed spatiotemporal pattern of stimulus deliveries. Complex stimulus shapes can induce local connectivity patterns in the vicinity of the stimulation sites. In contrast, in networks with longer synaptic length scales, the spatiotemporal sequence of stimulus deliveries is of major importance for synaptic downregulation. In particular, rapid shuffling of the stimulus sequence is advantageous for synaptic downregulation. Conclusion: Our results suggest that CR stimulation parameters can be adjusted to synaptic connectivity to further improve the long-lasting effects. Furthermore, shuffling of CR sequences is advantageous for long-lasting desynchronization effects. Our work provides important hypotheses on CR parameter selection for future preclinical and clinical studies.

A mean-field model of gamma-frequency oscillations in networks of excitatory and inhibitory neurons.

Gamma oscillations are widely seen in the cerebral cortex in different states of the wake-sleep cycle and are thought to play a role in sensory processing and cognition. Here, we study the emergence of gamma oscillations at two levels, in networks of spiking neurons, and a mean-field model. At the network level, we consider two different mechanisms to generate gamma oscillations and show that they are best seen if one takes into account the synaptic delay between neurons. At the mean-field level, we show that, by introducing delays, the mean-field can also produce gamma oscillations. The mean-field matches the mean activity of excitatory and inhibitory populations of the spiking network, as well as their oscillation frequencies, for both mechanisms. This mean-field model of gamma oscillations should be a useful tool to investigate large-scale interactions through gamma oscillations in the brain.

Spatial memory requires hypocretins to elevate medial entorhinal gamma oscillations.

The hypocretin (Hcrt) (also known as orexin) neuropeptidic wakefulness-promoting system is implicated in the regulation of spatial memory, but its specific role and mechanisms remain poorly understood. In this study, we revealed the innervation of the medial entorhinal cortex (MEC) by Hcrt neurons in mice. Using the genetically encoded G-protein-coupled receptor activation-based Hcrt sensor, we observed a significant increase in Hcrt levels in the MEC during novel object-place exploration. We identified the function of Hcrt at presynaptic glutamatergic terminals, where it recruits fast-spiking parvalbumin-positive neurons and promotes gamma oscillations. Bidirectional manipulations of Hcrt neurons' projections from the lateral hypothalamus (LHHcrt) to MEC revealed the essential role of this pathway in regulating object-place memory encoding, but not recall, through the modulation of gamma oscillations. Our findings highlight the significance of the LHHcrt-MEC circuitry in supporting spatial memory and reveal a unique neural basis for the hypothalamic regulation of spatial memory.

Combining aperiodic 1/f slopes and brain simulation: An EEG/MEG proxy marker of excitation/inhibition imbalance in Alzheimer's disease.

INTRODUCTION: Accumulation and interaction of amyloid-beta (Aß) and tau proteins during progression of Alzheimer's disease (AD) are shown to tilt neuronal circuits away from balanced excitation/inhibition (E/I). Current available techniques for noninvasive interrogation of E/I in the intact human brain, for example, magnetic resonance spectroscopy (MRS), are highly restrictive (i.e., limited spatial extent), have low temporal and spatial resolution and suffer from the limited ability to distinguish accurately between different neurotransmitters complicating its interpretation. As such, these methods alone offer an incomplete explanation of E/I. Recently, the aperiodic component of neural power spectrum, often referred to in the literature as the '1/f slope', has been described as a promising and scalable biomarker that can track disruptions in E/I potentially underlying a spectrum of clinical conditions, such as autism, schizophrenia, or epilepsy, as well as developmental E/I changes as seen in aging. METHODS: Using 1/f slopes from resting-state spectral data and computational modeling, we developed a new method for inferring E/I alterations in AD. RESULTS: We tested our method on recent freely and publicly available electroencephalography (EEG) and magnetoencephalography (MEG) datasets of patients with AD or prodromal disease and demonstrated the method's potential for uncovering regional patterns of abnormal excitatory and inhibitory parameters. DISCUSSION: Our results provide a general framework for investigating circuit-level disorders in AD and developing therapeutic interventions that aim to restore the balance between excitation and inhibition.

Empirical mode decomposition of local field potential data from optogenetic experiments.

Introduction: This study investigated the effects of cocaine administration and parvalbumin-type interneuron stimulation on local field potentials (LFPs) recorded in vivo from the medial prefrontal cortex (mPFC) of six mice using optogenetic tools. Methods: The local network was subject to a brief 10 ms laser pulse, and the response was recorded for 2 s over 100 trials for each of the six subjects who showed stable coupling between the mPFC and the optrode. Due to the strong non-stationary and nonlinearity of the LFP, we used the adaptive, data-driven, Empirical Mode Decomposition (EMD) method to decompose the signal into orthogonal Intrinsic Mode Functions (IMFs). Results: Through trial and error, we found that seven is the optimum number of orthogonal IMFs that overlaps with known frequency bands of brain activity. We found that the Index of Orthogonality (IO) of IMF amplitudes was close to zero. The Index of Energy Conservation (IEC) for each decomposition was close to unity, as expected for orthogonal decompositions. We found that the power density distribution vs. frequency follows a power law with an average scaling exponent of ~1.4 over the entire range of IMF frequencies 2-2,000 Hz. Discussion: The scaling exponent is slightly smaller for cocaine than the control, suggesting that neural activity avalanches under cocaine have longer life spans and sizes.

Online spike-based recognition of digits with ultrafast microlaser neurons.

Classification and recognition tasks performed on photonic hardware-based neural networks often require at least one offline computational step, such as in the increasingly popular reservoir computing paradigm. Removing this offline step can significantly improve the response time and energy efficiency of such systems. We present numerical simulations of different algorithms that utilize ultrafast photonic spiking neurons as receptive fields to allow for image recognition without an offline computing step. In particular, we discuss the merits of event, spike-time and rank-order based algorithms adapted to this system. These techniques have the potential to significantly improve the efficiency and effectiveness of optical classification systems, minimizing the number of spiking nodes required for a given task and leveraging the parallelism offered by photonic hardware.

Boost event-driven tactile learning with location spiking neurons.

Tactile sensing is essential for a variety of daily tasks. Inspired by the event-driven nature and sparse spiking communication of the biological systems, recent advances in event-driven tactile sensors and Spiking Neural Networks (SNNs) spur the research in related fields. However, SNN-enabled event-driven tactile learning is still in its infancy due to the limited representation abilities of existing spiking neurons and high spatio-temporal complexity in the event-driven tactile data. In this paper, to improve the representation capability of existing spiking neurons, we propose a novel neuron model called "location spiking neuron," which enables us to extract features of event-based data in a novel way. Specifically, based on the classical Time Spike Response Model (TSRM), we develop the Location Spike Response Model (LSRM). In addition, based on the most commonly-used Time Leaky Integrate-and-Fire (TLIF) model, we develop the Location Leaky Integrate-and-Fire (LLIF) model. Moreover, to demonstrate the representation effectiveness of our proposed neurons and capture the complex spatio-temporal dependencies in the event-driven tactile data, we exploit the location spiking neurons to propose two hybrid models for event-driven tactile learning. Specifically, the first hybrid model combines a fully-connected SNN with TSRM neurons and a fully-connected SNN with LSRM neurons. And the second hybrid model fuses the spatial spiking graph neural network with TLIF neurons and the temporal spiking graph neural network with LLIF neurons. Extensive experiments demonstrate the significant improvements of our models over the state-of-the-art methods on event-driven tactile learning, including event-driven tactile object recognition and event-driven slip detection. Moreover, compared to the counterpart artificial neural networks (ANNs), our SNN models are 10× to 100× energy-efficient, which shows the superior energy efficiency of our models and may bring new opportunities to the spike-based learning community and neuromorphic engineering. Finally, we thoroughly examine the advantages and limitations of various spiking neurons and discuss the broad applicability and potential impact of this work on other spike-based learning applications.

Fast-Spiking Interneurons of the Premotor Cortex Contribute to Initiation and Execution of Spontaneous Actions.

Planning and execution of voluntary movement depend on the contribution of distinct classes of neurons in primary motor and premotor areas. However, timing and pattern of activation of GABAergic cells during specific motor behaviors remain only partly understood. Here, we directly compared the response properties of putative pyramidal neurons (PNs) and GABAergic fast-spiking neurons (FSNs) during spontaneous licking and forelimb movements in male mice. Recordings centered on the face/mouth motor field of the anterolateral motor cortex (ALM) revealed that FSNs fire longer than PNs and earlier for licking, but not for forelimb movements. Computational analysis revealed that FSNs carry vastly more information than PNs about the onset of movement. While PNs differently modulate their discharge during distinct motor acts, most FSNs respond with a stereotyped increase in firing rate. Accordingly, the informational redundancy was greater among FSNs than PNs. Finally, optogenetic silencing of a subset of FSNs reduced spontaneous licking movement. These data suggest that a global rise of inhibition contributes to the initiation and execution of spontaneous motor actions.SIGNIFICANCE STATEMENT Our study contributes to clarifying the causal role of fast-spiking neurons (FSNs) in driving initiation and execution of specific, spontaneous movements. Within the face/mouth motor field of mice premotor cortex, FSNs fire before pyramidal neurons (PNs) with a specific activation pattern: they reach their peak of activity earlier than PNs during the initiation of licking, but not of forelimb, movements; duration of FSNs activity is also greater and exhibits less selectivity for the movement type, as compared with that of PNs. Accordingly, FSNs appear to carry more redundant information than PNs. Optogenetic silencing of FSNs reduced spontaneous licking movement, suggesting that FSNs contribute to the initiation and execution of specific spontaneous movements, possibly by sculpting response selectivity of nearby PNs.

An explainable artificial intelligence approach to spatial navigation based on hippocampal circuitry.

Learning to navigate a complex environment is not a difficult task for a mammal. For example, finding the correct way to exit a maze following a sequence of cues, does not need a long training session. Just a single or a few runs through a new environment is, in most cases, sufficient to learn an exit path starting from anywhere in the maze. This ability is in striking contrast with the well-known difficulty that any deep learning algorithm has in learning a trajectory through a sequence of objects. Being able to learn an arbitrarily long sequence of objects to reach a specific place could take, in general, prohibitively long training sessions. This is a clear indication that current artificial intelligence methods are essentially unable to capture the way in which a real brain implements a cognitive function. In previous work, we have proposed a proof-of-principle model demonstrating how, using hippocampal circuitry, it is possible to learn an arbitrary sequence of known objects in a single trial. We called this model SLT (Single Learning Trial). In the current work, we extend this model, which we will call e-STL, to introduce the capability of navigating a classic four-arms maze to learn, in a single trial, the correct path to reach an exit ignoring dead ends. We show the conditions under which the e-SLT network, including cells coding for places, head-direction, and objects, can robustly and efficiently implement a fundamental cognitive function. The results shed light on the possible circuit organization and operation of the hippocampus and may represent the building block of a new generation of artificial intelligence algorithms for spatial navigation.

A general pattern of non-spiking neuron dynamics under the effect of potassium and calcium channel modifications.

Electrical activity of excitable cells results from ion exchanges through cell membranes, so that genetic or epigenetic changes in genes encoding ion channels are likely to affect neuronal electrical signaling throughout the brain. There is a large literature on the effect of variations in ion channels on the dynamics of spiking neurons that represent the main type of neurons found in the vertebrate nervous systems. Nevertheless, non-spiking neurons are also ubiquitous in many nervous tissues and play a critical role in the processing of some sensory systems. To our knowledge, however, how conductance variations affect the dynamics of non-spiking neurons has never been assessed. Based on experimental observations reported in the biological literature and on mathematical considerations, we first propose a phenotypic classification of non-spiking neurons. Then, we determine a general pattern of the phenotypic evolution of non-spiking neurons as a function of changes in calcium and potassium conductances. Furthermore, we study the homeostatic compensatory mechanisms of ion channels in a well-posed non-spiking retinal cone model. We show that there is a restricted range of ion conductance values for which the behavior and phenotype of the neuron are maintained. Finally, we discuss the implications of the phenotypic changes of individual cells at the level of neuronal network functioning of the C. elegans worm and the retina, which are two non-spiking nervous tissues composed of neurons with various phenotypes.

Voltage slope guided learning in spiking neural networks.

A thorny problem in machine learning is how to extract useful clues related to delayed feedback signals from the clutter of input activity, known as the temporal credit-assignment problem. The aggregate-label learning algorithms make an explicit representation of this problem by training spiking neurons to assign the aggregate feedback signal to potentially effective clues. However, earlier aggregate-label learning algorithms suffered from inefficiencies due to the large amount of computation, while recent algorithms that have solved this problem may fail to learn due to the inability to find adjustment points. Therefore, we propose a membrane voltage slope guided algorithm (VSG) to further cope with this limitation. Direct dependence on the membrane voltage when finding the key point of weight adjustment makes VSG avoid intensive calculation, but more importantly, the membrane voltage that always exists makes it impossible to lose the adjustment point. Experimental results show that the proposed algorithm can correlate delayed feedback signals with the effective clues embedded in background spiking activity, and also achieves excellent performance on real medical classification datasets and speech classification datasets. The superior performance makes it a meaningful reference for aggregate-label learning on spiking neural networks.

A surrogate gradient spiking baseline for speech command recognition.

Artificial neural networks (ANNs) are the basis of recent advances in artificial intelligence (AI); they typically use real valued neuron responses. By contrast, biological neurons are known to operate using spike trains. In principle, spiking neural networks (SNNs) may have a greater representational capability than ANNs, especially for time series such as speech; however their adoption has been held back by both a lack of stable training algorithms and a lack of compatible baselines. We begin with a fairly thorough review of literature around the conjunction of ANNs and SNNs. Focusing on surrogate gradient approaches, we proceed to define a simple but relevant evaluation based on recent speech command tasks. After evaluating a representative selection of architectures, we show that a combination of adaptation, recurrence and surrogate gradients can yield light spiking architectures that are not only able to compete with ANN solutions, but also retain a high degree of compatibility with them in modern deep learning frameworks. We conclude tangibly that SNNs are appropriate for future research in AI, in particular for speech processing applications, and more speculatively that they may also assist in inference about biological function.

Synaptic alterations in visual cortex reshape contrast-dependent gamma oscillations and inhibition-excitation ratio in a genetic mouse model of migraine.

BACKGROUND: Migraine affects a significant fraction of the world population, yet its etiology is not completely understood. In vitro results highlighted thalamocortical and intra-cortical glutamatergic synaptic gain-of-function associated with a monogenic form of migraine (familial-hemiplegic-migraine-type-1: FHM1). However, how these alterations reverberate on cortical activity remains unclear. As altered responsivity to visual stimuli and abnormal processing of visual sensory information are common hallmarks of migraine, herein we investigated the effects of FHM1-driven synaptic alterations in the visual cortex of awake mice. METHODS: We recorded extracellular field potentials from the primary visual cortex (V1) of head-fixed awake FHM1 knock-in (n = 12) and wild type (n = 12) mice in response to square-wave gratings with different visual contrasts. Additionally, we reproduced in silico the obtained experimental results with a novel spiking neurons network model of mouse V1, by implementing in the model both the synaptic alterations characterizing the FHM1 genetic mouse model adopted. RESULTS: FHM1 mice displayed similar amplitude but slower temporal evolution of visual evoked potentials. Visual contrast stimuli induced a lower increase of multi-unit activity in FHM1 mice, while the amount of information content about contrast level remained, however, similar to WT. Spectral analysis of the local field potentials revealed an increase in the ß/low γ range of WT mice following the abrupt reversal of contrast gratings. Such frequency range transitioned to the high γ range in FHM1 mice. Despite this change in the encoding channel, these oscillations preserved the amount of information conveyed about visual contrast. The computational model showed how these network effects may arise from a combination of changes in thalamocortical and intra-cortical synaptic transmission, with the former inducing a lower cortical activity and the latter inducing the higher frequencies É£ oscillations. CONCLUSIONS: Contrast-driven É£ modulation in V1 activity occurs at a much higher frequency in FHM1. This is likely to play a role in the altered processing of visual information. Computational studies suggest that this shift is specifically due to enhanced cortical excitatory transmission. Our network model can help to shed light on the relationship between cellular and network levels of migraine neural alterations.

A neuroscience-inspired spiking neural network for EEG-based auditory spatial attention detection.

Recent studies have shown that alpha oscillations (8-13 Hz) enable the decoding of auditory spatial attention. Inspired by sparse coding in cortical neurons, we propose a spiking neural network model for auditory spatial attention detection. The proposed model can extract the patterns of recorded EEG of leftward and rightward attention, independently, and uses them to train the network to detect auditory spatial attention. Specifically, our model is composed of three layers, two of which are Integrate and Fire spiking neurons. We formulate a new learning rule that is based on the firing rate of pre- and post-synaptic neurons in the first and second layers of spiking neurons. The third layer has 10 spiking neurons and the pattern of their firing rate is used in the test phase to decode the auditory spatial attention of a given test sample. Moreover, the effects of using low connectivity rates of the layers and specific range of learning parameters of the learning rule are investigated. The proposed model achieves an average accuracy of 90% with only 10% of EEG signals as training data. This study also provides new insights into the role of sparse coding in both cortical networks subserving cognitive tasks and brain-inspired machine learning.

Programming Molecular Systems To Emulate a Learning Spiking Neuron.

Hebbian theory seeks to explain how the neurons in the brain adapt to stimuli to enable learning. An interesting feature of Hebbian learning is that it is an unsupervised method and, as such, does not require feedback, making it suitable in contexts where systems have to learn autonomously. This paper explores how molecular systems can be designed to show such protointelligent behaviors and proposes the first chemical reaction network (CRN) that can exhibit autonomous Hebbian learning across arbitrarily many input channels. The system emulates a spiking neuron, and we demonstrate that it can learn statistical biases of incoming inputs. The basic CRN is a minimal, thermodynamically plausible set of microreversible chemical equations that can be analyzed with respect to their energy requirements. However, to explore how such chemical systems might be engineered de novo, we also propose an extended version based on enzyme-driven compartmentalized reactions. Finally, we show how a purely DNA system, built upon the paradigm of DNA strand displacement, can realize neuronal dynamics. Our analysis provides a compelling blueprint for exploring autonomous learning in biological settings, bringing us closer to realizing real synthetic biological intelligence.

Multidimensional Dynamical Systems with Noise : Population Density Techniques for Neuroscience.

The problem of how to create efficient multi-scale models of large networks of neurons is a pressing one. It requires a balance between computational efficiency and a reduction of the number of parameters involved against biological realism. Simulations of point-model neurons show very realistic features of neural dynamics but are very hard to configure and to analyse. Instead of using hundreds or thousands of point-model neurons, a population can often be modeled by a single density function in a way that accurately reproduces quantities aggregated over the population, such as population firing rate or average membrane potential. These techniques have been widely applied in neuroscience, mainly on populations comprised of one-dimensional point-model neurons, such as leaky-integrate-and-fire neurons. Here, we present very general density methods that can be applied to point-model neurons of higher dimensionality that can represent biological features not present in simpler ones, such as adaptation and bursting. The methods are geometrical in nature and lend themselves to immediate visualisation of the population state. By decoupling the neural dynamics and the stochastic processes that model inter-neuron communication, an efficient GPGPU implementation is possible that makes the study of such high-dimensional models feasible. This decoupling also allows the study of different noise models, such as Poisson, white noise, and gamma-distributed interspike intervals, which broadens the application domain considerably compared to the Fokker-Planck equations that have traditionally dominated this approach. We will present several examples based on high-dimensional neural models. We will use dynamical systems that represent point-model neurons, but inherently there is nothing to restrict the approach presented here to neuroscience. MIIND is an open-source simulator that contains an implementation of these techniques.

A Heterogeneously Integrated Spiking Neuron Array for Multimode-Fused Perception and Object Classification.

Multimode-fused sensing in the somatosensory system helps people obtain comprehensive object properties and make accurate judgments. However, building such multisensory systems with conventional metal-oxide-semiconductor technology presents serious device integration and circuit complexity challenges. Here, a multimode-fused spiking neuron (MFSN) with a compact structure to achieve human-like multisensory perception is reported. The MFSN heterogeneously integrates a pressure sensor to process pressure and a NbOx -based memristor to sense temperature. Using this MFSN, multisensory analog information can be fused into one spike train, showing excellent data compression and conversion capabilities. Moreover, both pressure and temperature information are distinguished from fused spikes by decoupling the output frequencies and amplitudes, supporting multimodal tactile perception. Then, a 3 × 3 MFSN array is fabricated, and the fused frequency patterns are fed into a spiking neural network for enhanced tactile pattern recognition. Finally, a larger MFSN array is simulated for classifying objects with different shapes, temperatures, and weights, validating the feasibility of the MFSNs for practical applications. The proof-of-concept MFSNs enable the building of multimodal sensory systems and contribute to the development of highly intelligent robotics.

Sharp Tuning of Head Direction and Angular Head Velocity Cells in the Somatosensory Cortex.

Head direction (HD) cells form a fundamental component in the brain's spatial navigation system and are intricately linked to spatial memory and cognition. Although HD cells have been shown to act as an internal neuronal compass in various cortical and subcortical regions, the neural substrate of HD cells is incompletely understood. It is reported that HD cells in the somatosensory cortex comprise regular-spiking (RS, putative excitatory) and fast-spiking (FS, putative inhibitory) neurons. Surprisingly, somatosensory FS HD cells fire in bursts and display much sharper head-directionality than RS HD cells. These FS HD cells are nonconjunctive, rarely theta rhythmic, sparsely connected and enriched in layer 5. Moreover, sharply tuned FS HD cells, in contrast with RS HD cells, maintain stable tuning in darkness; FS HD cells' coexistence with RS HD cells and angular head velocity (AHV) cells in a layer-specific fashion through the somatosensory cortex presents a previously unreported configuration of spatial representation in the neocortex. Together, these findings challenge the notion that FS interneurons are weakly tuned to sensory stimuli, and offer a local circuit organization relevant to the generation and transmission of HD signaling in the brain.

Stochastic Spiking Behavior in Neuromorphic Networks Enables True Random Number Generation.

There is currently a great deal of interest in the use of nanoscale devices to emulate the behaviors of neurons and synapses and to facilitate brain-inspired computation. Here, it is shown that percolating networks of nanoparticles exhibit stochastic spiking behavior that is strikingly similar to that observed in biological neurons. The spiking rate can be controlled by the input stimulus, similar to "rate coding" in biology, and the distributions of times between events are log-normal, providing insights into the atomic-scale spiking mechanism. The stochasticity of the spiking behavior is then used for true random number generation, and the high quality of the generated random bit-streams is demonstrated, opening up promising routes toward integration of neuromorphic computing with secure information processing.