Input Pulse Train Research Articles

Dynamics of Leaky Integrate‐and‐Fire Neurons Based on Oxyvanite Memristors for Spiking Neural Networks

Neuromorphic computing implemented with spiking neural networks (SNNs) based on volatile threshold switching is an energy‐efficient computing paradigm that may overcome future limitations of the von Neumann architecture. Herein, threshold switching in oxyvanite (V3O5) memristors and their application as a leaky integrate‐and‐fire (LIF) neuron are explored. The spiking response of individual neurons is examined as a function of circuit parameters, input pulse train, and temperature and reveals a pulse height‐dependent spike rate in which devices exhibit excitatory spiking behavior under low input voltages and protective inhibition spiking under high voltages. Resistively coupled LIF neurons are shown to exhibit additional neural functionalities (i.e., phasic, regular and adaptation, etc.) depending on the input voltage and circuit parameters. The behavior of both individual and coupled neurons is shown to be described by a physics‐based lumped element circuit model, which therefore provides a solid foundation for exploring more complex systems. Finally, the performance of a perceptron SNN employing these LIF neurons is assessed by simulating the classification of image recognition algorithm. These results advance the development of robust solid‐state neurons with low power consumption for neuromorphic computing.

Optical computation of discrete Fourier transform utilizing the temporal Talbot effect with input pulse trains of finite duration

Optical computation of discrete Fourier transform utilizing the temporal Talbot effect with input pulse trains of finite duration

Non-ideal program-time conservation in charge trap flash for deep learning

Training deep neural networks (DNNs) is computationally intensive, but arrays of non-volatile memories such as charge trap flash (CTF) can accelerate DNN operations using in-memory computing. Specifically, the resistive processing unit (RPU) architecture uses a voltage-threshold program with stochastic encoded pulse trains and analog memory features to accelerate vector-vector outer product and weight update for gradient descent algorithms. Although CTF, offering high precision, has been regarded as an excellent choice for implementing RPU, the accumulation of charge due to the applied stochastic pulse trains is ultimately of critical significance in determining the final weight update. In this paper, we report on the non-ideal program-time conservation in CTF through pulsing input measurements. We experimentally measure the effect of pulse width and pulse gap, keeping the total ON-time of the input pulse train constant, and report three non-idealities: (1) the cumulative shift reduces when total ON-time is fragmented into a larger number of shorter pulses, (2) the cumulative shift drops abruptly for pulse widths <2 s, (3) the cumulative shift depends on the gap between consecutive pulses and the shift reduction recovers for smaller gaps. We present an explanation based on a transient tunneling field enhancement due to blocking oxide trap-charge dynamics to explain these non-idealities. Identifying and modeling the responsible mechanisms and predicting their system-level effects during learning is critical. This non-ideal accumulation is expected to affect algorithms and architectures relying on devices for implementing mathematically equivalent functions for in-memory computing-based acceleration.

Amorphous indium–gallium–zinc–oxide memristor arrays for parallel true random number generators

True random number generators (TRNGs) can generate unpredictable binary bitstream by exploiting the intrinsic stochasticity in physical variables. In a threshold switching memristor, the stochastic forming/rupture of conducting pathway has been proved to be a good random source, while further improvement of high randomness and throughput is still a challenge. Here, a crossbar array of amorphous indium–gallium–zinc–oxide (a-IGZO)-based threshold switching memristors was designed for high-throughput TRNGs. The intrinsic stochasticity of Ag conductive filament in IGZO memristor and the stochastic sneak paths in the crossbar array are the two sources of randomness in our TRNGs. In our design, one input pulse train can produce multi-channel random bits, which enables a high scalability for such TRNGs. In addition, the average energy consumption of the TRNGs can be further reduced by increasing the integration scale of the memristors. Such IGZO-based TRNGs are of great significance for security applications.

Designing a single-mode anomalous dispersion silicon core fiber for temporal multiplet formation.

A highly nonlinear single-mode anomalous dispersion silicon core fiber (SCF) is suitably designed and optimized to generate a high repetition rate pulse train in the temporal domain from a single input pulse at a sufficiently shorter optimum length in comparison to silica-based standard fibers used for the same purpose. The large amount of Kerr-induced nonlinearity of a SCF is effectively utilized here such that input Gaussian pulses or pulse trains transform into a highly repetitive temporal multiplet. The effects of free-carrier generation-induced change in absorption and dispersion are included while studying the nonlinear pulse propagation through the SCF. To declare the generated pulse as a superior-graded triplet, a Q parameter, as a function of relative pulse parameters of the individual pulses of a triplet, is defined for the first time, to the best of our knowledge. Different pulse parameters are thoroughly optimized as well as the effect of external gain is examined from the perspective of requirement of shorter fiber length and development of quality triplets. Finally, the work is further extended for the formation of quadruplet pulses by the same type of SCF. It is to be mentioned here that such a methodical study for the generation of a temporal multiplet using a semiconductor core fiber has not been reported earlier.

Pulse optimization and device engineering of 3D charge-trap flash for synaptic operation

We investigate 3D charge-trap (CT) nand flash cells using device-physics based multi-scale simulations to explore their potential and optimum operating conditions as electronic synapses of the neuromorphic hardware. A set of figure of merits (FOMs) has been adopted to indicate their goodness of operation under incremental pulse inputs. The results of this study suggest that excellent synaptic FOMs can be attained from 3D CT nands by designing and calibrating the input pulse trains. The impact of variations of device dimensions on charge capture and release phenomena have been investigated and linked to output characteristics in order to obtain intuitive guidelines for attaining desired synaptic functionalities. By co-designing gate dielectric stack and input pulses, the threshold voltage (VT) of the 3D CT cell can be sequentially increased and decreased in a linear and symmetric fashion, providing a large number of distinct VT levels with good retention characteristics. Statistical simulations suggest that device-to-device variations of electrical responses have a negligible impact on the synaptic capabilities of these devices. It has also been shown that the incorporation of deeper traps through material engineering improves synaptic reliability of the 3D CT cells under prolonged operations.

Design and analysis of an all-optical NAND logic gate using a photonic crystal semiconductor optical amplifier based on the Mach–Zehnder interferometer structure

In this study, a high-speed all-optical NAND logic gate (AO-NAND-LG) was designed and numerically simulated. The simulation was performed using the photonic crystal-semiconductor optical amplifier (PC-SOA) structure based on the Mach–Zehnder interferometer and nonlinear cross-phase modulation mechanism. The input optical pulse sequence used in the design was the return-to-zero type. Moreover, two PC-SOAs were utilized in the design of the NAND-LG. The rate and propagation equations were solved using the finite difference method. The optimal mode was achieved for NAND-LG with an energy value of 2.2 fJ for input pulse trains A and B, and an injection current of 1 mA at a bit rate of 80 Gbps. Furthermore, the impacts of the pattern effect, conversion efficiency, extinction ratio, and gain recovery on the design of the AO-NAND-LG were analyzed for the first time in simulations to enhance the performance and efficiency of the PC-SOA. Furthermore, the results demonstrated that the PC-SOA exhibited better logical performance than the ordinary SOA, and it is a suitable candidate for integrated optical circuits because it is much shorter than the SOA.

Nonlinear fiber system for shot-noise limited intensity noise suppression and amplification.

We propose a nonlinear fiber system for shot-noise limited, all-optical intensity noise reduction and signal amplification. The mechanism is based on the accumulation of different nonlinear phase shifts between orthogonal polarization modes in a polarization-maintaining fiber amplifier in combination with an implemented sinusoidal transmission function. The resulting correlation between the input intensity fluctuations and the system transmission enables tunable intensity noise reduction of the input pulse train. In the experiment, the noise spectral density of a mode-locked oscillator is suppressed by up to 20 dB to the theoretical shot-noise limit of the measurement at -151.1dBc/Hz with simultaneous pulse amplification of 13.5 dB.

Nearly Self-Similar Pulse Compression of High-Repetition-Rate Pulse Trains in Tapered Silicon Waveguides

We numerically demonstrate a feasible way to generate a high-repetition-rate ultrashort soliton train from a dual frequency beat signal on a chip-scale silicon waveguide. A train of well-separated solitons is formed in the designed novel tapered silicon waveguide. Through the self-similar pulse compression, the pulse duration of the input pulse train is compressed from 1.5625 ps to 123.4 fs at the telecomm wavelength. Besides, the pulse propagation consists of two stages: pulse shaping and pulse compression, where the raised-cosine pulse train is evolved into the hyperbolic secant pulse train and further realize an efficient self-similar compression. The results deepen our understanding of nonlinear propagation in silicon waveguides, and open the possibility of high-repetition-rate pulse trains functionalities for ultrahigh-capacity optical communication and micro-machining in a monolithic CMOS platform.

A tunable repetition rate multiplier for multi-wavelength optical pulse trains

We propose a temporal Fresnel filtering based repetition rate multiplication (RRM) scheme suitable for multi-wavelength optical pulse trains. Residual chirp terms inside and outside the temporal Fresnel integral contribute to preserving of the multi-wavelength feature. Implementation of our scheme is very simple and compact, and only one linearly chirped Bragg fiber grating (LCBG) and one temporal grating (TG) are needed. Optical signals before and after the TG share the same LCBG by forward and backward passing to implement temporal original and inverse Fresnel transforms, respectively, avoiding possible mismatch between these two transforms and minimizing the influence of the third and higher order dispersion coefficients. Electrical programming of the TG contributes to the tunability of multiplication factor m. In numerical simulations, our scheme succeeds to preserve the multi-wavelength feature of an input pulse train, while the current temporal Fourier filtering and temporal Talbot effect based schemes fail, verifying the novelty of our scheme. Our work is especially useful for tuning the scanning resolution of steering angles of microwave beams, formed by photonic-assisted phased array antennas, in the application of five-generation (5G) wireless communications.

Real-time discrete Fourier transformer with complex-valued outputs based on the inverse temporal Talbot effect.

Discrete Fourier transform (DFT) plays an important role in digital signal processing. In this paper, we present a novel optical real-time discrete Fourier transformer with complex-valued outputs, which is enabled by the inverse temporal Talbot effect. In the system, an input pulse train is first quadratically phase-modulated as in an inverse temporal Talbot system and then split into two channels. In the first channel, the pulse train is further amplitude-modulated pulse-by-pulse by a discrete data sequence to be transformed. In the second channel, a reference signal modulates the pulse train, which is for removing the residual quadratic phase profile in the output pulse train. The pulse trains in the two channels propagate through a shared dispersion medium with a proper dispersion value determined by the inverse temporal Talbot effect. A 90-degree optical hybrid and two balanced photodetectors are employed to retrieve the real and imaginary parts of the DFT results. In this scheme, the pulse repetition rate of the output pulse train is equal to the input one. In addition, we present a full theoretical framework to explain exactly the DFT relationship. We also demonstrate that the input data sequence can be complex-valued with the help of an I/Q modulator.

Rotating Angle Estimation for Hybrid Stepper Motors With Application to Bearing Fault Diagnosis

Hybrid stepper motors, which convert the input pulse train into rotating angle increments by using an open-loop control strategy, are widely used in industrial applications. The bearing is a key and vulnerable component in stepper motors, and the bearing faults affect the positioning accuracy and output torque of motors. Order analysis is an effective method in bearing fault diagnosis under varying speed conditions. The rotating angle curve is necessary for order analysis, but angle or speed sensors are not installed on hybrid stepper motors. In this article, a method of estimating the rotating angle from motor current is proposed and applied to bearing fault diagnosis. First, a variational nonlinear chirp mode decomposition method is introduced to decompose the current signal and extract a dominant mode that has a strong correlation with motor rotation. Second, the rotating angle is calculated, and the synchronous vibration signal is resampled using the rotating angle curve. Finally, the bearing fault type is identified from the envelope order spectrum of the resampled signal. The proposed method can accurately diagnose bearing faults and has potential applications in condition monitoring of stepper motors.

Accelerating self-imaging effect for Airy pulse trains

We investigate the temporal accelerating self-imaging effect for a train of Airy pulses propagating in optical fibers. The group velocity of the Airy pulses is varying during propagation, resulting in a parabolic time-space trajectory of self-imaging. The acceleration is determined by the main-lobe width of the Airy pulses. Meanwhile, the self-imaging distance depends both on the main-lobe width and the time interval of the pulses. In addition, the trajectory of self-imaging can be modified by imposing a linearly varying phase on the input pulse train. By applying third-order dispersion, we also realize the temporally magnified self-imaging of the Airy pulse trains. As the Airy pulses possess infinite energy, the self-imaging can be observed for infinite times. If the pulses are truncated to have finite-energy, the self-imaging maintains only in a limited distance. The study provides a promising way to control the self-imaging of optical pulses and may find applications in optical communication and signal processing systems.

Rapid programmable pulse shaping of femtosecond pulses at the MHz repetition rate

We have shown experimentally the successful engineering of femtosecond pulse shaping at a 76 MHz repetition rate input pulse with an acousto-optic modulator (AOM). High repetition rate (HRR) femtosecond laser pulse shaping using an AOM in the Fourier plane was incomprehensible because of its intrinsic 100 kHz acoustic update limit. We demonstrate an effective way of pulse selection and a calibration routine (Fourier shift theorem), which enables generation of ∼10 MHz shaped output pulses from the HRR input pulse train. We have generated a temporally shifted rectangular shaped pulse profile by applying modulation on both the phase and amplitude of the ‘sinc’ RF modulation function.

Corrections to “Modulation Fading in Temporal Talbot Effect” [Aug 1, 2018 1376-1379

In the above paper [1] , the derived equations (8) and (9) are based on the condition that $k (k=N T^{2}f_{m})$ is less than $T$ , which are incomplete since the input pulse train is periodic. Considering the periodicity of the pulse train and assuming that the period of the input pulse train is much shorter than the inverse of modulating signal frequency, the correct expressions of (8) and (9) are \begin{equation*} e_{D} (t)\propto [1+\alpha \exp (j\psi)A_{F} (f_{m})\cos (2\pi f_{m} t)]\cdot g_{T} (t)\end{equation*} and \begin{equation*} i(t)\propto [1+2\alpha \cos (\psi)A_{F} (f_{m})\cos (2\pi f_{m} t)]\cdot g_{T}^{2}(t)\end{equation*} respectively, where $A_{F} (f_{m})$ is a periodic function as \begin{equation*} A_{F} (f_{m})=\sum \limits _{n=-\infty }^\infty {\exp \left[{-\frac {N^{2}T^{4}(f_{m} -nF)^{2}}{2\tau _{0}^{2}}}\right]}\end{equation*} with a period of $F={1} / {(NT)}$ . Therefore, the modulation transfer function should be expressed as \begin{align*} H(f_{m})=&\cos (\psi)A_{F} (f_{m}) \\=&\textrm {cos(}\pi NT^{2}f_{m}^{2})\!\!\sum \limits _{n=-\infty }^\infty {\exp \left[{-\frac {N^{2}T^{4}(f_{m} -nF)^{2}}{2\tau _{0}^{2}}}\right]}\end{align*}

Combination and Compression of Multiple Optical Pulses in Nonlinear Fibers with the Exponentially Decreasing Dispersion

We propose a simple scheme to generate high-energy ultrashort pulses by combination and compression of multiple input pulses which share the same chirp profile. First, the multiple raised-cosine pulses in the input pulse train are modulated by a phase modulator, in which each modulation cycle covers two, three, four, or five pulses. Then, the modulated pulses are launched into a nonlinear fiber with the exponentially decreasing dispersion. We find that these pulses initially coalesce into a single pulse whose pulse profile is nearly hyperbolic secant, which then undergoes self-similar compression. Thus in the proposed method, first the combination of the multiple optical pulses occurs and then self-similar compression takes over. Besides, we also report the generation of ultrashort pulses by combination and compression of multiple hyperbolic secant pulses with the same chirp. The numerical results reveal that the resulting ultrashort pulse possesses a large portion of the input pulses for both raised-cosine and hyperbolic secant pulses. However, the compression factor and energy ratio are relatively higher for the hyperbolic secant pulses when compared with the raised-cosine pulses.

Design of High Voltage and High Frequency Pulse Generator Using DC Choke

Pulsed Electric Field (PEF) processing has demonstrated a high potential for various applications ranging from extraction of valuable components from different biological objects to water treatment. Therefore, high voltage and high frequency pulse generator design is essential in order to fulfill the requirement of a PEF processing system. A high voltage and high frequency pulse generator using DC choke is proposed in this works as solution to increase the pulse train amplitude. The DC choke was verified using function generator to identify its operating frequency and voltage. Results showed that the operating frequency of the DC choke was between 24 kHz to 35 kHz and that it was able to retain the pulse shape of the input pulse train. These results show the ability of the DC choke as a potential solution to increase the pulse amplitude of a pulse train while retaining its shape and duty cycle by cascading it with other power topologies to produce high voltage and high frequency pulse generator for various applications.

LED pumped micron-scale all-silicon Raman amplifier

A micron-scale all-silicon Raman amplifier has been proposed in this paper, exploiting the giant Raman gain of silicon nanocrystal material along with the extreme optical confinement of slotted photonic crystal waveguide. Light Emitting Diode (LED) has been considered here for low-cost optical pumping and the possibility of on-chip integration. At the same time, LED pumping eradicates the temporal impairment of output pulses which is otherwise unavoidable in case of continuous wave laser pumping. An overall gain of 3.22 dB has been achieved for a 400 Gbps input pulse train with a waveguide length of the amplifier which is as small as 4 μm. Moreover, the strong electroluminescence of silicon nanocrystal opens up the possibility of integrating the pump source on the same platform and, hence, expedites the future scope of realizing micron-scale silicon Raman laser without external pump source.

High-speed pulse train amplification in semiconductor optical amplifiers with optimized bias current.

In this paper, we have experimentally investigated the optimized bias current of semiconductor optical amplifiers (SOAs) to achieve high-speed input pulse train amplification with high gain and low distortion. Variations of the amplified output pulse duration with the amplifier bias currents have been analyzed and, compared to the input pulse duration, the amplified output pulse duration is broadened. As the SOA bias current decreases from the high level (larger than the saturated bias current) to the low level, the broadened pulse duration of the amplified output pulse initially decreases slowly and then rapidly. Based on the analysis, an optimized bias current of SOA for high-speed pulse train amplification is introduced. The relation between the SOA optimized bias current and the parameters of the input pulse train (pulse duration, power, and repetition rate) are experimentally studied. It is found that the larger the input pulse duration, the lower the input pulse power or a higher repetition rate can lead to a larger SOA optimized bias current, which corresponds to a larger optimized SOA gain. The effects of assist light injection and different amplifier temperatures on the SOA optimized bias current are studied and it is found that assist light injection can effectively increase the SOA optimized bias current while SOA has a lower optimized bias current at the temperature 20°C than that at other temperatures.

A fast, stochastic, and adaptive model of auditory nerve responses to cochlear implant stimulation

Cochlear implants (CIs) rehabilitate hearing impairment through direct electrical stimulation of the auditory nerve. New stimulation strategies can be evaluated using computational models. In this study, a computationally efficient model that accurately predicts auditory nerve responses to CI pulse train input was developed. A three-dimensional volume conduction and active nerve model developed at Leiden University Medical Center was extended with stochasticity, adaptation, and accommodation. This complete model includes spatial and temporal characteristics of both the cochlea and the auditory nerve. The model was validated by comparison with experimentally measured single fiber action potential responses to pulse trains published in the literature. The effects of pulse rate and pulse amplitude on spiking patterns were investigated. The modeled neural responses to CI stimulation were very similar to the single fiber action potential measurements in animal experiments. The model's responses to pulse train stimulation with respect to spatial location were also investigated. Adaptation was stronger at the borders of the stimulated area than in the center. By combining spatial details with long-term temporal components and a broad implementation of stochasticity a comprehensive model was developed that was validated for long duration electric stimulation of a wide range of pulse rates and amplitudes. The model can be used to evaluate auditory nerve responses to cochlear implant sound coding strategies.