Search

In this work, a dynamic S-box is designed based on the chaotic neural network using the memristor model of LiNbO3 (LNO) single crystalline thin film. The LNO memristor, which shows good repeatability and the characteristic of gradual resistance change, is used as the self-feedback weight in the Transient Chaotic Neural Network (TCNN) model to form a chaotic neural network. Double pseudo-random digits are generated for dynamically adjusting the S-box elements according to the weight function. The experimental results show that the proposed design can achieve physical unclonability and has better nonlinearity and differential uniformity.

A transparent thin film transistor (TFT) based on the combination of an InGaZnO channel and a high-κ (the dielectric constant is about 42.6) TaO

In this work, we report a spike-timing-dependent plasticity (STDP)-based weight-quantized/binarized online-learning spiking neural network (SNN). The SNN uses bio-plausible integrate-and-fire (IF) neuron and conductance-based synapse as the basic building blocks and realizes online learning by STDP and winner-take-all (WTA) mechanism. Weight quantization/binarization is introduced into the online-learning SNN to reduce storage requirements and improve computing efficiency. After the training process with STDP and weight quantization on the MNIST training set, the quantized SNN with 4-bit weight achieves a recognition accuracy of 93.8% on the MNIST test set, showing little loss compared with the accuracy of the non-quantized 32-bit SNN (94.1%). The accuracy of the binarized SNN slightly decreases to 92.9%, which is cost-effective considering the reduction in the weight storage space by ~ 32 times, and the product of input and weight in the binarized SNN can be realized by computationally cheap 1-bit “AND” operation. The proposed weight quantization/binarization online-learning scheme can largely save hardware costs. The area of the quantized (8-bit and 4-bit) and binarized (1-bit) SNN-based hardware is evaluated to be 448,524, 179,263, and 162,129 μm2, respectively, which is much smaller than their non-quantized 32-bit competitor (area of ~ 5.862 × 108 μm2). The hardware resource evaluation also provides a guide to make a trade-off between computational cost and performance. Moreover, the quantized/binarized STDP training method can be further extended to train various types of SNNs. In this regard, a hybrid STDP SNN and a hybrid STDP convolutional SNN, which are trained by combining unsupervised quantized/binarized STDP and supervised backpropagation (BP) training methods, achieve high accuracy in facial expression recognition scenarios.

In recent years, the weight binarized neural network (BNN) technology has made great progress. However, neural networks with binarized inputs and binarized weights suffer from low accuracy in pattern recognition or inefficiency in hardware implementation. This work proposes a spatio-temporal binary neural network (STBNN) to solve this problem. STBNN has binary network input/output, binary neuron input/output, and binarized weights, and it integrates the computationally expensive batch normalization (BN) operation widely used in previous BNNs into the neuron threshold. STBNN can largely save computing resources and storage space while maintaining high accuracy (e.g., 98.0% on the MNIST test set). Using binary input (0 or 1) and binarized weight (±1), the product of input and weight can be realized by a 1-bit Signed AND operation instead of multiplication operation in hardware implementation, thus significantly reducing computing resources, memory requirements, and power consumption. The results show that compared with a 32-bit multi-layer perceptron (MLP)-based hardware design, the STBNN-based hardware design typically reduces these three indicators by 84.2%, 96.4%, and 96.7%, respectively. This work provides an effective method to construct hardware-friendly neural network models and a guide for designing an extremely hardware-saving neural network processor.

In this article, the electrical noise signal triggered switching of resistive random access memory (RRAM) device is investigated. As noise is also generated when powering up the light source, such a phenomenon can be easily mistaken as a light-activated event. Thus, it is necessary to conduct a 'dark test' on the light-related experimental work to make a correct judgment of the observation. Although noise is often undesirable, we show that the voltage amplitude of the noise signal which triggers the forming process is much lower than the unipolar pulse height owing to the facilitation effect of the negative voltage. As a result, we propose a novel bipolar pulse writing scheme to reduce the forming voltage and variability. Conversely, the reverse effect is demonstrated for the set process, as the role of the negative voltage changes with the state of the device. This study provides a guideline for the design and optimization of the operation signals of the RRAM devices. Agency for Science, Technology and Research (A*STAR) National Research Foundation (NRF) This work was supported in part by the National Research Foundation of Singapore Program under Grant NRF-CRP13-2014-02, in part by the Economic Development BoardIndustrial Postgraduate Program (EDB-IPP) Program under Grant RCA16/335, and in part by the RIE2020 ASTAR Advanced Manufacturing and Engineering (AME) IAF-ICP under Grant I1801E0030.

The thermal budget of the standard integrated circuits (ICs) manufacturing process is one of the key considerations in the development of resistive random access memory (RRAM). To investigate the thermal budget effect on device performance, three types of HfO x -based RRAM devices with different top electrodes are subjected to postmetalization annealing (PMA) at 400 °C for more than three hours. Electrical results show that the initial amount of oxygen vacancies in the HfO x dielectric layer is strongly dependent on the electrode configuration. A device with sandwich-like (TiN/Ti/TiN) top electrode exhibits better performance in thermal stability, forming, switching uniformity, and endurance as compared to other devices with a single layer (TiN) or bi-layer (TiN/Ti) top electrode. This is mainly due to the proper engineering of the oxidation reaction at the metal–metal oxide interface. We believe that this work provides a valuable guideline in the design and optimization of RRAM electrodes.

A transparent thin film transistor (TFT) based on the combination of an InGaZnO channel and a high-κ (the dielectric constant is about 42.6) TaOx gate dielectric layer is fabricated. The TFT shows robust anticlockwise hysteresis under DC voltage sweep and synaptic behaviors (i.e., excitatory postsynaptic current, short-term memory plasticity, short-term memory to long-term memory transition, and potentiation and depression) under voltage pulse stimulus. In addition, the TFT shows high responsivity to illumination of light with various wavelengths (ultraviolet and visible light). Synaptic behaviors in response to light pulse stimuli, which could be employed in vision-based neuromorphic applications, are demonstrated. Large conductance change (Gmax/Gmin > 10) and ultra-low non-linearity (α < 0.5) of the potentiation and depression can be inspired by either gate bias pulses or photoelectric pulses with short pulse widths and small amplitudes. Ministry of Education (MOE) Published version The research of the project was supported by the Ministry of Education, Singapore, under grant AcRF TIER 1-2020-T1-002- 008 (RG144/20).

In this Letter, an asynchronous spike-driven processor based on gated recurrent neural network algorithm for electrocardiogram (ECG) cardiac arrhythmias detection has been designed. Based on the processor, the proposed ECG detection model, containing a many-to-many gated recurrent unit layer and a fully connected layer, can achieve a high classification overall accuracy of 97.8% using the MIT-BIH arrhythmia database. The processor was fabricated in 55-nm 1P6M CMOS technology. It integrates about 1.4 million logic gates, including 40 KB of on-chip memory within a 4.0 mm 2 die area. The test chip consumes 6.28 mW at real-time operation frequency (200 MHz) with 1.2 V core supply voltage and 2.65 μ W at 1 kHz.

Transparent resistive switching random access memory (ReRAM) is of interest for the future integrated invisible circuitry. However, poor understanding of its working mechanism in transparent ReRAMs with the indium tin oxide (ITO) electrode is still a critical problem and will hinder its widespread applications. To reveal the actual working mechanism in transparent ReRAMs with the ITO electrode, we investigate the transparent ITO/SiOx/ITO memory devices (∼82% transmittance in the visible region) and compare it with ITO/SiOx/Au memory devices, which both can exhibit reproducible bipolar switching. The indium (In) filament evolution, which accounts for the bipolar switching behaviors in the ITO/SiOx/ITO (or Au) memories, is directly observed using transmission electron microscopy on samples with different memory states (electroformed, ON, and OFF). These studies uncover the microscopic mechanism behind the bipolar switching in SiOx-based ReRAM devices with the ITO electrode, providing a general guidance for the design of high-performance ReRAMs with large scalability and high endurance. Agency for Science, Technology and Research (A*STAR) National Research Foundation (NRF) This work is supported by the National Research Foundation Competitive Research Programme (grant no. NRF-CRP13- 2014-02) and the RIE2020 ASTAR AME IAF-ICP Grant (No. I1801E0030). K.Q. acknowledges the “Qilu young scholar” program (grant no. 11500089963022) of Shandong University, China.

Both nanoscale surface modification and structural control play significant roles in enhancing the electrochemical properties of battery electrodes. Herein, we design a novel binder-free anode via N-doped graphene quantum dot (N-GQD) decorated Na2Ti3O7 nanofibre arrays (Na2Ti3O7 NFAs) directly grown on flexible carbon textiles (CTs) for high-performance sodium-ion batteries (SIBs). Three dimensional (3D) hierarchical Na2Ti3O7 NFAs constructed from ultrathin Na2Ti3O7 nanosheets provide a large specific surface area and shorter diffusion paths for both ions and electrons. More importantly, the unique N-GQD soft protection produces greatly increased surface conductivity and imparts stability to the nanofibre array structure, leading to fast Na-ion diffusion kinetics. As a result, the flexible 3D hierarchical Na2Ti3O7@N-GQDs/CT electrode as a binder-free anode for a sodium half-battery delivers a high specific capacity of 158 mA h g−1 after 30 cycles and retains ∼92.5% of this capacity after 1000 cycles at a high rate of 4C (1C = 177 mA g−1). Furthermore, it can be assembled into a flexible full cell with Na3V2(PO4)3@NC/CTs as the cathode, which exhibits high levels of flexibility, excellent long-term cycling stability, and outstanding energy/power density. Our results open up a new approach for the surface modification strategy to enhance the performance of battery electrodes.

Neuromorphic hardware inspired by the brain has attracted much attention for its advanced information processing concept. However, implementing online learning in the neuromorphic chip is still challenging. In this paper, we present a bio-plausible online-learning spiking neural network (SNN) model for hardware implementation. The SNN consists of an input layer, an excitatory layer, and an inhibitory layer. To save resource cost and accelerate information processing speed during hardware implementation, online learning based on the spiking neural model is realized by trace-based spiking-timing-dependent plasticity (STDP). Neuron and synapse activities are digitalized, and decay behaviors of neuron and synapse parameters are realized by the bit-shift operation. After learning training set from the Modified National Institute of Standards and Technology (MNIST), the spiking neural model successfully recognizes the digits from the MNIST test set, showing the feasibility and capability of the model. The recognition accuracy increases significantly from 90.0% to 94.5% with the number of the excitatory/inhibitory neurons rising from 400 to 3,500, which provides a guide to make a trade-off between the recognition accuracy and the resource cost during hardware implementation. Encouragingly, compared to its corresponding floating-point model, the proposed model reduces the hardware resources and power consumption by 40.7% and 36.3%, respectively (under 55-nm CMOS process). Published version

We report a high-mobility transparent Indium-Gallium-Zinc-Oxide (IGZO) thin-film transistor (TFT) with sputtered AlOx passivation layer. The interfacial region between the IGZO layer and the AlOx layer played a crucial role in improving the field-effect mobility (the maximum field-effect mobility increased from 6.292 cm2 Vs-1 for the TFT without the AlOx layer to 69.01 cm2 Vs-1 for the TFT with the passivation layer) and the on/off current ratio (from ∼107 without the layer to ∼108 with the layer). The driving current of IGZO TFT was also significantly enhanced. The formation of the interfacial layer has been investigated and verified. The ion bombardment during the AlOx deposition broke the In-O bond in IGZO, generating oxygen ions (O2-). The segregation of the O2- was facilitated by the sputtered amorphous AlOx. A metallic In-rich layer with high oxygen vacancy concentration was formed at the interface, leading to an increase in the carrier concentration in the interfacial layer. Besides the electrical performance, the reliability tests, including long-term exposure in the ambient environment and positive bias illumination stress (PBIS), showed improved results as well. Ministry of Education (MOE) National Research Foundation (NRF) Published version This work was supported by the National Research Foundation of Singapore (Program Grant No. NRFCRP13-2014-02) and Ministry of Education (MOE) of Singapore (MOE Tier 1 Grant No. RG144/ 20). We also would like to acknowledge the Facility for Analysis, Characterization, Testing and Simulation, Nanyang Technological University, Singapore, for use of their electron microscopy/X-ray facilities.

Transparent resistive switching random access memory (ReRAM) is of interest for the future integrated invisible circuitry. However, poor understanding of its working mechanism in transparent ReRAMs with the indium tin oxide (ITO) electrode is still a critical problem and will hinder its widespread applications. To reveal the actual working mechanism in transparent ReRAMs with the ITO electrode, we investigate the transparent ITO/SiO

In this paper, tungsten oxide (WO3) thin films and hybrid polyelectrolyte have been synthesized by solution-based processes to form solid-state electrochromic devices (ECDs). Electrochromic properties of the ECDs have been investigated, focusing on the near-infrared (NIR) region of the light spectrum. By optimizing the applied impulse voltage and WO3 film thickness, a coloration efficiency of 78.8 cm2/C has been achieved in the ECD under the impulse voltage of ± 2.5 V. In addition, at the same voltage, the response time of 2.5 s and 15.8 s was obtained for bleaching and coloring, respectively. The fast switching between colored and bleached states is due to the porous structure of solution-processed WO3 films that facilitates injection/release of Li ions. Stable current-time profiles have been obtained during cyclic measurements with alternating impulse voltages within 6 × 103 s, indicating excellent reversibility in the ECDs. The solid-state ECDs based on WO3 films are promising to be used in applications requiring NIR-related optical modulation such as smart windows.

The ability to selectively scatter green light is essential for an RGB transparent projection display, and this can be achieved by a silver-core, titania-shell nanostructure (Ag@TiO2), based on the metallic nanoparticle's localized surface plasmon resonance. The ability to selectively scatter green light is shown in a theoretical design, in which structural optimization is included, and is then experimentally verified by characterization of a transparent film produced by dispersing such nanoparticles in a polymer matrix. A visual assessesment indicates that a high-quality green image can be clearly displayed on the transparent film. For completeness, a theoretical design for selective scattering of red light based on Ag@TiO2 is also shown. NRF (Natl Research Foundation, S’pore) Accepted version

In this paper, electrical characteristics of an HfO2-based resistive switching memory device are investigated before and after γ-ray radiation with various total ionizing doses (TIDs). The device can still function properly even if irradiated with a TID of 20 Mrad(Si). The small changes of resistance states and set/reset voltages induced by γ-ray radiation can hardly influence the proper function of the device. The γ-ray radiation does not significantly degrade both retention and endurance characteristics even after a high-TID exposure. The radiation effects on the resistive switching memory device show little dependence on the cell area. The results suggest that the HfO2-based resistive switching memory device has good γ-ray radiation-resistant capability.

This work aims at finding a HfO2-based resistive random-access memory (RRAM) structure suitable for the integration of one RRAM with one InGaZnO thin film transistor (TFT) for large-area applications such as flexible electronic circuits. One of the major concerns is that the compliance current (CC) required for the formation of stable and strong conductive filaments in the forming and set processes as well as the maximum current required in the reset process in a large-size RRAM should be lower than that of the maximum current a TFT can deliver. In this work, an ultrathin Al2O3 layer of 2 nm was inserted between the HfO2 switching layer and the reactive Ti layer of the top electrode in the RRAM with the structure of Pt (bottom electrode)/HfO2/Al2O3/Ti/TiN (top electrode). With the ultrathin Al2O3 layer, the forming voltage was greatly reduced, and the CC for stable forming and set operations and maximum reset current can reach a low current level that an InGaZnO TFT is able to provide, while the device-to-device variation of the forming operation and cycle-to-cycle resistance variations of the set and reset operations are improved significantly.

With its high energy efficiency and ultra-high speed, processing-in-memory (PIM) technology is promising to enable high performance in Reservoir Computing (RC) systems. In this work, we demonstrate an RC system based on an as-fabricated PIM chip platform. The RC system extracts input into a high-dimensional space through the nonlinear characteristic and randomly connected reservoir states inside the PIM-based RC. To examine the system, nonlinear dynamic system predictions, including nonlinear auto-regressive moving average equation of order 10 driven time series, isolated spoken digit recognition task, and recognition of alphabet pronunciation, are carried out. The system saves about 50% energy and requires much fewer operations as compared with the RC system implemented with digital logic. This paves a pathway for the RC algorithm application in PIM with lower power consumption and less hardware resource required. Published version This work was supported by NSFC under Project Nos. 61774028, 92064004, and 61771097.

By using double TiNxOy absorption layers with low N/O ratio (TiNO_L) and high N/O ratio (TiNO_H), a spectral selective solar absorber (SSA) with the structure of SiO2-TiO2-TiNO_L-TiNO_H-Cu has been proposed. Optical properties of the TiNO_L and TiNO_H layers are investigated with spectroscopic ellipsometry (SE) in the wavelength range of 0.3–2.5 μm using an optical dispersion model in which the absorption mechanisms in TiNxOy including the localized surface plasmon resonance (LSPR), free electron absorption and interband transitions are taken into account. Because TiNO_H has a higher free electron concentration than that of TiNO_L, the absorption due to both the LSPR and free electron absorption in TiNO_H is found larger than that in TiNO_L. With the model parameters of the dispersion model obtained from the SE modeling, the reflectance spectrum of the SSA in the wide wavelength range of 0.3–25 μm covering both the solar radiation and heat radiation of a hot surface has been simulated; and the simulated result agrees well with the experimental result.

In this paper, a reconfigurable and scalable spiking neural network processor, containing 192 neurons and 6144 synapses, is developed. By using deep compression technique in spiking neural network chip, the amount of physical synapses can be reduced to 1/16 of that needed in the original network, while the accuracy is maintained. This compression technique can greatly reduce the number of SRAMs inside the chip as well as the power consumption of the chip. This design achieves throughput per unit area of 1.1 GSOP/( $\text{s}\!\cdot\!\text{mm}^2$ ) at 1.2 V, and energy consumed per SOP of 35 pJ. A 2-layer fully-connected spiking neural network is mapped to the chip, and thus the chip is able to realize handwritten digit recognition on MNIST with an accuracy of 91.2%.

In this work, a resistive switching memory device with Ag/(NiO/Al2O3)3/fluorine-doped SnO2 structure was fabricated with solution-based process including spin-coating of triple-layered NiO/Al2O3 films and ink-jet printing of Ag electrodes. Bipolar resistive switching characteristic was observed in such a structure, with the resistance ratio between high and low resistance states over two orders and good cycling stability in voltage sweeping measurements. More importantly, the SET/RESET voltages, which were in the range of 0.8‒2.4 V and − 0.7 to − 2.8 V, respectively, showed significant improvement in voltage distribution (78.4% and 71.2% narrower) as compared to devices solely based on Al2O3 film. It is believed that the narrow distribution of SET and RESET voltages results from the reduced randomness of the formation and rupture of conductive filaments under applied voltages, since NiO and Al2O3 have different dielectric constants and the distribution of electric field in the multilayers can be varied to facilitate the formation and rupture of conductive filaments. Moreover, electroforming process was not required to activate the device based on triple-layered NiO/Al2O3 films. The characterization results especially the narrow SET/RESET voltage distribution and forming-free property make devices based on multilayered NiO/Al2O3 thin films promising for thin film-based nonvolatile memory applications.

Selective scattering of red, green, and blue light and transmitting other visible light to achieve a transparent projection screen has been proposed based on metallic nanoparticle's localized surface plasmon resonance (LSPR). However, of the many proposed structures, dielectric (TiO2) substrate/silver (Ag) nanocube structure was only demonstrated to selectively scatter blue light in the backward direction. Therefore, for this setup, on one hand, it is of interest to enhance selective backward scattering of red light, and on the other hand, it is of interest to investigate ways of selective forward scattering of both blue and red light. Selective backward scattering may be used in applications such as heads up display (HUD), while selective forward scattering may be applied to, for example, display advertisements on shop windows. In this work, through numerical simulation, forward and backward scattering properties of dielectric substrate/Ag (or gold) nanocube structures were investigated. Based on these properties, three designs are proposed which can achieve selective scattering of blue and red light in forward or backward or both directions. NRF (Natl Research Foundation, S’pore) Published version

Novel desalination technologies with high ion removal capacity and low energy consumption are urgently needed to solve the water scarcity problem. Here we report a novel energy efficient dual-ions electrochemical deionization (DEDI) system with Ag nanoparticles/reduced graphene-oxide (AgNPs/rGO) as chloride ion Faradaic electrode and NaTi2(PO4)3/reduced graphene-oxide (NTP/rGO) as sodium ion Faradaic electrode. During the intercalation process, the sodium ions and chloride ions in the feed solution will be chemically intercalated into NTP/rGO electrode and AgNPs/rGO electrode, respectively. The DEDI system shows a stable and reversible salt removal capacity of 105 mg g−1 for 50 cycles with applied voltage range of −1.2–1.4 V. More importantly, when applying from 0 V to 1.4 V, although the removal capacity is relatively low (35.8 mg g−1), the energy recovery of this system is higher than 30% and the energy consumption is as low as 0.127 Wh g−1. Considering the brackish water used here is 2500 ppm, the energy consumption can be estimated to be 0.254 Wh L−1 for desalination of brackish water to drinkable water (500 ppm). The excellent performance of this DEDI system has made it a promising commercial technology for desalination of brackish water even seawater in the future. Environment & Water Industry Development Council (EWI) National Research Foundation (NRF) Public Utilities Board (PUB) This research work is supported by the National Research Foundation of Singapore, Prime Minister’s Office under its Environment & Water Research Programme with Grant No. 1301-IRIS-17 and administered by the Environment & Water Industry Programme Office (EWI) of the PUB, Singapore’s national agency.

In this paper, a 2 GHz LC-VCO with neural network (Multilayer Perceptron) has been designed in a 0.13 ţm CMOS technology. With the integrated neural network, the linearity and tuning range of the LC-VCO has been substantially improved. Compared to a conventional VCO design, the proposed technique can improve the linearity by selecting optimized bias voltages obtained from the output of the neuron network. The result shows that the tuning nonlinearity of the proposed VCO is further optimized from 0.335% to 0.254%. Published version

In this paper, an RF low noise amplifier (LNA) with self-recovery capability has been designed and implemented. A degradation model of hot carrier injection (HCI) of n-channel MOSFETs is proposed to simulate the aging process of the RF circuits, and a method for monitoring the HCI degradation in the RF circuits has been developed. Self-recovery mechanism of the LNA is triggered automatically by monitoring the HCI degradation to compensate for the HCI degradation. With the self-recovery capability, the LNA can maintain its performance under HCI stress over time. The proposed LNA has been fabricated with a 0.13μm CMOS technology and the self-recovery capability has been experimentally demonstrated. Published version

Winner-takes-all (WTA), an important mechanism in neural networks of recurrently connected neurons, is a critical element of many models of cortical processing. However, few WTA neural networks have been realized physically, especially by memristor networks. In this work, we have designed and implemented a neural network with memristor-based synapses to realize the WTA in a neural system. Neuronal self-excitatory, excitatory, and inhibition by other neurons have been demonstrated. Competitions between two neurons, among three neurons, and between two groups of neurons are realized based on the memristive neural network. The winner neuron or winner group can suppress the other neuron(s) or other group(s) of neurons and dominate the neuronal firing. This work paves the way for further realization of complex models of cortical processing with memristive neural networks.Winner-takes-all (WTA), an important mechanism in neural networks of recurrently connected neurons, is a critical element of many models of cortical processing. However, few WTA neural networks have been realized physically, especially by memristor networks. In this work, we have designed and implemented a neural network with memristor-based synapses to realize the WTA in a neural system. Neuronal self-excitatory, excitatory, and inhibition by other neurons have been demonstrated. Competitions between two neurons, among three neurons, and between two groups of neurons are realized based on the memristive neural network. The winner neuron or winner group can suppress the other neuron(s) or other group(s) of neurons and dominate the neuronal firing. This work paves the way for further realization of complex models of cortical processing with memristive neural networks.

Transparent nonvolatile memory has great potential in integrated transparent electronics. Here, we present highly transparent resistive switching memory using stoichiometric WO3 film produced by cathodic electrodeposition with indium tin oxide electrodes. The memory device demonstrates good optical transmittance, excellent operative uniformity, low operating voltages (+0.25 V/–0.42 V), and long retention time (>104 s). Conductive atomic force microscopy, ex situ transmission electron microscopy, and X-ray photoelectron spectroscopy experiments directly confirm that the resistive switching effects occur due to the electric field-induced formation and annihilation of the tungsten-rich conductive channel between two electrodes. Information on the physical and chemical nature of conductive filaments offers insightful design strategies for resistive switching memories with excellent performances. Moreover, we demonstrate the promising applicability of the cathodic electrodeposition method for future resistive ...