Search

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 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.

Silicon nanoparticle (Si-NP)-embedded silicon nitride (Si3N4) thin films have been synthesized by implantation of Si ions into Si3N4 thin films followed by high-temperature thermal annealing. With different implant dosage of Si ions, the concentration of Si-NPs has been varied in the Si3N4 matrix. By forming an Al/Si-NP-embedded Si3N4/p-Si structure, memory behavior was observed through charging-caused modulation in the device current. The current–voltage measurements were then conducted to study the carrier transport mechanism and thus to understand the origin of charging-induced variation in device resistance. It was found that the current exhibited a hopping-based conduction mechanism at low electric field. While at high electric field, a Frenkel–Poole (F–P) emission was found to dominate the current conduction. As a result, the charging-caused electron trapping under positive voltage in Si-NPs of the nitride film enhances the F–P emission, leading to a significant reduction in resistance. However, negative voltage-caused hole trapping suppresses the current conduction. The two-terminal devices based on such Si-NP-embedded Si3N4 thin films are promising to be used as charging-controlled memory devices.

The resistive switching device based on a Ni-rich nickel oxide thin film exhibits an inherent learning ability of a neural network. The device has the short-term-memory and long-term-memory functions analogous to those of the human brain, depending on the history of its experience of voltage pulsing or sweeping. Neuroplasticity could be realized with the device, as the device can be switched from a high-resistance state to a low-resistance state due to the formation of stable filaments by a series of electrical pulses, resembling the changes such as the growth of new connections and the creation of new neurons in the brain in response to experience.

Strong visible electroluminescence (EL) with electrically tunable colors from violet to white has been observed from Si-implanted silicon nitride thin films. Influence of the implanted Si ion dose on both the current conduction and EL properties has been studied. With a larger excess Si concentration, the carrier transport is enhanced leading to a higher EL intensity, but the light emission efficiency is reduced. On the other hand, the increase of the excess Si concentration causes redshifts in the major EL bands and improves the transition of the EL colors with increasing current. The excess Si concentration is also found to have a significant influence on the EL degradation. These findings are important to the application of the Si-implanted thin films in light emitting devices.

Aluminum nitride (AlN) films have attractive physical properties, such as wide bandgap (6.2 eV), high thermal conductivity (320 W/mK), and good match of both thermal expansion coefficient and lattice constant to those of Si substrate . Based on these properties, AlN films can be applied in surface acoustic wave (SAW) devices, memories devices, and optoelectronic devices. It has been reported in our recent studies that amorphous AlN thin films containing aluminum nanocrystals (nc-Al) deposited on Si substrate by radio-frequency (rf) magnetron sputtering possess memory effect and exhibit interesting current conduction behaviors.

Conduction switching, i.e., a sharp change in the conduction from a lower-conductance state to a higher-conductance state or vice versa in aluminum nitride thin films embedded with Al nanocrystals (nc–Al) has been observed in the ramped-voltage and ramped-current current–voltage (I–V) measurements and the time-domain current measurement as well. Each state is well defined and its I–V characteristic follows a power law. It is observed that the conductance decreases (or increases) with charging (or discharging) in the nc–Al. It is shown that the conduction switching is due to the charging and discharging in the nc–Al at certain strategic sites. With the connecting (or breaking) of some conductive tunneling paths formed by the uncharged nc–Al due to the discharging (or charging) in the nc–Al at the strategic sites, a conduction switching occurs.

Although synaptic behaviours of memristors have been widely demonstrated, implementation of an even simple artificial neural network is still a great challenge. In this work, we demonstrate the associative memory on the basis of a memristive Hopfield network. Different patterns can be stored into the memristive Hopfield network by tuning the resistance of the memristors, and the pre-stored patterns can be successfully retrieved directly or through some associative intermediate states, being analogous to the associative memory behaviour. Both single-associative memory and multi-associative memories can be realized with the memristive Hopfield network.

Localized surface plasmon resonance (LSPR) and electron confinement effects on the interband transitions and free-electron absorption in ultrathin Au films with the film thicknesses of about 1–12 nm are investigated. A significant evolution of the LSPR, interband transition energies, plasma energy, and conductivity with the film thickness is observed. The evolution is attributed to the changes in the size, shape, and spacing of the self-assembled gold nanoparticles (Au NPs) with the sizes from several nanometers to 110 nm as well as the aggregation of the Au NPs in the Au films.

In the fabrication of micro-devices and systems, wafer bonding offers a unique opportunity for constructing complicated three-dimensional structures. In this paper, a wafer bonding technique, called transmission laser bonding (TLB), is studied with focus on the effects of interface oxidation and contact pressure on the bonding strength. The TLB is implemented for bonding Pyrex glass-to-silicon wafers, with and without interface oxide layers, using a Q-switch pulsed Nd:YAG laser. The tensile strengths of the TLB bonded specimens are comparable to those generated by the existing major wafer bonding techniques. The advantages of TLB are also discussed with some details. The oxide thickness is measured by spectro-reflectometry while the roughness of the oxidized surfaces is quantified using Atomic Force Microscopy (AFM). The bonded interfaces are analyzed by X-ray Photoelectron Spectroscopy (XPS) and Auger Electron Spectroscopy (AES) to study the migration and diffusion of different atoms across the bonding interface and to provide the necessary information for the understanding of the bonding mechanism. A thermal penetration analysis is also provided to validate the findings of the bond strength and spectroscopic evaluations.

Tunable light resonance transportation along a single long Au hemisphere nanoparticles (NPs) chain was studied. The realistic experimentally determined gold dielectric function was used for the simulation of Au localized surface plasmon polariton (LSPPs) effect. The resonance light energy with minimized attenuation and its bandwidth were quantitatively analyzed by inducing the effective mass which was observed to increase only with the length of Au NPs between the source and the test point. The geometric ratio g/r of NP size and gap were investigated at 5 µm far of NPs with different gaps from 0 to 70 nm. Strongest resonance can be achieved with g/r = 1.2 by the factor of 1.5 than the connected NPs. This resonance mode falls in the wavelength λ = 555 nm (green light), which is exactly the maximum sensitivity of a light-adapted eye of human beings.

In this work, Si ions are implanted into the gate oxide of MOSFETs with different implantation schemes, followed by a high-temperature annealing. The memory characteristics of the MOSFETs have been investigated for the following two excess Si distributions: (1) the excess Si is distributed in a narrow layer in the gate oxide near the Si substrate; and (2) the excess Si is distributed throughout the gate oxide. It is observed that both the excess Si distributions have good endurance of up to 106 program/erase cycles. The second excess Si distribution exhibits a better retention characteristic with less than 50% charge loss after 10 years. In contrast, the first excess Si distribution shows a complete charge loss after 1 year.