Search

In this article, a partially isolated dual work function (PIDWF) gate In-Ga-Zn-O (IGZO) thin-film transistor (TFT) is proposed to reduce the off-state current (Ioff) obviously, which also provides a feasible integration method for stacking IGZO TFT on Si-based devices. It is found that compared with the general back gate IGZO TFT structure, the Ioff of the proposed IGZO TFT reduces from $2.57\times 10{^{-}14 }$ A/ $\mu $ m to $7.57\times 10{^{-}16 }$ A/ $\mu $ m, achieving two orders of magnitude improvement. This breakthrough has the potential to increase the retention time of DRAM applications by nearly 100 times. Moreover, the pronounced novel structure has mitigated parasitic capacitance, thereby leading to a notable 47.7% reduction in write latency within dynamic-random-access-memory (DRAM) circuits. The relevant operation mechanism is carefully demonstrated and verified by the simulation of the electric field and potential barrier results by technical computer-aided design (TCAD). Furthermore, the impacts of the dual gate work function level, the length, and the type of isolation dielectric between dual work function gates are systematically investigated. The results show that the off-state leakage is further reduced by increasing the difference of the work function levels between in dual gates, the dielectric length (LD) and using the isolation layer with a lower dielectric constant. The PIDWF gate IGZO TFT exhibits scalability and is capable of achieving an 84.6% reduction in leakage current even with ultra-short channel lengths, which offers a promising application for future 3D DRAM applications with little extra cost.

Membranous nephropathy (MN) is a common pathological type of nephrotic syndrome. The characteristic of MN is the presence of immune complex deposits containing immunoglobulin G (IgG), called electron-dense deposits (EDD), which can be observed by transmission electron microscopy (TEM). Quantitative analysis of the morphology and location of EDD can provide an essential reference for diagnosing and staging MN. However, accurately identifying and quantifying EDD is challenging due to their different morphologies, sizes, and locations with varying amounts. This paper proposes a two-stage Deformable R-CNN detector that overcomes these challenges, which has two characteristics: 1) The detector employs InternImage as the feature extractor, which extracts different morphological features of EDD using the core operator deformable convolution v3 (DCNv3). 2) The detector utilizes the multi-scale deformable attention model (MSDAM) as the attention mechanism to detect EDD of different sizes and locations effectively. The proposed Deformable R-CNN was tested on the Electron-Dense Deposit Detection for Membranous Nephropathy (EDDD-MN) dataset and outperformed other popular detectors, including two-stage, one-stage, and transformer-based detectors in detection and quantification. It also exhibited excellent performance in TIDE error analysis. Thus, this method would enable precise detection and rapid quantification of EDD, thereby reducing the workload of pathologists and helping them gain a comprehensive understanding of MN.

With the development of brain-computer interfaces (BCI) technologies, EEG-based BCI applications have been deployed for medical purposes. Motor imagery (MI), applied to promote neural rehabilitation for stroke patients, is among the most common BCI paradigms that. The Electroencephalogram (EEG) signals, encompassing an extensive range of channels, render the training dataset a high-dimensional construct. This high dimensionality, inherent in such a dataset, tends to challenge traditional deep learning approaches, causing them to potentially disregard the intrinsic correlations amongst these channels. Such an oversight often culminates in erroneous data classification, presenting a significant drawback of these conventional methodologies. In our study, we propose a novel algorithmic structure of EEG channel-attention combined with Swin Transformer for motor pattern recognition in BCI rehabilitation. Effectively, the self-attention module from transformer architecture could captures temporal-spectral-spatial features hidden in EEG data. The experimental results verify that our proposed methods outperformed other state-of-art approaches with the average accuracy of 87.67%. It is implied that our method can extract high-level and latent connections among temporal-spectral features in contrast to traditional deep learning methods. This paper demonstrates that channel-attention combined with Swin Transformer methods has great potential for implementing high-performance motor pattern-based BCI systems.

Immersive virtual reality technology (IVR) can create contextualized learning environments that learners cannot easily access. It is widely used in education. More and more researchers are paying attention to IVR’s influencing factors. However, most of these researches focus on the aspects of technology and environment, ignoring the aspects of learners themselves. Therefore, this paper explores the impact of learners’ attention level (AL) and learning style (LS) on learning behavior in IVR. Firstly, the AL data monitored by EEG equipment allows correlation and difference analysis to explore the relationship among AL, LS and learning performance (LP). Then, according to the video-recorded learning behavior data, the lag sequence analysis method is used to analyze the learning behavior sequence transformation of the high-concentration group and low-concentration group so that it can explore the problem learning behaviors of learners with different concentrations. The results show that: in the virtual learning environment (Vir-LE), there is a strong positive correlation between AL and LP. There is no significant difference in LP with different LS, but the AL of visual learners is higher than that of verbal learners. Through this experiment, it is helpful to eliminate the interference factors in the range of human subjective perception and improve the accuracy of the measurement of learning effect.

Motor-modality-based brain computer interface (BCI) could promote the neural rehabilitation for stroke patients. Temporal-spatial analysis was commonly used for pattern recognition in this task. This paper introduced a novel connectivity network analysis for EEG-based feature selection. The network features of connectivity pattern not only captured the spatial activities responding to motor task, but also mined the interactive pattern among these cerebral regions. Furthermore, the effective combination between temporal-spatial analysis and network analysis was evaluated for improving the performance of BCI classification (81.7%). And the results demonstrated that it could raise the classification accuracies for most of patients (6 of 7 patients). This proposed method was meaningful for developing the effective BCI training program for stroke rehabilitation.

In this paper, we consider the performance of exclusive-OR (XOR) rule in detecting the presence or absence of a deterministic signal in bivariate Gaussian noise. Signals, when present at the two sensors, are assumed unequal, whereas the noise components have identical marginal distribution but are correlated. The sensors send their one-bit quantized data to a fusion center, which then employs the XOR rule to arrive at the final decision. Here we prove that, in the limit as the correlation coefficient $r$ approaches 1, the optimum fusion rule for both parallel and tandem topologies is XOR with identical, alternating partitions (XORAP) of the observations at the sensors. We further quantify the asymptotic decrease of the Bayes error of XORAP towards zero as a constant multiplied by $\sqrt {1-r}$ , as $r$ approaches 1. When compared to the asymptotic Bayes error of CLRT, which decreases to zero exponentially fast, as a function of $1/(1-r)$ , the Bayes error of XORAP decreases to zero much slower.

A new approach of narrowing sub-fin with little extra process cost for suppressing parasitic-channel-effect (PCE) on vertically-stacked horizontal gate-all-around (GAA) Si nanosheet field-effect-transistors (NS-FETs) is proposed. The proposed sub-fin design demonstrates systematical technical advantages by calibrated 3D TCAD simulation, including 70% reduction in sub-channel gate-induced drain leakage (GIDL) current, over 20% promotion for on-off current ratio ( $\text{I}_{\mathrm{ on}}/\text{I}_{\mathrm{ off}}$ ) as well as improvement in sub-threshold slope (SS). The revealed narrow sub-fin offers nearly 10% on-state current promotion and gate controllability improvement for the NS-FETs with relatively lower ground-plane-concentration. The narrow sub-fin technique provides a new approach for suppressing PCE in the NS-FETs and indicates a promising supplementary technology adopted for the optimization of NS-FET fabrication process in sub-3nm technology node.

Most of the existing knowledge representation learning methods project the entities and relations represented by symbols in the knowledge graph into the low-dimensional vector space from the perspective of the structure and semantics of triples, and express the complex relations between entities and relations with dense low-dimensional vectors. However, triples in the knowledge graph not only contain relation triples, but also contain a large number of attribute triples. Existing knowledge representation methods often confuse these two kinds of triples and pay little attention to the semantic information contained in attributes and attribute values. In this paper, a novel representation learning method which makes use of the attribute information of entities is proposed. Specifically, deep convolutional neural network model is used to encode attribute information of entities, and both attribute information and triple structure information are utilized to learn knowledge representation, and then generate attribute-based representation of entities. The knowledge graph completion task was used to evaluate this method, and the experimental results on open data sets FB15K and FB24k showed that the attribute-embodied knowledge representation learning model outperforms the other baselines.

Multi-agent deep reinforcement learning (MDRL) is an emerging research hotspot and application direction in the field of machine learning and artificial intelligence. MDRL covers many algorithms, rules and frameworks, it is currently researched in swarm system, energy allocation optimization, stocking analysis, sequential social dilemma, and with extremely bright future. In this paper, a parallel-critic method based on classic MDRL algorithm MADDPG is proposed to alleviate the training instability problem in cooperative-competitive multi-agent environment. Furthermore, a policy smoothing technique is introduced to our proposed method to decrease the variance of learning policies. The suggested method is evaluated in three different scenarios of authoritative multi-agent particle environment (MPE). Multiple statistical data of experimental results show that our method significantly improves the training stability and performance compared to vanilla MADDPG.

In recent years, deep reinforcement learning methods have achieved impressive performance in many different fields, including playing games, robotics, and dialogue systems. However, there are still a lot of restrictions here, one of which is the demand for massive amounts of sampled data. In this paper, a hierarchical meta-learning method based on the actor-critic algorithm is proposed for sample efficient learning. This method provides the transferable knowledge that can efficiently train an actor on a new task with a few trials. Specifically, a global basic critic, meta critic, and task specified network are shared within a distribution of tasks and are capable of criticizing any actor trying to solve any specified task. The hierarchical framework is applied to a critic network in the actor-critic algorithm for distilling meta-knowledge above the task level and addressing distinct tasks. The proposed method is evaluated on multiple classic control tasks with reinforcement learning algorithms, including the start-of-the-art meta-learning methods. The experimental results statistically demonstrate that the proposed method achieves state-of-the-art performance and attains better results with more depth of meta critic network.

Tunnel field-effect transistors (TFETs) are promising for use in ultralow-power applications owing to their distinct band-to-band tunneling operation. However, the ON-state current of these Si-based tunnel devices is considerably lower than that of MOSFETs. Furthermore, a high-κ dielectric is used as a gate insulator or spacer to increase the TFET's tunneling field strength; this usually causes fringe-induced barrier lowering and deterioration of its subthreshold characteristics. Here, in this paper, a gate field plate TFET (GFP-TFET) structure is proposed to enhance the driving current and suppress the current kink induced by fringe-induced barrier lowering. A detailed investigation into the effects of a gate field plate on TFET performance was conducted with the help of extensive device simulations. The gate oxide and gate field plate oxide could be used to tailor the electric field lines to increase the electric field in the tunnel junction. Meanwhile, the fringing field concentration was reduced by using a metal with a relatively high work function as the field plate electrode. Thus, the electric field strength decreased near the gate edge and increased in the channel. The GFP-TFET exhibited high driving current and low subthreshold characteristics at the cost of a reduction in cutoff frequency. A GFP-TFET structure with optimized parameters had a 103× higher ON-state current compared with a normal silicon tunnel FET. The minimum subthreshold swing decreased from 146.3 to 21.4 mV/dec and the ON/OFF current ratio increased from 107 to 1013 compared with a normal TFET.

Speller had been proved that it's a kind of well interactive manners for brain computer interface (BCI) system. In this study, we proposed a novel steady-state visual evoked potential (SSVEP) BCI speller developed for numerical input. Based on a previous off-line method of SSVEP recognition, a sliding control protocol was used for our real-time spelling task. For ten subjects, on-line experiments of 10 consecutive number inputs were conducted for two different control conditions. In contrast to traditional static protocol of multi-phase SSVEP signal extraction, the average information transmission rate (ITR) of sliding control protocol reached 23.45 bits/min, higher than that of traditional static protocol (19.85 bits/min). The results showed the validity and high-efficiency of sliding control paradigm for a real-time multi-phase SSVEP speller.

Despite deep reinforcement learning has attained performance beyond human beings in many domains, including games, dialogue systems and robotics, sample inefficient is still a limitation in the application of deep reinforcement learning. This paper develops a novel and simple gradient-based meta learning method suitable for improving learning efficiency of deep reinforcement learning methods. Rather than designing complex network or adding excessive fine tuning parameters, the proposed method is simple and does not introduce any learned parameters for meta learning. Specifically, according to the characteristic of different trajectories, this paper proposes to weight every trajectory in model-agnostic meta-learning for meta updating gradient effectively. The key idea underlying this method is to take advantage of the relationship between different trajectories and the direction of parameter update. Additionally, an end-to-end training approach is also introduced so that the proposed model can attain good results with a small amount of training data on new tasks. The statistical results of the experiments indicate that the proposed algorithm delivers state-of-the-art performance on both discrete and continuous control tasks.

Benefiting from the research of machine learning (ML) and deep learning(DL), multivariate methods based on ML and DL have been the mainstream and successful analysis methods in Neural Engineering or Neuroimaging research, for example, assisting diagnosis based on brain Magnetic Resonance Imaging (MRI). However, many existing methods based on traditional ML methods cannot sufficiently extract discriminative features, especially feature patterns across long-distance brain areas, resulting in unsatisfactory classification performance. Designing an effective and robust classifier for different MRI images remains a challenge. In this paper, we introduced dilated 3D CNN method for classifying 3D MRI images combining CNN structure and dilated convolution with a small number of feature maps. We also presented a methodology framework based on dilated 3D CNN method, which can classify both single MRI images and image sequences. Our method and framework were evaluated on the structural MRI images of ADHD-200 dataset and fMRI images of a Schizophrenia dataset, demonstrating better performances than some other state-of-the-art methods.

Nowadays, somatosensory devices were widely used for developing the rehabilitation systems for limb-injured patients. However, professional evaluation was rarely studied in this field. In our study, we presented a novel hybrid deep network combined long short-term memory (LSTM) network and convolutional neural network (CNN) for rehabilitation evaluation referred to Brunnstrom Scale. In the identification task of 3-class Brunnstrom stages (III, IV, V), the mean accuracy of our proposed model was up to 80% (84.1%). The experimental result validated the reliability of our proposed evaluation method. And the comparison result of three machine learning algorithms indicated that the superiority of our hybrid model for Kinect-based 3D data analysis.