Search

Exposure to varied noisy environments impairs the recognition performance of artificial intelligence-based speech recognition technologies. Degradedperformance services can be utilized as limited systems that assure good performance in certain environments, but impair the general quality of speech recognition services. This study introduces an audiovisual speech recognition (AVSR) model robust to various noise settings, mimicking human dialogue recognition elements. The model converts word embeddings and log-Mel spectrograms into feature vectors for audio recognition. A dense spatial–temporal convolutional neural network model extracts features from log-Mel spectrograms, transformed for visual-based recognition. This approach exhibits improved aural and visual recognition capabilities. We assess the signalto-noise ratio in nine synthesized noise environments, with the proposed model exhibiting lower average error rates. The error rate for the AVSR model using a three-feature multi-fusion method is 1.711%, compared to the general 3.939% rate. This model is applicable in noise-affected environments owing to its enhanced stability and recognition rate.

Metal-ferroelectric-metal-insulator-semiconductor (MFMIS) FeFETs have significant potential for use in non-volatile memory applications. This is primarily due to their compatibility with CMOS technology and reliable switching characteristics. Previous studies have primarily concentrated on the endurance and memory window properties, while this study focuses on the short-term (< $1~\mu $ s) retention region of MFMIS FeFETs. Specifically, we examine the impact of the capacitance ratio of the ferroelectric capacitor (CFE) and the MOS capacitor (CDE) on short-term retention. Additionally, we conducted simulations to validate the experimental observations and investigate the interaction of the depolarization field with the charge trapping and polarization of the MFMIS structure. This study emphasizes the crucial role of controlling the CDE: ${\mathrm { C}}_{\mathrm { FE}}$ ratio in enhancing the short-term retention of MFMIS FeFETs. Its findings enhance our understanding of short-term retention mechanisms and provide a pathway for improving performance and functionality in non-volatile memory technology design.

Abstract The development of artificial tactile receptor systems is important in the fields of prosthetic devices, interfaces for the metaverse, and sensors. A pressure sensor and memory device may be used in this system to replicate the tactile detecting capabilities of human skin. The implementation of systems that take into account mass production and miniaturization is still difficult. Here, a flexible artificial tactile receptor built using conventional semiconductor processes that combine a vertically stacked piezoelectric sensor with neuromorphic memory is presented. As a fundamental component for both sensors and memory, hafnium zirconium oxide (HZO) formed by using semiconductor deposition technique is introduced. Due to its exceptional piezoelectric performance, the morphotropic phase boundary of HZO is studied. The entire materials and processes are highly compatible with conventional semiconductor processes, including microwave annealing‐based low‐temperature crystallization. Even after 10,000 times of bending stress, the sensor and memory constructed on a flexible substrate exhibit consistent pressure detection characteristics over a wide range of 2–25 kPa. The feasibility of the approach is further demonstrated by a deep neural network simulation, which reached 90.8% braille recognition accuracy. Wearable electronics and medical devices are two examples of industrial domains that can use these flexible, exceptionally durable devices.

Abstract In the present hyper-scaling era, memory technology is advancing owing to the demand for high-performance computing and storage devices. As a result, continuous work on conventional semiconductor-process-compatible ferroelectric memory devices such as ferroelectric field-effect transistors, ferroelectric random-access memory, and dynamic random-access memory (DRAM) cell capacitors is ongoing. To operate high-performance computing devices, high-density, high-speed, and reliable memory devices such as DRAMs are required. Consequently, considerable attention has been devoted to the enhanced high dielectric constant and reduced equivalent oxide thickness (EOT) of DRAM cell capacitors. The advancement of ferroelectric hafnia has enabled the development of various devices, such as ferroelectric memories, piezoelectric sensors, and energy harvesters. Therefore, in this review, we focus the morphotropic phase boundary (MPB) between ferroelectric orthorhombic and tetragonal phases, where we can achieve a high dielectric constant and thereby reduce the EOT. We also present the role of the MPB in perovskite and fluorite structures as well as the history of the MPB phase. We also address the different approaches for achieving the MPB phase in a hafnia material system. Subsequently, we review the critical issues in DRAM technology using hafnia materials. Finally, we present various applications of the hafnia material system near the MPB, such as memory, sensors, and energy harvesters. Graphical Abstract

In recent years, substantial research efforts have been devoted toward the development of non-volatile memory devices (NVM). In this regard, devices utilizing interfacial dipole modulation (IDM) have been considered. However, the narrow memory window of current IDM stack structures (HfO2-TiO2-SiO2) has been a major constraint. Herein, we propose a structure that exploits SiOx interfacial dipole layers, thereby overcoming the memory window limitation of conventional IDM stack structures. A memory window up to 8.05 V was achieved by using SiOx switching species together with HfO and TiO encapsulating layers owing to the bidirectional oxygen relocation process, resulting in numerous electric dipoles. Furthermore, we found that the memory window increases as the ratio of SiOX to Si2p and varies with the position of the SiOX layer, implying that the SiOx layer drives the switching mechanism. This IDM structure with an enhanced memory window can be used to develop practical NVM devices.

Triacylglycerols (TG) play an important role in skin homeostasis including the synthesis of ω-O-acylceramides (acylCER) required for skin barrier formation by providing linoleic acid (C18:2n6). However, the overall relationships of TG species with various ceramides (CER) including CER-NP, the most abundant CER, ω-O-acylCER, and another acylCER, 1-O-acylCER in human SC, remain unclear. Therefore, we investigated these relationships and their influence on skin health status in healthy Korean adults. Twelve CER subclasses including two ω-O-acylCER and two 1-O-acylCER were identified with CER-NP consisting of approximately half of the total CER. The ω-O-acylCER species exhibited positive relationships with TG 52:4 and TG 54:2 containing C18:2, while interestingly, 1-O-acylCER containing ester-linked C14:0 and C16:0 demonstrated positive relationships with TG 46–50 including C14:0 and C16:0, respectively. In addition, CER-NP and CER-NH showed positive correlations with TG 52–54 containing C18:2 or C18:3. A lipid pattern with higher levels of CER including CER-NP and ω-O-acylCER with TG 54 and TG with 5–6 double bonds was related to good skin health status, especially with acidic skin pH. Collectively, TG with increased chain length and unsaturation seemed to improve CER content, and profiles such as higher acylCER and CER-NP improved skin health status by fortifying skin barrier structure.

Speech is a commonly used interaction-recognition technique in edutainment-based systems and is a key technology for smooth educational learning and user–system interaction. However, its application to real environments is limited owing to the various noise disruptions in real environments. In this study, an audio and visual information-based multimode interaction system is proposed that enables virtual aquarium systems that use speech to interact to be robust to ambient noise. For audio-based speech recognition, a list of words recognized by a speech API is expressed as word vectors using a pretrained model. Meanwhile, vision-based speech recognition uses a composite end-to-end deep neural network. Subsequently, the vectors derived from the API and vision are classified after concatenation. The signal-to-noise ratio of the proposed system was determined based on data from four types of noise environments. Furthermore, it was tested for accuracy and efficiency against existing single-mode strategies for extracting visual features and audio speech recognition. Its average recognition rate was 91.42% when only speech was used, and improved by 6.7% to 98.12% when audio and visual information were combined. This method can be helpful in various real-world settings where speech recognition is regularly utilized, such as cafés, museums, music halls, and kiosks.

Abstract In recent years, MoS2 has emerged as a prime material for photodetector as well as phototransistor applications. Usually, the higher density of state and relatively narrow bandgap of multi-layer MoS2 give it an edge over monolayer MoS2 for phototransistor applications. However, MoS2 demonstrates thickness-dependent energy bandgap properties, with multi-layer MoS2 having indirect bandgap characteristics and therefore possess inferior optical properties. Herein, we investigate the electrical as well as optical properties of single-layer and multi-layer MoS2-based phototransistors and demonstrate improved optical properties of multi-layer MoS2 phototransistor through the use of see-through metal electrode instead of the traditional global bottom gate or patterned local bottom gate structures. The see-through metal electrode utilized in this study shows transmittance of more than 70% under 532 nm visible light, thereby allowing the incident light to reach the entire active area below the source and drain electrodes. The effect of contact electrodes on the MoS2 phototransistors was investigated further by comparing the proposed electrode with conventional opaque electrodes and transparent IZO electrodes. A position-dependent photocurrent measurement was also carried out by locally illuminating the MoS2 channel at different positions in order to gain better insight into the behavior of the photocurrent mechanism of the multi-layer MoS2 phototransistor with the transparent metal. It was observed that more electrons are injected from the source when the beam is placed on the source side due to the reduced barrier height, giving rise to a significant enhancement of the photocurrent.

Flexible metal for Electronic Skin Amorphous alloys have never been such a viable approach for flexible electronics because now their excellent properties are not compromised by the structure and morphology. A collaborative team of scientists led by Prof Sanghun Jeon from Korea Advanced Institute of Science and Technology optimize the performance of a Fe33Zr67 metal alloy for multifunctional skin-like sensors. One critical step is that they understand and control the deposition condition to avoid the porous structure and obtain uniform amorphous alloy films. As a result, the Fe33Zr67 metal alloy possesses higher strength, lower Young's modulus and can be made into pressure sensors, temperature sensors and photosensors when integrated with In-Zn-O thin film transistors. This highly versatile and high performance alloy can be used in many scenarios, such as medical, healthcare and wearable electronics.

Concomitant with the recent advances in deep learning, automatic speech recognition and visual speech recognition (VSR) have received considerable attention. However, although VSR systems must identify speech from both frontal and profile faces in real-world scenarios, most VSR studies have focused solely on frontal face pictures. To address this issue, we propose an end-to-end sentence-level multi-view VSR architecture for faces captured from four different perspectives (frontal, 30°, 45°, and 60°). The encoder uses multiple convolutional neural networks with a spatial attention module to detect minor changes in the mouth patterns of similarly pronounced words, and the decoder uses cascaded local self-attention connectionist temporal classification to collect the details of local contextual information in the immediate vicinity, which results in a substantial performance boost and speedy convergence. To compare the performance of the proposed model for experiments on the OuluVS2 dataset, the dataset was divided into four different perspectives, and the obtained performance improvement was 3.31% (0°), 4.79% (30°), 5.51% (45°), 6.18% (60°), and 4.95% (mean), respectively, compared with the existing state-of-the-art performance, and the average performance improved by 9.1% compared with the baseline. Thus, the suggested design enhances the performance of multi-view VSR and boosts its usefulness in real-world applications.

Deep learning technology has encouraged research on noise-robust automatic speech recognition (ASR). The combination of cloud computing technologies and artificial intelligence has significantly improved the performance of open cloud-based speech recognition application programming interfaces (OCSR APIs). Noise-robust ASRs for application in different environments are being developed. This study proposes noise-robust OCSR APIs based on an end-to-end lip-reading architecture for practical applications in various environments. Several OCSR APIs, including Google, Microsoft, Amazon, and Naver, were evaluated using the Google Voice Command Dataset v2 to obtain the optimum performance. Based on performance, the Microsoft API was integrated with Google’s trained word2vec model to enhance the keywords with more complete semantic information. The extracted word vector was integrated with the proposed lip-reading architecture for audio-visual speech recognition. Three forms of convolutional neural networks (3D CNN, 3D dense connection CNN, and multilayer 3D CNN) were used in the proposed lip-reading architecture. Vectors extracted from API and vision were classified after concatenation. The proposed architecture enhanced the OCSR API average accuracy rate by 14.42% using standard ASR evaluation measures along with the signal-to-noise ratio. The proposed model exhibits improved performance in various noise settings, increasing the dependability of OCSR APIs for practical applications.

Hf0.5Zr0.5O2 films are one of the most attractive HfO2-based ferroelectric films because of good ferroelectricity, extreme thinness, and excellent compatibility with silicon devices. The origin of the ferroelectricity of Hf0.5Zr0.5O2 films is the noncentrosymmetric orthorhombic phase (space group Pca21). The effects of process temperature, annealing temperature, thickness, and doping to increase the portion of the orthorhombic phase, which contributes to ferroelectricity, have been studied extensively. However, although most studies have used atomic layer deposition, no study has been reported on the effect of the deposition rate on the ferroelectricity of Hf0.5Zr0.5O2 films. In this work, the influences of the deposition rate on the ferroelectricity and crystal structure of Hf0.5Zr0.5O2 films were examined. In order to conduct systematic and quantitative analysis, measurements of switching transient current, ferroelectric P-E curve, dielectric constant, deconvolution of grazing angle incidence X-ray diffraction, and piezoresponse force microscopy were performed. Hf0.5Zr0.5O2 films with a deposition rate of 1.1 Å/cycle have a more ideal hysteresis curve shape, higher remanent polarization (initial state: 16 μC/cm2, wake up state: 22 μC/cm2), and a higher orthorhombic phase portion than other deposition rates.

In visual speech recognition (VSR), speech is transcribed using only visual information to interpret tongue and teeth movements. Recently, deep learning has shown outstanding performance in VSR, with accuracy exceeding that of lipreaders on benchmark datasets. However, several problems still exist when using VSR systems. A major challenge is the distinction of words with similar pronunciation, called homophones; these lead to word ambiguity. Another technical limitation of traditional VSR systems is that visual information does not provide sufficient data for learning words such as “a”, “an”, “eight”, and “bin” because their lengths are shorter than 0.02 s. This report proposes a novel lipreading architecture that combines three different convolutional neural networks (CNNs; a 3D CNN, a densely connected 3D CNN, and a multi-layer feature fusion 3D CNN), which are followed by a two-layer bi-directional gated recurrent unit. The entire network was trained using connectionist temporal classification. The results of the standard automatic speech recognition evaluation metrics show that the proposed architecture reduced the character and word error rates of the baseline model by 5.681% and 11.282%, respectively, for the unseen-speaker dataset. Our proposed architecture exhibits improved performance even when visual ambiguity arises, thereby increasing VSR reliability for practical applications.

Abstract The carrier transport and device instability of amorphous oxide semiconductor devices are influenced by defects that are exponentially distributed in energy, because of amorphous phase channels and front/back interfaces with a large number of sub-gap states. Thus, understanding defects and charge trapping in oxide semiconductor transistors is required for being core device element in reliable production lines. In this paper, we present the transient charging effect, the charge trapping mechanism, and the dynamic charge transport of high-mobility bilayer oxide semiconductor transistors. To this end, we exploited microsecond ramps, pulse ID–VG, transient current, and discharge current analysis methods. The mobility enhancement rate of single HfInZnO (HIZO) and bilayer HfInZnO-InZnO (HIZO-IZO) were 173.8 and 28.8%, respectively, in the charge-trapping-free environment. Transient charge trapping can be classified to temperature insensitive fast charging and thermally activated slow charging with two different trap energies. Insignificant fast transient charging of a bilayer-oxide high-mobility thin film transistor(TFT) can be explained by the low density of sub-gap states in the oxide semiconductor. Understanding defects and transient charging in the oxide semiconductor helps to determine the origin of device instability of oxide TFTs, and finally, to solve this problem.

The negative capacitance (NC) of ferroelectric materials has paved the way for achieving sub-60-mV/decade switching feature in complementary metal-oxide-semiconductor (CMOS) field-effect transistors, by simply inserting a ferroelectric thin layer in the gate stack. However, in order to utilize the ferroelectric capacitor (as a breakthrough technique to overcome the Boltzmann limit of the device using thermionic emission process), the thickness of the ferroelectric layer should be scaled down to sub-10-nm for ease of integration with conventional CMOS logic devices. In this paper, we demonstrate an NC fin-shaped field-effect transistor (FinFET) with a 6-nm-thick HfZrO ferroelectric capacitor. The performance parameters of NC FinFET such as on-/off-state currents and subthreshold slope are compared with those of the conventional FinFET. Furthermore, a repetitive and reliable steep switching feature of the NC FinFET at various drain voltages is demonstrated.