Search

Fine pitch Cu/SiO2 hybrid bonding has been successfully demonstrated at a low temperature of 120 °C, a breakthrough, using Au passivation method in this work. To explore the bonding mechanism of passivation structures for hybrid bonding in details, Cu-Cu direct bonding with Au passivation on both wafer-level and chip-level has been discussed, including analyses of AFM, SAT, TEM, electrical measurements, and reliability test. Cu/SiO2 hybrid bonding with the fine pitch structure with stable electrical performance can be achieved at low bonding temperature under an atmospheric environment. Accordingly, this Au passivation scheme for Cu/SiO2 hybrid bonding with excellent bonding quality, low thermal budget, and high reliability shows a great feasibility for the 3D IC and heterogenous integration applications.

Objective: In this research, we propose a human motion recognition method based on multi-feature fusion of millimeter Wave(mmWave) radar, which is implemented by CentralNet. Aiming at solving the problems that the method based on camera is susceptible to light and weather, and the flexibility of wearable devices is poor. Methods: Firstly, Radar data of moving human body is collected by millimeter-wave radar, and then we obtain Micro-Doppler Spectrogram(MDS) and Candence-Velocity Diagram(CVD) through time-frequency analysis of received signal. Next, the MDS and CVD were respectively put into the subnetwork of CentralNet which is a neural network designed to achieve feature fusion. Finally, we can implement human motion recognition through CentralNet. Result: The test data set verified that the proposed method could achieve classification of five kinds of human motions with accuracy of 98.96%. Conclusion: MDS and CVD are suited for extracting time-frequency features of signal. And CentralNet can effectively fuse multiple features, which leads to well classification performance. Significance: Compared to camera and wearable devices, human motion recognition based on radar has shown advantages in terms of accuracy, privacy, and robustness.

The main development challenge of low temperature hybrid bonding technology is the oxidation of Cu surface. The schemes to achieve the high-quality surface condition of Cu before bonding is urgently demanded. In this study, four acid solutions, including sulfuric acid (H 2 SO 4 ), hydrochloric acid (HCl), acetic acid (CH 3 COOH) and citric acid (C 6 H 8 O 7 ), were chosen as wet pretreatment approaches to remove Cu oxides based on the respective chemical reactions. To investigate the influence of these acid solutions on bonding surface, different acid immersion times are examined and the XPS analysis is conducted to check the surface component composition. Next, thermal compression bonding (TCB) is conducted after the Cu oxide removal process, followed by various analyses, including SAT analysis to validate bonding results, FIB and SEM analyses to inspect bonding interface, and pull test to measure bonding strength. The results of analyses verify the ability of different wet pretreatment to improve Cu-Cu bonding.

In this study, using metal passivation (Au or Pd), chip-level Cu to Cu bonding with $15\times 15\ \mu\mathrm{m}^{2}$ bump size and 15 µm bump pitch has been achieved at 180°C for 30 seconds under the atmosphere. The good bonding results have been carefully validated by analyzing SEM, TEM, EDS, and shear test. Despite the small bonding pad and fine pitch, the high bonding strength (27.9 MPa) can be still obtained, showing the high feasibility of this bonding structure for chiplet heterogeneous integration.

In this study, electroplated Cu-Cu bonding with electroless plating passivation on Cu surface can be achieved at a low bonding temperature. Compared with other low temperature bonding schemes, electroplated Cu and electroless plated passivation in this work is time-saving, high throughput, and high yield. After a pretreatment process, chip-level bonding with the good bonding quality can be achieved at 180°C under atmosphere. The morphology of the electroless plated passivation surface was characterized by the energy dispersive X-ray spectroscopy (EDX), and the bonding quality was investigated by the analyses of scanning electron microscope (SEM). In addition, electrical measurements were performed to evaluate the electrical properties of the bonding structure.

In this work, a novel biocompatible flexible heterogeneous integration platform has been developed. Using thermal compression bonding and conductive adhesive, three different functional devices, including sensing array, analog front-end chip, and LED array, are integrated on the flexible biocompatible substrate. Moreover, the whole exposed area of the platform is designed to be processed with biocompatible material. Thus, this structure is a promising integration scheme for the biomedical applications. Bending tests and reliability tests have been also performed to validate the feasibility of this flexible heterogeneous integration platform.

This paper concerns the stability analysis of discrete-time systems(DTS) with a time-varying delay(TVD). Firstly, by adding double summation terms, an appropraite Lyapunov-Krasovskii functional (LKF) is constructed. Then, the generalized reciprocally convex matrix inequality(GRCMI), together with the Jensen-based summation inequality(JBI) and the Wirtinger-based summation inequality(WBI), is used to estimate the forward difference of the LKF. Finally, a new delay-dependent stability criterion is established. The stability criterion show lower conservatism and require lower computational burden than results previously proposed, whose advantages are demonstrated via a numerical instance.

Research has shown the potential of device-free WiFi sensing for human identification. Each and every human has a unique gait and prior works suggest WiFi devices are able to capture the unique signature of a person’s gait. In this article, we show for the first time that the monitored gait could be inconsistent and have mirror-like perturbations when individuals walk through WiFi devices in different directions, provided that the WiFi antenna array is horizontal to the walking path. Such inconsistent mirrored patterns are to negatively affect the uniqueness of gait and accuracy of human identification. Therefore, we propose a system called Gate-ID for accurately identifying individuals’ identities irrespective of different walking directions. Gate-ID employs theoretical communication model and real measurements to demonstrate that antenna array orientations and walking directions contribute to the mirror-like patterns in WiFi signals. A novel heuristic algorithm is proposed to infer individual’s walking directions. A set of methods are employed to extract and augment the representative spatial–temporal features of gait and enable the system performing irrespective of walking directions. We further propose a novel attention-based deep learning model that fuses various weighted features and ignores ineffective noises to uniquely identify individuals. We implement Gate-ID on commercial off-the-shelf devices. Extensive experiments demonstrate that our system can uniquely identify people with average accuracy of 90.7%–75.7% from a group of 6–20 people, respectively, and improve the accuracy by 12.5%–43.5% compared with baselines.

Computing-In-Memory (CIM) using Flash memory is a potential solution to support a heavy-weight DNN inference accelerator for edge computing applications. Flash memory provides the best high-density and low-cost non-volatile memory solution to store the weights, while CIM functions of Flash memory can compute AI neural network calculations inside the memory chip. Our analysis indicates that Flash CIM can save data movements by ~85% as compared with the conventional Von-Neumann architecture. In this work, we propose a detail device and design co-optimizations to realize Flash CIM, using a novel vertical split-gate Flash device. Our device supports low-voltage (

In this study, an asymmetric Cu-polymer hybrid bonding technology was successfully developed at 200 °C, which uses SU-8 for polymer bonding and Au passivation for Cu bonding. The polymer bonding performance was investigated by the analyses of sat, SEM and pulling test, indicating that bonding with the best quality was achieved at 200 °C. With the Au passivation layer capping on the Cu surface, a low temperature Cu-Cu bonding process can be realized at 200 °C. After demonstrating the bonding process with single material, a chip-level hybrid bonding structure was fabricated and bonded successfully. The hybrid bonding interface was observed by the SEM images, showing that both of the metal and polymer region were well bonded. Furthermore, the electrical properties and reliability performance were analyzed by electrical measurements with a modified Kelvin structure and after un-biased highly accelerated stress test for 168 hours. The bonded structure can still maintain excellent electrical property about 10−7 Ω•cm2. Based on the bonding results, this research provides a practical concept for low temperature hybrid bonding in 3D integration.

With the positive effect of big data on promoting education reform and innovation, student management education in universities moves on to the age of information and intelligence. Collecting, integrating and analyzing data from "Intelligent Student management education Platform" of University of Electronic Science and Technology of China (UESTC), this paper first figures out the association among students’ behaviors, then discusses the applications of big data analysis via four cases in practice. Our paper shows that big data analysis can not only help teachers keep informed about students’ daily routine, learning habits and expense records duly and thoroughly, but also promote our work towards operating more dynamically, warning in advance, subsidizing students more accurately, and teaching students more individualizedlly.

We present an in-memory-searching (IMS) architecture which features ultra-high parallel operation. It is based-on high-density, low-power, and low-cost 3D-NAND technology. The IMS unit cell composition, operation algorithm, and array structures are discussed in detail. Reliability concerns including program/read disturbance and retention issues are evaluated and solutions provided. The proposed novel coding scheme can not only provide flexibility to the searching function but also help to improve reliability. An application example on machine learning with k-NN model is demonstrated with good results. The IMS searching efficiency can evolve along with the 3D-NAND technology and is promising to play a key role in the future memory-centric computing systems.

The analog neuromorphic circuits with functional memory devices are considered as an ultimate ideal approach to mimic the human brain for artificial intelligence (AI). The spiking neural network (SNN) with integrate-and-fire (IF) circuit is the classic building block theoretically, but so far it is very difficult to find ideal devices to realize the SNN circuit. In this work, we propose a novel functional memory that is enabled by a novel thyristor, which features super-steep slope (S.S.~0.015mV/dec), large ON/OFF ratio (> 5 orders), and tunable V th range (0~3V). These are very ideal to meet the IF circuit requirements. Circuit and network simulations indicate that the gate-controlled thyristor (GCT) device for the IF circuit can realize high accuracy and performance for image recognition SNN. Our novel SNN architecture with the GCT device can provide good energy efficiency (equivalent to 181TOPS/W for accumulation operations), good error tolerance to V th variations (~10% range), and substantially smaller circuit area than that using conventional CMOS devices for IF circuit.

Most traditional recommendation algorithms rely on common scoring items between users when making product recommendations. The data is highly sparse and the recommendation effect is not good. Therefore, an improved collaborative filtering recom-mendation algorithm is proposed. Based on user purchase rec-ords, the algorithm uses a representation learning method to construct a user product network, obtains the low-dimensional em-bedded semantic relationship between users and product no-des, and uses cosine similarity to measure the semantic similarity between products. Then, according to the hidden Dirichlet topic distribution model, the topic features of the products are obtained, and the cosine similarity is used to calculate the similarity of the topic features between the products. The linear fusion method is adopted to effectively alleviate the data sparse problem and improve the recommendation performance. Through Amazon product reviews Data set to verify the effectiveness of the recommendation algorithm.

Bonding technology is the key to the success of vertical stacking in 3D integration, heterogeneous integration, and advanced packaging applications. Furthermore, developing a bonding technology which can be performed at low temperature becomes the top priority to avoid the concerns of low thermal budget, low mechanical stress, and high reliability in many applications. In addition, for material to be used as the bonding medium, Cu has always been the best candidate with its excellent electrical performances, thermal properties, and readiness in the backend and packaging fabrication, as well as cost consideration. This talk will describe the progress and advance of current Cu-based direct bonding, based on parameter optimization, structure design, and material selection. Technology development of low temperature Cu-Cu bonding and corresponding solutions will be described. Finally, Cu-Cu hybrid bonding, an enabling technology for fine pitch/size bonding in multiple 3D integration and advanced packaging applications, will be introduced and further discussed.

Nowadays, in the era of big data, consumers generate tens of thousands of data on e-commerce platforms every day. How to use the massive hidden feedback information generated by network consumers on e-commerce platforms to predict consumers’ purchasing behavior, thus improving the purchasing conversion rate of e-commerce enterprises is of great significance to enterprises. Therefore, aiming at the prediction of online consumer purchase behavior, this paper will propose a LSTM-RF prediction model based on the combination of Long Short-Term Memory (LSTM) and Random Forest (RF) by analyzing the consumer history shopping data of the e-commerce platform, so as to improve the learning ability of the prediction model and the prediction effect of the model. The experimental results show that from the overall evaluation index of the model, compared to the use of a single machine learning method to create the prediction model, the LSTM-RF prediction model combined with the LSTM algorithm and the RF algorithm has better learning ability and prediction effect.

Accurate merchandise sales forecasts increase the turnover rate for merchants’ benign inventory stocking, formulate cost budget items, formulate effective strategies according to the time and local conditions, and provide theoretical basis for reasonable merchandise sales strategies. However, the traditional methods of forecasting are not comprehensive and the actual effect is not ideal. Use the ARIMA model to fit the original product sales time series, extract the main deterministic information in the product sales time series through the deterministic factor decomposition method, and subtract or remove the trend effect fitting polynomial and seasonal effect fitting from the sales time series The remaining part of the polynomial obtains the residual sequence. The similarity distance function of the comparison layer in the adaptive resonance theory neural network is improved to measure the degree of similarity between a subsequence of the residual sequence and another subsequence, and the weight is adjusted according to the similarity. Each neuron in the competition layer of the network compares the input subsequences accordingly to generate the most similar subsequences of each subsequence, and iteratively eliminates all sequences related to the total residual sequence to obtain the final residual sequence that meets the requirements. Experimental comparison shows that the model after optimization by residual error can reflect the sales law more accurately and comprehensively, and improve the accuracy of product sales forecast.

This paper is concerned with the stability analysis of discrete time-delay system. Firstly, an improved augmented functional form is proposed and the positive definite condition of functional is derived. Then, the forward difference of functional is estimated by applying summation inequalities and a state-connecting-based zero-value equation. As a result, an improved stability criterion is established. Finally, a numerical example is given to show the efficiency and merit of the proposed method.

Switchable glass, which can electronically change its state between opaque and transparent, is finding applications in modern buildings as a replacement for traditional curtains or blinds to enable dynamic privacy control in physical spaces. In this work, we propose to exploit the electronic configurability of switchable glass to modulate natural light, which can be demodulated by a nearby receiver with the light-sensing capability to realise data communication over natural light. We call this natural light communication (NLC). A key advantage of NLC is that no energy is used to generate light, as it simply modulates the existing light in nature. The slow response rate of the switchable glass, however, limits the achievable data rates for NLC. We propose and optimise an ON-OFF modulation/demodulation technique for such slow-response switchable glass and implement it using off-the-shelf Polymer Dispersed Liquid Crystal (PDLC) glass. We demonstrate that the proposed NLC can achieve 33.33 bits per second communication with a bit error rate below 1% under a wide range of ambient luminance.

We address the problem of human posture detection with casual loose-fitting smart garments by fabricating a new type of highly sensitive, stretchable, optical transparent and low-cost strain sensor enabled by uniquely designed microcracks within a hybrid conductive thin film. In terms of sensitivity and stretchability, the developed sensor outperformed most of the works reported in recent literature, and has a gauge factor of 103 at the high strain of 58%. By attaching these sensors to an off-the-self casual jacket, we implement E-Jacket, a smart loose-fitting sensing garment prototype. To detect postures from sensor data, we implement a conventional deep learning model, CNN-LSTM, capable of overcoming the noise induced by the loose-fitting of the sensors to the human skin. To evaluate E-Jacket, we conducted three case studies in experimental environments: recognition of daily activities, recognition of stationary postures with random hand movements, and slouch detection. Our evaluation results demonstrate the feasibility of the proposed E-Jacket smart garment system for different posture recognition applications.

Each year, hundreds of thousands of people contract Healthcare Associated Infections (HAI). Poor hand hygiene compliance among healthcare workers is thought to be the leading cause of HAIs and methods were developed to measure compliance. Surprisingly, human observation is still considered the gold standard for measuring compliance by World Health Organization (WHO). Moreover, no automated solutions exist for monitoring hand hygiene techniques, such as "how to hand rub" technique by WHO. In this work, we introduce RFWash; the first radio-based device-free system for monitoring Hand Hygiene (HH) technique. On the technical level, HH gestures are performed back-to-back in a continuous sequence and pose a significant challenge to conventional two-stage gesture detection and recognition approaches. We propose a deep model that can be trained on unsegmented naturally-performed HH gesture sequences. RFWash evaluation demonstrates promising results for tracking HH gestures, achieving gesture error rate of < 8% when trained on 10-second segments, which reduces manual labelling overhead by ≈ 67% compared to fully supervised approach. The work is a step towards practical RF sensing that can reliably operate inside future healthcare facilities.

We demonstrate a human activity detection with casual loose-fitting smart garment system. By employing a new type of highly sensitive, stretchable, optical transparent and low-cost strain sensor and a deep learning model enabled by CNN-LSTM, the loose-fitting jacket is able to recognize 5 activities with 90.9% accuracy when the system is trained with the user data, and 73.5% accuracy when an unseen user wears the smart jacket, which is comparable with tight-fitting smart garment system. In the demonstration, we will showcase activity recognition of three activities: walk, sit, and stand.

Face recognition (FR) has become increasingly important in many real-world IoT applications like public safety surveillance camera system or CCTV, person re-identification system and face-based authentication system. The privacy of face images has been a growing concern, not only in the collected face dataset stored in the cloud platforms but also in its everyday use. However, most existing schemes (e.g., deep learning with differential privacy [1]) build privacy-preserving analytics models from the stored face data while ignoring the privacy concern in end devices. In this paper, we propose a novel efficient privacy-preserving face representation scheme in the Bloom filter space, which can satisfy the resource limits from IoT devices. Our solution allows analytics tasks on privacy-preserving face data representation but retains the high data utility on analytics (e.g., classification).