Search

As the market increases for Artificial Intelligence and High-Performance Computing applications, the geometry of 3-Dimensional Integrated Circuit packages becomes more complicated; therefore, predicting the thermal distributions of the structures becomes not only more important but also more challenging. The physics governing the thermal distribution is a 3-dimensional partial differential equation. In order to predict the thermal distributions, various approaches such as the layer modeling method have been invented. While practical, these approaches solve a simplified version of the differential equation placing an inherent limitation on their capabilities which may be improved upon. In this research we solve the actual differential equation using Finite Element Analysis. Finite Element Analysis is known to produce accurate solutions, albeit being computationally intensive, and may take days to run even with a powerful computer, making it impractical for some applications. In this paper, we present case studies of computing the thermal distributions of large scale 3-dimensional integrated circuit models using XSim. We find that XSim’s solving capabilities are capable of computing accurate thermal distributions of large-scale models in minutes instead of days. This suggests a new way to compute the thermal distributions rapidly and accurately for large 3-Dimensional Integrated Circuit packages, which may bring potential benefits in research, design and development of 3-Dimensional Integrated Circuit packages.

Line of Sight (LOS) deformation based on Differential Interferometric Synthetic Aperture Radar (DInSAR) techniques cannot be used in traditional probability integration method (PIM) parameter inversion. To improve the accuracy of parameter inversion, a model based on 3D deformation was proposed. The model simulates 3D deformation using PIM directly. The inverse of the Sum of the Squared Errors (SSE) of the PIM results and the measured deformation results was used as a fitting function within the GA. Reliable PIM parameters can be obtained based on this GA model. To identify the surface movement law of the Jinfeng coal mine, 6 Global Navigation Satellite System (GNSS) monitor points were established over the 011207 and 011809 working panels. Due to the limited number of points and the large distance between the points, it is not sufficient to obtain reliable PIM parameters using GNSS only. As a complement, 83 Sentinel-1A images were analyzed with small baseline subset (SBAS) DInSAR, and the LOS direction deformation was obtained. The reliable PIM parameters were calculated with the 3D inversion model based on the combination of LOS direction deformation and GNSS-monitored deformation. Then, those parameters were used to predict the coal mine deformation of panels 011207 and 011809, which demonstrated that the prediction results coincide with the measured results. The model can be used to study the laws of mining subsidence combined with DInSAR and GNSS, which can reduce the requirements of the number of GNSS points and the impact of radar decoherence. This provides a new technical approach for studying the law of surface movement in mining subsidence research.

In this paper, passive generation of the multi-wavelength parabolic pulses (PPs) in tapered silicon nanowires (TSNs) is numerically investigated. The coupled inhomogeneous nonlinear Schrödinger equation (INLSE) is derived in the context of TSN to model the multi-wavelength PP generation. The TSN is designed based on the well-known self-similar theory that the group-velocity dispersion is decreased while the nonlinear coefficient is unchanged along the propagative direction. While the pump wavelengths for three input pulses are specially set with equal intervals of 18, 12, and 6 nm, the time-domain profiles are aligned at the same position. Simulation results show that except for wavelength interval, the waveguide length is also a critical factor that influences the quality of generated PP. Both of them reshape the pulse profile by the means of walk-off effect which depends on the cross-phase modulation between multiple pulses. By properly optimizing the input parameters of Gaussian pulses as well as the center wavelength interval and TSN length, we prove that the generation of two- even three-wavelength PPs with high-quality is feasible.

In recent years, with the opening of the “smart age” curtain, smart devices dominated by technologies such as robots, drones, and intelligent perception have gradually moved to the center of the Intelligent CPSS stage. However, the new security risks of the Intelligent CPSS have also become increasingly prominent. Especially in recent years, in Ukraine and Venezuela's power attack incidents, a series of related attacks always occur simultaneously. This is a multi-task compound attack. This paper designs a set of Cloud-Fog-Edge closed-loop feedback security risk prediction strategies for multi-task compound attacks based on the offensive and defensive ideas of intelligent games, combining classified deep Boltzmann machines and Markov time-varying models. This strategy can be used for various types of power intelligent system terminals, and realizes security risk prediction with modularity, interoperability, open interfaces and compliance with open standards. Interoperability with other safety equipment can also be achieved through standardized interfaces to form system security protection capabilities to meet the actual needs of the industry Internet system. Experimental results show that the method is superior to the typical traditional method.

Supercontinuum (SC) with broad bandwidth and high coherence is crucial in the SC-based frequency comb source generation. In this paper, we numerically investigate the mid-infrared (MIR) SC generations with the three optical modes (TE00, TE10, and TE20) in a multimode chalcogenide (As2Se3) strip waveguide. The waveguide structure is carefully engineered to ensure that the pump pulses are propagated in the normal dispersion regions of the considered three optical modes. Highly coherent and octave-spanning MIR SCs are generated when the optimized pump pulse with 80-fs pulse duration, 3-kW peak power, and 3-μm center wavelength is used. Moreover, the nonlinear dynamics of the SC generation are numerically analyzed. Finally, the SC-based frequency combs are numerically demonstrated when a pulse train with a repetition rate of 50 MHz is used as the pump source and launched into the multimode As2Se3-based strip waveguide. It is believed that the generated MIR SC and SC-based frequency comb sources have important applications in biophotonics, metrology, and sensing.

A simple optic-fiber Fabry-Perot (FP) sensing technique was proposed and experimentally investigated by using polymer material to connect the ends of two singlemode fibers. Four different polymer materials (benzocyclobutene (BCB), UV88 (Relentless, China), Loctite3525 (HenKel, Germany), and NOA68 (Norland, USA)) filling the FP cavity were used to comparatively study the sensing performance of temperature, strain and refractive index. The result shows that the FP sensor with BCB has excellent repeatability with good linear response to temperature in a wide range from room temperature to 250 °C, which is much larger than that of other three materials (2)) and the best repeatability among the four polymer materials.

At present, low emissivity coating has been widely used in various fields, but damage will greatly reduce the efficiency of low emissivity coating, so the damage detection of low emissivity coating becomes an important work. Based on convolution neural network, a model for automatic identification of coating damage with low emissivity is proposed. Firstly, the optical image data set of low emissivity coating is constructed and extended by means of data enhancement. After that, VGG-19 and ResNet-50 models are built based on tensorflow, and the cross entropy loss function is used in the models. Then, SGD, momentum, RMSprop and Adam are used to optimize the model. In the process of model optimization, the learning rate is adjusted to get the optimal model. The results show that when the learning rate is $5\times 10^{-5}$ and Adam method is used to optimize the model, the recognition accuracy of VGG-19 model is 90.64%, while that of ResNet-50 model is 94.14%. This paper is of great significance for the study of automatic damage identification of low emissivity coatings.

An attitude controller with finite time convergence rate for satellite attitude tracking issue considering the flexible dynamic based on a finite time modal state observer is proposed in this article. The deformation of flexible appendages is taken into consideration and a finite-time state observer with auxiliary state to offset the initial error is proposed to get the flexible vibration estimation. The modal state and its derivative could be estimated by a full dimensional feedback observer. A finite time sliding mode based on the fraction order feedback of MRP is constructed and the finite time controller is proposed. The modal estimation is implied in controller design hence the counteraction of flexible system could be offset. System global finite-time stability is proved by Lyapunov method and the superiority is demonstrated by simulation results.

A microfiber interferometer based on a single-mode tapered-no-core single-mode (STNCS) fiber structure coated with a thin layer of graphene oxide (GO) was proposed and demonstrated as a relative humidity (RH) sensor. The STNCS fiber structure has a strong evanescent field, which can sensitively response to changes in the surrounding refractive index. The moisture-sensitive material GO was used to form the film on the no-core fiber by a dip impregnation method. The effects of GO concentration and the tapered waist diameter on RH sensitivity were studied experimentally. The experimental results showed that a thinner waist diameter provides higher RH sensitivity. When the concentration of GO solution is 0.01 mg/mL, the STNCS fiber structure with a waist diameter of 2.9 μm performs better in RH sensing; a high sensitivity of 0.461 nm/%RH in the range of 40-98% was achieved. The sensor has good stability with wavelength fluctuations of ±0.009 nm and ±0.008 nm over 60 mins at RHs of 50% and 80%, respectively. The proposed fiber optic humidity sensor has a high sensitivity over a wide measurement range and offers good stability showing promising potential application in the field of RH sensing.

Diffusion tensor imaging (DTI) is currently the unique imaging technique that can detect the structure of in-vivo human myocardium without invasivity and radiation. However, it is particularly sensitive to motions, especially respiratory motion that results in serious signal loss in diffusion-weighted (DW) images. This makes it impossible to accurately measure cardiac microscopic structural properties. To cope with such problem, this paper proposes an unsupervised dense-encoder-fusion-decoder network (DEFD-net) to compensate for signal loss in cardiac DW images, which allows investigating in-vivo myocardium structure more accurately. The DEFD-net consists of three modules, namely dense-encoder, fusion module and decoder module. The dense-encoder and decoder are trained firstly with DW images acquired at different trigger delays in an unsupervised manner for extracting local and global features. A fusion strategy is then designed to fuse the extracted features. Finally, the well-trained decoder is used to reconstruct the fused DW image from the fused features. To validate the superiority of the proposed method, comparison with existing methods such as PCAMIP, WIF and U2Fusion is performed on both simulated and acquired datasets. The experimental results showed that the proposed method effectively compensates for motion-induced signal loss in DW images, thus leading to much better DW image quality with respect to existing methods. Moreover, the subsequently derived myocardium fiber structure is more regular.

Advanced driver assistance systems (ADAS) based on monocular vision are rapidly becoming a popular research subject. In ADAS, inter-vehicle distance estimation from an in-car camera based on monocular vision is critical. At present, related methods based on a monocular vision for measuring the absolute distance of vehicles ahead experience accuracy problems in terms of the ranging result, which is low, and the deviation of the ranging result between different types of vehicles, which is large and easily affected by a change in the attitude angle. To improve the robustness of a distance estimation system, an improved method for estimating the distance of a monocular vision vehicle based on the detection and segmentation of the target vehicle is proposed in this paper to address the vehicle attitude angle problem. The angle regression model (ARN) is used to obtain the attitude angle information of the target vehicle. The dimension estimation network determines the actual dimensions of the target vehicle. Then, a 2D base vector geometric model is designed in accordance with the image analytic geometric principle to accurately recover the back area of the target vehicle. Lastly, area-distance modeling based on the principle of camera projection is performed to estimate distance. The experimental results on the real-world computer vision benchmark, KITTI, indicate that our approach achieves superior performance compared with other existing published methods for different types of vehicles (including front and sideway vehicles).

Precise control of a coupling effect is critical for space-limited communication system applications. Dielectric metamaterials, which derive their electromagnetic properties from subwavelength structures, have emerged as a promising way to tune coupling effect due to their complex resonant modes. However, it has not yet achieved quantitative control of coupling effects in metamaterial designs, which is important for advanced engineering applications such as artificial intelligence antenna and reconfigurable camouflage. In this paper, all dielectric left-handed metamaterial operating in THz band based on the deep coupling of higher mode Mie resonance is presented. Distinguished from those designs published before, this left-handed metamaterial is composed only by one same structure, not two separated parts providing negative permittivity and permeability, respectively. By utilizing different Mie resonant modes, negative permittivity and permeability are accomplished via high order Mie resonant modes and the coupling between cubes. This work opens a gate to highly control of electromagnetic wave in dielectric metamaterial by using the coupling between the unit cells.

A subspace tracking technique has drawn a lot of attentions due to its wide applications. The main objective of this approach is to estimate signal or noise subspace basis for the sample covariance matrix. In this paper, we focus on providing a fast, stable, and adaptive subspace tracking algorithm that is implemented with low computational complexity. An alternative realization of the fast approximate power iteration (FAPI) method, termed modified FAPI (MFAPI), is also presented. Rather than solving an inverse square root of a matrix employed in the FAPI, the MFAPI applies the matrix product directly to ensure the orthonormality of the subspace basis matrix at each recursion. This approach yields a simpler derivation and is numerically stable while maintaining a similar computational complexity as compared with that of the FAPI. Furthermore, we present a detailed mathematical proof of the numerical stability of our proposed algorithm. Computer simulation results indicate that the MFAPI outperforms many classical subspace tracking algorithms, particularly at the transient-state step.

Functional magnetic resonance imaging (fMRI) has increasingly come to dominate brain mapping research, as it provides a dynamic view of brain matter. Feature selection or extraction methods play an important role in the successful application of machine learning techniques to classifying fMRI data by appropriately reducing the dimensionality of the data. While whole-brain fMRI data contains large numbers of voxels, the curse of dimensionality problem may limit the feature selection/extraction and classification performance of traditional methods. In this paper, we propose a novel framework based on a tensor neural network (TensorNet) to extract the essential and discriminative features from the whole-brain fMRI data. The tensor train model was employed to construct a simple and shallow neural network and compress a large number of network weight parameters. The proposed framework can avoid the curse of dimensionality problem, and allow us to extract effective patterns from the whole-brain fMRI data. Furthermore, it reveals a new perspective for analyzing complex fMRI data with a large numbers of voxels, through compressing the number of parameters in a neural network. Experimental results confirmed that our proposed classification framework based on TensorNet outperforms traditional methods based on an SVM classifier for multi-class fMRI data.

This paper develops an adaptive radical basis function neural-network (NN)-based controller design strategy that uses integral Lyapunov functions for a class of non-strict-feedback nonlinear systems subject to perturbations. The design difficulty caused by the non-strict-feedback system structure is handled by using the inherent property of the square of neural network's base vector. The design procedure of the adaptive NN tracking controller is presented by using backstepping technique, which can update the adaptive laws at any time and solve the design problem derived from the correlation degree of the controlled plant. The uniform ultimate boundedness and good tracking performance of the derived closed-loop system are ensured with the design controller. Finally, a comparative simulation example is carried out to prove the effectiveness of the proposed control method.