Compact Model Research Articles

Environmental sound classification on an embedded hardware platform

Convolutional neural networks (CNNs) have exhibited state-of-the-art performance in various audio classification tasks. However, their real-time deployment remains a challenge on resource constrained devices such as embedded systems. In this paper, we analyze how the performance of large-scale pre-trained audio neural networks designed for audio pattern recognition changes when deployed on a hardware such as a Raspberry Pi. We empirically study the role of CPU temperature, microphone quality and audio signal volume on performance. Our experiments reveal that the continuous CPU usage results in an increased temperature that can trigger an automated slowdown mechanism in the Raspberry Pi, impacting inference latency. The quality of a microphone, specifically with affordable devices such as the Google AIY Voice Kit, and audio signal volume, all affect the system performance. In the course of our investigation, we encounter substantial complications linked to library compatibility and the unique processor architecture requirements of the Raspberry Pi, making the process less straightforward compared to conventional computers (PCs). Our observations, while presenting challenges, pave the way for future researchers to develop more compact machine learning models, design heat-dissipative hardware, and select appropriate microphones when AI models are deployed for real-time applications on edge devices.

Deep learning-based construction and demolition plastic waste classification by resin type using RGB images

The construction and demolition sector generates a substantial portion of Australia's total waste, with plastics being a key recyclable component. The perceived financial impracticality of sorting and separating waste, coupled with the simplicity of landfilling processes often contribute to mixed material loads sent directly to landfills. Therefore, developing a commercially feasible system that can accurately separate the generated waste is imperative. This paper presents a comprehensive study on using RGB images for deep learning-based construction and demolition plastic waste classification by resin type. A large and specialised dataset of end-of-life plastic waste images is gathered. This dataset comprises four commonly used plastic types in construction projects—ABS, HDPE, PS, and PVC. Leveraging Transfer Learning with models pre-trained on ImageNet, highly accurate models tailored to this classification task are developed in this paper. Advanced Convolutional Neural Network and Vision Transformer-based models, including ResNet, ResNeXt, RegNet, and Swin Transformer, are trained and evaluated on this dataset. Another contribution of this work is Knowledge Distillation from a large, computationally intensive, and accurate model to enhance the accuracy of fast and compact models specifically designed for deployment on edge devices. This study applies Knowledge Distillation by using the output class probabilities of the large, computationally intensive Swin Transformer model to enhance the accuracy of the fast and lightweight MobileNetV3 models. The results demonstrate that RGB images offer a practical alternative to other costly and complex systems for effective plastic identification, due to their availability, low cost, ease of use, simple setups, and robustness to variations in operational conditions.

Study of the thermal measurement depth of a skin calorimeter using simple RC and TF models

An experimental and theoretical study of the heat capacity thermal measurement depth of a skin calorimeter has been carried out. This calorimeter consists of a thermopile placed between a thermostat and an aluminum measuring plate, which is placed on the sample to be measured. We performed simulations using the finite element method (FEM) with inert samples of different sizes. From these simulations, Transfer Functions (TFs) between the calorimeter's thermostat temperature and the temperature of the measuring plate were determined. Compact thermal models (RC models) able to determine the heat capacity of the sample were also determined. We conclude that, in the case of heat-conducting materials, a single time constant is enough to represent the TF. However, at least two time constants are required for heat-insulating materials. In this work we also studied the time dependence of the thermal measurement depth, concluding that this dependence is exponential. Finally, we present some experimental measurements performed on inert samples and on human skin, which are coherent with the results of the simulation.

Nonvolatile logic gate and full adder based on tri-terminal oxide resistive switching devices

Today's on-chip computing power is constrained by the “memory wall” and “power wall” caused by the Von Neumann bottleneck. As a potential solution, this work has developed nonvolatile logic gates based on field-effect tri-terminal oxide resistive switching memory devices (3T-RRAM). A compact circuit model using a polynomial control source (PCS) is proposed to describe the behavior of the fabricated 3T-RRAM. The 3T-RRAM can be regarded as a nonvolatile transmission gate for constructing nonvolatile logic gates. Additionally, a full adder with input storage functionality has been designed using only eight 3T-RRAMs (four nonvolatile logic gates), and a binarized neural network (BNN) based on 3T-RRAM logic gate arrays has been proposed. This demonstrates the great potential of nonvolatile logic gates in computing-in-memory applications.

An adaptive learning strategy for surrogate modeling of high-dimensional functions - Application to unsteady hypersonic flows in chemical nonequilibrium

Many engineering applications rely on the evaluation of expensive, non-linear high-dimensional functions. In this paper, we propose the RONAALP algorithm (Reduced Order Nonlinear Approximation with Adaptive Learning Procedure) to incrementally learn, on-the-fly as the application progresses, a fast and accurate reduced-order surrogate model of a target function. First, a combination of nonlinear auto-encoder, community clustering, and radial basis function networks allows us to learn an efficient and compact surrogate model with limited training data. Secondly, an active learning procedure overcomes any extrapolation issues during the online stage by adapting the surrogate model with high-fidelity evaluations that fall outside its current validity range. This approach results in generalizable, fast, and accurate reduced-order models of high-dimensional functions. The method is demonstrated on three direct numerical simulations of hypersonic flows in chemical nonequilibrium. Accurate simulations of these flows rely on detailed thermochemical gas models that dramatically increase the cost of such calculations. Using RONAALP to learn a reduced-order thermodynamic model surrogate on-the-fly, the cost of such simulations was reduced by up to 75% while maintaining an error of less than 10% on relevant quantities of interest.

Applications of knowledge distillation in remote sensing: A survey

With the ever-growing complexity of models in the field of remote sensing (RS), there is an increasing demand for solutions that balance model accuracy with computational efficiency. Knowledge distillation (KD) has emerged as a powerful tool to meet this need, enabling the transfer of knowledge from large, complex models to smaller, more efficient ones without significant loss in performance. This review article provides an extensive examination of KD and its innovative applications in RS. KD, a technique developed to transfer knowledge from a complex, often cumbersome model (teacher) to a more compact and efficient model (student), has seen significant evolution and application across various domains. Initially, we introduce the fundamental concepts and historical progression of KD methods. The advantages of employing KD are highlighted, particularly in terms of model compression, enhanced computational efficiency, and improved performance, which are pivotal for practical deployments in RS scenarios. The article provides a comprehensive taxonomy of KD techniques, where each category is critically analyzed to demonstrate the breadth and depth of the alternative options, and illustrates specific case studies that showcase the practical implementation of KD methods in RS tasks, such as instance segmentation and object detection. Further, the review discusses the challenges and limitations of KD in RS, including practical constraints and prospective future directions, providing a comprehensive overview for researchers and practitioners in the field of RS. Through this organization, the paper not only elucidates the current state of research in KD but also sets the stage for future research opportunities, thereby contributing significantly to both academic research and real-world applications.

TCAD‐enabled machine learning framework for DC and RF performance evaluation of InGaAs sub‐channel DG‐HEMTs

AbstractThis research presents a machine learning (ML)‐based model that determines the DC and RF characteristics of InGaAs sub‐channel double gate high electron mobility transistors (DG‐HEMTs) to optimize the device structure. We employ technology computer‐aided design (TCAD) simulations to analyze the DC and RF performance of InGaAs sub‐channel DG‐HEMTs, generating a range of datasets by varying the material composition, layer width, and thickness of different layers in the device structure. We then train and optimize support vector regression (SVR) models using 5‐fold cross‐validation, varying the kernel function and degree parameters, and achieve better performance with the radial basis function (RBF) kernel. The simulated results indicate that the ML model predicts physical parameters more effectively than experimental analysis, offering a compact modeling solution that requires fewer computing resources than traditional methods.

Fabrication and Compact Modeling of Low-Voltage Flexible Organic TFT Using Self-Assembly of Conductive Polymer Channel Over High-k PMMA/SrZrOₓ Dielectric

Fabrication and Compact Modeling of Low-Voltage Flexible Organic TFT Using Self-Assembly of Conductive Polymer Channel Over High-<i>k</i> PMMA/SrZrO<i>ₓ</i> Dielectric

Compact Modeling of the Pulse Modulation of the Resistive Switching Memory for Closed-Loop Multibit Write Operation

Compact Modeling of the Pulse Modulation of the Resistive Switching Memory for Closed-Loop Multibit Write Operation

Application of artificial intelligence in the new generation of underwater humanoid welding robots: a review

Underwater welding robots play a crucial role in addressing challenges such as low efficiency, suboptimal performance, and high risks associated with underwater welding operations. These robots face a dual challenge encompassing both hardware deployment and software algorithms. Recent years have seen significant interest in humanoid robots and artificial intelligence (AI) technologies, which hold promise as breakthrough solutions for advancing underwater welding capabilities. Firstly, this review delves into the hardware platforms envisioned for future underwater humanoid welding robots (UHWR), encompassing both underwater apparatus and terrestrial support equipment. Secondly, it provides an extensive overview of AI applications in underwater welding scenarios, particularly focusing on their implementation in UHWR. This includes detailed discussions on multi-sensor calibration, vision-based three-dimensional (3D) reconstruction, extraction of weld features, decision-making for weld repairs, robot trajectory planning, and motion planning for dual-arm robots. Through comparative analysis within the text, it becomes evident that AI significantly enhances capabilities such as underwater multi-sensor calibration, vision-based 3D reconstruction, and weld feature extraction. Moreover, AI shows substantial potential in tasks like underwater image enhancement, decision-making processes, robot trajectory planning, and dual-arm robot motion planning. Looking ahead, the development trajectory for AI in UHWR emphasizes multifunctional models, edge computing in compact models, and advanced decision-making technologies in expansive models.

경차 브랜드의 헤드라이트에 적용된 디자인 조형성이 소비자 만족도 및 구전의도에 미치는 영향

Design is an important factor in the compact car market beyond economic feasibility and practicality, especially headlights that determine the first impression of the vehicle and greatly influence the brand image. The purpose of this study is to analyze the effect of design formativeness applied to the headlights of compact car brands on consumer satisfaction and word-of-mouth intention. To this end, case studies and empirical studies were conducted on major domestic compact car models, and structural equation model were used for the data collected through the survey. As a result of the study, it was found that formative factors such as form, size, and space of compact car headlights had a positive effect on consumer satisfaction, and this satisfaction led to positive word-of-mouth intentions of consumers. On the other hand, the material factor had no significant effect. This result suggests that compact car manufacturers should emphasize visual elements and move toward improving the consumer experience when developing design strategies. In addition, it was confirmed that there is a need to strengthen strategies to promote positive word-of-mouth by efficiently allocating design resources. In conclusion, in order to secure competitiveness in the compact vehicle market, a strategic approach that considers design originality and technical efficiency at the same time is needed, and it will be important to consider customized marketing strategies that consider the characteristics and strengths of each vehicle.

Modeling the Influence of Thermal Phenomena in Inductors and Capacitors on the Characteristics of the SEPIC Converter

The paper presents the results of measurements and calculations of the SEPIC converter characteristics, taking into account thermal phenomena in semiconductor devices and passive elements. Compact electrothermal models of the MOSFET transistor, diode, capacitor, and inductor are proposed. Parasitic phenomena are also included in these models. The form of the developed models and the method of determining the values of their parameters are presented. The correctness of the formulated models was verified experimentally. Calculations and measurements of the characteristics of SEPIC converters containing inductors with ferromagnetic cores made of different materials were carried out. The obtained results of the investigations are discussed, and the range of applicability of the formulated models is described. It was shown that, at the considered operating conditions at an ambient temperature equal to 22 °C, the temperature of capacitors can exceed 40 °C, whereas the temperatures of inductors can even reach 50 °C.

IGZO TFT Backplane Integration for µLED Flat-Panel Display

The focus of this work is the process integration of Indium Gallium Zinc Oxide (IGZO) transistors as a µLED backplane for row/column addressing. A single pixel is composed of a µLED driven by an arrangement of two transistors and a storage capacitor. The pixels are then arrayed on a glass substrate to support active-matrix control of monochrome and full color (RGB) displays from 1x1cm (50 x 50 pixels) up to 7.6 x 7.6 cm (380 x 380 pixels). Optimization of circuit parameters considering size and scan frequency was modeled using existing TFT and µLED electrical device compact models. New process parameters and procedures were determined for the proper integration of µLEDs in an existing TFT fabrication process. Red, green, and blue µLED devices were fabricated at Tyndall National Institute on native substrates and were transferred to TFT pixels using an X-Display MTP-1003 micro-transfer printer. The performance of individual test cells was assessed using an Agilent B1500, revealing a voltage transfer characteristic indicating the ability to modulate and control µLED current.

Improved Nonlinear Self-Heating Compact Modeling in SiGe HBTs Using Mextram

The linear and nonlinear SH models in the latest versions of Mextram are evaluated, with an emphasis on high-power biases and ambient temperature scalability. The Mextram nonlinear SH model is shown to have limitations in fitting high-power regions at all temperatures. An investigation of the dynamic thermal resistance rTH reveals that it is only a function of Tjunc, which limits the fitting capability of the current nonlinear SH model. A new nonlinear SH model is proposed based on analysis of the rTH characteristics to improve high-power region fitting at multiple ambient temperatures. The model is demonstrated on a SiGe HBT from −25◦C to 125◦C.

Detecting emerald ash borer boring vibrations using an encoder-decoder and improved DenseNet model.

Forest ecosystems are under constant threat from wood-boring pests such as the Emerald ash borer (EAB), which remain elusive owing to their hidden life cycles within tree trunks. Early detection is vital to mitigate economic and ecological damage. The main current monitoring method is manual detection which is ineffective at early stages of infestation. This study introduces VibroEABNet, a deep learning-based joint recognition network designed to enhance the detection of EAB boring vibration signals, with a novel approach integrating denoising and recognition modules. The proposed VibroEABNet model demonstrated exceptional performance, achieving an average accuracy of 98.98% across multiple signal-to-noise ratios (SNRs) in test datasets and a remarkable 97.5% accuracy in real forest datasets, surpassing traditional models and other deep learning networks evaluated in this study. These findings were supported by rigorous noise resistance analysis and real dataset evaluation, indicating the model's robustness and reliability in practical applications. Furthermore, the model's efficiency was highlighted by its inference time of 26 ms and a compact model size of 8.43 MB, underscoring its suitability for deployment in resource-limited environments. The development of VibroEABNet marks a significant advancement in pest detection methodologies, offering a scalable, accurate and efficient solution for early monitoring of wood-boring pests. The integration of a denoising module within the network structure addresses the challenge of environmental noise, one of the primary limitations in acoustic monitoring of pests. Currently, this research is limited to a specific pest. Future work will focus on the applicability of this network to other wood-boring pests. © 2024 Society of Chemical Industry.

Extended spherical harmonic expansion using multipolar bases of equivalent magnetic sources on arbitrary surfaces

Purpose The purpose of this paper is to propose a compact model to represent the magnetic field outside the sources. This model provides the multipolar ordering of a spherical harmonic expansion far from the source while being valid in its close proximity. Design/methodology/approach The authors investigate equivalent surface sources that enable to compute the field very close to any chosen surface that encloses the source. Then the authors present a method to find an appropriate initial basis and its associated inner product that allow to construct multipolar harmonic bases for these equivalent sources, where any vector of order k produces a field that decreases at least as fast as the field produced by a multipole of order k. Finally, those bases are numerically implemented to demonstrate their performances, both far from the source and in its close proximity. Findings The charge distribution and normal dipole distribution are well-suited to construct multipolar harmonic bases of equivalent sources. These bases can be described by as few parameters as the decreasing spherical harmonic expansion. Comparison with other numerical models shows its ability to compute the field both far from the source and close to it. Originality/value A basis for normal dipole distribution has already been described in the literature. This paper presents a general method to construct a multipolar basis for equivalent sources and uses it to construct a basis for single-layer potential.

String-level compact modeling of erase operations in the body-floated vertical channel of 3D charge trapping flash memory

In this study, we examined the use of string-level compact modeling as an effective framework for circuit simulation focusing on erase operation in 3D charge trapping flash (CTF) memory devices. We analyzed the behaviors of the accumulated hole from p-type bulk (p-well) and the corresponding difference of channel electrostatic potential in the CTF cell string, which is attributed to the variation in hole barrier height in the channel of the ground select line (GSL) transistor during erase operation. We derived a formula for the hole current delivering positive potential from the p-well to the channel region and established a modeling procedure. Technology computer-aided design (TCAD) simulation results were used to extract model parameters and analyze channel electrostatic potential during the erase operation. Additionally, experimental data for erase speed were verified using simulation program with integrated circuit emphasis (SPICE) results. Because the erase efficiency is strongly related to hole behaviors based on the conditions of the GSL transistor, the proposed compact modeling is an effective tool for circuit designers and system architects to achieve better performance in erase execution and optimizing the design of 3D CTF memory devices.

Anisotropic Durgapal-Fuloria compact stars in f(R) gravity

This study presented a new exact solution for anisotropic compact stellar objects within the framework of f(R)=R+αR2 gravity. In this context, the Durgapal-Fuloria metric potential has been employed to solve the field equation derived for f(R) theory. Furthermore, we have derived the generalized Darmois-Israel junction condition necessary for seamlessly connecting the interior region to the Schwarzschild exterior metric across the boundary hypersurface of the star in the context of f(R) gravity, and the interior solution is matched with the Schwarzschild exterior metric over the bounding surface of a compact star. These junction conditions stipulate that the pressure must not be zero at the boundary and should be proportional to the non-linear terms of f(R) gravity, a crucial aspect often overlooked by many researchers when investigating compact stellar models. Additionally, we derived the values of these parameters by using observational data of various compact stars (CSs), namely Her X-1, SAX J1808.4-3658, SMC X-1, LMC X-4, Cen X-3, 4U 1820-30, PSR J1903+327, 4U 1608-52, Vela X-1, and PSR J1416-2230. This approach enables us to investigate the comprehensive analysis of solutions numerically and graphically. We conducted various physical tests, including gradient of energy density and pressures, anisotropy, stability, equilibrium conditions, energy-density constraints, mass function, compactness, redshift, and adiabatic index, to assess the feasibility of our models. Our findings demonstrate the consistent behavior of our models provides a satisfactory physical situation as far as the observational results are confirmed.

Multilevel attention imitation knowledge distillation for RGB-thermal transmission line detection

Transmission line detection (TLD) plays a crucial role in ensuring the safety and stability of electricity supply. Applying RGB-thermal convolutional neural networks (CNNs) to unmanned aerial vehicles (UAVs) photography is a valuable alternative for diagnosing transmission line faults. However, existing CNNs struggle to generalize to TLD due to the clustered backgrounds and variable weather conditions. In addition, the limited computational resources and storage space of UAVs pose challenges for the lightweight design of models. In the present study, we developed a novel multilevel attention imitation knowledge distillation structure comprising a high-performing teacher model called MAINet-T and a compact student model called MAINet-S. We aimed to 1) improve the accuracy and robustness of TLD and 2) optimize the performance and capacity of the model for deployment on UAVs. The MAINet-T has a three-stage feature aggregation module and a detailed enhancement module to facilitate the processes of multi-modal and multilevel feature complement and interaction. To balance model performance and capacity for deployment, we proposed a novel KD strategy, including response distillation and feature distillation, to obtain an optimized model called MAINet-S*. Within feature distillation, we proposed a multilevel attention imitation module to integrate the advantages of the attention maps in different stages of the encoder. In experiments based on the VITLD dataset, MAINet-S* outperformed 15 state-of-the-art methods, with a 66.2% reduction in the number of weight parameters (Params) and a 69.9% increase in floating-point operations (FLOPs) compared with MAINet-T.

Efficient large-scale scene representation with a hybrid of high-resolution grid and plane features

Existing neural radiance fields (NeRF) methods for large-scale scene modeling require days of training using multiple GPUs, hindering their applications in scenarios with limited computing resources. Radiance field models based on explicit dense or hash grid features have been proposed for fast optimization, but their effectiveness has mainly been demonstrated in the context of object-scale scene representation. In this paper, we point out that the low feature resolution in explicit representation is the bottleneck for large-scale unbounded scene representation. To address this problem, we introduce a new and efficient hybrid feature representation for NeRF that fuses the 3D hash-grids and high-resolution 2D dense plane features. Compared with the dense-grid representation, the resolution of a dense 2D plane can be scaled up more efficiently. Based on this hybrid representation, we propose a fast optimization NeRF variant, called GP-NeRF, that achieves better rendering results while maintaining a compact model size. Extensive experiments on multiple large-scale unbounded scene datasets show that our model can converge in 1.5 h using a single GPU while achieving results comparable to or even better than the existing method that requires about one day’s training with 8 GPUs. Our code can be found at https://zyqz97.github.io/GP_NeRF/.