检索结果-内蒙古大学图书馆

24th IEEE International Conference of Young Professionals in Electron Devices and Materials, EDM 2023

作者： Koloskov, Dmitry B. Dragunov, Valeriy P. Gluhov, Alexander V. Novosibirsk State Technical University Dept. of Semiconductor Devices and Microelectronics Novosibirsk Russia Siberian State University of Telecommunications and Information Science Dept. of Computer Aided Design Novosibirsk Russia

ISBN: (纸本)9798350336870

The paper deals with the problem of counteracting parasitic layout elements in precise analog circuits, in particular, precision voltage references and high-bit DACs and ADCs. We described the two main methods of countering them - minimization and compensation ones. On the example of a 14-bit 1 GHz DAC with embedded Bandgap voltage reference, we evaluated the effect of parasitic resistances on the device performance and its static characteristics. We showed that parasitic resistances could increase the TC of the Bandgap voltage reference source by more than 4 times, from 4 ppm/C to 17.3 ppm/C. This impact was assessed as critical for studied 14-bit DAC functioning. With the compensation method of countering the parasitic resistances, we were able to reduce the TC of the reference to 4.8 ppm/C. We concluded that dealing with parasitic elements is mandatory for precision Bandgap voltage references and high-bit DACs, and the compensation method is effective for this task. © 2023 IEEE.

关键词： Energy gap

来源：评论

学校读者我要写书评

暂无评论

3D Vehicle Detection for Autonomous Buses

3D Vehicle Detection for Autonomous Buses

引用

2023 International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2023

作者： Chiu, Yi-Kai Chen, Xiu-Zhi Chen, Yen-Lin Huang, Chieh-Sheng National Taipei University of Technology Dept. of Computer Science & Information Engineering Taipei Taiwan Elan Microelectronics Corporation Artificial Intelligence Research and Development Department Hsinchu Taiwan

ISBN: (纸本)9798350324174

The development of automatic driving technology and Advanced Driver Assistance Systems (ADAS) has become more and more mature, and has led to the development of 3D object detection technologies and discussions. Most of the researches are focus on the applications of small cars, but not common in large vehicles, such as buses and trucks, although such applications are more important for them. Therefore, this study mainly uses the bus data to verify the feasibility of the existed model, and by adjusting the loss function, the prediction accuracy of the class and bounding box can be improved. Based on our experimental results, we found that the model trained by KITTI dataset can predict acceptable result refers to the input images from the bus viewpoint. As a result, we can expect that the ADAS for large vehicles can be implemented based on the sample model structure, and its accuracy is able to be improved by fine-tuning based on more large vehicle viewpoint data. © 2023 IEEE.

关键词： Buses

来源：评论

学校读者我要写书评

暂无评论

Compensation Method of Quantized Deep Learning Models for Edge Devices

Compensation Method of Quantized Deep Learning Models for Ed...

引用

2023 International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2023

作者： Chen, Xiu-Zhi Li, Jhen-Hao Chen, Yen-Lin Huang, Chieh-Sheng National Taipei University of Technology Dept. of Computer Science & Information Engineering Taipei Taiwan Elan Microelectronics Corporation Artificial Intelligence Research and Development Department Hsinchu Taiwan

ISBN: (纸本)9798350324174

Quantization is one of the optimization methods for developing deep learning models for edge devices. Through converting the floating-point into 8bit integer or even lower bitwidth, the model's storage size can be reduced. As the rounding error exists during the quantization process, the model performance decreases. As a result, a method that can recover model performance is needed. In this research, a compensation method for improving the performance of quantized deep learning models is proposed, which make the quantized model can achieve equal or even better performance compared to the original floating-point model. © 2023 IEEE.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

3D object detection with low resolution dept. map

3D object detection with low resolution depth map

引用

2023 International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2023

作者： Huang, Wei-Ting Chen, Xiu-Zhi Chen, Yen-Lin Huang, Chieh-Sheng National Taipei University of Technology Dept. of Computer Science & Information Engineering Taipei Taiwan Elan Microelectronics Corporation Artificial Intelligence Research and Development Department Hsinchu Taiwan

ISBN: (纸本)9798350324174

Obtaining high-resolution dept. map from radar and lidar for training dept. estimation network is a common approach in recent years, although the expensive price and high computational complexity cause it hard to be apply in real-world applications. In this research, we propose a novel concept that only requires low resolution dept. map for training, but able to reach similar performance comparing to high-resolution dept. map-based approaches. Refers to the experimental results, which evaluated on KITTI dataset, our proposed concept decreases 7% and 2% on the mAP of car and pedestrian comparing to the original approach in CaDDN, although it achieves similar performance on visualization result. © 2023 IEEE.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Scalable Event-by-Event Processing of Neuromorphic Sensory Signals with Deep State-Space Models

Scalable Event-by-Event Processing of Neuromorphic Sensory S...

引用

2024 International Conference on Neuromorphic Systems, ICONS 2024

作者： Schone, Mark Sushma, Neeraj Mohan Zhuge, Jingyue Mayr, Christian Subramoney, Anand Kappel, David Ruhr University Bochum Institut für Neuroinformatik Germany Highly-Parallel VLSI-Systems and Neuro-Microelectronics TU Dresden Dresden Germany Dresden Germany with Human-in-the-Loop Dresden Germany University of London Royal Holloway Dept. of Computer Science Egham United Kingdom

ISBN: (纸本)9798350368659

Event-based sensors feature valuable properties for real-time processing such as fast response times and encoding of the sensory data as successive temporal differences. However, most current methods either collapse events into frames, therefore suppressing these properties, or cannot scale up when processing the event data directly event-by-event. In this work, we address the key challenges of scaling up event-by-event modeling of the long event streams emitted by such sensors, which is a particularly relevant problem for neuromorphic computing. While prior methods can process up to a few thousand time steps, our model, based on modern recurrent deep state-space models, scales to event streams of millions of events for both training and inference. We leverage their stable parameterization for learning long-range dependencies, parallelizability along the sequence dimension, and their ability to integrate asynchronous events effectively to scale them up to long event streams. We further augment these with novel event-centric techniques enabling our model to match or beat the state-of-the-art performance on several event stream benchmarks. In the Spiking Speech Commands task, we improve state-of-the-art by a large margin of 7.7 % to 88.4 %. On the DVS128-Gestures dataset, we achieve competitive results without using frames or convolutional neural networks. Our work demonstrates, for the first time, that it is possible to use fully event-based processing with purely recurrent networks to achieve state-of-the-art task performance in several event-based benchmarks. © 2024 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Multi-Task Deep Learning Model for Autonomous Driving: Object Detection, Semantic Segmentation, and dept. Estimation

Multi-Task Deep Learning Model for Autonomous Driving: Objec...

引用

2023 International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2023

作者： Wang, Yih-Chen Lin, Jhe-Li Chen, Yen-Lin Huang, Chieh-Sheng Lai, Ming-Liang National Taipei University of Technology Dept. of Computer Science & Information Engineering Taipei Taiwan Elan Microelectronics Corporation Artificial Intelligence Research and Development Department Hsinchu Taiwan National Taipei University of Technology Graduate Institute of Intellectual Property Taipei Taiwan

ISBN: (纸本)9798350324174

In the field of autonomous driving, many models based on deep learning methods have been constructed to solve computer vision tasks related to this domain, such as object detection, semantic segmentation, and dept. estimation. Each model can provide different information about the surrounding environment of a vehicle to assist in driving. However, for applying in the real world, more detailed information about the surrounding environment is essentially required. In this study, we propose a model based on the concept of multi-task learning. This model is an encoder-decoder architecture mainly consisting of an encoder using the hard-parameter sharing technique and three decoders for individual tasks. Therefore, this model can handle object detection, semantic segmentation, and dept. estimation at the same time. Our proposed multi-task model has been verified to perform well on the public dataset Cityscapes and has higher generalizability than other models. © 2023 IEEE.

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

Weight Sparsity Complements Activity Sparsity in Neuromorphic Language Models

Weight Sparsity Complements Activity Sparsity in Neuromorphi...

引用

2024 International Conference on Neuromorphic Systems, ICONS 2024

作者： Mukherji, Rishav Schone, Mark Nazeer, Khaleelulla Khan Mayr, Christian Kappel, David Subramoney, Anand Birla Institute of Technology and Science Pilani Goa Campus Goa India TU Dresden Highly-Parallel VLSI-Systems and Neuro-Microelectronics Dresden Germany Dresden Germany with Human-in-the-Loop Dresden Germany Ruhr University Bochum Institut für Neuroinformatik Germany Royal Holloway University of London Dept. of Computer Science Egham United Kingdom

ISBN: (纸本)9798350368659

Activity and parameter sparsity are two standard methods of making neural networks computationally more efficient. Event-based architectures such as spiking neural networks (SNNs) naturally exhibit activity sparsity, and many methods exist to sparsify their connectivity by pruning weights. While the effect of weight pruning on feed-forward SNNs has been previously studied for computer vision tasks, the effects of pruning for complex sequence tasks like language modeling are less well studied since SNNs have traditionally struggled to achieve meaningful performance on these tasks. Using a recently published SNN-like architecture that works well on small-scale language modeling, we study the effects of weight pruning when combined with activity sparsity. Specifically, we study the tradeoff between the multiplicative efficiency gains the combination affords and its effect on task performance for language modeling. To dissect the effects of the two sparsities, we conduct a comparative analysis between densely activated models and sparsely activated event-based models across varying degrees of connectivity sparsity. We demonstrate that sparse activity and sparse connectivity complement each other without a proportional drop in task performance for an event-based neural network trained on the Penn Treebank and WikiText-2 language modeling datasets. Our results suggest sparsely connected event-based neural networks are promising candidates for effective and efficient sequence modeling. © 2024 IEEE.

关键词： Modeling languages

来源：评论

学校读者我要写书评

暂无评论

Application of Compensation Method for Layout Parasitic Elements Control in Precision DACs

Application of Compensation Method for Layout Parasitic Elem...

引用

Annual Siberian Russian Workshop on Electron Devices and Materials (EDM)

作者： Dmitry B. Koloskov Valeriy P. Dragunov Alexander V. Gluhov Dept. of Semiconductor Devices and Microelectronics Novosibirsk State Technical University Novosibirsk Russia Dept. of Computer Aided Design Siberian State University of Telecommunications and Information Science Novosibirsk Russia

关键词：

来源：评论

学校读者我要写书评

暂无评论

SigDLA: A Deep Learning Accelerator Extension for Signal Processing

arXiv

引用

arXiv 2024年

作者： Fu, Fangfa Zhang, Wenyu Jiang, Zesong Zhu, Zhiyu Li, Guoyu Yang, Bing Liu, Cheng Xiao, Liyi Wang, Jinxiang Li, Huawei Li, Xiaowei Department of Microelectronics Science and Technology Harbin Institute of Technology Harbin China SKLP Institute of Computing Technology Chinese Academy of Sciences Beijing China Dept. of Computer Science University of Chinese Academy of Sciences Beijing China Institute of Advanced Technology University of Science and Technology of China Hefei China Dept. of Computer Science and Technology Harbin University of Science of Technology Harbin China

Deep learning and signal processing are closely correlated in many IoT scenarios such as anomaly detection to empower intelligence of things. Many IoT processors utilize digital signal processors (DSPs) for signal processing and build deep learning frameworks on this basis. While deep learning is usually much more computing-intensive than signal processing, the computing efficiency of deep learning on DSPs is limited due to the lack of native hardware support. In this case, we present a contrary strategy and propose to enable signal processing on top of a classical deep learning accelerator (DLA). With the observation that irregular data patterns such as butterfly operations in FFT are the major barrier that hinders the deployment of signal processing on DLAs, we propose a programmable data shuffling fabric and have it inserted between the input buffer and computing array of DLAs such that the irregular data is reorganized and the processing is converted to be regular. With the online data shuffling, the proposed architecture, SigDLA, can adapt to various signal processing tasks without affecting the deep learning processing. Moreover, we build a reconfigurable computing array to suit the various data width requirements of both signal processing and deep learning. According to our experiments, SigDLA achieves an average performance speedup of 4.4×, 1.4×, and 1.52×, and average energy reduction of 4.82×, 3.27×, and 2.15× compared to an embedded ARM processor with customized DSP instructions, a DSP processor, and an independent DSP-DLA architecture respectively with 17% more chip area over the original DLAs. © 2024, CC BY-NC-SA.

关键词： Digital signal processing

来源：评论

学校读者我要写书评

暂无评论

Wideband Circularly Polarized FOWLP Antenna in Package for Satellite Communications

Wideband Circularly Polarized FOWLP Antenna in Package for S...

引用

Electronic Components and Technology Conference (ECTC)

作者： Kai Bo Zheng Mei Sun Yong Xin Guo IME attached student from Dept. of Electrical and Computer Engineering National University of Singapore (NUS) Singapore A*STAR (Agency for Science Technology and Research) Institute of Microelectronics (IME) Singapore Electrical and Computer Engineering Dept. National University of Singapore Singapore

This paper presents a wideband circularly-polarized Antenna-in-Package (AiP) element implemented in Fan-Out Wafer-Level-Packaging (FOWLP) technology for satellite communication. The AiP element features a meta-surface radiator with a single-ended aperture coupled feeding structure to achieve a wideband and circular polarization performance. By stacking a top mold of 500μm, a middle mold of 90μm and a 42μm RDL layer at the bottom, with a size of 10 mm×10 mm×0.652 mm, the AiP element achieves a Return Loss > 17.95 dB, an Axial Ratio < 2.67 dB, and a Gain of 7.19-7.88 dBi over 27–31 GHz. The antenna design concept is further confirmed using printed circuit board (PCB) technology for fast prototyping and evaluation.

关键词：

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：