113年10-12月
ACS Nano (2024/12)
HfO2 Memristor-Based Flexible Radio Frequency Switches
電子研究所 連德軒 ( ■ 通訊作者) / JIF:15.8 / Rank:15/231
本研究成功運用記阻技術開發出適用於6G通訊系統的射頻開關。此裝置可適用於軟性應用基板,為第六代網路通訊下的穿戴式裝置提供一個解決方案。此射頻開關具備低插入損耗(2 dB)與超過840 GHz的截止頻率,能在機械彎折中操作仍保持穩定的性能與可靠性。此外,射頻開關可整合至可加工之光敏聚酰亞胺(PSPI)基板,實現與射頻天線及互連線等元件的高效2.5D整合。
Advanced Material (2024/12)
Breaking the Trade‐Off Between Mobility and On–Off Ratio in Oxide Transistors
電子研究所 連德軒 (▓ 通訊作者 ) / JIF:27.4 / Rank:NANOSCIENCE & NANOTECHNOLOGY (2.8% 4/141)
We successfully address the fundamental limitation of oxide semiconductors—namely, the trade-off between mobility and on-off ratio—through a simple mild CF4 plasma treatment. This approach achieves a high mobility of 104 cm2 V-1 s-1 and an on-off ratio exceeding 108 in In2O3 transistors. Furthermore, a subsequent low-temperature annealing process significantly enhances the reliability and stability of CF4-doped In2O3. We also demonstrate the application of CF4 plasma treatment in logic devices, highlighting its potential for advanced logic circuits. The combination of low-temperature processing and logic compatibility further establishes this technique as highly suitable for monolithic three-dimensional integration.
IEEE Transactions on Instrumentation and Measurement (2024/10)
Mitigating Negative Transfer Learning In Source Free-Unsupervised Domain Adaptation for Rotating Machinery Fault Diagnosis
電子研究所 陳坤志 (■ 通訊作者) / JIF:5.6 / Rank:11.8
在智慧製造領域中,馬達故障診斷為一個非常重要的研究議題,因為馬達被廣泛的應用在許多日常生活與產業中。雖然透過遷移式學習技術可以提高應用AI技術進行馬達故障診斷的泛用性,但由於不同馬達間的結構性差異,使得遷移式學習技術經常遭遇到嚴重的負面遷移學習效應(Negative Transfer Learning, NTL),進一步降低其泛用性優勢。此外,由於蒐集不同馬達種類資料不易,也造成馬達故障診斷泛用性技術的開發遭遇到很大的挑戰。此論文為世界首次透過非監督式學習方式來進行資料域自適應(domain adaptation)分群,並可進行虛擬標註(pseudo-labeling),可有效改善傳統遷移式學習的NTL效應。總結來說,此篇論文提出一種泛用性的方式,可以讓馬達AI診斷技術可以使用更低運算成本的方式讓技術更加普及,是一項極具產業應用性的技術。
IEEE Trans. Biomedical Engineering (2024/11)
Coordinate-Independent 3-D Ultrasound Principal Stretch and Direction Imaging
電子研究所 鄭耿璽 (■第一作者 ■通訊作者) / JIF:4.756 / Rank:SJR Q1 (7/407) Biomedical Engineering
1. 此論文獲得11月front cover, 以及featured article 2. This study is the first in cardiac elastography to manipulate 3-D principal stretch and direction information to identify myocardial anomalies, potentially revolutionizing how clinicians visualize and interpret myocardial deformations.
Small (2024/12)
Enhanced Performance of Sn-Based Perovskite Photodetectors Through Double-Sided Passivation for Near-Infrared Applications
電子研究所 譚至善 (■ 通訊作者) / JIF:13 / Rank:16/140 (Nanoscience & Nanotechnology)
近紅外光由於波長長、能量低,具有穿透生物組織更深、對樣本損傷更小的優勢,同時在活體生物系統中幾乎不受背景自發螢光的干擾。這些特性使其成為醫學診斷、手術和治療的理想工具。用來製作光電探測器的材料—「錫基鈣鈦礦」則因其無鉛、環境友好的特性,以及適合吸收近紅外光的能隙而備受矚目。然而,該材料暴露在大氣中容易氧化,穩定性差,材料迅速降解降低光電元件性能,阻礙了實際應用的發展。 本研究成功開發了一項創新技術改善錫基鈣鈦礦的穩定性。研究團隊在錫基鈣鈦礦材料的頂部與底部引入大型烷基銨中間保護層,通過雙面鈍化有效抑制氧化,讓材料保持穩定及維持光電性能,也顯著降低材料缺陷密度、增強載流子遷移率,並最大限度減少界面處的非輻射能量損失,從而顯著提升薄膜品質。實驗結果說明經過改善後的探測器,在弱光偵測中的信號辨別能力極為卓越。 這項技術能提升醫療檢測儀器的精度與效率,特別適合用於650–900 nm波段,該範圍內的光穿透性強且對生物組織的損傷低,有助於生物醫學影像及診斷技術的發展。
ACS Sensors (2024/09)
Achieving Room-Temperature ppb-Level H2S Detection in a Au-SnO2 Sensor with Low Voltage Enhancement Effect(Front Cover)
光電工程學系 冉曉雯 (■ 通訊作者) / JIF:8.3 / Rank:0.037
我們開發一種低功耗的Au/SnO₂ H₂S氣體感測器,製作出花蕊 奈米柱表面且引入介面電偶特性,來提升此感測器在室溫(24 ± 1°C)以及超低電壓(0.5伏特)下的偵測靈敏度,其檢測極限(LOD)可達到2 ppb,並可快速於兩分鐘恢復,為近年H2S感測器文獻中,能同時具有低功耗和高靈敏度的優秀結果,獲選為期刊封面(Front Cover,免付費的高競爭性封面),未來能應用於環境監測、食品安全檢測和呼氣健康監測領域。
Advanced Optical Materials (2024/11)
Directionally Asymmetric in Orbital Angular Momentum Generation Using Single-Layer Dielectric Janus Metasurface
光電工程學系 黃耀緯 (■ 通訊作者) / JIF:8 / Rank:10/119, JIF Percentile = 92.0%, OPTICS
首個展示能在任意自旋態下生成方向不對稱性軌道角動量的雅努斯超穎介面。以二氧化鈦為材料的超穎原子實現了在可見光範圍內的操作,超越了以往在微波及中紅外波段的研究。超穎原子提供局部/全域旋轉及寬度控制,透過瓊斯矩陣達成全域方向性多功能特性。此單層設計中的多功能性超越了基於空間多工或垂直堆疊的先前方法,拓展了其在多個領域的應用潛力。
Elsevier’s Vehicular Communications (2024/10)
Assessing the Impact of Communication Delays for Autonomous Intersection Management Systems
電機工程研究所 溫宏斌 (■ 共同作者(作者順序 2 )) / JIF:5.8 / Rank: 82.8% (TELECOMMUNICATIONS)
論文探討通訊延遲對於自動化路口管理系統的影響,並透過實驗數據佐證「降低通訊延遲對於路口管理的重要性」以及「自動化路口管理系統對於通訊延遲的最低需求」。針對十字路口與圓環進行實驗模擬與模型分析,我們發現(1)  通訊延遲在 0~100ms 的區間中,通訊延遲與路口效率呈線性關係(2)  當通訊延遲得以從 100ms 降低至 10ms 時,受影響的十字路口吞吐量,可以從 7-10% 降低至 0.7-1.0%。證實了「降低通訊延遲對於路口管理的重要性」(3)  通訊延遲要低於 247 ms,才能確保「自動化路口管理系統」能優於現今既有的紅綠燈系統,展示了「自動化路口管理系統對於通訊延遲的最低需求」
IEEE Transactions on Pattern Analysis and Machine Intelligence (2024/12)
Latent Semantics and Disentangled Attention
電機工程學系 簡仁宗 (■第一作者 ■通訊作者) / JIF:20.8 / Rank:1/54 (Computer Science, Hardware & Architecture)
轉換器(Transformer)為生成式人工智慧最重要的基石模型,生成式預訓練轉換器(GPT)更是現代大型語言模型(LLM)的核心架構,本研究鑽研大型語言模型關鍵技術並提出轉換器之潛在語意解纏以強化模型高效率表達能力,透過機率型解纏學習及遮罩式注意機制演算法,發展出編碼解碼過程並有效改善機器翻譯及語音辨識效能。本研究在潛在空間中的解纏表示對於文本理解及文句學習具有關鍵性貢獻。本論文發表在資訊科學領域排名第一的國際期刊為生成式機器學習的代表著作之一。
Psychiatry Research (2024/10)
Electrophysiological functional connectivity and complexity reflecting cognitive processing speed heterogeneity in young children with ADHD
電控工程研究所 柯立偉 (■ 通訊作者) / JIF:4.2 / Rank:Top 17.92% of PSYCHIATRY
Early intervention is imperative for young children with attention-deficit/hyperactivity disorder (ADHD) who manifest heterogeneous neurocognitive deficits. The study investigated the functional connectivity and complexity of brain activity among young children with ADHD exhibiting a fast cognitive processing speed (ADHD-F, n = 26), with ADHD exhibiting a slow cognitive processing speed (ADHD-S, n = 17), and typically developing children (n = 35) using wireless electroencephalography (EEG) during rest and task conditions. During rest, compared with the typically developing group, the ADHD-F group displayed lower long-range intra-hemispheric connectivity, while the ADHD-S group had lower frontal beta inter-hemispheric connectivity. During task performance, the ADHD-S group displayed lower frontal beta inter-hemispheric connectivity than the typically developing group. The ADHD-S group had lower frontal inter-hemispheric connectivity in broader frequency bands than the ADHD-F group, indicating ADHD heterogeneity in mental processing speed. Regarding complexity, the ADHD-S group tended to show lower frontal entropy estimators than the typically developing group during the task condition. These findings suggest that the EEG profile of brain connectivity and complexity can aid the early clinical diagnosis of ADHD, support subgrouping young children with ADHD based on cognitive processing speed heterogeneity, and may contain specific novel neural biomarkers for early intervention planning.
IEEE/ACM Transactions on Audio, Speech, and Language Processing (2024/11)
Multi-Domain Emotion Recognition Enhancement: A Novel Domain Adaptation Technique for Speech-Emotion Recognition
電控工程研究所 戴立嘉 (■ 通訊作者) / JIF:4.1 / Rank:9% (71/797)
As artificial intelligence advances, speech-emotion recognition (SER) has become a critical research area. We introduce a novel domain adaptation technique, multi domain emotion recognition enhancement (MDERE), which utilizes a non-negative matrix to reduce the inflexibility of the conventional binary label matrix for source domain data. Extensive testing on multiple cross-domain SER tasks has shown that MDERE substantially improves recognition accuracy, surpassing several state-of-the-art algorithms
MDPI Sensors (2024/12)
Modal Analysis with Asymptotic Strips Boundary Conditions of Skewed Helical Gratings on Dielectric Pipes as Cylindrical Metasurfaces for Multi-Beam Holographic Rod Antennas
電信 黃謀勤 (■第一作者 ■通訊作者) / JIF:3.4 / Rank:SJR (SCImago Journal Rank): 0.764
A dielectric pipe wound by a helical strip grating that is skewed upon stretching two sides of the coil in opposite directions along the axis and sheathed over a core dielectric rod has been analyzed by the asymptotic strips boundary conditions along with the technique of coordinate transformation, all under the analytical framework of vector potentials. The concept of the approach begins with the rolling up of a coordinate-transformed plane in the Cartesian system into that of the cylindrical system, just like rolling up a piece of paper. Results of modal dispersion, field distributions, and radiation patterns computed by the code written based on this method are compared favorably with those simulated by an independent software solver. The latter radiated far-fields are acquired via eigen-modal field solutions provided by the approach that facilitate the Fourier integration of the cylindrical aperture equivalent currents. The modal field solutions provided by the analysis enable the study and design of holographic cylindrical leaky-wave antennas, which comprise differently grated rod sections that represent the various surface properties required by the cylindrical-type hologram with a modulated tensor reactance surface. Our analysis is distinguished from most other related ones in the literature on holographic cylindrical antennas, which in large proportions, have entailed extensions of initial analyses of conventional planar topologies to their conformal or cylindrical counterparts. Computed results of radiation patterns for such connections of differently skew-grated rods for emitting single and dual beams towards various directions and of different polarizations are compared well with simulations by the software. Prototypes were also manufactured and measured in terms of modal dispersion and radiation patterns. Results from these experiments agree reasonably with those expected from theory of the analysis and simulations. This work has opened the doors to this type of cylindrical leaky-wave antenna whose surface can be readily retextured by the proposed skew-gratings via mechanical sliding and stretching without the need to change the underlying rod structure, thus offering the possibility of beam reconfiguration and steering.
IEEE Transactions on Information Theory (2024/11)
Semidefinite programming bounds on the size of entanglement-assisted codeword stabilized quantum codes
電信工程研究所 賴青沂 (■第一作者 ) / JIF:2.2 / Rank:138/250 Computer Science
這篇論文的主要貢獻在於將半定規劃(SDP)技術應用於糾 纏輔助量子碼(EA-QECC)的理論界限分析,並揭示其結構特性。與傳統的線性規劃技術相比,SDP方法提供了更準確的性能評估,有助於進一步提升對糾纏輔助碼的理解。此外,論文提出了Shor-Laflamme和shadow權重計數器的新詮釋,進一步闡明量子錯誤更正碼的特性,為量子碼的理論研究與實際應用奠定了更為堅實的基礎。