Learn decision trees with deep visual primitives

In this paper, we strive to propose a self-interpretable framework, termed PrimitiveTree, that incorporates deep visual primitives condensed from deep features with a conventional decision tree, bridging the gap between deep features extracted from deep neural networks (DNNs) and trees’ transparent decision-making processes. Specifically, we utilize a codebook, which embeds the continuous deep features into a finite discrete space (deep visual primitives) to distill the most common semantic information. The decision tree adopts the spatial location information and the mapped primitives to present the decision-making process of the deep features in a tree hierarchy. Moreover, the trained interpretable PrimitiveTree can inversely explain the constituents of the deep features, highlighting the most critical and semantic-rich image patches attributing to the final predictions of the given DNN. Extensive experiments and visualization results validate the effectiveness and interpretability of our method.     

基于深度学习的ELM实时识别研究

基于深度学习的方法，在HL-2A装置上开发出了一套边缘局域模(ELM)实时识别算法。算法使用5200次放电数据(约24.19万数据切片)进行学习，得到一个深度为22层的卷积神经网络。为衡量算法的识别能力，识别了HL-2A装置自2009年实现稳定ELMy H模放电以来所有历史数据(约26000次放电数据)，共识别出1665次H模放电，其中误识别35次，误报率为2.10%。在实际的1634次H模放电中，漏识别4次，漏识别率为0.24%。该误报率和漏报率可以满足ELM实时识别的精度要求。识别算法在实时控制环境下，对单个时间点的平均计算时间为0.46ms，可以满足实时控制的计算速度要求。     

深亚微米MOSFET衬底电流的模拟与分析

利用器件模拟手段对深亚微米ＭＯＳＦＥＴ的衬底电流进行了研究和分析，给出了有效的道长度，栅氧厚度，源漏结深，衬底掺杂浓度以及电源电压对深亚微米ＭＯＳＦＥＴ衬底电流的影响，发现电源电压对深亚微米ＭＯＳＦＥＴ的衬底电流有着强烈的影响，热载流子效应随电源电压的降低而迅速减小，当电源电压降低到一定程度时，热载流子效应不再成为影响深亚微米ＭＯＳ电路可靠性的主要问题。     

CFD在燃煤细粒子凝聚过程中的应用

燃煤产生的细粒子富集了大量的有毒痕量元素,对大气环境和人类健康造成严重危害。本文针对燃煤细粒子的形成过程,在计算流体力学(CFD)的基础上,结合气溶胶动力学理论,模拟了烟气中细粒子在圆柱形流场中由于碰撞而凝聚的过程,并通过计算结果分析细粒子颗粒特征参量(速度、质量、直径和颗粒数目)随流场的变化趋势,为深入研究燃煤细粒子的形成演化机理提供理论基础。     

深亚微米条件下的多层VLSI／MCM层分配的遗传算法

本文提出了一个深亚微米条件下的多层ＶＬＳＭＣＭ有约束分层层分配的遗传算法。该算法分为两步：首先进行超层分配，使各线网满足Ｃｒｏｓｓｔａｌｋ约束，且超层数目最少；然后进行各超层的通孔最少化二分层。与目前的层分配算法相比，该遗传算法具有目标全面，全局优化能力强等特点，是一种可应用于深亚微米条件下的ＩＣ ＣＡＤ的有效分层方法。     

用迁移率和载流子浓度研究碲镉汞中的深能级

获得了一种研究碲镉汞深能级的方法。通过分析迁移率 载子浓度与温度的关系，可以得到关于深能级的重要依据。     

Deep levels,electrical and optical characteristics in SnTe-Doped GaSb Schottky Diodes

The GaSb layers investigated were grown directly on GaAs substrates by molecular beam epitaxy (MBE) using SnTe source as the n-type dopant. By using admittance spectroscopy, a dominant deep level with the activation energy of 0.23-0.26 eV was observed and its concentration was affected by the Sb₄/Ga flux ratio in the MBE growth. A lowest deep-level concentration together with a highest mobility was obtained for GaSb grown at 550°C under a Sb₄/Ga beam equivalent pressure (BEP) ratio around 7, which should correspond to the lowest ratio to maintain a Sb-stabilized surface reconstruction. In the Hall measurement, an analysis of the temperature-dependent mobility shows that the ionized impurity concentration increases proportionally with the sample’s donor concentration, suggesting that the ionized impurity was introduced by an SnTe source. In addition, optical properties of an undoped p-, a lightly and heavily SnTe-doped GaSb layers were studied by comparing their photoluminescence spectra at 4.5K.     

深能级杂质对光导半导体开关非线性特性的影响

建立了非线性GaAs光导开关深能级杂质瞬态模型的基本方程,获得了与实验现象定性吻合的电流输出,给出了平均载流子随时间演化的情况.分析结果表明,在考虑了深能级杂质的俘获、发射和碰撞电离后,有可能对非线性光导开关中发生的一系列现象做出解释,进一步的仔细分析将对非线性光导开关的设计和制作提供理论指导.     

Controlled oxygen incorporation in indium gallium arsenide and indium phosphide grown by metalorganic vapor phase epitaxy

The defect engineering in metalorganic vapor phase epitaxy In_xGa_1-xAs and InP by controlled oxygen doping using diethyl aluminum ethoxide (DEALO) was developed in this study. DEALO doping has led to the incorporation of Al and O, and the compensation of shallow Si donors in In_xGa_1−xAs: Si with 0 ≤ x ≤ 0.25. With the same DEALO mole fraction during growth, the incorporation of Al and O was found to be independent of x, but the compensation of Si donors decreases with increasing In content. Deep level transient spectroscopy analysis on a series of In_xGa_1-xAs: Si. samples with 0 ≤ x ≤ 0.18 revealed that oxygen incorporation led to a set of deep levels, similar to those found in DEALO doped GaAs. As the In composition was increased, one or more of these deep levels became resonant with the conduction band and led to a high electron concentration in oxygen doped In_0.53Ga_0.47As. Low temperature photoluminescence emission measurements at 12K on the same set of samples revealed the quenching of the near-band edge peak, and the appearance of new oxygen-induced emission features. DEALO doping in InP has also led to the incorporation of Al and O, and the compensation of Si donors due to oxygen-induced multiple deep levels.     

LB技术制备大面积超薄PMMA抗蚀层的研究

采用自行研制的ＬＢ自动提膜装置，制备出大面积（１０×８ｃｍ２）、高质量的ＰＭＭＡ超薄抗蚀剂膜，并将其用于高分辨率铬掩模版的研制。通过电子束曝光，湿法蚀刻，制作了分辨率优于０．５μｍ，特征线宽０．３８μｍ的４（１００ｍｍ）铬掩模版。