A two-dimensional analytical model for channel potential and threshold voltage of short channel dual material gate lightly doped drain MOSFET

An analytical model for the channel potential and the threshold voltage of the short channel dual-material-gate lightly doped drain （DMG-LDD） metal-oxide-semiconductor field-effect transistor （MOSFET） is presented using the parabolic approximation method. The proposed model takes into account the effects of the LDD region length, the LDD region doping, the lengths of the gate materials and their respective work functions, along with all the major geometrical parameters of the MOSFET. The impact of the LDD region length, the LDD region doping, and the channel length on the channel potential is studied in detail. Furthermore, the threshold voltage of the device is calculated using the minimum middle channel potential, and the result obtained is compared with the DMG MOSFET threshold voltage to show the improvement in the threshold voltage roll-off. It is shown that the DMG-LDD MOSFET structure alleviates the problem of short channel effects （SCEs） and the drain induced barrier lowering （DIBL） more efficiently. The proposed model is verified by comparing the theoretical results with the simulated data obtained by using the commercially available ATLASTM 2D device simulator.

A two-dimensional analytical modeling for channel potential and threshold voltage of short channel triple material symmetrical gate Stack(TMGS) DG-MOSFET

In the present work, a two-dimensional（2D） analytical framework of triple material symmetrical gate stack（TMGS）DG-MOSFET is presented in order to subdue the short channel effects. A lightly doped channel along with triple material gate having different work functions and symmetrical gate stack structure, showcases substantial betterment in quashing short channel effects to a good extent. The device functioning amends in terms of improved exemption to threshold voltage roll-off, thereby suppressing the short channel effects. The encroachments of respective device arguments on the threshold voltage of the proposed structure are examined in detail. The significant outcomes are compared with the numerical simulation data obtained by using 2D ATLAS;device simulator to affirm and formalize the proposed device structure.

Comparative study on the performance of ConvLSTM and ConvGRU in classification problems-taking early warning of short-duration heavy rainfall as an example

卷积长短期记忆单元ConvLSTM和卷积门控循环单元ConvGRU是两种广泛应用的深度学习单元,通过将循环机制与卷积运算相结合,常常用于时空序列的预测.为了明确上述两种模型的收敛速度和分类能力,需要使用相同的模型架构对相同的分类问题进行预测.本研究将北京短时强降水区级预警问题看作深度学习中的二分类问题,使用京津冀雷达网的组合反射率数据和北京区域内的自动气象站降雨数据进行深度学习模型的训练和评估.结果表明,ConvGRU的收敛速度比ConvLSTM快约25%.ConvLSTM和ConvGRU的预警性能随地区,时间,降雨强度的变化趋势相似,但大部分ConvLSTM的得分较高,少数情况下ConvGRU的得分较高.

Fabrication and characterization of groove-gate MOSFETs based on a self-aligned CMOS process

N and P-channel groove-gate MOSFETs based on a self-aligned CMOS process have been fabricated and characterized. For the devices with channel length of 140nm, the measured drain induced barrier lowering （DIBL） was 66mV/V for n-MOSFETs and 82mV/V for p-MOSFETs. The substrate current of a groove-gate n-MOSFET was 150 times less than that of a conventional planar n-MOSFET, These results demonstrate that groove-gate MOSFETs have excellent capabilities in suppressing short-channel effects. It is worth emphasizing that our groove-gate MOSFET devices are fabricated by using a simple process flow, with the potential of fabricating devices in the sub-100nm range.

Study on two-dimensional analytical models for symmetrical gate stack dual gate strained silicon MOSFETs

Based on the exact resultant solution of two-dimensional Poisson＇s equation, the novel two-dimensional models, which include surface potential, threshold voltage, subthreshold current and subthreshold swing, have been developed for gate stack symmetrical double-gate strained-Si MOSFETs. The models are verified by numerical simulation. Besides offering the physical insight into device physics, the model provides the basic designing guidance of further immunity of short channel effect of complementary metal-oxide-semiconductor （CMOS）-based device in a nanoscale regime.

Snapback-free shorted anode LIGBT with controlled anode barrier and resistance

A novel shorted anode lateral-insulated gate bipolar transistor(SA LIGBT)with snapback-free characteristic is proposed and investigated.The device features a controlled barrier V_(barrier)and resistance R_(SA)in anode,named CBR LIGBT.The electron barrier is formed by the P-float/N-buffer junction,while the anode resistance includes the polysilicon layer and N-float.At forward conduction stage,the V_(barrier)and R_(SA)can be increased by adjusting the doping of the P-float and polysilicon layer,respectively,which can suppress the unipolar mode to eliminate the snapback.At turn-off stage,the low-resistance extraction path(N-buffer/P-float/polysilicon layer/N-float)can quickly extract the electrons in the N-drift,which can effectively accelerate the turn-off speed of the device.The simulation results show that at the same V_(on) of 1.3 V,the E_(off)of the CBR LIGBT is reduced by 85%,73%,and 59.6%compared with the SSA LIGBT,conventional LIGBT,and TSA LIGBT,respectively.Additionally,at the same Eoffof 1.5 m J/cm^(2),the CBR LIGBT achieves the lowest V_(on) of 1.1 V compared with the other LIGBTs.

Two-dimensional analytical models for asymmetric fully depleted double-gate strained silicon MOSFETs

This paper develops the simple and accurate two-dimensional analytical models for new asymmetric double-gate fully depleted strained-Si MOSFET. The models mainly include the analytical equations of the surface potential, surface electric field and threshold voltage, which are derived by solving two dimensional Poisson equation in strained-Si layer. The models are verified by numerical simulation. Besides offering the physical insight into device physics in the model, the new structure also provides the basic designing guidance for further immunity of short channel effect and draininduced barrier-lowering of CMOS-based devices in nanometre scale.

基于CEEMD-SE的CNN&LSTM-GRU短期风电功率预测

为进一步提升短期风电功率的预测精度,提出了一种基于互补集合经验模态分解-样本熵(complementary ensemble empirical mode decomposition-sample entropy,CEEMD-SE)的卷积神经网络(convolutional neural network,CNN)和长短期记忆-门控循环单元(longshorttermmemory-gatedrecurrentunit,LSTM-GRU)的短期风电功率预测模型。首先,利用互补集合经验模态分解将原始风电功率序列分解为若干本征模态函数(intrinsic mode function,IMF)分量和一个残差(residual,RES)分量,利用样本熵算法将相近的分量进行重构;其次,搭建卷积神经网络和长短期记忆网络的并行网络结构,提取数据的局部特征和时序特征,并将特征融合后输入门控循环单元网络中进行学习预测;最后,通过算例进行验证,结果表明采用该模型后预测精度得到了有效提升,其均方根误差降低了15.06%、平均绝对误差降低了15.22%、决定系数提高了1.91%。

融合Mar-GLSTM的流程生产工艺质量预测算法

针对流程生产连续性强、时序耦合复杂等特点,传统神经网络不具备长期记忆能力,且在深层次网络训练时易出现训练参数灾难、梯度爆炸等问题,提出基于马尔可夫优化的融合门控循环单元(GRU)与长短期记忆网络(LSTM)的组合预测模型(Mar-G LSTM)。首先在循环神经网络结构中融入门控机制构建深度LSTM神经网络模型,对流程生产时序数据信息进行选择性记忆,学习时序数据序列的信息依赖,进而解决训练过程中的梯度爆炸问题;同时结合马尔可夫链对GRU-LSTM模型的预测结果进行修正优化,在降低模型的复杂度的情况下进一步提高了模型的预测精度。最后,结合某流程生产线的工艺数据进行分析验证,结果表明,Mar-G LSTM算法在预测精度上较随机森林模型、门控循环单元神经网络模型(GRU)、长短期记忆神经网络模型(LSTM)和卷积神经网络与门控循环单元网络组合模型(CNN-GRU)分别提高了37.42%、21.32%、17.91%和12.56%,所提Mar-G LSTM算法可实现流程生产质量的准确预测,为降低工艺参数调控任务的完成时间提供了思路和实现途径。

A threshold voltage analytical model for high-k gate dielectric MOSFETs with fully overlapped lightly doped drain structures

We investigate the influence of voltage drop across the lightly doped drain（LDD） region and the built-in potential on MOSFETs,and develop a threshold voltage model for high-k gate dielectric MOSFETs with fully overlapped LDD structures by solving the two-dimensional Poisson＇s equation in the silicon and gate dielectric layers.The model can predict the fringing-induced barrier lowering effect and the short channel effect.It is also valid for non-LDD MOSFETs.Based on this model,the relationship between threshold voltage roll-off and three parameters,channel length,drain voltage and gate dielectric permittivity,is investigated.Compared with the non-LDD MOSFET,the LDD MOSFET depends slightly on channel length,drain voltage,and gate dielectric permittivity.The model is verified at the end of the paper.

基于CNN-BiGRU-Attention的短期电力负荷预测

针对目前电力负荷数据随机性强,影响因素复杂,传统单一预测模型精度低的问题,结合卷积神经网络(Convolutional neural network,CNN)、双向门控循环单元(Bi-directional gated recurrent unit,BiGRU)以及注意力机制(Attention)在短期电力负荷预测上的不同优点,提出一种基于CNN-BiGRU-Attention的混合预测模型。该方法首先通过CNN对历史负荷和气象数据进行初步特征提取,然后利用BiGRU进一步挖掘特征数据间时序关联,再引入注意力机制,对BiGRU输出状态给与不同权重,强化关键特征,最后完成负荷预测。试验结果表明,该模型的平均绝对百分比误差(Mean absolute percentage error,MAPE)、均方根误差(Root mean square error,RMSE)、判定系数(R-square,R~2)分别为0.167%、0.057%、0.993,三项指标明显优于其他模型,具有更高的预测精度和稳定性,验证了模型在短期负荷预测中的优势。

基于二次分解双向门控单元新型电力系统超短期负荷预测

在新型电力系统中,电力负荷随机性和波动性较强,现有预测方法难以对其实现高精度预测。为此,提出一种基于二次分解和双向门控循环单元的超短期负荷预测模型。首先,针对电力负荷的强随机性和强波动性,利用自适应噪声完备经验模态分解对电力负荷历史序列进行初步分解,使负荷序列更加平稳。随后,对初步分解得到的强非平稳分量运用连续变分模态分解进行二次分解,降低其预测难度。最后,为充分学习电力负荷的时序特征,在预测过程构建基于双向门控循环单元的超短期电力负荷预测模型。实验结果表明,该模型相较于现有优秀预测模型有更高的预测精度。

GPU异构计算环境中长短时记忆网络模型的应用及优化

随着深度学习的广泛应用及算力资源的异构化,在GPU异构计算环境下的深度学习加速成为又一研究热点。文章探讨了在GPU异构计算环境中如何应用长短时记忆网络模型,并通过优化策略提高其性能。首先,介绍了长短时记忆网络模型的基本结构(包括门控循环单元、丢弃法、Adam与双向长短时记忆网络等);其次,提出了在GPU上执行的一系列优化方法,如CuDNN库的应用及并行计算的设计等。最终,通过实验分析了以上优化方法在训练时间、验证集性能、测试集性能、超参数和硬件资源使用等方面的差异。

面向文本识别的CRNN模型的改进

复杂场景下文本识别因阴影、残缺、模糊、虚化等因素会出现识别精度下降问题。鉴于此,提出一种基于特征融合与双向简化门结构的CRNN模型。首先引入特征融合机制改进卷积神经网络(CNN)模型,利用特征金字塔结构,多加一条自底向上的路径,将低层特征与高层特征融合在一起,以保留更多低层细节特征,提高场景文本识别精度;其次通过合并遗忘门与输入门,得到结构更简单、计算量和参数量更少的简化门结构替换长短期记忆(LSTM)网络改进循环神经网络(RNN)模型部分;最后设计消融实验验证改进后模型的有效性。三个数据集的测试结果表明:在ResNet50做主干网络时,与原始模型相比,改进后模型准确率提升了1.5%以上;在MobileNetV3做主干网络时,准确率提升了1.4%以上。

高压大功率SiC MOSFETs短路保护方法

碳化硅(SiC)MOSFETs短路承受能力弱,研究其短路保护方法成为保障电力电子设备安全运行的重要课题。现有方法大多围绕低压小功率SiC MOSFETs,然而随着电压和功率等级的提升,器件特性有所差异,直接套用以往设计难以实现高压大功率SiC MOSFETs的快速、可靠保护。该文首先详细研究了几种常用短路检测方法;其次基于高压大功率SiC MOSFETs器件特性,深入对比分析了不同短路检测方法的适用性,提出一种阻容式漏源极电压检测和栅极电荷检测相结合的短路保护方法;最后搭建了实验平台验证所提方法的可行性。结果表明,提出的方法在硬开关短路故障(hard switching fault,HSF)下,保护响应时间缩短了1.4μs,短路能量降低了62.5%;且能可靠识别负载短路故障(fault under load,FUL)。

面向语法加权图文本的方面情感三元组抽取

方面情感三元组抽取包括方面抽取、意见抽取和方面情感分类3项任务,以管道方式解决该任务的研究方法无法利用元素之间的交互信息,同时也会造成错误传播和冗余训练。基于此,提出一种基于门控注意力和加权图文本的方面情感三元组抽取方法。采用双向长短时记忆网络学习句子的序列特征表示;利用门控注意力单元学习单词之间的线性联系;利用语法距离加权图卷积网络增强三元组元素之间的交互;利用网格标记推理策略预测三元组。在4个公开数据集上进行实验,结果表明:所提方法可以有效增强三元组元素之间的交互,提高三元组抽取的准确率;同时,所提方法的F1值分别为57.94%、70.54%、61.95%和67.66%,与基准模型相比均有所提高。

基于Bo-BiLSTM网络的IGBT老化失效预测方法

针对绝缘栅双极型晶体管(IGBT)受热应力冲击后对其进行老化失效预测精度不高的情况,提出一种基于贝叶斯优化(Bo)-双向长短期记忆(BiLSTM)网络的IGBT老化失效预测方法。首先分析IGBT模块老化失效原理,然后基于NASA老化实验数据集建立失效特征数据库,最后利用Matlab软件构造Bo-BiLSTM网络预测失效特征参数数据。选取常用回归预测性能评估指标将长短期记忆(LSTM)网络模型、BiLSTM网络模型与Bo-BiLSTM网络模型的预测结果进行对比分析。结果表明,Bo-BiLSTM网络的模型拟合精度更高,基于Bo-BiLSTM网络的IGBT老化失效预测方法具有较好的预测效果,能够应用于IGBT的失效预测。

结合注意力机制和Mengzi模型的短文本分类

如何使用短文本分类技术挖掘有用的文本信息,是当前热门的研究方向之一。为了解决短文本特征信息稀疏和特征信息难以提取的问题,提出一种Mengzi-ADCBU短文本分类模型,该模型利用Mengzi预训练模型将输入的文本信息转化为相应的文本表示,再将获得的文本向量分别输入改进的深度金字塔卷积神经网络和融合了多头注意力机制的双向门控单元中提取文本特征信息,将两者提取到的特征信息进行融合之后,输送给全连接层和Softmax函数完成短文本分类。在公开的短文本数据集THUCNews和SougouCS上分别进行多组模型对比实验,实验结果表明本文提出的MengziADCBU模型在短文本分类的准确率、精确度、召回率和F1值等评价指标上都比现在的主流模型性能更优,具有较好的短文本分类能力。

基于注意力机制的CNN-BiLSTM的IGBT剩余使用寿命预测

针对绝缘栅双极型晶体管(IGBT)可靠性问题,提出了一种融合卷积神经网络(CNN)、双向长短期记忆(BiLSTM)网络和注意力机制的剩余使用寿命(RUL)预测模型,可用于IGBT的寿命预测。模型中使用CNN提取特征参数,BiLSTM提取时序信息,注意力机制加权处理特征参数。使用IGBT加速老化数据集对提出的模型进行验证。结果表明,对比自回归差分移动平均(ARIMA)、长短期记忆(LSTM)、多层LSTM(Multi-LSTM)、 BiLSTM预测模型,在均方根误差和决定系数等评价指标方面该模型的性能最优。验证了提出的寿命预测模型对IGBT失效预测是有效的。

针对光伏发电功率预测的LSTformer模型

为了提高光伏发电功率预测精度,提出了一种基于长短期时序数据融合的Transformer生成式预测模型:LSTformer,能准确有效地预测光伏发电功率。LSTformer创新性地提出了时序分析模块(time series analysis,TSA)、时序特征融合模块(time series feature fusion,TSFF)和多周期嵌入模块(cycleEmbed),利用数据融合解决难以提取多时间尺度时序特征问题。设计时间卷积前馈(time convolution feedforward,TCNforward)单元,在编解码的过程中进一步提取时序特征。利用某光伏电站实际历史发电数据,通过实验验证LSTformer模型在光伏发电功率预测领域得到最低的均方误差(mean squared error,MSE)、平均绝对误差(mean absolute error,MAE),并通过消融实验验证了各模块的有效性。