In this paper, a new pre-alignment approach based on Four-Quadrant-Photo-Detector （FQPD） for IC mask is presented. The voltage outputs from FQPDs are the functions of alignment mark＇s position offsets with respect ...In this paper, a new pre-alignment approach based on Four-Quadrant-Photo-Detector （FQPD） for IC mask is presented. The voltage outputs from FQPDs are the functions of alignment mark＇s position offsets with respect to FQPDs. The functions are obtained with least squares error （LSE）-based polynomial fitting after the normalization of experimental data. As the acquired functions are not monotonic about their variables, the alignment mark＇s position offset cannot be given by direct inverse operation on the obtained functions. However, the piecewise polynomial fitting gives the inverse function, with which the alignment mark＇s position offset can be predicted according to the voltage outputs of FQPDs. On the basis of prediction, a pre-alignment control strategy is proposed. The feasibility and robustness of the pre-alignment approach is shown by experiments. Furthermore, the results demonstrate that the maximum error of mask＇s position offset in the X- and Y- directions is less than 15μm after coarse pre-alignment. Keywords： Four-Quadrant-Photo-Detector （FQPD）, pre-alignment, IC mask, polynomial fitting展开更多
近年来,随着以ChatGPT为代表的大语言模型(Large Language Models,LLMs)在通用人工智能方向上取得突破性进展,国内外掀起了大模型应用的研究热潮。人类获取和处理信息的方式往往涉及视觉、听觉、文本等多种模态,单纯依赖文本的语言模型...近年来,随着以ChatGPT为代表的大语言模型(Large Language Models,LLMs)在通用人工智能方向上取得突破性进展,国内外掀起了大模型应用的研究热潮。人类获取和处理信息的方式往往涉及视觉、听觉、文本等多种模态,单纯依赖文本的语言模型难以充分理解和表达复杂的现实世界信息。因此,研究者开始探索将LLMs扩展到多模态领域,通过统一建模文本、图像、视频等不同类型的数据,构建具有跨模态理解能力的多模态大模型(Multimodal Large Models,MLMs)。对MLMs的发展现状进行了全面梳理,重点介绍了当前主流的模型架构、训练策略以及评估方法,并分析了该领域面临的挑战和未来发展方向。随着模型参数规模和训练数据的大幅扩展,MLMs在跨模态任务中的性能显著超越了传统方法,为通用人工智能的发展奠定了重要基础。这些模型在视觉问答(Visual Question Answering,VQA)、图像描述、多模态对话等典型任务中展现出卓越的理解与生成能力。然而,当前MLMs仍存在长序列处理效率、计算资源需求以及模型可靠性等方面的技术瓶颈。未来研究将致力于在保持模型性能的前提下提升计算效率,并推动技术从通用框架向领域专用解决方案的转化,为通用人工智能的实现和产业智能化转型提供关键技术基础。展开更多
漂移管直线加速器(Drift Tube Linac,DTL)是中国散裂中子源(China Spallation Neutron Source,CSNS)直线加速器的主要部分,负责将脉冲流强为15 m A的负氢离子从3 Me V加速到80 Me V,再注入到快循环同步加速器(Rapid Cycling Synchrotron...漂移管直线加速器(Drift Tube Linac,DTL)是中国散裂中子源(China Spallation Neutron Source,CSNS)直线加速器的主要部分,负责将脉冲流强为15 m A的负氢离子从3 Me V加速到80 Me V,再注入到快循环同步加速器(Rapid Cycling Synchrotron,RCS)中实现进一步加速。DTL加速器本身技术工艺复杂,要求极高的加工精度和准直安装精度,是CSNS的关键技术之一。本文介绍了中国散裂中子源漂移管的预准直方法,从最初的预研到正式安装,解决了一系列难题,包括漂移管的标定、安装和准直调整,形成一整套流水线式的预准直流程,最终漂移管预准直的精度优于物理设计指标,可为同类别的准直测量提供参考。展开更多
基金This work was supported by National High Technology Research and Development Program of PRC (No. 2002AA420040)National 973 Program of PRC (No. 2002CB312200).
文摘In this paper, a new pre-alignment approach based on Four-Quadrant-Photo-Detector （FQPD） for IC mask is presented. The voltage outputs from FQPDs are the functions of alignment mark＇s position offsets with respect to FQPDs. The functions are obtained with least squares error （LSE）-based polynomial fitting after the normalization of experimental data. As the acquired functions are not monotonic about their variables, the alignment mark＇s position offset cannot be given by direct inverse operation on the obtained functions. However, the piecewise polynomial fitting gives the inverse function, with which the alignment mark＇s position offset can be predicted according to the voltage outputs of FQPDs. On the basis of prediction, a pre-alignment control strategy is proposed. The feasibility and robustness of the pre-alignment approach is shown by experiments. Furthermore, the results demonstrate that the maximum error of mask＇s position offset in the X- and Y- directions is less than 15μm after coarse pre-alignment. Keywords： Four-Quadrant-Photo-Detector （FQPD）, pre-alignment, IC mask, polynomial fitting
文摘近年来,随着以ChatGPT为代表的大语言模型(Large Language Models,LLMs)在通用人工智能方向上取得突破性进展,国内外掀起了大模型应用的研究热潮。人类获取和处理信息的方式往往涉及视觉、听觉、文本等多种模态,单纯依赖文本的语言模型难以充分理解和表达复杂的现实世界信息。因此,研究者开始探索将LLMs扩展到多模态领域,通过统一建模文本、图像、视频等不同类型的数据,构建具有跨模态理解能力的多模态大模型(Multimodal Large Models,MLMs)。对MLMs的发展现状进行了全面梳理,重点介绍了当前主流的模型架构、训练策略以及评估方法,并分析了该领域面临的挑战和未来发展方向。随着模型参数规模和训练数据的大幅扩展,MLMs在跨模态任务中的性能显著超越了传统方法,为通用人工智能的发展奠定了重要基础。这些模型在视觉问答(Visual Question Answering,VQA)、图像描述、多模态对话等典型任务中展现出卓越的理解与生成能力。然而,当前MLMs仍存在长序列处理效率、计算资源需求以及模型可靠性等方面的技术瓶颈。未来研究将致力于在保持模型性能的前提下提升计算效率,并推动技术从通用框架向领域专用解决方案的转化,为通用人工智能的实现和产业智能化转型提供关键技术基础。
文摘漂移管直线加速器(Drift Tube Linac,DTL)是中国散裂中子源(China Spallation Neutron Source,CSNS)直线加速器的主要部分,负责将脉冲流强为15 m A的负氢离子从3 Me V加速到80 Me V,再注入到快循环同步加速器(Rapid Cycling Synchrotron,RCS)中实现进一步加速。DTL加速器本身技术工艺复杂,要求极高的加工精度和准直安装精度,是CSNS的关键技术之一。本文介绍了中国散裂中子源漂移管的预准直方法,从最初的预研到正式安装,解决了一系列难题,包括漂移管的标定、安装和准直调整,形成一整套流水线式的预准直流程,最终漂移管预准直的精度优于物理设计指标,可为同类别的准直测量提供参考。
