【作者】 项天；

【导师】 林子敬； 金西；

【作者基本信息】 中国科学技术大学 ， 微电子学与固体电子学， 2014， 博士

【摘要】 数据处理包括对数据的采集、存储、检索、加工、变换和传输。随着信息科学技术的飞速发展,计算机已经从初期的计算工具发展成集接收、存储、发布于一体、与人类社会发展密不可分的工具。随着对物质世界及自身探索的逐步深入,人类对信息处理系统的需求也逐步提高。在数据接收平台中,探测物理量动态范围要求越来越大,时间测量精度要求越来越高；在数据存储平台中,存储能力要求越来越高,处理方式要求越来越灵活；在数据处理平台中,人们希望更小的系统能实现以往复杂系统才能实现的功能。真实世界的物理量经过采集过程转化为模拟信号,模拟信号是在幅度和相位上都连续的电信号。为了利用数字计算机处理模拟信号中承载的信息,必须将模拟信号数字化。由于模拟信号易受干扰,当前数据采集系统尽可能地保证信号的数字化靠近采集事件的发生处,即数据采集及数字化的过程在空间上是靠近的。本论文将数据采集及数字化发生的场所定义为现场。从空间分布来讲,数据处理单元与数据采集单元可以位于同一块芯片内、同一块PCB板中或相同的机箱内,它们之间通过总线或其它简单的数据传输协议实现通信；它们也可能相距甚远,需要通过网络协议等方式进行通信。从时间分布来讲,数据处理单元所处理的数据可以是实时采集得到的；也可以是从存储单元读取、在较远的过去采集的。本论文将空间分布较小、时间分布较近的数据采集和处理过程定义为数据的现场处理。为了实现高效能的现场数据处理,需要对系统的各个单元进行优化,包括数据采集单元、电子学系统及计算系统。本文中所针对的复杂数据,就是指处理它的系统必须经过上述优化才能高效工作的数据。本论文系统地研究了复杂数据的现场处理阶段中接收、存储、发布技术的前沿理论和实现的关键技术,提出了大动态范围光电转换器(光电倍增管)的标定、高精度时间-数字转换器(TDC)的设计、分布式海量数据采集系统的数据存储转发平台设计和现场图形处理加速平台解决方案。本论文的主要研究工作和创新点包括：(1)研究了PMT(光电倍增管)的原理及读出方法,针对大动态范围BGO(锗酸铋)量能器读出电子学的标定和线性度分析展开深入研究。BGO量能器中单根晶体希望探测的粒子能量达到2TeV量级,且动态范围可达105倍量级,用地面加速器不能实现完全的标定。因此设计了一种大动态范围荧光粒子模拟器,该模拟器可以真实地模拟BGO晶体发光,且动态范围达4.11×105倍,已用于暗物质粒子探测卫星PMT地面检测及标定。(2)研究了时间-数字转换的原理及实现方法,针对现有TDC(时间-数字转换器)精度-资源占用量的矛盾,采用Multi-Phase Clock架构,利用FPGA(现场可编辑逻辑门阵列)IO逻辑中的ISERDES模块,设计了时间精度达56ps的高精度TDC。被测信号通过IO管脚输入到Xilinx FPGA,通过IO逻辑中的IBUFDS_DIFF_OUT模块,产生2路互补输出,并作为输入连接到相邻的2个ISERDES模块。ISERDES模块被配置为oversample模式,相当于4相位的串并转换器,两个ISERDES形成8相位采样器,相当于采用8倍采样时钟来采样被测信号。由于专用器件设计紧凑,以达到高时钟速率。在Xilinx Artix-7XC7A100T-1型号FPGA中,采样时钟高达800MHz。且该结构占用逻辑极少,适于集成大量测量通道。(3)针对分布式数据采集节点的数据存储转发需求,提出了现场数据传输原型设计。某型高能粒子探测器阵列中安置了3600个PMT,由400个FEE(前端电子学)平台负责读出数据。由于采用了分布式数字化和无触发架构,FEE需要将全部信号数字化,并将数据传输到主控系统。本原型设计提出了SoC系统架构模型、数据存储模型,以及基于无操作系统的软件传输模型,设计调试了适用于现场数据处理的原型验证板。(4)研究了计算机系统发展的规律,提出计算机系统沿着数据处理位置由远端向现场、信息交互能力从松散到紧密的发展规律。提出可穿戴设备为数据处理现场化的未来发展方向。总结了处理器和图形处理器的技术发展路线,针对增强现实技术在现场数据处理系统的需求,提出了处理器系统的解决方案,并针对图形任务提出了实现案例。针对国家专用集成电路设设计了支持2维图形加速的图形处理引擎(GPE)原型。更多还原

【Abstract】 The data processing includes data reception, storage, conversion and dis-tribution. With the rapid development of information science and technology, computers have evolved from the early days computing tools into an inseparable partner of the human and society. With the deepening of exploration of the ma-terial world and the human themselves, the demand for information processing system is also gradually increased. On the data acquisition platform, the require-ment of dynamic range of the charge measurement and the accuracy of the time measurement continue to increase; On the data storage platform, requiring higher storage capacity, and more flexible ways to store data are required; On the data processing platform, people want a smaller system which can achieve the similar performance comparing with the complex systems in old ways.The real physical world through the acquisition process into analog signals, analog signals are continuous on both the amplitude and phase of the electrical signal. To take advantage of computer processing analog signals carrying in-formation, the analog signals must be digitized. Since the analog signal is easily interfered, the data acquisition system will now digitized signal as close as possible the occurrence of the event acquisition, data acquisition and digitization process is spatially close. This paper defines On-Site as the place of data collection and digitalization.In terms of the spatial distribution of the data processing unit and the data acquisition unit may be located within a same chip, a same PCB board or block within the same chassis, the communication between them via a bus or other simple data transfer protocol; They may also be far from each other, the commu-nication between them via network protocols. In terms of the time distribution, the data processed by the data processing unit may be obtained in real-time ac- quisition, or read from the storage unit, in the distant past collection. In this paper, we define on-site data processing as the data acquisition and processing procedure within small spatial distribution and close time distribution.To achieve high performance on-site data processing, the system needs to be optimized for each cell, including the data acquisition unit, the electronic system and the computing system. In this thesis, the complex data refers to the data which its processing system must be optimized in order to work efficiently.This thesis systematically studied the on-site processing of frontier theory and key technology implementation of complex data in the stage of receiving, s-toring, transforming, publishing, proposed large dynamic range media converter (photomultiplier tube) calibration, precision time to digital convertor (TDC) de-sign, data store and forward platform design in the distributed data acquisition system, and on-site graphics acceleration platform solution.The main research work and innovation of this thesis include:(1)This thesis studied the principle and the reading method of the PMT (pho-tomultiplier tube), focus on the large dynamic range BGO calorimeter calibration and linearity analysis of readout electronics system. Since BGO calorimeter in a single crystal particle energy exploration hope to reach2TeV magnitude with dynamic range up to105times, which is difficult of calibration by the artificial accelerator. Therefore a large dynamic range fluorescent particle simulator had been designed, the simulator can simulate real BGO crystal fluorescent, and the dynamic range of up to4.11x105times, where has been used for ground testing and calibration of PMT.(2)This thesis studied the principle and implementation method of time to digital convertor with the Multi-Phase Clock architecture, using the ISERDES module of the IO logic in the FPGA(Field programmable logic gate array), which archives a56-ps accuracy. The signal input through the IO pin to the Xilinx FPGA, through the IO logic IBUFDS_DIFF_OUT modules produce two com-plementary outputs, and as an input connected to an adjacent two ISERDES mod-ules. ISERDES module is configured to oversample mode, equivalent to4-phase serial-parallel convertor, while2-ISERDES equivalent to8-phase serial-parallel convertor. Dedicated devices such as ISERDES can achieve high clock rates. In the Xilinx Artix-7XC7A100T-1model FPGA, the sampling clock up to800MHz. The structure and logic rarely occupied, suitable for a large number of measuring channels integrated.(3)Forwarding demand for data storage nodes in distributed data acquisition, on-site data transmission prototype was proposed. A certain type of detector array places3600PMT, and400FEE(front-end electronics) platform for reading data. Hance distributed digitalization and non-trigger architecture has been used, FEE needs to digitalize all the signals and transfer them to the host system. I proposed SoC prototyping system architecture model, data storage model, and software-based transmission model no operating system, and design prototyping board.(4)This thesis studied the laws of computer systems development, proposed with the development of computer systems, data processing position from remote location to on-site, information exchange capability from loose to tight, and pro-posed that the future direction of development of on-site of data processing is the wearable device. This thesis summarizes the technology roadmap for proces-sors and graphics processors. For the requirements of applying augmented reality technology in the on-site data processing system, this thesis proposed processor system solution, and a implementation scheme for graphic application.更多还原