【作者】 王珊；

【导师】 吴华强；

【作者基本信息】 清华大学 ， 集成电路工程（专业学位）， 2018， 硕士

【摘要】 当今时代,半导体存储器的需求量越来越大。传统的flash存储器面临着尺寸缩小特性的物理瓶颈,随着器件尺寸的减小,串扰问题严重及电荷保持性能下降等问题凸显出来,发展新型的存储技术成为必然趋势。阻变存储技术具有结构简单、与CMOS工艺兼容、良好的尺寸缩小特性、操作电压低、功耗低、擦写速度快、易于3D集成等优点,成为下一代新型非易失性存储器的有力竞争者。然而,阻变存储器作为大规模存储应用时也面临着许多问题,如器件在小电流下工作时操作电压和高低阻值均一性差、采用多次校验(verify)的操作方法时擦写速度慢等问题。可以从多个角度改善这些问题,如材料体系、制备工艺及操作方式等。针对这些问题,本文主要从操作方式上优化阻变存储器的均一性和擦写速度。本文对TiN/HfO_x/TaO_x/TiN材料体系的阻变存储器件和阵列进行了操作方式的优化。为了优化阻变存储器在小电流下工作时的均一性,本文首先提出了电流和电压共同调节的verify操作方式。电流和电压共同调节的verify操作方式与传统的脉冲幅度按步长递增操作方式(incremental step pulse programming,ISPP)的不同之处在于,传统的ISPP操作方式通过改变操作电压单方面来控制导电细丝的通断,在电流和电压共同调节的verify操作方式中,电压产生的电场和电流限流(current compliance,CC)共同控制阻变存储器导电细丝的通断。基于电流和电压共同调节的verify操作方式,首先,大大提高了阻变存储器高低阻值的均一性。此外,得到了7个不同的阻态,提高了阻变存储器的多值存储能力。擦写速度和均一性之间需要折中,因此在提高擦写速度的同时需要兼顾均一性。因此,为了提高阻变存储器的擦写速度,本文提出了擦写速度和均一性协同优化的verify操作方式。速度和均一性协同优化的verify操作方式主要从三方面进行改善,分别是施加多个脉冲后校验一次、适当增加脉冲幅度及步长、改变脉冲宽度。首先,在三种不同脉宽下的测试结果表明,多个短脉冲要优于相同长度的一个长脉冲的作用效果。此外,该操作方式大大提高了阻变存储器的擦写速度,对TiN/HfO_x/TaO_x/TiN材料体系的存储器来说,在10ns脉冲的激励下,SET过程最多需要3个脉冲,过编程比率(over programmed bit rate)为7%;RESET过程最多需要4个脉冲,实现了擦写速度和均一性之间的trade off。更多还原

【Abstract】 Nowadays,the demand for semiconductor memory is increasing.The traditional flash memory faces the physical bottleneck of the size scalability.As the size of the device decreases,the problem of crosstalk and the decline of the charge retention performance are highlighted.Therefore,it is an inevitable trend to develop a new type of memory technology.The resistive random access memory has the advantages of simple structure,compatibility with CMOS process,excellent scalability,low operation voltage,low power consumption,fast switching speed,feasibility for 3D integration and so on.It will become a strong competitor of the next generation of new non-volatile memory.However,resistive random access memories are also faced with many problems when used as large-scale storage applications,such as the uniformity of operating voltage and resistance is poor and slow switching speed when using verify operation methods.We can improve these issues from multiple perspectives,such as material systems,manufacturing processes,and operating methods.To solve these problems,this paper mainly optimizes the uniformity and switching speed from the operating methods.In this paper,the operating methods of the single device and array with TiN/HfO_x/TaO_x/TiN material systems are optimized.In order to optimize the uniformity of the resistive memory under low current,we first proposed a current and voltage joint verification strategy.The current and voltage joint verification strategy is different from the traditional incremental step pulse programming method.The traditional ISPP method control the conductive filaments only by changing the voltage.In the current and voltage joint verification strategy,the electric field and current compliance(current compliance,CC)jointly control the conductive filaments of the resistive random access memory,therefore,it greatly improves the uniformity of the resistance.Based on the current and voltage joint verification strategy,7 different resistance states are obtained,which improves the multilevel storage capability of the resistive random access memory.Since there is tradeoff between switching speed and uniformity,we need to consider uniformity when improving switching speed.Therefore,in order to improve the switching speed of the resistive random access memory,we proposed a verify operation method in which the switching speed and uniformity are cooperatively optimized.The new verification method is mainly improved from three aspects,namely,the verification number,the pulse amplitude and the step,and the pulse width.The test results under three kinds of different pulse widths show that multiple short pulses are better than a long pulse with the same length.For the memory of the TiN/HfO_x/TaO_x/TiN material system,the SET process needs 3 pulses at most and the over programming bit rate is 7%under the excitation of the 10ns pulse.The RESET process needs 4 pulses at most,which improves the switching speed of the resistive random access memory greatly,and the tradeoff between the switching speed and uniformity is achieved.更多还原