纳米SONOS存储器件多值多位存储技术研究

发布时间：2018-01-12 13:09

本文关键词：纳米SONOS存储器件多值多位存储技术研究　出处：《南京大学》2013年硕士论文　论文类型：学位论文

【摘要】：如今非挥发性快闪存储器在人们生活工作的各个领域都发挥着巨大的作用,其中局部俘获型多晶硅-氧化硅-氮化硅-氧化硅-硅(SONOS)存储器因为可以通过多值／多位技术实现高密度存储,而受到人们的广泛的重视和深入的研究。但是随着器件尺寸不断缩小,当工艺节点缩小到纳米量级后,在SONOS存储器中实现多值多位存储面临着诸多挑战。例如由于在编程擦除过程中注入的电子和空穴的空间分布不匹配性加速了器件耐受特性的退化；在纳米SONOS存储器中分别在源极与漏极实现两位存储时,这两位相互间的干扰也变得更严重；同时人们对于存储器在保持特性中电荷的主要流失机制也一直存在争论。本篇论文主要针对在纳米SONOS存储器中出现的这些问题,通过对比传统的沟道热电子注入编程、脉冲激发衬底热电子注入编程以及改进的衬底加正偏压沟道热电子注入编程,这三种局部注入编程方法在90nm SONOS存储器中的耐受特性与保持特性,并利用测量电荷泵电流等表征方法来分析它们对纳米SONOS存储器多值多位存储特性的提高。在耐受性实验中,发现相对于传统的沟道热电子注入编程,另外两种编程方法都有明显改善,其中使用改进的衬底加正偏压沟道热电子注入编程方法的样品耐受特性表现最好。这与用测量电荷泵电流等方法表征出其中残余电荷最少的结果相一致,说明改进的衬底加正偏压沟道热电子注入编程方法有效地抑制了编程中二次离子注入,减少了存储层中因为注入电荷分布不匹配引起的耐受特性退化。同时电子注入范围变窄还抑制了第二位比特效应,我们将这种编程方法应用于4Bit/4Level操作中,发现样品在经过10K次反复编程擦除后仍有着足够大的编程窗口,并且将保持特性外推十年后仍然有足够大的读取窗口。同时这种编程方法还避免了衬底与漏极之间的PN结正偏降低了功耗,并且与产品编程中递增步长脉冲编程方法相兼容。为研究器件在保持过程中主要的电荷流失机制,我们对比经过三种不同编程方法10K次编程擦除操作后样品的保持特性,因为这三种编程方法中电子注入的空间分布不同,那么经过反复编程擦除操作后必然会造成存储层中残余电荷分布不同。假设电荷横向迁移是主要流失机制,那么经历长时间后,样品的阈值漂移必然会有差别。但最终实验结果中三者并没有明显的差异,而且对比在保持特性实验前后三组样品电荷泵电流曲线也没有明显的漂移,这与横向迁移模型明显矛盾。因此相对电荷横向迁移,纵向电荷丢失对常温下存储器保持特性影响更大。
[Abstract]:Nowadays, non-volatile flash memory plays an important role in every field of people's life and work. The locally trapped polysilicon-silicon oxide-silicon nitride-silicon oxide-silicon oxide (SONOS) memory can be used to achieve high density storage through multi-value / multi-bit technology. However, with the continuous reduction of device size, when the process node is reduced to nanometer order of magnitude. The realization of multi-valued multi-bit memory in SONOS memory faces many challenges. For example, due to the mismatch between the spatial distribution of the injected electrons and holes in the programming erasure process, the degradation of the device's tolerance is accelerated. The interference between the two bits becomes more serious when the two bits are stored in the source and drain respectively in the nanometer SONOS memory. At the same time, there has been a debate about the main mechanism of charge loss in memory retention characteristics. This paper mainly aims at these problems in nanocrystalline SONOS memory by comparing the traditional channel hot electron injection programming. Pulse excited substrate hot electron injection programming and improved substrate plus positive bias channel hot electron injection programming. The tolerance and retention characteristics of these three local injection programming methods in 90nm SONOS memory. The methods of measuring charge pump current and other characterization methods are used to analyze the improvement of the multi-valued multi-bit memory characteristics of nanocrystalline SONOS memory. In the tolerance experiment, compared with the traditional channel hot electron injection programming, the other two programming methods are obviously improved. The improved substrate plus positive bias channel hot electron injection programming method has the best resistance performance, which is consistent with the measurement of charge pump current and other methods to characterize the least residual charge. It is shown that the improved substrate bias channel hot electron implantation method can effectively suppress the secondary ion implantation in the programming. At the same time, the narrow range of electron injection also suppresses the second bit effect. We applied this programming method to the 4bit / 4 level operation and found that the sample still had a large enough programming window after 10 K repeated programming erasures. After extrapolating the retention characteristics for ten years, there are still large read windows. Meanwhile, this programming method also avoids the PN junction between the substrate and the drain and reduces the power consumption. And it is compatible with the incremental step pulse programming method in product programming. In order to study the main mechanism of charge loss during the device retention, we compared the retention characteristics of the sample after three different programming methods after 10K programming erasure operation. Because the spatial distribution of electron injection in these three programming methods is different, the residual charge distribution in the storage layer will be different after repeated programming erasure operations. It is assumed that the transverse charge migration is the main loss mechanism. After a long period of time, the threshold shift of the sample will be different, but there is no significant difference among the final experimental results. In addition, there is no obvious drift in the current curve of the three groups of samples before and after the retention characteristic experiment, which contradicts the transverse migration model obviously. Therefore, the relative charge transversal migration is not obvious. The longitudinal charge loss has a greater effect on the memory retention characteristics at room temperature.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333

【相似文献】