Semiconductor memory, which is an electronic device used for data storage, becomes a key component in today’s electronic systems. The dominant memory technologies of the present time are Dynamic Random Access Memory (DRAM) and Flash. Both these memories store data as electronic charge and are known to suffer from further scaling issues [1–. The DRAM, which is a volatile memory, offers extremely long endurance (~1014 program/erase cycles) and fast (<10 ns) switching operation [3. On the other hand, Flash is a nonvolatile memory, but possesses limited endurance of typically 104–105 cycles and slow program/erase time of (μs up to ms) [2–.
To overcome the scaling and operational limitations of these conventional memories, various next-generation memory technologies have been proposed and are being intensively investigated. Some of these memories are: Ferroelectric Random Access Memory (FeRAM) [4–, Spin Transfer Torque-Magnetic Random Access Memory (STTMRAM) [7–, Phase-change Random Access Memory (PRAM) [10–, and Resistive Random Access Memory (ReRAM) [13–. One of the basic differences between these next-generation memories and conventional memories is in the way these two types of memories store data; the former stores the data as electronic charge whereas the latter as resistance. Among other next-generation memories, ReRAM is more promising owing to its simple Metal-Insulator-Metal (MIM) structure, excellent scalability (<10 nm), fast switching speed (<10 ns), low power operation, three-dimensional (3D) stackability and good complementary metal-oxide-semiconductor (CMOS) compatibility [13–.
2 Resistive random access memory (ReRAM)
Resistive switching was first observed by Hickmott in 1962 in binary oxides [23, but it was in early 2000s only when the resistive switching effect caught huge interest triggered by the search of an alternative memory technology [1524 27]. Generally, a ReRAM device consists of two electrically conducting metal electrodes and a sandwiched insulating switching layer making it a two-terminal passive device. Such a typical MIM structured ReRAM device is schematically shown in Figure 1(a). It was found that the resistance of such a device can reversely be switched between the high and low states simply by applying external bias across the MIM stack [13, 24, 25]. The electrode on which the external stimulus is applied is referred to as the top electrode (TE) and the other electrode, kept at electrical ground is called as the bottom electrode (BE) [Figure 1(a)]. Usually, prior to the reversible resistance switching, an electroforming (or forming) process is generally required in which a relatively higher bias is applied across a pristine device. After the forming process, the initial high resistance of the pristine device irreversibly changes to the low resistance [13–. A more elaborate discussion on the forming process and its mechanism can be found in the references [28–.
2.1 Resistive switching modes
The resistance of a ReRAM device can be switched between the high resistance state (HRS) and the low resistance state (LRS) by two different ways depending on the applied voltage polarity on the TE. When both the resistance states of HRS and LRS can be obtained by applying either positive voltage or negative voltage on TE, it is called unipolar mode switching [13. On the other hand, when both positive as well as negative polarity voltage is needed to switch the device between HRS and LRS, it is called bipolar mode switching as schematically shown in Figure 1(b) [13–. Although unipolar switching is beneficial in terms of requiring only single polarity voltage, which may minimize the circuit complexity, the resistive switching performance of unipolar device is commonly poorer compared to that of bipolar device. Therefore, in this chapter, we will discuss ReRAM with only bipolar mode switching. When a positive (or negative) sweep voltage is applied on TE, the current gradually increases until it reaches to a certain voltage at which the current abruptly increases and attains a predefined compliance current (IC) as shown in Figure 1(b). At this stage, the resistance of the device switches from HRS to LRS. This is called set process and the voltage at which the current suddenly increases is termed as set voltage (Vset). The IC is generally needed to avoid the permanent breakdown of the device. The device remains in the LRS state until applied negative (or positive) sweep voltage reaches beyond a certain value at which the current starts to decrease [Figure 1(b)]. This voltage is called reset voltage (Vreset) and the process is termed as reset process. After the reset process, the device switches to the HRS. As both the memory states of LRS and HRS retain their respective states even after the removal of applied voltage, ReRAM is a nonvolatile memory.
2.2 Resistive switching mechanism
The switching mechanism of ReRAM can be broadly classified as (i) filamentary type and (ii) interface type and is schematically shown in Figure 2(a) and (b), respectively [21. In the filamentary model, the resistance switching occurs owing to the formation and rupture of localized conductive filament (CF) in the sandwiched insulating layer also termed as switching layer upon the application of appropriate external electrical stimulus. The filamentary mechanism is schematically shown in Figure 2(a) [13, 14, 21]. As the CF can generate anywhere in the switching layer during bias application, the resistance switching in ReRAM is intrinsically random or stochastic. It is generally required that the switching material, especially the oxide, should have sub-stoichiometric composition in order to obtain stable resistance switching [21. During forming/set operation, the filamentary paths formed due to the anion (oxygen ion)/cation (Cu+ or Ag+) migration depend on the electrode and switching materials used and rupture during the reset process [Figure 2(a)]. The device in which switching occurs due to metallic (Cu or Ag) filament formation is called as conductive bridging random access memory (CBRAM) [14. Electrochemical migration of oxygen ions/cations and redox reaction near the metal/oxide interface is widely accepted as the possible mechanism of filament formation and rupture [14–. A more detailed discussion on ReRAM switching mechanism can be found in refs. [13, 14, 29, 31]. However, accurate universal switching model for ReRAM is still to be proposed. In addition, although studies involving high resolution transmission electron microscopy showed the CFs in different systems [32–, their in-situ clear visualization in the switching material have yet to be achieved. The lack of clear understanding of the ReRAM switching mechanism poses one of the biggest hindrances in its advancement towards commercial availability. On the other hand, in the interface type mechanism, the switching occurs at the interface of the metal and switching material owing to the movement of oxygen vacancies/ ions or trapping/de-trapping of electrons or holes under the applied electric field [Figure 2(b)] [21.
One of the main differences between the filamentary switching and interface type switching is that the later occurs at the whole area of the device in contrast to the former where switching is highly localized. This can be identified by measuring the device area dependence of both the resistance states of LRS and HRS of a resistive memory device. If, in general, both LRS and HRS vary with device area, the switching is interface type. In contrast, in filamentary switching, though HRS increases with decreasing device area, LRS remains unchanged with device area variation. Due to the highly localized switching, the filamentary ReRAM shows excellent scalability potential and the devices with <10 nm physical size have already been demonstrated [34–. A comprehensive discussion on ReRAM can be found in the refs. [13, 14, 19, 21, 22, 29, 37].
Although ReRAM shows excellent memory traits, it suffers from some drawbacks, such as its switching mechanism is not fully understood and it shows resistance variability (temporal as well as spatial) in both LRS and HRS [38. Figure 3(a) shows all the current–voltage (I–V) curves of 20 consecutive direct current switching cycles obtained from a ReRAM device with Ta/TaOx/Pt structure. As can be seen, the resistance of both LRS and HRS varies with the switching cycles. Owing to this variation, the overall memory window (HRS/LRS) decreases, which can be seen more clearly in the cumulative probability plot shown in Figure 3(b).
The observed variability is attributed to the intrinsic randomness in the CF formation and rupture during the switching events [39–. In other words, exactly same amount of filament cannot be formed or ruptured due to the stochastic nature of ReRAM switching mechanism, resulting in a cycle-to-cycle resistance variability. A more detailed discussion on resistance variability will be provided later in this chapter.
3 Multilevel per cell (MLC) storage
In order to reduce cost, the density of memory devices should be increased, which in turn minimizes the use of Si substrate area. One of the most obvious ways to increase the storage density is to decrease the physical size of the device to nanoscale dimensions. In this regard, ReRAM is very attractive as it shows excellent scalability potential with the architectural feasibility of the smallest possible cell area of 4F2 (F: feature size of the fabrication technology) in two-dimensional layout [41. ReRAM with <10 nm device size has already been reported with excellent memory performance [34–. Although the density can be increased by reducing the device size, this method requires complex experimental processes and is limited by patterning techniques as well. The other interesting approach is to stack the devices three-dimensionally (3D), that is, on top of each other vertically. For ReRAM, two kinds of feasible architectures of “crossbar” and “vertical ReRAM” are suggested [41–. Though both the architectures are appealing for future 3D integration, they also have patterning and integration complexities. Another alternative and simpler way to increase storage density is to use MLC storage technology in which more than one bit per cell can be stored without further decreasing the physical device size [43–. In principle, the storage of “n” bits per cell will result in “n” times (n×) increase in the storage density with 2n distinct memory levels, thereby scaling the Si die area with 1/n [46. However, the precise control over the resistance of the different resistance levels should be assured in order to achieve reliable MLC operation especially for ReRAM, which suffers from resistance variability and reliability issues mainly due to its intrinsic randomness in the switching process [39, 40, 47, 48]. By combining cell scalability, MLC, and 3D process architecture, ReRAM can have great potential for future ultrahigh density, nonvolatile memory applications [49. We will focus only on the MLC storage in ReRAM in this chapter.
4 MLC modes in ReRAM
One of the important traits of ReRAM, which makes it useful for high density application, is its MLC behavior. There are mainly three ways to obtain MLC characteristic: (i) changing compliance current, (ii) controlling reset voltage, and (iii) varying pulse width of program/erase operation.
4.1 MLC by changing compliance current
MLC characteristic in a ReRAM device with 1-ReRAM (1R) cell configuration can be obtained by changing the current compliance (IC) during “set” operation [44, 50 – 55]. In 1-Transistor 1-ReRAM (1T–1R) cell configuration, which is more practically viable, the current during “set” operation can be controlled by varying the applied voltage at the gate of the transistor [50–. The typical MLC I–V curves of a ReRAM device in TiN/Ti/HfOx/TiN structure in 1R cell configuration are shown in Figure 4(a) [56. When the IC is increased sequentially from 100 to 250 and 500 μA, the respective current of LRS (ILRS) also increases resulting in three different levels of LRS whereas the HRS current (IHRS) remains same for all the LRS levels. These three distinct LRS levels with the same HRS, leading to a total of 4-resistance levels can be used in 2-bit per cell storage and can enhance the storage density up to 2× higher as compared to a single level cell with the same Si die area. Recently, Prakash et al. have demonstrated 3-bit per cell storage by stack engineering in a TaOx-based ReRAM [52. Furthermore, it is observed that the maximum reset current (Ireset) also increases with IC whereas the set voltage almost remains unchanged [44–. The dependence of LRS resistance (RLRS) and Ireset on IC is shown in Figure 4(b) where RLRS and Ireset were plotted as a function of IC. Generally, the RLRS shows the universal dependence on IC when reset voltage is constant and can be well-fitted by an equation of the form RLRS = C/IC, where C is a constant, with the slope of “−1” [39–. However, different values of the slope other than the universal value of “−1” are also reported especially in the devices in which the reset voltage also varies with IC as in Figure 4(a) [50–. In contrast to the inverse dependence of RLRS on IC, the Ireset depends linearly on IC as can be seen in Figure 4(b).
The mechanism of MLC in IC control mode can be attributed to the formation and subsequent lateral widening of the CF with increasing IC as schematically shown in Figure 5 [52–. As the CF size (or diameter) increases, its resistance becomes smaller and hence, results multiple LRS levels. This argument is well-supported by the fact that the Ireset also increases with IC because it needs higher power to break the CF having larger diameter. Although MLC behavior in IC control mode is relatively easier to be obtained, it may have limited applications especially in passive cross-point array architecture owing to the difficulty in limiting the current by the “ on chip” circuit.
4.2 MLC by controlling reset voltage
The MLC characteristics can also be obtained by controlling the applied maximum reset voltage for “reset” operation [43, 44, 55, 58–61]. Figure 6(a) shows the typically observed I–V curves in semilog plot obtained from a HfOx-based ReRAM for three different reset voltages of −0.9, −1.1, and −1.3 V. As the applied voltage for the “ reset” operation increases, the HRS current (IHRS) decreases, leading to different resistance levels of HRS with same LRS resistance as shown in Figure 6(b). Further, in addition to HRS resistance, the “set” voltage (Vset) also increases with increasing reset voltage whereas Ireset remains almost constant [44–. Some studies have shown the possibility of 3-bit per cell storage in this mode of MLC operation by using modified program-verify technique already being used in Flash memory [59–.
The decrease in IHRS current is attributed to the increase in the gap between CF tip and metal electrode when higher reset voltage is applied as schematically illustrated in Figure 7 [43, 58, 61, 62]. It is also argued that the observed change in the resistance of HRS can be due to CF thinning (decrease in CF size) with increasing reset voltage, especially in the devices that show gradual reset current change, rather than abrupt during “reset” operation [63. This mode of MLC operation is more practically feasible especially for passive cross-point array in terms of comparatively less circuit complexity.
4.3 MLC by changing program/erase pulse width
Another suggested method to obtain MLC characteristics is to vary the program/erase pulse width while keeping the pulse amplitude constant [58–. Three distinct HRS resistance levels were successfully obtained by exponentially changing the reset pulse width from 50 ns to 5 μs in a HfOx-based ReRAM [58. The origin of the HRS resistance modulation and its equivalence with the reset voltage control scheme may be attributed to the universal voltage-time relationship in the switching dynamics of ReRAM where the ions migration velocity exhibit the hyper-sinusoidal dependence on the applied electric field [58. Although it is relatively easier to generate the program/erase pulses with varying widths, this scheme has the disadvantage of being energy inefficient [58. The comparison of the transient responses between reset pulse amplitude and pulse width control confirmed the higher energy consumption for the later scheme owing to the unwanted energy dissipation in the switching material as thermal energy [58.
5 Resistance variability and MLC operation
Although the multiple resistance levels can be easily obtained in ReRAM by the above-mentioned methods, the successful implementation of the MLC technology mainly depends on the ability to precisely control the resistance margin between the two resistance levels. There exist various factors that can hinder the MLC operation in ReRAM by degrading the resistance margin and eventually leading to its failure [48–. Therefore, enhanced understanding of the origin of these factors is required along with the search of the feasible approaches to minimize the effects. We will discuss about these factors and their effect on MLC margin more elaborately in the following subsections.
5.1 Cycle-to-cycle variability
One of the most critical factors that can degrade the margin between the two resistance levels and that can result in an erroneous MLC operation, especially in ReRAM, is the cycle-to-cycle (or temporal) switching variability [31, 48, 65–67]. Figure 8(a) shows the multiple switching I–V curves obtained from a HfOx- based device when operated in MLC mode by changing IC [56. The resistance distribution of all the 100 consecutive switching cycles for all the four resistance levels (three LRS and one HRS levels) are also shown in Figure 8(b) [56. The resistance margin is reduced owing to the cycle-to-cycle switching variability. The origin of the cycle-to-cycle variability at a particular switching current is mainly attributed to the dynamic change in the number of constituent oxygen vacancy defects present in the CF with each switching event because of the intrinsic stochastic nature of CF formation and rupture during switching in ReRAM [47, 66, 67]. This compositional variation in the conductive channel leads to the conductivity fluctuation. More fundamentally, it can be understood in terms of having different activation energy of defect migration due to the random location as well as different surroundings of the defects within the conductive channel [66. In addition, the variability becomes more severe when the IC is reduced [65–. It is observed that not only the variation in the LRS resistance increases, but also the variations in Ireset and reset voltage increase when the IC is decreased as shown in Figure 9 where the statistical analysis of RLRS, Ireset, and reset voltage as a function of IC is presented [56. The IC dependent resistance variability behavior in ReRAM can be related to the change in the total number of oxygen vacancy defects (N) contributing to CF formation with varying IC [39. Figure 10 shows the resistance–voltage (R–V) curves of LRS corresponding to the I–V curves of positive bias shown in Figure 8 and the schematic illustration of the CF defects change with IC. The smaller relative LRS resistance spread (ratio of standard deviation (σ) and average resistance (μ)) at higher IC (see Figures 9 and 10) is attributed to the increased number of defects present in the CF rendering well-defined path for current conduction. Balatti et al. have theoretically shown a quantitative relation between relative LRS resistance spread and IC from Poisson statistics, assuming ohmic conduction in CF [39. It is demonstrated that (σ/μ)LRS = (N)−0.5 is proportional to (IC)−0.5 [39. Various methods including materials engineering, different electrical methods and device operation conditions modification have been suggested to improve the cycle-to-cycle resistance variability in ReRAM [20, 62, 68–72].
5.2 Device-to-device variability
In addition to the cycle-to-cycle variability, ReRAM also exhibits device-to-device or spatial non uniformity in the switching parameters which additionally degrades the usable resistance margin between the memory states and hence, can hinder the MLC operation [15, 16, 38, 69, 73–76]. The device-to-device variability can originate from the nonuniformities in the fabrication processes such as variation in the switching film thickness, surface roughness of the electrodes, etching damages, etc., as well as the lack of precise control over the defect generation and filament formation during the electroformation step of a pristine device [73, 74, 77, 78]. Lee et al. have demonstrated significant improvement in the resistance distribution in an array by adopting chemical mechanical polish to smoothen the bottom electrode in HfOx-based device [73. Chen et al. suggested the application of ramping voltage instead of constant voltage across the gate in 1T–1R cell to improve device-to-device variations [15. Recently, Hayakawa et al. have reported improved resistance distribution in a nanoscale TaOx-based ReRAM by developing the fabrication process technologies, such as low-damage etching, cell side oxidation and proper cell encapsulation [78. On the other hand, one of the main origins of the electroformation-induced resistance variation is attributed to the overshoot current due to the parasitic effects[69, 76, 77]. The overshoot current that can be different for different devices leads to the uncontrolled defects generation and subsequent CF formation in the switching layer resulting in device-to-device resistance variation. To minimize this, various methods, such as constant voltage forming [77 and forming at higher temperature, so-called hot forming [69, are advocated, and improvement in the resistance distribution is reported.
5.3 Effect of operation temperature
In ReRAM, it was observed that depending on the programmed resistance state (LRS or HRS) and the operation current (low or high), the resistance of the different multilevel varies differently with changing the operation (ambient) temperature [45, 63, 79, 80]. The resistance of HRS generally decreases when the operation temperature is increased, whereas the LRS resistance can increase or decrease depending on the programming current[45, 63, 80]. Goux et al. have shown that the LRS resistance programmed at a high current (>1 mA) in an HfO2-based device increases with temperature, showing metallic-like behavior with negative activation energy [63, whereas Wei et al. observed that the LRS resistance of a TaOx-based device programmed at a relatively lower current (<100 μA) decreased with increasing temperature [79. Although the resistance of HRS decreases with increasing operation temperature, the degree (or slope) of the resistance change is found to be different for different HRS levels [45–. Owing to the different behavior of the multiple resistance levels on operation temperature, the resistance margin between the resistance levels deteriorates (with respect to room temperature) when the operation temperature is raised as shown in Figure 11, which may limit the temperature range of MLC operation. In addition, as the ambient temperature of the device can be increased during its operation, this can result in an erroneous MLC operation.
5.4 Effect of random telegraph noise (RTN)
Random telegraph noise (RTN) that can cause severe fluctuations in the read current during read operation of a programmed resistance level in a ReRAM cell leads to a reduction in the detectable margin between the resistance levels [43, 81–84]. The origin of the RTN in ReRAM is attributed to the capture and emission (trap and detrap) of electrons in the trap (more probably oxygen vacancy trap) near the filament in LRS state and in the tunneling gap in the HRS state[43, 81, 83]. It is also reported that the RTN can be governed by the oxygen vacancy perturbation in the conducting channel during switching [84. A typical two-step RTN observed in a Ta/ TaOx/Pt stacked device is shown in Figure 12. Ielmini et al. have shown a universal relation between the signal-to-noise ratio or relative amplitude (Δ R/R) of the RTN and the average resistance (R) of the memory state based on the size-dependent depletion of the carriers [81. The relative amplitude (Δ R/R) increases with average resistance of the programmed memory state. In other words, the RTN amplitude increases with decreasing the operation current and is higher for those multiple resistance levels that are programmed to the higher resistance. In addition, the HRS levels show higher RTN amplitude than the LRS levels [43. Therefore, additional resistance margin should be ensured between the two resistance levels especially with higher resistance in order to avoid the RTN effect and to achieve a reliable MLC operation.
5.5 Interstate switching variability
It was observed that the resistance of the memory levels can also vary when a device operating in the MLC mode is switched from one resistance level to the other especially from the highest resistance level to the lowest and vice versa. This can further reduce the margin between the resistance levels for the MLC operation. Figure 13 shows that the two values of LRS resistance corresponding to IC of 100 μA before and after increasing the IC to 200 μA are not the same [43. The variation can be due to the failure to form the same amount of filament to that of the previous switching event of the same resistance level in case of IC-controlled MLC mode or to generate exactly the same gap between the CF tip and the electrode in case of reset voltage control MLC scheme.
As the above discussed factors can degrade the usable resistance margin among the various multiple resistance levels as schematically shown in Figure 14 and hence, can limit the MLC implementation, proper care should be taken while designing the MLC storage in ReRAM.
6 Reliability of multiple resistance levels
6.1 Retention characteristics
ReRAM is a nonvolatile memory because it can retain the memory state (LRS or HRS) after being programmed or erased. The nonvolatile behavior is due to the fact that in ReRAM, once the filament is formed or ruptured, it remains as it is even when the electric field is removed. The retention time of LRS and HRS can be evaluated by monitoring the resistance of the respective resistance state with time by applying a small constant read voltage pulse with much smaller amplitude than that of the program/erase voltage pulse as shown in Figure 15. The retention loss is mainly due to the out diffusion/annihilation of the oxygen vacancies from the CF resulting CF dissolution in LRS while in HRS, the retention loss can be attributed to the oxygen vacancy accumulation/generation within the gap between CF and electrode [85–. The driving force for vacancy diffusion is the concentration gradient between the filament and bulk; and vacancy generation can be due to the electrical/thermal stress [85, 86, 88, 89]. The argument is supported by the fact that retention loss accelerates when ambient temperature is increased due to enhanced diffusion [85–87, 90]. In a previous study, the LRS retention loss was remarkably improved in a TaOx-based device operating at low current by generating the CF having higher vacancy concentration or denser filament using a two-step forming technique [65–. In another study on HfOx-based ReRAM, significant improvement in the LRS retention at small current has been achieved by limiting the oxygen diffusion with post-annealing the ReRAM stack [90. It was also observed that stronger forming operation resulted in better retention [90. In addition, it is commonly observed that the retention time especially for LRS is a strong function of programming current and the retention loss can be severe when operating at a small current (<50 μA)[65, 90, 91]. This retention degradation dependence on programming current can be related to the total number of vacancies present in the CF and their diffusion/annihilation. As at higher IC, the CF is thicker and hence has higher number of vacancies, the retention time is longer. This means that in case of MLC operation, the resistance levels will have a different rate of retention loss depending on the IC with the worst data retention for the resistance level programmed at the lowest current level. This can limit the MLC operation in a way that the MLC retention time should be determined according to the retention time of the highest resistance LRS level, which may be significantly lower than the other resistance levels. This is also valid for reset voltage controlled MLC mode with the only exception that the lowest HRS level (obtained from the smallest rest voltage) exhibits the worst data retention [87.
The switching endurance, which tells how many times a memory device can be switched between the resistance states without degradation is one of the most important performance metrics of a memory device. In ReRAM, an endurance of over 1012 cycles in a bilayer TaOx-based device has been demonstrated [92. The endurance is known to be highly dependent on the program/erase conditions of a resistive memory device [93– and various types of endurance failure mechanisms are discussed [95–. The existence of a trade-off relation between the endurance and retention characteristics in ReRAM has been reported [98. Figure 16 shows reset voltage-dependent endurance of a HfSiOx-based device with constant pulse width of 1 μs [93. As the reset voltage is increased, the endurance degrades sharply due to the initiation of filament growth (negative set) from the bottom electrode [93–. In light of the above discussion, it can be said that the endurance of multiresistance levels obtained by controlling reset voltage may not be the same for all the resistance levels and hence the endurance of MLC operation will be limited by the endurance of the HRS level erased with the largest reset voltage. Similarly, it may also be possible that the endurance of the LRS state programmed with the largest set voltage will be the shortest due to the increased stress in the switching material. Therefore, the endurance of a single level cell is much higher than that of the multilevel cell.
7 Summary and future work
ReRAM is particularly appealing for ultrahigh density and low-cost, nonvolatile memory applications owing to its excellent cell scalability, MLC storage capability and 3D stackability. MLC technology is an alternative way to increase the storage density without the requirement of decreasing the physical device size or 3D stacking. The MLC storage in ReRAM along with the resistance variability and reliability of multiple resistance levels were discussed in detail in this chapter. Then, various MLC operation schemes and their physical mechanisms have been explored. A comprehensive analysis of variability of multiple resistance levels by identifying various factors such as cycle-to-cycle and device-to-device nonuniformity, RTN, effect of ambient temperature and interstate switching variability have been provided, which can help in estimating the safe resistance margin to ensure the successful MLC operation. The temporal and spatial switching variability contribute most in resistance margin degradation. Finally, the reliability characteristics such as retention and endurance of MLC levels and the impact on the MLC operation have been assessed. The retention and endurance of the MLC operation is limited by that resistance level which has the worst endurance/retention time.
Although comprehensive analysis on MLC storage is provided for a single ReRAM cell, a similar and practically useful analysis on the MLC characteristics and resistance variability can be performed in a cross-point array in 1selector-1ReRAM cell configuration. Furthermore, ways to combine the ReRAM cell physical salability, MLC characteristics, and 3D stackability should be researched in order to fully exploit the benefits of ReRAM for ultrahigh density and inexpensive nonvolatile memory applications.
Waser R. Nanotechnology: Volume 3: Information Technology. Wiley; 2008. Google Scholar
Keeney SN. A 130 nm generation high density Etox flash memory technology. Int Electron Devices Meet Tech Dig 2001. Google Scholar
Durlam M, Chung Y, DeHerrera M, et al. MRAM memory for embedded and stand alone systems. in: Proceedings 2007 IEEE International Conference on Integrated Circuit Design and Technology, ICICDT. 2007. pp. 75–8. Google Scholar
Parkin SSP. Spintronic materials and devices: Past, present and future!IEDM Tech Dig IEEE Int Electron Devices Meet 2004 2004.Google Scholar
Parkin SSP, Roche KP, Samant MG, et al. Exchange-biased magnetic tunnel junctions and application to nonvolatile magnetic random access memory (invited). J Appl Phys 1999;85(8):5828–33. CrossrefGoogle Scholar
Qiao B, Feng J, Lai Y, et al. Effects of Si doping on the structural and electrical properties of Ge2Sb2Te5 films for phase change random access memory. Appl Surf Sci 2006; 252(24):8404–9. CrossrefGoogle Scholar
Chen ACA, Haddad S, Wu Y-C, et al. Non-volatile resistive switching for advanced memory applications. IEEE Int Devices Meet 2005 IEDM Tech Dig 2005. Google Scholar
Wong HSP, Lee HY, Yu S, et al. Metal-oxide RRAM. in: Proceedings of the IEEE. 2012. pp. 1951–70.Google Scholar
Liu Q, Long S, Lv H, et al. Controllable growth of nanoscale conductive filaments in solid-electrolyte-based ReRAM by using a metal nanocrystal covered bottom electrode. ACS Nano 2010;4(10):6162–8. CrossrefPubMedGoogle Scholar
Baek IG, Lee MS, Seo S, et al. Highly scalable nonvolatile resistive memory using simple binary oxide driven by asymmetric unipolar voltage pulses. IEDM Tech Dig IEEE Int Electron Devices Meet 2004 2004. Google Scholar
Sim HSH, Choi HCH, Lee DLD, et al. Excellent resistance switching characteristics of Pt/SrTiO/ sub 3/schottky junction for multi-bit nonvolatile memory application. IEEE Int Devices Meet 2005 IEDM Tech Dig 2005.Google Scholar
Baek IG, Kim DC, Lee MJ, et al. Multi-layer cross-point binary oxide resistive memory (OxRRAM) for post-NAND storage application. IEEE Int Devices Meet 2005 IEDM Tech Dig 2005. Google Scholar
Ho C, Hsu CL, Chen CC, et al. 9nm half-pitch functional resistive memory cell with <1uA programming current using thermally oxidized sub-stoichiometric WOx film. in: Technical Digest – International Electron Devices Meeting, IEDM. 2010. Google Scholar
Govoreanu B, Kar GS, Chen YY, et al. 10×10nm2 Hf/HfOx crossbar resistive RAM with excellent performance, reliability and low-energy operation. in: Technical Digest – International Electron Devices Meeting, IEDM. 2011.Google Scholar
Park J, Lee W, Choe M, et al. Quantized conductive filament formed by limited Cu source in sub-5nm era. in: Technical Digest – International Electron Devices Meeting, IEDM. 2011. Google Scholar
Pan F, Gao S, Chen C, Song C, Zeng F. Recent progress in resistive random access memories: Materials, switching mechanisms, and performance. Mater Sci Eng R Reports 2014;83(1):1–59. CrossrefGoogle Scholar
Chen A, Lin MR. Variability of resistive switching memories and its impact on crossbar array performance. in: IEEE International Reliability Physics Symposium Proceedings.2011. Google Scholar
Fantini A, Goux L, Degraeve R, et al. Intrinsic Switching Variability in HfO2 RRAM. 2013;1–4. Google Scholar
Zhirnov VV., Cavin RK, Menzel S, et al. Memory devices: Energy-space-time tradeoffs. In: Proceedings of the IEEE. 2010. pp. 2185–200. Google Scholar
Park SG, Yang MK, Ju H, et al. A non-linear ReRAM cell with sub-1μA ultralow operating current for high density vertical resistive memory (VRRAM). in: Technical Digest – International Electron Devices Meeting, IEDM. 2012. Google Scholar
Terai M, Sakotsubo Y, Kotsuji S, Hada H. Resistance controllability of Ta2O5/TiO2 stack ReRAM for low-voltage and multilevel operation. IEEE Electron Device Lett 2010;31(3):204–06. CrossrefGoogle Scholar
Lee HY, Chen PS, Wu TY, et al. Low power and high speed bipolar switching with a thin reactive ti buffer layer in robust HfO2 based RRAM. in: Technical Digest – International Electron Devices Meeting, IEDM. 2008. Google Scholar
Chien WC, Chen YC, Chang KP, et al. Multi-level operation of fully CMOS compatible WOx resistive random access memory (RRAM). in: 2009 IEEE International Memory Workshop, IMW ‘09. 2009. Google Scholar
Yu S, Guan X, Wong HSP. On the stochastic nature of resistive switching in metal oxide RRAM: Physical modeling, Monte Carlo simulation, and experimental characterization. in: Technical Digest – International Electron Devices Meeting, IEDM. 2011. Google Scholar
Chen A, Lin MR. Electrical characterization of resistive switching memories. AIP Conf Proc 2011;1395:139–47. Google Scholar
Ielmini D. Resistive-switching memory. in: Wiley Encyclopedia of Electrical and Electronics Engineering. John Wiley & Sons, Inc; 2014. pp. 1–32. Google Scholar
Wu M-C, Jang W-Y, Lin C-H, Tseng T-Y. A study on low-power, nanosecond operation and multilevel bipolar resistance switching in Ti/ZrO2/Pt nonvolatile memory with 1T1R architecture. Semicond Sci Technol 2012;27(6):065010. CrossrefGoogle Scholar
Long B, Li Y, Jha R. Switching characteristics of Ru/HfO2/TiO2-x/Ru RRAM devices for digital and analog nonvolatile memory applications. IEEE Electron Device Lett 2012;33(5):706–8.CrossrefGoogle Scholar
Prakash A, Park J, Song J, Woo J, Cha E, Hwang H. Demonstration of low power 3-bit multilevel cell characteristics in a TaOx-based RRAM by stack engineering. IEEE Electron Device Lett 2015;36(1):32–4. CrossrefGoogle Scholar
Russo U, Kamalanathan D, Ielmini D, Lacaita AL, Kozicki MN. Study of multilevel programming in Programmable Metallization Cell (PMC) memory. IEEE Trans Electron Devices 2009;56(5):1040–7. CrossrefGoogle Scholar
Prakash A, Maikap S, Banerjee W, Jana D, Lai C-S. Impact of electrically formed interfacial layer and improved memory characteristics of IrOx/high-κx/W structures containing AlOx, GdOx, HfOx, and TaOx switching materials. Nanoscale Res Lett 2013;8(1):379. CrossrefPubMedGoogle Scholar
Prakash A, Park J, Song J, et al. Multi-state resistance switching and variability analysis of HfOx based RRAM for ultrahigh density memory applications. in: Next-Generation Electronics (ISNE), 2015 International Symposium on. Taipei, Taiwan: IEEE; pp. 1–2. Google Scholar
Chien WC, Chen YR, Chen YC, et al. A forming-free WOx resistive memory using a novel self-aligned field enhancement feature with excellent reliability and scalability. in: Technical Digest – International Electron Devices Meeting, IEDM. 2010. Google Scholar
Lee SR, Kim YB, Chang M, et al. Multi-level switching of triple-layered TaOx RRAM with excellent reliability for storage class memory. in: Digest of Technical Papers – Symposium on VLSI Technology. 2012. pp. 71–2. Google Scholar
Prakash A, Maikap S, Lai CS, et al. Improvement of uniformity of resistive switching parameters by selecting the electroformation polarity in IrOx/TaOx/WOx/W structure. Jpn J Appl Phys 2012;51(4 PART 2). Google Scholar
Muraoka S, Osano K, Kanzawa Y, et al. Fast switching and long retention Fe-O ReRAM and its switching mechanism. in: Technical Digest – International Electron Devices Meeting, IEDM. 2007. pp. 779–82. Google Scholar
Ninomiya T, Takagi T, Wei Z, et al. Conductive filament scaling of TaOx bipolar ReRAM for long retention with low current operation. Dig Tech Pap – Symp VLSI Technol 2012;48(2011):73–4. Google Scholar
Ambrogio S, Member S, Balatti S, et al. Statistical fluctuations in HfOx resistivesSwitching memory: Part I – Set/reset variability. IEEE Trans Electron Devices. 2014;61(8):2912–9. CrossrefGoogle Scholar
Raghavan N, Degraeve R, Fantini A, et al. Stochastic variability of vacancy filament configuration in ultrathin dielectric RRAM and its impact on OFF-state reliability. Tech Dig – Int Electron Devices Meet IEDM 2013;21.1.1–21.1.4. Google Scholar
Banerjee W, Maikap S, Rahaman SZ, et al. Improved resistive switching memory characteristics using core-shell IrOx Nano-Dots in Al₂O₃/WOx Bilayer Structure. J Electrochem Soc 2012;159(2):H177. CrossrefGoogle Scholar
Butcher B, Bersuker G, Young-Fisher KG, et al. Hot forming to improve memory window and uniformity of low-power HfOx-based RRAMs. 2012 4th IEEE Int Mem Work IMW 2012 2012;(V):4–7. Google Scholar
Prakash A, Deleruyelle D, Song J, Bocquet M, Hwang H. Resistance controllability and variability improvement in a TaOx-based resistive memory for multilevel storage application. Appl Phys Lett 2015;106(23):233104.CrossrefGoogle Scholar
Lee HY, Chen YS, Chen PS, et al. Evidence and solution of over-RESET problem for HfOx based resistive memory with sub-ns switching speed and high endurance. in: Technical Digest – International Electron Devices Meeting, IEDM. 2010. Google Scholar
Yu S, Guan X, Wong H-SP. On the switching parameter variation of metal oxide RRAM – Part II: Model corroboration and device design strategy. Electron Devices, IEEE Trans 2012;59(4):1183–8. CrossrefGoogle Scholar
Yu S, Guan X, Wong H-SP. Understanding metal oxide RRAM current overshoot and reliability using Kinetic Monte Carlo simulation. in: Electron Devices Meeting (IEDM), 2012 IEEE International. 2012. pp. 26.1.1–26.1.4.Google Scholar
Kinoshita K, Tsunoda K, Sato Y, et al. Reduction in the reset current in a resistive random access memory consisting of NiOx brought about by reducing a parasitic capacitance. Appl Phys Lett 2008;93(3):1–4. Google Scholar
Kalantarian A, Bersuker G, Gilmer DC, et al. Controlling uniformity of RRAM characteristics through the forming process. IEEE Int Reliab Phys Symp Proc 2012;6C.4.1–6C.4.5. Google Scholar
Hayakawa Y, Himeno A, Yasuhara R, et al. Highly reliable TaOx ReRAM with centralized filament for 28-nm embedded application. in: VLSI Technology (VLSIT), 2015 Symposium on. Kyoto, Japan: pp. 12–3. Google Scholar
Wei Z, Takagi T, Kanzawa Y, et al. Demonstration of high-density ReRAM ensuring 10-year retention at 85°C based on a newly developed reliability model. in: Technical Digest – International Electron Devices Meeting, IEDM. 2011. Google Scholar
Wang Y, Liu Q, Long S, et al. Investigation of resistive switching in Cu-doped HfO2 thin film for multilevel non-volatile memory applications. Nanotechnology 2010; 21(4):045202. PubMedCrossrefGoogle Scholar
Ambrogio S, Member S, Balatti S, et al. Statistical fluctuations in HfOx resistive-switching memory : Part II – Random telegraph noise. IEEE Trans Electron Devices. 2014;61(8):2920–7. CrossrefGoogle Scholar
Veksler D, Bersuker G, Chakrabarti B, et al. Methodology for the statistical evaluation of the effect of random telegraph noise (RTN) on RRAM characteristics. Tech Dig IEDM 2012;(ii):9.6.1–9.6.4.Google Scholar
Puglisi FM, Pavan P, Larcher L, Padovani A. Analysis of RTN and cycling variability in HfO2 RRAM devices in LRS. in: Solid State Device Research Conference (ESSDERC), 2014 44th European. Venice: pp. 246–9. Google Scholar
Raghavan N, Degraeve R, Fantini A, et al. Modeling the impact of reset depth on vacancy-induced filament perturbations in HfO2 RRAM. IEEE Electron Device Lett 2013; 34(5):614–6. CrossrefGoogle Scholar
Yu S, Yin Chen Y, Guan X, Philip Wong H-S, Kittl J A. A Monte Carlo study of the low resistance state retention of HfOx based resistive switching memory. Appl Phys Lett 2012; 100(4):043507. CrossrefGoogle Scholar
Wei Z, Takagi T, Kanzawa Y, et al. Retention model for high-density ReRAM. in: 2012 4th IEEE International Memory Workshop, IMW 2012. Milan: 2012. pp. 1–4. Google Scholar
Chen YY, Komura M, Degraeve R, et al. Improvement of data retention in HfO2/Hf 1T1R RRAM cell under low operating current. in: Electron Devices Meeting (IEDM), 2013 IEEE International. Washington, DC: 2013. pp. 10.1.1–10.1.4. Google Scholar
Ninomiya T, Wei Z, Muraoka S, Yasuhara R, Katayama K, Takagi T. Conductive filament scaling of TaOx bipolar ReRAM for improving data retention under low operation current. IEEE Trans Electron Devices 2013;60(4):1384–9. CrossrefGoogle Scholar
Lee M-J, Lee CB, Lee D, et al. A fast, high-endurance and scalable non-volatile memory device made from asymmetric Ta2O(5-x)/TaO(2-x) bilayer structures. Nat Mater 2011; 10(8):625–30. CrossrefPubMedGoogle Scholar
Balatti S, Ambrogio S, Wang Z, et al. Pulsed cycling operation and endurance failure of metal-oxide resistive (RRAM). in: Electron Devices Meeting (IEDM), 2014 IEEE International. San Francisco, CA: 2014. pp. 14.3.1–14.3.4. Google Scholar
Chen CY, Goux L, Fantini A, et al. Understanding the impact of programming pulses and electrode materials on the endurance properties of scaled Ta2 O5 RRAM cells. in: Electron Devices Meeting (IEDM), 2014 IEEE International. San Francisco, CA: 2014. pp. 14.2.1–14.2.4. Google Scholar
Huang P, Chen B, Wang YJ, et al. Analytic model of endurance degradation and its practical applications for operation scheme optimization in metal oxide based RRAM. in: Electron Devices Meeting (IEDM), 2013 IEEE International. Washington, DC:2013.pp.22.5.1–22.5.4. Google Scholar
Chen B, Lu Y, Gao B, et al. Physical mechanisms of endurance degradation in TMO-RRAM. In: Technical Digest – International Electron Devices Meeting, IEDM. 2011.Google Scholar
Chen YY, Degraeve R, Clima S, et al. Understanding of the endurance failure in scaled HfO2-based 1T1R RRAM through vacancy mobility degradation. in: Technical Digest – International Electron Devices Meeting, IEDM. 2012. Google Scholar