Ying-Chen Chen*a,
Yao-Feng Changa,
Xiaohan Wua,
Fei Zhoua,
Meiqi Guoa,
Chih-Yang Linb,
Cheng-Chih Hsieha,
Burt Fowlera,
Ting-Chang Changb and
Jack C. Leea
aMicroelectronics Research Center, Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, Texas 78758, USA. E-mail: yingchenchen@utexas.edu
bDepartment of Physics, National Sun Yat-Sen University, Kaohsiung 804, Taiwan
First published on 24th February 2017
Characteristics of HfOx-based resistive switching memory (RRAM) in Al/HfOx/Al and Al/AlOx/HfOx/Al structures were studied using a dynamic conductance method. Step-like RESET behaviors as well as pre- and post-RESET regions of operation were characterized. The results indicated that defects at the oxide interface caused cycling issues in the Al/AlOx/HfOx/Al structure. No such RESET behavior was observed for the Al/HfOx/Al structure. Current induced over-heating, which caused an early RESET event, could be avoided using current-sweep technique that caused less electrical and thermal stress in localized regions. The experimental results not only provided insights into potential reliability issues and power management in HfOx-based RRAM, but also helped clarifying the resistive switching mechanisms.
To investigate RESET mechanisms, I–V curves were analyzed using the dynamic conductance method (i.e. ∂I/∂V) in two regions (i.e., pre-RESET and post-RESET). During the RESET sweep to ∼0.8–1 V, current began to drop at the RESET voltage (VRESET) and the device was programmed to HRS. With identical SET CCL of 1 mA, multi-step-like and one-step-like RESET were observed in RRAM with and without AlOx (Fig. 1(c) and (d)). In other words, by adding AlOx in the switching layers, the RESET I–V transition above VRESET changed from one-step fall-off to multiple fall-off in current resulting in different HRS states. This difference could be contributed to the high defect density at the interface between the AlOx and HfOx layers, leading to the abrupt current drop during filament rupturing in the RESET process.15,16 In contrast, the SET process showed no multiple SET behaviors in these two structures. Analyzing plots of I–V dynamic conductance (or 2nd derivative) can help understand the RESET mechanism and further clarify the fall-off process where the filament degrades/ruptures during the RESET process. Fig. 2(a) shows a schematic of the temperature dependence in the pre-RESET region, cycling capability, and device structure dependence observed in the 1st derivative/2nd derivative plots for devices with (green curve) and without AlOx (blue curve). These dynamic conductance characteristics (1st derivative/2nd derivative analyses) can help to verify that the fundamental filament degradation behaviors were enhanced in devices without AlOx compared to devices with AlOx for these variables under cycling analyses (more discussion in Fig. 2(c), shows that the unnecessary heat resulting in filament degradation in the pre-RESET region can be avoided). For the post-RESET region, temperature dependent dynamic conductance characteristics could further confirm the statement of unnecessary heat induced filament degradation in the pre-RESET region (called an early RESET event in LRS), and help to clarify the one-step fall-off RESET process and the multi-step fall-off current drop between these two structures (more discussion in Fig. 3, diffusion-driven and electrical-driven RESET behavior). Fig. 2(b) shows the I–V curve and the dynamic conductance plot of RRAM devices with AlOx, which provides the switching parameters, such as intercept of dynamic conductance (at 0 V), the tangential slope of the dynamic conductance at the intercept (at 0 V), and the slope of dynamic conductance (>zero bias region), which is described in further detail in Fig. 2(c) and 3.
![]() | ||
Fig. 3 (a) Filament degradation as a function of temperature pre- and post-RESET, (b) non-zero counts in 2nd derivative. |
Fig. 2(c) shows the cycling effect on the initial rate of ILRS increment (i.e. the first derivative) and filament degradation for the two RRAM devices. The initial rate of ILRS increment is defined as the intercept of dynamic conductance curve, which is shown by the red curve in Fig. 2(b), and filament degradation is defined as the tangential slope of the dynamic conductance at the intercept. The initial rate of ILRS increment was higher for RRAM devices without AlOx, which indicated that the RRAM without AlOx reached the RESET current faster than RRAM with AlOx (see the black curve in Fig. 2(c)). The initial rate of ILRS increment values in Al/AlOx/HfOx/Al structure were quite close to zero for all 10 cycles, which indicated that the filament was weaker than the Al/HfOx/Al structure even in the zero-bias region due to the growth of filament being limited by the internal filament effect, which potentially triggered the RESET event earlier (in Al/HfOx/Al, the values were positive, which means that the LRS or filament continues increasing or growing).17 A similar scenario could be observed in the filament degradation analysis (blue curve in Fig. 2(c)); RRAM with AlOx having worse filament degradation after eight cycles (the larger of negative values (e.g. −0.2)). This implies that the filament was already in the self-compliance region (the tangential slope of the dynamic conductance has three cases: slope > 0, the filament continues growth; slope ∼ 0, filament self-limiting behavior; and slope < 0, RESET transition beginning) and would run into the RESET transition at the zero-bias region, which indicated that the generated current-induced Joule heating by DC cycling severely affected the filament degradation rate in the Al/AlOx/HfOx/Al structure. With a higher initial rate for the ILRS increment, the unnecessary heat resulting in filament degradation in the pre-RESET region could be avoided for RRAM without AlOx. In other words, the RRAM without AlOx is more desirable due to potentially less electrical or thermal stress than in the Al/AlOx/HfOx/Al structure and less cycle-induced filament degradation. In addition, this filament degradation of RESET behaviors has an important correlation with endurance failure.
Fig. 3(a) shows the filament degradation as a function of operating temperature for the pre- and post-RESET regions on RRAM without AlOx. The filament degradation (i.e. the slope of dynamic conductance) in the pre-RESET region (>zero bias region, as shown in Fig. 2(a)) is caused by a sequence of random defect events, i.e. a defect injection into the filament or extraction from the filament. The value of 2nd derivative due to dynamic resistance change of filament (Fig. 2(c)) was then analyzed using a linear fitting method to obtain the overall slope, which was negative in the pre-RESET region, as shown in Fig. 2(a). The LRS filament degradation was caused by discrete defects migration, which correlates to a random value of activation energy (Ea) and a corresponding migration rate.18,19 With increasing operating temperature up to 60 °C, filament degradation begins to occur in the pre-RESET region. The filament degradation was much more severe in post-RESET (ruptured) than in the pre-RESET processes at room temperature (value of −3 vs. 0). For the pre-RESET region, the filament degradation deteriorated with increasing temperature after 50 °C (from −0.01 to −0.05). However, for the post-RESET region, the filament degradation became less negative with increasing temperature from 40 °C to 70 °C. This can be explained by diffusion-driven (i.e. highly temperature-dependent) RESET behavior in post-RESET process, which means gradual-RESET behavior occurs other than an abrupt RESET.20–22 On the other hand, the electrical-driven RESET behavior can explain the post-RESET region, i.e. the abrupt RESET with larger filament degradation (value of −3) at 20 °C in Fig. 3(a).
Fig. 3(b) shows the non-zero fluctuation counts in the 2nd derivative of the RESET I–V curve from 20 to 70 °C. The non-zero fluctuation counts indicate the observation of smooth curve-like continuous gradual RESET switching behavior instead of an abrupt RESET one. The more non-zero fluctuation indicated highly gradual RESET-like behavior. In the pre-RESET region, the counts increased with increasing temperature due to thermal disturbance, leading to filament degradation and increasing probability of early RESET on LRS. In other words, the increasing fluctuation may result from the increasing probability to either inject into or extract out of the conductive filament even under a thicker filament condition, i.e. LRS. In the post-RESET region, the counts increased with increasing temperature up to 50 °C and showed the temperature-induced gradual RESET. This is due to the fact that the rate of defect injection exceeded the rate of extraction among the discrete defect migration during the RESET transition. This phenomenon retarded after 50 °C, possibly due to current induced over-heating (or self-accelerated by Joule heating) in a localized region,23 which in turn caused abrupt RESET and highly efficient rupture of the filament. Thus, by suppressing the accelerated self-heating process in unipolar-type resistive switching, a precise control of filament temperature and localized switching phenomena could be achieved. This provides a possible solution for an early RESET event on LRS.24,25
Fig. 4 and 5 show the data obtained by the current-sweep technique during the SET process. The current sweep technique is known to prevent device hard-breakdown and current overshoot failures (i.e. less electrical and thermal stress).26 The typical current-sweep I–V curves of Al/HfOx/Al and TaN/HfOx/Al devices are shown in Fig. 4. In contrast to the I–V curves using voltage-sweep, there were multiple states observed between the LRS and HRS in both Al/HfOx/Al and TaN/HfOx/Al RRAM devices. These multiple intermediate states can potentially be used for multi-level memory application by controlling the CCL in the SET process and RESET stop voltage in the RESET process, respectively. Moreover, the snap-back voltage, i.e. the maximum sensing voltage, could be observed during the current-sweep process where the resistance state begins to change from HRS to LRS. Through the SET process, the voltage across the device continued to increase as the current was swept up until a certain current endpoint was reached, i.e. LRS, where the “snap back voltage” is. The multiple voltage snap-back observation reflected the gradual RESET process of the conductive filament, suggesting that the broken conductive filament built up again gradually in the HfOx layer.27 The SET voltage was relatively higher in Al/HfOx/Al than in TaN/HfOx/Al devices potentially due the competitive oxygen vacancies affinity between the Al electrode and TaN electrode, e.g. the enthalpy of formation for of TaN and Al, −252.3 kJ mol−1 and 577.5 kJ mol−1, respectively.28,29
![]() | ||
Fig. 5 Device area effect of Al/HfOx/Al RRAM (a) on the HRS and LRS, (b) on forming and switching voltage by voltage-sweep (VS) and current-sweep (CS) methods. |
The effect of device area by current-sweep in the Al/HfOx/Al device compared to the voltage-sweep measurement technique is shown in Fig. 5(a) and (b). It was found that LRS and HRS currents (Fig. 5(a)), as well as the SET and RESET voltages (Fig. 5(b)) were independent of device area, illustrating that charge transport and RS occurred in a localized region along a conductive filament.30 Consistently, it provided further switching voltage reduction using current-sweep (Fig. 5(b), 0.6–1.4 V for the current-sweep and 1.5–2.8 V for the voltage-sweep). This voltage reduction may be a result of heat-induced defect injection that could improve the programming efficiency than an electrical-field driven process in a localized region along a conductive filament. The structure of filament changes was due to defect injection/extraction during programing.31 The current sweep technique not only improves data acquisition and enhances the understanding of the physics related to the SET process, but also provides the power management capability in a localized region for further reduction of current induced over-heating in an early RESET event on LRS.
This journal is © The Royal Society of Chemistry 2017 |