Conductance Quantization Behavior in Pt/SiN/TaN RRAM Device for Multilevel Cell

Abstract

In this work, we fabricated a Pt/SiN/TaN memristor device and characterized its resistive switching by controlling the compliance current and switching polarity. The chemical and material properties of SiN and TaN were investigated by X-ray photoelectron spectroscopy. Compared with the case of a high compliance current (5 mA), the resistive switching was more gradual in the set and reset processes when a low compliance current (1 mA) was applied by DC sweep and pulse train. In particular, low-power resistive switching was demonstrated in the first reset process, and was achieved by employing the negative differential resistance effect. Furthermore, conductance quantization was observed in the reset process upon decreasing the DC sweep speed. These results have the potential for multilevel cell (MLC) operation. Additionally, the conduction mechanism of the memristor device was investigated by I-V fitting.

1. Introduction

Resistive switching random access memory (RRAM) is a promising candidate for next-generation non-volatile memory owing to its high scalability, low-power operation, high switching speed, long retention time, and high endurance [1,2,3]. In a metal-insulator-metal structure, resistive switching occurs by bias voltage and optical stimulation and results in a change in the internal resistance. In particular, metal electrodes play an important role in resistive switching. For example, diffusive metals such as Ag and Cu can enter the insulator and form a conducting filament [4,5,6]. In the case of inert metals such as Pt and Au, intrinsic switching of the insulating layer occurs [7].

On the other hand, the use of a reactive-type metal with oxygen such as TiN and TaN results in the formation of an interfacial layer between the metal and the insulating layer, which can improve the resistive switching properties [8,9,10,11]. For the insulating layer acting as resistive switching, a lot of materials could be used. Metal-oxide-based RRAM has been the most studied RRAM and shows excellent memory performance in terms of features such as endurance, retention, stability, repeatability, and reproducibility [1,3,12]. Nitride-based RRAM also shows good memory performance but has been relatively less researched [13,14,15,16,17,18]. Further research of SiN-based RRAM may be worthwhile. Among nitride-based RRAM devices, SiN devices could be the most promising, since SiN is not only a well-known material but also shows good complementary metal-oxide-semiconductor (CMOS) compatibility. Furthermore, the large number of traps in a SiN film is advantageous for achieving resistive switching.

The switching of RRAM can be classified into three types: unipolar, filamentary bipolar, and interface bipolar types. In the unipolar type [19], the set and reset processes occur at the same polarity voltage. The reset process is induced by the joule heating of the conducting filament because of a high current passing through it [19]. In the filamentary bipolar type, opposite electric fields should be applied to the device for the set and reset processes, to connect and disconnect the conducting filament in the insulating layer [13]. In the interface bipolar type, conducting defects are distributed throughout the metal electrode-switching layer interface [20]. Gradual resistive switching is observed in the interface type, and it is beneficial for achieving multilevel states [21,22]. Low-current operation is necessary to realize high-density in a cross-point RRAM array. Nonlinear I-V characteristics and low current operation can minimize the sneak current and increase the array size. In previous works, we have demonstrated that the current in SiN-based RRAM devices can be reduced through appropriate device scaling and by adjusting the ratio of Si to N and the number of depositions. Here, we show that the current in a Pt/SiN/TaN device can be reduced by inducing a negative differential resistance (NDR) effect of an initial I-V curve of the device. A metal electrode with an asymmetric work function could facilitate fluent resistive switching because the use of a reactive electrode such as a TiN or TaN electrode on one side can confine the conductive filament [23,24]. We also compare the resistive switching behavior between normal bipolar resistive switching and low current mode by NDR mode [25]. Finally, we present the results of an investigation into the quantized conductance effect on the reset process for multilevel operation; understanding the effect of quantized conductance on the reset process is important for the implementation of neuromorphic systems.

2. Materials and Methods

Pt/SiN/TaN devices were fabricated as follows. A 100 nm thick TaN bottom electrode was deposited on a SiO₂/Si wafer by DC sputtering. Ar (19 sccm) and N₂ (1 sccm) were used as the sputtering gases, the sputtering power was 65 W (DC), and the deposition pressure was 5 mTorr. SiN film was deposited by plasma-enhanced chemical vapor deposition (PECVD) at a temperature of 250 °C and an RF power of 100 W, and SiH₄ (5%)/He (95%) (320 sccm), NH₃ (9 sccm), and N₂ (450 sccm) were used as reactive gases. A 100 nm thick Pt top electrode was deposited on the SiO₂/Si wafer by DC sputtering. The flow rate of the sputtering gas (Ar) was 20 sccm, the sputtering power was 65 W (DC), and the deposition pressure was 6 mTorr. Each memory cell was separated by a shadow mask, which contained a 100 μm diameter circular pattern. A Keithley 4200-SCS semiconductor parameter analyzer (Keithley Instrumnets, Cleveland, OH, USA) and a 4225 pulse measurement unit (Keithley Instrumnets, Cleveland, OH, USA) in the probe station were used to measure electrical characteristics in the DC sweep mode and transient features. While a bias was applied to the Pt electrode, the TaN electrode was grounded. X-ray photoelectron spectroscopy (XPS) analysis was conducted using a Nexsa X-ray photoelectron spectrometer (Thermo Fisher Scientific, Waltham, MA, USA) with a microfocus monochromatic X-ray source (Al-Kα, 1486.6 eV); additionally, a sputtering source (Ar⁺), an ion energy of 2 kV, a sputtering area of 1 mm × 1 mm, a sputtering rate of 0.5 nm/s for SiO₂, and a beam size of 100 µm were used for the XPS analysis.

3. Results and Discussion

Figure 1 shows a transmission electron microscope (TEM) image of the Pt/SiN/TaN memory device. The thickness of the amorphous SiN layer was about 6 nm. We investigated the XPS spectra of the SiN and TaN layers, and Ar⁺ etching was used to obtain chemical and material information about these two layers. The sputtering rate was 0.5 nm/s for SiO₂. Figure 1b shows Si 2p spectra after an Ar⁺ etching time of 2 s. The peak binding energy of Si 2p is centered at 102.2 eV for the Si-N bond [14], indicating that the SiN film formed through PECVD did not have the stoichiometry of Si₃N₄. Figure 1c shows the N 1s spectra after etching times of 2 and 6 s. For the 2 s etching time, the peak of the binding energy was 497.85, corresponding to the Si-N bond. Interestingly, for the 6 s etching time, another peak from 402 to 404 eV was observed for the Ta-N bond [26]. Figure 1d shows Ta 4f spectra for the etching time of 6 s, and doublet peaks of 23.67 and 25.57 eV, corresponding to TaN 4f_7/2 and TaN 4f_5/2 [26], are evident.

Figure 2a–c show typical I-V curves of the Pt/SiN/TaN device by different switching. This device showed a forming-free process that can eliminate the first process with different voltage conditions because of the presence of numerous traps in SiN, and therefore, the initial state was close to a high-resistance state (HRS). In this study, the conventional sweep mode (no delay between steps, step voltage = 0.05 V) was used. Typical bipolar resistive switching was observed for high-current operation (Figure 2a). The set process occurred at about −1.3 V, and it changed the HRS to a low-resistance state (LRS). The current increased until the compliance current reached 5 mA with an additional voltage sweep. It is noteworthy that a self-compliance current can prevent surges in the current. By contrast, the reset process changed the LRS to a HRS. A decrease in the current was observed for a reset bias sweep. The reset current was comparable to the compliance current. Resistive switching of a SiN-based RRAM can occur by the trapping and de-trapping of an electron in nitride-related traps [18]. Figure 2b shows the lower current operation achieved by limiting the compliance current for typical bipolar resistive switching. We also demonstrate the lowest current operation by using an NDR with a negative bias, and it is shown in Figure 2c. An initial reset was observed, and it could be attributed to the presence of a sufficient number of initial traps in the SiN film. Subsequently, the set process was observed for a positive bias. Similar characteristics have been previously observed for a Pt/Al₂O₃/TiN memory device [27].

Figure 3a shows I-V curves for the slow sweep measurement mode (delay time = 0.1 s, step voltage = 0.002 V); the quantized effect on the reset process is evident. A decrease in the current is observed over several steps during the reset process. Figure 3b shows conductance-voltage curves for the reset process. G₀ = 2e²/h is the quantum of conductance (77.5 μs) [28], where e is the electron charge and h is Planck’s constant. Although the behavior of the reset process is quite different owing to large randomness, the quantized effect can be clearly observed in a few cycles. The red curve in Figure 3b shows the conductance decreasing gradually in a quantized manner (4G₀ → 3G₀ → 2G₀ → 1G₀ → 0.5G₀). Similarly, the blue curve shows the conductance decreasing (5G₀ → 4G₀ → 2G₀ → 1G₀) in a stepwise manner. Multiple conductance states are useful to hardware neuromorphic systems [29,30,31,32,33,34,35].

Figure 4 shows the pulse train to clearly observe current modulation by the applied voltage for practical operation of the memory devices. First, the pulses were applied after the switching type was determined on the basis of the DC sweep. The total pulse number was 100 for each case, the width of each pulse was 0.1 ms, and the rise and fall times were 0.01 ms. Figure 4a,b show the set and reset processes for type 1 at pulse voltages of −1.5 and 2 V, respectively. An abrupt transition occurred for both processes. Figure 4c,d show the transient characteristics of the pulse train for type 2. Compared with the set and reset processes of type 1, type 2 showed a more gradual transition.

Next, the conduction mechanism was investigated by I-V fitting for the aforementioned three switching types. Figure 5a shows a log–log fit to the type 1 I-V curve for the LRS. A slope of 1 is observed for Ohmic conduction, indicating that the current could flow through a strong silicon conducting filament in the SiN film. Before the reset process, the slope for the LRS for type 2 exceeded 1. On the other hand, the Poole-Frenkel emission fitting (In(I/V) versus V^1/2) [36] matched the HRS of types 1 and 2 well (Figure 5d). An electron was exited from traps in the SiN film to the conduction band. Figure 5c shows the Schottky emission fitting (ln(I) versus V^1/2) [36] for low current switching in the LRS and HRS. A gradual resistance change was achievable through the modulation of the Schottky barrier for MLC.

4. Conclusions

In summary, a Pt/SiN/TaN memristor device was fabricated and characterized by DC sweep and pulse operation. Component analysis of the SiN/TaN stack was performed, and the SiN thickness was determined from a high-resolution TEM image. Three typical resistive switchings were classified by compliance current and polarity control. Gradual resistive switching was achieved at a compliance current of 1 mA, and the gradual increase and decrease of the current are verified by the pulse train. Moreover, the effect of quantized conductance on the multistep reset process was investigated by reducing the DC sweep speed. The resulst are expected to be applicable to hardware based neuromorphic applications. Finally, the conduction mechanisms of different switching types were determined by I-V fitting.