Background

Resistive random access memory (RRAM) has attracted great attention due to its potential for the replacement of flash memory in next-generation nonvolatile memories [1,2,3]. Resistive switching (RS) phenomenon has been widely discovered in transition metal oxides, solid electrolytes, and organic polymers [4,5,6,7]. RRAM devices based on transition metal oxides have been extensively explored because of its simple composition and compatible processing with conventional complementary metal-oxide-semiconductor (CMOS) fabrication [8,9,10]. The filament model of oxygen vacancy migration is used to elucidate the switching behavior [1, 11]. A unified microscopic principle is proposed to quantify both unipolar and bipolar switching characteristics of transition metal oxide-based RRAM, which are correlated with the distribution of localized oxygen vacancies in the oxide switching layer [12, 13].

However, memory cells using transition metal oxides suffer from nonuniformity of resistive switching parameters, such as unstable resistance values of low and high resistance states (LRS and HRS), dispersed set and reset voltages, impeding the commercial applications. Recently, the trilayer-structure oxide-based RRAM devices have been demonstrated to improve the dispersion of resistive switching parameters. The cells with a structure of Al2O3/HfO2/Al2O3 exhibited fantastic uniformity of set and reset voltages, and excellent endurance of switching between the LRS and HRS [14]. The linkage or rupture of the conductive filaments more easily occurred in two interfacial layers between Al2O3/IL/HfO2/IL/Al2O3. Meanwhile, the unit of trilayer-structure of TaOx/TiO2/TaOx showed good performance in one selector-one resistor arrays, which was ascribed to the fact that the energy band of the TiO2 film was symmetrically bent at the top and bottom TaO x /TiO2 interfaces and modified as a crested oxide barrier due to some Ta atoms diffusion into TiO2 film [15].

Moreover, the RS behavior of a given oxide storage medium can be significantly affected by the electrode materials [1, 16, 17]. However, the existing models based on the free energy of interfacial oxide formation and the metal work functions are insufficient to completely explain the results. Meanwhile, the work on the electrodes dependent RS phenomenon of the trilayer-structure RRAM is also rather lacking at present.

Atomic layer deposition (ALD) is a new type of thin film deposition technology based on sequential self-limited and complementary surface chemisorptions reactions using precursor vapor with simple and precision thickness control, large area uniformity, and excellent three-dimensional conformality, especially for deposition of nano-laminated structure [18, 19].

In this work, the HfO2/TiO2/HfO2 trilayer-structure RRAM devices have been prepared on Si/SiO2/Ti/Pt and Si/TiN substrates with Pt top electrodes by ALD. The impact of the bottom electrodes of Pt and TiN on the RS behaviors of HfO2/TiO2/HfO2 devices has been investigated carefully. The related explanation has been proposed.

Methods

In this experiment, we used two different bottom electrodes, including commercial Si/SiO2/Ti/Pt and home-made Si/SiO2/TiN. Conductive TiN was deposited by plasma-enhanced atomic layer deposition (PEALD) in our laboratory.

ALD was performed in a commercial Picosun SUNALETMR-200 advanced reactor (Picosun, Finland). P type Si (100) wafers with a resistivity of 1~10 Ω cm were used as the starting substrates. After the conventional RCA cleaning of the Si wafers without removing native oxide, 30-nm-thick TiN was deposited on Si as the bottom electrode at 400 °C by PEALD, using room temperature TiCl4 and NH3 plasma gas as the Ti and N precursors, respectively. Liquid NH3 was selected as NH3 plasma source at room temperature. The plasma power and NH3 gas flow rate were 2500 W and 150 sccm, respectively.

Subsequently, 5 nm HfO2/10 nm TiO2/5 nm HfO2 stacking structures were deposited in turn on Pt- and TiN-coated Si substrates at 250 °C by thermal ALD using Hf[N(C2H5)CH3]4 (TEMAH), TiCl4, and H2O as the Hf, Ti, and O precursors, respectively, where one oxide cycle consisted of 0.1s metal source injection, 4s N2 purging, 0.1s H2O injection, and 4s N2 purging. TEMAH was evaporated at 150 °C. Pure N2 (99.999%) was used as carrier gas and purging gas. Then, 100-nm-thick Pt top electrodes were DC sputtered through a shadow mask with a diameter of 150 μm using the Q150T system.

The growth per cycle (GPC) of pure HfO2 or TiO2 on Si was determined by spectroscopic ellipsometer (GES-5, Sopra). The topography and surface roughness of the films and bottom electrodes were analyzed by atomic force microscopy (AFM, Cypher, Asylum Research). The root-mean-square (RMS) roughness values were recorded from 1 μm × 1 μm areas. The composition and chemical state of the stack structures were examined by X-ray photoelectron spectroscopy (XPS, Thermo Fisher K-Alpha) with a monochromatic Al Kα source (hν = 1486.6 eV) for excitation of photoelectrons. The charge effect was calibrated by setting the C 1s photoemission at 284.6 eV. The XPS depth profile of HfO2/TiO2/HfO2 on Pt- and TiN-coated Si was obtained by Ar ion etching. The electrical properties of the HfO2/TiO2/HfO2 trilayer-structure RRAM devices were measured by Keithely 4200 semiconductor characterization system on probe station (CasCade Summit 12000 B-M). A current compliance of 10 mA was imposed to protect the fabricated device units from damages of high current during set processes. The bias voltage was applied to the Pt top electrode with the grounded bottom electrodes of Pt or TiN.

Results and Discussion

The schematic of the RRAM device of HfO2/TiO2/HfO2 trilayer-structure by ALD is illustrated in Fig. 1. The surface morphology and roughness of the bottom electrodes and trilayer-structure of HfO2/TiO2/HfO2 on Pt- and TiN-coated Si have been examined. The Pt bottom electrode has smaller RMS value of 0.39 nm than PEALD-derived TiN of 0.87 nm. Hence, the sample of HfO2/TiO2/HfO2 on Pt-coated Si also exhibits relatively smoother surface with RMS of 0.68 nm than that on TiN-coated Si with 1.3 nm.

Fig. 1
figure 1

The schematic of the RRAM device of HfO2/TiO2/HfO2 trilayer-structure by ALD

Full size image

The DC I-V curves of Pt/HfO2/TiO2/HfO2/Pt and Pt/HfO2/TiO2/HfO2/TiN devices containing the initial electroforming process are plotted in Fig. 2a, b, respectively, indicating a typical bipolar resistive switching characteristic. For almost all the samples, larger forming voltage is needed to form conductive filaments before the switching test. When applying a positive bias voltage on the Pt top electrode, the Pt/HfO2/TiO2/HfO2/Pt device unit shows a forming voltage of +7 V in Fig. 2a. With continuing the voltage sweeping, a reset voltage of −0.8 V is measured for unit cell from LRS to HRS and a set voltage of 2.0 V from HRS to LRS. The corresponding ratio of high and low resistances is about 105. The electroforming and rest process can also be completed by applying a negative voltage of −11 V and a positive one of +4 V, respectively, which are much larger than the positive forming and negative reset voltages. Moreover, the device cell only switches from LRS to HRS for several cycles after the negative forming process and then fails to reset to HRS due to the irreversible breakdown (not shown here). In Fig. 2b, compared to that with symmetric Pt top and bottom electrodes, the tri-layer structure RRAM devices with asymmetric TiN bottom electrode and Pt top electrode show an opposite lower forming voltage of about −3.7 V, set voltage of −1.5 V, reset voltage of +1.5 V, and relatively smaller ratio of high and low resistances of 102. When imposing the positive electroforming voltage, the RS phenomenon cannot be observed in the Pt/HfO2/TiO2/HfO2/TiN cell and the device is permanently broken down at +14 V without the following efficient reset from LRS to HRS at negative voltage (not shown here).

Fig. 2
figure 2

The typical bipolar resistive switching characteristics of the RRAM devices. a Pt/HfO2/TiO2/HfO2/Pt. b Pt/HfO2/TiO2/HfO2/TiN

Full size image

The trilayer-structure RRAM devices with symmetric Pt top/bottom electrodes and asymmetric TiN bottom electrode/Pt top electrodes exhibit such different electroforming polarity preference. The bottom electrode of chemically inert Pt or relatively active TiN seems to play a key role. The related reason on electrode-dependent electroforming polarity will be discussed later after considering the XPS depth profiles of trilayer structures of HfO2/TiO2/HfO2 on Pt- and TiN-coated Si.

For high-density memory application, uniformity of RS parameters is very important. Figure 3a, b plots the statistical results of distribution of the set and reset voltages measured from a single device unit of the Pt/HfO2/TiO2/HfO2/Pt and Pt/HfO2/TiO2/HfO2/TiN for 200 times tests, respectively. Figure 3c, d records the I–V curves of 10 randomly selected device units of the Pt/HfO2/TiO2/HfO2/Pt and Pt/HfO2/TiO2/HfO2/TiN, respectively. The trilayer-structure RRAM device units with symmetric Pt top and bottom electrodes show a broad distribution for set voltage from 1.2 to 2.8 V and reset voltage from −0.5 to −1 V (Fig. 3a) and dispersive I–V curves (Fig. 3c). Whereas, the device units with asymmetric TiN bottom and Pt top electrodes display better RS behavior, such as relatively concentrated distribution of set voltage from −0.8 to −1.8 V and reset voltage from 1.3 to 1.8 V, and stable reproducibility in I–V curves (Fig. 3b, d). Simultaneously, compared to those on Pt-coated Si, different device units on TiN-coated Si also behave improved monodispersion in RS parameters, beneficial to RRAM practical applications.

Fig. 3
figure 3

The statistical results of distribution of the set and reset voltages measured from a single device unit for 200 times tests. a Pt/HfO2/TiO2/HfO2/Pt. b Pt/HfO2/TiO2/HfO2/TiN. The I–V curves of 10 randomly selected device units. c Pt/HfO2/TiO2/HfO2/Pt. d Pt/HfO2/TiO2/HfO2/TiN

Full size image

The endurance and retention properties of the device units of Pt/HfO2/TiO2/HfO2/Pt and Pt/HfO2/TiO2/HfO2/TiN have been examined, as seen in Fig. 4a–d, respectively. In Pt/HfO2/TiO2/HfO2/Pt, the sweeping voltage was applied from 0 to 3 V for set and 0 to −1.5 V for reset. In Pt/HfO2/TiO2/HfO2/TiN, the sweeping voltage was applied from 0 to −2 V for set and 0 to 2 V for reset. The ON and OFF resistance values were read using 0.2 V at room temperature. The retention tests were measured at room temperature with the reading voltage of 0.2 V. After 200 program/erase cycles, Pt/HfO2/TiO2/HfO2/Pt device units show relatively stable resistance ratio of OFF/ON states above 105 (Fig. 4a); however, the endurance characteristic of Pt/HfO2/TiO2/HfO2/TiN memory cells seem not to be as good as that of Pt/HfO2/TiO2/HfO2/Pt (Fig. 4b). The ON and OFF states in devices with Pt-TiN electrodes are not very steady with smaller resistance ratio of OFF/ON states of about 102 during switching cycle test, especially significant HRS fluctuation than that of LRS. Based on the physical model in Ref. [12], the endurance will be improved by increasing the formation energy of oxygen vacancy and interface O2− amount. We speculate one plausible explanation. In our previous work [14], excellent bipolar resistive switching properties of ALD-derived Al2O3/HfO2/Al2O3 trilayer-structures with asymmetric TiN bottom and Pt top electrodes have been demonstrated, including better switching endurance up to 103 cycles with stable ON/OFF resistance ratio. Herein, we adopted HfO2/TiO2/HfO2 configure instead of Al2O3/HfO2/Al2O3. The metal ions in HfO2 and TiO2 have the same identical chemical valence of +4, leading to the less interface charged defects such as oxygen vacancies between two interfacial layers (ILs) of trilayer HfO2/TiO2/HfO2. Whereas, the metal ions in Al2O3 and HfO2 have different chemical valence of Al3+ and Hf4+, producing more interface charged defects of oxygen vacancies between two ILs of Al2O3/HfO2/Al2O3. It can be deduced that the interface O2− amount by adding two ILs between Al2O3/HfO2/Al2O3 should be higher than between HfO2/TiO2/HfO2. In addition, the formation energy of oxygen vacancy in storage layer of Al2O3/HfO2/Al2O3 is also higher than that of HfO2/TiO2/HfO2 (formation energy of oxygen vacancy 7.08 eV (Al2O3), 6.53 eV (HfO2), and 4.35 eV (TiO2) [20, 21]). After considering these factors, the memory cell of Pt/HfO2/TiO2/HfO2/TiN exhibits endurance degeneration, compared to Pt/Al2O3/HfO2/Al2O3/TiN.

Fig. 4
figure 4

The endurance and retention characteristics of the device units. a, c Pt/HfO2/TiO2/HfO2/Pt. b, d Pt/HfO2/TiO2/HfO2/TiN.

Full size image

For Pt/HfO2/TiO2/HfO2/Pt and Pt/HfO2/TiO2/HfO2/TiN, the resistance level of the ON and OFF states has no evident changes after cumulative waiting time of 104 s at room temperature in Fig. 4c–d, indicating a better retention property of both memory cells over 10 years based on the extrapolation method.

Pt/HfO2/TiO2/HfO2/TiN devices have much lower resistance ratio of OFF/ON states than Pt/HfO2/TiO2/HfO2/Pt, which can be attributed to the difference of the bottom electrodes in work function and conductivity. The work function of Pt and TiN bottom electrodes is 5.65 and 4.5 eV, respectively; hence, the interface barrier between TiN and underlying HfO2 is relatively lower in the Pt/HfO2/TiO2/HfO2/TiN device. During a reset process, the needed energy for electron to cross the barrier is also smaller. So Pt/HfO2/TiO2/HfO2/TiN manifests lower resistance value in OFF state than Pt/HfO2/TiO2/HfO2/Pt correspondingly. During a set process, RRAM device changes from HRS into LRS owing to the formation of the conducting channels. The Pt bottom electrode has higher conductivity than TiN bottom one; accordingly, the resistance value in ON state for Pt/HfO2/TiO2/HfO2/TiN is higher than that for Pt/HfO2/TiO2/HfO2/Pt. As a result, Pt/HfO2/TiO2/HfO2/TiN devices show a smaller resistance ratio of OFF/ON states of 102. However, the resistance ratio of 102 has already met the requirement of RRAM applications.

To clarify the conductive mechanism during resistive switching, the typical I–V curves are replotted in double-logarithmic scale. Figure 5a, b shows the linear fitting of the I–V curves for the voltage sweeping regions of Pt/HfO2/TiO2/HfO2/Pt and Pt/HfO2/TiO2/HfO2/TiN devices. Both memory cells behave similar conductive mechanism. When the device is switched to the LRS, the curves of log (I)-log (V) are linear with slope close to 1 (0.94 in Fig. 5a, 0.98 in Fig. 5b), indicating that the I–V at the LRS is dominated by the Ohmic law. The filament model of oxygen vacancy migration can be used to explain the switching behavior. For the HRS, at low-voltage region (absolute value <0.11 V), the I–V is dominated by the Ohmic law with the linear relationship of current and voltage (slope 1.05 in Fig. 5a, 1.09 in Fig. 5b). At higher voltage region (6.8 V > absolute value >0.11 V in Fig. 5a, 1.85 V > absolute value >0.11 V in Fig. 5b), the slope of the log (I)-log (V) lines are around 2 and the current is dependent of approximate square of applied voltage (I/V2). At critical voltage (absolute value 6.8 V in Fig. 5a, 1.85 V in Fig. 5b), a steep current rise suddenly appears with a very large slope. This result consists of three regions in HRS, basically obeying the typical trap-controlled space-charge-limited conduction (SCLC) injection [11].

Fig. 5
figure 5

The typical I–V curves plotted in double-logarithmic scale of a Pt/HfO2/TiO2/HfO2/Pt and b Pt/HfO2/TiO2/HfO2/TiN

Full size image

In order to further understand the underlying resistive switching mechanism of HfO2/TiO2/HfO2 on Pt-coated and TiN-coated Si, we performed the XPS narrow scans and depth analyses on trilayer structures with symmetric and asymmetric bottom/top electrodes. XPS spectra were fitted with Gaussian-Lorentzian (G-L) functions after smart-type background subtraction.

Figure 6a–d illustrates the narrow-scan XPS spectra of Hf 4f, Ti 2p, and O 1s peaks in HfO2 and TiO2 layers on TiN-coated Si. The Hf 4f and Ti 2p spectra from HfO2 and TiO2 layers can be deconvoluted into two group peaks. The stronger Hf 4f 5/2 and Hf 4f 7/2 peaks at 18.4 and 16.7 eV with a spin–orbit splitting of 1.7 eV are assigned to Hf–O bonding (Hf4+) from HfO2 (Fig. 6a). The weaker spin–orbit doublet peaks at lower binding energy of 17.6 and 14.8 eV possibly result from the low-chemical valence state of Hfn+–O (n < 4), indicating the presence of oxygen vacancies in HfO2 layer. The calculated percentage concentration of Hfn+ (n < 4) is about 3.7% in Hf ions. In Fig. 6b, a stronger doublet corresponds to Ti 2p3/2 and 2p1/2 features at 458.8 and 464.5 eV, belonging to the Ti–O bonding (Ti4+) from TiO2. A weaker doublet locates at 456.1 and 462.0 eV, assigning to the Ti3+–O bonding. The calculated percentage concentration of Ti3+ is about 21% in Ti ions. This implies the existence of oxygen vacancies in TiO2 layer.

Fig. 6
figure 6

Narrow-scan XPS spectra from trilayer-structure of HfO2/TiO2/HfO2 on TiN-coated Si. a Hf 4f, b Ti 2p peaks of HfO2/TiO2/HfO2. O 1s peaks of c HfO2 and d TiO2 layers

Full size image

The O 1s spectra from HfO2 and TiO2 layers can also be deconvoluted into two peaks, as seen in Fig. 6c, d. The relatively lower binding energies of the O 1s peak at ~530.9 and 531.2 eV are assigned to Ti-O and Hf-O bonding in TiO2 and HfO2 layers, respectively, belonging to lattice oxygen without oxygen vacancies. Whereas the slightly higher energy of 532.1 eV in the O 1s spectra of Fig. 6c, d are ascribed to the oxygen atoms near oxygen vacancies in HfO2 and TiO2 layers based on the literature reports [22,23,24,25]. The relative oxygen vacancy concentration in the oxide layer can be roughly evaluated by calculating the area proportion of each peak [22, 23]. The calculated percentage concentration of oxygen vacancy in HfO2 and TiO2 layers is about 2.3 and 19.4%, respectively, in accord with the results of Hfn+ and Ti3+.

Figure 7a, b shows the XPS depth profiles of HfO2/TiO2/HfO2 samples on Pt- and TiN-coated Si by Ar ion etching, respectively. The trilayer-structure of HfO2/TiO2/HfO2 on Pt- and TiN-coated Si can be recognized easily, although the significant interfacial diffusion between HfO2/TiO2 and HfO2/TiN has been observed. Usually the filament model of oxygen vacancy migration dominates the RS behavior in RRAM devices based on transition metal oxides [11]. However, simply increasing the oxygen vacancies content is not fully effective. How to effectively control the distribution of oxygen vacancy filaments is a key issue to finally improving the RS uniformity [20]. Lots of work has shown that usually a nonuniform distribution of oxygen vacancies is beneficial to the RS behaviors, including decreasing the forming voltage, improving the switching stability and endurance ability [1, 14, 20, 26,27,28,29]. An initial nonuniform distribution of oxygen vacancies in a storage oxide layer is often obtained by the use of a chemically active electrode with a relatively high oxygen affinity (e.g., Ta, Ti, Al, or TiN) or by deliberately introducing an oxygen vacancy-rich interfacial layer by interface engineering [1, 17, 30].

Fig. 7
figure 7

XPS depth profiles of HfO2/TiO2/HfO2 on Pt- and TiN-coated Si by Ar ion etching. a HfO2/TiO2/HfO2 on Pt-coated Si. b HfO2/TiO2/HfO2 on TiN-coated Si. The depth distribution of oxygen vacancy concentration (V O 2+/O) determined from XPS spectra for HfO2/TiO2/HfO2 on Pt- and TiN-coated Si. c HfO2/TiO2/HfO2 on Pt-coated Si. d HfO2/TiO2/HfO2 on TiN-coated Si

Full size image

Figure 7c, d presents the distribution curves of oxygen vacancy concentration of HfO2/TiO2/HfO2 on Pt- and TiN-coated Si based on the above depth XPS profiles, respectively. The oxygen vacancy concentration was evaluated using the method mentioned above. It can be found that the highest oxygen vacancy concentration (~34%) appears in TiO2 intermediate layer. Another noteworthy feature is that the underlying HfO2 layer near Pt or TiN bottom electrodes has higher oxygen vacancy concentration than upper HfO2 layer (~2%). Besides, the oxygen vacancy concentration (~9%) of underlying HfO2 layer on TiN-coated Si is obviously higher than that (~6%) on Pt-coated Si.

Figure 8 shows the schematic diagrams of electroforming and reset of Pt/HfO2/TiO2/HfO2/Pt at positive and negative bias voltages. For chemically inert Pt top and bottom electrodes, the bias polarity-dependent electroforming phenomena are related to the O2 gas released from the oxide layer at the anode as a product of electro-reduction. As known, during electroforming process, oxygen vacancies are created by high electric field and migrate to the cathode, forming localized conducting filaments in the oxide layer, i.e., O O →V O 2+ + O 2−, V O 2+ + 2e −→V O [12] (Fig. 8b, e). Simultaneously, O2− ions drift towards the anode, releasing their charge and evolving O2 gas, i.e., O 2−→1/2O 2  + 2e − (Fig. 8b, e), which possibly cause physical deformation of the cell and even formation of crack or hole [31]. When applying positive forming voltage to Pt top electrode, we postulate the O2 gas released above the upper HfO2 film easily escapes at the top electrode edge or through small nanopore in Pt top electrode (Fig. 8b), which causes weak physical deformation. Meanwhile, considering the relatively higher oxygen vacancy concentration in underlying HfO2 (~6%) nearby Pt bottom electrode than the upper HfO2 layer (~2.3%) (Fig. 7c), the conductive filament readily forms, leading to smaller positive forming voltage of +7 V. With reverse voltage of −0.8 V, the reverse reaction leads to the filament ruptures, and the device is easily switched back to the HRS state (Fig. 8d).

Fig. 8
figure 8

Schematic diagrams of electroforming and reset of trilayer-structure of HfO2/TiO2/HfO2 with symmetric Pt top and bottom electrodes. a–d Electroforming at positive voltage and reset at negative voltage. e–g Electroforming at negative voltage and reset at positive voltage

Full size image

Conversely, with a negative forming voltage on the top electrode, oxygen gas forms under the underlying HfO2 layer and above Pt the bottom electrode (Fig. 8e). The O2 releasing becomes more difficult, which will impede the formation of conductive filaments. Moreover, because the oxygen vacancy concentration of the upper HfO2 layer (~2.3%) is lower than that of the underlying HfO2 (~6%) (Fig. 7c), the higher negative forming voltage of −11 V is needed to form filaments. When more O2 accumulates to a certain pressure under underlying HfO2, it possibly erupts from the mechanically weakest part of the thin films, leading to the hole in oxide films or separation between oxide films and the bottom electrode induced by bubble cracking. Although the device can be reset to HRS at applied +4 V bias (Fig. 8g), the memory cell is degraded after several switching cycles. In our samples, a part of top electrode was blown off after the forming process. Similar electroforming polarity preference scenario has also been observed in Pt/TiO2−x /Pt bipolar RRAM cells with various physical deformation evidences of the junctions [31, 32].

Pt/HfO2/TiO2/HfO2/TiN devices exhibit quite different electroforming polarity from Pt/HfO2/TiO2/HfO2/Pt. The LRS caused by electroforming can be observed at both negative and positive bias voltage; however, the effective reset from LRS to HRS can be achieved only at positive bias voltage in Pt/HfO2/TiO2/HfO2/TiN device. Similar electroforming preference with asymmetric Pt and TiN electrodes has been observed in some literatures with various storage oxide layers such as HfO2 [33, 34], TiO2 [35], ZrO x /HfO y bilayer [22], and Al2O3/HfO2/Al2O3 trilayer [14]. However, the related explanations are divergent or lacking.

After considering the TiN electrode’s chemical activity with oxygen [1, 30, 36] and the nondistribution of oxygen vacancy concentration in trilayer-structure of HfO2/TiO2/HfO2 based on the XPS depth profiles (Fig. 7), a possible mechanism on electroforming polarity preference of Pt/HfO2/TiO2/HfO2/TiN cells is proposed. Figure 9 shows the schematic diagrams of electroforming and reset of Pt/HfO2/TiO2/HfO2/TiN at negative and positive bias voltages. The TiN bottom electrode plays a key role in the electroforming polarity. Kwak et al. reported that the relatively active TiN electrode would easily absorb oxygen ions from oxide films to form TiO x N1−x (TiON) interfacial layer [36]. The severe oxygen diffusion of underlying HfO2 layer into TiN bottom electrode has been confirmed in our sample by the XPS depth profile (Fig. 7b). For Pt/HfO2/TiO2/HfO2/TiN device, the TiN electrode with high oxygen affinity [34, 36] produces a lot of oxygen vacancies in the underlying HfO2 layer near the TiN bottom electrode. The oxygen vacancies concentration of ~9% of underlying HfO2 layer is much higher than that of ~6% in the Pt/HfO2/TiO2/HfO2/Pt device.

Fig. 9
figure 9

Schematic diagrams of electroforming and reset of trilayer-structure of HfO2/TiO2/HfO2 with asymmetric Pt top electrode and TiN bottom electrode. a–d Electroforming at negative voltage and reset at positive voltage. e–g Electroforming at positive voltage and reset at negative voltage

Full size image

For chemically inert Pt top electrode and relatively active TiN bottom electrode, when applying negative forming voltage, O2− ions drift towards the TiN anode and are absorbed to form TiO x N1−x layer (Fig. 9b), i.e., TiN + xO2−→TiO x N 1−x  + x2e, avoiding the O2 release and possible damage to cells. The TiN electrode might act as a reservoir for oxygen ions drifting under an applied voltage [1, 36]. Simultaneously, due to the higher oxygen vacancy in the underlying HfO2 layer (~9%) and intermediate TiO2 layer (~10–34%) than the upper HfO2 layer (~1.9%), the migration of the rich oxygen vacancies from the underlying HfO2 layer through TiO2 layer towards the the upper HfO2 layer, directly producing conductive oxygen vacancy filaments with a smaller forming voltage of −3.7 V (Fig. 9b). When performing reverse reset process at +1.5 V, the bottom electrode of TiN layer acting as oxygen reservoir supplies oxygen ions to react with oxygen vacancy, i.e., O2− + V O 2+→O O (lattice oxygen), which is beneficial to the reset operation from LRS to HRS with rupture of conductive filament (Fig. 9d). Above all, the oxygen reservoir effect of TiN bottom electrode contributes the forming voltage reduction as well as better set/reset voltage variation [1, 29, 36].

On the other hand, when imposing the positive forming voltage on top electrode Pt, the oxygen vacancies migrate from the upper HfO2 into the underlying HfO2 on TiN, accumulate in its vicinity, and develop into filaments. Owing to the lower oxygen concentration in the upper HfO2 (~1.9%) than the underlying HfO2 layer (~9%) and intermediate TiO2 layer (~10–34%), this leads to further increase of the oxygen vacancy concentration in underlying HfO2 and intermediate TiO2 layer. In the same time, due to slight oxygen existence in PEALD-derived TiN films (about 10%), some oxygen ions in TiN drift into underlying HfO2 layer to form insulate lattice oxygen with nearby oxygen vacancy, preventing from the growth of conductive filaments (Fig. 9e). Finally, when the bias voltage attains +14 V, the coarsening conductive filaments in oxides is formed. However, when a negative voltage is applied on the device, the conductive filament in trilayer structure is too large to be ruptured (Fig. 9g). Therefore, the device cannot switch to HRS by applying a negative voltage, indicating that an irreversible hard breakdown occurs in Pt/HfO2/TiO2/HfO2/TiN device.

Finally, on account of the fact that the inserted TiO2 layer stores more oxygen vacancies than HfO2 layer, the distribution of oxygen vacancies in trilayer structure is not uniform, especially in the two interfacial layers between HfO2/IL/TiO2/IL/HfO2, which might affect the growth position, direction, and overlapping of conductive filaments. The linkage or rupture of the conductive filaments corresponds to the set process from HRS to LRS or the reset process from LRS to HRS, respectively, which more easily happens in two interfacial layers. Further, the shape and position of the conductive filaments in HfO2 and TiO2 layers change less in the set and reset processes.

Conclusions

In summary, RRAM devices based on trilayer-structure of Pt/HfO2/TiO2/HfO2/Pt and Pt/HfO2/TiO2/HfO2/TiN have been prepared by ALD. Both memory cells show typical bipolar resistive switching characteristics, and Ohmic and SCLC dominant conduction mechanisms in LRS and HRS, respectively. It is found that the bottom electrodes of Pt and TiN have great influence on the electroforming polarity preference, the ratio of high and low resistances and dispersion of the operating voltage of trilayer-structure memory cells. Compared to with symmetric Pt top/bottom electrodes, the RRAM cells with asymmetric Pt top/TiN bottom electrodes show smaller negative forming voltage of −3.7 V, relatively narrow distribution of the set/reset operation voltages and lower ratio of high and low resistances of 102. The electrode-dependent electroforming polarity can be explained by considering electrodes’ chemical activity with oxygen, the related reactions at anode, and the nonuniform distribution of oxygen vacancy concentration in trilayer-structure of HfO2/TiO2/HfO2 on Pt- and TiN-coated Si. Furthermore, the TiN electrode as oxygen reservoir plays an important role in forming voltage reduction and better dispersion of RS parameters for Pt/HfO2/TiO2/HfO2/TiN devices. Considering the modulation effect of electrode and trilayer-structure on resistive switching performance, this work provides a new device design route for future RRAM applications.