Introduction

Recently, resistive random access memory (RRAM) has attracted much attention because of its potential to replace three-dimensional (3-D) flash memory in future1,2,3. In this regard, many research groups have proposed different switching materials like binary metal oxides and transition metal oxide4 such as Ta2O5 3, 5, TiO2 6, HfO2 7, etc. Among them, the perovskite oxides such as SrTiO3 8, 9, SrZrO3 10 and BaTiO3 11, 12 have drew enormous attention towards the application since last decade. Importantly, BaTiO3 has high dielectric constant of 100–60013 and large band gap of 3.42 eV14, which is one of the potential resistive switching materials. The demand of high density data storage can effectively be achieved by multi-level resistive memory cell. Multi-level resistive switching operation of five resistance states using different materials in ITO/RGO/ITO15 and TiN/Ta2O5/Pt16 structures have been demonstrated. Several research groups have reported multi-level switching operation with multiple states for brain-inspired neuromorphic applications17, 18. However there is no report on tunable multi-level resistive switching characteristics of the BaTiOx switching material in Cr/BaTiOx/TiN structure and its transport mechanism in each level is not reported yet. In addition, quantum conductance occurs due to movement of the oxygen vacancies19 when the contact point of filament is reduced to atomic scale. Nowadays, this phenomenon is at the center of attraction due to its possible application in multi-level and neuromorphic resistive memory20,21,22. Chen et al.19 have reported anion migration based quantum conductance in Ti/Ta2O5/Pt structure. Younis et al.20 have reported voltage sweep rate dependent quantum conductance in Au/SnO2–CeO2/FTO structure. In this report, negative voltage dependent quantum conductance in a novel Cr/BaTiO3/TiN structure has been reported for the first time. The oxidation-reduction (redox) process and change in oxidation state of Ba is responsible for multi-level and quantum conductance phenomenon. The staircase oxidation of Ba in switching material, BaTiO3 is justified by the sensing of hydrogen-peroxide (H2O2) in electrolyte-insulator-semiconductor (EIS) structure, which is also completely novel approach presented in this paper.

The negative stop voltage modulated multi-level resistive switching in Cr/BaTiOx/TiN structure is observed due to gradual dissolution of oxygen vacancy filament. Quantum conductance is observed at staircase RESET in which the experimental result fits very well with the simulated curve. The X-ray photoelectron spectroscopy shows oxidation states of Ba and Ti. The oxidation-reduction of Ba is responsible for resistive switching mechanism and multi-level resistance states are due to more generation of Ba2+ ions under staircase RESET. The rate of dissolution of filament with negative stop voltage i.e. the increase of high resistance state (HRS) with negative stop voltage is uniform and controllable moderate value of 217.39 mV/decade is obtained for lower thickness (2.5 nm) of BaTiOx. The devices with 0.4 × 0.4 μm2 size exhibit better resistive switching than devices with 4 × 4 μm2 size. In both positive and negative voltage cycles of low resistance state (LRS), the Ohmic conduction is observed in low field where as the hopping conduction is observed in high field for both devices. In HRS, Poole-Frenkel and hopping conduction are observed in moderate and high field, respectively. Moreover, the Fowler-Nordheim tunneling is observed in very high negative voltage. The switching mechanism including multi-level operation and quantum conductance in staircase RESET is explained through evidence of H2O2 sensing with concentration of 1 nM to 1000 nM in electrolyte/BaTiOx/SiO2/p-Si structure. In addition, the BaTiOx membrane shows good sensitivity of 48 mV/pH. The devices also exhibit high resistance ratio of 2000, high speed program/erase endurance of more than 107 cycles with 100 ns pulse width and 3 hours data retention at 85 °C. This unique presentation of switching mechanism through H2O2 sensing shows a path towards combination of resistive memory and bio-sensor.

Results and Discussion

A schematic view of resistive switching memory device is shown in Fig. 1a. An optical microscope image with a size of 4 × 4 µm2 is also shown in Fig. 1b. Figure 1c shows the cross-sectional transmission electron microscopy (TEM) image with 5 nm-thick BaTiOx layer. Plane-view TEM image shows amorphous BaTiOx film (Fig. 1d). Detection of pH and H2O2 is performed by using electrolyte/BaTiOx/SiO2/p-Si structure, which is shown in Fig. 2 schematically.

Figure 1
figure 1

(a) Schematic view of the device and measurement setup with Agilent 4156C. (b) Optical microscope (OM) image of a device. (c) Cross-sectional HRTEM image confirms the amorphous BaTiOx layer. (d) Plane-view TEM image confirms amorphous BaTiOx SM also.

Full size image
Figure 2
figure 2

Schematic view of pH and H2O2 sensing using BaTiOx membrane in EIS structure.

Full size image

XPS characteristics

Figure 3 shows the XPS analysis of the switching materials (SMs). The doublet spectra of Ba3d 5/2 and Ba3d 3/2 are fitted at 780 eV and 795.2 eV, respectively (Fig. 3a). Forster et al.23 have reported similar binding energy peaks at 779.6 eV and 795 eV for the Ba3d 5/2 and Ba3d 3/2 peaks, respectively. The Ba 3d 5/2 peak consists of two components, which correspond to BaO at 779.8 eV and BaO2 at 781.6 eV. Droubay et al.24 have reported similar observation where the peak binding energy at 775.5 eV corresponds to BaO and the peak binding energy at 779 eV corresponds to BaO2. For both BaO and BaO2 species, Ba has ionic state of 2+. The oxidation state of oxygen (O) in BaO is 2- and it is 1- for BaO2. Figure 3b shows XPS of Ti doublet spectra at 458 eV and 464 eV, which corresponds to Ti2p 3/2 and Ti2p 1/2 peaks. These values are close to our previous reported values, 458.8 eV for Ti2p 3/2 peak and 464.4 eV for Ti2p 5/2 peak25. These Ti2p peaks correspond to TiO2, where the oxidation state of Ti is 4+26. In addition, there is no oxidation state of Ti for 3+ or 2+, where Ti2p 3/2 peaks are centered at 457.6 eV and 456.4 eV, respectively. The O1s spectrum is de-convoluted into three peaks at 529.5 eV, 531 eV and 532.4 eV, where first two peaks correspond to BaO at 529.5 eV and BaO2 at 531 eV. These values are close to the reported values of 528.9 eV for BaO and 531.1 eV for BaO2 27. Hashimoto et al.28 have reported that the energy peak centered at 528.9 eV is owing to TiO2. The energy peak entered at 532.4 eV is due to hydroxide (OH) groups on BaTiOx surface. Chu et al.29 have reported that O1s peak centered at 532.9 eV is hydroxide groups on TiO2 nanotube’s surface. From the XPS data the composition of switching material is BaTiOx (1.98 < x < 3), which shows less oxygen or oxygen vacancies in SM. The O1s peak located at 531 eV may be due to possible oxygen deficiencies in TiO2 film, which is similar to the reported peak binding energy of 531.3 eV29. By considering Gibbs free energy at 300 K (−1114.1 kJ/mol for BaO and −887.62 kJ/mol for TiO2 30), there are strong Ti-O bonds or stoichiometric TiO2 than the Ba-O bonds or BaO2, i.e., BaOx. To check thermal stability, the BaTiOx films were annealed at 450 °C, 600 °C, and 750 °C. There is negligible change of Ba oxidation state and the composition is stable up to 600 °C. The details of XPS analysis at high temperature are given in the supplementary information (Fig. S1).This will lead to good resistive switching memory characteristics as well as H2O2 sensing, which have been explained below.

Figure 3
figure 3

XPS characteristics of (a) Ba3d, (b) Ti2p, and (c) O1s spectra from the BaTiOx SM.

Full size image

Negative voltage modulated multi-level and transport mechanism

Figure 4a shows typical bipolar resistive switching characteristics of the S1 devices with size of 0.4 × 0.4 µm2. The SET and RESET voltages are 0.84 V and −1.3 V, respectively. The device is operated under ± 1.5 V. The sweeping bias is shown by arrows 1 to 4. Cumulative distributions of the leakage current and formation voltage are shown in Fig. 4b and c, respectively. Larger (4 × 4 µm2) devices of S1 show lower mean value of leakage current than that of the S2 devices (1.347 pA vs. 0.167 µA) while those values for smaller devices (0.4 × 0.4 µm2) have the similar trend (0.153 pA vs. 0.434 pA). This indicates that both the larger size devices with thinner SM have higher leakage current owing to more defect paths and shorter lengths31. Accordingly, the mean value of formation voltage is lower for the larger (4 × 4 µm2) S2 devices than the S1 devices (1.295 V vs. 2.19 V) and those values for the smaller (0.4 × 0.4 µm2) devices show a similar trend also (2.287 V vs. 4.207 V). S. Yazdanparast32 has reported the reduction of formation voltage by increasing device size for the electrodeposited cuprous oxide as switching material. The standard deviation of formation voltage for the smaller (0.4 × 0.4 µm2) S1 devices is higher than the S2 devices (0.673 V vs. 0.06 V). Therefore, the smaller size devices need thinner SM which has benefit of lower operation voltage of <2.287 V with high uniformity as well as scaling further the RRAM device. However, the current transport mechanism is one of the important issues to develop the RRAM device in future. So the I-V curves are fitted with all possible transport mechanisms (Fig. 4a). The LRS currents show Ohmic conduction at low field regions for both positive (+Ve) and negative (−Ve) sides. By plotting ln(J)-ln(E), the slope values are found to be the same 1.01 (Fig. 5a). The HRS current is fitted with Poole-Frenkel (P-F), hopping, and Fowler-Nordheim (F–N) tunneling in sequence of moderate field to high field regions. At higher field, the P-F is fitted (Fig. 5b) and corresponding value of dielectric permittivity is calculated by using equation (1) as given below33, 34,

$${\varepsilon }_{PF}=\frac{{q}^{3}}{{({k}_{B}T\times {S}_{PF})}^{2}\pi {\varepsilon }_{0}}$$
(1)

where q is the electronic charge, kB is the Boltzmann’s constant, T is absolute temperature, ε0 is free-space permittivity, and SPF is the slope of the fitted line. The εPF values are found to be 219 and 80 for both positive and negative bias, respectively and those values are in the range of reported values of 100–600 for BaTiO3 film13. Oxygen vacancies in SM play a crucial role in the electrical characteristics of the BaTiO3 films35, 36. At moderate field regions of both HRS and LRS currents, the hopping conduction is observed by fitting ln(J) vs. E curves (Fig. 5c) and corresponding hopping distance (a) is expressed as equation (2)5, 37,

$$a=\frac{{k}_{B}T}{q}\times {S}_{H}$$
(2)

where SH is the slope. The hopping distances at HRS and LRS of +Ve bias are found to be 0.69 nm and 0.29 nm, while those values are 0.47 nm and 0.21 nm for -Ve bias, respectively. P. J. Freud et al.38 have reported a hopping distance of 0.3 nm for localized charge carrier conduction through the Ni0.6Mn2.4O4 material. In our previous report34, the hopping distance of 0.56 nm was reported for IrOx/GdOx/Al2O3/TiN resistive switching memory device. The hopping distance at LRS is shorter than the value at HRS owing to thin oxygen-rich layer formation at the Cr/BaTiOx interface. For HRS, the F-N tunneling is observed at high field and the barrier height (Φb) is calculated by plotting ln(J/E2) vs. 1/E as represented in equation (3) below39,

$${\phi }_{b}=\frac{S{{}_{FN}}^{2/3}}{3.6\times {10}^{6}\times {\chi }^{1/3}}$$
(3)

where SFN is the slope, m* ( = χ  ×  m0) is the effective mass of electron, m0 is the rest mass of electron, and the value of χ is considered to be 0.4. The critical field (Ec) is 2.8 MV/cm (Fig. 5d). The typical value (>2.6MV/cm) of Ec is reported for F-N tunneling40, 41. The φb value is calculated to be 0.59 eV, which is close to the difference of Cr work function (4.5 eV42) and electron affinity of BaTiO3 (3.9 eV36). Similar conduction mechanism and negative voltage modulated multi-level operation under RESET are also observed for the S2 devices as discussed below.

Figure 4
figure 4

(a) Typical bipolar resistive switching characteristics of the S1 devices. Cumulative probability of (b) leakage current and (c) formation voltage for randomly measured many S1 and S2 devices.

Full size image
Figure 5
figure 5

Current transport characteristics of S1 device with a size of 0.4 × 0.4 µm2. (a) Ohmic conduction, (b) Poole-Frenkel (P-F), (c) Hopping conduction, and (d) Fowler-Nordheim (F-N) tunneling.

Full size image

First, the device sets and the conductance is decreased by increasing value of negative stop voltage (VSTOP) at RESET (Fig. 6), where I–V curves are plotted in log scale (Fig. 6a) and linear scale (Fig. 6b). Then VSTOP value is started −0.5 V and step voltage is −50 mV until complete RESET is achieved at −1.3 V. The current value is read out at −50 mV. The resistance value is increased from 3.15 kΩ (VSTOP = −0.5 V) to 1820 kΩ (VSTOP = −1.3 V), due to VSTOP modulated gradual RESET as well as multi-level of the device is observed. The multi-level operation can easily be realized in resistance vs. VSTOP plot for both the S1 and S2 devices (Fig. 7). A numerical parameter is introduced which is ratio of change in VSTOP to change in resistance (mV/decade), which is similar to the subthreshold swing (SS) of a metal-oxide-semiconductor filed-effect-transistor (MOSFET). The SS values of the larger (4 × 4 µm2) sizes are found to be 375.14 (Fig. 7a) and 215.51 (Fig. 7b) mV/decade, while those values of the smaller (0.4 × 0.4 µm2) sizes are found to be 84.74 (Fig. 7c) and 217.39 (Fig. 7d) mV/decade for the S1 and S2 devices, respectively. Higher SS value is needed to design multiple resistance states. From those SS values, it is inferred that larger size device with thicker SM or smaller size device with thinner SM is useful for multi-level operation. On the other hand, smaller device size with thicker SM has larger filament diameter because of hard breakdown of SM. This will have larger value of RESET voltage as compared to the other devices (−1.6 V vs. −1.3 V). Therefore, the S2 devices with smaller size of 0.4 × 0.4 µm2 have higher SS value that will give an opportunity for tuning multi-level high resistance states, even the smaller SM thickness of 2.5 nm has been used. Ten states with resistances of approximately 3.2, 4.3, 7.2, 17, 42, 93, 128, 288, 685, and 1700 kΩ are obtained. Typical data retention characteristics for LRS and four HRS at VSTOP of −0.7 V, −0.8 V, 0.9 V and −1 V are also shown in Fig. 8a. These resistance states are stable. Similarly, the LRS value is decreased with increasing current compliances from 300 µA to 3 mA (Fig. 8b) because of the increment of filament diameter43. The HRS value is also increased with increasing CC owing to generation of more BaO2 in BaTiOx SM. A high resistance ratio ranging from 34 to 3200 are obtained with CCs from 300 µA to 3 mA. To explore the transport mechanism further in multi-level operation (Fig. 6), I–V curves have been fitted at different VSTOP ranging from −0.5 V to −1.3 V. The fitting parameters are listed in Fig. 9. In the segment I, when the VSTOP value is in the range of −0.5 V to −0.7 V, the slope value is 1.12–1.16, which is owing to Ohmic conduction. In segment II, when the VSTOP values are from −0.75 V to −1.1 V, P-F emission at mid field region (−0.08 V to −0.46 V) and hopping conduction at high field region (−0.48 V to −1.06 V) are observed. The slope values are >1.2, which indicates no Ohmic conduction at moderate field regions. The εPF values are found to be 308 to 96 and ‘a’ values are from 0.31 nm to 0.47 nm. When the applied VSTOP is in the range from −1.15 V to −1.3 V (segment III), the εPF values are from 117 to 163, ‘a’ values are from 0.35 nm to 0.40 nm, and the Φb values are from 0.63 eV to 0.76 eV. Therefore, the transport mechanism has been modulated by values of VSTOP. By applying bias of more than SET voltage, the oxygen ions (O2−) migrate towards Cr TE and leaving oxygen vacancy from TiN BE forming conducting filament (CF) and an oxygen-rich layer at the Cr/BaTiOx interface is also formed. On the other hand, the CF is dissolved in stair-case by applying negative bias of less than RESET voltage. This shows multiple states because of gradual oxidation of CF or generation of more Ba2+ ions, which has been explained by quantum conductance through evidence of H2O2 sensing as follows.

Figure 6
figure 6

One time SET and staircase RESET by varying negative stop voltage of the S2 devices with a size of 0.4 × 0.4 µm2. I-V characteristics using (a) log and (b) linear scales.

Full size image
Figure 7
figure 7

Variation of HRS with negative stop voltage of the S1 devices with (a) 4 × 4 µm2 and (b) 0.4 × 0.4 µm2. The S2 devices with (c) 4 × 4 µm2 and (d) 0.4 × 0.4 µm2.

Full size image
Figure 8
figure 8

(a) Data retention for multi- levels corresponding to stop voltages of −0.7 V, −0.8 V, −0.9 V and −1 V. (b) Current compliance dependent multi-level resistance changes in both LRS and HRS.

Full size image
Figure 9
figure 9

Parameter obtained from current transport characteristics under RESET. Typical I-V curve changing negative voltage of the S2 devices with a size of 0.4 × 0.4 µm2.

Full size image

Quantum conductance and H2O2 sensing mechanism

An interesting phenomenon is observed by arranging multi-level I-V curves under gradual RESET of voltage ranging from −0.65 V to −1.05 V (Fig. 6). We can see that the RESET process is consisted of several abrupt current changes, as shown in the inset of Fig. 10a. The conductance (G = I/V) has been calculated in terms of quantum conductance phenomenon, G0 ( = 2q2/h), where h is Plank’s constant (h = 6.62 × 10−34 m2 kg/s). Figure 10a shows conductance vs. negative voltage which is changed in a staircase fashion. The values of conductance at −0.8 V, −0.85 V, −0.9 V and −0.95 V are found to be 10G0, 9G0, 7.5G0, and 6.5G0, respectively. So the conductance values are integer or half-integer multiple of G0 21, 22. The step voltage is higher (50 mV) than the thermal energy of 26 meV ( = kBT/q) at 300 K. The I-V curves of staircase RESET using quantum conductance phenomenon have been simulated. The conductance is decreased by multiple of G0 in the interval of 50 meV (Fig. 10a), which is owing to generation of Ba2+ ions. To investigate the ionic states of the BaTiOx SM, the EIS structure was fabricated (Fig. 2). The BaTiOx sensing membrane shows good pH sensitivity of 48 mV/pH and linearity of 99.15% from pH values of 2 to 10 (Fig. 10b). Similar pH sensitivity values by using EIS structure are also reported with different sensing membranes such as 52.3 mV/pH for HfO2 44, 56 mV/pH for TiO2 45, 55 mV/pH for Gd2O3 46, 56 mV/pH for Ta2O5 47, 57.1 mV/pH for Al2O3 48. Basically, the C-V curves are shifted towards positive direction with increasing pH value owing to deprotonation of BaTiOx surface or adsorption of OH− ions in the sensing membrane. On the other hand, the C-V curves are shifted towards negative direction owing to protonation (H+) of the BaTiOx surface. Therefore, protonation/deprotonation49 of a sensing membrane shows pH sensitivity or reference voltage changes. The higher pH sensitivity of BaTiOx membrane is observed than the bare SiO2 (48 vs. 35 mV/pH50) and comparable pH sensitivity (50 mV/pH) of TiO2 membrane (not shown here) owing to different oxidation states of Ba+ and Ba2+ 42. The oxidation-reduction (redox) of the sensing membrane is observed by measuring H2O2 and switching mechanism is explored as well. In contact of H2O2, the Ba surface is oxidized to Ba2+ (2BaO + O2 → 2BaO2 27) or Ba changes to Ba2+ ions, and provides two electrons (e−) after H2O2 reduction. Possible surface reactions at Ba-sites are shown below.

$$B{a}^{o}\leftrightarrow B{a}^{2+}+2{e}^{-}$$
(4)
$${H}_{2}{O}_{2}\leftrightarrow O{H}^{-}+O{H}^{\ast }$$
(5)
$$O{H}^{\ast }+{e}^{-}\leftrightarrow O{H}^{-}$$
(6)
$$O{H}^{-}+{H}^{+}\leftrightarrow {H}_{2}O$$
(7)
Figure 10
figure 10

(a) Experimental and simulated conductance curve as a function of voltage. Inset shows the I–V of the RESET region. (b) C–V characteristics with pH values of 2 to 10. (c) Reference voltage shift with H2O2 concentration from 1 nM to 1000 nM. (d) Experimental and simulated I-V curves.

Full size image

From equations (4)–(7), the generation of Ba2+ ions increases with increasing H2O2 concentration from 1 nM to 1000 nM (Fig. 10c). By considering doping of 1015 cm−3 in p-Si, the work function is approximately 4.9 eV whereas the work function of Ba is 2.52–2.7 eV42. Droubay et al.24 have reported that the work function of BaO2 on Ag(001) is higher than the value of BaO (3.5 eV vs. 2.5 eV). This suggests that the work function of Ba increases after oxidation as well as the work function difference in between Si and BaO is reduced with increasing H2O2 concentration or Si band bending is reduced with H2O2 concentration. Due to work function modulation of BaO by oxidation/reduction, the reference voltage shifts towards positive direction with increasing H2O2 concentration as well as energy band bending of Si is decreased. Therefore, the reference voltage is needed to have flat-band of Si. This implies that the Ba2+ ion increases as well as work function increases with increasing the H2O2 concentration. On the other hand, the work function of Ti decreases after oxidation (ФTi = 4.33 eV42 and ФTiO2 = 4.13 eV51). Both TiO2 and SiO2 membrane do not show H2O2 sensing because of no oxidation/reduction of those membranes (Fig. 10c) due to strong bonding of Ti-O and Si-O. The reference voltage shift increases linearly with increasing H2O2 concentration from 10 nM to 1000 nM. However, pH value of 7 is unchanged during successive addition of H2O2. The BaTiOx membrane detects also a low concentration of 1 nM with reference voltage shift of 6 mV, which is comparable with the reported H2O2 sensing of 2 to 10 nM by using different materials and methods52,53,54. From this H2O2 sensing, the oxidation of a single Ba atom confirms the evidences of the quantum conductance under RESET. The experimental and simulated I-V curves are given in Fig. 10d. The regions with different conduction mechanism are denoted by different colors in the simulated I-V curve. The parameters taken in the simulation i.e. the value of mobility of electron (μ = 2 × 104 m2/V.s, close to reported value of 1 × 104 m2/Vs40), density of state of conduction band (Nc = 8 × 1024 m−3, close to reported value of 8.5 × 1024/m[3 40), permittivity of BaTiO3 (εr = 85), hopping distance (a = 0.6 nm), frequency of thermal vibration (ν = 1013 Hz41), activation energy (EA = 0.5 eV, close to the reported value of 0.8 eV42), and barrier height of 0.7 eV have been used in MATLAB simulator. Considering negative voltage modulated transport mechanisms and quantum conductance under staircase RESET, simulated and experimental I-V curves are filled well. This is very useful way to understand switching mechanism as well as multilevel operation of resistive switching phenomena. In addition, the S2 device shows excellent 1000 consecutive repeatable DC cycles, high speed (100 ns) program/erase endurance more than 107 cycles (Fig. S2), and multi-level data retention at 85 °C (Fig. S3). The time-response of H2O2 sensing is also shown in Fig. S4, which promises repeatedly used of the sensor (supplementary information). This negative voltage modulated multi-level resistive switching with high resistance ratio and understanding of resistive switching mechanism of quantum conductance through H2O2 sensing will show a path towards solution for high density 3D multi-level cross-point memory application.

Conclusion

In summary, the negative voltage modulated multi-level resistive switching and quantum conductance have been reported in simple Cr/BaTiOx/TiN structure. Amorphous nature of BaTiOx switching material has been confirmed by the cross-sectional and plain view HRTEM. Different oxidation states of Ba are revealed from XPS. The switching characteristics follow the P-F and hopping conduction mechanism in HRS whereas Ohmic and hopping conduction have been observed in LRS. It is interesting to note that F-N conduction in very high field of negative cycle of HRS and quantum conductance in staircase RESET has been observed. For the realization of multi-level resistive switching, change in resistance with VSTOP voltage has been demonstrated and quantized by a parameter in the unit of mV/decade. The device with 2.5 nm thickness of BaTiOx and 0.4 × 0.4 µm2 size exhibits a tunable, moderate change of HRS with stop voltage (217.39 mV/decade). By exploring the conduction mechanism of each HRS corresponding to each VSTOP voltage, the multi-level phenomenon has been explained through a gradual dissolution of oxygen vacancy filament. The devices also show current compliance dependent multi-level operation. Quantum conductance and multi-level switching phenomenon have been explained through evidence of H2O2 sensing mechanism. The BaTiOx sensing membrane shows low concentration of H2O2 detection (1 nM) and good pH sensitivity of 48 mV/pH. The device with 2.5 nm thickness of BaTiOx shows also high speed program/erase endurance of 107 cycles with 100 ns pulse width and data retention of more than 3 hours at 85 °C. Oxygen migration based tunable multi-level RRAM with 2000 resistance ratio has been reported, which will be useful for high density 3-D cross-bar application. The explanation of switching mechanism using H2O2 sensing of the same material indicates the possibility of using memory and bio-sensor in bio-memory chip.

Methods

Memory device fabrication

At first, 8-inch Si wafer was cleaned by Radio Corporation of America (RCA) process. Then, 200 nm-thick SiO2 film was grown by thermal oxidation. A 160 nm-thick Ti as an adhesive layer and 40 nm-thick TiN as a bottom electrode (BE) were deposited on SiO2/Si substrate. Then, 150 nm-thick SiO2 as an insulating layer was deposited on BE. This SiO2 layer was patterned and etched out to create via-holes with size ranging from 0.4 × 0.4 µm2 to 4 × 4 µm2. The BaTiOx as a switching material (SM) and 200 nm-thick Cr as a top electrode (TE) were deposited by rf sputtering. Both BaTiO3 and Cr targets were used. The deposition was done in the same sputtering system. Deposition parameters such as power, Ar flow rate, and deposition pressure were 100 W, 10 sccm, and 6 mTorr, respectively for both materials. The SMs with different thicknesses of 5 nm (S1) and 2.5 nm (S2) were deposited in a Cr/BaTiOx/TiN structure. Finally, lift-off process was performed to get the Cr/BaTiOx/TiN device, as shown in Fig. 1a. The electrical characteristics were obtained by applying bias on the TE, whereas the BE is connected to ground. Memory characteristics were measured using Agilent 4156C and Agilent B1500A semiconductor parameter analyzers.

H2O2 sensor fabrication

To confirm the oxidation states of the BaTiOx SM, the H2O2 sensing was performed by using electrolyte-insulator-semiconductor (EIS) structure. First, 4-inch p-type Si wafer was cleaned by RCA process. Then, the wafer was dipped into dilute HF solution and the SiO2 layer with a thickness of 40 nm was deposited by hot horizontal furnace at a temperature of 950 °C. Oxygen gas was used during growth of SiO2. Backside SiO2 layer was removed by using buffer oxide etching solution (BOE). Then, aluminum (Al) with a thickness of 200 nm was deposited on backside by thermal evaporation. Post-metal annealing (PMA) was performed in a hot horizontal furnace at a temperature of 450 °C for 10 min. Nitrogen gas with a flow rate of 2.5 SLM (standard liter per minute) was used during annealing. Then, the BaTiOx sensing membrane with a thickness of approximately 2 nm was deposited on 40 nm-thick SiO2 layer by using the same deposition recipe of the SM. For comparison, the TiO2 and bare SiO2 membranes in EIS structure were also deposited. The sensing area was defined by lithography. Negative photo-resist, SU8 was used. The sensing area was 3.14 mm2. Then, one sensor was cut properly and fixed on Cu printed circuit board by using silver (Ag) paste. Epoxy was used to isolate in between Cu electrode and sensing area. Schematic view of a sensor is shown in Fig. 2. A reference electrode was used during capacitance-voltage (C–V) measurement in aqueous solutions with pH values from 2 to 10. The C-V characteristics with measurement frequency of 100 Hz were measured by using Agilent HP 4184 A LCR meter. The reference voltage was measured at capacitance (C) of 0.6Cox, where Cox is accumulation capacitance.

pH and H2O2 solution preparation

Sodium phosphate monobasic monohydrate (NaH2PO4.H2O), sodium phosphate dibasic anhydrous (Na2HPO4), sodium chloride (NaCl) were purchased from J. T. Baker Co. Ltd, (PA, USA) and hydrogen chloride (HCl) was purchased from Sigma-Aldrich (USA). Different pH buffer solutions (pH 2 to pH 10) were bought from Alfa- Aesar Co. Ltd (MA, USA). Hydrogen peroxide (H2O2, 31%) was purchased from BASF Co. Ltd. (Taipei, Taiwan). Detection of H2O2 was carried-out in phosphate buffer solution (PBS, 5 mM, pH 7) and the solution was prepared by adjusting the pH by 0.1 mol/l HCl in mixing appropriate amount of Na2HPO4 and NaH2PO4. To increase the ionic strength of solution, an amount of 4.6-gm of NaCl was dissolved in to buffer solution. The capacitance-voltage response was performed in 5 ml of PBS solution and further the concentration range of H2O2 from 1 nM to 1000 nM was increased by successive addition from stock solution of 1 µM H2O2. After preparation of each concentration of H2O2, pH value was checked by market available pH meter and the pH value of 7 was unchanged throughout H2O2 mixing in PBS solution. To obtain repeatable data, every concentration was measured by using three different sensors.