• 沒有找到結果。

Chapter 2 Experimental Details

2.3 Electrical Analyses

2.3.4 Nondestructive Readout Measurement

The nondestructive readout measurement is applying a DC voltage bias to read the current of various memory devices to estimate the maximum read time and the stability of the memory states. The nondestructive readout measurement is measured by Agilent 4156A.

Chapter 3

Result and Discussion

3.1 Physical Analysis of Multi-layer Resistive Switching Film

We show our physical analysis result in this section. The reasons why we use glass for substrate and ITO for electrode are making fully transparency device. It is named transparent resistive random access memory (TRRAM). In order to investigate how resistive switching works in amorphous oxide layer, we design our resistive switching layer into a series of multi-layer oxide insulator by Atomic layer deposition. A ratio 6:1 for HfO2 : Al2O3 has been chosen and make it to 38-cycle to reach about 25nm oxide thickness. Because this sample changes its material 76 times in about 25nm, it is not enough thickness for getting crystallized. Then we double the thickness of each cycle but deposited half cycles of the multi-layer insulator.

Then do it again. Here comes our three samples, Sample 1 (HfO2:Al2O3=6:1, 38-cycle, 260 Å), Sample 2 (HfO2:Al2O3=12:2, 19-cycle, 260 Å) and Sample 3 (HfO2:Al2O3=24:4, 9-cycle, 260 Å). In this thesis, we named these samples by its cycles, 9-cycle, 19-cycle and 38-cycle. After depositing the resistive switching layer, we use DC sputtering to deposit 100 nm ITO as the top electrode of resistive memory devices. Also we use the tungsten probe as top electrode of resistive memory devices. In this thesis, we discuss the tungsten probe top electrode memory device mostly.

The transmittances of the three samples are measured in the visible light region from 400 nm to 800 nm wavelength. Fig. 3.1.1. The transmittance including the substrate is approximately 81% averaged, maximum: 86%, minimum: 76%.

The cross-section HR-TEM images of the three samples with ITO top electrode are shown in Fig. 3.1.2, Fig. 3.1.3 and Fig. 3.1.4. Obviously, we can see the multi-layer configuration of 9-cycle and 19-cycle in Fig. 3.1.2 and Fig.

3.1.3. In addition, because of each Al2O3 layer has only one monolayer in 38-cycle, it has no interface configuration in oxide layer as we expected before and is shown in Fig. 3.1.4. Each sample demonstrates analogous amorphous phase type. Moreover, the bottom electrode ITO films showed nearly polycrystalline configuration. The oxide layer thickness of each sample is about 26nm.

The surface roughness is measured by AFM, as shown in Fig. 3.1.5, Fig.

3.1.6 and Fig. 3.1.7. The mean roughness of the samples is about 2~3 nm. As a result, 9-cycle has the smoothest surface. And the Fig. 3.1.8 shows the XRD patterns of these samples. The four peaks at 30.2°, 35.3°,50.6°, and 60.1° are caused by the bottom electrode ITO film. Therefore, according to these XRD patterns the multi-layer resistive switching layers are amorphous films, as we expected before.

Fig. 3.1.1 Optical transmittance spectrum of 9-, 19- and 38-cycle devices

400 500 600 700 800

0

ITO/HfO2+Al2O3...(19 cycle)/ITO/Glass

Transmittance(%)

Wavelength(nm) (b)

400 500 600 700 800

0

ITO/HfO2+Al2O3+...(9 cycles)/ITO/Glass

Transmittance(%)

Wavelength(nm) (a)

400 500 600 700 800

0

Fig. 3.1.2 The cross section HR-TEM images of ITO/9-cycle/ITO memory devices

ITO

ITO

9-cycle

Fig. 3.1.3 The cross section HR-TEM images of ITO/19-cycle/ITO memory devices

ITO

ITO

19-cycle

Fig. 3.1.4 The cross section HR-TEM images of ITO/38-cycle/ITO memory devices

ITO ITO

38-cycle

Fig. 3.1.5 The surface AFM images of 9-cycle

9-cycle

9-cycle

Fig. 3.1.6 The surface AFM images of 19-cycle

19-cycle

19-cycle

Fig. 3.1.7 The surface AFM images of 38-cycle

38-cycle

38-cycle

20 30 40 50 60 70 80

ITO (2 1 -5)

ITO (1 2 -4)

ITO (2 1 -2)

Intensity (a.u)

2 theta (θ)

ITO/Glass HfO2+Al2O3+...(38 cycles)/ITO/Glass HfO2+Al2O3+...(19 cycles)/ITO/Glass

HfO2+Al2O3+...(9 cycles)/ITO/Glass

ITO (0 0 3)

Fig. 3.1.8 The XRD patterns of 9-cycle/19-cycle/38-cycle

3.2 Electrical characteristics of resistive switching layer

In order to turning the insulator film into a resistive switching layer, we have to apply the forming process. In previous report [21]-[23], the forming process is defined as applying a high voltage bias to make a soft breakdown (SBD) happened to the insulator film, and this SBD switches it form original state to LRS. After forming process, multi-layer resistive switching memory device can be switched between LRS and HRS. Fig 3.2.1, Fig 3.2.2 and Fig 3.2.3 show the I-V curve of these devices. Obviously, the hysteresis can be observed for The-9-cycle devices /19-cycle/38-cycle.

Fig 3.2.1 Resistive switching behavior of 9-cycle

Fig 3.2.2 Resistive switching behavior of 19-cycle

Fig 3.2.3 Resistive switching behavior of 38-cycle

3.3 Resistive Switching Properties of Multi-layer Resistive Switching Film

In this section, we demonstrate the electrical properties of basic resistive switching characteristics, set/reset voltage distribution dispersion, retention time, endurance cycles and resistances of LRS and HRS

3.3.1 Electrical Property

At the first time we applying a negative voltage bias on the device, the forming process, filaments have been formed in the device, the device has been switched to LRS. Then a positive voltage bias is applied on the device, this reverse bias could switch the device back into the HRS, this is called the reset process. After the forming process and the first reset process, a negative voltage bias apply to the device again and this time it could make the device switch to LRS again, this is called the set process. Then we could keep doing set and reset process to make the device switch between LRS and HRS again and again.

When set process is happening the device switched to LRS. It needs a current compliance in order to protect the device from the permanent damage by sudden high current. This current compliance is also a very important

operation parameter of the device. Different compliance currents cause different LRSs. As shown in Fig. 3.3.1 , the larger compliance currents enhance hysteresis curve. This means the larger compliance current cause lower LRS, as shown in Fig. 3.3.2. The lower LRS also induce lager HRS/LRS ratio. This phenomenon might be attributed to that because lager compliance current forms stronger conducting filaments and cause the LRS becoming lower. In this reset process, the stronger conducting filaments need more energy power to rupture [24]. However, X. CaO et al. [25] believed that the switching current and voltage depend on the microstructures and stoichiometric of the materials. According to Kim et al [26], this can be related to the average power dissipated at SET and RESET processes.

The reason of filament rupture can be considered of the temperature of forming filaments, using the steady state temperature model, where the equation is given by Tm=(T04

+J2ργ/2Pw)1/4 [27]. Tm is the filaments temperature raised by Joule heating, T0 is the room temperature (=300K), J is the current density, ρ is the sample resistivity. γ is the filament radius, and PW is the radiative loss parameter of the filament. This result about the Joule heating effect is important factor for the RESET. However, there is another way to enlarge the HRS/LRS ratio. As shown in Fig. 3.3.3 and Fig. 3.3.4, the lager stop voltage can also enhance the hysteresis curve by making the HRS higher. These two phenomena help us to decide what compliance current and stop voltage we should chose for operating the device. And these two phenomena had been discovered by HY. Lee et al [11].

We chose 500 uA for compliance current and 5 V for stop voltage to

endurance experiment and read HRS/LRS resistances by 0.1 V. The results of these devices are shown in Fig. 3.3.5, Fig. 3.3.6 and Fig. 3.3.7. Obviously, these three kinds of devices have very different result. The-9-cycle devices has the best performance over 1250 cycle times. The 19-cycle could be operated to about 950 cycle times. And the 38-cycle has a result about 300 cycle times. Therefore, the set/reset voltage distributions are shown in Fig.

3.3.8, the average set voltage and the standard deviation are -1.258 V, 0.162 V (9-cycle), -1.927 V, 0.92 V (19-cycle) and -1.706 V, 0.932 V (38-cycle), the average reset voltage and the standard deviation are 1.275 V, 0.193V (9-cycle), 1.867 V, 0.375 V (19-cycle) and 1.412 V, 1.131 V (38-cycle).

Obviously 9-cycle has the tighter distribution of set/reset voltage and the operation voltage is also the smallest. According to the above result, 9-cycle must have some good property that can help the resistive switching become more stable. And the comparison of these devices is shown in Table. 3.3.1.

3.3.2 Resistive Switching Localization Test

We wonder what mechanism is of the resistive switching and what cause the difference among the three devices. So we start with the resistive switching localization test. Here we have to make a description of how we operate the device. Generally, the typical set/reset process is applying a voltage bias to the top electrode and the bottom electrode is grounding. In this test, another tungsten probe is used for second top electrode and is grounding as the bottom electrode used in typical set/reset process, as shown in Fig. 3.3.9. Unlike the typical operation, this test is just like applying a

apply the voltage bias on the twice thicker resistive switching layer with a conducting ITO layer in the middle and this middle conducting layer could split the resistive switching layer into two sides, the anode side and the cathode side. Where the anode side is defined by the voltage bias we applied for the forming process. Here we use negative bias for the forming process, so the side we applied bias is the cathode side and the other side, which is grounding, is the anode side. We can apply a small read voltage bias to get to know the resistance of both side and find out which side is the resistive switching happening.

This experiment had been done by KM. Kim et al [28]. The READ 1, READ 2 and READ 3 are shown in Fig. 3.3.10. READ 1 finds out the resistance of the cathode side and READ 2 is going to read the resistance of the anode side. READ 3 helps us to know if the device has been separated into anode side and cathode side by calculate the READ 1 resistance plus READ 2 resistance is equal to the READ 3 resistance. The results of this resistive switching localization test are shown in Fig. 3.3.11, Fig. 3.3.12 and Fig. 3.3.13. Obviously, all the READ 1 resistances are not switching, but the READ 2 resistances are switching its resistance after each set/reset process.

Even though the conducting filament is going through the entire resistive switching layer, the resistive switching could still be localized on the anode side.

3.3.3 Retention Time Property

As a nonvolatile memory, storing data without applying bias is a very important property. The retention time means how long does the data can be stored in the memory. The devices had been put in our retention time experiment and could stored data more than 104 seconds at room temperature, as shown in Fig. 3.3.14, Fig. 3.3.15 and Fig. 3.3.16, and are without apparently degradation. Therefore, these three devices could be developed as nonvolatile memories.

3.3.4 Nondestructive Readout Property

The previous section mentioned that we use 0.1V bias to read the resistance. Therefore, a no destructive read out experiment gives to the devices to find out does the small read voltage really not affect the resistance state. Each device is applying a voltage bias of 0.6V, which is much larger than the 0.1V read voltage, to read the current of various memory states. All the devices can survive more than 2000 seconds at RT, based on this result, as shown in Fig. 3.3.17, Fig. 3.3.18 and Fig. 3.3.19, the maximum read times and the stability of the memory states can be estimated.

-2 -1 0 1 2 3 ITO/HfO2+Al2O3+...(19cycles)/ITO/Glass

Current (A)

Voltage (V)

A

Fig. 3.3.1 (I-V) Different LRS due to different compliance current for 19-cycle

1 2 3 4 5

Fig. 3.3.2 (R-I) Different LRS due to different compliance current for 19-cycle

-2 -1 0 1 2 3 ITO/HfO2+Al2O3+...(19cycles)/ITO/Glass

Current (A)

Voltage (V)

Vstop

Fig. 3.3.3 (I-V) different stop voltage cause different HRS for 19-cycle

1.6 2.0 2.4 2.8 3.2 3.6 4.0

Fig. 3.3.4 (R-V) different stop voltage cause different HRS for 19-cycle

Fig. 3.3.5 Endurance experiment of 9-cycle

Fig. 3.3.6 Endurance experiment of 19-cycle

Fig. 3.3.7 Endurance experiment of 38-cycle

Fig. 3.3.8 Set/Reset voltage distribution of 9cycles/19-cycle/38-cycle

Fig. 3.3.9 The set/reset process of resistive switching localization test

Fig. 3.3.10 READ 1, READ 2 and READ 3 in resistive switching localization test

Fig. 3.3.11 READ 1/READ 2 resistances of 9-cycle by resistive switching localization test

Fig. 3.3.12 READ 1/READ 2 resistances of 19-cycle by resistive switching localization test

Fig. 3.3.13 READ 1/READ 2 resistances of 38-cycle by resistive switching localization test

Fig. 3.3.14 Retention time experiment of 9-cycle

Fig. 3.3.15 Retention time experiment of 19-cycle

Fig. 3.3.16 Retention time experiment of 38-cycle

Fig. 3.3.17 Nondestructive readout experiment of 9-cycle at 0.6V

Fig. 3.3.18 Nondestructive readout experiment of 19-cycle at 0.6V

Fig. 3.3.19 Nondestructive readout experiment of 38-cycle at 0.6V

Table. 3.3.1 Comparisons of 9-cycle, 19-cycle and 38-cycle device

3.4 Conduction Mechanism of Resistive Switching Layer

Fig. 3.4.1, Fig. 3.4.2 and Fig. 3.4.3 show the double-logarithmic plots of current-voltage curve for negative bias in set process for 9-cycle, 19-cycle and 38-cycle with the fitting results. Obviously they show I∝V at first in low voltage region and then show I∝V2 characteristics with the increase of voltage bias. This is the typical behavior of trap-controlled space-charge-limited current (SCLC) behavior. Because it happens before set process gets finished, this SCLC behavior stands for HRS carrier conduction mechanism[11][14].

And Fig. 3.4.4, Fig. 3.4.5, Fig. 3.4.6 show the double-logarithmic plots of current-voltage curve for positive bias in reset process for 9-cycle, 19-cycle and 38-cycle with the fitting results and I∝V relation can be observed. So the Ohmic conduction mechanism could stands for the carrier conduction mechanism of LRS in the resistive memory. Also the HRS fit well as SCLC here, this echo the statement in the above paragraph.

Fig. 3.4.1 SCLC conduction fitting of 9-cycle in HRS by set process

Fig. 3.4.2 SCLC conduction fitting of 19-cycle in HRS by set process

Fig. 3.4.3 SCLC conduction fitting of 38-cycle in HRS by set process

Fig. 3.4.4 Ohmic conduction fitting of 9-cycle in LRS by reset process and SCLC fitting in HRS

Fig. 3.4.5 Ohmic conduction fitting of 19-cycle in LRS by reset process and SCLC fitting in HRS

Fig. 3.4.6 Ohmic conduction fitting of 38-cycle in LRS by reset process and SCLC fitting in HRS

3.5 Resistive Switching Mechanisms for Multi-layer Resistive Switching Film

The evolution of the resistive switching hysteresis is demonstrated in previous section. Even though the basic physical resistive switching mechanisms are not clear in details, we still work on that. Overall, the memory devices fabricated by multi-layer stacking as resistive switching films could be treated as the transition-metal oxides such as HfO2 mixed with Al2O3 in different level. In other words, the Al2O3 has different distribution in resistive switching layers. And 38-cycle device has the widest distribution of Al2O3. According to the research by X. F. Wang et al [29], Al atoms in HfO2 thin films would introduce a significant amount of oxygen vacancies. And it was reported that the single Al2O3 films usually show random resistive switching behaviors during repeated current–voltage measurements [30].

S. Yu et al [31] confirmed that the Al diffusion into HfO2 may be mainly responsible for the improved switching uniformity. Therefore, the oxygen vacancies in transition-metal oxides plays an important role, and it is widely believed that resistive switching is triggered by an electrical migration of anions, such as oxygen anions, which are typically described by the motion of corresponding vacancies [9]. The resistive switching from HRS to LRS is associated with the enrichment of oxygen vacancies arranged in chains under an electric field. Thus, stabilizing the generation of oxygen vacancies is the key issue for stabilizing the switching behaviors. The ab initio calculations

reveal that the oxygen vacancy formation energy in a HfO2 supercell is effectively reduced from 6.53 to 4.09 eV in the vicinity of Al atoms [32], while it is almost unaffected at locations several atoms away from the impurities. Therefore, the chains of oxygen vacancy are easier to generate and be localized along the Al atoms, and more stable conducting filaments are expected to be formed during different switching processes, which could account for the experimentally observed the difference among these devices of switching parameters as shown in previous section.

For our devices, we have known that 38-cycle meets Al2O3 every 6 monolayer and 9-cycle meets Al2O3 every 24 monolayer as we mentioned in chapter 2. Because the chains of oxygen vacancy are easier to generate and be localized along the Al atoms, the 38-cycle device could have so many oxygen vacancies that when electron current flows in the oxide after the forming process. These oxygen vacancies generate unstable conducting filaments that electron current will spread out very easy, because there are oxygen vacancies almost everywhere. For the reason that Al2O3 spread out more in resistive switching layers of the 38-cycle device, it is more difficult to form stable conducting filaments. Therefore, the performance of 38-cycle could be attributed to the random formation of the conducting filaments in the resistive switching layer during each switching process. On the contrary, The-9-cycle device does not happen this way. The Al toms in 9-cycle device distribute tighter and it helps us to control the oxygen vacancies generation, so the conducting filaments are more stable after the forming process, as illustrated in Fig. 3.5.1. Accordingly, 9-cycle has better performance than 38-cycle and 19-cycle is between two parties.

Fig. 3.5.1 Abridged general view of proposed conducting filaments distribution in 9-cycle and 38-cycle

Chapter 4 Conclusion

We first demonstrate the resistive switching characteristics of Transparent Resistive Random Access Memory Devices (TRRAM) with HfO2/Al2O3 multi-layer stacking via different ALD deposition cycles as resistive switching films for 9-cycle, 19-cycle and 38-cycle devices respectively. The endurance times, average set operation voltage, standard deviation of set voltage distribution; average reset operation voltage, standard deviation of reset voltage distribution for the three devices are: 1250 times, -1.258V, 0.162V; 1.275V, 0.193V (9-cycle), 950 times, -1.927V, 0.92V;

1.867V, 0.375V (19-cycle) and 300 times, -1.706V, 0.932V; 1.412V, 1.131V (38-cycle). The-9-cycle device has the best performance than the other two (19-cycle and 38-cycle) from the electrical characteristics shown above. For the reason that Al2O3 spread out more in resistive switching layers of the 38-cycle device, it is more difficult to form stable conducting filaments.

In addition, our 9-cycle device with endurance times up to 1250 times, -1.258V, 1.275V set/reset operation voltage and 0.162V, 0.193V standard deviation of reset voltage distribution is relatively excellent compare to recent published transparency memory devices [10]. It was claimed that oxygen vacancy forming energy could be lower down by Al atoms which leads conducting filaments formed more stable in previous non-transparent RRAM [11]. Even though, the performances of TRRAM in this thesis are still not comparable to non-transparent RRAM. To the best of our knowledge, we first

apply this mechanism to control the conducting filament formation in TRRAM. Not only echo the previous study of the mechanism but also demonstrate TRRAM with the ever highest performance. The multi-layer stacking as the resistive switching films in TRRAM shows great potential in the future modern electronics.

Chapter 5 Future Work

Even though the performance of TRRAM in this thesis is excellent, it still does not match the criterion of NVM application today. Due to the proposed resistive switching mechanism in this thesis, changing the deposition cycles could improve the performance and changing the Al2O3/HfO2 ratio may give a big help too. Furthermore, this thesis use tungsten probe as the top electrode might not be the best choice, deposition a metal between the resistive switching layer and ITO should make the performance better. Hopefully the future research could focus on these views to further make the modern TRRAM in commercial applications come true.

Reference:

[1] W. W. Zhuang, W. Pen, B. D. Ulrich, J. J. Lee, L. Stecker, A. Burmaster, D. R. Evans, S.T. Hsu, M. Tajiri, A. Shimaoka, K. Inoue, T. Naka, N.

Awaya, K. Sakiyama, Y.Wang, S. Q. Liu, N.J. Wu, and A. Ignatiev,”

Novel colossal magnetoresistive thin film nonvolatile resistance random

Novel colossal magnetoresistive thin film nonvolatile resistance random

相關文件