中 華 大 學

中 華 大 學 碩 士 論 文

混合四氟甲烷/氧氣電漿處理對氧化鉿電阻式 記憶體元件之電阻轉換影響研究

Influence of mixed CF

₄

₂

_x

random access memory device

系 所 別：電機工程學系碩士班 學號姓名：M09801044 張悳舜 指導教授：張國明 博士

賴瓊惠 博士

中 華 民 國 1 0 0 年 8 月

i

混合四氟甲烷/氧氣電漿處理對氧化鉿電阻式記 憶體元件之電阻轉換影響研究

摘要

置。現今快閃式記憶體雖然是目前的主流非揮發性記憶體，但仍然存在一些需要 克服的問題，如其高操作電壓、記憶能力隨著元件微縮而下降、低存取速度等。

因此新興的非揮發性記憶體也逐漸受到關注。其中電阻式記憶體具有簡單結構、

低操作電壓、功率消耗低、操作速度快、與半導體製程相容和高微縮能力的優點，

故很有機會成為下一個世代主流的非揮發性記憶體。

在本論文中我們使用電子束蒸鍍沉積電極與薄膜，並使用四氟甲烷/氧氣高 密度電漿對薄膜表面進行處理，分別比較未處理與處理後對電阻式記憶體上的影 響，其中又以四氟甲烷/氧氣比例為 10:1 的元件出現最佳的特性，比起未處理元 件，如耐久度約增加 40 次、開啟操作電壓分佈約少 2.5 伏特、關閉操作電壓約 少 2.5 伏特與高低阻態比約增加 1 倍且阻值穩定性較佳，這對本身以電子束蒸鍍 所沉積品質較差的薄膜電阻式記憶體元件有著特性改善的效果，此處理在未來或 許具有潛力的運用。

ii

Influence of mixed CF

/O

plasma treatment on resistance switching in the Ni/HfO

/Ni resistive

random access memory device Abstract

Due to the popularity of digital equipment, nonvolatile memory (NVM) plays an important role in our life, flash memory is the mainstream among the NVM nowadays, but several challenge emerge, such as high operation voltage, scaling problem and low operation speed. Therefore, the novel NVM attract much attention in recent years.

One of next generation NVM is the resistive random access memory (RRAM), RRAM has many advantages including simple structure, low operation voltage, high operation speed, compatible with CMOS processes and high scalability, etc. so RRAM is possible to become the next generation NVM.

In this work, the electrode and thin film were deposited by electron beam evaporator, HfOx surface was performed with CF4/O2 plasma treatment by high density plasma chemical vapor deposition (HDPCVD). Then compare difference between without treatment and after plasma treatment to observe influence of RRAM.

Among these device, the sample of CF₄:O₂=10:1 shown optimum characteristics.

Compared to sample without treatment, the sample with treatment of CF4:O2=10:1 exhibits endurance of over about 40 cycles, distribution of set voltage reduce about 2.5 V, distribution of reset voltage reduce about 2.5 V and better stability of resistance.

The results indicate plasma treatment have an improve effect in the poor quality thin film by electron beam evaporator, in the future the plasma treatment maybe has the potential on the RRAM device.

iii

誌 謝

在這碩士生涯中，我不論在學業上或是人際關係相處上都學習到非常多，對 於對事情的看法也改變了非常多，在這兩年中，首先要感謝賴瓊惠教授在這期間 提供了很好的學習環境和到交通大學的難得學習機會，對於研究方面也會適時提 出建議，使得我擁有更多的想法，克服困難的能力。接著要特別感謝 交通大學 的張國明教授，提供了一個氣氛融洽的研究環境，也非常照顧學生，讓所有人感 到一股暖意，在專業知識和待人處事上也讓我學習到非常多。

其次非常感謝曾文賢學長，在研究過程中總是傾囊相授，引導我對遇到困難 時所需要考慮的事情，提供我解決事情的一些想法，也將自己在專業領域的知識 做為討論的基礎，在實驗過程中也提供不少建議，讓我在半導體這領域真的學習 到非常多。

另外非常感謝交通大學實驗室裡的菘宏學長、庭嘉學長、博鈞等同學，讓我 在學習過程中有歡樂與鼓勵，讓實驗更為順利。

還要感謝我們實驗室裡的好同學，俊哲、瑞陽、聖文、竹均、宜羚，在我的 實驗上提供了非常多的幫助，在這兩年中帶來許多歡笑與支持，一路走來沒有你 們我的實驗就不可能完成。

最後要謝謝我的家人，一直以來支持與鼓勵是我繼續堅持下去的原動力。

iv

Contents

Chinese Abstract...i

English Abstract...ii

Acknowledgment...iii

Contents...iv

Table Captions...vii

Figure Captions...viii

Chapter 1 Introduction

1.2 Introduction to next generation nonvolatile...4

1.2.1 Magnetroresistive random access memory...4

1.2.2 Phase change random access memory...5

1.2.3 Ferroelectric random access memory...7

1.2.4 Resistive random access memory...8

1.3 Introduction to resistive random access memory...9

1.3.1 Basic resistive switching current-voltage curves...9

1.3.2 Electrodes...11

1.3.3 Resistive switching films...11

1.4 Resistive switching mechanisms...13

1.4.1 Schottky emission...14

1.4.2 Frenkel-Poole mission...14

1.4.3 Tunnel or field emission...15

1.4.4 Space charge limited current...15

1.4.5 Ohmic conduction...15

v

1.4.6 Ionic conduction...16

1.5 Resistive switching models...17

1.5.1 Oxygen movement model...17

1.5.2 Filament model...17

1.5.3 Charge trapping and detrapping model...21

1.5.4 Schottky barrier modulation model...22

1.6 Motivation...24

Chapter 2 Experiment Details

2.2 Different plasma proportion of CF_4/O₂...27

2.3 ICP system...27

2.4 Material analysis...27

2.4.1 X-ray photoelectron spectroscopy (XPS)...27

2.4.2 X-ray diffraction (XRD)...28

2.5 Electrical measurement...28

2.5.1 Current-voltage measurement...28

2.5.2 Endurance measurement...29

2.5.3 Data retention time measurement...29

Chapter 3 Results and discussion

3.1.1 ON/OFF state...30

3.1.2 Compliance current...30

3.2 The sample without treatment...30

3.3 The sample with treatment of CF₄:O₂=10:1...38

3.4 The sample with treatment of CF4:O2=2:8...45

3.5 XRD analysis...52

vi

3.6 AES analysis...53 3.7 XPS analysis...55 3.8 Driving mechanisms...60

Chapter 4 Conclusions

4.1 Conclusions...64 4.2 Future Work...66

References

vii

Table Captions

Table 1-1 NVMs for next generation...3

Table 1-2 Basic conduction mechanisms in insulator...13

Table 2-1 Parameters of plasma treatment...26

Table 4-1 Operation results of three samples...65

viii

Figure Captions

Fig. 1-1 Floating gate and dielectric compose as a sandwich structure...2

Fig. 1-2 Magnetic tunneling junction with two magnetic layers...4

Fig. 1-3 Standard memory cell of PCRAM...6

Fig. 1-4 Time temperature relationship of the PCRAM...6

Fig. 1-5 Metal ferroelectric semiconductor field effect transistor (MFSFET) structure...7

Fig. 1-6 Sandwich structure of metal-insulator-metal (MIM)...8

Fig. 1-7 RS operation polarization of unipolar...10

Fig. 1-8 RS operation polarization of bipolar...10

Fig. 1-9 Nature of oxide breakdown and amount that the defects existing in an oxide bulk...19

Fig. 1-10 Vacancies of oxygen or metals constitute the leakage current path...19

Fig. 1-11 Conducting atomic force microscopy (CAFM) mapping results and the insensitivity of LRS and HRS...20

Fig. 1-12 Concept of tree shaped filament structure...20

Fig. 1-13 Charge transfer with trapping and detrapping for positive voltage..21

Fig. 1-14 Charge transfer with trapping and detrapping for negative voltage.22 Fig. 1-15 Schematic diagram of Schottky junction based on an intermediate region...23

Fig. 1-16 Plot presented this Schottky barrier modulation mechanism...23

Fig. 2-1 Fabrication of Ni/HfOx/Ni MIM structure...25

Fig. 3-1 Typical I-V characteristic of Ni/HfO_x/Ni of the sample without treatment...33

Fig. 3-2 About 20 repeats switching cycles of the sample without treatment....33

ix

Fig. 3-3 Endurance measurement of the sample without treatment...34 Fig. 3-4 Voltage distribution of the sample without treatment...34 Fig. 3-5 Different cycles switching distribution in reset process of the sample

without treatment...35 Fig. 3-6 Different cycles switching distribution in set process of the sample

without treatment...35 Fig. 3-7 Cumulative probability distributions of the switching resistance of the

sample without treatment...36 Fig. 3-8 Double logarithmic plots of I-V curve for positive bias in set process of

the sample without treatment...36 Fig. 3-9 Double logarithmic plots of I-V curve for negative bias in reset process

of the sample without treatment...37 Fig. 3-10 Typical I-V characteristic of Ni/HfOx/Ni of the sample with treatment

of CF₄:O₂=10:1...40 Fig. 3-11 About 60 repeats switching cycles of the sample with treatment of

CF₄:O₂=10:1...40 Fig. 3-12 Endurance measurement of the sample with treatment of

CF₄:O₂=10:1...41 Fig. 3-13 Voltage distribution of the sample with treatment of CF4:O2=10:1...41 Fig. 3-14 Different cycles switching distribution in reset process of the sample

with treatment of CF4:O2=10:1...42 Fig. 3-15 Different cycles switching distribution in set process of the sample

with treatment of CF4:O2=10:1...42 Fig. 3-16 Cumulative probability distributions of the switching resistance of the sample with treatment of CF4:O2=10:1...43 Fig. 3-17 Double logarithmic plots of I-V curve for positive bias in set process

x

of the sample with treatment of CF4:O2=10:1...43

Fig. 3-18 Double logarithmic plots of I-V curve for negative bias in reset process of the sample with treatment of CF4:O2=10:1...44

Fig. 3-19 Typical current-voltage (I-V) characteristic of Ni/HfO_x/Ni of the sample with treatment of CF4:O2=2:8...47

Fig. 3-20 About 10 repeats switching cycles of the sample with treatment of CF4:O2=2:8...47

Fig. 3-21 Endurance measurement of the sample with treatment of CF4:O2=2:8...48

Fig. 3-22 Voltage distribution of the sample with treatment of CF₄:O₂=2:8...48

Fig. 3-23 Different cycles switching distribution in reset process of the sample with treatment of CF₄:O₂=2:8...49

Fig. 3-24 Different cycles switching distribution in set process of the sample with treatment of CF₄:O₂=2:8...49

Fig. 3-25 Cumulative probability distributions of the switching resistance of the sample with treatment of CF₄:O₂=2:8...50

Fig. 3-26 Double logarithmic plots of I-V curve for positive bias in set process of the sample with treatment of CF₄:O₂=2:8...50

Fig. 3-27 Double logarithmic plots of I-V curve for positive bias in reset process of the sample with treatment of CF₄:O₂=2:8...51

Fig. 3-28 Crystal structure of the HfOx film by XRD analysis...52

Fig. 3-29 AES analysis of the sample without treatment...53

Fig. 3-30 AES analysis of the sample with treatment of CF4:O2=10:1...54

Fig. 3-31 AES analysis of the sample with treatment of CF₄:O₂=2:8...54

Fig. 3-32 Hf 4f ESCA spectra of CF4 plasma treated 5 min sample...55

Fig. 3-33 Hf 4f ESCA spectra of CF₄ plasma treated sample...56

xi

Fig. 3-34 XPS spectra of Hf 4f of HfOx/Ni substrate...57 Fig. 3-35 XPS spectra of Hf 4f of HfO_x/Ni substrate by the sample with

treatment of CF4:O2=10:1...57 Fig. 3-36 XPS spectra of Hf 4f of HfO_x/Ni substrate by the sample with

treatment of CF4:O2=2:8...58 Fig. 3-37 XPS spectra of F 1s of HfO_x/Ni substrate by the sample with

treatment of CF4:O2=10:1...58 Fig. 3-38 XPS spectra of F 1s of HfO_x/Ni substrate by the sample with

treatment of CF4:O2=2:8...59 Fig. 3-39 Schematic diagrams of driving mechanism the sample without

treatment...61 Fig. 3-40 Schematic diagrams of limited conductive filament driving

mechanism of the sample with treatment of CF4:O2=10:1...62 Fig. 3-41 Schematic diagrams of conductive filament driving mechanism of the

sample with treatment of CF4:O2=2:8...62 Fig. 3-42 Schematic diagrams of metal conductive filament driving mechanism

of the sample with treatment of CF4:O2=2:8...63

1

Chapter 1 Introduction

1.1 Introduction to nonvolatile memory

The nonvolatile memory (NVM) need to store information without power supply for a long time. Since 1967, D. Kahng and S. M. Sze invented the floating gate nonvolatile semiconductor memory at Bell Labs. In recent years, the portable electronic products can provide the convenient life for human. Due to the popularity of portable devices, such as mobile phone, digital camera, mp3 player, personal digital assistant and USB memory stick, the requirements of the NVM increase significantly in semiconductor industry. An ideal NVM should have the properties of low power consumption, low operation voltage, low operation current, high operation speed, high endurance, long retention time, nondestructive readout, simple structure, small size, low cost, and high cell density. However, there is no such NVM meet the above properties up to now.

Nowadays, the most popular of NVM is flash memory. There are two kinds of flash memories, included the NOR flash and NAND flash. The NOR flash has high operation speed, which is suitable for the operation system storage, DVD player, and mobile phone applications. Another one is the NAND flash which with higher density is used for large data storage application, such as memory cards, MP3 players, digital cameras, and USB flash personal disc etc. A common structure of flash memory is like a MOSFET with a stacking gate which include control gate, dielectric, floating gate and dielectric to compose as a sandwich structure, shown in Fig. 1-1. The logic high or low is determined by charges injected in the floating gate are maintained there, allowing the difference between threshold voltages of the cell transistor for NVM application.

However, the flash memory has some issues still needed to be solved, such as

2

high operation voltage, low operation speed, poor retention time, and coupling interference effect during scaling down [1-2]. Especially because of the coupling interference effect, it will set a scaling down limitation. Although some different flash memories, such as charge trapping layer (SONOS) flash are proposed to replace the tradition flash memory [3-4] but still have some problems. So many researchers want to find some next generation NVM to replace DRAM, SRAM, and flash memory.

There are some possible candidates for next generation NVM, such as magnetroresistive random access memory (MRAM) [8-10], phase change random access memory (PCRAM) [11-15], ferroelectric random access memory (FeRAM) [5-7] and resistive random access memory (RRAM). A brief comparison of flash memory with these next generation NVMs are shown in Table 1-1. These next generation NVMs are discussed in the following section.

Fig. 1-1 Floating gate and dielectric compose as a sandwich structure.

3 Table 1-1 NVMs for next generation [47].

4

1.2 Introduction to next generation nonvolatile memories

1.2.1 Magnetroresistive random access memory

The Magnetroresistive Random Access Memory (MRAM) [8-10] cell is the magnetic tunneling junction which consists of two magnetic layers sandwiching a thin tunneling layer as shown in Fig. 1-2 The magnetization of one magnetic layer is fixed and kept in a specific direction. The other magnetic layer can be switched to parallel or anti-parallel to the reference layer by applying a specific magnetic field. This two magnetic states cause two different resistance states, so the logic high or low is generated by the resistance of parallel or anti-parallel state.

Although the MARM has many merits, the tunneling layer causes the scaling limitation of the device and becoming the most important challenge of MRAM.

Fig. 1-2 Magnetic tunneling junction with two magnetic layers.

5

1.2.2 Phase change random access memory

The phase change memory (PCRAM), also named Ovonic unified memory (OUM), is reported by Neale et al. in 1970. It is a hopeful technology to fit the requirements of the ideal NVM. The standard memory cell of PCRAM is shown in Fig. 1-3 [12], The primary material of the PCM is GeSbTe (GST). The PCRAM exhibits two different structural phases of the GST, amorphous and polycrystalline, for data storage. In reset process, a high magnitude current pulse with short tailing edge is applied on the programmable volume of the phase change material. The temperature of the material exceeds the melting point which eliminates the polycrystalline order in the volume. When the reset pulse is terminated, the device quenches to “freeze in” the disordered structural state (amorphous state). This quench time (cooling time) about several nanoseconds is determined by the thermal environment of the memory cell device and the fall time of the reset pulse. In set process, a moderate magnitude current pulse with sufficient duration is applied to maintain the device temperature for crystal growth.

The amorphous structural state (high resistance state) or the polycrystalline structural state (low resistance state) is read by applying a low magnitude and long duration current pulse. The time temperature relationship of the PCRAM is shown in Fig. 1-4 [12]. During the set and reset processes, large Joule heating is applied on the phase change material, and hence, very large power consumption is generation. How to reduce the power consumption during the PCRAM operation is becoming the primary challenge.

6

Fig. 1-3 Standard memory cell of PCRAM [12].

Fig. 1-4 Time temperature relationship of the PCRAM [12].

7

1.2.3 Ferroelectric random access memory

Ferroelectric random access memory (FeRAM) [16-17] is a device with ferroelectric material. The ferroelectric material is a material with a spontaneous polarization and the polarization can be altered by applying electric field. A standard structure of the ferroelectric material is the ABOstructure (perovskite structures), where the A, B, and O atoms are located at corner, body center, and face center of the cubic, respectively. The B atom has two thermodynamically stable positions which depend on the polarity of the applied electric field. The polarity hysteresis curve of the ferromagnetic material can be used for NVM application and this is so called ferroelectric random access memory.

One of the typical structures of FeRAM is the metal ferroelectric semiconductor field effect transistor (MFSFET) structure as shown in Fig. 1-5. The structure is very similar to the MOSFET except the oxide film is replaced by the ferroelectric film. The polarizations (+Pr or –Pr) of the ferroelectric material film can affect the drain current of the transistor, and the memory effect of the typical MFSFET structure FeRAM is nonvolatile and with nondestructive readout property.

Fig. 1-5 Metal ferroelectric semiconductor field effect transistor (MFSFET) structure.

8

1.2.4 Resistive random access memory

The resistive random access memory (RRAM) [18] with metal-insulator-metal (MIM) sandwich structure is mostly designed, as shown in Fig. 1-6. It exhibits different resistance states by applying voltage bias, called resistive switching (RS).

The RRAM which have two memory states, the low resistance state (LRS) and the high resistance state (HRS). But the RS may generate more than two different resistance states by different compliance current or different reset voltage. However, RRAM has advantage of high operation speed, low operation voltage, retention time, endurance, very simple MIM sandwich structure, excellent scalability and high density integration. It is very convenient to proposing into mass production in the market. Nowadays, the main issue of RRAM physical mechanism of RS is not well revealed. If the RS mechanism is cleared, we can expect the well design RRAM to becoming the most ideally NVM.

Fig. 1-6 Sandwich structure of metal-insulator-metal (MIM).

9

1.3 Introduction to resistive random access memory

1.3.1 Basic resistive switching current-voltage curves

In order to understand the influences of each stack in the RS device structure, a brief introduction to the primary electrical properties is necessary. The RRAM device can operate between two different resistance states, the lower resistance state is called LRS or on state and the higher one is called HRS or off state. In this paper, the process that makes the RRAM device change its resistance from LRS to HRS is called set process. And the reset process turns the RRAM device from HRS into LRS. There are three sorts of RS operation polarization, unipolar, bipolar, nonpolar [18-20], as shown in Figs. 1-7 and 1-8. The unipolar RS means the set and reset process are happened in the same electrical polarity. And the bipolar RS means the set and reset process are happened in different electrical polarity. As for bipolar RS, it depicts that the RS phenomenon is observed as applying the specific polarity of the voltage to do the set process, and the reset process has to be done by applying the opposite polarity of the voltage. The nonpolar RS means that the device can be switched to another resistance state by applying a voltage no matter what the polarity.

The RRAM with the binary metal oxide films need forming process to form the filament of the device. This phenomenon is similar to the dielectric breakdown, named it “soft breakdown”. The filament model is discussed in the following sections.

But some RRAM devices start in the LRS. Because it does not need the forming process, it is called forming free RRAM.

10

Fig. 1-7 RS operation polarization of unipolar [18].

Fig. 1-8 RS operation polarization of bipolar [18].

11

1.3.2 Electrodes

The RRAM stores the data by resistive switching behaviors, resulting from the resistance change between metal and insulator interface, and the different work function, electronegativity, oxygen affinity property of metal, even the interface reaction and the inter-diffusion. For example, inter-diffusion of metal may cause transport mechanism changed. So it is very important to pick up the suitable combination of the electrodes for better operation.

1.3.3 Resistive switching films

The resistive switching phenomena have been found in many materials. The research mainstream is focused on several groups, including binary transition metal oxides, complex metal oxide, perovskite oxides, manganites, and other alloy or polymers. The transition metal oxide based resistive random access memory (TMO-RRAM) devices have been extensively studied due to its excellent characteristics such as low power, high speed, good scalability, and high density integration [21-24]. The materials of binary transition metal oxides such as NiO, TiO₂, Al2O3, Nb2O5, HfO2 and etc. [25-27], the materials of complex metal oxides such as:

Gd_1-xCa_xBaCo₂O₅, Ba_0.7Sr_0.3TiO₃ and etc, the carrier-doped manganites with the perovskite structure, such as Pr1-xCaxMnO3 (PCMO), La1-xCaxMnO3 (LCMO) and etc.

Various processing techniques can be used for the preparation of composite and multilayer resistive switching thin films by modifying the apparatus and the conditions for deposition, such as atomic layer deposition (ALD), electron beam evaporator (E-GUN), electron cyclotron resonance (ECR) sputtering, plasma-enhanced atomic layer deposition (PEALD), plasma-enhanced chemical vapor deposition (PECVD), pulsed laser deposition (PLD), metallorganic chemical vapor deposition (MOCVD), reactive DC sputtering, RF sputtering, sol-gel method, and

12

solid-state reaction method. For example, the MOCVD usually use to deposit the high quality film, but process parameters such as deposition temperature, gas partial pressure are introduced to elucidate the influence on the resistive switching properties.

In addition, the doping species and the crystallinity of the resistive switching films also determine the resistive switching behaviors, which have been discussed as well.

13

1.4 Resistive switching mechanisms

Nowadays the RS mechanisms are not totally clear. The discussion of carrier conduction mechanisms may be another way to discover RS mechanisms. The carrier conduction mechanisms of RRAM device operation are dependent on the species and the combinations of electrodes. In this section, six kind of basic conduction mechanisms in insulator are summarized in Table 1-2 [28].

Table 1-2 Basic conduction mechanisms in insulator [28].

14

1.4.1 Schottky emission

The Schottky emission is thermionic emission of carriers across the interface between metal and insulator. Therefore, the Schottky emission is also named it thermionic emission. If carriers transport in an insulator is by this mechanism, a plot of Ln|J/T²| versus 1/T in a specific voltage is a straight line, and the slope can determine the permittivity of the insulator. The formula of Schottky emission is expressed as below:

(1-1) where A^* denotes Richardson constant, Φ_b is the Schottky barrier height, ε₀ is the permittivity of free space, εr is the dynamic dielectric constant, V is the external applied voltage and d is the insulator thickness.

1.4.2 Frenkel-Poole emission

The Frenkel-Poole (F-P) emission is carrier hopping between defect states in the dielectric material, causing trapped electrons into the conduction band due to field enhancement thermal excitation and the barrier lowering as twice as that in Schoottky emission mechanism. If the carriers transport in this mechanism, the plot of ln|J|

versus 1/T in a specific voltage and ln|J/V| versus V^1/2 at a specific temperature are straight lines. The formula of FP emission is expressed as below:

(1-2) Where B denotes a material-related constant, Φt is the trap level, ε0 is the permittivity

15

of free space, εr is the dynamic dielectric constant, V is the external applied voltage and d is the insulator thickness.

1.4.3 Tunnel or Field emission

The tunnel or field emission is carriers through the barrier, corresponding to electrons tunneling from the metal Fermi level into the conduction band of insulator or field ionization of trapped electrons into the conduction band of insulator. This emission is strongly dependent on the applied voltage, but is essentially independent on the temperature. If the carriers transport in this mechanism, the plot of Ln|J/V²| versus 1/V is a straight line.

(1-3)

1.4.4 Space charge limited current

The theory of space charge limited current (SCLC) was first given by Mott and Gurney (1940) and has been extended by many authors. This mechanism is attributed to defects in the dielectric. Trap states is the key factor that carrier can conduct in different degrees. Carriers injected into the insulator and no compensation charge is present. The density of SCLC is direct proportion to V², and is essentially independent on the temperature.

(1-4)

16

1.4.5 Ohmic conduction

The Ohmic conduction is caused by thermionic carriers. The thermally excited electrons are hopping from one isolated state to the next state. Temperature is another effect in this mechanism. The plots of ln|J| versus ln|V| at fixed temperature and ln|J|

versus 1/T in a specific voltage are straight lines The formula of Ohmic conduction is expressed as below:

(1-5) Where a and c denote constants, V is the external voltage, and T is the absolute temperature.

1.4.6 Ionic conduction

The ionic conduction is similar to a diffusion process. In this mechanism, the relationship between voltage and current is Ohmic conduction at a fixed temperature, and a plot of ln|JxT| versus 1/T in a specific voltage is a straight line with unity slope.

In addition, each conduction mechanism may dominate in some voltage and temperature ranges, and there are possibly two or more mechanisms causing the conduction in an insulator.

(1-6)

17

1.5 Resistive switching models

1.5.1 Oxygen movement model

In many transition metal oxides, oxygen ion defects are much more mobile than the cations. When low voltage is applied on the electrode, the slope of I-V curve is straight line according to the Ohmic principle. If larger voltage is applied, the line will not obey the Ohmic conduction due to large change of oxygen concentration dramatically. When larger bias is applied, oxygen ions drift toward the anode, causing the region near the anode become p-conductivity, and the region near the cathode become n-conductivity. The set of mass action equations leads to a strong depletion of the oxygen vacancy concentration in a somewhat larger anodic region. In this situation, the large region of n-conductivity is called “virtual cathode”. This “virtual cathode” moves towards the anode and will finally form a conductive path. When the virtual cathode approaches the anode, the resistance of the dielectric decreases dramatically, and the forming process is completed. Once the electroforming is completed, the bipolar resistance switching takes place between the virtual cathode and the anode when reverse external bias is applied [29-31].

1.5.2 Filament model

The filament model has been proposed since 1970’s. For resistive switching, the major possible models having been proposed by many researchers, including filament model, charge tapping and detrapping model, phase change, and solid state electrolyte.

Each model may be applied for certain combinations of electrodes and oxide materials.

However, every model was proposed up to now, but nowadays the mainstream of each model is filament model [31-35].

The filament model is attributed to the oxide soft breakdown and the amount of the defects existing in an oxide bulk, such as the interstitials, vacancies of oxygen,

18

and the conducting path constitute of metals, as shown in Figs. 1-9 and 1-10 [36].

Recently, due to nonvolatile memories have been extensively studied, more detailed studies have been carried out to investigate the possible resistive switching mechanisms. The most obvious evidences for the filamentary model are conductive atomic force microscopy (CAFM) mapping results and the insensitivity of LRS and HRS, as shown Fig. 1-11 [37]. CAFM is one of the best tools to investigate the local conductivity throughout the film.

Debate this model for the resistive switching process. It is reasonable to assume that the oxygen deficient region (filaments) of a film is much more conducting than the stoichiometric region. Generally the mobility of oxygen ions in binary oxides is much higher than metal ions, and the formation of the filaments could be induced by the rejection of O2ions. As for the filament rupture, it is the current induced effect which Joule heating effect resulting from the current flow through the tiny filaments generates large heat and raises temperature at the local section, providing enough energy for oxygen ions to reflow and return to insulator to rupture the filaments.

Recently, the concept of tree shaped filament structure has been proposed as shown in Fig. 1-12 [36]. The filaments are formed in a dielectric when the carriers are injected from a local point at the electrode interface by forming process. When one of the tree shaped gets to the anode, the resistance of the dielectric decreases significant, and the resistance switching to LRS is formed. Basically, the common base region of the filament is strong, but the branch of the filament is weak. The weakly formed branches of the filament can easily be thermally ruptured after subsequent I-V measurements, resulting in an HRS of the resistance switching.

19

Fig. 1-9 Nature of oxide breakdown and amount that the defects existing in an oxide bulk [36].

Fig. 1-10 Vacancies of oxygen or metals constitute the leakage current path [36].

20

Fig. 1-11 Conducting atomic force microscopy (CAFM) mapping results and the insensitivity of LRS and HRS [37].

Fig. 1-12 Concept of tree shaped filament structure [36].

21

1.5.3 Charge trapping and detrapping model

The trapping and detrapping model is demonstrated by the charge transfer, trapping and detrapping in the insulator as shown Figs. 1-13 and 1-14 [38]. It is widely explained for RS binary oxides and perovskite oxides. Defects play an important role in this mechanism. When defect states are empty, they can capture the current electrons, leading to low electron concentration and HRS. After all the defect states are filled, the current electrons are free to drift through the oxide, leading to high electron concentration and LRS. Hence, the resistance change is dominated by the bulk defect states.

Fig. 1-13 Charge transfer with trapping and detrapping for positive voltage [38].

22

Fig. 1-14 Charge transfer with trapping and detrapping for negative voltage [38].

1.5.4 Schottky barrier modulation model

Schottky barrier between oxide and metal electrodes is different from various materials. For p-type semiconductor, metal with lower work function has higher Schottky barriers for holes. For n-type semiconductor, metal with high work function has higher Schottky barriers for electrons. For Schottky barrier mechanism, the interface between the oxide and metal plays an important role for resistive switching.

When the external bias applied on the metal, electrons were injected into the oxide, and a large amount of electrons are accumulated into the interface states. The net charges in the interface states resulted in a modulation of the Schottky barrier high, because the band bending was affected by the net charges, leading to HRS or LRS as shown in Fig. 1-15 and 1-16 [39-40].

23

Fig. 1-15 Schematic diagram of Schottky junction based on an intermediate region [39].

Fig. 1-16 Plot presented this Schottky barrier modulation mechanism [40].

24

1.6 Motivation

Nowadays, many materials that are not compatible with CMOS technology are not applicable to practical applications. Therefore, we adopt HfO2 as the material in our device for RRAM device. HfO₂ is a promising high-κ dielectric to replace SiO₂ gate dielectrics in the CMOS technology, which means not only the compatibility with CMOS technology is good, but also a deeper understanding on its characteristics has been achieved. Also, because of its wide band gap (approximately 6 eV), a large storage window (On/Off ratio) can be achieved. Furthermore, HfO₂ based RRAM is already fabricated by different deposition ways in previous research, such ALD, ECR sputtering, PECVD, MOCVD, reactive DC sputtering, RF sputtering, but very few researcher use E-GUN to deposit the thin film for insulator of RRAM, so we hope to use E-GUN to deposition of insulator and observe phenomenon of resistive switching of different quality of thin film. In additional, the characteristics of electrode/oxide interface are critical issue to the switching properties of RRAM in previous research, and plasma treatment can improve some material properties, such as defect density, leakage current, so we want to improve properties of interface by surface plasma treatment which may influence characteristics of resistive switching.

25

Chapter 2 Experiment Details

2.1 Experimental procedure

The metal-insulator-metal (MIM) structure of the device with Ni/HfO_x/Ni is as shown in Fig. 2-1.

First, the p-Si substrate was rinsed in the deionization water (DI water), then dipped in dilute HF solution (HF:DI water =1:100) to remove the native oxide. After the RCA clean procedure, 500-nm-thick silicon dioxide was grown by furnace at 1000

°C for 1 hr for isolation purpose.

The 30-nm-thick HfO_x thin film was deposited on Ni/SiO₂/Si substrates by electron beam evaporator (E-GUN). Followed that the HfOx film was performed by high density plasma chemical vapor deposition (HDPCVD) system at 150 °C for 90 s under ICP power of 500 W, bias power of 0 W, a pressure of 100 mTorr by plasma treatment with the CF₄/O₂ ratios of 10/1 and 2/8, respectively. For comparison, the control sample was prepared without any plasma treatment. Finally, 50-nm-thick Ni top and bottom electrodes were deposited by E-GUN using a shadow mask at room temperature.

Fig. 2-1 Fabrication of Ni/HfO_x/Ni MIM structure.

26 Table 2-1 Parameters of plasma treatment.

27

2.2 Different plasma proportion of CF

/O

The plasma treatment of mixed CF₄/O₂ gas is performed by HDPCVD with two proportion of plasma treatment. First the deposition source CF4 flow rate was 50 sccm and O₂ flow rate was 5 sccm, ICP power was 500 W, bias was 10 W, process pressure was 100 mTorr, temperature was 150 °C and process time was 90 sec.

Another plasma process was the deposition source CF₄ flow rate was 10 sccm and O2 flow rate was 40 sccm, ICP power was 500 W, bias was 10 W, process pressure was 100 mTorr, temperature was 150 °C and process time was 90 sec, which summarized is table 2-1.

2.3 ICP system

Inductively coupled plasma (ICP) system is used in the high density plasma source of HDPCVD. When an RF current flows in the coils, it generates a changing magnetic field. The inductively coupled electric field accelerates electrons and causes ionization collisions. Since the electric field is in the angular direction, electrons are accelerated in the angular direction, which allows electrons to travel a long distance without collisions with the chamber wall or electrode.

2.4 Material analysis

X-ray diffractometer (XRD) and X-ray photoelectron spectrometer (XPS) analyses are used to understand the related and the corresponding to electrical properties, investigating the dopant effects and the resistive switching mechanism.

2.4.1 X-ray photoelectron spectroscopy (XPS)

When the light illuminates the matter, the inner electrons are going to be excited and ionized. The ionized electrons, which close to the surface of the material, have the ability to escape into vacuum, named photoelectron, and the phenomenon is called

28

photoelectric effect. Due to the different components and chemical states of the matters, photoelectrons have characteristic kinetic energy, which can be used to identify the composed elements and the chemical states of surface atoms. When the excitation source is soft x-ray, the photoemission spectroscopy is termed by X-ray photoelectron spectroscopy.

2.4.2 X-ray Diffraction (XRD)

The crystal structure of the RS layer is investigated by the X-ray diffraction. The scanning step was 0.02°, and the scanning speed was 4°/min. According to the theory of X-ray diffraction, the average grain size of each orientation can be estimated by using Scherrer’s equation.

2.5 Electrical measurement

The most important parts of all are the electrical measurements. The electrical properties of the devices from I-V curve are demonstrated. The electrical measurement system consisted of a probe station, an Agilent 4156A semiconductor parameter analyzer, and GPIB controller. Our electrical measurement was sorted into three items: current-voltage measurement by DC voltage sweeping, endurance measurement, and data retention time measurement. The three initial items are tested for criteria of our memory devices.

2.5.1 Current-voltage measurement

The measurement is performed by Agilent 4156A which provides a DC voltage sweeping between two specific voltages to observe the resistive switching. We could observe the relations between the bias voltage and the current of the high/low resistive states.

29

2.5.2 Endurance measurement

Before used in commercial applications, it is important to investigate that how many times that the memory devices can be operated before failure. That is so called endurance measurement. A memory device is expected with excellent endurance characteristics due to the requirements for users.

2.5.3 Data retention time measurement

Retention time is how long the stored data can be maintained, and the data is not able to be distinguished beyond retention time. The data retention time measurement is performed by reading the stored data after certain periods to find out if there occurs data loss or not. The data retention time measurement performed at high temperature can accelerate the degradation of the device to investigate the thermal stability of the memory devices.

30

Chapter 3 Results and Discussion

3.1 Some resistive switching introduction

3.1.1 ON/OFF state

The unipolar or bipolar resistive switching is performed by sweeping current bias.

The forming process is the common phenomenon in the binary metal oxide films which is similar to the dielectric breakdown, named it “soft breakdown”. In general, two memory states are exhibits. The LRS, also named high conducting state (ON state) and the HRS, also named low conducting state (OFF state), which can be altered by applying bias voltage, and utilizes the two different resistance values to storage digital data. R_HRS (R_OFF) is defined the resistance of HRS. R_LRS (R_ON) is defined the resistance of LRS. ON process is defined as the transformation from LRS into HRS.

OFF process is defined as the transformation from HRS into LRS. V_Set (V_ON) is the voltage which is the highest slope of current-voltage curve. During ON process, a sudden increase in current occurs at V_Set and limited at compliance current. V_Reset (VOFF) is the voltage which the lowest slope of I-V curve. During OFF process, a sudden decrease in current occurs at V_Reset.

3.1.2 Compliance current

An important part of RRAM is the compliance current. Compliance current imposed by the measure instruments or the electrical circuit is imposed on the resistive switching memory device to avoid the device breakdown.

3.2 The sample without treatment

Fig. 3-1 shows the typical I-V characteristics of Ni/HfO_x/Ni for the sample without treatment memory cell measured under the dc voltage sweep. A positive bias was applied to the top electrode while the bottom electrode was grounded during the

31

I-V measurements. The I-V curve exhibits bipolar resistance switching between HRS and LRS which induced by the applied voltage. The devices was required a forming process with current compliance to prevent fatal destruction of the device, Hence compliance current is necessary to activate the nonvolatile switching properties of a RRAM device. The forming voltage of device was approximately 8 V with current compliance of 10 µA. After forming process, the device became a LRS. When a sweep voltage from zero to negative. The resistance decrease from the LRS to HRS was observed at approximately -1.3 V. Then sweep voltage from zero to positive. A suddenly increase in resistance from the HRS to LRSat approximately 1.1 V. Fig. 3-2 shows about 20 repeats switching cycles of the sample without treatment. The device exhibits unstable reset voltage and had different resistance with each time. On the other hand, the higher process current was observed at low positive voltage region in the HRS. These results may relate with thin film which exist some defects such as oxygen vacancies, metallic defects, and dislocations. According to the filament model, the defects extended to form conducting filaments in the HRS, these conducting filaments gather together to form stronger and more conducting filaments leading to the transition to the LRS. If a large number of defects exist in the oxide film, the operation voltage very likely become uncontrollable, owing to the filaments is not formed localized, so the filaments may can’t to completely disruption. These reasons may cause to higher process current and unstable operation voltage in this device.

Although the device could be changed between LRS and HRS for about 20 switching cycles without switching failure, but there has a slight fluctuation of resistance of LRS by different process current as shown in Fig. 3-3.

Fig. 3-4 shows the distribution of operation voltage of the sample without treatment which both V_Set and V_Reset were obtained different operation voltage range around 2 ~ 9 V and 0 ~ -5 V, respectively. The sketch appeared average distribution

32

of operation voltage either set or reset processes which the distribution result indicated unstable operation voltage in this device. This result may different amount of defects extend to form conducting filament which cause unstable filament path with each operation process by extra bias.

Fig 3-5 and Fig. 3-6 show the resistance distribution of different cycles, the different cycle curves were obtained unstable resistance and operation voltage distribution. This result can clear obtain the different amount of defects extend to form conducting filament by extra bias, so unstable characteristics occur with different operation process.

Fig. 3-7 plots the distributions of the switching resistance, including LRS and HRS. For the sample without treatment, a larger variation in LRS and HRS values were observed. It shows similar result of unstable resistance distribution.

Fig. 3-8 shows the double logarithmic plots of I-V curve for positive bias in set process for fitting results of the sample without treatment. Obviously it shows 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 related SCLC behavior. Because it happens before set process completed, so this SCLC behavior indicates for HRS carrier conduction mechanism [41-42].

Fig. 3-9 shows the double logarithmic plots of I-V curve for negative bias in reset process for the fitting results of the sample without treatment and I ∝ V relation can be observed. So the Ohmic conduction mechanism can indicate for the carrier conduction mechanism of LRS in this sample.

33

-5 0 5 10 15

10^-13 10^-11 10^-9 10^-7 10^-5 10^-3 10^-1

Without treatment

C u rr e n t( A )

Voltage(V)

Forming Reset Set

Fig. 3-1 Typical I-V characteristic of Ni/HfOx/Ni of the sample without treatment.

Fig. 3-2 About 20 repeats switching cycles of the sample without treatment.

-10 -5 0 5 10 15

10^-14 10^-12 10^-10 10^-8 10^-6 10^-4 10^-2 10⁰

Without treatment, 21 cycles

C u rr e n t( A )

Voltage(V)

Set Reset

3

2

1

2

1

34

Fig. 3-3 Endurance measurement of the sample without treatment.

-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 0

2 4 6 8 10 12

Without treatment

N u m b e rs

Voltage

V_set V_reset

Fig. 3-4 Voltage distribution of the sample without treatment.

0 5 10 15 20 25

10⁰ 10² 10⁴ 10⁶ 10⁸ 10¹⁰ 10¹² 10¹⁴

Without treatment, read@0.2V

R e s is ta n c e (  )

Cycle Number

LRS HRS

35

Fig. 3-5 Different cycles switching distribution in reset process of the sample without treatment.

Fig. 3-6 Different cycles switching distribution in set process of the sample without treatment.

-7 -6 -5 -4 -3 -2 -1 0

10^-10 10^-9 10^-8 10^-7 10^-6 10^-5 10^-4 10^-3 10^-2 10^-1

Without treatment

C u rr e n t( A )

Voltage(V)

Cycle 1 Cycle 10 Cycle 20

-2 0 2 4 6 8 10 12 14

10^-13 10^-11 10^-9 10^-7 10^-5 10^-3

Without treatment

C u rr e n t( A )

Voltage(V)

Cycle 1 Cycle 10 Cycle 20

36

Fig. 3-7 Cumulative probability distributions of the switching resistance of the sample without treatment.

Fig. 3-8 Double logarithmic plots of I-V curve for positive bias in set process of the sample without treatment.

2.05

1.27

-3 -2 -1 0 1 2 3

-26 -24 -22 -20 -18 -16 -14 -12 -10

L n ( |J |)

Ln (|V|)

Set

10⁰ 10¹ 10² 10³ 10⁴ 10⁵ 10⁶ 10⁷ 10⁸ 10⁹10¹⁰10¹¹10¹²10¹³10¹⁴ 0

20 40 60 80 100

Without treatment

C u m u la ti v e p ro b a b il it y (% )

Resistance(  )

LRS HRS

37

Fig. 3-9 Double logarithmic plots of I-V curve for negative bias in reset process of the sample without treatment.

0.95

-3 -2 -1 0 1

-20 -18 -16 -14 -12 -10 -8 -6 -4

L n ( |J |)

Ln (|V|)

Reset

38

3.3 The sample with Treatment of CF

:O

=10:1

Fig. 3-10 shows the typical I-V characteristics of the sample with treatment of CF4:O2=10:1 in Ni/HfOx/Ni memory cell measured under the dc voltage sweep. The forming voltage of device was approximately 13 V with current compliance of 1 µA.

After forming process, the device became a LRS. When a sweep voltage from zero to negative. The resistance decrease from the LRS to HRS was observed at approximately 0.8 V. Then sweeping from zero to positive, a sudden increase in resistance from the HRS to LRSat approximately 2.3 V. Figs. 3-11 and 3-12 shows about 60 repeat switching cycles and could be changed between a LRS and HRS for about 60 switching cycles without switching failure. Compared to the sample without treatment, the device exhibits more stable resistance with LRS and lower process current with HRS. These results may some defects be restored by fluorine atoms at metal/insulator surface and limited the conducting filaments in some region. Hence the forming voltage slightly increase, stable resistance and process current was obvious to decrease with less oxide defect.

Fig. 3-13 shows the distribution of operation voltage of the sample with treatment of CF4:O2=10:1 in Ni/HfOx/Ni device which both VSet and VReset were obtained different operation voltage range around 2 ~ 7 V and 0 ~ -2.5 V, respectively.

The sketch appeared more uniform distribution of operation voltage than the sample without treatment. Especially the distribution of reset voltage exhibits more concentrate operation voltage in this device. This result may reduced defects of surface by fluorine atoms, providing the more stable conducting filament path which cause more stable voltage distribution with each operation process.

Fig 3-14 and Fig. 3-15 show the switching distribution of different cycles.

Compared to sample without treatment, the different cycle curves were obtained more

39

stable resistance and distribution of operation voltage. These results may the sample characteristics improved by plasma treatment which reduced defects of surface by fluorine atoms, providing the more stable conducting filament path, so more stable resistance and distribution of voltage with different operation process.

Figure 3-16 plots the distributions of the switching resistance, including LRS and HRS. For the sample with treatment of CF4:O2=10:1, a smaller variation in LRS values were observed. Compared to the sample without treatment, it shows similar result of stable distribution of resistance.

Fig. 3-17 shows the double logarithmic plots of I-V curve for positive bias in set process with the fitting results of the sample with treatment of CF4:O2=10:1.

Obviously it shows the same mechanism with the sample without treatment that 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 related SCLC behavior. Because it happens before set process completed, so this SCLC behavior indicates for HRS carrier conduction mechanism.

Fig. 3-18 shows the double logarithmic plots of I-V curve for negative bias in reset process with the fitting results of the sample with treatment of CF₄:O₂=10:1 and I ∝ V relation can be observed. So the Ohmic conduction mechanism can indicate for the carrier conduction mechanism of LRS in this sample.

40

0 5 10 15 20

10^-12 10^-10 10^-8 10^-6 10^-4

10^-2 Treatment with CF₄:O₂ = 10:1

C u rr e n t( A )

Voltage(V)

Forming Reset Set

Fig. 3-10 Typical I-V characteristic of Ni/HfO_x/Ni of the sample with treatment of CF4:O2=10:1.

Fig. 3-11 About 60 repeats switching cycles of the sample with treatment of CF4:O2=10:1.

2

3

1

2

1

-4 -2 0 2 4 6 8 10

10^-14 10^-12 10^-10 10^-8 10^-6 10^-4 10^-2 10⁰

Set Reset Treatment with CF₄:O₂ = 10:1, 55 cycles

C u rr e n t( A )

Voltage(V)