Temperature dependences of ferroelectricity and resistive switching behavior of epitaxial BiFeO3 thin films

This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 218.94.142.65

This content was downloaded on 14/11/2015 at 03:12

Please note that terms and conditions apply.

Temperature dependences of ferroelectricity and resistive switching behavior of epitaxial BiFeO3 thin films

View the table of contents for this issue, or go to the journal homepage for more 2015 Chinese Phys. B 24 107705

(http://iopscience.iop.org/1674-1056/24/10/107705)

Temperature dependences of ferroelectricity and resistive switching behavior of epitaxial BiFeO ₃ thin films ^∗

Lu Zeng-Xing(芦增星)^a), Song Xiao(宋 骁)^a), Zhao Li-Na(赵丽娜)^a), Li Zhong-Wen(李忠文)^a), Lin Yuan-Bin(林远彬)^a), Zeng Min(曾 敏)^a), Zhang Zhang(张 璋)^a), Lu Xu-Bing(陆旭兵)^a), Wu Su-Juan(吴素娟)^a), Gao Xing-Sen(高兴森)^a)†, Yan Zhi-Bo(严志波)^b), and Liu Jun-Ming(刘俊明)^b)‡

a)Institute for Advanced Materials and Laboratory of Quantum Engineering and Quantum Materials, South China Normal University, Guangzhou510006, China

b)Laboratory of Solid State Microstructures and Innovation Center of Advanced Microstructures, Nanjing University, Nanjing210093, China

(Received 10 March 2015; revised manuscript received 14 May 2015; published online 20 August 2015)

We investigate the resistive switching and ferroelectric polarization properties of high-quality epitaxial BiFeO3

thin films in various temperature ranges. The room temperature current–voltage (I–V) curve exhibits a well-established polarization-modulated memristor behavior. At low temperatures (<253 K), theI–V curve shows an open circuit voltage (OCV), which possibly originates from the dielectric relaxation effects, accompanied with a current hump due to the polar- ization reversal displacement current. While at relative higher temperatures (>253 K), theI–Vbehaviors are governed by both space-charge-limited conduction (SCLC) and Ohmic behavior. The polarization reversal is able to trigger the conduc- tion switching from Ohmic to SCLC behavior, leading to the observed ferroelectric resistive switching. At a temperature of

>298 K, there occurs a new resistive switching hysteresis at high bias voltages, which may be related to defect-mediated effects.

Keywords:ferroelectric, memristor, resistive random access memory

PACS:77.84.–s, 77.80.–e, 85.25.Hv, 73.50.–h DOI:10.1088/1674-1056/24/10/107705

1. Introduction

In recent years, ferroelectric resistive switching or mem- ristor effects^[1]have aroused a great interest because it enables the polarization modulation of leakage^[2,3] and/or tunneling currents.^[4,5] This makes it promising to be used in resistive random access memories (RRAMs). Compared with the con- ventional RRAMs that are based on defect-mediated (ionic or electronic) process, ferroelectric RRAMs are based on the intrinsic ferroelectric domain switching instead of defects immigration,^[6] which enables fast, stable resistive switch- ing, without the need for a power consuming electroforming process.^[7,8]

Among the ferroelectric resistive switching materials, the well-known multiferroic BiFeO3 (BFO) has become a pop- ular topic of research because of its relatively low bandgap of ∼2.8 eV.^[2,6,9,10] This enables relatively large conduc- tivity and easy bandgap modulation, in addition to its su- perior multiferroic,^[8,10]photovoltaic,^[11–14]and piezoelectric properties.^[15] Several mechanisms have been proposed to explain the resistive switching behaviors in BFO films, in- cluding formation/rupture of conductive filaments,^[16]carrier- and/or defect-control of the depletion layer thickness,^[17]

charge trapping/detrapping,^[18]ferroelectric tunneling,^[4,5]po-

larization modulation of interface Schottky barrier,^[2,19,20]and space-charge-limited conduction.^[14] However, the origin of resistive switching is still controversial, and the reported re- sistive behaviors in BFO films are also greatly varied, depend- ing on the film fabrication process.^[21,22]Therefore, a system- atic investigation on the resistive switching mechanism in BFO film is still needed.

In this work, we systematically study the temperature- dependent ferroelectricity and conductive mechanisms in BFO thin films. To reduce the artifacts from defects or grain bound- aries, we employ epitaxial BFO films with excellent ferro- electric hysteresis loops. It is easier to correlate the ferro- electricity with the conductive behaviors. We identify vari- ous conductive mechanisms in different temperature and volt- age ranges, such as open circuit voltage (OCV) from dielectric relaxation, polarization-modulated Ohmic, and space-charge- limited conductive (SCLC) behaviors, as well as the defect- mediated charge trapping/detrapping process.

2. Experimental procedure

The BFO film with a thickness of∼450 nm was grown on a∼60-nm-thick SrRuO₃(SRO)-covered SrTiO₃(STO) sub- tract (shown in Fig. 1(d)) by pulsed laser deposition (PLD,

∗Project supported by the National Natural Science Foundation of China (Grant Nos. 51272078 and 51332007), the State Key Program for Basic Research of China (Grant No 2015CB921202), the Guangdong Provincial Universities and Colleges Pearl River Scholar Funded Scheme, China (2014), the Interna- tional Science & Technology Cooperation Platform Program of Guangzhou, China (Grant No. 2014J4500016), and the Program for Changjiang Scholars and Innovative Research Team in University of China (Grant No. IRT1243).

†Corresponding author. E-mail:[email protected]

‡Corresponding author. E-mail:[email protected]

© 2015 Chinese Physical Society and IOP Publishing Ltd http://iopscience.iop.org/cpb　　　http://cpb.iphy.ac.cn

KrF excimer laser,λ =248 nm) with a stoichiometric BFO ceramic target. The deposition was carried out at an oxygen pressure of 4.0 Pa, an ambient temperature of 650^◦C, and a pulse energy of 400 mJ/pulse with a repetition rate of 5 Hz.

After deposition, the sample was cooled down to room tem- perature (RT) at an ambient oxygen pressure of 500 Pa. After- wards, small Pt electrodes (60µm×60µm) were deposited on BFO by PLD at 180^◦C through a metal shadow mask.

The structures of the film were examined by scanning electron microscope (SEM, Ultra 55, Zeiss) and x-ray diffrac- tion (XRD, X’Pert PRO, Pan-Analyzer). The current–voltage (I–V) characteristics were measured by a computer-controlled Keithley 6430 meter. The ferroelectric–polarization (P–

V) hysteresis loops were tested by a Radiant ferroelectric tester (Multiferroics, Premier II, Radiant). The temperature- dependent I–V and P–V measurements were carried out on a Jannis temperature variable probe station (Jannis ST-500-1- 4CX), which was connected with a Radiant tester and a Keith- ley meter. For theI–V testing, the voltage sweeps upwards and downwards in a staircase manner. It is worth mentioning that all of the tests are conducted in a vacuum environment in

order to eliminate the artifacts induced by moisture.

3. Results and discussion

Figure1(a)shows the XRD diffraction pattern for BFO film grown on SRO/STO layers, which exhibits a pure phase of (001)-orientated BFO and an epitaxial structure. When tested at room temperature, they exhibit well-square hystereticP–V loops with a remanent polarization of 58µC/cm²at a maxi- mum bias voltage (Vm) of 14 V. This is very close to the pre- viously reported value of∼60µC/cm²for high quality BFO single-crystal film.^[23]The hysteresis loops also show apparent asymmetric imprint behaviors, which probably originate from the dissimilar electrodes or non-uniform trapped charge dis- tribution adjacent to the metal-oxide interfaces.^[14,15]Typical topological and cross-sectional scanning electron microscopy (SEM) images are shown in Figs.1(c)and1(d), respectively.

The top view SEM image shows a squarish crystalline struc- ture. From the cross-sectional image, we can clearly iden- tify a sharp double-layer structure with thickness values of

∼450 nm and∼60 nm for the BFO and SRO layers, respec- tively.

20 30 40 50 60 70

10² 10³ 10⁴ 10⁵

-15 -5 5 15

-60 -30 0 30 60

♣ BFO

♦ STO

♣

♣

♦

♦

Intensity/arb. units

♣

♦

(a) (b)

Polarization/mCScm-2

Voltage/V

6 V 8 V 10 V 12 V 14 V 16 V

@ 5 kHz

2θ/(Ο)

800 nm 800 nm

BFO SRO STO

(c) (d)

Fig. 1.(color online) (a) XRD diffraction pattern for the epitaxial BFO/SRO/STO heterostructure. (b) Room temperatureP–Vloops measured at 5 kHz with various maximum bias voltages. (c) and (d) SEM images for surface and cross-section, respectively.

Figure 2 shows the polarization and resistive switching properties at room temperature. During the electrical testing, a bias voltage is applied between the Pt top electrode and SRO bottom electrode, as shown in the inset of Fig.2(a). At a low maximum bias voltage of 4 V, noI–V hysteresis is observed.

TheI–V hysteresis loop starts to develop at 6 V, and the loop grows with further increasingV_m. Finally, it stabilizes atV_m of 12 V. It is worth noting that we do not need a power con- suming electroforming process to realize the resistive switch- ing, which is usually required in conventional filament-like re-

sistive switching.^[16,24] Figure 2(b)shows the replotted I–V curves in semi-logarithmic scale, which exhibit rather appar- ent resistive switching at positive bias. In contrast, the resis- tance variation at negative bias range is rather small, indicat- ing apparent asymmetric resistive behaviors. This is likely to be due to the effects of dissimilar electrodes or non-uniform space charge distribution.^[3] We mainly focus on the resis- tive behaviors at positive bias thereafter. To further study the polarization-modulated resistive switching behavior, we use a sequence of pulsed voltages and subsequently record the small bias resistance/current at 3 V. It is revealed that the read cur- rent/resistance as a function of pulsed voltage shows a squar- ish hysteresis loop, as shown in Fig.2(c), indicating a typical memristor behavior. It is interesting that the shape of current-

pulse voltage curve closely matches with the ferroelectric po- larization loop, implying a strong correlation between the re- sistive switching characteristics and the ferroelectric polariza- tion states. To examine the stability of the resistive switching, we also monitor the retention properties, as shown in Fig.2(d).

We find that the resistance (R_L) for the low resistance state (LRS) increases slightly with time rising in logarithmic scale, while the resistance (R_H)for high resistance state (HRS) re- mains rather stable. After 3000 s, the On/Off ratio ofR_H/R_L decreases slightly from 64 to 61. When we extend the time to 10 years by data extrapolation, the ratio still maintains a con- siderable value of 50, which is large enough for practical data storage application.

-15 -5 5 15

10¹ 10³

10¹ 10² 10³ 10⁷10⁸10⁹ 0.1

1 10

-15 -5 5 15

0 100 200

-15 -5 5 15

-4 0 4

8 (b)

Current/nA

Voltage/V -6 V~6 V

-8 V~8 V -10 V~10 V -12 V~12 V -14 V~14 V

(d)

Resistance/GW

Time/s

50

10 years after -14 V~20 ms pulse

after +14 V~20 ms pulse RH/RL

♦ 100

RH/RL ratio (c)

Current/nA

Voltage/V

-60 -30 0 30 60

Polarization/mCScm-2

PtPt 3Current/10 nA Pt

Voltage/V (a)

or BFO SRO STO

Pt A

signal䠖 Pt

Fig. 2.(color online) Room temperature resistive switching properties for a BFO capacitor: (a)I–Vcurves with different maximum bias voltages.

Arrows indicate the sweep sequence, and the inset shows schematic diagram for the device structure for electrical characterizations; (b) semi- logarithmicI–Vcurves; (c) loop of conductive current versus written pulse voltage, measured at 3 V, showing memristor behaviors, which is also compared with a saturatedP–Vloop; (d) resistive switching retention against logarithmic time. The two resistance states (RHandR_L)are read at 3 V which is previously polarized downwards (upwards) by applying 20-ms-pulse voltage of±14 V. The blue dots represent the On/Off resistance ratio (RH/RL) against retention time.

To further understand the conductive mechanisms, tem- perature dependences of resistive switching characteristics and polarization properties are also investigated. Figure3 shows theI–V loops in semi-logarithmic scale along with ferroelec- tric hysteresis loops at different temperatures. The curves are tested by sweeping the bias between ±12 V, which is con- sidered as being large enough to obtain the saturated resistive switching behavior. We find three distinct characteristics in different temperature ranges, which are listed as follows.

(i) At low temperatures ranging from 173 K to 253 K,

theI–V loops are rather small, which seem to be irrelevant to the ferroelectric polarization. Besides, there are two distinct open circuit voltages (OCVs) in both the positive and nega- tive bias ranges. Such OCVs are most probably due to di- electric relaxation-associated charge carriers on both surfaces, which build up open circuit voltages. In addition, we find a hump close to the polarization switching voltage (vanishing at 253 K), which is probably due to the transient displacement current from polarization switching,^[22] as will be discussed later. (ii) From 253 K to 298 K, the leakage current increases

significantly, while both the OCVs and current hump greatly diminish. These indicate that the great increases in mobility of charge carriers and leakage current^[3]have greatly suppressed both the dielectric relaxation-related OCVs and the displace- ment transient current. With increasing temperature, a resis- tive switching hysteresis occurs in theI–Vcurve. At this stage, the resistive switching behaviors can be greatly modulated by the ferroelectric polarization. (iii) From 313 K to 373 K, the low voltageI–V curves are similar to those of stage (ii). Nev- ertheless, a new resistive switching hysteresis (an abnormal) occurs at a large positive voltage (>7 V) in addition to the normal ferroelectric resistive switching. The hysteretic behav- ior becomes more and more apparent with temperature rising.

From Fig.3(b), we can see a well-established squarish polar- ization hysteresis loop for each individual temperature, indi- cating that ferroelectric polarization exists in the whole tem- perature range. It is found that the low temperatureI–V hys- teresis loops and OCVs do not correlate with the ferroelectric polarization, and the polarization-modulated resistive switch- ing only occurs at relatively high temperatures (above 253 K).

-15 -10 -5 0 5 10 15

10^-2 10¹ 10⁴

-15 -10 -5 0 5 10

-60 -30 0 30 60

NRS

PS

|Current|/nA

Voltage/V (a)

OCV PS

173 K 233 K 253 K 273 K 298 K 313 K 373 K (b)

Polarization/mCScm-2

Voltage/V

Fig. 3.(color online) Temperature dependences ofI–VandP–Vloops:

(a) semi-logarithmicI–V curves at a bias sweeping-rate of 3 V/s, and (d)P–Vhysteresis loops at 5 kHz. Here, OCV stands for open circuit voltage, PS denotes the current hump due to the displacement transient current related to polarization switching, and NRS represents the new resistive switching at high bias voltage range.

To further understand the conductive mechanisms, we ex- amine carefullyI–V and polarization properties at different

temperatures. Figure4(a)shows theI–V loops in linear scale, tested at three different bias sweeping speeds (6 V/s, 3 V/s, 1.5 V/s). It shows a clockwise I–V rotation and a current hump, along with a typical characteristic of OCV for eachI–V cure. The OCV voltages decrease as the sweeping speed de- creases. To further illustrate the pulsed voltage effect, we use pulsed voltages of ±14 V to pole the polarization first, and then test the small biasI–V curve (Fig.4(b)). It is found that the OCV can still be observed at a small value ofV_m (3 V) after removing the pulsed voltage. This is very similar to what has been observed by Watanabe in Pb(Ti,Zr)O3/Nb:SrTiO3[25]

and (Pb,La)(Ti,Zr)O₃/Nb:SrTiO₃,^[26] and by Sarafet al. in Nd:STO/Fe:STO^[27] heterostructures. Such OCVs are of- ten attributed to the so-called dielectric relaxation effect, in which charge carriers can be trapped by both electrodes of the capacitor.^[25,26,28]The detrapping of carriers can occur in the process of the band bending, leading to the slow release of charge carriers, and consequently to a relaxation time of

∼10³s.^[26]Sarafet al.^[27]has proposed applying the OCV in data storage. However, from our observation, such an OCV possesses very poor retention property, which is not stable enough for application in memory devices.

It is also interesting to see that the current hump becomes increasingly apparent as the voltage sweep speed rises, which most likely originates from ferroelectric displacement current due to polarization reversal.^[3,21,22] The correlation between the current maximum and polarization switching is further il- lustrated in Fig.4(c). Here, we first use two pulsed voltages (±14 V) to polarize the BFO upwards and downwards, respec- tively, and then measure the I–V curves. By comparing the I–Vcurve and saturatedP–V curve, it can be seen that the cur- rent hump is closely related to the coercive voltage, support- ing the result of our previous work that the current maximum is associated with polarization switching.^[3,21,22]However, it is generally received that the transient displacement current normally occurs on a nanosecond/microsecond time scale at room temperature,^[29]which may not be observable on a time scale of milliseconds, as in our case. To verify whether the displacement current is detectable on a time scale of millisec- onds at low temperatures, we perform a PUND testing to mea- sure the polarization switching time. Figure 4(d)shows the switched polarization value (∆P) dependences of pulsed du- ration at different pulsed voltages. It is found that at a fixed pulsed voltage,∆Pincreases with pulsed duration rising, un- til it reaches the saturated ∆P. The pulse duration required for stabilizing∆Pcan be as long as 200 ms at 173 K, which is much slower than that at room temperature. This indicates that the polarization reversal process at low temperature is signif- icantly slower than that at room temperature. Therefore, it is possible to produce the observed current maximum in theI–V curves at around 173 K.

-3 -2 -1 0 1 2 3 -0.05

0 0.05 0.10

-15 -10 -5 0 5 10 15 -3

-2 -1 0 1

-15 -10 -5 0 5 10 15 -0.4

0 0.4

8 V 10 V 12 V 14 V 16 V

0 100 200 300 400 500 0

100 200

-8 -4 0 4 8 -0.2

0 0.2

polarized down at +14 V polarized up at -14 V (b)

Current/nA

Voltage/V

OCV OCV

Current/nA

Voltage/V

(a) PS

P

Current/nA

Voltage/V

(c) (d)

P

-30 0 30

Polarization/mCScm-2 DP/mCScm-2

Pulse time/ms 6 V/s

3 V/s 1.5 V/s

Fig. 4.(color online) Resistive and ferroelectric polarization properties measured at 173 K: (a)I–Vloops under three different bias sweep-rates of 1.5 V/s, 3 V/s, 6 V/s, respectively; (b) small voltageI–V curves measured at polarization downward and upward states after poling by pulsed voltages of±14 V; (c) a comparison betweenI–Vand saturatedP–Vcurves; (d) variations of switched polarization (∆P) with pulse duration for different bias voltages, obtained by PUND testing. Here, OCV stands for open circuit voltage and PS represents the displacement transient current.

We also measure theI–V behaviors at different bias volt- age sweeping speeds for three relative higher temperatures (shown in Figs.5(a)–5(c)). At 253 K, both the current humps for displacement current and the OCVs almost vanish, and no resistive switching can be observed in a forward bias range.

The current density decreases as the bias sweeping speed rises, which is unlike that at 173 K, exhibiting a normal current re- laxation characteristic. As the temperature increases to 298 K, we can see a well-established ferroelectric resistive switching.

At a high temperature (373 K) (as seen in Fig.5(c)), the re- sistive hysteresis at low voltage is similar to that at 298 K.

Nonetheless, there arises a new resistive switching hystere- sis in a large bias range (>7 V). The new resistive switch- ing loop can be explained by the defect-mediated charge trap- ping/detrapping mechanism.^[30]At a higher temperature, more defect-related charge carriers and charge traps can be acti- vated, which may also produce resistive switching through trap filling and unfilling process. From the inset in Fig.5(c), we can see the distorted polarization loops, which imply that the effects of leakage and charge defects are rather significant.

It is suggested that the high trap filling level shifts the quasi- Fermi level and reduces the interface barriers,^[14,30] hence greatly reducing the resistivity and leading to LRS. With a fur- ther increase of voltage, some of the trapped charges can be de- trapped, resulting in the high resistivity and HRS. It is interest- ing that the new resistive switching hysteresis starts at around the polarization reversal voltage, indicating that it is likely to

be prompted by the polarization reversal. As the ferroelectric polarization bound charge can help to stabilize the trap fill- ing at polarization downward state, once the polarization state switches upwards some of the trapped charges become unsta- ble and will be detrapped at higher voltages. This can also contribute to the formation of the new resistive switching.

To systematically investigate the conductive mechanisms, we analyze theI–V curves in different temperature and bias ranges. For simplicity, we focus on the positive bias range, in which relatively large resistive switching could be obtained.

In ferroelectric oxide films, there may occur some typical conducting mechanisms, including thermionic emission lim- ited (TEL), Poole–Frenkel emission limited (PFL), Fowler–

Nordheim tunneling (FNT), and space charge limited (SCLC), and Ohmic behaviors.[14,15,29,31,32] We replot theI–V curves for both HRS and LRS to fit the individual conducting mech- anisms, and find that theI–V behaviors could fit better to the SCLC relation. The SCLC can be described by the following equation^[14,31–34]

I=

θ

θ+1

9µ ε_rε0V²

8d³ , (1)

whereµis the electronic mobility,ε₀is the permittivity of free space,ε_ris the optical dielectric permittivity,Vis the bias volt- age,d is the film thickness, andθ is a constant related to the charge trap filling status, which is a function of temperature.

-60 -40 -20 0 20

-300 0 300

-12 -8 -4 0 4 8 12 -4000

0 4000

-8 0 8

-60 0 60

Current/nACurrent/nACurrent/nA

6 V/s 3 V/s 1.5 V/s (a) 253 K

373 K 298 K (b)

NRS

Voltage/V (c)

P/mCScm-2

0.1 ms 0.2 ms 0.5 ms

Fig. 5.(color online)I–Vresistive switching loops each measured as a function of bias voltage sweep-rate for three different temperatures: (a) 253 K; (b) 298 K; (c) 373 K. The inset in panel (c) shows theP–Vloops tested at different testing durations. Here, NRS stands for new resistive switching.

We plot both the LRS (upward polarization) and HRS (downward polarization) I–V curves for different tempera- tures in Fig.6. The LRS curves in Fig.6(a)show well-linear relations as described by I ∝Vⁿ.^[15] At a low temperature (∼173 K), the curve deviates from linear relation at low bias, which is likely to be affected by the OCV from dielectric re- laxation. At a moderate bias, the curve conforms to the SCLC relation withn∼2. While in a large bias range, n exceeds 2, which may be related to other conducting mechanisms, e.g.

FNT, or partially filled deep trap state. As temperature in- creases from 253 K to 298 K), n varies from 2.2 to 1.8 in a low voltage range (lnV <1.8), which is nearly consistent with the SCLC relation. For higher voltages (lnV >1.8), it has a largern>2. At high temperatures (313 K and 373 K, Fig.6(a)), indexngradually reduces to below 1.6 at low bias (lnV <1.6), which is between SCLC and Ohmic behaviors.

This may be related to the partially unfilled trap states, ow- ing to the increase of the number of charge traps activated at elevated temperatures. In this case (>298 K), the SCLC rela- tion shifts toward higher voltages (lnV>1.8), which shows a nearly trap-free conductive behavior withn∼2 (1.9–2.1).

Figure6(b) shows the I–V curves for the HRS (down- ward polarization) on a logarithmic scale. The curves fit to the Ohmic and SCLC relations fairy well, similar to those reported by Lee et al.^[14] At low temperatures (<253 K), the curve conforms to the Ohmic conduction (I∝V) below bias voltage of lnV ≈1.8, suggesting that the shallow traps are not fully filled by charge carriers injected from the electrodes. When the bias voltage increases (to lnV >1.9), there occurs an ab- normal hump at around ferroelectric coercive voltage due to the displacement transient current from polarization reversal, which were discussed previously. At moderate temperatures (253 K–273 K), the conduction exhibits Ohmic behavior at low bias (lnV<1.1), beyond which the conduction takes place via a partially deep-trap-filled state, which is reflected with a large slope (I∝Vⁿ,n>3.5) in the graph. At a high temper- ature (<298 K), a similar transition of Ohmic conduction to partially deep-trap-filled conduction (n>2) can be observed below a bias voltage of lnV ≈2.1. While at bias voltages (lnV>2.1), the indexndeclines to 2.3–1.4, which indicates that the conduction behavior evolves from deep-trap-filled to partially filled trap states as temperature increases.

173 K 213 K 253 K 273 K 298 K 313 K 373 K

-2 -1 0 1 2 3

-25 -20 -15

0 0.5 1.0 1.5 2.0 2.5

-20 -15 -10

Ohmic: n~

SCLC: n~2 2.0 2.2 1.8 1.6 1.2

2.6 4.1 4.0 2.6 2.0 1.9

ln|U/V

ln|U/V

2.1

OCV I∝Uⁿ

SCLC: n~

Ohmic

SCLC (a)

(b)

LRS

~5.5 HRS

~1.1

1.3 4.0

3.5 2.3 1.5 1.4

ln|I|/Aln|I|/A

Fig. 6. (color online)I–Vcurves as a function of temperature, which are plotted by logarithmically scaling the fitting relationI∝Vⁿ, for (a) LRS (upwards polarization state) and (b) HRS (downward polarization state). SCLC refers to the case ofn∼2; Ohmic conduction is denoted asn∼1; OCV stands for open circuit voltage.

Finally, we also plot the curves of resistance versus tem- perature, measured at small voltage of 3 V in Fig.7(a). Both

the LRS (downward polarization) and HRS (upward polariza- tion) gradually decrease with the rising of temperature. The On/Off resistance ratio first rises up and then drops down as temperature increases, resulting in a maximum value at around 313 K. It is also worth mentioning that at a low temperature (<253 K) there is also some extent of resistive-switching- like behavior. However, such a resistive variation is related to the OCV and dielectric relaxation effects, which is not stable.

Only above 253 K can we see a robust ferroelectric resistive switching, which is applicable for memory application.

HRS LRS

150 200 250 300 350 400

0.1 10 1000

HRS LRS

3 4 5 6

-3 0 3 6

Resistance/GW

Temperature/K

0 10 20 30 313 K

RH/RL ratio (a)

(b)

0.57 eV

0.47eV 0.04 eV

0.04 eV

ln|resistance|/GW

(1000/T)/K^-1

Fig. 7.(color online) Temperature dependences of resistive properties for both HRS (polarization down) and LRS (polarization down): (a) curves of resistance and On/Off resistance ratioR_H/RLversus temper- ature; (b) curves of ln(R)versus 1000/T, which can be fitted to the Arrhenius equationR∝exp(−E_a/kBT).

The resistance as a function of 1/T is also plotted in Fig.7(b). The lnR–(1000/T)curve shows linear relations in different temperature ranges. The behavior well conforms to the Arrhenius relation(R∝exp(−E_a/k_BT)), whereE_ais the activation energy.^[29,35–37]For the LRS, the activation energy (E_a) deduced from the slope of the linear plot is very small (E_a=0.04 eV) at low temperatures (<253 K). Such a small activation energy indicates that the majority of the charge car- riers are close to free electrons. From Fig.6(a), we can see that in similar temperature and bias ranges for the small activa- tion energy, the conduction behaviors are mainly dominated by Ohmic behavior or dielectric relaxation-related OCVs. When the temperature increases above 253 K, the activation energy rises to 0.57 eV, which corresponds to the SCLC conduction

region shown in Fig.6(a). From the SCLC relation in Eq. (1), the constant related to the trap filling state can be written as θ = (N_c/N_te^{−(Ec−Et)/kBT}, whereN_c andN_t are the effective density of states in the conduction band and the density of states for charge traps, and Ec andEt are the electron ener- gies at the minimum of conduction band and energy of trapped charge, respectively;kBis the Boltzmann constant; and,T is the absolute temperature,^[33,34]Therefore, the SCLC conduc- tion also produces an exponential relation against 1/T, and the activation energy derived from the Arrhenius relation can correspond toE_c−E_t. Thus, the relatively large activation en- ergy (0.57 eV) at above 253 K for the LRS is most probably attributed toE_c−E_tfor the SCLC. For HRS (upward polariza- tion), the activation energy is∼0.04 eV below 198 K, which may be also related to electron conduction, above which its ac- tivation energy increases to 0.47 eV, and it is likely to be cor- respondent toE_c−E_tfor the SCLC. It is also found that there is an apparent variation in activation energy for the SCLCs be- tween the LRS (0.57 eV) and HRS (0.47 eV). The variation of activation energy is associated with the different polariza- tion states, which generate dissimilar bound charge states that are able to attract or repulse charge traps. As a result, we are able to observe an apparent resistive change with a maximum On/Off ratio located at 313 K.

Therefore, at low temperatures (<253 K) the resistive behaviors in the Pt/BFO/SRO structure are mostly dominated by dielectric relaxation current related to OCVs. At rela- tively high temperatures (>253 K), the HRS conducive mech- anisms are governed by both the Ohmic conduction in a low voltage range and the charge-trap-related SCLC behav- ior in a high voltage range. While the LRS is mainly dom- inated by the SCLC conduction. The polarization reversal is able to change the low voltage conductive behavior from Ohmic conduction to SCLC. This causes a significant vari- ation in resistivity, which is likely to be due to the change of trap filled status which can be affected by the polariza- tion bound charge-related interface energy band bending. At high temperatures (>298 K), an additional resistive switch- ing appears in a high bias voltage range. This is likely to be related to the defect-mediated trapping/detrapping effects.

Our observations indicate that in a high quality BFO film, the general ferroelectric resistive switching is more likely to originate from the polarization-modulated SCLC conduction, rather than from the defect-mediated trapping/detrapping pro- cess. In highly defective films, other conduction mechanisms like trapping/detrapping process can also occur. Our work may help to clarify some of previous controversies regarding the ferroelectric resistive switching mechanism.

However, since SCLC is a bulk effect, it is unlikely to pro- duce the observed asymmetric conductive behaviors between the forward and backwardI–V curves. This discrepancy can be accounted for by the existence of a defective layer (e.g.

oxygen vacancy layer) adjacent to Pt/BFO interface, which could be produced in a film deposition process and which acts as charge carrier traps for the SCLC behavior.^[14]Therefore, it is critical to control the defect layer to improve the resistive switching properties for certain applications.

4. Summary

We have analyzed the temperature-dependent ferroelec- tric and resistive switching behaviors at different bias voltages and sweeping rates for the epitaxial BFO. At temperatures below 253 K, the conductive behaviors exhibit ferroelectric displacement current together with OCVs related to dielectric relaxation. Above 253 K, theI–V curves show both SCLC and Ohmic behaviors, which could be modulated by polar- ization switching. While at temperatures higher than 298 K, there occurs an additional resistive switching in a high voltage range in addition to the normal ferroelectric resistive switch- ing, which is likely to be associated with the defect-mediated trapping/detrapping effects.

References

[1] Chua L O 1971IEEE Trans. Circuit TheoryCT-18507

[2] Wang C, Jin K J, Xu Z T, Wang L, Ge C, Lu H B, Guo H Z, He M and Yang G Z2011Appl. Phys. Lett.98192901

[3] Lin Y B, Yan Z B, Lu X B, Lu Z X, Zeng M, Chen Y, Gao X S, Wan J G, Dai J Y and Liu J M2014Appl. Phys. Lett.104143503

[4] Chanthbouala A, Crassous A, Garcia V, Bouzehouane K, Fusil S, Moya X, Allibe J, Dlubak B, Grollier J, Xavier S, Deranlot C, Moshar A, Proksch R, Mathur N D, Bibes M and Barth´el´emy A2012Nature Nan- otech.7101

[5] Yamada H, Garcia V, Fusil S, Boyn S, Marinova M, Gloter A, Xavier S, Grollier J, Jacquet E, Carr´et´ero C, Deranlot C, Bibes M and Barth´el´emy A2013ACS Nano75385

[6] Jim´enez D, Miranda E, Tsurumaki-Fukuchi A, Yamada H, Su˜n´e J and Sawa A2013Appl. Phys. Lett.103263502

[7] Waser R, Dittmann R, Staikov G and Szot K2009Adv. Mater.212632 [8] Sando D, Barthelemy A and Bibes M2014J. Phys.: Condens. Matter

26473201

[9] Clark S J and Robertson J2007Appl. Phys. Lett.90132903

[10] Catalan G and Scott J F2009Adv. Mater.212463 [11] Ji W, Yao K and Liang Y C2010Adv. Mater.221763

[12] Li H, Jin K X, Yang S H, Wang J, He M, Luo B C, Wang J Y, Chen C L and Wu T2012J. Appl. Phys.112083506

[13] Choi T, Lee S, Choi Y J, Kiryukhin V and Cheong S W2009Science 32463

[14] Lee D, Baek S H, Kim T H, Yoon J G, Folkman C M, Eom C B and Noh T W2011Phys. Rev. B84125305

[15] Hu Z Q, Li Q, Li M Y, Wang Q W, Zhu Y D, Liu X L, Zhao X Z, Liu Y and Dong S X2013Appl. Phys. Lett.102102901

[16] Li M, Zhuge F, Zhu X J, Yin K B, Wang J Z, Liu Y W, He C L, Chen B and Li R W2010Nanotechnology21425202

[17] Liu G Z, Wang C, Wang C C, Qiu J, He M, Xing J, Jin K J, Lu H B and Yang G Z2008Appl. Phys. Lett.92122903

[18] Chen X M, Wu G H, Zhang H L, Qin N, Wang T, Wang F F, Shi W Z and Bao D H2010Appl. Phys. A100987

[19] Wang L, Wang Z, Jin K J, Li J Q, Yang H X, Wang C, Zhao R Q, Lu H B, Guo H Z and Yang G Z2013Appl. Phys. Lett.102242902 [20] Wu S X, Ren L Z, Yu F M, Yang K G, Yang M, Wang Y J, Meng M,

Zhou W Q and Li S W2014Appl. Phys. A1161741

[21] Tsurumaki-Fukuchi A, Yamada H and Sawa A2013Appl. Phys. Lett.

103152903

[22] Tsurumaki A, Yamada H and Sawa A2012Adv. Funct. Mater.221040 [23] Lebeugle D, Colson D, Forget A, Viret M, Bonville P, Marucco J F and

Fusil S2007Phys. Rev. B76024116

[24] Gao R L, Fu C L, Cai W, Chen G, Deng X L, Yang H W, Sun J R and Shen B G2014Chin. Phys. B23097702

[25] Watanabe Y and Okano M2003J. Appl. Phys.947187 [26] Watanabe Y1995Appl. Phys. Lett.6628

[27] Saraf S, Markovich M, Vincent T, Rechter R and Rothschild A2013 Appl. Phys. Lett.102022902

[28] Watanabe Y1999Phys. Rev. B5911257

[29] Pabst G W, Martin L W, Chu Y H and Ramesh R2007Appl. Phys. Lett.

90072902

[30] Jiang X L, Zhao Y G, Zhang X, Zhu M H, Zhang H Y, Shang D S and Sun J R2013Appl. Phys. Lett.102233501

[31] Chen Z H, He L, Zhang F, Jiang J, Meng J W, Zhao B Y and Jiang A Q 2013J. Appl. Phys.113184106

[32] Shang D, Wang Q, Chen L D, Dong R, Li X M and Zhang W Q2006 Phys. Rev. B73245427

[33] Lampert M A and Mark P 1970Current Injection in Solids(New York:

Academic Press) p. 363

[34] Lampert M A1956Phys. Rev. B1031648 [35] Mead C1962Phys. Rev.1282088

[36] Zubko P, Jung D J and Scott J F2006J. Appl. Phys.100114113 [37] Ranjith R, Prellier W, Cheah J W, Wang J and Wu T2008Appl. Phys.

Lett.92232905