Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Home Search Collections Journals About Contact us My IOPscience

Scaling and disorder analysis of local I–V curves from ferroelectric thin films of lead zirconate

titanate

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

2011 Nanotechnology 22 254031

(http://iopscience.iop.org/0957-4484/22/25/254031)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 193.140.28.22
This content was downloaded on 06/11/2014 at 22:46

Please note that terms and conditions apply.


IOP PUBLISHING NANOTECHNOLOGY
Nanotechnology 22 (2011) 254031 (11pp) doi:10.1088/0957-4484/22/25/254031

Scaling and disorder analysis of local I –V


curves from ferroelectric thin films of lead
zirconate titanate
Peter Maksymovych1,3, Minghu Pan1 , Pu Yu2 ,
Ramamoorthy Ramesh2 , Arthur P Baddorf1 and Sergei V Kalinin1
1
Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge,
TN 37831, USA
2
Department of Materials Sciences and Engineering and Department of Physics, University of
California Berkeley, Berkeley, CA 94720, USA

E-mail: maksymovychp@ornl.gov

Received 29 October 2010, in final form 30 March 2011


Published 16 May 2011
Online at stacks.iop.org/Nano/22/254031

Abstract
Differential analysis of current–voltage characteristics, obtained on the surface of epitaxial
films of ferroelectric lead zirconate titanate (Pb(Zr0.2 Ti0.8 )O3 ) using scanning probe
microscopy, was combined with spatially resolved mapping of variations in local conductance
to differentiate between candidate mechanisms of local electronic transport and the origin of
disorder. Within the assumed approximations, electron transport was inferred to be determined
by two mechanisms depending on the magnitude of applied bias, with the low-bias range
dominated by the trap-assisted Fowler–Nordheim tunneling through the interface and the
high-bias range limited by the hopping conduction through the bulk. Phenomenological
analysis of the I –V curves has further revealed that the transition between the low- and
high-bias regimes is manifested both in the strength of variations within the I –V curves
sampled across the surface, as well as the spatial distribution of conductance. Spatial variations
were concluded to originate primarily from the heterogeneity of the interfacial electronic barrier
height with an additional small contribution from random changes in the tip–contact geometry.
(Some figures in this article are in colour only in the electronic version)

Electron transport through wide bandgap semiconductors and The common feature of wide bandgap oxides is small
nearly insulating materials has been thoroughly investigated band dispersion, low carrier mobility and an inherently large
to optimize the quality of gate dielectrics in transistors density of shallow and deep levels in the bandgap. This is
and develop viable high-k alternatives to SiO2 [1, 2]. particularly true for ternary oxides, which may exhibit disorder
The recent resurgence of interest in such materials at on A, B and oxygen sublattices [9]. Current sensing scanning
nanoscale dimensions has been stimulated in large part by probe microscopy, where a metal-coated tip acts as one of
the phenomenon of resistive switching [3], where electronic the electrodes in the two-terminal transport measurement,
conductance is altered by current and/or electric field, allowing is emerging as a potent method to investigate electron
implementation of non-volatile memory and logic [4] on a transport properties of such materials, because the localization
variety of length scales. Resistive switching in binary oxides of the probed volume to less than tens of nanometers
is believed to arise from ionic motion in an applied electric allows one to differentiate between intrinsic and defect-
field. In more complex ternary oxides, electron transport can mediated transport [10, 11]. Examples of new phenomena
be modulated by the order parameters, such as spontaneous discovered and characterized by conducting atomic force
polarization [5–7], magnetization [8] and strain, enabling a rich microscopy (cAFM), often in combination with other
variety of intrinsic nonlinear transport phenomena. types of scanning probe microscopy, include polarization-
controlled tunneling across ferroelectric surfaces [7] and tunnel
3 Author to whom any correspondence should be addressed.
junctions [5, 6], local conductivity of ferroelastic domain

0957-4484/11/254031+11$33.00 1 © 2011 IOP Publishing Ltd Printed in the UK & the USA
Nanotechnology 22 (2011) 254031 P Maksymovych et al

Figure 1. (A) X-ray diffraction analysis of the 50 nm Pb(Zr0.2 Ti0.8 )O3 film, representative of all the films studied here. (B) Several I –V
curves obtained on the surface of a 30 nm Pb(Zr0.2 Ti0.8 )O3 film in ultrahigh vacuum at 294 K. The arrows show the direction of the linear
tip-bias ramp. The values on the bottom curve correspond to measured current, while the upper two curves were offset for clarity. (C) Time
dependence of local current measured on two different surface locations at a tip bias of −3.2 V and at T = 300 K. (D) Schematic band
alignment and bending of the Pt-tip/PZT/La(Sr0.3 Mn0.7 )O3 junction and the equivalent circuit diagram (parallel capacitance ignored). PZT is
assumed to be n-doped in this case.

walls in multiferroic oxides [12], filamentary conduction d I /dV measurements with variable-temperature statistics to
through perovskite oxides [13], heterogeneity and breakdown identify the transport-limiting mechanism behind the local I –
of ultrathin dielectrics [14], local manipulation of two- V curve. Our analysis is reminiscent of works on macroscopic
dimensional electron gas at perovskite interfaces [15] and local capacitor measurements that relied on differential transport
variations in catalytic activity of ionic membranes for fuel characteristics [19, 20]. However, the key distinction here is
cells [16]. an emphasis on the temperature dependence, on the one hand,
Despite the numerous advantages of cAFM as a local and on spatially resolved analysis of the transport properties,
technique and a large amount of experimental work on the on the other. Systematic analysis of variations within I –
subject to date, relatively few attempts have been made to V curves obtained at different locations has also revealed
quantify and systematically analyze the cAFM data, and to coherent features that have been attributed to local changes in
answer the fundamental question of the dominant conduction the interfacial electronic barrier height.
mechanism. The difficulties in data interpretation arise due Experiments were carried out in a customized ultrahigh
to a large number of unknown parameters in the cAFM vacuum atomic force microscope (VT STM/AFM, Omicron)
experiment, such as the tip shape and contact area, the using Pt-coated cantilever tips (Mikromasch, CSC37). The
dielectric structure of the contact, and even the tip material. I –V measurements were performed at a pressure of
The localized nature of the measurement may further limit 3 × 10−10 Torr after transferring the sample from ambient
the applicability of conventional 1D semiconductor models without any subsequent treatment, using a variable-bandwidth
for metal–semiconductor interfaces, as recently demonstrated current preamplifier (FEMTO DPLCA-200) with a noise floor
for the analysis of space-charge-limited conduction in polymer of ∼100 fA at the highest gain setting of 109 V A−1 . The 30
films [17]. Moreover, many parameters of the material itself, and 50 nm ferroelectric films of Pb(Zr0.2 Ti0.8 )O3 (100) were
such as the doping and dopant profile, trap and surface state grown by pulsed-laser deposition on a low miscut (<0.1◦ )
density, the type of traps or effective electron mass, may not be SrTiO3 (001) substrate, buffered by a 50 nm La0.7 Sr0.3 MnO3
known or reliable [18]. (LSMO) conducting electrode [7]. The crystal structures
In this paper, we present a statistical analysis of local I –V of the PZT films have been studied by x-ray diffraction
curves that aims to differentiate between candidate conduction (XRD) (Panalytical X’Pert MRD PRO). A typical XRD 2θ –
mechanisms behind local conductivity through a ferroelectric θ scan of PZT films (figure 1(A)) reveals the crystallinity and
semiconductor PbZr0.2 Ti0.8 O3 (PZT) films grown epitaxially crystallographic orientation of the PZT films. The epitaxial
on La0.7 Sr0.3 MnO3 (LSMO) oxide electrodes on a SrTiO3 growth was established from the presence of only 00L type
substrate. We have combined local current–voltage ( I –V ) and diffraction in the diffraction pattern in the wide scan range. The

2
Nanotechnology 22 (2011) 254031 P Maksymovych et al

Table 1. 1D transport equations for the candidate conduction mechanisms in PZT films.
√ 3/2
2m PZT φB
E 2 exp(− 8π
3
Fowler–Nordheim tunneling I = A eff 8π hm
e m Pt
PZT φB 3he E
)

Schottky emission I = A eff A ∗ T 2 exp(− kTe (φB0 − 4πeEε0mεh ))
Low-field migration I = eμn E bulk exp(−E F /kT ), n —dopant volume density,
E F —Fermi level, E bulk —electric field in the bulk of the film
Poole–Frenkel I = eμn E bulk exp(− kTe (E i0 − eE bulk
4π ε0 εh
)), E i0 —trap ionization energy
2
9εμE bulk
Space-charge-limited (Mott–Gurney law) I = A eff 8L
, L —thickness of the dielectric, V —applied bias

out-of-plane lattice constant was measured to be ∼4.22 Å for mechanisms can therefore be partitioned into interface-limited
PZT films (c/a ∼ 1.07), which indicates that the PZT layers contributions, for example by Schottky emission and Fowler–
are highly strained on the STO substrates with the c axis along Nordheim tunneling, and bulk-limited contributions, such as
the (001) direction. low-field migration, Poole–Frenkel hopping and space-charge-
Despite having a relatively large bandgap (>3 eV) and limited conduction (table 1) [24–26].
being as thick as 30–50 nm, PZT films exhibited significant In the equations given in table 1, e—the elementary
local conductance in the current range from 1 pA to 100 nA charge, k —Boltzmann constant, ε —low frequency dielectric
at a negative tip bias less than −2 V (figure 1(B)). The constant, εh —high frequency dielectric constant, ε0 —vacuum
conductance strongly depended on the sign of the spontaneous permittivity, φB —barrier height, μ—electron mobility (can
polarization in PZT as detailed in our earlier publication [7]. be field-dependent), Aeff —effective transport area of the tip–
In all the experiments presented here, the films were poled surface junction, h —Planck constant and m i —electron (hole)
to ascertain upward polarization orientation relative to the effective mass.
film’s surface. Typical I –V curves (figure 1(B)) were highly At a given bias polarity, electron transport will be limited
rectifying and exhibited negligible hysteresis between forward primarily by the reverse-biased interfacial Schottky barrier
and backward directions of the tip-bias ramp. The conductance (here we assume that it is localized at the tip–surface junction),
also showed negligible time dependence on the time scales up but it may also have a strong contribution from the nonlinear
to 100 s (figure 1(C)). Finally, acquisition of I –V curves was (e.g. Poole–Frenkel) resistance of the bulk. It is common
not associated with irreversible changes of surface topography to attempt to linearize the I –V curves in the normalized
(not shown). Together these measurements indicate that the coordinates that correspond to the respective mechanisms.
electronic state of the probed volume is unchanged in the Here we will assume the abrupt junction approximation for
course of the measurement, ruling out the scenarios of mixed the metal–ferroelectric contact, making it equivalent to the
electron–ionic transport or effects due to polarization dynamics standard Schottky model [24]. Therefore the maximum electric
(including displacement current) and charge injection during field in the contact region, which factors in the above Fowler–
I –V scans. Parenthetically, a simple estimate of the oxygen Nordheim
 and Schottky emission mechanisms, is E m =
vacancy displacement can be derived from the Nernst–Einstein 2eND
εε0
(V+ Vbi ), where Vbi is the built-in bias due to the
equation as  = τ eD E/kT , where τ —diffusion time, difference between metal and ferroelectric work functions and
D —diffusion constant, e—electron charge, E —electric field, the presence of charged species on the surface, etc, ND is
k —Boltzmann constant and T —temperature. Using D ∼ the donor (or acceptor) density (combined shallow and deep
10−18 m2 s−1 , conservatively estimated from the data on a levels) and V is applied bias. This relationship represents
related SrTiO3 [21], the ionic displacement during a 1 s voltage the limiting case when the contact area is comparable to or
pulse would be ∼0.4 nm, or only about a unit cell in the electric larger than the film thickness (the 1D approximation) and the
field as high as 106 V m−1 at 300 K. width of the Schottky barriers is significantly smaller than
At the same time, the magnitude of local current varied by the film thickness. Although the properties of the metal–
almost one order of magnitude across the surface (see below), ferroelectric contacts are still under debate [23], there is indeed
implying significant variations of the material parameters a growing awareness that the Schottky barrier width could be
responsible for the interfacial barrier height (the density of as thin as several nanometers [18], in part due to an extremely
surface states) and the shape of the local tip potential (vacancy large density of vacancies in the skin region necessary to
density) across the surface [22]. Analysis of the variability stabilize ferroelectric polarization. On the other hand, the
in I –V curves enables spatially resolved mapping of these characteristic diameter of our tip–contact area is ∼40–60 nm,
parameters. as judged from the ∼30 nm width of a ferroelectric domain
As is now well accepted [18, 23], an equivalent circuit wall [27], typically observed on the surface of the 50 nm
diagram of the PZT film (and related perovskite oxides) PZT film (for more details see appendix A in [28]). We
consists of two back-to-back Schottky diodes, one due to the also note that the validity of the above assumption requires
buried interface between the PZT film and bottom metallic a relatively fast response of the donor and/or acceptor levels
electrode (La0.7 Sr0.3 MnO3 ) and the other one due to the in the ferroelectric (primarily charging/discharging) to applied
top tip–surface junction, separated by a resistor representing bias. While this is not generally the case for PZT [29], our
the bulk of the film (figure 1(D)). The electron transport observation of time-independent conductance on the time scale

3
Nanotechnology 22 (2011) 254031 P Maksymovych et al

to assume a different relationship between applied potential


and electric field at the interface (e.g. E ∝ V ), the ability
of the differential normalized analysis presented below to
distinguish between candidate mechanisms does not diminish
and requires only accounting for the respective changes in the
scaling exponents.
In the abrupt junction approximation, the Schottky
emission mechanism should yield a linear I –V curve in
log(I ) = f (V 0.25 ) coordinates, and Fowler–Nordheim
tunneling in log(I /V ) = f (V 0.5 ) coordinates. As seen
in figure 2, the observed I –V curves are moderately well
linearized using both systems of coordinates, although Fowler–
Nordheim tunneling provides a slightly better linearization.
Local I –V curves are also strongly temperature-dependent
(figure 3(A)), with average local current (extracted from 100
to 200 individual I –V curves acquired on a grid) increasing
exponentially between ∼50 and ∼ 350 K (figure 3(B)). From
the analysis of the temperature-dependent I –V curves using
standard Arrhenius (log(I ) ∼ −φ  B /kT , figure 3(C)) and
Schottky (log(I /T 2 ) ∼ (−[φB0 − 4πeEε0mεh ]/kT ), figure 3(D))
Figure 2. Representative I –V curves plotted in coordinates that
linearize the I –V curves due to Fowler–Nordheim (left) and
coordinates, it is also evident that the I –V curves cannot
Schottky emission (right) transport mechanisms. The original curves be described by a single activation energy (Arrhenius) or
are shown in the inset. The electric field at the interface is inferred interfacial barrier height (Schottky) over the whole temperature
within the abrupt junction approximation (see text for details). range. Indeed, as seen in figure 3(C) there is a marked decrease
of the slope of the I –V curves below ∼150 K in the Arrhenius
of 100 s (figure 1(C)), together with the relatively small thermal coordinates (from 100 to 230 meV at high temperatures to
activation energy of ∼0.2 eV (suggesting relatively shallow 60–100 meV at low temperatures), and the structure is even
levels) supports this assumption. On the other hand, if we were more complicated in the Schottky analysis revealing a similar

Figure 3. (A) Temperature-dependent I –V curves averaged over 100–200 random locations across PZT surface at each temperature.
(B) Temperature dependence of current at three values of the tip bias. (C) Arrhenius analysis of average current values at several values of tip
bias. (D) Schottky analysis of the average current values.

4
Nanotechnology 22 (2011) 254031 P Maksymovych et al

Table 2. Differently normalized conductance for each of the


mechanisms in table 1, within the abrupt junction approximation (see
text).
 3/2
π 2m PZT εε0 φB
Fowler–Nordheim d log I
dV
= V +V
1
bi
+ 34he 2eND (V +Vbi )3/2
tunneling
3
Schottky emission d log I
dV
= 4kT
e
( 8πN2Dε3eε2 ε )1/4 (V + Vbi )−3/4
0 h
−1
Low-field migration d log I
dE
= E bulk 
e3 −0 . 5
Poole–Frenkel d log I
dE
= 1
E bulk
+ 1
kT 4π ε0 εh
E bulk
−1
Space-charge-limited d log I
dE
= 2 E bulk
(Mott–Gurney law)

d log(I )/dV has the following advantages:


(1) It can be measured by modulating the applied bias with
a small magnitude sinusoidal AC voltage, filtering the
current signal with a lock-in amplifier and recording
its first harmonic simultaneously with the I –V curve
(figure 4(A)). The obvious advantage of using lock-in is
to increase the signal-to-noise ratio up to 100-fold [31].
d log(I )/dV is then obtained numerically as (d I /dV )/I ,
figure 4(B).
(2) d log(I )/dV is independent of a significant number of
unknown parameters from the transport equations as seen
in table 2.
Simplification of the governing equations is manifested
by the fact that nearly identical d log(I )/dV = f (V )
dependences are obtained for I –V curves that differ by
as much as an order of magnitude in the absolute current
(figure 4(B)), implying a significant contribution of the
prefactor (e.g. contact area) to such a difference. Such
‘normalization’ is also strong evidence to support the
Figure 4. (A) I –V curves and their respective derivatives obtained at validity of d log(I )/dV as a benchmark observable for
294 K using lock-in amplifier detection with an AC modulation of assigning the dominant conduction mechanism. More
10 mV at a frequency of 1.2 kHz. The inset shows lock-in derivatives examples of the normalization are presented in the last
overlaid on top of numerical derivatives. This procedure was used to
section of this paper, dedicated to the analysis of the
calibrate the output of the lock-in amplifier. (B) d log(I )/dV curves
obtained from four I –V curves (inset) using the lock-in amplifier. spatial variations among the I –V curves on the surface.
(3) Temperature and bias dependence of d log(I )/dV provide
a much more robust test for the transport mechanisms
change of the slope below 150 K as well as a smaller, but still compared to conventional fitting approaches. In particular,
significant, kink at ∼200 K (figure 3(D)). temperature dependence is observed only for Schottky and
Thus, conventional analysis of I –V curves does not Poole–Frenkel mechanisms. Note that, fundamentally,
yield conclusive assignment of the mechanism, neither by the temperature dependence of d log(I )/dV in the above
way of fitting of the bias dependence nor the temperature mechanisms arises from the Schottky effect [24], i.e. the
dependence of the I –V data. To address this problem, electric-field-induced lowering of the trapping energy (as
we turned to current-normalized differential conductance, in the Poole–Frenkel mechanism) or the interfacial barrier
d log(I )/dV , as an observable that simplifies the transport height (as in Schottky emission). The various properties
equations and provides a more rigorous test for the candidate of the current-normalized differential conductance are
mechanisms. Conceptually, using d log(I )/dV is similar to exemplified in figure 5, where model parameters were
several previous works, where instead the field dependence used to calculate I –V characteristics and numerically
of d log(I )/d log(E) [19, 20] (or d log(I )/d log(V ) [30]) compute d log(I )/dV .
was proposed as a descriptor of the mechanism. Here
we focus not only on a different functional form of the To analyze the temperature-dependent behavior of
differential conductance, but also emphasize its temperature experimental d log(I )/dV we have sampled 100–200 I –V
dependence and spatial variations as key observables to curves at random locations on the PZT surface and numerically
differentiate between mechanisms. As we will show below, evaluated d log(I )/dV for each of them (this was done to
the bias dependence of differentially normalized conductance increase the speed of data acquisition at the expense of
is strongly affected by the built-in field and must therefore be increased noise). Subsequently, individual d log(I )/dV curves
used with caution. were averaged across the ensemble at each temperature,

5
Nanotechnology 22 (2011) 254031 P Maksymovych et al

Figure 5. (A) I –V curves calculated using equations for Fowler–Nordheim tunneling and Schottky emission (see text) using the following
parameters: FN: φB = 0.8 eV, A eff = 314 nm2 , ND = 1021 cm−3 , ε = 30 SE : φB = 0.65 eV, Vbi = 0, ND = 1020 cm−3 , ε = 30,
εh = 6.5 [38], T = 330 K and A eff = 314 nm2 (effective tip radius 10 nm). Inset shows linearization of the calculated I –V curves in their
respective coordinates. The parameters were chosen to approximate experimental current levels and the overall shape of the I –V curves. The
analysis presented in this paper does not depend on the exact choice of these parameters. (B) d log(I )/dV = f (V ) as a function of
temperature for Schottky emission mechanism and temperature dependence of d log(I )/dV at 1.7 V (inset). (C) d log(I )/dV = f (V ) for
various values of built-in field within the Fowler–Nordheim tunneling transport. (D) The exponent of the larger bias linear segment of
d log(I )/dV as a function of built-in potential for Fowler–Nordheim and Schottky emission. In all calculations the sign of the tip bias is
opposite to experimental negative values for convenience.

which significantly reduced the noise level due to digital emission or Poole–Frenkel conduction at |V | > 4 V, compare
differentiation. This experiment was repeated three times, in figures 5(B) and (D). Notably, d log(I )/dV scales as T n ,
the temperature range from ∼90 to 320 K, with the results with n = −0.8:−1.3 at high bias (figure 6(D)), in excellent
shown in figures 6(A)–(C). The following trends are observed: agreement with the anticipated n = −1 (see above). The
(1) the exponent α in d log(I )/dV = f (V α ) is the same crossover between the two mechanisms can be seen not only
and roughly constant when the magnitude of the tip bias from the temperature dependence, but also from the exponent
exceeds ∼4 V; (2) the absolute value of d log(I )/dV at each α . It is particularly apparent for the corresponding curves
voltage in this range is temperature-dependent and increases at T = 252 K in figure 6(A), T = 291 K in figure 6(B)
with decreasing measurement temperature, figures 6(A)–(D) and T = 260 K in figure 6(C). In each case, the slope of
and (3) α at |V | < 4 V is larger than that at |V | > 4 V and d log(I )/dV decreases by 1.2–1.5 units at |V | = 3.5–4 V. The
d log(I )/dV is virtually temperature-independent in this range, decrease of the magnitude of the slope by a similar amount is
figure 6(B). indeed expected for the crossover between Fowler–Nordheim
A model that fits the observed trends in normalized tunneling and Poole–Frenkel hopping (figure 5(D)). We can
conductance is that of two rather than one conductance- further rule out low-field migration and space-charge-limited
limiting mechanisms and a crossover between the mechanisms conduction as contributing at any point of the I –V curve,
at a certain value of the tip bias. The statistically negligible simply because the slope of d I /dV is not constant, figure 4(A),
temperature independence of d log(I )/dV at low tip bias as expected for low-field migration and its magnitude is 4–5,
|V | < 4 V (figure 6(D)) points to Fowler–Nordheim tunneling much larger than unity expected for the Mott–Gurney law.
as the most likely interfacial mechanism. On the other hand, As the temperature dependence of Schottky emission
the increase of the magnitude of d log(I )/dV with decreasing and Poole–Frenkel hopping is indistinguishable without a
temperature almost unambiguously points to either Schottky knowledge of the material-specific constants, one may suggest

6
Nanotechnology 22 (2011) 254031 P Maksymovych et al

by the relative increase of the overall conductivity of the 30 nm


PZT film compared to a 50 nm PZT film grown by the same
methods [7].
The absolute values of the exponent α ranges from
−2.5 to −3 for the low bias, interfacially limited region
of the I –V curve, and from −1.2 to −1.5 for the high-
bias region and low-temperature I –V curves below ∼200 K.
These values are significantly larger than the expectation for
the Fowler–Nordheim tunneling (−1.42), Schottky emission
(−0.75) and Poole–Frenkel hopping (−0.66) in the abrupt
junction approximation (in the case of a linear relationship
between electric field and applied potential E ∝ V , the values
would be ∼−1.9 for Fowler–Nordheim tunneling and −0.5
for Schottky emission). We believe that the reason for this
discrepancy is the value of the built-in potential that factors into
the total applied field (see table 2) and can be very large (up to
2 V) for PZT films and other ferroelectrics in vacuum [7, 36].
Indeed, estimating the value of d log(I )/dV as a function of
built-in field reveals a very strong dependence of the numerical
value (figure 5(D)) and in Fowler–Nordheim tunneling the
slope can be well in excess of −2.2. Furthermore, as seen in
figure 5(C), the d log(I )/dV cannot be simply characterized by
a single exponent, but rather the slope significantly increases
with decreasing tip bias and can even become as high as −5.
Intriguingly, the sign of the deviation from the expected values
at zero built-in field points to the sign of the built-in potential.
Figure 6. (A)–(C) Temperature-dependent d log(I )/dV = f (V )
extracted from 100 to 200 I –V curves randomly acquired on the In our case, observation of the consistently large values reveals
surface. (A)–(C) represent measurements done on different days and that the built-in potential has an opposite direction from the
using different AFM tips. (D) d log(I )/dV as a function of applied bias relative to the bottom electrode. This is consistent
temperature at specific values of tip bias extracted from (A). Based with the observed shift of the ferroelectric hysteresis loop,
on the least-squares fit, the values at −3.3 V scale approximately as
d log(I )/dV ∝ T −0.2±0.1 while those at −7.0 V as
where the ferroelectric switching is induced at smaller positive
d log(I )/dV ∝ T −0.8±0.1. tip bias compared to negative tip bias [7].
According to our analysis, the temperature-dependent
low-bias conductance between 250 and 330 K should
that the crossover takes place between Fowler–Nordheim correspond to Fowler–Nordheim tunneling. This mechanism in
tunneling and Schottky emission. Both Fowler–Nordheim its original inception describes field emission of electrons from
tunneling in series with either Schottky emission or Poole– the metal surfaces [34] and is indeed a largely temperature-
Frenkel hopping are consistent with our data. In the former independent mechanism, because of the negligible thermal
case, the crossover would occur between two mechanisms that broadening of the Fermi distribution function in the metal from
are both interface-limited, while in the latter, the crossover 0 to 300 K. This was recently observed for a ultrathin tunnel
occurs between the interface and bulk-limited mechanism. junction of SrTiO3 [35], which is electronically similar to PZT.
We find the first scenario highly unlikely. In fact, Fowler– However, electron tunneling from the metal electrode into the
Nordheim tunneling transitions into the Child–Langmuir law insulator’s conduction band (or hole into the valence band)
of space-charge-limited conduction at high fields [32]. Also, can be mediated by empty (or filled) trap states energetically
the spatially resolved analysis presented below indicates a located in the bandgap of an insulator, in the so-called trap-
different degree of disorder for the low- and high-bias regions assisted tunneling [34]. The Fowler–Nordheim tunneling
of the I –V curve, indicating a difference in the material equation in this case is modified by a field-independent
properties that are responsible for each regime. We therefore Boltzmann term, exp(−ϕi /kT ) [35], where ϕi is the trapping
conclude that the model that best fits the observations is that of energy. We estimate ϕi to be approximately 0.2 eV based on
a Fowler–Nordheim tunneling in series with a Poole–Frenkel- the Arrhenius fit. Note that, because the electron’s escape from
limited bulk region. The crossover occurs because at a tip bias the trap is not field assisted in this case [33], d log(I )/dV is still
larger than 4–5 V in magnitude the resistance of the interfacial temperature-independent despite the temperature dependence
region becomes significantly smaller than that of the bulk of current itself.
region. A similar crossover was suggested to occur for silicon One key advantage of scanning probe microscopy is the
oxide thin films [33] as well as other insulators [20], although ability to resolve the variations of the observables across the
to our knowledge local transport measurements have not yet surface. In the case of cAFM, a spatial resolution of ∼10 nm
been discussed in this context. Our conclusion regarding the can be readily achieved. Here we use the spatial correlations of
involvement of the bulk-limited conductivity is also supported the I –V curves to judge the onset of the crossover between two

7
Nanotechnology 22 (2011) 254031 P Maksymovych et al

Figure 7. (A) Statistical distribution of I –V curves based on an array acquired on a 30 × 30 grid with a resolution of 10 nm, plotted as a 2D
histogram. (B) Individual histograms at several values of tip bias.

Figure 8. Spatial distribution of current as a function of increasing negative tip bias. The pseudocolor maps represent log(I ) as a function of
tip position on a 28 pixel × 30 pixel grid with a spatial resolution of ∼10 nm. The colorbar scale is log(I × 10−1 nA).

mechanisms, which supports the conclusions derived above shows an average I –V curve, which was plotted as a histogram
using the temperature dependence of the current-normalized of observed current values at each bias value. The magnitude of
differential conductance. This analysis can be eventually current varies by ∼1 order of magnitude across the PZT surface
extended for systematic mapping of the material properties that (figure 7(B)), most likely due to inherent electrostatic disorder
enter the corresponding transport equations. in the film [36, 37], with the logarithm of current following
Spatially resolved analysis of the variations among the I – approximately a Gaussian distribution (figure 7(B)). What is
V curves was based on 900 I –V curves acquired on a 30 × 30 rather striking, however, is that the standard deviation is bias-
grid of points with a pixel resolution of ∼10 nm. Figure 7(A) dependent and there is a relatively abrupt crossover from

8
Nanotechnology 22 (2011) 254031 P Maksymovych et al

Figure 9. Spatial distribution of: (A) reverse current at −3.9 V; (B) the respective d log(I )/dV values (d log(I )/dV was numerically obtained
from the I –V curves, then fitted with fourth-order polynomial to calculate the values at −3.9 V); (C) reverse current at −3.9 V normalized by
the current values at −5.5 V for each I –V curve and (D) scaling exponent α of d log(I )/dV as a function of bias, obtained by fitting
d log(I )/dV in the bias range from −3 to −4.5 V (approximate window of interfacially limited conductance). The x and y axes in each case
are pixels (with a resolution of ∼10 nm) and the colorbars are the mapped values in their respective units.

wider distributions at V > −3.5 V to significantly narrower on the data in figure 8: (1) simple normalization, where the
distributions at V < −3.5 V, figures 7(A) and (B) (from current at a chosen bias (−3.9 V in figure 9) was divided by
σ = 0.28 at −5.5 V to σ = 0.47 at −2.8 V). Furthermore, the current value at another bias value (−5.5 V in figure 9(C))
the variations between I –V curves are not random, but rather and (2) d log(I )/dV analysis in the low-bias region (from
exhibit strong correlations across the surface, figure 8. In the −2 to −4.5 V) which we assigned to interfacially limited
low-bias region from −2.3 to −2.8 V, spatially coherent and conductance. d log(I )/dV was first numerically calculated for
discrete regions of enhanced conductance (∼30 nm diameter) each pixel, then fitted to a fourth-order polynomial, and then
are clearly visible (figure 8). Between −2.8 and −4 V, these the absolute values of d log(I )/dV (at −3.9 V in figure 9(B)),
‘hot-spot’ features begin to blend in with each other and with as well as the exponent α were calculated from the fit.
the surrounding areas, effectively disappearing. For −4 to The goal of simple normalization, I (−3.9 V)/I (−5.5 V),
−6 V, the spatial distribution of conductance is significantly is to reveal whether the disorder originates from the bias-
more uniform than in the low-bias region and it is also independent terms of the corresponding transport equations.
distinct because the surface areas that exhibited different These include the prefactor (e.g. contact area), as well as zero-
eφ 0
conductivity, e.g. the left versus right half of the image, field barrier-height terms, such as in exp(− kTB ) the Schottky
become approximately similarly conducting. The observed emission equation. Normalization should reduce the strength
variation of the disorder with tip bias is consistent with a of the disorder and change its spatial pattern if its origin is
crossover between two mechanisms, particularly since the tip bias-independent. The comparison of panels in figure 9(A)
bias corresponding to the narrowing of the current distribution (current) versus figure 9(C) (normalized current) reveals that
is nearly identical to the value where the exponent α is reduced the disorder pattern does remain after normalization. However,
(figures 6(A), (C) and (D)). The coherent and distinct features its details are slightly altered, particularly at the top and bottom
in the low-bias range may signify either extended defects or of the map. Upon careful examination one can see that the map
defect gradients at the interface or surface, the origin of which in figure 9(A) has several ‘streaks’ crossing the image (also
is subject to further investigation. The effect of these defects seen in figure 8 at −2.35, −2.52 V). These streaks clearly
on the overall conductance diminishes with increased bias, as manifest changes in the tip–contact geometry (e.g. contact
the bulk-limiting mechanism (and apparently more uniform area) along the fast x axis of the scan, so that along the slow-
distribution of the relevant bulk parameters) becomes more scan y axis the consecutive lines are abruptly different from
dominant at high bias. each other. We can therefore infer that changes revealed in
To gain more insight into the origins of the observed simple normalization, as well as the corresponding disorder,
disorder, we have carried out two types of numerical analysis originates from the tip–contact effects.

9
Nanotechnology 22 (2011) 254031 P Maksymovych et al

in figures 9(A)–(C). These trends are confirmed by the 2D


histograms in figure 10: d log(I )/dV spatially anti-correlates
with the raw current (figure 10(A)), exhibits a much cleaner
anti-correlation with the normalized current (figure 10(B)),
while α does not exhibit any significant correlation with the
raw current.
The spatial correlation between d log(I )/dV and the
absolute current yet again emphasizes that not all disorder
originates from prefactors in the transport equation, such
as the contact area. However, a stronger correlation of
d log(I )/dV with normalized current (figure 10(B)), both of
which eliminate the prefactor, compared to that of d log(I )/dV
with the raw current (figure 10(A)) confirms our earlier
conclusion that the prefactor contributes some of the disorder,
particularly in the lower half of the image. We can even assert
that normalizations ‘recover’ some of the intrinsic features,
such as two clear depressions at the bottom of the map in
figure 9(B).
At the same time, the lack of significant spatial correlation
of the exponent α with the disorder pattern observed in the
current identifies that the spatially correlated variations in
d log(I )/dV originate primarily from the (near) zero-field
terms, such as the interfacial barrier height in the Fowler–
Nordheim equation, rather than the changes in the built-in
field (which would change the exponent, figure 5(D)). It is
straightforward to assert that changing the barrier height does
indeed shift the magnitude of the d log(I )/dV = f (V ) curve,
but has a smaller effect on its exponent. Furthermore, the
higher barrier yields a higher absolute value of d log(I )/dV
and a lower value of current, which explains why d log(I )/dV
is anti-correlated with both the raw and normalized current
values (figure 10). Finally, within our approximations for the
electric field at the interface and 1D transport equations, the
absolute values of d log(I )/dV in figure 9(B) are consistent
with Fowler–Nordheim tunneling, a built-in bias from −1.5
to −2.5 V, total donor/acceptor density of 1024 –1026 m−3
and zero-field barrier height from 0.4 to 0.8 eV. In the
future, macroscopic characterization of the film properties
(particularly the respective trap distributions and response
times), combined with more detailed mapping of the I –V
curves, can help in making these estimates more accurate.
Figure 10. 2D histograms of the data from figure 9, revealing In summary, we have identified the dominant conduction
significant correlation between: (A) reverse current (measured at
−3.9 V) and d log(I )/dV ; (B) unambiguous anti-correlation mechanisms behind local I –V curves through a 30 nm film
between reverse normalized current and d log(I )/dV and (C) lack of of PbZr0.2 Ti0.8 O3 using a combined analysis that relies on
observable correlation between reverse current and scaling of the use of current-normalized differential conductance and
d log(I )/dV with applied bias (scaling exponent α extracted as spatial variations of the I –V curves across the surface.
described before). Scale bar in each case is counts.
The observed transport characteristics are consistent with a
model where I –V curves are interfacially limited by trap-
assisted Fowler–Nordheim tunneling at low bias and bulk
The spatial analysis of d log(I )/dV has likewise turned limited by Poole–Frenkel hopping at high bias, with a
out to be very informative. It is easily seen that the crossover near −4 V. As the temperature is lowered, the
spatial variations of the absolute magnitude of d log(I )/dV experimentally detectable conductance shifts into the high-
in figure 9(B) reveal the same overall pattern as figures 9(A) field regime, eventually dominated by Poole–Frenkel hopping
and (C), albeit in an anti-correlated fashion—the magnitude of below ∼200 K. Spatially resolved analysis of the I –V
d log(I )/dV is smaller in the regions of higher current. At curves has further revealed that the two mechanisms exhibit
the same time, the scaling of d log(I )/dV with bias (exponent significantly different degrees of disorder. The disorder, in
α in d log(I )/dV = f (V α )) is spatially heterogeneous, but turn, originates partly from the changes in the interfacial
without a clear pattern, and with little correlation to the patterns barrier height and partly from the changes in the contact

10
Nanotechnology 22 (2011) 254031 P Maksymovych et al

area. The two effects can be reasonably well separated using [14] Sune J, Nafria M, Miranda E, Oriols X, Rodriguez R and
differential conductance. The methodology presented here is Aymerich X 2000 Failure physics of ultra-thin SiO2 gate
applicable to any wide bandgap semiconductor or polymer oxides near their scaling limit Semicond. Sci. Technol.
15 445
material, as well as to a variety of transport mechanisms
[15] Cen C, Thiel S, Mannhart J and Levy J 2009 Oxide
provided their functional form is known. We therefore nanoelectronics on demand Science 323 1026
envision that the systematic analysis of normalized I –V [16] O’Hayre R 2004 Ionic and electronic impedance imaging using
curves will shed new light on the rapidly growing number atomic force microscopy J. Appl. Phys. 95 8382
of local transport experiments on wide bandgap oxides and [17] Reid O G, Munechika K and Ginger D S 2008 Space charge
enable more quantitative characterization of their dielectric and limited current measurements on conjugated polymer films
interfacial properties and phase transitions, many of which are using conductive atomic force microscopy Nano Lett.
8 1602
relevant to applications in information technology and energy
[18] Scott J F 2000 Ferroelectric Memories (New York: Springer)
harvesting and storage. [19] Niklasson G and Brantervik K 1986 Analysis of
current–voltage characteristics of metal–insulator composite
Acknowledgments films J. Appl. Phys. 59 980
[20] Mikhaelashvili V, Betzer Y, Prudnikov I, Orenstein M,
Research was conducted at the Center for Nanoscale Materials Ritter D and Eisenstein G 1998 Electrical characteristics of
Sciences (PM, MP, APB, SVK), sponsored at the Oak metal–dielectric–metal and metal–dielectric–semiconductor
Ridge National Laboratory by the Division of Scientific User structures based on electron beam evaporated Y2 O3 , Ta2 O5
and Al2 O3 thin film J. Appl. Phys. 84 6747
Facilities, US Department of Energy. The work at Berkeley
[21] Noll F, Munch W, Denk I and Maler J 1996 SrTiO3 as a
is partially supported by the SRC-NRI-WINS program as well prototype of a mixed conductor: conductivities, oxygen
as by the Director, Office of Science, Office of Basic Energy diffusion and boundary effects Solid State Ion. 86 711
Sciences, Materials Sciences Division of the US Department [22] Simmons J G 1971 Conduction in thin dielectric films J. Phys.
of Energy under contract no. DE-AC02-05CH1123. D: Appl. Phys. 4 613
[23] Kohlstedt H et al 2008 arXiv:0810.4272v1
[24] Sze S M 1981 Physics of Semiconductor Devices (New York:
References Wiley–Interscience)
[25] Pabst G W, Martin L W, Chu Y and Ramesh R 2007 Leakage
[1] Simmons J G 1967 Poole–Frenkel effect and Schottky effect in
metal–insulator–metal systems Phys. Rev. 155 657 mechanisms in BiFeO3 thin films Appl. Phys. Lett.
[2] Wilk G D, Wallace R M and Anthony J M 2001 High-kappa 90 072902
gate dielectrics: current status and materials properties [26] Pintilie L, Vrejoiu I, Hesse D, LeRhun G and Alexe M 2007
considerations J. Appl. Phys. 89 5243 Ferroelectric polarization-leakage current relation in high
[3] Sawa A 2008 Resistive switching in transition metal oxides quality epitaxial Pb(Zr, Ti)O3 films Phys. Rev. B 75 104103
Mater. Today 11 28 [27] Kalinin S V, Jesse S, Rodriguez B J, Eliseev E A,
[4] Borghetti J, Snider G S, Kuekes P J, Yang J J, Stewart D R and Gopalan V and Morozovska A N 2007 Quantitative
Williams R S 2010 Memristive’ switches enable ‘stateful’ determination of tip parameters in piezoresponse force
logic operations via material implication Nature 464 873 microscopy Appl. Phys. Lett. 90 212905
[5] Garcia V, Fusil S, Bouzehouane K, Enouz-Vedrenne S, [28] Maksymovych P et al 2011 Ultrathin limit and dead-layer
Mathur N D, Barthélémy A and Bibes M 2009 Giant tunnel effects in local polarization switching of BiFeO3 Phys. Rev.
electroresistance for non-destructive readout of ferroelectric B at press
states Nature 460 81 [29] Stolichnov I and Tagantsev A 1998 J. Appl. Phys. 84 3216
[6] Gruverman A et al 2009 Tunneling electroresistance effect in [30] Blom P, Wolf R, Cillessen J and Krijn M 1994 Ferroelectric
ferroelectric tunnel junctions at the nanoscale Nano Lett. Schottky diode Phys. Rev. Lett. 73 2107
9 3539 [31] Klaassen K B and Gee S 1996 Electronic Measurement and
[7] Maksymovych P, Jesse S, Yu P, Ramesh R, Baddorf A P and Instrumentation (Cambridge: Cambridge University Press)
Kalinin S V 2009 Polarization control of electron tunneling [32] Koh W and Ang L 2006 Transition of field emission to
into ferroelectric surfaces Science 324 1421 space–charge-limited emission in a nanogap Appl. Phys.
[8] Gajek M, Bibes M, Fusil S, Bouzehouane K, Fontcuberta J, Lett. 89 183107
Barthélémy A and Fert A 2007 Tunnel junctions with [33] Simmons J 1968 Transition from electrode-limited to
multiferroic barriers Nat. Mater. 6 296 bulk-limited conduction processes in metal–insulator–metal
[9] Chambers S A 2009 Epitaxial growth and properties of doped
systems Phys. Rev. 166 912
transition metal and complex oxide films Adv. Mater. 22 219
[34] Fowler R H and Nordheim L 1928 Proc. R. Soc. 119 173
[10] Rabson D A, Jonsson-Akerman B J, Romero A H, Escudero R,
[35] Son J and Stemmer S 2009 Resistive switching and resonant
Leighton C, Kim S and Schuller I K 2001 Pinholes may
mimic tunneling J. Appl. Phys. 89 2786 tunneling in epitaxial perovskite tunnel barriers Phys. Rev. B
[11] Kohlstedt H, Petraru A, Szot K, Rudiger A, Meuffels P, 80 035105
Haselier H, Waser R and Nagarajan V 2008 Method to [36] Maksymovych P, Balke N, Jesse S, Huijben M, Ramesh R,
distinguish ferroelectric from nonferroelectric origin in case Baddorf A P and Kalinin S V 2009 Defect-induced
of resistive switching in ferroelectric capacitors Appl. Phys. asymmetry of local hysteresis loops on BiFeO3 surfaces
Lett. 92 062907 J. Mater. Sci. 44 5095
[12] Seidel J et al 2009 Conduction at domain walls in oxide [37] Jesse S et al 2008 Direct imaging of the spatial and energy
multiferroics Nat. Mater. 8 229 distribution of nucleation centres in ferroelectric materials
[13] Szot K, Speier W, Bihlmayer G and Waser R 2006 Switching Nat. Mater. 7 209
the electrical resistance of individual dislocations in [38] Boerasu I, Pintilie L, Pererira M, Vasilevskiy M I and
single-crystalline SrTiO3 Nat. Mater. 5 312 Gomes M J M 2003 J. Appl. Phys. 93 4776

11

You might also like