See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/358287355

Negative Capacitance from the Inhomogenous Stray Field in a Ferroelectric–

Dielectric Structure

Article  in  Advanced Functional Materials · February 2022

DOI: 10.1002/adfm.202200389

CITATIONS READS

0 40

3 authors, including:

Hyeonwoo Park
Seoul National University
21 PUBLICATIONS   379 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Negative Capacitance View project

domain dynamics View project

All content following this page was uploaded by Hyeonwoo Park on 11 March 2022.

The user has requested enhancement of the downloaded file.

Research Article
Negative Capacitance from the Inhomogenous Stray Field
in a Ferroelectric–Dielectric Structure
Hyeon Woo Park, Minsik Oh, and Cheol Seong Hwang*
The phenomenological Landau–Ginzburg–Devonshire model provides a
fundamental background for an understanding of the peculiar charge–voltage
behavior of ferroelectric (FE) materials. However, the model cannot explain
the polarization behavior of multidomain FE materials. The experimentally
observed negative capacitance (NC) effect, which is interpreted as an emer-
gence of the Landau barrier effect, involves particular conceptual difficulty.
This work provides a new conceptual framework to explain the quasi-static
NC effect based on the energy formula for a stacked dielectric/ferroelectric
(DE/FE) layer structure with the multidomain configuration with arbitrary
shape. The presence of such domain configuration causes the energy–dis-
placement curve of the inhomogenous Helmholtz energy term to have
negative curvature. This is caused by the stray field between the neighboring
domains. The model can be further extended to the DE/FE system with poly-
crystalline FE grains using the advanced phase-field analysis. It is determined
that the NC effect from the stray field is a universal phenomenon. These
models provide quantitative explanations for the previously reported short-
pulse measurement results for various DE/FE material systems, which have
lacked accurate interpretations.
1. Introduction
The phenomenological Landau–Ginzburg–Devonshire (LGD)
model[1–3] has provided a theoretical foundation for the explana-
tion of the peculiar internal energy–polarization (U–P) relation-
ships of ferroelectric (FE) materials. In normal dielectric (DE)
materials, the polarized state involves electrostatic energy; there-
fore, in the absence of an electric field, DE materials remain
depolarized. In contrast, FE materials have a polarized state
as their minimum energy state, the atomistic origin of which
has been ascribed to extreme softening of its optical phonons.
Within the LGD theory framework, the depolarized state (P = 0)
coincides with the local energy maximum, at which the local
curvature has a negative value (the yellow curve in Figure  1a).
H. W. Park, M. Oh, C. S. Hwang
Department of Materials Science and Engineering
and Inter-University Semiconductor Research Center
College of Engineering
Seoul National University
Seoul 08826, Republic of Korea
E-mail: cheolsh@snu.ac.kr
The ORCID identification number(s) for the author(s) of this article
can be found under https://doi.org/10.1002/adfm.202200389.
DOI: 10.1002/adfm.202200389
Adv. Funct. Mater. 2022, 2200389 2200389  (1 of 12)
www.afm-journal.de
This corresponds to a negative permittivity
region when the U–P curve is converted
into a polarization–voltage (P–V) curve.
This peculiar property of FE materials has
generated several novel ideas,[4] including
the negative capacitance (NC) effect[5–11] in
DE/FE stacked structures.
The FE material’s polarized state
induces an additional electrostatic energy
effect when its crystal bound charges
are not compensated, which was not
addressed well in the original LGD theory.
This additional electrostatic energy could
be mitigated in bulk-FE materials by FE
domain formation due to the lateral stray
field effect.[12–14] In FE thin films with
intimate metal contacts (i.e., metal–fer-
roelectric–metal structures), the high car-
rier (electron and hole) concentrations in
the metals compensate efficiently for the
bound charges. Therefore, in both cases,
the FE materials can remain in their polar-
ized state stably.
Salahuddin and Datta argued[5] that
such a polarized state could be converted into a depolarized
state by hindering the charge compensation effect that occurs
when a DE layer intervenes between the FE and metal layers.
In terms of LGD theory, this can be understood as having a
positive curvature for the entire DE/FE stack, thus making
the stack thermodynamically stable (blue curve in Figure  1a).
This indicates that the FE layer’s energy contribution to the
total energy is lower than that of the DE layer (red curve in
Figure  1a) because of its small thickness (< critical thickness).
This effect could be used to boost the stack voltage internally,
which may be useful in improving the performance of field-
effect transistors.[15,16]
However, it has been proven theoretically that the DE/
multidomain-polarized FE stacked system is energetically more
favorable than the DE/depolarized-FE system as long as the
domain wall (DW) energy is not too high.[12–14,17–20] In a multi-
domain FE thin film, all parts of the FE film remain almost
fully polarized. Therefore, the original concept of the NC effect
from the depolarized FE material in the DE/FE stacked struc-
ture becomes difficult to prove. Nevertheless, the introduction
of a superlattice (SL) structure composed of atomically thin
epitaxial DE/FE layers, e.g., SrTiO3/PbTiO3 (STO/PTO), pro-
vided a new avenue to use this particular idea and indeed has
shown enhancement of the static dielectric constant (ε) when
compared with that of a single DE layer.[8,21–23] It was initially
suggested that this represented proof of the depolarized state
© 2022 Wiley-VCH GmbH
www.advancedsciencenews.com www.afm-journal.de

Figure 1.  Free energy density versus displacement field diagrams for the two models. a) Helmholtz energy density versus displacement field diagram
that describes stabilization of the NC in the 1D LGD model. b) Helmholtz energy density versus displacement field diagram in the ISE model. The
green curve represents the homogenous electrostatic energy density of the DE/FE stacked device. The red curve represents the electrostatic energy
density of the DE layer. The yellow curve represents the inhomogenous electrostatic energy of the DE/FE stacked device. The blue curve represents the
total electrostatic energy of the DE/FE stacked device. c) Distributions of the inhomogenous potential (except for the mean potential) inside the DE/
FE stacked structure. Each figure corresponds to the distribution with the relative ratio of domain width ε denoted by the corresponding purple dot in
(b). The arrows in these figures represent the polarization directions in each domain. See the Experimental Section for detailed derivations.

of the FE material within the SL.[8] However, more recent atom-
istic structural investigations[21–23] showed that the thin FE layer
retains the multidomain structure, even below the critical thick-
ness, but the DWs have diffused configurations. In such dif-
fused DWs, which take a significant portion of the material, the
effective ε value should be negative because the DWs are natu-
rally depolarized. However, the SL is of lower relevance to many
applications, particularly microelectronics.
In this context, two recent reports on the charge-boosting
effect in DE/FE stacked structures are notable; the first considers
an 8 or 10 nm thick amorphous Al2O3/150 nm thick epitaxial
BaTiO3 (AO/BTO) system,[24] and the second considers 4 nm
thick amorphous AO/7.7 nm thick polycrystalline (Hf,Zr)O2 (AO/
HZO) and 13.5 nm thick amorphous Ta2O5/11.6 nm thick poly-
crystalline HZO (TO/HZO)[25] systems. The former represents a
conventional DE/FE stack structure where the FE layer thickness
is much greater than the critical thickness, while the latter is a
complementary metal–oxide–semiconductor ­(CMOS)-compatible
DE/FE system. Both works acknowledged the role of the
injected charge (more precisely, the change with FE polarization
switching) involved at the DE/FE interface, which was described
previously by Jiang et  al.[26,27] (called the tunnel-switch mecha-
nism). In this mechanism, the FE bound charge compensation,
with which the intervening thin DE layer was supposed to inter-
fere, could be achieved fluently using the trapped charges at the
DE/FE interface, which could be varied synchronously by polari-
zation switching via the tunneling mechanism. Because the NC
effect, or internal voltage boosting, must suppress such fluent
polarization switching,[28] the two reports above[24,25] used a short
pulse measurement technique to suppress the trapped charge
exchange. The experimental results, which will be discussed in
greater detail later, demonstrated the temporal NC effect, where
charge boosting was observed over the single DE layer. The two
previous works then interpreted their results based on the con-
ventional LGD theory and presented Landau-type potential curves
(W-shaped, as shown in Figure 1a) and S-shaped P–V curves.

Adv. Funct. Mater. 2022, 2200389 2200389  (2 of 12) © 2022 Wiley-VCH GmbH
www.advancedsciencenews.com
However, several critical problems arise when the results are
examined in a more quantitative manner. In the former case
(AO/BTO), fitting the experimental data to the LGD theory only
worked within a limited voltage range; the low-voltage region
showed a severe mismatch. In the latter cases (AO/HZO, TO/
HZO), the extracted potential barrier height was lower than the
theoretical value for the HfO2-based FE material from the den-
sity functional theory (DFT) calculation by one or two orders of
magnitude,[29–31] indicating that the estimated potential barrier
could not be the Landau potential barrier from LGD theory.
In this work, a new conceptual framework was suggested
to solve the problems that arise in explaining the fundamental
mechanism of the observed NC effect from the multidomain
ferroelectric film. Section  2.1 describes the theoretical back-
ground of the analytical model based on the total energy argu-
ments containing the inhomogenous stray-field energy term
at the domain walls. Section 2.2 shows that the suggested ana-
lytical model can explain well the observed NC effect in AO/
BTO system and AO (TO)/HZO system. In Section 2.3, the NC
effect in a practically viable AO/polycrystalline-multidomain-
HZO system is thoroughly investigated by using the phase-field
model. The coincidence with the numerical simulation further
validates the accuracy and generality of the suggested analytical
model. Furthermore, this numerical simulation elucidates the
importance of nucleating domain walls in the P–V behavior,
which is consistently explained in the analytical model.
2. Results
2.1. Theoretical Model of Quasi-Static NC from
Antiparallel Domains
Park et  al. recently reported that a stripe domain configura-
tion for the FE layer in the DE/FE structure could produce a
hysteresis-free S-shaped P–V curve, according to numerical
simulations based on the time-dependent Ginzburg–Landau
formalism.[19] The present work uses this theoretical back-
ground further to unveil the mechanism underlying the
W-shaped potential profile of the multidomain FE system,
in which the material remained in the fully polarized state
throughout the entire voltage-sweep step and thus could not
follow the Landau-type potential profile. In this context, the ear-
lier work by Bratkovsky and Levanyuck (BL)[32,33] was notable.
They derived theoretical backgrounds showing that the FE layer
can have a negative ε value within the DE/FE stacked structure
when the FE layer has an antiparallel domain pattern from the
Gibbs free energy formalism of the system. Nonetheless, the
BL model can only explain the (static) negative ε value for the
case where there are equal portions of the antiparallel domains.
The Gibbs free energy density f of the entire structure can
be divided into three terms as follows[33]
f = f LGD (Ps ) + f DW + f es (1)
where the first term refers to the LGD bulk free energy of the
FE layer. Unlike the 1D LGD model, this value is independent
of the voltage because it assumes that there is a fully developed
spontaneous polarization (Ps) in each domain. The second term
Adv. Funct. Mater. 2022, 2200389 2200389  (3 of 12)
www.afm-journal.de
is the energy density corresponding to the DW surface ten-
sion. The third term represents the electrostatic energy density
generated by the electric fields. Because of the multidomain
structure, this electrostatic energy fes can be subdivided into the
average homogenous term fes,h and the inhomogenous term
fes,ih, depending on the DW location. The inhomogenous fes,ih is
composed of an inhomogenous term fes,δ that is dependent on
δ, which is the relative fraction of the antiparallel domains (i.e.,
the relative difference between the upward domain width a1
a1 − a 2
and the downward domain width a2; δ = ), and the Kittel
a1 + a 2
energy density fes,K, which corresponds to fes,ih in the antipar-
allel-polarized state with a1 = a2 (δ = 0), as follows
f es ,ih = f es ,K + f es ,δ (2)
Although the BL model helped understand the static nega-
tive ε effect near equilibrium ( δ ∼ 0 ), it cannot explain the
dynamic behavior that occurs under varying or pulsing voltage
conditions. The present work extends the theoretical BL model
further to the cases where the system is under a varying elec-
tric potential (or voltage), based on the minimization of the
total energy. In this model, which is called inhomogenous stray
energy (ISE) model, the DW moves following the application
of the driving force to minimize the total energy with a velocity
proportional to its magnitude. This approach is consistent with
the drift motion in both the dynamic formalism of the Landau–
Khalatnikov model[34] and the more generalized phase-field
model.[35] In the ISE model, the variable that represents the
overall switching corresponds to the relative fraction δ and the
average polarization Pavg corresponds to Psδ. Therefore, in the
ISE formalism, the switching kinetics equation is derived in the
form of a drift motion, as follows
∂Pavg ∂ f
= −Γ avg (3)
∂t ∂Pavg
where Γavg is the macroscopic conductivity of the averaged
polarization Pavg. Equation  (3) can then be developed further
into the following form
 C d  U  1  Psδ 
∂δ t

 C + C  t +  C + C  t 
Ps = −Γ avg f  d fb f d fb f  (4)
∂t tf + td  ∞ 8Ps 
 + ∑ ( −1) sin (π nδ )  
n
  n =1 ε 0kn2Ttf Dn  
where ε0 is the vacuum permittivity, tf (td) is the FE (DE) thick-
ness, U is the applied voltage, T is the domain period, kn is
the wave number, and Dn is the coupled effective dielectric
constant. Cfb and Cd are the background capacitance along
the vertical direction of the FE layer and the DE layer linear
capacitance, respectively. A detailed derivation and related dis-
cussion can be found in Section S1 (Supporting Information).
This analytical formula is similar to the previously introduced
NC model,[19] but the derivation method of the equations
in this model is different. While Equation  (4) was directly
derived from the local field applied on the DW in the prior
work, the ISE model indicates that they can be formulated by
a more generalized approach based on the total free energy.
© 2022 Wiley-VCH GmbH
www.advancedsciencenews.com
In Equation (4), the antiparallel domain period T is a variable
that is not predetermined, which is one of the crucial varia-
bles for the theoretical modeling, although the other variables
were predetermined for the given material system. In the ISE
model, the domain period can be determined by the general-
ized Kittel’s law. A detailed discussion of the determination
of T can be found in Section S2 and Figure S1 (Supporting
Information).
It should be noted that Kittel’s law finds the optimum T
based on the balance between the electrostatic energy due to
the uncompensated polarization charge and total DW energy.
Therefore, it requires information on the DW energy of the
material under discussion. The DW energy has been reported
for BTO, rendering applying the ISE model to the AO/BTO
system feasible. However, the available DW energy data of FE
HfO2 (or HZO) are highly anisotropic, i.e., dependent on the
specific DW type.[36] It is even negative for one DW type, and
other DW types have one to two orders of magnitude higher
than BTO, making applying the ISE model to the AO/HZO and
TO/HZO cases challenging. However, this was not a serious
problem because the general HZO film has a polycrystalline
structure with a grain size smaller than the equilibrium T. This
will be discussed in detail in Section 2.2.
Interestingly, the ISE model resulted in an energy curve
with negative curvature near P  ∼ 0, (more precisely, the dis-
placement field D ∼ 0, because this model treats the entire DE/
FE system as a combined entity; therefore, the P of the only
FE layer can hardly be separated) as indicated by the yellow
curve in Figure 1b. For ease of comparison with Figure 1a, the
Helmholtz energy h was plotted, although f was introduced
(methods for the conversion between these quantities via the
Legendre transformation). Among the energy contributions,
hih has negative curvature, whereas hh (green curve) and hDE
(red curve) have normal positive curvatures. Therefore, htotal
(blue curve) could have a minimum near D  ∼ 0. Figure  1c,
which shows the inhomogenous potential distribution within
the device for different δ (or D), hints at the negative curva-
ture of hih. As the figure shows, the overall magnitude of the
inhomogenous potential (apart from the mean internal poten-
tial) decreases as the polarization is poled into one direction.
This is because the stray field intensity decreases as the width
of the domains oriented according to the poling direction
increases.
In the 1D LGD model, the energies of the DE and FE layers
were treated independently, and the NC effect was attributed
to the FE layer’s double-well potential. In contrast, the ISE
model shows that the total energy generated by the DE/FE
stacked structure is a coupled property of both layers, and
thus, cannot be divided into independent energies for each
layer. In this case, the NC phenomenon originates from the
inhomogenous stray field distributed within both layers. This
indicates the possibility that the recently observed double-
well potential was not caused by the LGD-type anisotropic
energy of the FE layer. Instead, it could be originated from
the inhomogenous potential in the antiparallel domain struc-
ture. Another significant merit of the ISE model is that it can
quantitatively treat the effects of the various parameters of
the DE and FE layers, including their thicknesses, ε values,
and Ps values.
Adv. Funct. Mater. 2022, 2200389 2200389  (4 of 12)
www.afm-journal.de
2.2. Revisiting the Experimental ‘S’-Curves in
DE/FE Stacked Structures
Based on this new finding, the previous works by Kim et al.[24]
for AO/BTO and Hoffmann et  al.[25] for AO/HZO and TO/
HZO are reinterpreted. The material parameters in Table S1
(Supporting Information) were used to perform short pulse sim-
ulations for the AO/BTO using the ISE model. The spontaneous
polarization and the compensation charge at the interface were
assumed to be 0.175 and −0.170 C m−2, respectively, indicating
that the interfacial charge initially compensated almost fully
for the FE polarization charge. Short pulse simulations were
conducted by assuming the same circuit and voltage source as
the actual measurements. The results and schematics for the
polarization states at the first and last data points are shown
in Figure  2a. The model fitted the experimental results accu-
rately over the entire tested voltage region. This demonstrates
that the charge boosting over the single AO layer was due to
the reversible motion of the DW and not caused by the homog-
enous polarization switching behavior of the FE layer according
to LGD theory. The P–E relationship of the FE layer immedi-
ately before termination of the pulses can be extracted from
Figure 2a and is plotted in Figure 2b. A detailed derivation of the
P–E values is addressed in the Experimental section. Surpris-
ingly, the experimental S-shaped P–E curves that were extracted
from two different samples with different DE thicknesses had
almost identical shapes that can be reproduced precisely using
the ISE model. The independence of P–E curve for the different
DE thicknesses was theoretically driven by defining an effective
dielectric constant, which corresponds to the slope in P–E curve
at P  = 0 (Section S3 in Supporting Information). Figure S2 in
the Supporting Information shows that the effective dielectric
constant and the shape of the P–E curve do not vary with DE
thicknesses, which is consistent with the extracted result from
the experiment. Moreover, U–P curves can be obtained from
the experiment as shown in Figure 2c, and they were explained
well by the ISE model as well. Section S4 (Supporting Informa-
tion) identifies that these free energy curves are not related to
the intrinsic LGD barrier but the inhomogenous electrostatic
energy. Therefore, these curves must be different from the DFT
result, as shown in Figure S4a in the Supporting Information.
The results in Figure 2 show that the ISE model, represented
by Equation (4), which assumes the simple antiparallel domain
structure, can fully simulate the NC effect in the AO/BTO
system where the BTO film is epitaxial on the SuRuO3/DyScO3
substrate. The epitaxial BTO film had a compressive strain
(ε = −0.01) due to the lattice mismatch with the substrate, which
favored the formation of the antiparallel DW.[24] On the other
hand, to apply this model for the DE/polycrystalline FE struc-
ture, it should be guaranteed that a similar NC effect can be
reproduced in the more complex domains. This was proven for
the AO/HZO and TO/HZO structures, as shown in Section 2.3.
The results for the AO/HZO and TO/HZO structures are
summarized in Figure  3. The spontaneous polarization was
assumed to be 0.14 C m−2 for both samples, and the compen-
sation charges were assumed to be −0.135 and −0.132 C m−2
for AO/HZO and TO/HZO samples, respectively. As discussed
before, the equilibrium domain width (aeq) values for these sam-
ples were predetermined to be 5–7 nm due to its considerable
© 2022 Wiley-VCH GmbH
www.advancedsciencenews.com www.afm-journal.de

Figure 2.  Analysis of Kim et al.’s work[24] (AO/BTO system) using the ISE model. a) Discharging amounts of AO/BTO capacitors for various pulse
heights and schematics of the polarization states at two data points. Dotted lines represent the calculated charge density–voltage curves of the AO
single layer. b) Extracted FE layer electric field versus polarization curves for the AO/BTO capacitors. c) Extracted polarization–free energy curves for
the AO/BTO capacitors. In all subgraphs, the data for the 8 nm AO/150 nm BTO capacitor are plotted in blue, and the data for the 10 nm AO/150 nm
BTO capacitor are plotted in red. Circles represent the experimental results, and solid lines represent the simulation results from the model.

DW energy and small grain size, and these domain width
values are small enough for the antiparallel domains to be
present within a single grain.[37,38] The other material para­­­­
meters shown in Section S5 and Table S2 (Supporting Informa-
tion) were used to perform the fitting. Note that the original
data were plotted until charge injection was induced by the
high voltage. The results show that the model again produces
excellent predictions, even for the polycrystalline FE layer, in
which the domain configuration cannot be single and homog-
enous. Also, the model can predict all the electrical behavior,
although the DE and FE layers have different thicknesses and ε
values. As shown in Figure 3c, the S-curves of the two samples
are quite different, even when the same FE material was used.
This difference in the S-curves cannot be explained using the
LGD model, which regards the S-curve as an intrinsic property
of the FE material. In the ISE model, however, this critical dif-
ference was anticipated based on the different FE layer thick-
nesses. Figure S2b in the Supporting Information shows how
the S-curves vary with FE thicknesses, which is consistent with
the experimental result. Figure S3 in the Supporting Infor-
mation shows the dependence of the effective dielectric con-
stant and P–E curve on the dielectric constants of both layers.
Because the effective dielectric constant is almost constant for
different dielectric constants of the DE layer (Figure S3c, Sup-
porting Information), it can be concluded that the difference
in the S-curves in Figure  3c resulted from the different FE
thicknesses, not from the different types of DE materials. The
energy barrier shown in Figure 3d is also dependent on the FE
layer thickness, which cannot be explained by the LGD model
but can only be explained by the ISE model. Additionally, as
illustrated in Figure S4b in the Supporting Information, the
value of the barrier height generally being much smaller than
the theoretical LGD barrier height is no longer problematic
because it does not represent the intrinsic LGD barrier. Instead,
it represents the inhomogenous electrostatic energy for DW
migration with the antiparallel domain configuration.
Additional proof for accurate application of the model to
another AO/HZO system, reported by Kim et  al.,[39] can be

Adv. Funct. Mater. 2022, 2200389 2200389  (5 of 12) © 2022 Wiley-VCH GmbH
www.advancedsciencenews.com www.afm-journal.de

Figure 3.  Analysis of Hoffmann et al.’s work[25] using the ISE model. a) Discharging amounts of AO/HZO capacitor for various pulse heights. b) Dis-
charging amounts of TO/HZO capacitor for various pulse heights. In (a) and (b), dotted lines indicate the calculated charge density–voltage curves
for the DE (AO or TO) single layer. c) Extracted FE layer electric field versus polarization curves for the two capacitors. d) Extracted polarization versus
free energy curves for the two capacitors. In all subgraphs, circles represent experimental results, and solid lines represent simulation results from
the model.

found in Section S6 and Figure S6 (Supporting Information),
where the HZO film has a Hf:Zr ratio of 3:7 and severely
imprinted behavior. The AO layers had thicknesses of 5.5, 7.3,
and 10.1 nm. Even for these materials, the ISE model accurately
predicts the material behavior and provides valuable informa-
tion about the NC properties.
It should be noted that the discussions in this section
assumed the simple antiparallel domain configuration as for
the epitaxial BTO case, which may be impractical for the poly-
crystalline-multidomain HZO system. However, the following
section justifies applying the ISE model even for this case.
2.3. Numerical Model Analysis on the NC Effect in
AO/HZO System
In polycrystalline ferroelectric HZO thin films, domains with
random polar axes exist due to independent crystal orienta-
tion in each grain. Also, nonferroelectric phases, such as
nonpolar monoclinic (P21/c) and tetragonal (P42/nmc) phases,
may coexist with the polar orthorhombic phase (Pca21) in the
HZO thin film.[41] In this regard, the validity of the analytical
model based on the simple antiparallel domain structure in the
out-of-plane direction might be limited. Therefore, to examine
the validity of this claim, a more realistic numerical model for
AO/polycrystalline-multidomain-HZO (AO/p-m-HZO) thin
films was developed, and their charge–voltage responses were
numerically simulated. For this purpose, the previous phase-
field model[19] was modified to incorporate the multigrain, mul-
tiphase, and multidomain configurations by taking the crystal
orientation and the coefficients of the free energy function of
the phases into consideration. In this modified simulation, the
crystal angle and the first coefficient of the LGD polynomials
are the additional order parameters that distinguish each order
(or phase) from others.[42] A detailed description of this model
can be found in Section S7 (Supporting Information). The NC
effect in the AO/p-m-HZO bilayer structure can be system-
atically investigated by adding each order parameter. In this

Adv. Funct. Mater. 2022, 2200389 2200389  (6 of 12) © 2022 Wiley-VCH GmbH
www.advancedsciencenews.com
Figure 4.  Simulated NC effect in the phase-field model of the single-crystal ferroelectric HZO thin film with slanted domains (AO/HZO system).
a) Polarization–electric field (P–E) curves of the FE layer in the local coordinate of the crystal orientation. b) Magnified version of the P–E curves.
c) Snapshot images of the domain distribution in each step denoted by the circles on the P–E curves. White arrows in the images indicate the direction
of the FE polarization. The local conductivity, Γloc was assigned as 1.14 S m−1 in the model so that the scale of the domain wall velocity (≈0.15 m s−1)
is consistent with the experiment.[40]
section, two model cases were dealt with; One is slanted multi-
domain in a single-crystal HZO (Figure 4), and the other is ran-
domly oriented multiple domains in polycrystalline HZO with
columnar grain shape (Figure 5).
Figure  4a shows that the NC effect can also be found in
the slanted domain structure, in which the polar axis of the
single-crystal ferroelectric HZO thin film has an angle of 60°
with the out-of-plane direction in a 5 nm  AO/10 nm  HZO
structure. The P–E curve (the sky-blue curve in Figure  4a)
following the LGD formula was extracted from the intrinsic
double-well potential of HZO obtained from the DFT calcula-
tion.[36,43,44] Other material parameters for the numerical sim-
ulation are listed in Table S3 (Supporting Information). DFT
studies of HZO revealed that the intrinsic coercive field (Ec)
representing homogenous switching of HZO (> ≈10 MV cm−1)
is larger than the experimentally observed Ec (≈1 MV cm−1)
by approximately one order of magnitude.[36,43] The purple
dotted curve in Figure 4a (and a magnified version, Figure 4b)
shows the P–E curve of the FE layer when a voltage sweep
of ± 20  V  is applied from the initial depoled state. The P–E
curves were calculated by averaging the local polarization and
electric field along the crystal orientation. Figure  4c shows
the snapshots of the (slanted) domain configuration at the
corresponding points in Figure  4b. In this case, the initially
(mostly) pointing-upward-polarized multidomain structure
(state (1)) is being polarized pointing-downward by applying
a voltage of +20 V to the top part of the figure. This operation
transits the multidomain (MD) state to the single domain (SD)
state (state (5)). Once the SD state is reached, the P–E curve
follows the homogenous switching path (sky-blue curve).
Note that at state (4) of Figure 4c, the stripe domain structure
disappears through a vertical DW motion. At this point, an
abrupt and hysteretic NC effect (blue dot in Figure  4b), devi-
ating from the original switching path, appears. However,
when the sweep voltage range is decreased from ± 20  to ±
6.72  V,  the (slanted) stripe domain structure remains. This
renders the hysteresis-free NC operation (red dotted curve) by
the reversible lateral domain wall motion possible, as shown
in Figure  4b,c (state (1) to (3), and vice versa). Such an NC
Adv. Funct. Mater. 2022, 2200389
www.afm-journal.de
effect appears due to the stray field between domains regard-
less of the orientation of the domains. As a result, the quasi-
static NC effect exists in this AO/HZO thin film but within a
limited sweep range (± 6.72 V).
An analytical NC model which explained such a circum-
stance could also be derived through the energy framework
of the ISE model. This extended model mathematically veri-
fied that the same principle holds even for a domain structure
having an arbitrary direction. A detailed description of this ana-
lytical model and the quantitative analysis on the tilted angle
are addressed in Section S8 (Supporting Information).
Next, the possible NC effect in the 5 nm  AO/10 nm  p-m-
HZO thin films was investigated by adding a spatial order
parameter for crystal orientation to this model. Figure  5a
shows the hysteretic S-shaped P–E curves of a polycrystal-
line HZO thin film with five columnar grains (each width
of 20  nm). Figure  5b shows the domain distribution during
sweeping the voltage (Vpol) from −20 to +20  V. As shown
in state (1) of Figure  5b, most domains are initially poled
pointing upward by a highly negative voltage of −20 V except a
few reversed domains in the outermost grains. As the voltage
increases toward the positive value, the domains with effec-
tively pointing downward polarization (blue region) expand
while the effectively pointing upward polarization (red region)
shrinks. Nonetheless, the corresponding P–E curve still
shows a positive slope, suggesting the positive capacitance
between the states (1) and (2), even though the DW motion
occurs from the residual domains in outer grains. As shown
in Figure 5c, this is because alternating bound charges are not
generated at the DE/FE interface from these residual domains.
Thus, the influence of the stray field is insignificant at state
(1) in Figure  5c. As the positive bias increases, a reversed
domain is nucleated, denoted by the green circle in state
(2) of Figure  5b. At this state, notably, the capacitance of the
FE HZO film is converted from positive to negative. At this
state, a strong stray field is generated between the domains
grown from the nucleus and the adjacent opposite domains.
Hence, a distinct NC effect can be attributed to the inho-
mogenous electrostatic energy. Note that the nucleus of this
2200389  (7 of 12) © 2022 Wiley-VCH GmbH
www.advancedsciencenews.com www.afm-journal.de

Figure 5.  Simulated NC effect in the phase-field model of the polycrystalline-multidomain ferroelectric HZO thin film (AO/p-m-HZO layer). a) Dis-
placement field–electric field curve of the polycrystalline HZO layer in the bilayer structure during the voltage sweep from −20 to +20 V. b) Snapshot
images of the domain distributions. Each step is denoted by the yellow square with the number on the blue square curves in (a). c) Snapshot images
of the electrical potential. Note that the lateral stray field distribution is denoted by the magenta arrows. d) Displacement field–electric field curves
of the polycrystalline HZO layer in the bilayer structure during the voltage sweep from −15 to +15 V. e) Snapshot images of the domain distributions.
Each step is denoted by the sky-blue circle with the number on the red dotted curve in (d). f) Snapshot images of the electrical potential. g) Profile
of the energy density of the lateral electric field (Ex) in each step of (d). The figures below the distribution of the energy density represent the domain
distribution in each step. h) Averaged electric energy density of the lateral stray field (Ex) for the displacement field (D). The dashed line corresponds
to the energy density with the opposite sweep direction.

reversed domain is formed at the tail to tail charged domain
wall (CDW), which has been observed in polycrystalline HZO
thin films.[36,45] Since the opposing bound charges in the CDW
are not compensated, a depolarization field can be enhanced
in this region. Therefore, a nucleus of the reversed domain
is mostly formed at the CDWs. Comparing the P–E curve
(navy square curve) in Figure 5a with the P–E curve based on
the LGD theory (sky-blue curve) in Figure  4a shows that the
polarization switching occurs at a field much smaller than
the intrinsic Ec through a nucleation and growth mechanism
in this p-m-HZO. When such a new domain is nucleated, a
strong stray field is produced (states (2)–(4) in Figure  5c) at
the DE/FE interface, which induces the inhomogenous field
energy, and thus, the NC effect.

Adv. Funct. Mater. 2022, 2200389 2200389  (8 of 12) © 2022 Wiley-VCH GmbH
www.advancedsciencenews.com
Nonetheless, in this case, the S-curve occurs with hysteresis
because the nucleation field (≈  ±3.74 MV cm−1) is required in
both sweep directions once the FE film reached a uniformly
polarized state. Interestingly, this hysteresis is significantly
diminished as the Vpol decreases to ± 15 V, as shown by the green
diamond curve in Figure 5d. During the simulation, the intrinsic
Ec is fixed at ≈12 MV cm−1. This result indicates that the hys-
teresis of the S-curve is determined by not only the (apparent)
Ec but also Vpol, which is consistent with general experiments.
Figure 5e shows the change of domain distribution during this
voltage sweep. State (1) retains a much larger portion of the
reversed domains than state (1) in Figure 5b. In particular, most
residual domains are found near the grain boundaries because
they have a relatively large depolarization field. This result also
justifies the presumption that the domain period is determined
by the grain size of the p-m-HZO film. As the bias voltage
increases into the positive direction up to 15 V, switching in each
grain proceeds by the DW motion of the residual 2–3 domains
without involving the nucleation and growth of new domains.
These results are also consistent with the recent ab initio calcula-
tion results, which showed that fast DW motion is possible even
in HZO via the specific topological path of DW motion.[46] Nev-
ertheless, there is an NC region between the states (2) and (4).
Because there is no involvement of reverse domain nucleation
and growth, the switching process, in this case, can be described
by Equation  (4). Equation  (4) assumes a drift motion in which
the domain wall velocity is proportional to the electric field.[47]
Therefore, the voltage of the FE layer can be expressed as follows
dQ f  ∞ 8Ps  Q f  
Vf = ρDW + ∑ ( −1) sin  π n
n
 (5)
ε 0knTDn  Q s  
2
dt  n =1
where A is the area of the device, ρDW  = (tf  + td)/ΓavgA is the
macroscopic damping resistance of the DW motion, Qs  = PsA
is the saturated charge, and Qf  = PsδA  = Qsδ is the ferroelec-
tric polarization charge. As shown in the first term of Equa-
tion (5), the damping effect resulting from the DW motion with
finite mobility induces an additional voltage drop, which can be
expressed as a series resistance (ρDW) in Figure S12 in Section S9
(Supporting Information). The second term in Equation (5) rep-
resents the nonlinear change of the voltage as the Qf changes
through DW motion, which can be expressed as a nonlinear
capacitor (Cinh(Qf )) in the equivalent circuit of Figure S12a in
the Supporting Information. Cinh(Qf ) is responsible for the
S-curve in the P–E curve. The parallel capacitor (Cfb) is attrib-
uted to the background dielectric constant of the FE layer. The
red dotted curve in Figure S12b in the Supporting Information
shows the D–V curve of the equivalent circuit. The hysteresis
of the D–V curve is attributed to the first term in Equation (5),
meaning that the hysteretic green diamond curve in Figure 5d
originates from the damping effect by the movement of the
DW, not by the reversed domain nucleation. If the fraction of
multidomain changes without involving the damping effect
 dQ f  tive bound charges can also be located at the interface between
∼ 0  , the D-internal voltage (Vint) relationship appears as
 dt  the polar and the nonpolar grains, generating a vertical stray
the blue S-curve (Figure S12b, Supporting Information) from
the nonlinear capacitor. Therefore, Equation  (5) indicates that
the FE layer shows the hysteresis-free NC behavior when the
Adv. Funct. Mater. 2022, 2200389 2200389  (9 of 12)
www.afm-journal.de
damping effect by DW motion of the multidomain in the DE/
FE bilayer can be minimized. This circumstance can be defined
as a “quasi-static” NC in the multidomain structure.
It is possible to reach a quasi-static state  f ∼ 0  for a
dQ
 dt 
given voltage condition by increasing the mobility of the DW.
Reducing the voltage sweep rate also provides a similar effect.
The red dotted curve in Figure 5d shows that the hysteresis by
the damping effect can be substantially reduced as the conduc-
tivity of local polarization (Γloc), which is proportional to DW
mobility, increases by ten times. This indicates the possibility
of a hysteresis-free NC effect even in the p-m-HZO. The solid
red curve in Figure 5d shows that this quasi-static S-curve can
be safely approximated by the analytical ISE model assuming
the simple antiparallel domain configuration, presented in Sec-
tion  2.2. Again, this consistency stresses that the NC effect is
essentially due to the consistent change of the inhomogenous
stray field energy during the switching process. The following
explains this NC effect more precisely.
Figure 5f shows the variation in the electric potential during
the switching process denoted by the sky-blue circles on the
quasi-static S-curve in Figure 5d. The stray field (indicated by the
magenta arrows) is concentrated at the DE/FE interface, where
alternative FE bound charges are exposed. While all the stray
fields are not necessarily along the x-axis (Ex), it is clear that Ex
constitutes a large portion of the stray field. Therefore, to visu-
alize the change of inhomogenous energy due to the stray field,
1
the electrostatic energy density  h = ε 0 ε f E x2  associated with
 2 
the Ex in each state was displayed three-dimensionally. Figure 5g
shows that the field energy density increases rapidly when the
one polarization portion of the domain is much smaller than the
other (states (1) to (2) in Figure  5g). However, as it approaches
the depoled state (the two portions of the opposite domains were
similar), the rate of increase of the stray field energy decreases
and eventually reaches the maximum value (States (2) to (3) in
Figure 5g). Subsequently, the stray field energy decreases again
as the switching progresses further to the SD state with the
opposite polarity. Consequently, the change in inhomogenous
electrostatic energy manifests as a negative curvature concerning
the displacement field near the depoled state, as in Figure  5h.
This result is also consistent with the energy curve of the simple
model in Figure 1b. It should be noted that the NC effect is not
necessarily attributed to only the appearance of the stray field
but how the field energy evolves during the switching. Since the
underlying physics must be identical, the NC effect of a com-
plex domain structure can also be approximated by adjusting the
parameters of the simple analytical model.
Furthermore, the NC effect due to the stray field in the DE/
FE bilayer, where the FE layer contains a nonpolar phase, can
be addressed in essentially the same manner. Of course, the
average NC effect must be decayed due to the parallel posi-
tive capacitance of the nonpolar phase. However, as shown in
Figure S8 in Section S7 (Supporting Information), the alterna-
field. Because of this reason, the NC effect due to the inho-
mogenous electrostatic energy can be expected not only at the
DE/FE interface but also at the different phase boundaries.
© 2022 Wiley-VCH GmbH
www.advancedsciencenews.com www.afm-journal.de

3. Conclusion In the LGD model, the Helmholtz free energy of the stacked structure
could be expressed as follows
In conclusion, this study developed the analytical model for the
htot = hFE + hDE (9)
quasi-static NC phenomenon in the DE/FE stacked structure
with the antiparallel domain configuration. It also provided
quantitative explanations for the several experimental results 1 
hFE = HFE = a1Pf2 + a11Pf4 + a111Pf6 +
(D − Pf )2  t
(10)
and revealed that the origin of the observed NC effects was not A 2ε 0ε c  f
 
the intrinsic LGD barrier of the FE layer, but the inhomogenous
electrostatic energy caused by the stray field between the neigh-
D2
hDE =
boring domains. Moreover, the advanced phase-field model anal- 2C d (11)
ysis showed that the NC effect from the stray field is a universal
The electric field in the FE layer for which Pf can be stabilized
phenomenon. The analytical model could further be extended ∂P
(i.e., ∂tf = 0 ) could be obtained in the form of Equation  (12), and the
to the DE/FE system with polycrystalline and multidomain FE
film. It could also be adopted to identify the possible NC effect relationship between D and Pf could then be obtained by inserting
Equation (12) into Equation (8)
from the thin films composed of polar and nonpolar phases.
Finally, the experimental circumstances under which this new E f = 2a1Pf + 4a11Pf3 + 6a111Pf5 (12)
ISE model cannot be applied are described briefly. This model
assumes that if the charge injection occurs, the interfacial trap
charge follows the FE layer’s polarization switching behavior D = ε 0ε c ( 2a1Pf + 4a11Pf3 + 6a111Pf5 ) + Pf (13)
instantaneously. Therefore, if there is a time lag between the Insertion of Equation (13) into Equations (10) and (11) allowed every
changes, a hysteretic P–V curve would be achieved that would Helmholtz free energy term and displacement field to be expressed
complicate the application of the model. Additionally, if the FE in terms of Pf. Therefore, using Pf as a parameter, the relationships
layer is fully and homogenously switched into one direction, between the displacement field D and the Helmholtz free energy
then switching into the reverse direction requires nucleation terms htot, hFE, and hDE could be plotted as shown in Figure  1a. The
of the reverse domains, which requires nucleation time (or a following values for BTO were used for effective display of the results:
a1 = − 2.26 × 107 V ⋅ m C−1, a11 = 1.37 × 108 V ⋅ m5 C−2, a111 = 2.76 × 109 V ⋅ m9 C−5,
delay). This was not taken into account in the model since the εc = 50, εd = 8.9, tf = 100 nm, and td = 1 nm.
actual FE layer almost always retains nuclei of opposite Ps, even In the model, the Helmholtz free energy could be expressed as follows
in the almost fully switched configuration, which then serves as using Equation (S1.2) and Equation (S1.3) (Supporting Information)
nuclei for reverse domain growth. In the short pulse measure-
ments, the assumptions of unvaried interfacial trap charges and 1 1  1 1 
h = H = FLGD (Ps ) + FDW + ∫ dAσϕ + ∑e aϕ a  (14)
retained reverse domain nuclei in even saturation-level switched A A

2 F /D 2
a 
FE films were fulfilled well, and the model thus applies without
The only difference between the expressions for the Helmholtz
any conceptual difficulty. The ISE model fundamentally assumes
free energy and the Gibbs free energy was the sign of the last term in
a hard domain (abrupt domain wall). For the material with the Equation (14). This difference led to the change in the sign of the second
dominant soft domain (gradual domain wall), the depoled state term in fes,h (Equation (S1.5) in Supporting Information), while the other
of the domain walls may constitute the main reason for the NC Helmholtz free energy terms, which corresponded to the LGD energy,
effect, not the stray field-induced inhomogenous energy. the DW surface energy, and the inhomogenous electrostatic energy,
respectively, were the same as the corresponding terms for the Gibbs
free energy. Only hes,h, and hes,δ were considered here because the other
terms were independent of δ and thus did not affect the capacitance in
4. Experimental Section this case
Derivation of the Helmholtz Free Energy Density–Displacement Field htot = hh + hih (15)
Curves in Terms of Energy: To extract the capacitance of the structure from
the free energy density–displacement field curve, the Gibbs free energy
must be transformed into the Helmholtz free energy using the Legendre ε 0ε dU + λPsδ td  t f  1 ε ε U 2 − λPsδ t f U
hh = hes,h = λPsδ + ε ε 0 c (16)
transformation as follows[48] ε 0ε c td + ε 0ε dt f  2  2 0 d ε 0ε c td + ε 0ε dt f
−1
 ∂ 2 h 
4( λPs )2 a ∞ ( −1) (1− cos (π nδ ))
n
C =  2 (6)
 ∂D  hih = hes,δ = ∑
ε 0π n=1 n
3 3 Dn
(17)

The displacement field D could be obtained from the applied voltage

1 1 
h = H =  F + ∑e aϕ a  (7) U and the equation for the displacement continuity at the DE/FE
A A  interface, as follows
a

where C is the differential capacitance of the structure, D is the U = −E dtd − E f t f (18)

displacement field in the structure, H is the Helmholtz free energy of
the structure, and h is the Helmholtz free energy of the structure per
unit area. The displacement field D could be obtained using the equation ε 0ε dE d = ε 0ε cE f + λPsδ (19)
for the displacement continuity at the DE/FE interface, as follows

D = ε 0ε dE d = ε 0ε cE f + Pf (8)
D = ε 0ε dE d =
Cd
(
C d + C fb
λPsδ − C fbU ) (20)

Adv. Funct. Mater. 2022, 2200389 2200389  (10 of 12) © 2022 Wiley-VCH GmbH
www.advancedsciencenews.com www.afm-journal.de

The final step here was to determine the value of U at which the value initial state, the equations then became the same as those reported in
of δ can be stabilized  i.e., ∂∂δt = 0 using Equation (S1.21) (Supporting the previous work.[25]
Information) The double-well potential of the stacked structure could be obtained
from the S-curve as follows

 Cd  U  1  λPsδ  ∞ 8λPs  1 Pf
 C d + C fb  t f +  C d + C fb  t f + ∑ ( −1) ε k 2Tt D sin(π nδ )  = 0 (21) f (Pf ) = f 0 +
L ∫0 f
n
V dP (29)
 n=1 0 n f n 
where f is the free energy per unit volume of the stacked structure.

∞
λPsδ C d + C fb  
U=− −  ∑ ( −1) n 8λ2Ps sin(π nδ ) (22)
Cd Cd 
 n=1 ε 0 k n TDn  Supporting Information
Supporting Information is available from the Wiley Online Library or
Insertion of Equation  (22) into Equations  (16) and (20) allowed
from the author.
all the Helmholtz free energy terms and the displacement field to be
expressed using δ. Therefore, the relationship between the displacement
field D and the Helmholtz free energy terms htot, hh, and hih could
be acquired, as shown in Figure  1b, by using δ as a parameter. hDE Acknowledgements
in Figure  1b was plotted using Equation  (11) while assuming that the
single DE layer had the same displacement field. The following values H.W.P. and M.O. contributed equally to this work. This study
were used for effective display of the results: a1 = − 2.26 × 107 V ⋅ m C−1, was supported by the National Research Foundation of Korea
a11 = 1.37 × 108 V ⋅ m5 C−2, a111 = 2.76 × 109 V ⋅ m9 C−5, εa = 50, εc = 50, (2020R1A3B2079882). The authors thank David MacDonald, MSc, from
εd = 8.9, tf = 25 nm, td = 5 nm, Ps = 0.2 C m−2, and λ = 0.1. The value of Edanz Group (https://en-author-services.edanz.com/) for editing a draft
εa was intentionally set to be equal to that of εc for the sake of schematic of this manuscript.
clarity. Note that the values of the displacement field and the Helmholtz
energy in Figure 1b were quite small because a considerable amount of
charge compensation (λ = 0.1) was assumed for effective display.
Derivation of S-Curve and Double-Well Potential from the Short Pulse Conflict of Interest
Measurements:[25] From the pulse height-discharging amount graph
The authors declare no conflict of interest.
shown in Figure  2a, the voltage across the FE layer and the apparent
polarization at the end of each short pulse could be extracted as follows.
Before application of a short pulse, the voltage across the stacked
structure was zero
Data Availability Statement
Vf + Vd = 0 (23) Research data are not shared.
where Vf  (Vd) is the voltage across the FE (DE) layer. From the equation
for the displacement continuity at the DE/FE interface

σ comp + C dVd = C fbVf − Ps (24) Keywords

where − Ps is the initial spontaneous polarization of the FE layer. In domain wall, ferroelectric devices, inhomogenous energy, negative
this section, a positive sign for the polarization referred to the direction capacitance, stray field
in which the polarization was switched when a positive voltage was
applied, which was opposite to the way used in the main results section Received: January 11, 2022
above. Therefore, the amount of charge in the electrode near the DE Published online:
layer, which was denoted by Qd, was as follows

Qd = −
Cd
(
P + σ comp
C d + C fb s ) (25)
[1] L. D. Landau, Zh. Eksp. Teor. Fiz. 1937, 7, 19.
[2] V. L. Ginzburg, J. Exp. Theor. Phys. 1945, 15, 739.
Because charge injection did not occur during the short pulse, the
charge that flowed into the sample was stored in the electrodes. Qd at [3] A. F. Devonshire, London, Edinburg Dublin Philos. Mag. J. Sci. 1949,
the end of the short pulse could be determined as follows, given that the 40, 1040.
charging amounts were the same as the discharging amount Qdis [4] Y. J.  Kim, M. H.  Park, W.  Jeon, H. J.  Kim, T.  Moon, Y. H.  Lee,
K. D. Kim, S. D. Hyun, C. S. Hwang, J. Appl. Phys. 2015, 118, 224105.
[5] S. Salahuddin, S. Datta, Nano Lett. 2008, 8, 405.
Q d = Q dis −
Cd
( P + σ comp
C d + C fb s ) (26)
[6] A. I.  Khan, D.  Bhowmik, P.  Yu, S. J.  Kim, X.  Pan, R.  Ramesh,
S. Salahuddin, Appl. Phys. Lett. 2011, 99, 161106.
Therefore, the voltage across the FE layer (Vf) and the apparent
[7] D. J. R.  Appleby, N. K.  Ponon, K. S. K.  Kwa, B.  Zou, P. K.  Petrov,
polarization (Pf) at the end of the short pulse could be obtained as follows
T. Wang, N. M. Alford, A. O’Neill, Nano Lett. 2014, 14, 3864.
Ps + σ comp [8] W.  Gao, A.  Khan, X.  Marti, C.  Nelson, C.  Serrao, J.  Ravichandran,
Qd Q
Vf = U − = U − dis + R. Ramesh, S. Salahuddin, Nano Lett. 2014, 14, 5814.
Cd Cd C d + C fb (27)
[9] A. I.  Khan, K.  Chatterjee, B.  Wang, S.  Drapcho, L.  You, C.  Serrao,
S. R. Bakaul, R. Ramesh, S. Salahuddin, Nat. Mater. 2015, 14, 182.
Pf ≈ Q d + σ comp = Q dis + σ comp −
Cd
( P + σ comp
C d + C fb s ) (28) [10] C. S. Hsu, S. C. Chang, D. E. Nikonov, I. A. Young, A. Naeemi, Phys.
Rev. Appl. 2021, 15, 034048.
where U is the pulse height. If the spontaneous polarization was [11] Z. Liu, H. Jiang, B. Ordway, T. P. Ma, IEEE Electron Device Lett. 2020,
compensated completely by the interfacial charge that existed in the 41, 1492.

Adv. Funct. Mater. 2022, 2200389 2200389  (11 of 12) © 2022 Wiley-VCH GmbH
www.advancedsciencenews.com
[12] C. Kittel, Phys. Rev. 1946, 70, 965.
[13] E. V. Chenskii, V. V. Tarasenko, Zh. Eksp. Teor. Fiz. 1982, 83, 1089.
[14] L. D. Landau, E. Lifshitz, Phys. Z. Sowjet. 1935, 8, 153.
[15] C. H. Cheng, A. Chin, IEEE Electron Device Lett. 2014, 35, 274.
[16] M. Si, C. J. Su, C. Jiang, N. J. Conrad, H. Zhou, K. D. Maize, G. Qiu,
C. T. Wu, A. Shakouri, M. A. Alam, P. D. Ye, Nat. Nanotechnol. 2018,
13, 24.
[17] S. Kasamatsu, S. Watanabe, C. S. Hwang, S. Han, Adv. Mater. 2016,
28, 335.
[18] M.  Hoffmann, M.  Pešić, S.  Slesazeck, U.  Schroeder, T.  Mikolajick,
Nanoscale 2018, 10, 10891.
[19] H. W.  Park, J.  Roh, Y. B.  Lee, C. S.  Hwang, Adv. Mater. 2019, 31,
1805266.
[20] T.  Rollo, F.  Blanchini, G.  Giordano, R.  Specogna, D.  Esseni,
Nanoscale 2020, 12, 6121.
[21] P.  Zubko, J. C.  Wojdel, M.  Hadjimichael, S.  Fernandez-Pena,
A. Sené, I. Luk’yanchuk, J. M. Triscone, J. Íñiguez, Nature 2016, 534,
524.
[22] A. K. Yadav, K. X. Nguyen, Z. Hong, P. García-Fernández, P. Aguado-
Puente, C. T.  Nelson, S.  Das, B.  Prasad, D.  Kwon, S.  Cheema,
A. I. Khan, C. Hu, J. Íñiguez, J. Junquera, L. Q. Chen, D. A. Muller,
R. Ramesh, S. Salahuddin, Nature 2019, 565, 468.
[23] S.  Das, Z.  Hong, V. A.  Stoica, M. A. P.  Gonçalves, Y. T.  Shao,
E. Parsonnet, E. J. Marksz, S. Saremi, M. R. McCarter, A. Reynoso,
C. J.  Long, A. M.  Hagerstrom, D.  Meyers, V.  Ravi, B.  Prasad,
H.  Zhou, Z.  Zhang, H.  Wen, F.  Gómez-Ortiz, P.  García-Fernández,
J.  Bokor, J.  Íñiguez, J. W.  Freeland, N. D.  Orloff, J.  Junquera,
L. Q.  Chen, S.  Salahuddin, D. A.  Muller, L. W.  Martin, R.  Ramesh,
Nat. Mater. 2021, 20, 194.
[24] Y. J.  Kim, H.  Yamada, T.  Moon, Y. J.  Kwon, C. H.  An, H. J.  Kim,
K. D.  Kim, Y. H.  Lee, S. D.  Hyun, M. H.  Park, C. S.  Hwang, Nano
Lett. 2016, 16, 4375.
[25] M.  Hoffmann, F. P. G.  Fengler, M.  Herzig, T.  Mittmann, B.  Max,
U.  Schroeder, R.  Negrea, P.  Lucian, S.  Slesazeck, T.  Mikolajick,
Nature 2019, 565, 464.
[26] Q.  Jiang, H. J.  Lee, G. H.  Kim, C. S.  Hwang, Adv. Mater. 2009, 21,
2870.
[27] H. W.  Park, S. D.  Hyun, I. S.  Lee, S. H.  Lee, Y. B.  Lee, M.  Oh,
B. Y. Kim, S. G. Ryoo, C. S. Hwang, Nanoscale 2021, 13, 2556.
Adv. Funct. Mater. 2022, 2200389 2200389  (12 of 12)
View publication stats
www.afm-journal.de
[28] Y. J. Kim, M. H. Park, Y. H. Lee, H. J. Kim, W. Jeon, T. Moon, K. D. Kim,
D. S. Jeong, H. Yamada, C. S. Hwang, Sci. Rep. 2016, 6, 19039.
[29] T. D.  Huan, V.  Sharma, G. A.  Rossetti, R.  Ramprasad, Phys. Rev. B
2014, 90, 064111.
[30] R. Materlik, C. Kunneth, A. Kersch, J. Appl. Phys. 2015, 117, 134109.
[31] T.  Maeda, B.  Magyari-Köpe, Y.  Nishi, IEEE International Memory
Workshop (IMW), Monterey, CA 2017, https://doi.org/10.1109/
IMW.2017.7939087.
[32] A. M. Bratkovsky, A. P. Levanyuk, AIP Conf. Proc. 2000, 535, 218.
[33] A. M. Bratkovsky, A. P. Levanyuk, Phys. Rev. B 2001, 63, 2.
[34] L. D. Landau, I. M. Khalatnikov, Dokl. Akad. Nauk SSSR 1954, 96, 469.
[35] N.  Provatas, K.  Elder, Phase-Field Methods in Materials Science and
Engineering, Wiley-VCH, Weinheim, Germany 2010.
[36] W. Ding, Y. Zhang, L. Tao, Q. Yang, Y. Zhou, Acta Mater. 2020, 196,
556.
[37] H. J.  Kim, M. H.  Park, Y. J.  Kim, Y. H.  Lee, W.  Jeon, T.  Gwon,
T. Moon, K. D. Kim, C. S. Hwang, Appl. Phys. Lett. 2014, 105, 3.
[38] M. H.  Park, Y. H.  Lee, H. J.  Kim, T.  Schenk, W.  Lee, K. D.  Kim,
F. P. G. Fengler, T. Mikolajick, U. Schroeder, C. S. Hwang, Nanoscale
2017, 9, 9973.
[39] K. D.  Kim, Y. J.  Kim, M. H.  Park, H. W.  Park, Y. J.  Kwon, Y. B.  Lee,
H. J. Kim, T. Moon, Y. H. Lee, S. D. Hyun, B. S. Kim, C. S. Hwang,
Adv. Funct. Mater. 2019, 29, 1808228.
[40] P.  Buragohain, C.  Richter, T.  Schenk, H.  Lu, T.  Mikolajick,
U. Schroeder, A. Gruverman, Appl. Phys. Lett. 2018, 112, 222901.
[41] M. H.  Park, Y. H.  Lee, H. J.  Kim, Y. J.  Kim, T.  Moon, K. D.  Kim,
S. D. Hyun, C. S. Hwang, ACS Appl. Mater. Interfaces 2018, 10, 42666.
[42] S. Choudhury, Y. L. Li, C. Krill, L. Q. Chen, Acta Mater. 2007, 55, 1415.
[43] Y. Qi, K. M. Rabe, Phys. Rev. B 2020, 102, 214108.
[44] Y. W.  Chen, S. T.  Fan, C. W.  Liu, J. Phys. D: Appl. Phys. 2021, 54,
085304.
[45] X.  Li, C.  Li, Z.  Xu, Y.  Li, Y.  Yang, H.  Hu, Z.  Jiang, J.  Wang, J.  Ren,
C. Zheng, C. Lu, Z. Wen, Phys. Status Solidi RRL 2021, 15, 2000481.
[46] D. H.  Choe, S.  Kim, T.  Moon, S.  Jo, H.  Bae, S. G.  Nam, Y. S.  Lee,
J. Heo, Mater. Today 2021, 15, 8.
[47] J. Y. Jo, S. M. Yang, T. H. Kim, H. N. Lee, J. G. Yoon, S. Park, Y. Jo,
M. H. Jung, T. W. Noh, Phys. Rev. Lett. 2009, 102, 045701.
[48] M. E. Lines, A. M. Glass, Principles and Applications of Ferroelectrics
and Related Materials, Oxford University Press, Oxford 2010.
© 2022 Wiley-VCH GmbH