ArticlePDF Available

Synthesis and Characterization of Conducting Polyaniline Nanostructured Thin Films for Solar Cell Applications


Abstract and Figures

Optical-quality transparent, conducting polyaniline (PANI) thin films are suitable candidates for efficient counter electrodes for high-performance solar cells. In the first part of this work, the synthesis of highly uniform and homogenous nanostructured PANI films is reported. The film properties were assessed via scanning electron microscopy, atomic force microscopy, optical profilometry, spectrophotometry, and conductimetry. Simultaneous modeling, optimization and physical characterization of the PANI nanostructured films have not received much attention in the literature. Hence, in the second part, a multi-objective optimization approach with three objectives, namely minimum film thickness, maximum transparency, and maximum conductivity, was performed based on artificial neural network models with a novel k-fold cross-validation technique. The developed models can accurately predict the film characteristics in a wide range of design variables with most residuals remarkably less than 1.0%. Furthermore, after optimization, conductivity was increased three-fold (~ 2.2 × 10−1 S/cm) at a good level of transparency (~ 55%), which suit solar cell applications.
Content may be subject to copyright.
Synthesis and Characterization of Conducting Polyaniline
Nanostructured Thin Films for Solar Cell Applications
1.—Department of Chemical Engineering, Hamedan University of Technology, P.O. Box 65155-
579, Hamedan, Iran. 2.—e-mail:
Optical-quality transparent, conducting polyaniline (PANI) thin films are
suitable candidates for efficient counter electrodes for high-performance solar
cells. In the first part of this work, the synthesis of highly uniform and
homogenous nanostructured PANI films is reported. The film properties were
assessed via scanning electron microscopy, atomic force microscopy, optical
profilometry, spectrophotometry, and conductimetry. Simultaneous modeling,
optimization and physical characterization of the PANI nanostructured films
have not received much attention in the literature. Hence, in the second part,
a multi-objective optimization approach with three objectives, namely mini-
mum film thickness, maximum transparency, and maximum conductivity,
was performed based on artificial neural network models with a novel k-fold
cross-validation technique. The developed models can accurately predict the
film characteristics in a wide range of design variables with most residuals
remarkably less than 1.0%. Furthermore, after optimization, conductivity was
increased three-fold (2.2 910
S/cm) at a good level of transparency
(55%), which suit solar cell applications.
Electrically conductive polymer thin films are
extensively used in numerous types of solar cells,
such as perovskite,
architectures. The catalytic activity
and transparency of conducting polymers are also
important as they are supposed to substitute costly
platinum-based and opaque carbon-based
Conducting polymers have typically good electri-
cal and mechanical properties, and can be produced
continuously as flexible films by chemical coating or
electrochemical techniques. Polyaniline (PANI) is
one of the most interesting conductive polymers
with an electrical conductivity between 1 S/cm and
100 S/cm, and is used as a hole injection layer in
thin films because of its good environmental stabil-
ity and high structural resistance.
The perfor-
mance of PANI thin films in solar cells is affected by
their thickness, optical transparency and stability,
and electrical conductivity.
There has been a breakthrough in the modeling
and analysis of the properties of nanostructured
materials by first-principle models.
These mod-
els intrinsically give rise to complex and nonlinear
relationships between design variables and target
characteristics of the conductive polymers. In com-
puter simulations, these models require relatively
cumbersome and time-consuming runs. Hence, a
simple and straightforward methodology is still
required for including, but not limited to, optimiza-
tion purposes, where numerous solutions must be
checked for the best outcome. In this regard, a feed-
forward backpropagation artificial neural network
(ANN) model
has been developed in this work to
study the interrelationships of the design variables
and desired characteristics, as mentioned above.
The performance of an ANN model greatly
depends on its learning process, which is posed as
a dedicated optimization technique for supervised
networks. A typical approach when the number of
samples is small is k-fold cross-validation,
is modified here with a heuristic iterative process to
improve the network accuracy and generalization
(Received April 9, 2020; accepted August 26, 2020)
2020 The Minerals, Metals & Materials Society
Downloaded from
Since a minimum film thickness, maximum opti-
cal transparency (and optical stability), and maxi-
mum conductivity are simultaneously desired, a
multi-objective optimization approach is required.
Furthermore, as the solutions tend not to be opti-
mum at a single point, they appear as a so-called
Pareto front.
Therefore, the multi-objective ver-
sion of a genetic algorithm (NSGA-II)
is utilized
for the optimization task.
In this study, the experimental work will be
presented as accounts of the variations in film
thickness, optical transparency (and stability), and
conductivity of nanostructured PANI thin films
with respect to the design variables. The results
have been used for training the ANN models. After
demonstrating the ability of the networks to simu-
late the experimental results, they have been used
for the optimization task with the aims of minimiz-
ing film thickness and simultaneously maximizing
transparency and conductivity.
Aniline (99.5%, ACS reagent grade), ammonium
peroxydisulfate, hydrochloric acid (37%, ACS
reagent grade), m-cresol, and ethanol were all
purchased from Sigma-Aldrich.
Synthesis of PANI
First, the polymerization of aniline (0.55 mM)
was carried out by mixing ammonium peroxydisul-
fate as the oxidant and hydrochloric acid as the
catalyst at room temperature for 12 h. The initial
concentrations of ammonium peroxydisulfate and
hydrochloric acid were 0.25 M and 0.5 M, respec-
tively. The PANI precipitate was initially washed
with ethanol and then with deionized water several
times. The precipitate was then heated under
vacuum condition at 328 K. Finally, m-cresol was
added to emeraldine-base PANI powder and stirred
for 2 h at room temperature to achieve a 15 wt.%
homogenous solution.
Preparation of Thin Films
Ultrasonically cleaned indium tin oxide (ITO)
glass was used for the preparation of PANI nanos-
tructured thin films. The thin films were coated on
the substrate through immersion in the polymeric
solution using the dip-coating method.
withdrawal velocity of the substrate varied from
2.0 910
m/s to 2.4 910
m/s. The procedure was
repeated several times with intermittent heat treat-
ment under vacuum condition at drying tempera-
tures of 298–373 K for 30 min. In order to study the
effects of the thickness on the transparency and
conductivity of the thin films, PANI thin films with
different thicknesses were prepared in the experi-
ments. The samples were cooled to room tempera-
ture and stored for further analysis.
Characterization of Thin Films
The morphology and structure of the PANI
nanostructured thin films were examined by scan-
ning electron microscopy (SEM; Cam Scan MV2300
microscope). The surface morphology and height
profile of the films were studied by atomic force
microscopy (AFM; DME DS-95-50). Data analysis of
the AFM images was performed by the image
analysis software package, Scanning Probe Image
Processor (Image Metrology). The height profile and
thickness of the thin films were measured by a
ZeScope optical profilometer. The optical trans-
parency of the samples was determined using a
UV–Vis spectrophotometer (Hitachi U-3140) in the
wavelength range of 300–900 nm, and the conduc-
tivity of the thin films was measured by the four-
point probe technique.
Artificial Neural Network
ANN is a general-purpose modeling tool, which
can be used for various applications, including static
and dynamic modeling, clustering, and pattern
ANN is useful especially when mod-
eling with fundamental governing equations is
costly, time-consuming, or both. Regarding opti-
mization problems, since the number of function
calls (i.e., employing the system model) is enormous,
even with today’s powerful computers, simpler
models based on ANNs are favored. In the field of
nanostructured materials, ANNs have also been
popular and successful for the prediction of material
A popular performance evaluation function for
ANNs is the mean squared error (MSE). The
training algorithm aims at minimizing MSE, but
the main difference between a usual optimization
problem and supervised training of an ANN model
is the risk of overfitting too few actual data by using
a too-complex model. In such cases, the ANN loses
its generalization ability if not carefully trained.
One way of dealing with such problems is early
in a typical supervised training method,
the experimental data is divided into three subsets
and provided to the ANN, from which the larger
subset is used for training, the second subset for
validation, and the third subset for testing. The
validation error is monitored for the occurrence of
overfitting while the training is in progress. The
training stops as soon as the prediction error starts
to grow on the validation set. This process enhances
the generalization ability of the network. On the
other hand, the testing subset is used to check if the
validation dataset is properly randomly distributed.
This gives an extra level of confidence in the
generalization ability of the designed network, since
poor distribution of data in the validation set
renders the entire process inefficient.
Medi, Bahramian, and Nazari
Downloaded from
Despite the outstanding performance of ANNs in
numerous problems, it is not easy to achieve
acceptable accuracy, due to the multi-dimensional
search space for optimal weights and biases in
addition to poor data division. In general, the
solution starts with assigning a set of random initial
values and a random distribution of data in the
abovementioned subsets (i.e., training, validation,
and testing). However, commonly, not much effort is
made to study the effects of changing initial values
on the final solution, while, in our experience, a
good initial guess is as important as a good training
procedure for many networks.
In this study, we have improved the common
approach by recalling both new random values and
new random distributions many times (i.e., in the
order of 1000) until the performance error drops to
an acceptable level. We observe that this method
performs very well on almost all experimental data.
This process is schematically depicted in Fig. S-1 of
the Supplementary Materials, where the main
training algorithm is encompassed by the heuristic
iterations proposed in this work as the outer loop.
It is worthy of attention that the proposed method
resembles the well-known k-fold cross–validation,
in that the data are shuffled many times. However,
in the standard method, typically, the weights and
biases are not randomly re-initialized. Hence, the
method may lose the chance of exploring unseen
regions of the search space even with multiple
shuffling of the dataset. In our approach, however,
the search space is also shuffled in terms of initial
Optimization Problem Formulation
In general, a constrained multi-objective opti-
mization problem is defined as:
Minimize f1xðÞ;f2xðÞ;...;fpxðÞ ð1Þ
Subject to:
lb xub
where f
(x) are the objective
functions, xis the n-dimensional vector of the
design variables, g(x) is the vector of inequality
constraints with m components, and. similarly, h(x)
is a k-dimensional vector of equality constraints.
There are also nlower lb and upper ub bounds on
the set of design variables, x.
We have proposed three objective functions for a
minimization problem as follows. We assume the
first objective function as the normalized film
thickness, which must be minimized:
f1xðÞ¼ h
hnom is an arbitrary value close to the average
expected value for film thickness.
On the other hand, since a maximum value for
transparency is desired, the negative value of
transparency is used in the normalized form as
the second objective function:
100 ð4Þ
And for conductivity, we have again chosen the
negative of the normalized conductivity (r)to
comply with the minimization problem:
f3xðÞ¼ r
We have taken the withdrawal velocity (V), the
drying temperature (T), and the number of depos-
ited layers (N) as the three design variables. It is
noteworthy that these design variables are identi-
cally the same inputs as the network models. The
ranges of design variables are given in Table I,
noting that these ranges act as both lower and
upper bounds in the optimization problem formula-
tion, which coincides with the experimental
As mentioned earlier, we have used the non-
dominated sorting genetic algorithm (NSGA-II),
which efficiently employs population ranking and
crowding distance measures for the selection oper-
ator. Further details can be found elsewhere.
In the following paragraphs, we elaborate on
important morphological characteristics of the
PANI thin films with the help of multiple imaging
and profiling techniques, while optical and electrical
properties are also investigated by their dedicated
methods. The complete set of the experimental data
is given in Table S-I of the Supplementary Materi-
als. We will then briefly present the modeling
results and conclude this section with the optimiza-
tion outcomes.
Experimental Results
Effects of Withdrawal Velocity
Figure 1shows the SEM images of PANI nanos-
tructured thin films prepared by two deposited
layers at two withdrawal velocity levels of
2.0 910
(a, b) and 2.4 910
m/s (c, d) heated
at T= 323 K. As can be seen in Fig. 1b, the surface
morphology of the PANI thin films is relatively
uniform, but some irregular nanoparticles with the
dimensions of 21.00–35.10 nm can be observed for
the lower withdrawal velocity, while, for the higher
withdrawal velocity, these values rise to 24.61–
40.37 nm (Fig. 1d), which suggests an increase in
the formation of agglomerate structures on the
PANI films. In fact, larger agglomerates with sizes
Synthesis and Characterization of Conducting Polyaniline Nanostructured Thin Films for
Solar Cell Applications
Downloaded from
close to 200 nm can be seen, which spread randomly
on the ITO glass substrate with a relatively small
surface coverage. This indicates that PANI has
limited adhesion to the substrate compared to the
rate that PANI particles attach to their own
From the fluid dynamics point of view, the
increase in the size of the particles (hence average
film thickness) with the withdrawal velocity can be
explained by the well-known Landau–Levich theory
for Newtonian fluids,
and also its extensions for
non-Newtonian systems.
On the other hand, a
mass transfer gradient may exist under the effects
of physicochemical adsorption. Surface reactions
are controlled by limited mass transfer coefficients.
Hence, a higher withdrawal velocity results in a
higher film mass transfer coefficient and, conse-
quently, higher adsorption/reaction rates.
Figure 2shows the AFM images of PANI nanos-
tructured thin films with two deposited layers,
which were dried at 298 K at two withdrawal
velocity levels of 2.0 910
(a, b) and 2.4 910
m/s (c, d). The surface of the films appears to be
crack-free over the studied section of 5 95lm. The
height analysis given in Fig. 2indicates that the
mean thicknesses of the films are 79.3 nm and
88.4 nm, respectively, for the abovementioned
velocities. As can be seen, the film thickness
increases with the increase in the withdrawal
velocity. It is noteworthy that the mean thickness
measurements in Table S-I of the Supplementary
Materials are from the AFM analysis, and that they
exhibit a similar trend.
As shown in Fig. 2, the surface of the films
contains bright spots, indicating the presence of
peaks as assemblies of the PANI nanostructures.
The maximum peak heights relative to the deepest
groove are 86.2 nm and 97.1 nm, respectively, for
the velocity levels mentioned earlier. The dark holes
seen in the AFM images arise from non-
Table I. The ranges of design variables for optimization
Variable Minimum Maximum
Withdrawal velocity (910
m/s) 2.0 24
Drying temperature (K) 298 373
Number of deposited layers 2 4
Fig. 1. SEM images of PANI nanostructured thin films prepared at T= 323 K by two deposited layers at two withdrawal velocity levels of (a, b)
2.0 910
and (c, d) 2.4 910
m/s.(a) and (c) are low-magnification and (b) and (d) are high-magnification images.
Medi, Bahramian, and Nazari
Downloaded from
homogeneities on the surface of the thin films,
which increase with the increase in the withdrawal
Effects of Drying Temperature
The optical properties of PANI thin films are
greatly influenced by their thickness and morphol-
ogy. The surface roughness is a key factor for
obtaining optimal transparency and optical stabil-
Figure 3shows the AFM images of four-layer
(n= 4) PANI nanostructured thin films prepared
under a withdrawal velocity of 4.0 910
m/s and
drying temperatures of 298 K (a, b) and 373 K (c, d).
Non-homogeneity on the film surface is observable
for the latter case, which is associated with surface
roughness. The existence of voids suggests solvent
entrapment in the PANI film during drying at
temperatures below its glass transition, which is
reported as 378 K.
A low evaporation rate of m-
cresol trapped in the film structure led to the
formation of small holes in the film surface (Fig. 3a
and b). As can be seen in Fig. 3c and d, a high
evaporation rate at T= 373 K leads to localized
non-homogeneities and coalescing of the PANI
nanoparticles, thereby yielding an overall less-ho-
mogeneous surface morphology.
Optical Properties
Figure 4presents the optical properties of four-
layer PANI nanostructured thin films prepared
under withdrawal velocity of 4.0 910
According to Fig. 4a, the wavelength spectra can
be divided into two regions: an exponential region
(wavelengths below 450 nm) and a zone of optical
stability and maximum transparency associated
with minor fluctuations (wavelength range 450–
900 nm). Moreover, transparency increases with the
increase in drying temperature up to T= 348 K.
The sudden loss of transparency at T= 348 K is
attributed to the coalescing and chain folding of the
PANI structures,
which is confirmed by the
AFM results (Fig. 3c and d). It must be noted that
this temperature is close to the glass transition
temperature of the PANI films (i.e., T= 378 K
Figure 4b shows the calculated values of the
refractive index n(k) versus wavelengths at different
drying temperatures. The mean refractive indices in
the wavelength range 450–900 nm for the PANI
Fig. 2. AFM images of two-layer PANI nanostructured thin films dried at 298 K at two withdrawal velocity levels of (a, b) 2.0 910
and (c, d)
2.4 910
m/s. (a) and (c) are two-dimensional, and (b) and (d) are three-dimensional images. Values in brackets and next to brackets are
average and maximum film thickness, respectively.
Synthesis and Characterization of Conducting Polyaniline Nanostructured Thin Films for
Solar Cell Applications
Downloaded from
films prepared at 298 K and 348 K are found to be
about 2.11 and 1.89, respectively.
As an example, the high-resolution transparency
of the PANI film prepared at 348 K is given in
Fig. 4c. If two envelopes are drawn through the
maxima TmaxðkÞand minima Tmin ðkÞof the oscillat-
ing transmittance, then the refractive index can be
calculated following the methods of Manifacier
et al.
and Swanepoel,
as described elsewhere.
The value of ahtwas calculated as a function of
the photon energy and is plotted in Fig. 4d. The
linear extrapolation in the transition region using
the Tauc’s plot
on the photon energy-axis gives
the band-gap energy of the PANI thin film. The
value of band-gap energy was found to increase
from 2.53 eV to 2.70 eV with increasing the drying
temperature from 298 K to 348 K, suggesting a
decrease in conductivity.
The decrease in the
band-gap energy may be due to a reduction in the
disorder of the system (e.g., better conjugation
Moreover, it is evident that the value of band-gap
energy is inversely related to the refractive
while the refractive index is a function
of transparency.
In general, conductivity is controlled by two
factors: the number of carrier electrons or holes
and the carrier mobility.
According to Table S-I of
the Supplementary Materials, the trend between
conductivity and temperature does not appear to
follow the monotonic Arrhenius behavior. The loss
in conductivity after 350 K is most likely due to
simple dehydration of the polymer and evaporation
of the free dopant, HCl. It has also been reported
that the loss in conductivity of thermally treated
PANI can be attributed to changes in the morphol-
ogy, cross-linking, and possible structural rear-
rangements or side reactions, such as chemical
interactions between the dopant or solvent with the
polymer that lead to the reduction of conductivity.
On the other hand, looking at Figs. S-3 and S-5 of
the Supplementary Materials, which are full-scale
simulations by the trained ANN models, an inter-
esting pattern in conductivity is observed; firstly,
the maximum conductivity falls mostly at the
regions of lower film thicknesses for any number
of deposited layers, which is in accordance with the
common effect of film thickness on the conductivity
Fig. 3. AFM images of four-layer PANI nanostructured thin films prepared under a withdrawal velocity level of 4.0 910
m/s and drying
temperatures of (a, b) 298 K and (c, d). 373 K. (a) and (c) are two-dimensional, and (b) and (d) are three-dimensional images. Values in brackets
and next to brackets are average and maximum film thickness, respectively.
Medi, Bahramian, and Nazari
Downloaded from
for thin films.
Secondly, the optimal conductivity
rises when moving from two to three deposited
layers, and then falls for n= 4 number of deposited
layers. This can be explained by the fact that PANI
may not have fully covered the underlying ITO
substrate surface. Therefore, an increase in the
number of deposited layers first enhances surface
coverage, which favors conductivity, and then it
drops due to excessive film thickness.
Effects of the Number of Deposited Layers
Figure 5shows the height profile of the PANI film
surfaces obtained by the ZeScope optical profilome-
ter for two (a) and four (b) numbers of deposited
layers dried at 298 K and prepared at a withdrawal
velocity of 2.4 910
m/s. The height profile anal-
ysis indicates that the mean film thickness was
increased from 90.0 nm to 140.6 nm when the
number of deposited layers is increased from two
to four, respectively. Moreover, it can be found that
the film with lower numbers of deposited layers
shows relatively uniform surface morphology and
smooth surface, while the film prepared with higher
numbers of deposited layers shows more surface
non-homogeneities. It is important to note that
these data are in good agreement with the AFM
results, as given in Table S-I of the Supplementary
The maximum conductivity was measured at
7.2 910
S/cm for n= 3 number of deposited
layers (see Supplementary Materials). It is difficult
to draw a general conclusion for the relationship
between conductivity and the number of deposited
layers directly from the experimental results. How-
ever, simulation results give a better insight into
understanding the conductivity variations versus
the input variables, as mentioned and discussed in
‘‘Conductivity’’ section.
Fig. 4. Optical properties of four-layer PANI nanostructured thin films prepared under a withdrawal velocity of 4.0 910
m/s: (a) transparency
and (b) refractive index of films prepared at different drying temperatures, (c) high-resolution transparency of PANI film prepared at 348 K, and
(d) band-gap estimation of films prepared at 298 K and 348 K.
Synthesis and Characterization of Conducting Polyaniline Nanostructured Thin Films for
Solar Cell Applications
Downloaded from
Modeling Results
We employed three independent feed-forward
ANNs for predicting layer thickness, transparency,
and conductivity. A multi-layer network with five
neurons in a single hidden layer was designed and
trained for each output. The inputs to each network
are withdrawal velocity, drying temperature, and
number of deposited layers. The type of transfer
function is log-sig for all the neurons in the hidden
layer. These settings were chosen based on some
trial and error, since the proposed training method
is quite efficient in terms of accuracy, prediction,
and computational time so no thorough optimiza-
tion is required on the network topology. The
complete set of the experimental and ANN results
are given in the Supplementary Materials.
The experimental dataset is divided into
60:20:20% subsets for training, validation, and
testing, respectively. The training algorithm takes
less than 10 min to complete the full k-fold cross-
validation procedure on a laptop with an Intel Core
i5-2430 M (2.40 GHz) processor and 3 Gb memory.
The MATLAB Artificial Neural Network Toolbox is
used for network design, training, and simulation.
The results of training the ANN for film thickness
prediction is shown in Fig. 6a. It is apparent that
the regression is quite good on all the datasets,
while the randomly selected data are broadly dis-
tributed for training, validation, and testing over
the experimental domain. Furthermore, the residu-
als are small with the errors mostly less than 1.0%,
while no particular trend or bias is observed in the
Figure 6b shows the results of the transparency
prediction via training the ANN. The linear regres-
sion between the network outputs and the experi-
mental data with small intercepts for all the three
sets of training, validation, and testing is a good
indication of both the accuracy and generalization
ability of the ANN model under consideration.
There is only one exception in the training set
(run number 4), which gives the maximum devia-
tion of 4.0% error as the residual, which may be an
outlier. Apart from this point, no particular trend or
bias can be observed in the residuals.
As for the conductivity prediction, the training
results are shown in Fig. 6c. It is apparent that,
with the help of the modified k-fold cross–validation
method, the accuracy and generalization ability of
the network are good, although the mean residual
value is larger compared to the other outputs, and
with one outlier (9.52%). This is actually expected
since the accuracy of the four-point probe is rela-
tively low due to local film thickness variations. It
must be mentioned that the results for runs 4, 11,
and 17 were not collected because of the limited
number of conductivity tests possible for the
authors. Nonetheless, no specific trend in the resid-
uals can be observed here, like the two other
Optimization Results
We have used the NSGA-II algorithm available in
MATLAB Global Optimization Toolbox, which pre-
sents results in order of their ranks. The subset of
the results in the objectives space with rank = 1, is
the Pareto front. The corresponding design vari-
ables comprise the Pareto set.
Thanks to the simple yet efficient ANN models,
the multi-objective optimization takes less than
3 min to complete on the same machine used for
the network training, provided that parallel pro-
cessing is utilized. The optimization solution param-
eters are summarized in Table S-II of the
Supplementary Materials. For a full description of
the parameters see the algorithm
The evolution of the resulting Pareto front is
given in Fig. 7. The absence of significant variations
in the generations shown is an indication of
approaching convergence, although the Pareto
spread is satisfactorily wide, suggesting a good
global search.
The analysis of the Pareto front given in Fig. 7
reveals that the results are scattered on two distinct
regions of conductivity values, as they are simulta-
neously drawn toward lower values of film thickness
and higher values of transparency. These results
confirm the pattern observed in the simulation
studies (results not shown here for brevity) that no
single point can be reported as the optimal solution
for this problem. Hence, a Pareto optimal solution
emerges as the result of the optimization. It must be
emphasized that the range of conductivity values
obtained in the optimization study is far beyond the
values obtained in the experiments, while trans-
parency is kept at an acceptable range (above 55%).
The corresponding Pareto set (design variables) is
given in Fig. S-2 of the Supplementary Materials.
Fig. 5. Height profile of the PANI film surfaces prepared by two (a)
and four (b) numbers of deposited layers at drying temperature of
298 K and withdrawal velocity of 2.4 910
Medi, Bahramian, and Nazari
Downloaded from
The absence of n= 4 for the number of layers in the
results suggests that it is unnecessary to proceed to
the fourth deposited layer to achieve higher con-
ductivity. In other words, with regulation of the
withdrawal velocity and drying temperature, we
can readily achieve a sufficiently high conductivity
with the extra benefit of reduced production time
and cost for fewer deposited layers.
Optically efficient polyaniline (PANI) nanostruc-
tured thin films were prepared by the sol–gel dip-
coating technique. The effects of design variables,
that is, withdrawal velocity, drying temperature,
and the number of PANI deposited layers on the
film morphology/film thickness, optical trans-
parency, and conductivity of the prepared films
were studied experimentally and then analytically
using artificial neural network (ANN) models. The
simulation results were produced based on the data
retrieved from AFM, UV–Vis spectrophotometry,
and four-point probe technique. The experimental
results can be further used for studying the rela-
tionships between band-gap energy and the refrac-
tive index, and between the refractive index with
transparency and film thickness to increase the
conversion efficiency of solar cells made from PANI
thin films.
The experimental results showed that the with-
drawal velocity played a major role in the surface
morphology and thickness of the prepared films,
while the drying temperature and the number of
deposited layers played key roles in the surface
homogeneity and thickness of the PANI films. It is
important to note that we have obtained morpho-
logically more homogeneous films compared to other
studies (e.g.,
), which directly affects the optical
and electrical properties of various types of solar
cells based on the PANI thin films. The combination
of the broadened spectrum of the PANI thin films in
the range of 450–900 nm and the high band-gap
energy (2.7 eV) contributes to the considerably
higher optical stability and conversion efficiency of
solar cells.
Fig. 6. ANN prediction results for (a) film thickness, (b) transparency, and (c) conductivity.
Synthesis and Characterization of Conducting Polyaniline Nanostructured Thin Films for
Solar Cell Applications
Downloaded from
Finding the best operating point is difficult for any
production or experimental scheme with a multi-
dimensional search space and/or nonlinear func-
tional relationships. A ubiquitous design of experi-
ments (DOE) method is not competent on extreme
nonlinearities. In addition, they are not developed by
default for multi-objective optimization purposes
where Pareto optimality is expected. In this regard,
the current approach of using versatile ANN models
in combination with the multi-objective genetic algo-
rithm is potentially superior and should outperform
DOE methods in search and optimization.
Thanks to the efficient ANN learning algorithm,
there was excellent agreement between the results
of the ANN models and the experimental data. It
must be emphasized that, despite the fundamental
complexities in modeling polymers compared to
inorganic thin films, we have obtained a good level
of accuracy in data regression and prediction for
solar cell applications.
Also, using the superior
optimization technique proposed in the present
study, several promising combinations of the design
variables were found for which at least a three-fold
increase in conductivity of the nanostructured PANI
film (2.2 910
S/cm) at a good level of trans-
parency (55%) is expected. Since no extrapolation
was carried out, the Pareto set is within the range of
experimental conditions.
The obtained Pareto front serves as a multi-
paradigm basis on which several different types of
solar cells can be optimized with regard to the
proposed objective functions. In particular, using
the proposed optimization framework, one can
optimally decide on film thickness for thin-films,
transparency for bifacials,
and conductivity for
heterojunction solar cells,
The authors would like to thank the Hamedan
University of Technology for financially supporting
this work through Grant No. 18-96-1-3.
The authors declare that they have no conflict of
Fig. 7. Pareto fronts for the minimization of film thickness and maximization of transparency and conductivity. The plots are for (a) 20th, (b) 40th,
and (c) 48th (c) (convergence) generations.
Medi, Bahramian, and Nazari
Downloaded from
The online version of this article (
10.1007/s11837-020-04361-8) contains supplemen-
tary material, which is available to authorized
1. J. Wei, F. Huang, S. Wang, L. Zhou, Y. Xin, P. Jin, Z. Cai, Z.
Yin, Q. Pang, and J.Z. Zhang, Mater. Res. Bull. 106, 35
2. A. Bahramian and D. Vashaee, Sol. Energy Mater. Sol. Cells
143, 284 (2015).
3. J. Wu, Y. Li, Q. Tang, G. Yue, J. Lin, M. Huang, and L.
Meng, Sci. Rep. 4, 4028 (2014).
4. J. Yuan, J. Gu, G. Shi, J. Sun, H.-Q. Wang, and W. Ma, Sci.
Rep. 6, 26459 (2016).
5. M. Asyraf, M. Anwar, L.M. Sheng, and M.K. Danquah, JOM
69, 2515 (2017).
6. S. Bhadra, D. Khastgir, N.K. Singha, and J.H. Lee, Prog.
Polym. Sci. 34, 783 (2009).
7. K.M. Molapo, P.M. Ndangili, R.F. Ajayi, G. Mbambisa, S.M.
Mailu, N. Njomo, M. Masikini, P. Baker, and E.I. Iwuoha,
Int. J. Electrochem. Sci. 7, 11859 (2012).
8. P. Kar and A. Choudhury, Adv. Polym. Technol. 32, E760
9. J.C. Yu, J.A. Hong, E.D. Jung, D.B. Kim, S.-M. Baek, S. Lee,
S. Cho, S.S. Park, K.J. Choi, and M.H. Song, Sci. Rep. 8,
1070 (2018).
10. J.N. Pereira, P. Vieira, A. Ferreira, A.J. Paleo, J.G. Rocha,
and S. Lanceros-Me
´ndez, J. Polym. Res. 19, 9815 (2012).
11. G. Wu, P. Tan, D. Wang, Z. Li, L. Peng, Y. Hu, C. Wang, W.
Zhu, S. Chen, and W. Chen, Sci. Rep. 7, 43676 (2017).
12. X.P. Chen, Q.H. Liang, J.K. Jiang, C.K.Y. Wong, S.Y.Y.
Leung, H.Y. Ye, D.G. Yang, and T.L. Ren, Sci. Rep. 6, 20621
13. J.C. Wang, R. Bruttini, and A.I. Liapis, Ind. Eng. Chem.
Res. 55, 6649 (2016).
14. J.B. de Lima Filho and A
´.A. Hidalgo, Synth. Met. 223, 80
15. S.R. Bhattacharyya, R.N. Gayen, R. Paul, and A.K. Pal,
Thin Solid Films 517, 5530 (2009).
16. J. Schmidhuber, Neural Netw. 61, 85 (2015).
17. R. Simon, Resampling strategies for model assessment and
selection. in Fundamentals of Data Mining in Genomics and
Proteomics, (Springer, New York, 2007), pp. 173–186.
18. Z. Wang and G.P. Rangaiah, Ind. Eng. Chem. Res. 56, 560
19. K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, IEEE
Trans. Evol. Comput. 6, 182 (2002).
20. A. Bahramian, Thin Solid Films 592, 39 (2015).
21. A. Bahramian, Surf. Interface Anal. 47, 1 (2015).
22. U. Ozdemir, B. Ozbay, S. Veli, and S. Zor, Chem. Eng. J.
178, 183 (2011).
23. G.H. Shafabakhsh, O.J. Ani, and M. Talebsafa, Constr.
Build. Mater. 85, 136 (2015).
24. M. Tanzifi, S.H. Hosseini, A.D. Kiadehi, M. Olazar, K.
Karimipour, R. Rezaiemehr, and I. Ali, J. Mol. Liq. 244, 189
25. X. X. Wu and J. G. Liu, A new early stopping algorithm
for improving neural network generalization. in: Proceed-
ings of the 2009 Second International Conference on Intel-
ligent Computation Technology and Automation, (2009)
pp. 15–18.
26. Mathworks, Genetic Algorithm Options, MATLAB Docu-
mentation Center 2014, (2014), No. accessed 3 June 2014.
27. Mathworks, Global Optimization Toolbox 3.2.5 User Guide,
28. X. Tang and X. Yan, J. Sol-Gel. Sci. Technol. 81, 378
29. M. Guglielmi, P. Colombo, F. Peron, and L.M. Degli Esposti,
J. Mater. Sci. 27, 5052 (1992).
30. W. Zhao, L. Ye, S. Zhang, B. Fan, M. Sun, and J. Hou, Sci.
Rep. 4, 6570 (2014).
31. D.S. Fryer, R.D. Peters, E.J. Kim, J.E. Tomaszewski, J.J. de
Pablo, P.F. Nealey, C.C. White, and W.L. Wu, Macro-
molecules 34, 5627 (2001).
32. A. Bahramian, J. Appl. Polym. Sci. 132, 41858 (2015).
33. J.C. Manifacier, J. Gasiot, and J.P. Fillard, J. Phys. E 9,
1002 (1976).
34. R. Swanepoel, J. Phys. E: Sci. Instrum. 16, 1215 (1983).
35. M.S. Ghamsari and A.R. Bahramian, High transparent sol–
gel derived nanostructured TiO
thin film. Mater. Lett. 62,
361 (2008).
36. J. Tauc and A. Menth, J. Non-Cryst. Solids 8–10, 569 (1972).
37. F. Kanwal, A. Batool, M. Adnan, and S. Naseem, Mater. Res
Innov. 19, S8-354 (2015).
38. Y.J. Cheng, S.H. Yang, and C.S. Hsu, Chem. Rev. 109, 5868
39. A.G. Baker, ARO Sci. J. Koya Univ. 7, 47 (2019).
40. E.M. Scherr, A.G. MacDiarmid, S.K. Manohar, J.G. Mas-
ters, Y. Sun, X. Tang, M.A. Druy, P.J. Glatkowski, V.B.
Cajipe, J.E. Fischer, K.R. Cromack, M.E. Jozefowicz, J.M.
Ginder, R.P. McCall, and A.J. Epstein, Synth. Met. 41, 735
41. Q. Qin and R. Zhang, Electrochim. Acta 89, 726 (2013).
42. Mathworks, Neural Network Toolbox 8.2 User Guide (2014).
43. Q. Tai, B. Chen, F. Guo, S. Xu, H. Hu, B. Sebo, and X.-Z.
Zhao, ACS Nano 5, 3795 (2011).
44. H. Bejbouji, L. Vignau, J.L. Miane, M.-T. Dang, E.M. Oua-
lim, M. Harmouchi, and A. Mouhsen, Sol. Energy Mater.
Sol. Cells 94, 176 (2010).
45. A.A.B. Baloch, S.P. Aly, M.I. Hossain, F. El-Mellouhi, N.
Tabet, and F.H. Alharbi, Sci. Rep. 7, 11984 (2017).
46. M. Sendova-Vassileva, H. Dikov, G. Popkirov, E. Lazarova,
V. Gancheva, G. Grancharov, D. Tsocheva, P. Mokreva, and
P. Vitanov, J. Phys: Conf. Ser. 514, 012018 (2014).
Publisher’s Note Springer Nature remains neutral with re-
gard to jurisdictional claims in published maps and institutional
Synthesis and Characterization of Conducting Polyaniline Nanostructured Thin Films for
Solar Cell Applications
Downloaded from
... Artificial neural networks estimate complex nonlinear associates existing between input and output variables to an arbitrary degree of accuracy. Recently, neural networks have been applied as a powerful modeling tool in different processes such as adsorption [6], flotation [7], liquid-liquid extraction [8] and many other fields of modeling problems [9,10]. They could be successfully used in regression and interpolating problems. ...
... Their results indicated that intelligent techniques outperform statistical approaches. Medi et al. [10] have used artificial neural network models with a novel k-fold crossvalidation technique to optimize minimum film thickness, maximum transparency and maximum conductivity. Their results showed that the developed model could predict the film characteristics with a reasonable error. ...
Full-text available
It is essential to know the process efficiency in the industrial magnetic separation process under different operating conditions because it is required to control the process parameters to optimize the process efficiency. To our knowledge, there is no information about using artificial intelligence for modeling the magnetic separation process. Hence, finding a robust and more accurate estimation method for predicting the separation efficiency and selectivity index is still necessary. In this regard, a feed-forward neural network was developed to predict the separation efficiency and selectivity index. This model was trained to present a predictive model based on the percentage of iron, iron oxide and sulfur in mill feed and cobber feed, 80% passing size in mill feed and cobber feed and plant capacity. Therefore, this work aims to develop an intelligent technique based on an artificial neural network and a hybrid neural-genetic algorithm for modeling the concentration process. Results indicated that the values of mean square error and coefficient of determination for the testing phase were obtained 0.635 and 0.86 for selectivity index and of 4.646 and 0.84 for separation efficiency, respectively. In order to improve the performance of neural network, genetic algorithm was used to optimize the weights and biases of neural network. The results of modeling with GA-ANN technique indicated that the mean square error and coefficient of determination for the testing phase were achieved by 0.276 and 0.95 for selectivity index and of 1.782 and 0.92 for separation efficiency, respectively. The other statistical criteria for the GA-ANN model were better than those of the ANN model.
... Artificial neural networks (ANNs) are general-purpose modeling tools that can be used for various applications, including static and dynamic modeling, clustering, and pattern recognition [8,9,30,35]. ANN is useful in particular when modeling with fundamental governing equations is costly, time-consuming or both [11,22]. Naregalkar and Subbulekshmi [25] proposed a novel approach using NARX (nonlinear autoregressive with exogenous input) model and enhanced moth flame optimization (EMFO) for pH neutralization of wastewater. ...
Full-text available
Chemical and biochemical processes generally suffer from extreme nonlinearities with respect to internal states, manipulated variables, and also disturbances. These processes have always received special technical and scientific attention due to their importance as the means of large-scale production of chemicals, pharmaceuticals, and biologically active agents. In this work, a general-purpose genetic algorithm (GA)-optimized neural network (NNARX) controller is introduced, which offers a very simple but efficient design. First, the proof of the controller stability is presented, which indicates that the controller is bounded-input bounded-output (BIBO) stable under simple conditions. Then the controller was tested for setpoint tracking, handling modeling error, and disturbance rejection on two nonlinear processes that is, a continuous fermentation and a continuous pH neutralization process. Compared to a conventional proportional-integral (PI) controller, the results indicated better performance of the controller for setpoint tracking and acceptable action for disturbance rejection. Hence, the GA-optimized NNARX controller can be implemented for a variety of nonlinear multi-input multi-output (MIMO) systems with minimal a-priori information of the process and the controller structure.
Full-text available
Nowadays, we are assisting in the exceptional growth in research relating to the development of wearable devices for sweat analysis. Sweat is a biofluid that contains useful health information and allows a non-invasive, continuous and comfortable collection. For this reason, it is an excellent biofluid for the detection of different analytes. In this work, electrochemical sensors based on polyaniline thin films deposited on the flexible substrate polyethylene terephthalate coated with indium tin oxide were studied. Polyaniline thin films were abstained by the potentiostatic deposition technique, applying a potential of +2 V vs. SCE for 90 s. To improve the sensor performance, the electronic substrate was modified with reduced graphene oxide, obtained at a constant potential of 􀀀0.8 V vs. SCE for 200 s, and then polyaniline thin films were electrodeposited on top of the as-deposited substrate. All samples were characterized by XRD, SEM, EDS, static contact angle and FT-IR/ATR analysis to correlate the physical-chemical features with the performance of the sensors. The obtained electrodes were tested as pH sensors in the range from 2 to 8, showing good behavior, with a sensitivity of 62.3 mV/pH, very close to a Nernstian response, and a reproducibility of 3.8%. Interference tests, in the presence of competing ions, aimed to verify the selectivity, were also performed. Finally, a real sweat sample was collected, and the sweat pH was quantified with both the proposed sensor and a commercial pH meter, showing an excellent concordance.
Full-text available
The beneficial use of a hole transport layer (HTL) as a substitution for poly(3,4-ethlyenedioxythiophene): polystyrene sulfonate (PEDOT:PSS) is regarded as one of the most important approaches for improving the stability and efficiency of inverted perovskite solar cells. Here, we demonstrate highly efficient and stable inverted perovskite solar cells by applying a GO-doped PEDOT:PSS (PEDOT:GO) film as an HTL. The high performance of this solar cell stems from the excellent optical and electrical properties of the PEDOT:GO film, including a higher electrical conductivity, a higher work function related to the reduced contact barrier between the perovskite layer and the PEDOT:GO layer, enhanced crystallinity of the perovskite crystal, and suppressed leakage current. Moreover, the device with the PEDOT:GO layer showed excellent long-term stability in ambient air conditions. Thus, the enhancement in the efficiency and the excellent stability of inverted perovskite solar cells are promising for the eventual commercialization of perovskite optoelectronic devices.
Full-text available
Advances in computational materials have paved a way to design efficient solar cells by identifying the optimal properties of the device layers. Conventionally, the device optimization has been governed by single or double descriptors for an individual layer; mostly the absorbing layer. However, the performance of the device depends collectively on all the properties of the material and the geometry of each layer in the cell. To address this issue of multi-property optimization and to avoid the paradigm of reoccurring materials in the solar cell field, a full space material-independent optimization approach is developed and presented in this paper. The method is employed to obtain an optimized material data set for maximum efficiency and for targeted functionality for each layer. To ensure the robustness of the method, two cases are studied; namely perovskite solar cells device optimization and cadmium-free CIGS solar cell. The implementation determines the desirable optoelectronic properties of transport mediums and contacts that can maximize the efficiency for both cases. The resulted data sets of material properties can be matched with those in materials databases or by further microscopic material design. Moreover, the presented multi-property optimization framework can be extended to design any solid-state device.
Full-text available
Supercapacitors, which store electrical energy through reversible ion on the surface of conductive electrodes have gained enormous attention for variously portable energy storage devices. Since the capacitive performance is mainly determined by the structural and electrochemical properties of electrodes, the electrodes become more crucial to higher performance. However, due to the disordered microstructure and low electrochemical activity of electrode for ion tortuous migration and accumulation, the supercapacitors present relatively low capacitance and energy density. Here we report a high-performance supercapacitor based on polyaniline/vertical-aligned carbon nanotubes (PANI/VA-CNTs) nanocomposite electrodes where the vertical-aligned-structure is formed by the electrochemical-induction (0.75 V). The supercapacitor displays large specific capacitance of 403.3 F g−1, which is 6 times higher than disordered CNTs in HClO4 electrolyte. Additionally, the supercapacitor can also present high specific capacitance (314.6 F g−1), excellent cycling stability (90.2% retention after 3000 cycles at 4 A g−1) and high energy density (98.1 Wh kg−1) in EMIBF4 organic electrolyte. The key to high-performance lies in the vertical-aligned-structure providing direct path channel for ion faster diffusion and high electrochemical capacitance of polyaniline for ion more accommodation.
Full-text available
This paper presents a review on dip-coating for fibrous materials, mainly concentrated on the mechanism, recently developed dip-coating methods and novel functional applications. The emphasis has been made here, to present theoretical basis of dip-coating-induced film deposition, especially, the reported works to predict the thickness based on various processing parameters. Different modified dip-coating techniques to fabricate deposited films for fibrous substrate have also been gathered. The scope of reviewed dip-coating methods are not only conventional solution and sol–gel-based dip-coating, but also recently developed vacuum-assisted, spin-assisted, photo-assisted and multi-layered dip-coating methods. An overview of reported and potential applications for coated fibrous materials has also been given, which mainly including self-cleaning, oil–water separation, conductive textiles, fibrous-based energy storage devices, and photonic crystals, etc. This review is intended to give readers a good horizon for the present status concerning variety of studies and applications related to dip-coating. An effort has been made here to report the important contributions in the area of dip-coating for fibrous substrate, and critical points regarding future research directions are outlined in the summary. Graphical Abstract
Methyl ammonium lead iodide (MAPbI3) perovskite film has been fabricated and modified with conductive polyanilines (C- PANI) and demonstrated for highly stable and efficient solar cells. For the C-PANI modified perovskite solar cells (PSCs), the average short circuit current density increased by 13.8% compared with pristine perovskite film (i.e., from 15.9 to 18.1 mA/cm² ) and the fill factor also improved, eventually leading to a power conversion efficiency (PCE) enhancement from 7.91% to 9.43% with high reproducibility. The C-PANI acts as a scaffold for crystallization of the perovskite layer, and improves the perovskite photovoltaic performance and stability, attributed to the favorable interfacial interaction between –NH2⁺– groups of C-PANI and I- of MAPbI3 perovskite. The C-PANI-rich surfaces of MAPbI3/PNAI perovskite film may also assists in the hole transport to the conductive carbon in the device. This work demonstrates that C-PANI can enhance the PCE and stability of PSCs.
Conductive polymers (CPs) have received significant research attention in material engineering for applications in microelectronics, micro-scale sensors, electromagnetic shielding, and micro actuators. Numerous research efforts have been focused on enhancing the conductivity of CPs by doping. Various conductive materials, such as metal nanoparticles and carbon-based nanoparticles, and structures, such as silver nanoparticles and graphene nanosheets, have been converted into polypyrrole and polypyrrole compounds as the precursors to developing hybrids, conjugates, or crystal nodes within the matrix to enhance the various structural properties, particularly the electrical conductivity. This article reviews nanomaterial doping of conductive polymers alongside technological advancements in the development and application of nanomaterial-doped polymeric systems. Emphasis is given to conductive nanomaterials such as nano-silver particles and carbon-based nanoparticles, graphene nano-sheets, fullerene, and carbon nanotubes (CNT) as dopants for polypyrrole-based CPs. The nature of induced electrical properties including electromagnetic absorption, electrical capacitance, and conductivities of polypyrrole systems is also discussed. The prospects and challenges associated with the development and application of CPs are also presented.
This study aims to synthesize a polyaniline nano-adsorbent and use it to adsorb methyl orange dye from the aqueous solution. The average particle size of nano-adsorbent was about 70 nm. The effects of various parameters have been analysed, as are pH, temperature, adsorption time, initial concentration and adsorbent dosage, and they were optimized by an artificial neural network model. The multilayer feed forward neural network with five inputs and one output has been trained with eight neurons in the hidden layer. A comparison of the experimental data with the dye adsorption efficiency predicted by the artificial neural network model showed that this model can estimate the behavior of the adsorption process of methyl orange dye on the polyaniline nano-adsorbent under different conditions. The study yielded the result that dye adsorption capacity of the nano-adsorbent increased from 3.34 to 32.04 mg/g and from 3.28 to 30.28 mg/g as the dye initial concentration was increased from 10 to 100 mg/L, at 65 °C and 25 °C, respectively. Also, dye adsorption equilibrium was achieved in 60 min for methyl orange dye. The adsorption kinetics was studied based on pseudo-first-order, pseudo-second-order, intraparticle diffusion and Elovich models. The results showed that the adsorption data at all levels of initial concentration have the best consistency with the pseudo-second order equation. Furthermore, two-parameter isotherm models (Langmuir, Freundlich and Temkin) and three-parameter isotherm models (Hill, Sips and R-P) were used to ascertain the nature of the adsorption isotherm. Based on this study, the maximum adsorption capacity was estimated to be 75.9 mg/g. Thermodynamic studies indicated that methyl orange dye adsorption was endothermic on the polyaniline nano-adsorbent.
Process optimization often has two or more objectives, which are conflicting. For such situations, Multi-Objective Optimization (MOO) provides many optimal solutions, which are equally good from the perspective of the given objectives. These solutions, known as Pareto-optimal front and as non-dominated solutions, provide deeper insights into trade-off among the objectives and many choices for implementation. In the past twenty years, researchers have applied MOO to numerous applications in chemical engineering. However, selection of an optimal solution from the set of non-dominated solutions has not received much attention in chemical engineering literature. In the present study, ten methods for selecting an optimal solution from the Pareto-optimal front are carefully chosen and implemented in an MS Excel-based program. Then, they are applied to the selection of an optimal solution in many benchmark/mathematical problems and chemical engineering applications, and their effectiveness and similarities are analysed. Results of analysis indicate that, among the ten methods studied, Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), Grey Relational Analysis (GRA) and Simple Additive Weighting (SAW) are better for choosing one of the Pareto-optimal solutions.
In this work, we have reported for the first time an efficient all-polymer tandem cell using identical sub-cells based on P2F-DO:N2200. A high power conversion efficiency (PCE) of 6.70% was achieved, which is among the highest efficiencies for all polymer solar cells and 43% larger than the PCE of single junction cell. The largely improved device performance can be mainly attributed to the enhanced absorption of tandem cell. Meanwhile, the carrier collection in device remains efficient by optimizing the recombination layer and sub-cell film thickness. Thus tandem structure can become an easy approach to effectively boost the performance of current all polymer solar cells.