sensors

Article
Sensing Mechanisms of Rough Plasmonic Surfaces for Protein
Binding of Surface Plasmon Resonance Detection
Treesukon Treebupachatsakul 1 , Siratchakrit Shinnakerdchoke 1 and Suejit Pechprasarn 2, *

1 Department of Biomedical Engineering, School of Engineering, King Mongkut’s Institute of Technology

Ladkrabang, Bangkok 10520, Thailand
2 College of Biomedical Engineering, Rangsit University, Pathum Thani 12000, Thailand
* Correspondence: [email protected]

Abstract: Surface plasmon resonance (SPR) has been utilized in various optical applications, includ-
ing biosensors. The SPR-based sensor is a gold standard for protein kinetic measurement due to its
ultrasensitivity on the plasmonic metal surface. However, a slight change in the surface morphology,
such as roughness or pattern, can significantly impact its performance. This study proposes a theo-
retical framework to explain sensing mechanisms and quantify sensing performance parameters of
angular surface plasmon resonance detection for binding kinetic sensing at different levels of surface
roughness. The theoretical investigation utilized two models, a protein layer coating on a rough
plasmonic surface with and without sidewall coatings. The two models enable us to separate and
quantify the enhancement factors due to the localized surface plasmon polaritons at sharp edges of
the rough surfaces and the increased surface area for protein binding due to roughness. The Gaussian
random surface technique was employed to create rough metal surfaces. Reflectance spectra and
quantitative performance parameters were simulated and quantified using rigorous coupled-wave
analysis and Monte Carlo simulation. These parameters include sensitivity, plasmonic dip position,
intensity contrast, full width at half maximum, plasmonic angle, and figure of merit. Roughness
can significantly impact the intensity measurement of binding kinetics, positively or negatively,
depending on the roughness levels. Due to the increased scattering loss, a tradeoff between sensitivity
Citation: Treebupachatsakul, T.; and increased roughness leads to a widened plasmonic reflectance dip. Some roughness profiles can
Shinnakerdchoke, S.; Pechprasarn, S. give a negative and enhanced sensitivity without broadening the SPR spectra. We also discuss how
Sensing Mechanisms of Rough the improved sensitivity of rough surfaces is predominantly due to the localized surface wave, not
Plasmonic Surfaces for Protein the increased density of the binding domain.
Binding of Surface Plasmon
Resonance Detection. Sensors 2023,
Keywords: sensing mechanisms; surface plasmon resonance; quantitative-sensing performance;
23, 3377. https://doi.org/10.3390/
binding-kinetics sensitivity; surface roughness; sensitivity-enhancement mechanisms
s23073377

Academic Editor: Nicole

Jaffrezic-Renault
1. Introduction
Received: 13 January 2023
Revised: 5 March 2023 Surface plasmon resonance (SPR) is a guided electromagnetic phenomenon that oc-
Accepted: 16 March 2023 curs when electrons oscillate between a noble metal and dielectric when illuminated by
Published: 23 March 2023 an external electric field, for example, an incident light with a matching momentum [1,2].
The coupling between the incident photon and electron oscillation leads to optical energy
absorption [3], resulting in a dark band in plasmonic reflectance spectra. The coupling
of SPR is a strong candidate for real-time monitoring of binding interactions [4], leading
Copyright: © 2023 by the authors.
to the development of sensors based on SPR along with numerous other optical sensing
Licensee MDPI, Basel, Switzerland.
technologies. Examples of applications using SPR include host–pathogen detection [5,6],
This article is an open access article
biochemical interactions [7,8], ultrasonic detection [9], refractive index sensing [10,11],
distributed under the terms and
voltage sensing [12], drug discovery [13,14], microscopic imaging [15], and therapeutic
conditions of the Creative Commons
Attribution (CC BY) license (https://
monitoring [16].
creativecommons.org/licenses/by/
A widely adopted SPR-based sensor consists of a glass substrate with a refractive
4.0/). index of n0 , a noble metal film with a refractive index of nm , and a sensing region with

Sensors 2023, 23, 3377. https://doi.org/10.3390/s23073377 https://www.mdpi.com/journal/sensors

dent angle θ0 at a fixed wavelength of λ, illuminating the thin plasmonic metal due to the
explained light-matter interaction between photons and electrons. It manifests as a dark
band in the reflectance spectra at a specific incident angle, known as the plasmonic angle
θsp, as depicted in Figure 1b. Furthermore, alterations on the metal film surface, such as
molecular binding at an active site on the sensor surface, resulting in a subsequent altera-
Sensors 2023, 23, 3377 tion in the resonant coupling condition and a shift in the plasmonic angle, as depicted 2 ofin
22
Figure 1b.
Figure 1b shows the plasmonic angle shift from the water-backing environment with
aarefractive
refractive index
index of
of 1.33 to a structure,
ns . This 5 nm thickalso
bovine
knownserum albumin
as the (BSA) with
Kretschmann a BSA refrac-
configuration [17],
tive index nin
is depicted of 1.351a.
BSAFigure forThere
an 80aremg/mL
severalconcentration
measurement[22] adhering
schemes to the
for SPR, metal surface,
including angular
respectively.
interrogationFigure 1b was
detection [18],calculated using
wavelength the computational
scanning method
detection [19], explained
intensity later[20],
detection in
Section 2. detection [21].
and phase

(a)The
Figure1.1.(a)
Figure TheKretschmann
Kretschmann configuration consisting
consistingofofaawater-based
water-basedsensing
sensingregion,
region,BSA
BSAbinding
bind-
ing layer,
layer, noble
noble metallic
metallic film,
film, a glass
a glass substrate,and
substrate, andincident
incidentlight,
light,and
and (b)
(b) the
the SPR sensor’s
sensor’s outputs
outputs
consisting of reflectance
consisting of reflectancecurves,
curves,measuring
measuringfromfrom
the the water-based
water-based environment
environment sensing
sensing regionregion
shown
shown in the
in the blue blue and
curve curve
a 5and a 5 nm-thick
nm-thick BSA coating
BSA coating layer inlayer in shown
water water shown in the
in the red red curve.
curve.

Multiple
The angular studies [23–28] have
interrogation reported
detection [18]that the surface
measures profile, while
reflectance including the thick-
varying the in-
ness
cidentand roughness
angle θ 0 at a of the wavelength
fixed metal film, significantly
of λ, illuminating affectsthe
the thin
performance
plasmonicofmetalthe SPR-
due
based
to thesensor
explained for bulk refractiveinteraction
light-matter index sensing.between For example,
photons andKurihara et al. [27]
electrons. explained
It manifests as
that theband
a dark SPR in dipsthecould be classified
reflectance spectra into large, small,
at a specific incidentandangle,
equivalent
knownrelationships
as the plasmonicbe-
angle absorption
tween θ sp , as depicted in Figure frequency,
and excitation 1b. Furthermore,depending alterations on the metal
on the frequency film surface,
relationship be-
such as molecular binding at an active site on the sensor surface,
tween the optical property of the light source and the metal layer thickness. Furthermore, resulting in a subsequent
alterationet
Hoffman in al.
the[26]
resonant coupling condition
demonstrated a significant and change
a shift ininthe plasmonic
dielectric angle, asand
functions depicted
SPR
in Figure
wave vectors1b. as the roughness amplitudes increased. Our recent publication [28] found
Figure
that the 1b shows
sensitivity, thewidth
full plasmonic angle
at half shift from
maximum, the water-backing
intensity contrast, and environment
plasmonic with dip
a refractive
position for index
measuringof 1.33the to arefractive
5 nm thick bovine
index usingserum albumin (BSA)
an SPR-based withwith
sensor a BSA refractive
rough gold
films nBSAtoofa 1.35
indexleads for an 80
degradation ofmg/mL concentration
the sensor’s [22] adhering to the metal surface,
overall quality.
respectively. Figure 1b was calculated using the
Several papers [29,30] reported that the sensitivity of binding computational method explained
kinetic later in
measurement
Section 2.
increases with the roughness level due to a larger surface area and higher surface protein
Multiple
in contact withstudies [23–28] have
the plasmonic metal.reported that to
It is crucial thenote
surface
that profile,
branching including the thickness
molecules, such as
and roughness
dextran, have been of the metal film,
employed significantly
in binding surfaceaffects the performance
preparation to increase theof the SPR-based
binding den-
sensor for bulk refractive index sensing. For example, Kurihara
sity. Unlike in rough plasmonic surfaces, for example, Liu et al. [30] employed a Ni seed et al. [27] explained that the
SPR dips could be classified into large, small, and equivalent relationships
layer on a silver-based SPR sensor to diminish the surface roughness, resulting in a smaller between absorp-
tion andIn
FWHM. excitation
addition,frequency,
Byun et depending on the the
al. [29] analyzed frequency
localizedrelationship between resonance
surface plasmon the optical
property of the light source and the metal layer thickness. Furthermore, Hoffman et al. [26]
demonstrated a significant change in dielectric functions and SPR wave vectors as the
roughness amplitudes increased. Our recent publication [28] found that the sensitivity, full
width at half maximum, intensity contrast, and plasmonic dip position for measuring the
refractive index using an SPR-based sensor with rough gold films leads to a degradation of
the sensor’s overall quality.
Several papers [29,30] reported that the sensitivity of binding kinetic measurement
increases with the roughness level due to a larger surface area and higher surface protein
in contact with the plasmonic metal. It is crucial to note that branching molecules, such
as dextran, have been employed in binding surface preparation to increase the binding
density. Unlike in rough plasmonic surfaces, for example, Liu et al. [30] employed a Ni
seed layer on a silver-based SPR sensor to diminish the surface roughness, resulting in
a smaller FWHM. In addition, Byun et al. [29] analyzed the localized surface plasmon
resonance (LSPR) on a nanowire-based structure with different degrees of roughness. They
Sensors 2023, 23, 3377 3 of 22

also concluded that there was a change in the extinction peak amplitude caused by the
LSPR effect.
Various surface treatment techniques have been employed as a postprocessing step
for metal film fabrication to reduce surface roughness. Conventionally, the metal layer of
the SPR-based sensor was manufactured using a sputter coating, which has a root-mean-
square (RMS) roughness of 1.4 nm to 1.5 nm [31]. The metal film can be further smoothed
through surface treatments, such as thermal annealing [32], helium ion beam [33], laser
ablation [34], chemically grown single-crystalline gold [35], mica substrate utilizing [36],
and chemical polishing [37,38]. Table 1 summarizes the roughness values for the mentioned
treatment methods.

Table 1. The root mean square (RMS) roughness of metal film after different surface treatments.

Surface Treatment Techniques RMS Roughness

Sputter Coating (No Treatment) [31] 1.4–2.5 nm
Thermal Annealing [32] <1 nm
Chemically Grown Single-Crystalline Gold [35] <1 nm
Chemical Polishing [37,38] 0.38 ± 0.05 nm
Helium Ion Beam [33] 0.267 nm
Mica Substrate Utilizing [36] 0.2 nm
Laser Ablation [34] 0.17 nm

This paper theoretically investigated multiple roughness levels on a uniform gold

sensor to analyze roughness’s effects on the SPR protein binding performance. This study
has chosen the gold material due to its chemical stability and biocompatibility for biochem-
ical and bioscience-based applications. A theoretical framework is proposed to analyze
comparative sensing performance parameters, including protein binding sensitivity (S), the
plasmonic dip position (n0 sinθ sp ), the full width at half maximum (FWHM), the intensity
contrast (∆I), the average plasmonic dip intensity (Isp ), and the figure of merit (FoM). The
theoretical investigation was based on rigorous coupled-wave analysis [39,40] and Monte
Carlo simulation [41].
It is significant to note that the research differs from previous literature regarding
protein binding to SPR sensors with different roughness levels. We systematically separate
and quantify the two enhancement schemes: the LSPR and the increased binding density.
The proposed method provides insight into the underlying performance enhancement of
SPR sensors for binding kinetic measurement. In addition, we believe that the research
provides a complete explanation of the impact of surface roughness on the molecular
binding application of the SPR-based sensor performance and its limitations.

2. Materials and Methods

2.1. Rough Surface Model and Simulated Structures
The rough gold surface was constructed using numerical modeling with Gaussian
random surfaces [42], which were adopted from Byun et al. [29], and characterized by
varying maximum roughness height (h) and correlation length (cl) from 2 nm to 20 nm and
5 nm to 50 nm, respectively. The roughness factors were selected to cover SPR excitation
and its cut-off positions. Different mode characteristics are present in the chosen roughness
scope, including effects from LSPR and enhanced protein binding density, and these will
be discussed later in the results section.
A function s(x) consisting of digital numbers 0 and 1 in a random sequence was first
developed, where the term x indicated the spatial distance along the x-axis of the sensor to
generate the rough surface. The function was then multiplied by the maximum roughness
height, h. The generated profile was then converted to the frequency domain using Fourier
transformation, expressed as S(ω) = F[s(x)], where S(ω) is the Fourier-transformed output,
ω is the frequency domain axis, and F is the Fourier-transform operator. The S(ω) function
was then multiplied with the Gaussian distribution function, G(ω) = exp(−x2 /(cl2 /2)).
The G(ω) function essentially acts as a low-pass filter for the signal, where the term cl
function was then multiplied with the Gaussian distribution function, G(ω) =
exp(−x2/(cl2/2)). The G(ω) function essentially acts as a low-pass filter for the signal, where
the term cl indicates the roughness correlation length. Next, the inverse Fourier transfor-
Sensors 2023, 23, 3377
mation was applied to convert the signal back to the spatial domain signal and 4rearrange of 22
the function into a 200 columns rectangular contour. It is worth noting that higher cl val-
ues indicate a smoother surface. Therefore, a theoretical uniform surface is a Gaussian
random surface
indicates with h equal
the roughness to zerolength.
correlation and clNext,
approaching
the inverse infinity.
Fourier transformation was
The rough surface was then rearranged into a rectangular
applied to convert the signal back to the spatial domain signal and rearrange coordinate for rigorous
the function
coupled-wave
into a 200 columns analysis simulation
rectangular by dividing
contour. It is worththe surface
noting into multiple
that higher cl valueslayers
indicate with a
a smoother
height of 0.1 nmsurface.
each.Therefore,
The rough a theoretical
layer wasuniform surface is awith
then assembled Gaussian random gold
the uniform surfacesurface
of the h equalThe
withsensor. to zero
rough cl approaching
and surface structure infinity.
profile consisted of three regions: (1) the yellow
The rough surface was then
area beneath the surface indicating the plasmonicrearranged into agold
rectangular
material coordinate for rigorous
with a refractive index of
coupled-wave analysis simulation by dividing the surface into multiple layers with a height
0.18 + 3.43i [43], (2) the light green contour indicating the 5 nm-thick BSA binding layer
of 0.1 nm each. The rough layer was then assembled with the uniform gold surface of
[44] with a refractive index of 1.35, equivalent to 80 mg/mL concentration [22], and (3) the
the sensor. The rough surface structure profile consisted of three regions: (1) the yellow
blue region
area beneath indicating
the surface theindicating
water-based environment
the plasmonic sensingwith
gold material areaawith a refractive
refractive index ofindex
of 1.33.
0.18 +BSA
3.43iwas
[43], chosen as agreen
(2) the light substrate
contour layer in thisthe
indicating investigation
5 nm-thick BSA duebinding
to its numerous
layer [44] ap-
plications, including
with a refractive indexcell
of culture and studies
1.35, equivalent of biochemical
to 80 mg/mL reactions.
concentration Furthermore,
[22], and (3) the blue BSA,
including in biomedical
region indicating research,environment
the water-based has been widelysensing areaadopted
with aas a protein
refractive concentration
index of 1.33.
BSA was
standard inchosen as a substrate
laboratory layer in this investigation due to its numerous applications,
experiments.
including cell culture and studies
The theoretical study was conducted of biochemical
usingreactions.
three setsFurthermore,
of samples: BSA, including
in biomedical research, has been widely adopted as a protein concentration standard in
(1) no BSA protein layer case, as shown in Figure 2a;
laboratory experiments.
(2) the second sample was a rough surface with only a BSA layer on top with no
The theoretical study was conducted using three sets of samples:
sidewall (1)coating, as depicted
no BSA protein in Figure
layer case, 2b. This
as shown model
in Figure 2a;indicated an equivalent amount of
protein(2)to the
thesecond
ideallysample
smoothwas metal,
a roughallowing
surfaceuswithto evaluate
only a BSA thelayer
enhancement
on top with due noto the
LSPR without
sidewall the influence
coating, as depictedofinthe increased
Figure 2b. Thisamount of protein
model indicated andue to the roughness;
equivalent amount of
(3) Conversely,
protein to the ideally the BSA layer
smooth metal,inallowing
the thirdussample typethe
to evaluate was coated on top
enhancement dueoftothe
thestruc-
LSPR
ture andwithout the influencesidewall,
its rough-surface of the increased amount of
as illustrated inprotein
Figuredue2c. to
Thetheinvestigation
roughness; of this
(3) Conversely, the BSA layer in the third sample type was
structure aimed to replicate the protein solution binding to the entire surface, leading to coated on top of the
an structure
increased and its rough-surface sidewall, as illustrated in Figure 2c. The investigation of this
protein sample binding concentration. However, the amount of BSA in the
structure aimed to replicate the protein solution binding to the entire surface, leading to an
third sample can be significantly denser than in the second sample when the surface is
increased protein sample binding concentration. However, the amount of BSA in the third
highly
samplerough,
can be assignificantly
explained later denser inthan
Section
in the3.3. Therefore,
second sample thewhen increased
the surface protein
is highlyand the
LSPR canasenhance
rough, explainedthe SPR
later response
in Section 3.3. in the third
Therefore, thesample,
increasedwhich
proteintheandsecond
the LSPR model
can can
suppress to quantify and separate the two enhancement mechanisms.
enhance the SPR response in the third sample, which the second model can suppress to
quantify and separate the two enhancement mechanisms.

Figure 2. 2.
Figure TheTheGaussian
Gaussianrandom
random rough surfacefor
rough surface forh h2020
nmnm of and
of and cl ofcl5of
nm 5 for
nm(a)forno(a) noprotein
BSA BSA protein
coating
coating case (bare gold case), (b) BSA protein without sidewall coating, and (c) BSA protein with with
case (bare gold case), (b) BSA protein without sidewall coating, and (c) BSA protein
sidewall
sidewallcoating.
coating.The
Thefigures
figures on
on the rightshow
the right showthe thezoomed-in
zoomed-in version
version of rough
of rough surfaces
surfaces for thefor the
three models.
three models.
Sensors 2023, 23, x FOR PEER REVIEW 5 of 23

Sensors 2023, 23, 3377 5 of 22

The sidewall coating behavior of proteins in nature was more complicated than in
the simulation. Even though sidewall coating is typically used to prevent edges from com-
ing intoThedirect
sidewall coating
contact with behavior
water andoftoproteins
increaseinbinding
nature was more
density, complicated
this assumptionthan mayin
thealways
not simulation. Even though
be accurate. Even with sidewall coating
sidewall is typically
coating, edges canused to prevent
remain in directedges from
contact
coming into direct contact with water and to increase binding density,
with water, which could affect the precision of the provided data. In such a way, the en- this assumption may
not always be accurate. Even with sidewall coating, edges can remain
hanced sensitivity and FoM will be between models (2) and (3). The results section will in direct contact
with water,
discuss which the
that despite could affect of
absence thesidewall
precision of theroughness
coating, provided profiles
data. Incan such a way,the
improve the
enhanced sensitivity and FoM will be between models (2) and
SPR measurement due to the LSPR. However, many studies on polymer technology [45– (3). The results section will
discuss
48] assistthat despite theprotein
in controlling absence of sidewallSaleem
fabrication. coating, roughness
et al. [45] andprofiles
Nuutinen canetimprove the
al. [46] in-
SPR measurement due to the LSPR. However, many studies on polymer
vestigate the application of polymeric material on resonant wave gratings (RWG). They technology [45–48]
assist inTiO
utilized controlling protein fabrication. Saleem et al. [45] and Nuutinen et al. [46] investigate
2 as a cover layer on the gratings structure, and then the protein layer was
the application of polymeric
uniformly bound to the material. material
Mayeton et resonant wave gratings
al. [47] published a similar (RWG).
studyThey utilized
in which an
TiO 2 as a cover
oxide-nitride-oxide layer on the gratings structure, and then the protein layer
passivation layer evenly coated the sidewalls of the gratings. Zhang was uniformly
etbound
al. [48]toalso
the conducted
material. Mayet et al. [47]
a theoretical published
study involvinga similar study
protein in which
sidewall coatingan oxide-nitride-
on a metal-
oxide passivation layer evenly coated the sidewalls of the gratings. Zhang et al. [48] also
layer-assisted double-grating (MADG) biosensor. These indicate that the experimental
conducted a theoretical study involving protein sidewall coating on a metal-layer-assisted
and theoretical information agree well and that the protein coating behavior can be con-
double-grating (MADG) biosensor. These indicate that the experimental and theoretical
trolled.
information agree well and that the protein coating behavior can be controlled.
Figure 3 illustrates the overall simulation diagram, which consists of a BK7 glass sub-
Figure 3 illustrates the overall simulation diagram, which consists of a BK7 glass
strate with a refractive index of 1.52, a uniform gold surface with a thickness of 50-h nm,
substrate with a refractive index of 1.52, a uniform gold surface with a thickness of 50-h nm,
a mean rough gold layer with a thickness and depth of h, a water-based environment
a mean rough gold layer with a thickness and depth of h, a water-based environment
sensing area with a refractive index of 1.33, a binding analyte (an 80 mg/mL concentrated
sensing area with a refractive index of 1.33, a binding analyte (an 80 mg/mL concentrated
BSA protein) with a refractive index of 1.35, and a thickness of 5 nm lining the top and
BSA protein) with a refractive index of 1.35, and a thickness of 5 nm lining the top and
sidewall of the rough surface layer, and a linearly p-polarized HeNe laser with a wave-
sidewall of the rough surface layer, and a linearly p-polarized HeNe laser with a wavelength
length of 633 nm. The laser illuminates the gold film at an incident angle, scanning from
of 633 nm. The laser illuminates the gold film at an incident angle, scanning from the critical
the critical
angle to 90 angle to 90°. According
◦ . According to the sampling
to the sampling theory, thetheory,
unit the
cell unit
lengthcellislength
1 µm is 1 μm
and and
consists
consists of multiple rows with 1 nm each and 200 columns to ensure
of multiple rows with 1 nm each and 200 columns to ensure that the unit cell can provide that the unit cell can
provide different levels of the employed roughness with the minimum
different levels of the employed roughness with the minimum sampling feature size of sampling feature
size
5 nm.of 5The
nm. The proposed
proposed 200 columns
200 columns to represent
to represent the roughness
the roughness in this
in this study willstudy will be
be discussed
discussed
and validated and validated later
later in the in the
results results section.
section.

Figure 3. Simulated rough SPR sensor based on the Kretschmann configuration.

Figure 3. Simulated rough SPR sensor based on the Kretschmann configuration.
As depicted in Figure 3, roughness profiles simulated in this study were simplified
As depicted in Figure 3, roughness profiles simulated in this study were simplified
to a one-dimensional roughness profile, where the y-axis was treated as an infinite exten-
to a one-dimensional roughness profile, where the y-axis was treated as an infinite exten-
sion of the x-z plane. The Krestschmann configuration in this study was a prism-based
sion of the x-z plane. The Krestschmann configuration in this study was a prism-based
Kretschmann configuration with TM polarization illumination. The TM incident electric
Kretschmann configuration
field can be resolved in thewith TM polarization
x-z plane, illumination.
with no electric Thethe
field along TMy-axis.
incident electric
That is the
field can be resolved in the x-z plane, with no electric field along the y-axis. That
reason for the uniform y-axis in this study. It is essential to point out that this assumption is the
reason
may notforhold
the uniform y-axis
if the SPR in this study.
illumination It is essential
is achieved using ato point
high NAout that this
objective assumption
lens; there will
may not hold if the SPR illumination is achieved using a high NA objective
be p-polarization components along the y-axis [49]. It will be shown in the results lens; there will
section
be p-polarization components along the y-axis [49]. It will be shown in the
that the proposed simplified simulation can give similar SPR reflectance spectra comparedresults section
Sensors 2023, 23, x FOR PEER REVIEW 6 of 23

Sensors 2023, 23, 3377 that the proposed simplified simulation can give similar SPR reflectance spectra compared
6 of 22
to the experimental results reported in the literature for the prism-based configuration
and fabricated two-dimensional rough-gold sensors at different roughness levels.
to the experimental results reported in the literature for the prism-based configuration and
2.2.fabricated
Simulationtwo-dimensional rough-gold
of the Surface Plasmon sensors Detection
Resonance at different roughness levels.
This
2.2. research
Simulation of theemployed the Resonance
Surface Plasmon rigorous Detection
coupled-wave analysis (RCWA) function
[40,50] This
andresearch
the Monte Carlo simulation to
employed the rigorous coupled-wave calculate the reflectance
analysis (RCWA) functioncurve under a
[40,50]
MATLAB 2022bCarlo
and the Monte environment,
simulation toutilizing parallel
calculate the computing
reflectance anda graphic
curve under MATLAB processing
2022b
toolboxes.
environment, utilizing parallel computing and graphic processing toolboxes. In addition, from
In addition, 151 diffraction orders were applied to ensure that all results
the151
studied surfaces
diffraction achieved
orders convergence,
were applied to ensurewhich willresults
that all be further discussed
from the studiedin Section 3.1.
surfaces
achieved
Since aconvergence,
single pair of which will
h and clbe
canfurther discussed
generate in Section
multiple 3.1. the Monte Carlo simu-
surfaces,
Sinceapplied
lation was a single to
pair of h and an
estimate cl can generate
average multipleof
response surfaces, the Monte
100 surface Carlofor
profiles simula-
the three
tion was applied to estimate an average response of 100 surface profiles for the three sample
sample types. These responses from the rough surfaces generated by h and cl were then
types. These responses from the rough surfaces generated by h and cl were then analyzed
analyzed using the RCWA. Thus, each Monte Carlo simulation consists of performance
using the RCWA. Thus, each Monte Carlo simulation consists of performance factors’ val-
factors’ values
ues from fromRCWA
100 times 100 times RCWA
simulation withsimulation
a hundred with a hundred
different generateddifferent generated
rough surfaces
rough surfaces using a single value of h and cl, as illustrated in the flow
using a single value of h and cl, as illustrated in the flow chart in Figure 4. This procedurechart in Figure 4.
This procedure
allows allows
for a high for aand
accuracy high accuracy
stability of theand stability
calculated of the calculated
performance performance
parameters across all pa-
rameters across
values of all valuesfactors.
the roughness of the roughness factors.

Figure
Figure 4. 4.
TheThe flowchartof
flowchart ofthe
the overall
overall simulation
simulationprocesses, including
processes, the rigorous
including coupled-wave
the rigorous coupled-wave
analysis simulation, Monte Carlo simulation, and the performance parameters calculation.
analysis simulation, Monte Carlo simulation, and the performance parameters calculation.
2.3. Quantitative Performance Parameters
The studied performance parameters of surface plasmon resonance detection for
protein binding applications included sensitivity (S), full width at half maximum (FWHM),
intensity contrast (∆I), average plasmonic dip intensity (Isp ), and figure of merit (FoM).
This section provided definitions and descriptions of the presented performance factors;
Sensors 2023, 23, 3377 7 of 22

these parameters will be computed to explain the SPR sensing mechanisms for the protein
binding in the results section.
(1) The sensitivity (S) was calculated by dividing the change in the plasmonic wave vector
(k), also known as the plasmonic shifting distance, by the product of the protein layer
thickness and the difference in the binding region’s refractive indices, as shown in
Equation (1) and Figure 5b.

∆k 2πn0 ∆sinθsp
S= = (1)
d∆ns λd BSA ∆ns

The terms n0 , θ sp , dBSA , λ, and ∆ns are the refractive index of the glass substrate (1.52),
the plasmonic angle, the 5-nm BSA protein layer, the wavelength of the incident light
(633 nm), and the difference in sample refractive indices of 0.02, respectively.
(2) The n0 sinθ sp indicates the angular position at the minimum plasmonic intensity dip
(Isp ), which can be used to analyze the sensor’s detection range.
(3) The FWHM in this manuscript was calculated as the average width of the two plas-
monic dips, measured at 50% of the reflectance spectra’s intensity, as expressed in
Equation (2). Figure 5a illustrates the FWHM of the SPR curves, depicted as a black
arrow considering the unsymmetrical nature of SPR dips.

∆k nor,water + ∆k nor,BSA
FW HM = (2)
2
The terms ∆k nor, water and ∆k nor,BSA are the FWHM of the normalized reflectance
curves, measured at half of the maximum magnitude when the refractive index of the
sensing region was 1.33 and 1.35, respectively. In addition, several techniques can be
applied to measure the FWHM of the unsymmetrical SPR dip, such as Lorentzian fit [51],
Green’s function [52], and Horner’s curve fitting [53].
(4) The ∆I is one of the unique parameters which helps to determine the quality of the
surface plasmon resonance-based sensor. It can be calculated as the average difference
between the reflectance at the plasmonic angle and the intensity at the critical angle of
the reflectance spectra, as expressed in Equation (3) and illustrated in Figure 5a.

|∆Iwater | + |∆IBSA |
∆I = (3)
2
The terms ∆Iwater and ∆IBSA are the difference in the intensity at the plasmonic angles
of the reflectance spectra detected when the 5 nm thick BSA binding layer is absent and
present, respectively.
(5) The Isp is another factor for considering the sensor’s performance. It can be defined as
the average of the reflectance values at the plasmonic angles as expressed in Equation
(4) and shown in Figure 5b.

Isp,water + Isp,BSA
Isp = (4)
2
The terms Isp,water , and Isp,BSA are the intensity at the plasmonic angles of the reflectance
spectra computed without and with the binding substrate.
(6) FoM encapsulates an overall sensor quality, usually defined as a ratio between the S
and the FWHM [54], as expressed in Equation (5).

S
FoM1 = (5)
FW HM
It is interesting to point out that the FoM depends on the detection mechanism for
intensity-based measurement; the FoM is calculated based on the S and FWHM and can
Sensors 2023, 23, x FOR PEER REVIEW 8 of 23

Sensors 2023, 23, 3377 8 of 22

It is interesting to point out that the FoM depends on the detection mechanism for
intensity-based measurement; the FoM is calculated based on the S and FWHM and can
also
also bebe relatedtotointensity
related intensity parameters,
parameters,such ∆IΔI
suchasas and Isp ,Iand
and considering
sp, and the shot
considering the noise
shot noise
model
model [55],asasexpressed
[55], expressed in
in Equation
Equation(6).
(6).
√
S ×𝑆 ×∆I
√∆𝐼
FOM
𝐹𝑂𝑀=
2 2 = p (6) (6)
HM × ×
FW𝐹𝑊𝐻𝑀 4 I4
sp 𝐼
√ 𝑠𝑝

Figure
Figure5. 5.
TheThecalculation
calculationprocedures
procedures of ofthe
thediscussed
discussed quantitative
quantitative performance
performance parameters
parameters for thefor the
ideal uniform
ideal uniformgoldgoldsurface
surface sensor,
sensor, including ∆I,I,
including(a)(a) FWHM,
FWHM, (b) (b) S, and
S, and Isp . Isp.

2.4.2.4. EnhancementRatio
Enhancement Ratio
Later in the result section, we will discuss that the FWHM, ∆I, and Isp are comparable
Later in the result section, we will discuss that the FWHM, ΔI, and Isp are comparable
for both the nonsidewall and sidewall models, indicating that the additional protein from
forthe
both the nonsidewall and sidewall models, indicating that the additional
roughness does not affect these parameters. The S and FoM are the only performance
protein from
theparameters
roughnessthat doescannot
beaffect these
affected by parameters. The S and
the sidewall protein. FoMthe
Here, areenhancement
the only performance
ratio
parameters
denoted by ER compares the roughness models’ relative change in S and FoM to ratio
that can be affected by the sidewall protein. Here, the enhancement the de-
noted
idealby ER compares
uniform gold sensortheresponse.
roughness models’ relative
Furthermore, the LSPRchange in S and
wave, protein FoM to and
increment, the ideal
roughness
uniform goldeffect will be
sensor quantifiedFurthermore,
response. and classified using ER as expressed
the LSPR in Equations
wave, protein (7) to and
increment,
(9). Equations
roughness effect(7)will
to (9)
beshow the calculations
quantified for the using
and classified roughness enhancement
ER as expressed effect, LSPR (7)
in Equations
wave effect, and protein increase, respectively.
to (9). Equations (7) to (9) show the calculations for the roughness enhancement effect,
LSPR wave effect, and protein increase, respectively.
Ssidewall
ERS,roughness = (7)
Suni𝑆f 𝑠𝑖𝑑𝑒𝑤𝑎𝑙𝑙
orm
𝐸𝑅𝑆,𝑟𝑜𝑢𝑔ℎ𝑛𝑒𝑠𝑠 = (7)
S 𝑆𝑢𝑛𝑖𝑓𝑜𝑟𝑚
ERS,LSPR = non−sidewall (8)
Suni f orm
𝑆𝑛𝑜𝑛−𝑠𝑖𝑑𝑒𝑤𝑎𝑙𝑙
S 𝐸𝑅
− S𝑆,𝐿𝑆𝑃𝑅 = (8)
non−sidewall 𝑆
ERS,protein = sidewall ERS,roughness − ERS,LSPR
=𝑢𝑛𝑖𝑓𝑜𝑟𝑚 (9)
Suni f orm
The terms ERS,roughness , 𝑆 ER S,LSPR ,−
𝑠𝑖𝑑𝑒𝑤𝑎𝑙𝑙 and𝑆𝑛𝑜𝑛−𝑠𝑖𝑑𝑒𝑤𝑎𝑙𝑙
ERS,protein indicate the sensitivity enhancement (9)
𝐸𝑅𝑆,𝑝𝑟𝑜𝑡𝑒𝑖𝑛 = = 𝐸𝑅𝑆,𝑟𝑜𝑢𝑔ℎ𝑛𝑒𝑠𝑠 − 𝐸𝑅𝑆,𝐿𝑆𝑃𝑅
ratios caused by roughness (the LSPR and the increased binding
𝑆𝑢𝑛𝑖𝑓𝑜𝑟𝑚 density), the LSPR wave
effect, and increased protein binding density, respectively. Furthermore, Suniform , Ssidewall ,
andThe terms ER
Snon-sidewall areS,roughness, ERS,LSPR,obtained
the sensitivities and ERS,protein
by theindicate the sensitivity
ideal uniform gold model,enhancement
BSA with theratios
caused by roughness
sidewall model, and (the BSA LSPR
withoutand
thethe increased
sidewall model, binding density),
respectively. Thethe LSPR wave
enhancement foreffect,
the FoM can be calculated using the relative change in the FoM, similar to the sensitivity.
and increased protein binding density, respectively. Furthermore, Suniform, Ssidewall, and Snon-
sidewall are the sensitivities obtained by the ideal uniform gold model, BSA with the sidewall
3. Results
model, and BSA without the sidewall model, respectively. The enhancement for the FoM
3.1. Convergence Test for Extreme Cases
can be calculated using the relative change in the FoM, similar to the sensitivity.
The convergence test on the optimal diffraction orders was employed in four extreme
cases of the study, consisting of the rough surfaces generated using (a) h of 2 nm and cl of
diffraction orders ranging from 1 to 171. The numerical fluctuation between the 149 and
151 diffraction orders was 7 × 104, 0.002, 0.020, and 0.003 for the four tested surfaces. Em-
ploying a higher number of diffraction orders will surely increase the accuracy of the re-
sults; however, it also requires longer computing time and more resources. Therefore, the
Sensors 2023, 23, 3377
diffraction orders of 151 were employed in this study’s simulation cases with a numerical 9 of 22
stability of 0.01 for the reflectance calculations.
The convergence test also inspected the adequate number of columns representing
the rough surfaces, as shown in Figure 6b. The number of columns was investigated from
510nm,
to 220 h ofthe
(b) for 20 average cl of 5 nm, intensity
nm and plasmonic (c) h of 2 of
nmthe noclBSA
and of 50 nm, and
model, BSA(d) h of 20 the
without nmside-
and
cl of 50 nm, as shown in Figure 6a. The test was performed using the rigorous
wall model, and BSA with the sidewall model. The tested roughness model was con- coupled-
wave analysis to calculate the p-polarized HeNe laser’s optical reflectance with multiple
structed based on the same four extreme cases. The plasmonic intensity’s differences be-
diffraction orders ranging from 1 to 171. The numerical fluctuation between the 149 and
tween the 200 and 220 columns for (a) h of 2 nm and cl of 5 nm, (b) h of 20 nm and cl of 5
151 diffraction orders was 7 × 104 , 0.002, 0.020, and 0.003 for the four tested surfaces.
nm, (c) h of 2 nm and cl of 50 nm, and (d) h of 20 nm and cl of 50 nm was 7 × 104, 0.002,
Employing a higher number of diffraction orders will surely increase the accuracy of the
0.018, and 0.005 for the proposed three models. The following paragraphs will discuss
results; however, it also requires longer computing time and more resources. Therefore, the
how this roughness level can enhance protein binding. The analysis demonstrates that 200
diffraction orders of 151 were employed in this study’s simulation cases with a numerical
columns are enough to simulate the roughness frameworks.
stability of 0.01 for the reflectance calculations.

Figure 6.
Figure 6. (a)
(a) The
The convergence
convergencetest
testfor
forthe
thefour
fourextreme
extremecases
casesofofthe
therough
roughsurfaces
surfacesconstructed
constructed with
with h
of 2 nm and cl of 5 nm (solid blue line), h of 2 nm and cl of 50 nm (dashed blue line), h of 20 nm andnm
h of 2 nm and cl of 5 nm (solid blue line), h of 2 nm and cl of 50 nm (dashed blue line), h of 20 cl
and cl of 5 nm (solid red line), and h of 20 nm and cl of 50 nm (dashed red line), tested with the
of 5 nm (solid red line), and h of 20 nm and cl of 50 nm (dashed red line), tested with the number of
number of diffractions orders up to 171, and (b) The convergence test on the simulated columns for
diffractions orders up to 171, and (b) The convergence test on the simulated columns for the rough
the rough surfaces at the same four extreme cases, illustrated as average plasmonic intensities of the
surfaces at the same
no BSA model, four extreme
the nonsidewall BSAcases, illustrated
model, and theas averageBSA
sidewall plasmonic
model. intensities of the no BSA
model, the nonsidewall BSA model, and the sidewall BSA model.
3.2. Comparison between the Proposed RCWA Simulation and Monte Carlo-Based Method and
The convergence test also inspected the adequate number of columns representing the
Reported Experimental Results in the Literature
rough surfaces, as shown in Figure 6b. The number of columns was investigated from 10 to
220 forOnethecrucial
averagequestion that we
plasmonic will discuss
intensity of theinnothis
BSAsection is how
model, BSA accurate or how
without the inac-
sidewall
curate the
model, andproposed
BSA withRCWA simulation
the sidewall andThe
model. Monte Carlo-based
tested roughnessmethod
model canwasbe compared
constructed
to the on
based experimental
the same four results. Yangcases.
extreme et al.The
[56]plasmonic
reported experimental SPR reflectance
intensity’s differences between meas-
the
200 and 220 columns for (a) h of 2 nm and cl of 5 nm, (b) h of 20 nm and cl of 5 nm, (c) hnm
urements for different roughness levels for 632.8 nm incident wavelength and 57.5 of
2gold
nm films.
and cl These goldand
of 50 nm, films
(d)were annealed
h of 20 nm and using
cl of 50thermal 7 × 104 , 0.002,
nm wasannealing at four different
0.018, tem-
and 0.005
peratures
for at 100 °C
the proposed to 400
three °C forThe
models. 15 min, allowing
following them to will
paragraphs varydiscuss
the roughness
how thisofroughness
the films.
Theircan
level roughness
enhanceparameters were The
protein binding. measured using
analysis an atomicthat
demonstrates force200
microscope
columns are(AFM).
enough
to simulate the roughness frameworks.

3.2. Comparison between the Proposed RCWA Simulation and Monte Carlo-Based Method and
Reported Experimental Results in the Literature
One crucial question that we will discuss in this section is how accurate or how inac-
curate the proposed RCWA simulation and Monte Carlo-based method can be compared
to the experimental results. Yang et al. [56] reported experimental SPR reflectance measure-
ments for different roughness levels for 632.8 nm incident wavelength and 57.5 nm gold
films. These gold films were annealed using thermal annealing at four different tempera-
tures at 100 ◦ C to 400 ◦ C for 15 min, allowing them to vary the roughness of the films. Their
roughness parameters were measured using an atomic force microscope (AFM).
Figure 7 shows SPR reflectance spectra calculated using the proposed procedure
using the roughness levels and the roughness heights reported by Yang et al. [56]. Table 2
describes the roughness levels, including the root-mean-square (RMS) roughness and the
roughness heights, experimental plasmonic angles, and the plasmonic angles calculated
Figure 7 shows SPR reflectance spectra calculated using the proposed procedure us-
ing the roughness levels and the roughness heights reported by Yang et al. [56]. Table 2
Sensors 2023, 23, 3377 describes the roughness levels, including the root-mean-square (RMS) roughness and 10 of
the22
roughness heights, experimental plasmonic angles, and the plasmonic angles calculated
from Figure 7. The plasmonic angles of the experimental results and the simulated SPR
dips
fromagree
Figurewell, withplasmonic
7. The slight discrepancies within
angles of the 0.09 degrees.
experimental Theand
results slight
theerror shownSPR
simulated in
Table 2 could
dips agree be with
well, due to the gold
slight used in the
discrepancies simulation
within may differ
0.09 degrees. The from
slightthe actual
error gold
shown in
properties in the
Table 2 could beexperiment.
due to the gold used in the simulation may differ from the actual gold
properties in the experiment.

Figure
Figure7.7. SPR
SPRreflectance
reflectancespectra
spectrasimulated
simulatedusing
usingthe
theproposed
proposedprocedure
procedureat
atdifferent
differentroughness
roughness
parameters reported by Yang et al. [56].
parameters reported by Yang et al. [56].

Table
Table 2.
2. Shows
Shows the
the roughness
roughness parameters
parameters and
and plasmonic
plasmonicangles
angles from
from the
the experimental
experimental results
results
reported
reportedby
byYang
Yang et
et al.
al. [56]
[56]and
and the
the plasmonic
plasmonic angles
angles from
from Figure
Figure 7.
7.
Thermal Annealing RMS of Roughness Roughness Experimental θsp Simulated θsp Using the Proposed Method
Simulated θsp Using
Thermal Annealing
Temoerature in °C RMS
(nm)of Roughness Roughness (Degrees)
Height (nm) Height Experimental θsp (Degrees)
◦ the Proposed Method
Temoerature
100 °C in C 1.260(nm) 6.1 (nm) 44.04 (Degrees) 44.04 (Degrees)
200 °C◦ 0.906 3.7 43.94 43.93
100 C 1.260 6.1 44.04 44.04
300 °C◦ 0.700 2.9 43.75 43.70
200 C 0.906 3.7 43.94 43.93
400
300°C◦C 0.4150.700 1.5 2.9 43.65 43.75 43.56 43.70
400 ◦ C 0.415 1.5 43.65 43.56
3.3. Quantitative Performance Parameters of SPR Sensors at Different Roughness Levels
Ideally smoothPerformance
3.3. Quantitative gold SPR-based sensor of SPR Sensors at Different Roughness Levels
Parameters
All quantitative performance
Ideally smooth gold SPR-based sensor parameters were initially calculated based on Equa-
tions (1) to (5) after the RCWA on
All quantitative performance parameters the ideally smooth gold layer
were initially for thebased
calculated SPR-based sensor
on Equations
was simulated. As a result, the computed S, plasmonic dip position
(1) to (5) after the RCWA on the ideally smooth gold layer for the SPR-based sensor was (n0 sinθ sp ), FWHM, ΔI,
Isimulated.
sp, FoM1, andAs FoM 2 were the
a result, 122.59 rad/µm S,
computed
2 , 1.440, 0.41 rad∙RIU/μm,
plasmonic dip position0.90, 0.007,
(n0 sinθ 299.00, and
sp ), FWHM, ∆I,
980.66,
Isp , FoM respectively.
1 , and FoM 2
These
were are the
122.59 theoretical
rad/µm limits
2 , 1.440, of
0.41 the
rad quantitative
· RIU/µm, performance
0.90, 0.007, pa-
299.00,
rameters of an ideally uniform SPR gold sensor calculated from Figure
and 980.66, respectively. These are the theoretical limits of the quantitative performance 5a.
Rough gold
parameters of ansurfaces
ideally uniform SPR gold sensor calculated from Figure 5a.
Figure
Rough 8a,b illustrate the S of the SPR on a nonsidewall BSA model and a sidewall
gold surfaces
BSA modelFigureto different
8a,b illustrateroughness
the S of using
the SPR RCWA
on a and Monte Carlo
nonsidewall BSA simulation
model and computed
a sidewall
using
BSA modelEquation (1). Furthermore,
to different roughness theusing
increment
RCWA of and
the BSA
Montebinding
Carlodomain,
simulation in percent,
computed in
the
usingsidewall
Equation model(1).compared
Furthermore,to thetheBSA-on-top
increment of model at different
the BSA bindingroughness
domain, inwas shown
percent, in
in
theFigure 8c. Two
sidewall model gratings’
compared models
to thewith equivalentmodel
BSA-on-top h andatcl were generated
different roughnessusing theshown
was same
principle
in Figureto8c. calculate the protein
Two gratings’ increment.
models The additional
with equivalent h andsidewall
cl wereprotein
generated can using
be calcu-
the
lated
sameby subtracting
principle the nonsidewall
to calculate the protein model from theThe
increment. sidewall model.sidewall
additional Later, the percent-
protein can
be calculated
age by subtracting
of protein increment thedetermined
can be nonsidewall bymodel
dividingfrom
thethe sidewall
sidewall model.
protein by Later, the
the non-
percentage
sidewall of protein
model’s proteinincrement
amount.canAccording
be determined to thebycontour
dividingrepresentation,
the sidewall protein by the
the protein
nonsidewall model’s protein amount. According to the contour representation, the protein
difference remained negligible at the low-to-moderately rough surface. However, the
increased amount of protein due to the roughness significantly increased as the roughness
height, h, escalated to over 10 nm for a short correlation length, cl.
Sensors 2023, 23, x FOR PEER REVIEW 11 of 23

difference remained negligible at the low-to-moderately rough surface. However, the in-
Sensors 2023, 23, 3377 11 of 22
creased amount of protein due to the roughness significantly increased as the roughness
height, h, escalated to over 10 nm for a short correlation length, cl.

Figure 8. S of the (a) nonsidewall BSA model, (b) sidewall BSA model at different roughness levels
Figure 8. S2 of the (a) nonsidewall BSA model, (b) sidewall BSA model at different roughness levels
in rad/µm , and (c) the increase of the protein density due to sidewalls in percent as the surface
in rad/µm2 , and (c) the increase of the protein density due to sidewalls in percent as the surface
roughness increased.
roughness increased.
The S can be classified into four major regions:
The S can be classified into four major regions:
(1) The “no sidewall protein enhancement” corresponds to the region where the letters
(1) The “no sidewall protein enhancement” corresponds to the region where the letters ‘a’
‘a’ and ‘b’ are in Figure 8. This area indicated the roughness levels in which the side-
and ‘b’ are in Figure 8. This area indicated the roughness levels in which the sidewall
wall BSA was negligible. The protein lying on top of the rough surfaces overlapped
BSA was negligible. The protein lying on top of the rough surfaces overlapped the
the sidewall protein. Moreover, two types of SPR results were investigated in this
sidewall protein. Moreover, two types of SPR results were investigated in this region:
region: the negative-sensitivity LSPR and the degraded-sensitivity reflectance spec-
the negative-sensitivity LSPR and the degraded-sensitivity reflectance spectra, as
tra, as illustrated as the operating point ‘a’ and ‘b’, respectively. The negative plas-
illustrated as the operating point ‘a’ and ‘b’, respectively. The negative plasmonic
monic dip shifting means the plasmonic angle shifts towards a lower plasmonic res-
dip shifting means the plasmonic angle shifts towards a lower plasmonic resonance
onance angle when the sample refractive index increases. It is established that this is
angle when the sample refractive index increases. It is established that this is due to
due to the adverse diffraction orders of gratings or scattering surfaces [57]. The re-
the adverse diffraction orders of gratings or scattering surfaces [57]. The reflectance
flectance spectra at the operating point ‘a’ (h and cl of 3 nm and 10 nm, respectively),
spectra at the operating point ‘a’ (h and cl of 3 nm and 10 nm, respectively), as shown
as shown in Figure 9a, had an S of −379.88 rad/μm2; approximately threefold of the
in Figure 9a, had an S of −379.88 rad/µm2 ; approximately threefold of the magnitude
magnitude of thefrom
of the S acquired S acquired
the ideal from the ideal
uniform uniform
sensor. Withinsensor.
the sameWithin
region thebut
same region
at rougher
but at rougher surfaces, with h of 10 nm and cl of 30 nm
surfaces, with h of 10 nm and cl of 30 nm (operating point ‘b’), the S returned to(operating point ‘b’), the
theS
returned to the positive value but degraded to
2 85.31 rad/μm
positive value but degraded to 85.31 rad/µm , indicating a 30.41% decrease in the
2, indicating a 30.41%

decrease
sensitivity. inThe
the averaged
sensitivity. The averaged
reflectance spectrareflectance
calculated from spectra100calculated from 100
structure profiles at
structure profileslevels
these roughness at these
areroughness levels are
shown in Figure 9b; shown in Figure 9b;
(2) The
(2) The “positive-sensitivity
“positive-sensitivityLSPR” LSPR”only onlydescribed
described thetheS affected
S affected by bythetheLSPR.
LSPR.Here, the
Here,
additional protein was presented but had an insignificant
the additional protein was presented but had an insignificant effect on the S. The effect on the S. The reflec-
tance curvescurves
reflectance spectra at the at
spectra operating point ‘c’
the operating pointobtained at h of at
‘c’ obtained 12hnm and
of 12 nmcl and
of 20clnmof
indicated the enhanced
20 nm indicated S due Stodue
the enhanced thetoLSPR wavewave
the LSPR effecteffect
only.only.
Figure 9c illustrates
Figure the
9c illustrates
reflectance
the reflectancespectra at this
spectra location.
at this In addition,
location. In addition, thethe
twotwo employed
employed models’
models’ curves
curvesin
this roughness
in this roughness level diddid
level notnot
look significantly
look significantly different
differentduedueto the
to the small
smallincrement
increment of
the BSA
of the amount;
BSA amount;
(3) The “LSPR and
(3) The “LSPR and protein”
protein” was investigated as the roughness rose. In this region, S of
the
the SPR
SPR spectra
spectracan canimprove
improvedue duetotothethelocalized
localized LSPR
LSPR effect
effectand andthethe
presence
presence of an
of
additional
an additional protein. The
protein. Themodels
models indicate
indicate a asignificant
significantdifference
differencein insensitivity
sensitivity values
at
at the
the operating
operating point
point ‘d’,
‘d’, located
located at h of 9 nm and cl of of 88 nm.
nm. For
For thethe non-sidewall
non-sidewall
BSA
BSA model, the S of 1191.13 rad/µm rad/μm22 was was investigated.
investigated. In In contrast,
contrast, the BSA sidewall
model
model achieved a significantly higher SS of
significantly higher of 1608.02
1608.02rad/µm rad/μm22,, having
having a 35.00% increase
from
from thethenonsidewall
nonsidewallBSA BSAmodel,
model,asasshownshown ininFigure
Figure 9d.9d.
TheThebinding sensitivity
binding sensitivityen-
hancement
enhancement due to the
due LSPR
to the LSPRwas was9.72 times
9.72 timesat atthis roughness
this roughness level.
level. In contrast, the
BSA
In sidewallthe
contrast, model
BSAenhanced
sidewall the modelS byenhanced
13.12 times themore
S bythan
13.12thetimes
ideally smooth
more thangold
the
sensor, smooth
ideally indicating gold the S enhancement
thatsensor, indicating that due thetoSthe increased protein
enhancement due towas the 3.68 times.
increased
In addition,
protein was 3.68thistimes.
crucially high S was
In addition, thisdue to thehigh
crucially deterioration
S was due of a plasmonic
to the deterioration dip
structure, resulting in a lower n0 sinθ sp for the nonprotein coated model, which will
be explained in the next part. The S enhancement factor due to additional protein
agreed with the estimated protein concentration at the roughness level, as shown in
Figure 8c;
Sensors 2023, 23, x FOR PEER REVIEW 12 of 23

of a plasmonic dip structure, resulting in a lower n0sinθsp for the nonprotein coated
Sensors 2023, 23, 3377 model, which will be explained in the next part. The S enhancement factor due 12 of to
22
additional protein agreed with the estimated protein concentration at the roughness
level, as shown in Figure 8c;
(4) At
Atthe
theextreme
extremeroughness
roughness levels (near
levels the the
(near bottom
bottomrightright
corner of theof
corner contour in Figure
the contour in
8), the plasmonic
Figure dip structure
8), the plasmonic deteriorated,
dip structure as indicated
deteriorated, by the “No
as indicated SPR”
by the “No region,
SPR”
due to the
region, due scattering loss [58]loss
to the scattering of the
[58]rough
of thesurface. The propagation
rough surface. length oflength
The propagation surfaceof
surface plasmon
plasmon polaritons polaritons
is stronglyis strongly
distorteddistorted due to roughness
due to roughness [59], resulting
[59], resulting in an unde- in
an undetectable
tectable region, as region, as indicated
indicated in Figure in9e
Figure 9e operating
for the for the operating
point ‘e’point ‘e’ (obtained
(obtained at h of
at h of 16 nm and cl of 15 nm). The plasmonic dips have virtually
16 nm and cl of 15 nm). The plasmonic dips have virtually no intensity contrast. The no intensity contrast.
The SPR
SPR reflectance
reflectance dipsdips weakly
weakly maintained
maintained their
their structureand
structure anddistorted
distortedatatthe
the higher
higher
roughness. The
roughness. The increase
increase in in protein
protein quantity
quantity due to roughness
roughness was not the only reason
for the
for improved SS but
the improved but also
also the
the enhancement
enhancement of of the
the LSPR
LSPR from
from the
the roughness
roughness peaks.
peaks.
There is
There is aa substantial
substantial tradeoff
tradeoff between
between the the roughness
roughness level,
level, the sensitivity, and the
scattering loss
scattering loss of
of the
the SPR
SPR dip.
dip.

Figure 9. 9. (a)
(a)the
thereflectance
reflectancespectra
spectra with a negative
with sensitivity
a negative at the
sensitivity atoperating point point
the operating ‘a’, (b)‘a’,
degraded-
(b) de-
graded-sensitivity
sensitivity reflectance
reflectance curves atcurves at the operating
the operating point ‘b’, point ‘b’, (c) enhanced-sensitivity
(c) enhanced-sensitivity reflectancereflectance
spectra at
spectra
the at the operating
operating point
point ‘c’, (d) ‘c’, (d)
highly highly enhanced-sensitivity
enhanced-sensitivity reflectancereflectance
spectra atspectra at the operating
the operating point ‘d’,
point(e)‘d’,
and theand (e) the reflectance
reflectance curves at thecurves
highly at rough
the highly rough
surface wheresurface where thecannot
the sensitivity sensitivity cannot be
be computed at
computed at the operating point ‘e’. Note that the dotted curves indicate the plasmonic
the operating point ‘e’. Note that the dotted curves indicate the plasmonic angles of the ideal uniform angles of
the ideal uniform gold cases and the black arrows indicate the plasmonic angle shift direction.
gold cases and the black arrows indicate the plasmonic angle shift direction.

Figure 10a
10a illustrates
illustratesthetheaverage
averagenn 0sinθsp of the SPR dip positions calculated from the
Figure 0 sinθ sp of the SPR dip positions calculated from
BSA without the sidewall model, the
the BSA without the sidewall model, the BSA BSA withwith
the sidewall model,
the sidewall and the
model, andwithout BSA
the without
coating model. The parameter can be used to identify the sensor’s
BSA coating model. The parameter can be used to identify the sensor’s detection range. detection range. An
increased surface roughness could result in a higher plasmonic angle
An increased surface roughness could result in a higher plasmonic angle and a shorter and a shorter detec-
tion range.range.
detection Moreover,
Moreover,a higher n0sinθ
a higher results
n0spsinθ in a bigger numerical aperture (NA) re-
sp results in a bigger numerical aperture (NA)
quirement ofofthe
requirement theillumination
illuminationlens
lensfor
forangular
angularscanning
scanningapplications.
applications. Figure
Figure 10b10b compares
compares
the reflectance
the reflectance curves
curves from
from the
the model
model without
without the the BSA
BSA coating
coating and
and their
their plasmonic
plasmonic dip dip
position at hh of
of 00 nm
nm(ideal
(idealsmooth),
smooth),55nm, nm,1010nm, nm,and
and 1515
nm,
nm,and thethe
and equivalent
equivalent cl of
cl 20
of
nm.nm.
20 As As
the the
theoretical
theoreticaluniform surface
uniform became
surface rougher
became to 10to
rougher nm,
10 the
nm,n0the
sinθnsp0 sinθ
shifted
sp grad-
shifted
ually. At this
gradually. At cl, thecl,shifting
this distances
the shifting fromfrom
distances h of 0hnmof 0tonm
5 nm
to 5and
nmhand
of 5hnmof 5tonm 10 nm
to 10were
nm
were very similar at approximately 0.05. However, when the h is increased to 15 nm, the
n0 sinθ sp moves slightly backward due to the deterioration of the plasmonic structure due
to surface roughness.
Sensors 2023, 23, x FOR PEER REVIEW 13 of 23

very similar at approximately 0.05. However, when the h is increased to 15 nm, the n0sinθsp
Sensors 2023, 23, 3377 13 of 22
moves slightly backward due to the deterioration of the plasmonic structure due to sur-
face roughness.

Figure 10. (a)

Figure 10. (a) the
the average
averageplasmonic
plasmonicdip
diplocation,
location,
n0nsinθ
0sinθsp, for the three models at different rough-
sp , for the three models at different roughness
ness levels, and (b) the reflectance spectra from the
levels, and (b) the reflectance spectra from the nonprotein nonprotein coated at
coated model model
h of 0at
nmh (ideal
of 0 nm (ideal
smooth),
smooth), 5 nm, 10 nm, and 15 nm, and the equivalent cl of 20 nm. (c) The FWHM of the SPR detection
5 nm, 10 nm, and 15 nm, and the equivalent cl of 20 nm. (c) The FWHM of the SPR detection at
at different degrees of roughness and (d) reflectance spectra for the rough gold surface of h and cl
different degrees of roughness and (d) reflectance spectra for the rough gold surface of h and cl of
of 18 nm and 40 nm, respectively, compared to the uniform gold surface. Note that the black arrows
18 nmindicate
in (e) and 40 nm, respectively,
the plasmonic compared
angle to the uniform
shift direction when the gold surface. Note that the black arrows in
h increased
(b) indicate the plasmonic angle shift direction when the h increased.
Figure 10c shows the FWHM of the reflectance spectra at the simulated rough sur-
Figure 10c shows the FWHM of the reflectance spectra at the simulated rough surfaces.
faces. The parameter was computed using Equation (3). The FWHM results for the side-
The parameter was computed using Equation (3). The FWHM results for the sidewall and
wall and no sidewall cases were similar; therefore, only the FWHM obtained from the non-
no sidewall cases were similar; therefore, only the FWHM obtained from the non-sidewall
sidewall BSA model was presented to shorten the manuscript. The growth of roughness
BSA model was presented to shorten the manuscript. The growth of roughness strongly
strongly
broadensbroadens
the size ofthe thesize of the reflectance
reflectance spectra. Atspectra.
slightlyAt slightly
rough rough
surfaces withsurfaces with than
an h lower an h
lower thanFWHM
5 nm, the 5 nm, the wasFWHM
affected was
by affected
h only. Forby hinstance,
only. For instance,point
operating operating
‘a’ had point ‘a’ had
an FWHM
an FWHM of 0.58 rad∙RIU/μm, 41.46% higher than the
of 0.58 rad·RIU/µm, 41.46% higher than the ideal uniform sensor’s FWHM. However,ideal uniform sensor’s FWHM.
However,
when h waswhen h was
greater thangreater
5 nm,than 5 nm, cl significantly
cl significantly affected the affected the alternation
alternation in FWHM.inAt FWHM.
rough
At rough surfaces, with a cl less than 35 nm, the reflectance spectra may
surfaces, with a cl less than 35 nm, the reflectance spectra may have a deteriorated structure, have a deterio-
rated structure,
reducing ∆I. Thereducing
cruciallyΔI. lowThe∆Icrucially
can result lowin ΔI can result
a slightly in aFWHM
lower slightlythan lowertheFWHM
region
than the
with region
a larger cl with a larger
(smoother cl (smoother
surface). Thesesurface).
changes Thesein ∆I willchanges in ΔI willfurther
be discussed be discussed
in the
further
followingin the followingInparagraphs.
paragraphs. In addition,
addition, operating pointsoperating points
‘b’, ‘c’, and ‘d’ ‘b’,
were ‘c’,inand
this‘d’ were in
region.
this region.
On the other hand, at the operating point ‘e’, there was a crucial loss in the SPR
On the
structure, other hand,
resulting in anatunidentified
the operating point ‘e’,
FWHM. there was
However, theaintensity
crucial loss
depth in the
Isp SPR
of thestruc-
SPR
ture, resulting in an unidentified FWHM. However, the intensity
dips considerably deteriorated due to the scattering loss compared to the smooth sensor depth I sp of the SPR dips
considerably
surface. Apartdeteriorated due to theregions,
from the mentioned scatteringit canloss
becompared
increased to upthe smooth
to 0.83 sensor sur-
rad·RIU/µm at
face. Apart
an h and a clfrom
of 18the
nmmentioned
and 30 nm, regions, it can be
respectively, increased
which up to 0.83 rad∙RIU/μm
was approximately two timesatwider an h
than the FWHM of the uniform gold SPR detection, as shown in Figure 10d. The rougher
surface leads to stronger scattering and a broader SPR dip.
Figure 11a,b indicate the average ∆I and Isp at various roughness levels. The results
obtained from the BSA without the sidewall model and the BSA with the sidewall model
Sensors 2023, 23, x FOR PEER REVIEW 14 of 23

and a cl of 18 nm and 30 nm, respectively, which was approximately two times wider than
Sensors 2023, 23, 3377 the FWHM of the uniform gold SPR detection, as shown in Figure 10d. The14rougher of 22 sur-
face leads to stronger scattering and a broader SPR dip.
Figure 11a,b indicate the average ΔI and Isp at various roughness levels. The results
obtained from the BSA without the sidewall model and the BSA with the sidewall model
were slightly distinct, with a maximum difference of approximately 0.1 for both parameters,
were slightly distinct, with a maximum difference of approximately 0.1 for both parame-
as shown in Figure
ters, as shownTherefore,
11a,b. onlyTherefore,
in Figure 11a,b. the average only values were
the average reported
values to shorten
were reported to shorten
the manuscript.theInmanuscript.
addition, Inthis indicated that the presence of a sidewall
addition, this indicated that the presence of a sidewall proteinprotein does does not
not significantly affect theaffect
significantly change in ∆I and
the change in ΔIIand
sp . However,
Isp. However, the
thesurface roughnessstill
surface roughness still
substan-
substantially impacted the sensor’s
tially impacted output.
the sensor’s output.AtAtananoperating
operating point point‘d’,
‘d’,forfor instance,
instance, the
the average ΔI
average ∆I waswas 0.089, which
0.089, whichwas
was10.08 timesless
10.08 times less intensity
intensity contrast
contrast than than ∆I achieved
ΔI achieved by
by the theoretical
the theoretical uniform
uniformgold sensor.
gold Furthermore,
sensor. the average
Furthermore, the average Isp crucially rose to 0.74rose
Isp crucially (105.71 times com-
to 0.74
pared to the ideally smooth gold sensor’s
(105.71 times compared to the ideally smooth gold sensor’s Isp ). I sp ).

Figure 11. (a) theFigure 11. ∆I

average (a) and
the average
(b) the ΔI and (b)Isp
average theofaverage Isp nonsidewall
both the of both the nonsidewall
BSA and BSA and the sidewall
the sidewall
BSA models.
BSA models.
The reported
The reported quantitative quantitativeparameters
performance performancewereparameters
combined were
to combined
compute the to compute
FoM, the
calculated based on Equation (5), at different roughness levels for BSA with no sidewallno side-
FoM, calculated based on Equation (5), at different roughness levels for BSA with
model and BSAwall withmodel and BSA
a sidewall with aas
model, sidewall
shownmodel,
in Figureas shown
12a,b,inrespectively.
Figure 12a,b,Figure
respectively.
12c,d Figure
12c,d show FoM2 calculated using the intensity-based FoM expressed in Equation (6) [55].
show FoM2 calculated using the intensity-based FoM expressed in Equation (6) [55]. The
Sensors 2023, 23, x FOR PEER
protein REVIEWThe protein
enhancement
enhancement due to the roughness neither distorts nor enhances
due to the roughness neither distorts nor enhances the LSPR 15 ofinthe LSPR in
23 the
the range of studied rough surfaces.
range of studied rough surfaces.

Figure 12. (a) theFigure

FoM112.
of(a)
thetheBSA
FoMwithout
1 of the BSA without the sidewall, (b) BSA with the sidewall model; (c) FoM2
the sidewall, (b) BSA with the sidewall model; (c) FoM2 for
for SPR detection [55] obtained from the BSA without the sidewall, and (d) BSA with the sidewall
SPR detection [55] obtained
model. from the BSA without the sidewall, and (d) BSA with the sidewall model.

The FoM1 in Equation (5) depends mostly on how far the SPR dip moves S, not the
optical intensity. Moreover, the results can be interpreted as the degradation and enhance-
ment trends due to LSPR relying on surface roughness. In contrast, the additional protein
can enhance the performance parameters, including FoM, at the expense of a more de-
manding NA requirement. The sensor achieved its maximum FoM1 and FoM2 of 1777.81
and 2400.02 at an h of 9 nm and a cl of 8 nm (operating point ‘d’) for the nonsidewall BSA
Sensors 2023, 23, 3377 15 of 22

The FoM1 in Equation (5) depends mostly on how far the SPR dip moves S, not
the optical intensity. Moreover, the results can be interpreted as the degradation and
enhancement trends due to LSPR relying on surface roughness. In contrast, the additional
protein can enhance the performance parameters, including FoM, at the expense of a more
demanding NA requirement. The sensor achieved its maximum FoM1 and FoM2 of 1777.81
and 2400.02 at an h of 9 nm and a cl of 8 nm (operating point ‘d’) for the nonsidewall
BSA and sidewall BSA models, respectively. The reflectance spectra at this point had
a significantly increased S and a slightly dropped FWHM due to both the LSPR effect
and the protein increment, resulting in the crucially high FoM1 . In addition, a negative
FoMs region resulted from the negative S values. The FoMs in some of these reported
operating positions could surpass the sensing capability of the ideal uniform gold sensor.
For instance, the operating point ‘a’ had an FoM1 of −654.97, about two times higher
than the theoretical smooth gold sensor’s FoM1 . Other operating points’ FoMs and other
performance parameters are indicated in Table 3.

Table 3. The quantitative performance parameters comparison between the surface plasmon reso-
nance detection employed with a uniform and rough gold layer.

h = 3 nm, h = 10 nm, h = 12 nm, h = 9 nm, h = 16 nm,

Quantitative Ideal
cl = 10 nm cl = 30 nm cl = 20 nm cl = 8 nm cl = 15 nm
Performance Uniform
(Operating Point (Operating Point (Operating Point (Operating Point (Operating Point
Parameters Surface
‘a’) ‘b’) ‘c’) ‘d’) ‘e’)
No Side- No Side- No Side- No Side- No Side-
Sidewall Sidewall Sidewall Sidewall Sidewall
wall wall wall wall wall
BSA BSA BSA BSA BSA
BSA BSA BSA BSA BSA
S
122.59 −379.88 −379.88 85.31 85.31 996.11 1280.00 1191.13 1608.02 - -
(rad/µm2 )
n0 sinθ sp
1.440 1.481 1.481 1.506 1.506 1.495 1.496 1.486 1.490 1.466 * 1.473 *
(unitless)
FWHM
(rad·RIU 0.41 0.58 0.58 0.79 0.79 0.66 0.65 0.67 0.67 0.77 0.75
/µm)
∆I
0.90 0.88 0.88 0.47 0.47 0.13 0.15 0.10 0.14 0.001 0.003
(unitless)
Isp
0.007 0.018 0.018 0.41 0.41 0.72 0.70 0.69 0.60 0.75 0.69
(unitless)
FoM1
299.00 −654.97 −654.97 107.98 107.98 1509.26 1969.23 1777.80 2400.02 - -
(unitless)
FoM2
980.66 −1677.42 −1677.42 92.52 92.52 590.75 833.81 616.84 1020.33 - -
(unitless)
* The n0 sinθ sp of the operating point ‘e’ can still be determined. However, the reflectance spectrum of the model
without a protein coating lost its plasmonic dip structure, resulting in an unidentified S and FoM.

The SPR detection’s FoM2 , calculated involving S, FWHM, ∆I, and Isp , had similar
results, as illustrated in Figure 12c,d for the BSA without the sidewall and BSA with the
sidewall model, respectively. The highest FoM2 for the SPR detection was at the same
operating point as the FoM1 , with an FoM2 of 616.84 for the nonsidewall BSA model and
1020.33 for the sidewall BSA model, respectively. Therefore, the surface roughness can
slightly improve the overall sensing performance in the intensity detection of the SPR-based
sensor for the binding kinetic measurement.
Table 3 compares the seven investigated performance parameters obtained from the
five operating points. It is seen that the surface roughness can result in an enhancement
(operating points ‘c’ and ‘d’), a degradation (operating point ‘b’), or a complete deterioration
(operating point ‘e’) of the SPR detection due to both the LSPR effect and the additional
protein. The S can be significantly increased with highly rough surfaces in exchange for
the substantial deterioration in the application relying on intensity detection. There is
a tradeoff between the S enhancement due to the increased protein concentration, in other
words, binding density, and the FWHM of the SPR reflectance dip. Despite this limitation,
the negative-sensitivity detection (operating point ‘a’) could also be an efficient alternative
Sensors 2023, 23, 3377 16 of 22

for utilizing the SPR intensity-measurement applications due to its improved sensitivity
and slightly degraded FWHM.
The ER and the increased protein density and surface roughness effects on the S and
FoMs are summarized in Table 4. The ER for the roughness effect, the LSPR effect, and the
protein increment were calculated based on Equations (6) to (8). The only factor affecting
the sensor’s performance for operating points ‘a’ and ‘b’ was the LSPR since the protein
density was negligible, as explained in the earlier section. Note that some ERs for the
operating points ‘b’, ‘c’, and ‘d’ are less than 1.00, indicating that the roughness level
experienced a degradation effect. On the other hand, the additional protein and the LSPR
affected the operating points ‘c’ and ‘d’.
Table 4. Shows the enhancement ratio of S and FoMs compared to an ideal uniform gold case and
their effect in percentage due to the additional protein and wave effect from the surface roughness.

h = 3 nm, h = 10 nm, h = 12 nm, h = 9 nm,

cl = 10 nm cl = 30 nm cl = 20 nm cl = 8 nm
(Operating Point ‘a’) (Operating Point ‘b’) (Operating Point ‘c’) (Operating Point ‘d’)
ERS,roughness −3.10 0.70 10.44 13.12
ERS,LSPR −3.10 (100%) 0.70 (100%) 8.13 (77.87%) 9.72 (74.07%)
ERS,protein 0 (0%) 0 (0%) 2.31 (22.13%) 3.40 (25.93%)
ERFoM1,roughness −2.19 0.36 6.59 8.03
ERFoM1,LSPR −2.19 (100%) 0.36 (100%) 5.05 (76.64%) 5.95 (74.07%)
ERFoM1,protein 0 (0%) 0 (0%) 1.54 (23.36%) 2.08 (25.93%)
ERFoM2,roughness −1.71 0.09 0.85 1.04
ERFoM2,LSPR −1.71 (100%) 0.09 (100%) 0.60 (70.85%) 0.63 (60.45%)
ERFoM2,protein 0 (0%) 0 (0%) 0.25 (29.15%) 0.41 (39.55%)

The roughness level of the h of 9 nm and the cl of 8 nm, corresponding to the high-
est FoM1 and FoM2 responses discussed earlier, was further analyzed to validate their
sensing response. One crucial performance parameter for protein kinetics is to check
whether the sensor has a linear responsivity for different protein concentrations, such as
the thickness deposited on the sensor surface. Figure 13 compares the linear responsitivity
of the three models for the h of 9 nm and the cl of 8 nm roughness sensor computed in
this study for the different deposited thicknesses of BSA from 0 nm (bare gold) to 10 nm.
The three models had linear responsivity covering the protein thickness, with R-square
values for the linear equation fitting of 0.9985, 0.9990, and 0.9990 for the ideal smooth
gold surface, the rough surface with no sidewall protein coating, and the rough surface
with a sidewall protein coating, respectively. The slopes (∆sinθ sp /dBSA ) of the three linear
lines were 8.13 × 10−4 , 1.69 × 10−3 , and 2.01 × 10−3 , for the three models, which can be
converted to the S in Equation (1) and it was found that the S values were 122.59 rad/µm2 ,
1202.43 rad/µm2 , and 1595.30 rad/µm2 for the uniform gold surface, the BSA without the
sidewall model, and the BSA with the sidewall model for the 5-nm thick BSA layer case.
The S values agree with the S discussed in Table 3, indicating that the rough surface can
provide a linear responsivity covering 0 nm to 10 nm. Having mentioned that the uniform
sidewall protein coating assumption may not be accurate, Figure 13 provides a feasible
approach to predicting the expected S if the protein does not fully cover the sidewall. The
partial sidewall coating’s S will be between the no sidewall coating case (red line) and the
uniform sidewall coating (yellow line).
responsivity covering 0 nm to 10 nm. Having mentioned that the uniform sidewall protein
coating assumption may not be accurate, Figure 13 provides a feasible approach to pre-
dicting the expected S if the protein does not fully cover the sidewall. The partial sidewal
coating’s S will be between the no sidewall coating case (red line) and the uniform side-
Sensors 2023, 23, 3377 17 of 22
wall coating (yellow line).

Figure13.
Figure The∆sinθ
13.The Δsinθ for
sp sp for different
different deposited
deposited thicknesses
thicknesses of BSAoffrom
BSA0from 0 nmgold)
nm (bare (bareto gold)
10 nmtofor10 nm for
idealsmooth
ideal smoothgold
gold surface,
surface, rough
rough goldgold surface
surface with with
h of 9hnm,
of 9and
nm,cl and cl offor
of 8 nm 8 the
nmno forsidewall
the no sidewall
and and
sidewallcoating
sidewall coatingcases.
cases.

4. Discussion
4.1. Effect of LSPR and Sidewall Protein from Roughness
Table 4 shows that the wave effect was responsible for most of the impact on sensor
quality. For S and FoM1 , all roughness levels that preserve the SPR spectrum have a percent
impact of wave effect greater than 70%. The FoM2 , which considers the ∆I and Isp , ex-
perienced a greater impact from the sidewall protein. Furthermore, the percentage of S
and FoM1 enhancement of the operating points ‘c’ and ‘d’ in Figure 8c were close to the
percentage of protein increment due to roughness. The calculations using the explained
framework using the RCWA and Monte Carlo simulation agree well with the additional
protein calculated directly from the grating profile. The agreement provides an indepen-
dent measurement of the protein density enhancement due to the roughness and tells us
that the enhancement due to the LSPR and the increased protein are independent. This
investigation implies that increasing the protein amount increases the S and the FoMs at
the same rate.
In this paper, we have explained and provided insight into the detection mechanisms
of rough surfaces. Rough surfaces can enhance the binding domain density, S, and FoM1
with an expense of FWHM and optical intensity contrast of the SPR dip. It is essential to
point out that the advantage of rough surfaces is that they provide only a single SPR dip
or SPR mode from scatterings of surface plasmon polaritons (SPPs). In other words, an
average effect of scattered SPPs and LSPR hotspots at the sharp edges [60]. However, rough
plasmonic surfaces also have drawbacks, including fabrication repeatability, nonlocalized
heat sources due to the localized hotspots at the sharp edges, and nonuniform distribution
of the SPPs, making it difficult to be employed in optical imaging applications. Another
alternative approach to achieve a similar enhancement of the binding domain density,
S, and FoMs is plasmonic grating. Plasmonic gratings allow the SPPs to propagate in
a well-defined path and support the additional sidewall protein. The plasmonic gratings
can support and excite multiple modes due to the diffraction orders leading to a limited
detection range due to crosstalks and overlapping modes [61].

4.2. Intensity and Phase Detection Schemes

Aside from angular interrogation, the SPR measurement can be done using the wave-
length scanning method through intensity and phase detection, as in the angular interro-
gation investigated in this paper. In addition, these schemes are not affected by scanning
wavelengths and angles. Here, the five discussed operation points and the ideal uniform
surface sensor were analyzed for their intensity detection and phase detection performances,
as illustrated in Figure 14. Figure 14a shows that the roughness profiles can generate a more
Sensors 2023, 23, 3377 18 of 22

significant optical intensity contrast than the theoretically smooth plasmonic layer case.
For instance, the operating point ‘b’, which has significantly low sensitivity; on the other
hand, can provide a crucial change in the plasmonic intensity of 0.27. The phase detection
results have an alternative outcome, as shown in Figure 14b. Only the operating point ‘a’
(negative sensitivity SPR) has an outstanding performance in phase detection, while the
other operating positions have degraded their phase responses. According to the analysis,
Sensors 2023, 23, x FOR PEER REVIEW 19 of 23
the negative sensitivity measurement of SPR with a rough surface can be a strong candidate
for binding kinetics measurement.

Figure 14. (a)

Figure 14. (a) Intensity
Intensity detection
detection and
and (b) phase detection
(b) phase detection of
of the
the operating
operating point
point ‘a’,
‘a’, ‘b’,
‘b’, ‘c’,
‘c’, ‘d’, and
‘d’, and
the theoretically uniform surface SPR-based sensor.
the theoretically uniform surface SPR-based sensor.

4.3. Field Plots on Exemplary Optical Structures

Light possesses
possesses both both ananelectric
electricfield
fieldand andaamagnetic
magneticfield, field,and
andthis property
this property of of
electro-
elec-
magnetic waves was used to explain the reason behind
tromagnetic waves was used to explain the reason behind the performance improvement the performance improvement of
thethe
of rough
rough SPR-based
SPR-based sensor.
sensor.ForForTM-polarized
TM-polarizedlight, light,thethepolarization
polarizationcan can excite
excite the SPR
effects, or or the
theelectric
electricfield
fieldwill
willaffect
affect thethe metallic
metallic surface
surfacealong the xthe
along andx zand
axes, whilewhile
z axes, there
are noare
there magnetic
no magneticfieldsfields
in theintwothe axes.
two axes.The total intensity
The total of the
intensity of electric fieldfield
the electric along the
along
x-axis
the (|Ex(|E
x-axis |2 )xand
|2) andz-axis
z-axis |2 )z|provides
(|Ez(|E 2) provides a good
a goodindicator for the
indicator forsummation
the summation of theof LSPR
the
strengths
LSPR acrossacross
strengths the surface.
the surface.
Figure 15 illustrates
Figure 15 illustrates the the electric
electric field
field inin the
the x-direction
x-direction and and z-direction
z-direction for for the
the TM-
TM-
polarization case. Figure 15a–c illustrate the total electric field
polarization case. Figure 15a–c illustrate the total electric field intensity of the SI unit, |Ex|2 intensity of the SI unit,
2 2
+|E|Ex|z|2+, in|Ethree
z | , in three different
different SPR-based SPR-based
sensor sensor
surfaces, surfaces,
includingincluding a theoretical
a theoretical smooth smooth
gold
gold afilm,
film, golda grating
gold grating structure,
structure, and aand rough a rough gold surface,
gold surface, respectively.
respectively. For theFor the uniform
uniform SPR-
SPR-based
based sensor, sensor, the electric
the electric field field intensity
intensity was evenly
was evenly distributed
distributed alongalong the surface
the gold gold surface
with
with a maximum magnitude of 11.28 × 10 3 , as shown in Figure 15a. On the other hand,
a maximum magnitude of 11.28 × 10 , as shown in Figure 15a. On the other hand, the
3
the total
total intensity
intensity of the ofelectric
the electric
field field
for a for a gratings
gratings sensorsensor can reach
can reach up to up to ×16.08
16.08 × 105 ,
105, which
which is a hundred
is a hundred times greater
times greater than the than
field theintensity
field intensity
of the of the smooth
ideal ideal smooth
sensorsensor case.
case. Note
Note that the color scales in Figure 15b,c have been adjusted
that the color scales in Figure 15b,c have been adjusted to clearly show the electric field to clearly show the electric
field intensity
intensity acrossacross the entire
the entire surfacesurface
(their(their
maximum maximum valuesvalues
couldcould
exceed exceed the provided
the provided color
colorlimit).
bar bar limit). The gratings’
The gratings’ height,
height, period,
period, andand fill fill factor
factor of of
2424 nm,nm,500500nm,
nm,andand0.50.5 were
were
selected according to the analyzed rough surface model.
selected according to the analyzed rough surface model.
Moreover, the electric field with the significantly high intensity only appears on the
Moreover, the electric field with the significantly high intensity only appears on the
sharp edges of the gratings, while the intensity in the groove area ranges from 6.00 × 1033
sharp edges of3 the gratings, while the intensity in the groove area ranges from 6.00 × 10
to 30.00 × 10 . The rough surface in the operating point ‘d’ (h and cl of 12 nm) was
to 30.00 × 103. The rough surface in the operating point ‘d’ (h and cl of 12 nm) was chosen
chosen to investigate the magnetic field behavior due to the highest achieved FoM. The
to investigate the magnetic field behavior due to the highest achieved FoM. The maximum
maximum electric field intensity obtained can be varied from approximately 14.92 × 1055 to
electric field intensity obtained can be varied from approximately 14.92 × 10 to
17.51 × 1055 , depending on the randomly simulated surface. Even though the gratings and
17.51 × 10 , depending on the randomly simulated surface. Even though the gratings and
the rough surfaces produced similar electric field intensities, the roughness model provided
the rough surfaces produced similar electric field intensities, the roughness model
significantly higher sharp edges, resulting in more LSPR hotspots and increased sensitivity.
provided significantly higher sharp edges, resulting in more LSPR hotspots and increased
sensitivity.
Sensors 2023, 23,
Sensors 2023, 23, 3377
x FOR PEER REVIEW 19 of
20 of 22
23

Figure 15.
Figure 15. The
The plots
plots of
of the
the total
total electric
electric field
field intensity
intensity in
in the
the SI
SI unit
unit in
in the
the x-direction
x-direction and
and z-direction
z-direction
(|E x|22+ |Ez|2) 2on the gold-based SPR sensor with (a) an ideal smooth surface; (b) gratings structure
(|Ex | + |Ez | ) on the gold-based SPR sensor with (a) an ideal smooth surface; (b) gratings structure
with a gratings’ height of 24 nm, a gratings’ period of 500 nm, and a fill factor of 0.5; and (c) a rough
with a gratings’ height of 24 nm, a gratings’ period of 500 nm, and a fill factor of 0.5; and (c) a rough
sensing surface with h and cl of 9 nm and 8 nm, respectively (operating point ‘d’).
sensing surface with h and cl of 9 nm and 8 nm, respectively (operating point ‘d’).
5. Conclusions
5. Conclusions
The SPR-based
The SPR-based sensor sensor waswas responsive
responsive to to changes
changes in in the
the plasmonic
plasmonic sensor sensor surface.
surface.
Therefore, a slight change in the surface structure, including the roughness, can strongly
affect the
the sensor’s
sensor’sperformance.
performance.This Thismanuscript
manuscript analyzes the effect
analyzes of roughness
the effect of roughness on bind-on
ing kinetic
binding measurement
kinetic measurement for SPRfor detection
SPR detection by employing
by employing the RCWAthe RCWAand the andMonte Carlo
the Monte
simulation
Carlo on the
simulation onconstructed
the constructed roughroughsurface based
surface based ononthe
theGaussian
Gaussianrandomrandom surfaces
method. The simulation structure consisted consisted of two models: the nonsidewall nonsidewall BSA BSA model,
model,
indicating the same amount of protein as the roughness changed, and the sidewall BSA
model, where
where the theprotein
proteinquantity
quantityincreased
increasedwith withthethe roughness.
roughness. The Theconvergence
convergence testtest
on
the the
on number
numberof required
of requireddiffraction orders
diffraction and simulated
orders and simulated columns to represent
columns the rough-
to represent the
roughness
ness and the andSPRthephenomenon
SPR phenomenon was alsowas applied
also applied to obtain
to obtain the optimum
the optimum environment
environment for
for
the the theoretical
theoretical study.
study. TheThe calculated
calculated reflectancespectra
reflectance spectrawere
werethen
then observed.
observed. Then, Then, the
performance factors were observed observed and and calculated, including S,
calculated, including sinθspsp, FWHM, ∆I,
S, nn00sinθ ΔI, IIsp
sp,
and FoMs.
The surface roughness can improve, degrade, or deteriorate deteriorate thethe SPR
SPR sensing
sensing perfor-
perfor-
mance due to the generated LSPR and the additional protein amount. The sensor can have
an exceptionally high S S with highly rough surfaces due to the distorted plasmonic dip
structure,
structure, which
which reduced
reduced the the nn0sinθ
sinθ sp and field
sp and field enhancement
enhancement effect effect from
from the
the surface
surface rough-
rough-
ness. However, utilizing
ness. However, utilizing thethe rough
rough surface
surface shows
shows aa tradeoff
tradeoff between
between the improved S
the improved S and
and
the
the crucial deterioration in the intensity detection capability. One suggested approach for
crucial deterioration in the intensity detection capability. One suggested approach for
the
the SPR
SPR application
application is is to
to utilize
utilize the
the negative-sensitivity
negative-sensitivity SPR SPR detection
detection at at slight
slight roughness
roughness
levels. The reflectance
levels. The reflectance spectra
spectra obtained
obtained from from this
this region
region can increase the
can increase the S S toto threefold
threefold of of
the ideal uniform gold sensor’s S while maintaining the FWHM, ∆I,
the ideal uniform gold sensor’s S while maintaining the FWHM, ΔI, and Ispsp. The operating and I . The operating
points
points for
for the highest FoM
the highest FoM11 and FoM22 enhanced
and FoM enhanced the FoMs by
the FoMs by 8.03
8.03 and
and 1.04
1.04 times
times compared
compared
to
to the sensor with a theoretically smooth surface. In addition, the roughness level could
the sensor with a theoretically smooth surface. In addition, the roughness level could
be
be utilized
utilized inin other
other SPRSPR measurement
measurement schemes, schemes, including
including intensity
intensity andand phase
phase detection.
detection.
Further
Further analysis
analysis indicates
indicates that the sensing
that the sensing improvement
improvement and and degradation
degradation were were primarily
primarily
affected by the LSPR effect of over 70% in all operating cases.
affected by the LSPR effect of over 70% in all operating cases. In conclusion, In conclusion, the knowledge the
of surface roughness
knowledge of surfacecan
be employed
roughness can be
toemployed
enhance the to protein
enhancebinding
sensing
the protein
performance
binding sensing
of the SPR-based sensor.
performance of the SPR-based sensor.
Author Contributions: Conceptualization, S.P. and S.S.; methodology, S.P.; software, S.S. and S.P.;
Author Contributions: Conceptualization, S.P. and S.S.; methodology, S.P.; software, S.S. and S.P.;
validation: S.P., S.S. and T.T.; formal analysis, S.P.; investigation, S.P., S.S. and T.T.; resources, S.P.; data
validation: S.P., S.S. and T.T.; formal analysis, S.P.; investigation, S.P., S.S. and T.T.; resources, S.P.;
curation, S.S.; writing—original draft preparation, S.S.; writing—review and editing, S.P. and T.T.;
data curation, S.S.; writing—original draft preparation, S.S.; writing—review and editing, S.P. and
visualization, S.S.; supervision, S.P. and T.T.; project administration, S.P. and T.T.; funding acquisition,
T.T.; visualization, S.S.; supervision, S.P. and T.T.; project administration, S.P. and T.T.; funding ac-
S.P. and T.T. All authors have read and agreed to the published version of the manuscript.
quisition, S.P. and T.T. All authors have read and agreed to the published version of the manuscript.
Sensors 2023, 23, 3377 20 of 22

Funding: This work was supported by the Research Institute of Rangsit University (RSU) and the
School of Engineering of King Mongkut’s Institute of Technology Ladkrabang (KMITL)—scholarship
for Siratchakrit Shinnakerdchoke, funding number KREF016601.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Acknowledgments: We would like to acknowledge the resources, fruitful discussion, and sugges-
tions from the College of Biomedical Engineering, Rangsit University, Thailand, and the School of
Engineering, KMITL, Thailand.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Brongersma, M.L.; Kik, P.G. Surface Plasmon Nanophotonics; Springer: Dordrecht, The Netherlands, 2007; Volume 131.
2. Somekh, M.G.; Pechprasarn, S. Surface plasmon, surface wave, and enhanced evanescent wave microscopy. In Handbook of
Photonics for Biomedical Engineering; Springer: Dordrecht, The Netherlands, 2017; pp. 503–543.
3. Homola, J.; Yee, S.S.; Gauglitz, G. Surface plasmon resonance sensors. Sens. Actuators B Chem. 1999, 54, 3–15. [CrossRef]
4. Suda, Y.; Arano, A.; Fukui, Y.; Koshida, S.; Wakao, M.; Nishimura, T.; Kusumoto, S.; Sobel, M. Immobilization and clustering
of structurally defined oligosaccharides for sugar chips: An improved method for surface plasmon resonance analysis of
protein-carbohydrate interactions. Bioconjug. Chem. 2006, 17, 1125–1135. [CrossRef]
5. Mariani, S.; Minunni, M. Surface plasmon resonance applications in clinical analysis. Anal. Bioanal. Chem. 2014, 406, 2303–2323.
[CrossRef] [PubMed]
6. Chen, J.; Park, B. Label-free screening of foodborne Salmonella using surface plasmon resonance imaging. Anal. Bioanal. Chem.
2018, 410, 5455–5464. [CrossRef] [PubMed]
7. Douzi, B. Protein–protein interactions: Surface plasmon resonance. In Bacterial Protein Secretion Systems: Methods and Protocols;
Humana Press: New York, NY, USA, 2017; pp. 257–275.
8. Drescher, D.G.; Selvakumar, D.; Drescher, M.J. Analysis of protein interactions by surface plasmon resonance. Adv. Protein Chem.
Struct. Biol. 2018, 110, 1–30.
9. Sangworasil, M.; Pechprasarn, S.; Learkthanakhachon, S.; Ittipornnuson, K.; Suvarnaphaet, P.; Albutt, N. Investigation on
feasibility of using surface plasmons resonance (SPR) sensor for ultrasonic detection: A novel optical detection of ultrasonic
waves. In Proceedings of the 2016 9th Biomedical Engineering International Conference (BMEiCON), Laung Prabang, Laos,
7–9 December 2016; pp. 1–3.
10. Shen, M.; Learkthanakhachon, S.; Pechprasarn, S.; Zhang, Y.; Somekh, M.G. Adjustable microscopic measurement of nanogap
waveguide and plasmonic structures. Appl. Opt. 2018, 57, 3453–3462. [CrossRef] [PubMed]
11. Suvarnaphaet, P.; Pechprasarn, S. Enhancement of long-range surface plasmon excitation, dynamic range and figure of merit
using a dielectric resonant cavity. Sensors 2018, 18, 2757. [CrossRef]
12. Suvarnaphaet, P.; Pechprasarn, S. Quantitative cross-platform performance comparison between different detection mechanisms
in surface plasmon sensors for voltage sensing. Sensors 2018, 18, 3136. [CrossRef]
13. De Mol, N.J.; Dekker, F.J.; Broutin, I.; Fischer, M.J.; Liskamp, R.M. Surface plasmon resonance thermodynamic and kinetic analysis
as a strategic tool in drug design. Distinct ways for phosphopeptides to plug into Src-and Grb2 SH2 domains. J. Med. Chem. 2005,
48, 753–763. [CrossRef]
14. Navratilova, I.; Hopkins, A.L. Emerging role of surface plasmon resonance in fragment-based drug discovery. Future Med. Chem.
2011, 3, 1809–1820. [CrossRef]
15. Chow, T.W.; Pechprasarn, S.; Meng, J.; Somekh, M.G. Single shot embedded surface plasmon microscopy with vortex illumination.
Opt. Express 2016, 24, 10797–10805. [CrossRef] [PubMed]
16. Englebienne, P.; Van Hoonacker, A.; Verhas, M. Surface plasmon resonance: Principles, methods and applications in biomedical
sciences. Spectroscopy 2003, 17, 255–273. [CrossRef]
17. Kretschmann, E.; Raether, H. Radiative decay of non radiative surface plasmons excited by light. Z. Nat. A 1968, 23, 2135–2136. [CrossRef]
18. Guo, J.; Keathley, P.D.; Hastings, J. Dual-mode surface-plasmon-resonance sensors using angular interrogation. Opt. Lett. 2008,
33, 512–514. [CrossRef] [PubMed]
19. Zeng, Y.; Zhou, J.; Wang, X.; Cai, Z.; Shao, Y. Wavelength-scanning surface plasmon resonance microscopy: A novel tool for real
time sensing of cell-substrate interactions. Biosens. Bioelectron. 2019, 145, 111717. [CrossRef] [PubMed]
20. Ran, B.; Lipson, S. Comparison between sensitivities of phase and intensity detection in surface plasmon resonance. Opt. Express
2006, 14, 5641–5650. [CrossRef]
21. Kabashin, A.V.; Patskovsky, S.; Grigorenko, A.N. Phase and amplitude sensitivities in surface plasmon resonance bio and chemical
sensing. Opt. Express 2009, 17, 21191–21204. [CrossRef] [PubMed]
22. Barer, R.; Tkaczyk, S. Refractive index of concentrated protein solutions. Nature 1954, 173, 821–822. [CrossRef]
Sensors 2023, 23, 3377 21 of 22

23. Crowell, J.; Ritchie, R. Surface-plasmon effect in the reflectance of a metal. JOSA 1970, 60, 794–799. [CrossRef]
24. Braundmeier, A., Jr.; Arakawa, E. Effect of surface roughness on surface plasmon resonance absorption. J. Phys. Chem. Solids 1974,
35, 517–520. [CrossRef]
25. Rahman, T.S.; Maradudin, A.A. Surface-plasmon dispersion relation in the presence of surface roughness. Phys. Rev. B 1980, 21,
2137. [CrossRef]
26. Hoffmann, A.; Lenkefi, Z.; Szentirmay, Z. Effect of roughness on surface plasmon scattering in gold films. J. Phys. Condens. Matter
1998, 10, 5503. [CrossRef]
27. Kurihara, K.; Suzuki, K. Theoretical understanding of an absorption-based surface plasmon resonance sensor based on Kretch-
mann’s theory. Anal. Chem. 2002, 74, 696–701. [CrossRef]
28. Treebupachatsakul, T.; Shinnakerdchoke, S.; Pechprasarn, S. Analysis of effects of surface roughness on sensing performance of
surface plasmon resonance detection for refractive index sensing application. Sensors 2021, 21, 6164. [CrossRef]
29. Byun, K.M.; Yoon, S.J.; Kim, D. Effect of surface roughness on the extinction-based localized surface plasmon resonance biosensors.
Appl. Opt. 2008, 47, 5886–5892. [CrossRef] [PubMed]
30. Liu, H.; Wang, B.; Leong, E.S.; Yang, P.; Zong, Y.; Si, G.; Teng, J.; Maier, S.A. Enhanced surface plasmon resonance on a smooth
silver film with a seed growth layer. ACS Nano 2010, 4, 3139–3146. [CrossRef]
31. Schug, C.; Schempp, S.; Lamparter, P.; Steeb, S. Surface roughness of sputter-deposited gold films: A combined x-ray technique
and AFM study. Surf. Interface Anal. 1999, 27, 670–677. [CrossRef]
32. Zhang, J.; Irannejad, M.; Yavuz, M.; Cui, B. Gold nanohole array with sub-1 nm roughness by annealing for sensitivity enhance-
ment of extraordinary optical transmission biosensor. Nanoscale Res. Lett. 2015, 10, 238. [CrossRef]
33. Zhang, C.; Li, J.; Belianinov, A.; Ma, Z.; Renshaw, C.K.; Gelfand, R.M. Nanoaperture fabrication in ultra-smooth single-grain gold
films with helium ion beam lithography. Nanotechnology 2020, 31, 465302. [CrossRef] [PubMed]
34. Ng, D.K.; Bhola, B.S.; Bakker, R.M.; Ho, S.T. Ultrasmooth gold films via pulsed laser deposition. Adv. Funct. Mater. 2011, 21,
2587–2592. [CrossRef]
35. Wieduwilt, T.; Kirsch, K.; Dellith, J.; Willsch, R.; Bartelt, H. Optical fiber micro-taper with circular symmetric gold coating for
sensor applications based on surface plasmon resonance. Plasmonics 2013, 8, 545–554. [CrossRef]
36. Diebel, J.; Löwe, H.; Samori, P.; Rabe, J. Fabrication of large-scale ultra-smooth metal surfaces by a replica technique. Appl. Phys. A
2001, 73, 273–279. [CrossRef]
37. Miller, M.S.; Ferrato, M.-A.; Niec, A.; Biesinger, M.C.; Carmichael, T.B. Ultrasmooth gold surfaces prepared by chemical
mechanical polishing for applications in nanoscience. Langmuir 2014, 30, 14171–14178. [CrossRef]
38. Leandro, L.; Malureanu, R.; Rozlosnik, N.; Lavrinenko, A. Ultrathin, ultrasmooth gold layer on dielectrics without the use of
additional metallic adhesion layers. ACS Appl. Mater. Interfaces 2015, 7, 5797–5802. [CrossRef]
39. Gaylord, T.K.; Moharam, M. Analysis and applications of optical diffraction by gratings. Proc. IEEE 1985, 73, 894–937. [CrossRef]
40. Moharam, M.; Grann, E.B.; Pommet, D.A.; Gaylord, T. Formulation for stable and efficient implementation of the rigorous
coupled-wave analysis of binary gratings. J. Opt. Soc. Am. A 1995, 12, 1068–1076. [CrossRef]
41. Thadson, K.; Visitsattapongse, S.; Pechprasarn, S. Deep learning-based single-shot phase retrieval algorithm for surface plasmon
resonance microscope based refractive index sensing application. Sci. Rep. 2021, 11, 16289. [CrossRef] [PubMed]
42. O’Donnell, K.; Mendez, E. Experimental study of scattering from characterized random surfaces. JOSA A 1987, 4, 1194–1205. [CrossRef]
43. Johnson, P.B.; Christy, R.-W. Optical constants of the noble metals. Phys. Rev. B 1972, 6, 4370. [CrossRef]
44. Davidson, A.M.; Brust, M.; Cooper, D.L.; Volk, M. Sensitive analysis of protein adsorption to colloidal gold by differential
centrifugal sedimentation. Anal. Chem. 2017, 89, 6807–6814. [CrossRef] [PubMed]
45. Saleem, M.R.; Ali, R. Polymer Resonant Waveguide Gratings. In Emerging Waveguide Technology; IntechOpen: London, UK, 2018.
46. Nuutinen, T.; Karvinen, P.; Rahomäki, J.; Vahimaa, P. Resonant waveguide grating (RWG): Overcoming the problem of angular
sensitivity by conical, broad-band illumination for fluorescence measurements. Anal. Methods 2013, 5, 281–284. [CrossRef]
47. Mayet, A.S.; Cansizoglu, H.; Gao, Y.; Ghandiparsi, S.; Kaya, A.; Bartolo-Perez, C.; AlHalaili, B.; Yamada, T.; Devine, E.P.;
Elrefaie, A.F. Surface passivation of silicon photonic devices with high surface-to-volume-ratio nanostructures. JOSA B 2018, 35,
1059–1065. [CrossRef]
48. Zhang, C.; Zhou, Y.; Mi, L.; Ma, J.; Wu, X.; Fei, Y. High performance of a metal layer-assisted guided-mode resonance biosensor
modulated by double-grating. Biosensors 2021, 11, 221. [CrossRef] [PubMed]
49. Berguiga, L.; Zhang, S.; Argoul, F.; Elezgaray, J. High-resolution surface-plasmon imaging in air and in water: V (z) curve and
operating conditions. Opt. Lett. 2007, 32, 509–511. [CrossRef]
50. Moharam, M.; Gaylord, T. Rigorous coupled-wave analysis of planar-grating diffraction. JOSA 1981, 71, 811–818. [CrossRef]
51. Chu, Y.; Schonbrun, E.; Yang, T.; Crozier, K.B. Experimental observation of narrow surface plasmon resonances in gold nanoparticle
arrays. Appl. Phys. Lett. 2008, 93, 181108. [CrossRef]
52. Zhang, J.; Pitter, M.C.; Liu, S.; See, C.; Somekh, M.G. Surface-plasmon microscopy with a two-piece solid immersion lens: Bright
and dark fields. Appl. Opt. 2006, 45, 7977–7986. [CrossRef]
53. Thadson, K.; Sasivimolkul, S.; Suvarnaphaet, P.; Visitsattapongse, S.; Pechprasarn, S. Measurement precision enhancement of
surface plasmon resonance based angular scanning detection using deep learning. Sci. Rep. 2022, 12, 2052. [CrossRef]
54. Meng, Q.-Q.; Zhao, X.; Lin, C.-Y.; Chen, S.-J.; Ding, Y.-C.; Chen, Z.-Y. Figure of merit enhancement of a surface plasmon resonance
sensor using a low-refractive-index porous silica film. Sensors 2017, 17, 1846. [CrossRef]
Sensors 2023, 23, 3377 22 of 22

55. Treebupachatsakul, T.; Boosamalee, A.; Chaithatwanitch, K.; Pechprasarn, S. Generalized figure of merit for plasmonic dip
measurement-based surface plasmon resonance sensors. Biomed. Opt. Express 2022, 13, 1784–1800. [CrossRef]
56. Yang, Z.; Liu, C.; Gao, Y.; Wang, J.; Yang, W. Influence of surface roughness on surface plasmon resonance phenomenon of gold
film. Chin. Opt. Lett. 2016, 14, 042401. [CrossRef]
57. Zhao, M.; Wang, J.; Zhang, Y.; Ge, M.; Zhang, P.; Shen, J.; Li, C. Self-referenced refractive index sensor based on double-dips
method with bimetal-dielectric and double-groove grating. Opt. Express 2022, 30, 8376–8390. [CrossRef]
58. Agarwal, S.; Prajapati, Y.; Singh, V. Influence of metal roughness on SPR sensor performance. Opt. Commun. 2017, 383, 113–118. [CrossRef]
59. Pechprasarn, S.; Chow, T.W.; Somekh, M.G. Application of confocal surface wave microscope to self-calibrated attenuation
coefficient measurement by Goos-Hänchen phase shift modulation. Sci. Rep. 2018, 8, 8547. [CrossRef] [PubMed]
60. Wy, Y.; Jung, H.; Hong, J.W.; Han, S.W. Exploiting plasmonic hot spots in Au-based nanostructures for sensing and photocatalysis.
Acc. Chem. Res. 2022, 55, 831–843. [CrossRef]
61. Sasivimolkul, S.; Pechprasarn, S.; Somekh, M.G. Analysis of open grating-based Fabry–Pérot resonance structures with potential
applications for ultrasensitive refractive index sensing. IEEE Sens. J. 2021, 21, 10628–10636. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.