Optics Communications 431 (2019) 142–150

Contents lists available at ScienceDirect

Optics Communications
journal homepage: www.elsevier.com/locate/optcom

Optimization methodology for structural multiparameter surface plasmon

resonance sensors in different modulation modes based on particle swarm
optimization
Yi Sun a,b , Haoyuan Cai c , Xiaoping Wang a,b,c , Shuyue Zhan a,b ,∗
a Ocean College, Zhejiang University, Zhoushan, 316021, China
b
Key Laboratory of Ocean Observation-Imaging Testbed of Zhejiang Province, Zhejiang University, Zhoushan, 316021, China
c
State Key Laboratory of Modern Optical Instrumentation, College of Optical Science and Engineering, Zhejiang University, Hangzhou, 310027, China

ARTICLE INFO ABSTRACT

Keywords: One of the main challenges in designing plasmonic biosensors is maximizing their sensing performance. This
Plasmonic biosensor study proposes heuristic algorithms based on surface plasmon resonance-particle swarm optimization (SPR-PSO),
Surface plasmon resonance which were investigated for the optimization of the sensing performance of structural multiparameter SPR sensors
Multiparameter
in four modulation modes (phase, intensity, wavelength, and angle). Different fitness functions were designed
Particle swarm optimization
for different modulation modes that comprised a variety of evaluation indicators (such as sensitivity, figure
Modulation mode
of merit, full-width-at-half-maximum, electric field intensity, and penetration depth). Four types of available
experimental structures representing the various modulation schemes were compared with the corresponding
optimized structure by algorithms. The results showed that the introduced algorithms have a considerable
efficiency. Furthermore, the algorithms also showed some potential in aiding the parametric design of negative
refractive index materials.

1. Introduction
Plasmonic biosensors possess real-time, rapid, and label-free char-
acteristics that allow the detection of the interaction of biological
molecules. The basis of the plasmonic sensing mechanism is dependent
on the excited charge density oscillation (surface plasmons) propagating
along the metal/dielectric interface that can resonate when the wave
vector of the incident light satisfies the resonance condition [1–4]. In
visible to near-infrared (NIR) wavelengths, the electric fields associated
with these oscillations are particularly sensitive to changes in the refrac-
tive index of the surrounding medium. Therefore, plasmonic sensors can
be used to monitor biological molecules participating in binding events
in real time, and they have also been used in important applications in
life sciences, food safety, drug screening, and in other areas [5–8].
Surface plasmon resonance (SPR) sensors based on surface plasmon
polaritons (SPP) are extensively used as plasmonic biosensors [9,10].
The SPR sensor usually adopts the Kretschmann structure to match the
momentum between SPP and the light beam. SPR sensors are usually
divided into four modulation methods. These are the angle, wavelength,
intensity, and phase modulations. The design of the SPR sensor usually
requires the determination of the sensor’s structural parameters. These
determine the sensitivity of the sensor, the full-width-at-half-maximum
(FWHM) value of the resonance curve, the limit of detection (LOD),
the electric field strength, and its penetration depth. In addition, the
resonance figure of merit (FOM) determined by the sensitivity and the
FWHM value can be used as a measure of the detection ability for
trace low-molecular-weight biomolecules [11]. However, the inability
to detect low-molecular-weight species is also a test bottleneck (< 500
Da) [12]. The lower penetration depth (200–300 nm) of SPR imposes
a large limitation when detecting analyte sizes equal to or larger than
micrometer-sized organisms, such as cells. In addition, the fabrication
process of nanoscale SPR biosensing chips is complicated and costly, and
there are specific issues related to this field [13]. Therefore, the design
optimization of SPR biosensors is highly meaningful, even though the
number of prior investigations has been limited. In order to solve these
problems, excellent optimization techniques must be introduced.
In recent years, conducted research on heuristic algorithms has
been applied to the numerical optimization of conventional surface
plasmon resonance (cSPR) sensor configurations [14–16]. However,
these studies only considered the angle-modulated cSPR structure (gold
film structure) as a model, which has fewer parameters and lower
versatility. At the same time, the fitness function of this model is not
well designed, and simply taking the lowest point of the single-mode
resonance curve as the basis of its variation limits its applicability.
With the development of plasmonic biosensors, novel sensing methods,

∗ Corresponding author.
E-mail addresses: sundayi@zju.edu.cn (Y. Sun), shuyue_zhan@zju.edu.cn (S. Zhan).

https://doi.org/10.1016/j.optcom.2018.09.027
Received 17 June 2018; Received in revised form 25 August 2018; Accepted 11 September 2018
Available online 20 September 2018
0030-4018/© 2018 Elsevier B.V. All rights reserved.
Y. Sun et al. Optics Communications 431 (2019) 142–150

such as long-range surface plasmon resonance (LRSPR) [17], hyper-

bolic metamaterial (HMM) [18], multilayer surface plasmon waveguide
(MSPW) [19], symmetrical optical waveguide (SOW) [20], and oth-
ers [21,22], has led to the gradual replacement of cSPR. Owing to the
presence of the sensor chip with multilayer films, these novel sensors
have the characteristics of a multiparametric structure, and a variety
of performance (sensitivity of multi-mode resonance, spatial detection
capability, and small molecular detection capability). Thus, a more
comprehensive and effective intelligent algorithm used to solve the
current structural design of plasmonic biosensors, thus yielding optimal
parameters and the best sensor performance, has become particularly
important.
In this work, the SPR particle swarm optimization (SPR-PSO) al-
gorithm was formulated by combining the Fresnel coefficient matrix
method with particle swarm optimization (PSO), so that the optimal SPR Fig. 1. Optical flow diagram. 𝑁𝑖 and 𝑁𝑗 represent the two refractive indices. 𝐸𝑖+ and 𝐸𝑖−
represent the forward electric field and the reverse electric field in 𝑁𝑖 , respectively. 𝐸𝑗+ and
sensor structural parameters could be calculated. This study systemati-
𝐸𝑗− represent the forward electric field and the reverse electric field in 𝑁𝑗 , respectively.
cally proposes a novel fitness function calculation model corresponding
to the four modulation modes listed previously. The fitness function
model includes sensitivity, FOM, and penetration depth as three sensor
performance indicators. These are normalized and then unified for In order to determine the electric field intensity of the 𝑗th interface,
evaluation. The evaluation model can be adjusted according to the we need to calculate the amplitude and phase shift of the forward and
sensor design requirements. The four types of SPR sensing structures backward waves.
in the literature are optimized and analyzed, and performance results The physical structure through which light wave propagates shows
that are much superior to the original model structure are obtained. that 𝐸0+ and 𝐸𝑗− are known parameters, and we can express them as
Correlation analyses are also conducted. {
𝐸𝑗+ = 𝑀11 𝐸0+ + 𝑀12 𝐸0−
(7)
2. SPR-PSO theory 𝐸𝑗− = 𝑀21 𝐸0+ + 𝑀22 𝐸0− .

2.1. Fresnel coefficient matrix method The values of 𝐸0+ and 𝐸𝑗− are determined by solving the matrix
equation. Thus, we can obtain the total reflection coefficient of the film
SPR involves a multilayer optical thin-film structure. Correspond- structure as:
ingly, in this study, the Fresnel matrix method [23] is used to calculate 𝑟 = 𝐸0−∕𝐸0+ . (8)
light transmission within the multilayer thin-film structure. When the
optical thin film is in the optical field, the electric field vector of each The amplitude and phase shift are
layer is based on the matrix equations, and the electric field distribution [ ]1
| +| 2
in the membrane layer is calculated based on the matrix method. |𝐸𝑗 | = (𝐸𝑗+𝑅 )2 + (𝐸𝑗+𝐼 )2 (9)
| |
The optical flow diagram is shown in Fig. 1, where 𝑁𝑖 and 𝑁𝑗
𝛿𝑗+ = tan −1
(𝐸𝑗+𝑅 ∕𝐸𝑗+𝐼 ) (10)
represent two refractive indices. Considering the optical flow balance
on the interface, it is easy to derive the following pair of equations: where 𝐸𝑗+ = 𝐸𝑗+𝑅
+ 𝐸𝑗+𝐼 .
𝐸𝑖− = 𝑟𝑖𝑗 𝐸𝑖+ + 𝑡𝑗𝑖 𝐸𝑗− | |
The parameters, |𝐸𝑗− | and 𝛿𝑗− can be obtained in a similar manner.
(1) | |
𝐸𝑗+ = 𝑡𝑖𝑗 𝐸𝑖+ + 𝑟𝑗𝑖 𝐸𝑗− ,
The electric field intensity 𝐸𝑗 is expressed as
where 𝑟𝑖𝑗 and 𝑡𝑖𝑗 are the Fresnel reflection and transmission coefficients,
| |2 | |2 | || |
respectively. One can also derive the following relations between 𝑟𝑖𝑗 and 𝐸𝑗2 = |𝐸𝑗+ | + |𝐸𝑗− | + 2 |𝐸𝑗+ | |𝐸𝑗− | cos(𝛿𝑗+ + 𝛿𝑗− ). (11)
| | | | | || |
𝑡𝑖𝑗 : By these means, we can obtain the electric field intensity throughout
𝑡𝑖𝑗 𝑡𝑗𝑖 + 𝑟2𝑖𝑗 = 1 (2) the film.

𝑟𝑖𝑗 = −𝑟𝑗𝑖 (3) 2.2. Particle swarm optimization

This expression is rewritten as
Kennedy and Oberheart first introduced this heuristic algorithm in
( )( )
⎛𝐸 + ⎞ 1995 [24]. These algorithms are intended to mimic the behavior of
⎜ 𝑗⎟=
𝑡𝑖𝑗 1 −𝑟𝑖𝑗 𝐸𝑖+
, (4) natural systems. The algorithm assumes that the particles are searching
⎜ 𝐸 − ⎟ 1 − 𝑟2 −𝑟𝑖𝑗 1 𝐸𝑖−
⎝ 𝑗⎠ 𝑖𝑗 the solution space for optimal solution points, and the search begins
The inverse matrix can be expressed as randomly in order to form particles. Thus, the best position of each
( −𝑖𝛿𝑗 ) memory (𝑃𝑏𝑒𝑠𝑡 ) is explored. Conversely, the total memory of the best
𝑒 0
, (5) position for the population (𝐺𝑏𝑒𝑠𝑡 ) is obtained by sharing the path of the
0 𝑒−𝑖𝛿𝑗
particle before the state, the best experience of the particles, and the
where 𝛿𝑗 denotes the phase thickness of the light through the film. social status of the individual particles in the next path. The particle
According to these ideas, we can obtain the forward and reverse motion relationship is as follows:
fields of the 𝑗th interface in the multilayer film as
𝑖 = 𝜔𝑉 𝑖 + 𝑐 ⋅ 𝑟𝑎𝑛𝑑 ⋅ (𝑃 𝑖 𝑖
( ) 𝑉𝑡+1 𝑡 1 𝑏𝑒𝑠𝑡 − 𝑋𝑡 ) + 𝑐2 ⋅ 𝑟𝑎𝑛𝑑 ⋅ (𝐺𝑏𝑒𝑠𝑡 − 𝑋𝑡 )
⎛𝐸 + ⎞ (12)
⎜ 𝑗⎟=
𝑡1 𝑡2 … 𝑡𝑗 𝑒−𝑖𝛿𝑗−1 −𝑟𝑗 𝑒𝑖𝛿𝑗−1 𝑖
𝑋𝑡+1 = 𝑋𝑡𝑖 + 𝑉𝑡+1
𝑖 .
…
⎜𝐸 − ⎟ (1 − 𝑟2 )(1 − 𝑟2 ) … (1 − 𝑟2 ) −𝑟 𝑒−𝑖𝛿𝑗−1 𝑒𝑖𝛿𝑗−1
⎝ 𝑗⎠ 1 2 𝑗 𝑗 𝑖 is the particle speed at time t + 1, 𝑉 𝑖 is the particle speed at
𝑉𝑡+1
( −𝑖𝛿 )( ) ( +) 𝑡
𝑒 1 −𝑟2 𝑒𝑖𝛿1 1 −𝑟𝑖𝑗 𝐸0 time t, 𝜔 is a weight inertia factor, 𝑐1 is the cognitive critical index,
(6)
−𝑟2 𝑒−𝑖𝛿1 𝑒𝑖𝛿1 −𝑟𝑖𝑗 1 𝐸0− 𝑐2 is the social critical index, 𝑟𝑎𝑛𝑑 is a normally distributed random
( ) number between zero and one, 𝑃𝑏𝑒𝑠𝑡 is the best experienced position by
⎛𝐸 + ⎞ ⎛ +⎞
⎜ 𝑗 ⎟ = 𝑀11 𝑀12 ⎜𝐸0 ⎟ 𝑖
the particle, 𝐺𝑏𝑒𝑠𝑡 is the best experienced position, and 𝑋𝑡+1 and 𝑋𝑡𝑖 are
⎜𝐸 − ⎟ 𝑀21 𝑀22 ⎜⎝𝐸0− ⎟⎠ the particle positions at times t and t + 1.
⎝ 𝑗⎠

143
Y. Sun et al.
2.3. Fitness function
The design of the fitness function directly affects the convergence
speed of PSO and the selection of the optimal solution. The PSO does
not use external information during the evolutionary search but only
the fitness function as a basis for searching. The PSO algorithm does not
need the fitness function to satisfy the continuous and differentiable
solution conditions. The only requirement is to calculate the non-
negative results from the input parameters.
Considering the multimode resonance phenomenon, the influence of
the multilayer structure, different modulation methods, and penetration
depth, the algorithm is designed to determine the optimal wavelength
function, intensity modulation, wavelength modulation, and angle mod-
ulation fitness functions. According to the formation mechanism, phase
and intensity modulations are defined as Type A. Correspondingly, this
type has a direct detection capability that allows the detection of a
linear response in intensity or phase change. Wavelength modulation
and angle modulation are defined as Type B. In turn, this type has an
indirect ability to detect linear changes of the resonance peak.
2.3.1. Type A fitness function
⎧ ∑𝑛−1 |𝐼 − 𝐼 |
⎪ 1 | 𝑖+1 𝑖|
× 10𝑅 𝑅 ≥ 𝑅𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑
𝑓𝐴1 =⎨ 𝑛−1 (13)
⎪ 0 𝑅 < 𝑅𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑
⎩ matrices are obtained. Subsequently, the corresponding subscript values
where 𝑓𝐴1 is the evaluation index for the resonance response of the Type
A fitness function, which combines the curve slope (sensitivity) with the
curve linearity, 𝑅 represents the linear correlation coefficient of discrete
′ ), 𝐼 ′ = |𝐼
data 𝐼 ′ (where 𝐼 ′ = 𝐼 ′ (𝐼1′ , 𝐼2′ , … , 𝐼𝑛−1 |
𝑖 | 𝑖+1 − 𝐼𝑖 | , 𝐼𝑖 represents the
light intensity or phase response value), and 𝑅𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 is the threshold
for data linearity and is equal to 0.95.
√ The evaluation index of the FOM in angle or spectral modulation is
∑𝑛
( )( ) √ √∑ 𝑛
( )2 ∑ 𝑛 ( )2
𝑅= 𝑟𝑖 − 𝑟 𝐼𝑖 − 𝐼 ∕√ 𝑟𝑖 − 𝑟 𝐼𝑖 − 𝐼
𝑖=1 𝑖=1 𝑖=1
where 𝑟𝑖 is the discretization value in the effective sensing index range
corresponding to the light intensity or phase 𝐼𝑖 , 𝑟 is the mean value
corresponding to its refractive index of 1.33–1.35, and 𝐼 is the average
value corresponding to 𝐼𝑖 .
( ) √
𝑓𝐴2 = 𝑒 𝑑𝑝𝑒𝑛𝑒 ∕1000 × |𝐸𝑚∕𝑑 2 | (15)
Correspondingly, 𝑓𝐴2 is the evaluation index of the Type A fitness
function and characterizes the electric field propagating in the film,
𝑑𝑝𝑒𝑛𝑒 is the penetration depth of the evanescent wave in the dielectric
structure, which is defined as the distance at which the amplitude decays
to 1/e of its surface value, and 𝐸𝑚∕𝑑 2 is the initial electric field intensity.
𝑓𝐴1 𝑓 unified as follows,
𝐹𝐴 = 𝑘𝐴1 + 𝑘𝐴2 𝐴2 (16)
𝐹𝐴1 𝐹𝐴2
𝑘𝐴1 + 𝑘𝐴2 = 1 (17)
where 𝐹𝐴 is a comprehensive evaluation index of the Type A fitness
function. The evaluation index unifies the two physical quantities of
the resonance response and the propagating evanescent wave. Addi-
tionally, 𝑘𝐴1 and 𝑘𝐴2 are the weight values of the different evaluation
indexes, which can be set according to the calculation requirements.
Furthermore, 𝐹𝐴1 and 𝐹𝐴2 denote the constraint indices of the resonance
response, and the propagating evanescent wave, respectively. Empirical
analysis shows that in phase modulation, 𝐹𝐴1 = 1 000 000 and 𝐹𝐴2 = 100.
For intensity modulation, 𝐹𝐴1 = 1000 and 𝐹𝐴2 = 100. The parameter data
of the particle swarm algorithm are listed in Table 1.
Optics Communications 431 (2019) 142–150
Table 1
Initial parameters of the PSO algorithm for Type A.
Learning factor 1 Learning factor 2 Inertia weight Population size Iterative number
1.77 0.63 0.754 50 200
2.3.2. Type B fitness function
𝑇𝑑 = ([𝑇2 , 𝑇3 , ⋯ , 𝑇𝑛−1 ] − [𝑇1 , 𝑇2 , ⋯ , 𝑇𝑛−2 ])
∙ ([𝑇2 , 𝑇3 , ⋯ , 𝑇𝑛−1 ] − [𝑇3 , 𝑇4 , ⋯ , 𝑇𝑛 ]) (18)
where 𝑇𝑑 is an intermediate variable matrix that is used to determine
the extreme point of angle modulation and wavelength modulation
resonance, and 𝑇𝑖 is the response value of the light intensity in angle
modulation or spectral modulation. The positive elements of the one-
dimensional matrix 𝑇𝑑 can be obtained based on screening (𝑇𝑑 > 0).
This forms a one-dimensional matrix 𝑇 + = [𝑇1+ , 𝑇2+ , ⋯ , 𝑇𝑚+ ]. The point
where the resonance peak is not significant in the one-dimensional
matrix 𝑇 + appears in the following relationship:
|𝑇𝑗+ − 𝑇𝑗−1
+
| > 𝜌 && |𝑇𝑗+1 + − 𝑇𝑗+ | > 𝜌 (19)
where 𝜌 is the resonance peak determination constant, and is de-
fined as 𝜌 = 0.5. According to this relationship, all the elements
are traversed in sequence in the one-dimensional matrix 𝑇 + , and the
+
element [𝑇𝑗+1 , 𝑇𝑗+ , 𝑇𝑗−1
+
] that satisfies the condition is input into a one-
′
dimensional matrix 𝑇 + to obtain the final resonance peak matrix. By
changing the different refractive index values, different resonance peak
in the resonance peak matrix are extracted, and the angles or spectral
sensitivities of different resonance peaks are calculated. The sensitivities
are used to obtain the variation in the maximum sensitivity in the
resonance peaks according to the calculation method𝑓𝐴1 in the Type A
fitness function,
𝑓𝐵1 = 𝑓𝐴1 (20)
calculated by
(14)
𝑆𝑚𝑎𝑥
𝑓𝐵2 = (21)
𝑊𝐹 𝑊 𝐻𝑀
where 𝑆𝑚𝑎𝑥 is the maximum sensitivity of all resonance peaks, and
𝑊𝐹 𝑊 𝐻𝑀 is the FWHM of the resonance peak at maximum sensitivity
(angle width or spectral width).
In addition, the electric field strength and the evanescent wave
penetration depth in angle modulation or wavelength modulation are
taken as the evaluation indexes. The calculation is performed according
to the method for 𝑓𝐴2 in the Type A fitness function:
𝑓𝐵3 = 𝑓𝐴2 (22)
The resonance response evaluation index, the quality factor eval-
uation index, and the evanescent wave electric field and propagation
evaluation index in angle modulation or wavelength modulation can be
𝑓 𝑓 𝑓
𝐹𝐵 = 𝑘𝐵1 𝐵1 + 𝑘𝐵2 𝐵2 + 𝑘𝐵3 𝐵3 (23)
𝐹𝐵1 𝐹𝐵2 𝐹𝐵3
𝑘𝐵1 + 𝑘𝐵2 + 𝑘𝐵3 = 1 (24)
where 𝐹𝐵 is a comprehensive evaluation index of the Type B fitness
function. The evaluation index unifies the three physical resonance
response quantities (the quality factor and the electric field of the
propagating evanescent wave). Equivalently, 𝑘𝐵1 , 𝑘𝐵2 , and 𝑘𝐵3 , are the
weight values of the different evaluation indices. Additionally, 𝐹𝐵1 , 𝐹𝐵2 ,
and 𝐹𝐵3 are the resonance response constraint indices, the quality factor,
and the evanescent wave electric field, respectively. Empirical analysis
shows that 𝐹𝐵1 = 2 000 000, 𝐹𝐵2 = 2000, and 𝐹𝐵3 = 1000 in wavelength
modulation , and that 𝐹𝐵1 = 1000, 𝐹𝐵2 = 3000, and 𝐹𝐵3 = 1000, in angle
modulation. The initial parameters of the particle swarm algorithm are
shown in Table 2.
144
Y. Sun et al.
Fig. 2. (a) Schematic diagram of the differential phase-sensitive SPR sensor based on a modified interferometric configuration. (b) Schematic diagram of the SPR polarization contrast
method and layers. (c) Schematic diagram of the wavelength-sensitive LRSPR sensor and structure. (d) Schematic diagram of the SOW-based SPR imaging system and layers.
Table 2
Initial parameters of the PSO algorithm for Type B.
Learning factor 1 Learning factor 1 Inertia weight Population size Iterative number
0.74 1.51 0.849 200 500
3. Results and discussion
In order to demonstrate the application advantages of the PSO–
SPR model in the SPR sensing design, the four different modulation
modes involved in the SPR sensor are divided into two models: Type A
(phase modulation and intensity modulation) and Type B (wavelength
modulation and angle modulation). For each of the four modulation
modes, the PSO-SPR algorithm was used to maximize the performance
of the sensor and verify the applicability of the PSO–SPR algorithm in
the SPR sensor model design. The optimization algorithm can provide
intelligent and efficient design ideas and design tools in the field of
plasmonic sensors.
3.1. Phase modulation
Graphene plasmonic structures can provide extreme phase features
as a result of electric field enhancement, offering much improved phase
sensitivity. Zeng et al. proposed an ultrasensitive plasma biosensor based
on phase modulation [22], which uses a graphene–gold metasurface
structure. As shown in Fig. 2(a), the sensitivity of the sensor reached
34 200 deg/RIU, and the refractive index resolution reached a value of
1 × 10−8 , and was higher than that of the cSPR structure. Additionally,
the initial field strength of the sensor was much higher than that of
the cSPR, which made the proposed structure very attractive, and could
be used to detect aromatic ring structure biomolecules, such as helical
peptides or proteins that can selectively bind graphene surfaces.
Using the algorithm design of phase modulation in Type A, fur-
ther optimization of the graphene sensing structure was performed to
145
Optics Communications 431 (2019) 142–150
identify and explore the best performance parameters it can achieve.
In the calculation, the incident wavelength (range: 300–1000 nm), the
incident angle (range: 0–90◦ ), the thickness of the gold film structure
(range: 0–100 nm), and the number of graphene layers (0–4 layer), are
the four parameters that were used as variables to be optimized. In the
evaluation index of fitness of the Type A algorithm, the parameters were
set as 𝑘𝐴1 = 0.5, 𝑘𝐴2 = 0.5, and the corresponding simulation results are
shown in the angle modulation column in Table 3. In the optimization
structure, the incident wavelength is 870 nm, the incident angle is 51.5◦ ,
the gold film thickness is 47 nm, and the graphene layers can be in a
single layer, as shown in Fig. 3(a). In this structure, the sensitivity can
reach 45 060 deg/RIU, which is much higher than the achieved refer-
ence structure sensitivity of 34 200 deg/RIU, thus indicating that the
optimized structure elicits a considerable effect. In addition, as shown
in Fig. 3(b), the penetration depth of the optimized structure reaches
207 nm compared to the penetration depth (147 nm) of the reference
structure, which is indicative of an enhancement. From these results,
the evaluation index of the fitness in the optimized algorithm reaches
0.62, which is higher than the value of 0.45 for the reference structure
evaluation index. The simulation results further demonstrate the ability
of the phase modulation algorithm to tap the sensing performance of
the sensing structure when calculating the complex structure.
3.2. Intensity modulation
The polarization control method was first proposed by Homola and
Yee in 1998 [25]. This method achieves sensitivity close to phase detec-
tion. Dong et al.proposed an improved polarization contrast method for
surface plasmon resonance imaging sensors by inert background gold
film extinction [26]. As shown in Fig. 2(b), in the intensity modulation
structure of the sensor, the detection sensitivity reached 35/RIU and the
refractive index resolution reached 1.7 × 10−6 /RIU. The results show
that the detection resolution is enhanced by one order of magnitude
compared to the traditional intensity modulation. However, this type
Y. Sun et al. Optics Communications 431 (2019) 142–150

Table 3
Simulation parameters and results in four kinds of modulation modes (phase modulation, intensity modulation, wavelength modulation, and angle
modulation). Related abbreviations are as follows: CS: conference structure, OS: optimum structure, RI: refractive index, T: thickness, W: wavelength,
RA: resonance angle, PIEF: the propagation intensity of electric field, and PD: the penetration depth.
SF11 Au graphene
CS
RI 1.76552 0.14891 + 4.783i 2.90 + 1.500i
T/nm / 50 0.34
SF11 Au graphene
Phase modulation OS
RI 1.761 0.1656 + 5.4873i 2.90 + 1.579i
T/nm / 47 0.34
W/nm RA/deg S/(deg/RIU) PIEF PD/nm
CS
785 52.1 34 200 48.7 147
W/nm RA/deg S/(deg/RIU) PIEF PD/nm
OS
870 51.5 45 060 63.23 207
K9 Au
CS
RI 1.512 0.1489 + 4.783i
T/nm / 40
K9 Au
Intensity modulation OS
RI 1.5081 0.2344 + 6.439i
T/nm / 29.8
W/nm RA/deg polar/deg wave/deg ana/deg S/(I/RIU) PIEF PD/nm
CS
740 68.3 45 −43 80 35 24.9 151
W/nm RA/deg polar/deg wave/deg ana/deg S/(I/RIU) PIEF PD/nm
OS
950 64.8 53.2 36.1 137.5 73 37.9 259
BK7 Teflon AF Au
CS
T/nm / 1200 24
BK7 Teflon AF Au
Wavelength modulation OS
T/nm / 1509 15.7
RA/deg FWHM S/(nm/RIU) FOM PIEF PD/nm
CS
52.2 51 57 000 1117 43.5 1400
RA/deg FWHM S/(nm/RIU) FOM PIEF PD/nm
OS
62.3 31 54 000 1741 64.1 1638
SF4 MgF2 Au MgF2
CS
RI 1.75 1.38 0.197 + 3.09i 1.38
T/nm / 505 39 645
SF4 MgF2 Au MgF2
Angle modulation OS
RI 1.74 1.3756 0.167 + 4.959i 1.3756
T/nm / 1320 60 810
W/nm FWHM/deg S/(deg/RIU) FOM PIEF PD/nm
CS
633 0.051 39 764 122.6 661.6
W/nm FWHM/deg S/(deg/RIU) FOM PIEF PD/nm
OS
741 0.025 44 1760 136.2 552

of structure is still a traditional gold film structure, and its penetration
depth is still similar to the traditional gold film.
For using the algorithm to design structures for intensity modulation
in Type A, further optimization of the polarization contrast sensing
structure was conducted to explore the best performance parameters.
In the calculation, the incident wavelength (range: 300–1000 nm),
the incident angle (range: 0–90◦ ), the polarizing angle (range: 0–90◦ ),
the waveplate angle (range: −90–90◦ ), the analyzer angle (range: 0–
180◦ ), and the thickness of the thin gold film (range: 0–100 nm),
are the variables intended to be optimized. In the evaluation index
of fitness for the Type A algorithm, 𝑘𝐴1 = 0.8 and 𝑘𝐴2 = 0.2. The
corresponding simulation results are shown in the intensity modulation
column in Table 3. In the optimized structure, the incident wavelength
is 950 nm, the incident angle is 64.8◦ , the polarizing angle is 53.2◦ ,
the waveplate angle is 36.1◦ , the analyzer angle is 137.5◦ , and the
thickness of the thin gold film is 29.8 nm, as shown in Fig. 4(a). The
sensitivity of this structure can reach 73/RIU, which is approximately
twice the reference structure sensitivity of 35/RIU. The optimization
of the structural sensitivity results in a corresponding electric field
enhancement, as shown in Fig. 4(b). The penetration depth in the
structure reaches 259 nm, which is greatly improved compared to
the reference structure penetration depth of 151 nm. Based on these
results, the fitness evaluation index in the algorithm reaches 0.67,
which is higher than the reference structure evaluation index value
of 0.33. The simulation results further demonstrate the ability of the
intensity modulation algorithm to optimize the sensing performance
when calculating the complex structure.
The experimental data, based on the home-built polarization contrast
modulation SPRi sensor with the optimized structure, is in the Support
Information.
3.3. Wavelength modulation
Long-range surface plasmons (LRSPs) are surface electromagnetic
waves that can be produced on thin metallic film imbedded between
a dielectric medium and the analyte medium with similar refractive
indices. The LRSPR is improved compared to cSPR and can effectively
reduce the FWHM of the resonance peak and improve the sensor
sensitivity and refractive index resolution. R. Slavík et al. proposed
ultra-high sensitivity surface plasmon resonance sensors based on long-
range surface plasmon polaritons [17]. The overall structure shown in

146
Y. Sun et al.
Fig. 3. (a) Schematic diagram of the response curve between the reference structure and
the optimum structure. The sensing range of the reference structure is 1.3320–1.3348,
and the sensing range of the optimized structure is 1.3325–1.3345. In the inset figure, the
sensitivity of the reference structure is 34 200 deg/RIU, and the sensitivity of the reference
structure is 45 060 deg/RIU. (b) Schematic diagram of the electric field distribution in the
reference structure and the optimum structure. In the upper figure, the penetration depth
is 147 nm, and the inset shows the electric field propagation intensity between the two
kinds of structures. In the lower figure, the penetration depth is 207 nm, and the inset
shows the penetration depth between the two kinds of structures.
Fig. 2(c). The sensor achieves a resolution of 5.7 × 104 nm/RIU and
a refractive index resolution of 2.5 × 10−8 , which is higher than that
achieved by the traditional SPR instrumentation. The field distribution
of LRSP reaches a penetration depth of 1400 nm, which is much higher
than the cSP penetration depth of 200 nm. These results make the
proposed structure very attractive for detecting large analytes, such as
particles, bacteria, spores, and viruses.
The best performance parameters for the further optimization of
the LRSPR sensing structure are obtained from the design algorithm
for wavelength modulation in Type B. In the calculation, the angle
of incidence (range: 0–90◦ ), the thickness of Teflon AF (range: 0–
3000 nm), and the thickness of the gold film (range: 0–100 nm) are
chosen as the variables to be optimized. Equivalently, 𝑘𝐵1 = 0.2,
𝑘𝐵2 = 0.3, and 𝑘𝐵3 = 0.5 are used in the evaluation index of the fitness
function in the B-type algorithm. The corresponding simulation results
are shown in the wavelength modulation column in Table 3. In the
optimized structure, the incident angle is 62.3◦ , the Teflon AF thickness
is 1509 nm, and the thickness of the gold film is 15.9 nm, as shown
147
Optics Communications 431 (2019) 142–150
Fig. 4. (a) Schematic diagram of the response curve between the reference structure and
the optimum structure. The sensing range of the reference structure is 1.331–1.346, and
the sensing range of the optimized structure is 1.330–1.339. In the inset, the sensitivity of
the reference structure is 35 deg/RIU, and the sensitivity of the reference structure is 73
deg/RIU. (b) Schematic diagram of the electric field distribution for the reference structure
and the optimum structure. In the upper figure, the penetration depth is 151 nm, and the
inset shows the electric field propagation intensity between the two kinds of structures.
In the lower figure, the penetration depth is 259 nm, and the inset shows the penetration
depth between the two kinds of structures.
in Fig. 5(a). The sensor sensitivity reaches 54 000 nm/RIU, which is
similar to the reference structure sensitivity, which can reach a value
of 57 000 nm/RIU. However, the sensor’s FWHM is 31 nm, which is
better than the FWHM of the reference structure of 51 nm. Therefore,
the FOM determined by the sensitivity and FWHM is 1741, which is
56% higher than the value of 1117 elicited by the reference structure.
As shown in Fig. 5(b), the penetration depth of the optimized structure
reaches 1638 nm. The initial propagation intensity of the electric
field reaches 88.6, which is much better than the initial propagation
intensity of the reference structure. Based on these results, the fitness
evaluation index of the algorithm reaches 0.48, which is higher than
the evaluation index of 0.31 of the reference structure. The simulation
results further demonstrate the ability of the wavelength modulation
algorithm to optimize the sensing performance when calculating the
complex structure.
Y. Sun et al.
Fig. 5. (a) Schematic diagram of the response curve between the reference structure
and the optimum structure. The FWHM of the reference structure is 51 nm, and the
FWHM of the optimum structure is 31 nm. The upper inset figure shows the sensitivity
between the two kinds of structures. The lower inset shows the FOM between the
two kinds of structures. (b) Schematic diagram of the electric field distribution in the
reference structure and the optimum structure. In the upper figure, the penetration depth
is 1400 nm, and the inset shows the electric field propagation intensity between the two
kinds of structures. In the lower figure, the penetration depth is 1638 nm, and the inset
shows the penetration depth between the two kinds of structures.
3.4. Angle modulation
A symmetric optical waveguide (SOW) sensor structure coats the
film with the same buffer layer material on the gold film surface of
the LRSPR structure, thus forming a symmetric structure. Specifically,
when effective refractive indices of the two mediums are close, the
two SPs will form a kind of wave modes referred to as LRSPs. Shi
et al. proposed a surface plasmon resonance biosensing system based
on the SOW structure [20]. The system is a simple angle modulation
SPR imaging system. Fig. 2(d) shows the LRSPR imaging principle. The
sensitivity of the structure sensing in the liquid reaches 39 deg/RIU
with a refractive index resolution of 8.1 × 10−8 RIU. These values are
better than the corresponding values of the traditional SPR instrument.
Moreover, the strength of SOW is approximately equal to three times
that of the LRSPR, and the electric field penetration depth into the
analyte increases by a factor of four.
The angular modulation design algorithm in Type B is used to obtain
the best performance parameters for further optimization of the SOW
148
Optics Communications 431 (2019) 142–150
Fig. 6. (a) Schematic diagram of the response curve between the reference structure and
the optimum structure. The FWHM of the reference structure is 0.051◦ and the FWHM of
the optimum structure is 0.025◦ . The upper inset shows the sensitivity between the two
kinds of structures, and the lower inset shows the FOM between the two kinds of structures.
(b) Schematic diagram of the electric field distribution for the reference structure and the
optimum structure. In the upper figure, the penetration depth is 552 nm, and the inset
shows the electric field propagation intensity between the two structures. In the lower
figure, the penetration depth is 661 nm, and the inset shows the penetration depth between
the structures.
sensing structure. In the calculation, the incident wavelength (range:
300–1000 nm), thickness of the first layer of MgF2 (range: 0–3000 nm),
thickness of the gold film (range: 0–100 nm), and the thickness of the
second layer of MgF2 (range: 0–3000 nm), are chosen as the variables
to be optimized. Equivalently, in Type B, 𝑘𝐵1 = 0.3, 𝑘𝐵2 = 0.4, and
𝑘𝐵3 = 0.3. The corresponding simulation results are shown in the
angle modulation column in Table 3. In the optimized structure, the
incident wavelength is 741 nm, the thickness of the first MgF2 layer is
1320 nm, the thickness of the gold film is 60 nm, and the thickness
of the second MgF2 layer is 810 nm, as shown in Fig. 6(a). In this
structure, a sensitivity of 44 deg/RIU can be achieved, which is slightly
higher than that of the reference structure sensitivity of 39 nm/RIU.
The optimized structure has FWHM of 0.025◦ , which is better than the
FWHM value of 0.051◦ in the reference structure. Therefore, the FOM
value is 1760, which is 130% higher than that of the FOM (764). The
optimized structure greatly improves the response. In addition, as shown
in Fig. 6(b), the optimized structure has a penetration depth of 661 nm,
which is 120% of the reference structure’s penetration depth (552 nm).
Y. Sun et al.
Fig. 7. Comparison of the fitness values and increase ratios for different modulation
structures (phase, intensity, wavelength, and angle).
From these results, the fitness index of the algorithm reaches 0.43,
which is higher than the evaluation index value of 0.28 in the reference
structure. The simulation results further demonstrate the ability of the
angle modulation algorithm to optimize the sensing performance of the
sensing structure when calculating the complex structure.
As shown in Fig. 7, the designed simulation algorithm optimizes the
sensor parameters according to different structures in the four sensors
with different modulation modes. It is shown that they greatly enhance
the sensor performance and fully excite the sensing potential of the
sensors in specific structures. The designed fitness function evaluation
index improves the optimized structure in phase modulation by 38%
compared to the reference structure. This also yields a 103% increase in
the intensity modulation structure compared to the reference structure.
The wavelength modulation is 48% more efficient than the reference
structure, and the angle modulation structure is 54% more efficient than
the reference structure. Also shown, are the sensitivity, FOM, penetra-
tion depth, and the other sensing indices involved in the optimization
of the structure. The improvement of these indices corresponds to an
increase of the fitness value, which confirms that the evaluation method
of the fitness function is helpful for evaluating the sensor performance.
In addition to the optimization of the structural parameters of the
known materials, they can also play a role in the parametric design of
unknown materials. Negative refractive index materials used in plasma
biosensors are one example, and we have reported related results in
previously published work [27]. The refractive index of these materials
can be optimized for design. The application of this type of material in
sensors can exhibit a non-traditional response mode (sensing intensity
singularity), thus providing relevant research ideas in the field of
material design.
4. Conclusions
A PSO–SPR optimization algorithm was proposed in this work. By
identifying the response signal and the electric field distribution, and by
designing the fitness function, the algorithm was applied to the optimal
design of four SPR modulation modes (phase, intensity, wavelength, and
angle). The PSO–SPR optimization algorithm was then used to simulate
and optimize the sensor parameters of the four modulation schemes. The
results showed that the fitness function value of the optimized structure
was improved by 38% in the phase modulation structure, 103% in
the intensity modulation structure, 48% in the wavelength modulation
structure, and 54% in the angle modulation structure. The remarkable
enhancement effect confirmed the considerable advantage and potential
of this type of optimization algorithm in the design of the SPR sensing
149
Optics Communications 431 (2019) 142–150
structures. Not only can it optimize the parameters of the known sensor
structure in order to exploit its maximum sensing capability, it also
showed considerable ability to design the unknown refractive index
of the material. This optimization algorithm will provide an effective
solution to the design and optimization of new and exotic plasmonic
biosensors in the future.
Acknowledgments
This research was financially supported by the National Natural Sci-
ence Foundation of China (NSFC) (No. 61775191) and (No. 61605169).
The authors wishes to thank Chunfei Shi and Zheng Shi (College of
Optical Science and Engineering of Zhejiang University) for useful
discussions and assistance.
Appendix A. Supplementary data
Supplementary material related to this article can be found online at
https://doi.org/10.1016/j.optcom.2018.09.027.
References
[1] J.N. Anker, W.P. Hall, O. Lyandres, N.C. Shah, J. Zhao, R.P. Van Duyne, Biosensing
with plasmonic nanosensors, Nature Mater. 7 (2008) 442–453.
[2] J. Xu, Y. Chen, Surface plasmon resonance sensing with adjustable sensitivity based
on a flexible liquid core coupling unit, Talanta 184 (2018) 468–474.
[3] M. Svedendahl, R. Verre, M. Käll, Refractometric biosensing based on optical phase
flips in sparse and short-range-ordered nanoplasmonic layers, Light Sci. Appl. 3
(2014) e220.
[4] H. Im, H. Shao, Y. Il Park, V.M. Peterson, C.M. Castro, R. Weissleder, H. Lee, Label-
free detection and molecular profiling of exosomes with a nano-plasmonic sensor,
Nat. Biotechnol. 32 (2014) 490–495.
[5] J.-F. Masson, Surface plasmon resonance clinical biosensors for medical diagnostics,
ACS Sensors 2 (2017) 16–30.
[6] H.-S. Lee, T.-Y. Seong, W.M. Kim, I. Kim, G. Hwang, W.S. Lee, K. Lee, Enhanced
resolution of a surface plasmon resonance sensor detecting C-reactive protein via
a bimetallic waveguide-coupled mode approach, Sensors Actuators B 266 (2018)
311–317.
[7] R. Liu, Q. Wang, Q. Li, X. Yang, K. Wang, W. Nie, Surface plasmon resonance
biosensor for sensitive detection of microRNA and cancer cell using multiple signal
amplification strategy, Biosens. Bioelectron. 87 (2017) 433–438.
[8] S. Pan, J. Xu, Y. Shu, F. Wang, W. Xia, Q. Ding, T. Xu, C. Zhao, M. Zhang, P. Huang,
S. Lu, Double recognition of oligonucleotide and protein in the detection of DNA
methylation with surface plasmon resonance biosensors, Biosens. Bioelectron. 26
(2010) 850–853.
[9] P. Singh, SPR biosensors: Historical perspectives and current challenges, Sensors
Actuators B 229 (2016) 110–130.
[10] J. Homola, S.S. Yee, G. Gauglitz, Surface plasmon resonance sensors: review,
Sensors Actuators B 54 (1999) 3–15.
[11] L. Wang, X.-J. Liu, J. Hao, L.-Q. Chu, Long-range surface plasmon resonance sensors
fabricated with plasma polymerized fluorocarbon thin films, Sensors Actuators B
215 (2015) 368–372.
[12] a.V. Kabashin, P. Evans, S. Pastkovsky, W. Hendren, G.a. Wurtz, R. Atkinson,
R. Pollard, V.a. Podolskiy, a.V. Zayats, Plasmonic nanorod metamaterials for
biosensing, Nature Mater. 8 (2009) 867–871.
[13] H. Neff, W. Zong, A.M.N. Lima, M. Borre, G. Holzhüter, Optical properties and
instrumental performance of thin gold films near the surface plasmon resonance,
Thin Solid Films 496 (2006) 688–697.
[14] D. Barchiesi, Adaptive non-uniform, hyper-elitist evolutionary method for the
optimization of plasmonic biosensors, in: 2009 Int. Conf. Comput. Ind. Eng. 2009,
pp. 542–547.
[15] P. Lecaruyer, E. Maillart, M. Canva, J. Rolland, Generalization of the Rouard
method to an absorbing thin-film stack and application to surface plasmon reso-
nance, Appl. Opt. 45 (2006) 8419.
[16] L.M. Cavalcanti, E. Fontana, Monte Carlo and particle swarm methods applied to
the design of surface plasmon resonance sensors, Proc. IEEE Sensors (2017) 31–33.
[17] R. Slavík, J. Homola, Ultrahigh resolution long range surface plasmon-based sensor,
Sensors Actuators B 123 (2007) 10–12.
[18] K.V. Sreekanth, Y. Alapan, M. ElKabbash, E. Ilker, M. Hinczewski, U.A. Gurkan, A.
De Luca, G. Strangi, Extreme sensitivity biosensing platform based on hyperbolic
metamaterials, Nature Mater. 15 (2016) 4–11.
[19] T. Srivastava, R. Jha, R. Das, High-performance bimetallic SPR sensor based on
periodic-multilayer- waveguides, IEEE Photonics Technol. Lett. 23 (2011) 1448–
1450.
Y. Sun et al.
[20] H. Shi, Z. Liu, X. Wang, J.J. Guo, L. Liu, L. Luo, J.J. Guo, H. Ma, S. Sun, Y.
He, A symmetrical optical waveguide based surface plasmon resonance biosensing
system, Sensors Actuators B 185 (2013) 91–96.
[21] J. Dostálek, R.F. Roskamp, W. Knoll, Coupled long range surface plasmons for the
investigation of thin films and interfaces, Sensors Actuators B 139 (2009) 9–12.
[22] S. Zeng, K.V. Sreekanth, J. Shang, T. Yu, C.K. Chen, F. Yin, D. Baillargeat, P. Coquet,
H.P. Ho, A.V. Kabashin, K.T. Yong, Graphene-gold metasurface architectures for
ultrasensitive plasmonic biosensing, Adv. Mater. 27 (2015) 6163–6169.
[23] J. Tang, P. Gu, X. Liu, H. Li, Modern Optical Thin Film Technology, first ed.,
Zhejiang University Press, Hangzhou, 2006.
[24] R. Eberhart, J. Kennedy, A new optimizer using particle swarm theory, in: MHS’95.
Proc. Sixth Int. Symp. Micro Mach. Hum. Sci. (n.d.), pp. 39–43.
150
Optics Communications 431 (2019) 142–150
[25] J. Homola, S.S. Yee, Novel polarization control scheme for spectral surface plasmon
resonance sensors, Sensors Actuators B B51 (1998) 331–339.
[26] W. Dong, K. Pang, Q. Luo, Z. Huang, X. Wang, L. Tong, Improved polarization con-
trast method for surface plasmon resonance imaging sensors by inert background
gold film extinction, Opt. Commun. 346 (2015) 1–9.
[27] Y. Sun, H. Cai, X. Wang, Theoretical analysis of metamaterial-gold auxiliary grating
sensing structure for surface plasmon resonance sensing application based on
polarization control method, Opt. Commun 405 (2017) 343–349.