*Article* **A High Sensitivity Temperature Sensing Probe Based on Microfiber Fabry-Perot Interference**

#### **Zhoubing Li 1, Yue Zhang 1, Chunqiao Ren 1, Zhengqi Sui <sup>1</sup> and Jin Li 1,2,\***


Received: 27 March 2019; Accepted: 15 April 2019; Published: 16 April 2019

**Abstract:** In this paper, a miniature Fabry-Perot temperature probe was designed by using polydimethylsiloxane (PDMS) to encapsulate a microfiber in one cut of hollow core fiber (HCF). The microfiber tip and a common single mode fiber (SMF) end were used as the two reflectors of the Fabry-Perot interferometer. The temperature sensing performance was experimentally demonstrated with a sensitivity of 11.86 nm/ ◦C and an excellent linear fitting in the range of 43–50 ◦C. This high sensitivity depends on the large thermal-expansion coefficient of PDMS. This temperature sensor can operate no higher than 200 ◦C limiting by the physicochemical properties of PDMS. The low cost, fast fabrication process, compact structure and outstanding resolution of less than 10−<sup>4</sup> ◦C enable it being as a promising candidate for exploring the temperature monitor or controller with ultra-high sensitivity and precision.

**Keywords:** fiber sensors; temperature sensors; Fabry-Perot interferometer; microfiber; PDMS; integrated optics

#### **1. Introduction**

As a typical physical parameter, the temperature must be carefully controlled and monitored in many fields, such as clinical medicine, biochemical reactions, industrial production, aviation safety and so on [1–3]. In recent years, optical fiber temperature sensors have aroused widespread research interest, because of their unique advantages compared with electrical ones, such as remote monitoring capability, high sensitivity, anti-electromagnetic interference properties, and intrinsic safety [4,5]. By combining the resonance enhancement effect of the optical coupling technique, multi-modes interference, optical evanescent field, optical time domain reflecting and optical ring-down technology produced by different special optical fiber structures, various optical fiber temperature sensors were realized [6–9]. Multi-modes interference is carried out by splicing together different kinds of fibers to excite the modes' interference. The splicing joints are fragile and the length for each section must be carefully controlled during the fabrication process; Optical evanescent fields can be obtained around micro/nanofibers with diameters comparable to the wavelength of the incident light. Although micro/nanofibers offer excellent performance, the sensor probes based on them are difficult to fabricate because of their thin diameter and environmentally sensitive properties. The optical time domain reflection technique was used to sense temperature and strain based on Raman or Brillouin scattering [10]. The sensitivity of the temperature sensor based on optical ring-down technology only can be increased by extending the fiber length.

In addition to the basic sensing mechanism, the sensing performance was further improved by means of temperature sensitive materials [11]. Many materials, such as polymers and metal oxides, have been reported to be elaborated by surface or inner coating, and used to encapsulate the whole fiber structure [12–14]. In addition to the temperature dependence, the effect of humidity, strain and other related parameters on the sensing performance must be determined and eliminated. At present, the most common commercial optical fiber temperature sensor is the fiber Bragg grating (FBG) having good repeatability and stable sensing characteristics [15]. It can be prepared by ultraviolet exposure or nano-etching technology to meet the working requirements of different temperature ranges [16]. However, high sensitivity or precision is difficult to obtain for FBG temperature sensors, which seriously hinders their commercial application [17].

The optical fiber temperature sensors based on multi-wavelength interference mainly include Mach–Zehnder and Fabry-Perot interferometers. The former typically perform as transmission structures, which separate and transmit an independent signal light and reference light by using different special optical fibers or structures, such as micro/nano fibers, photonic crystal fibers (PCFs), dislocation fusion fibers and multi-core fibers. A corresponding sensitivity of up to 6.5 nm/ ◦C was observed [18]. However, the structures of the Mach-Zehnder interferometers are complex due to their dual-optical-paths system [19–22]. To simplify the structures, the two optical paths can be revealed in single fiber, named the in-line Mach-Zehnder interferometer, such as C-typed PCFs [23], side-hole PCFs [24], D-shaped-hole fibers [25] and muti-core fibers [26]. These compact structures were precisely machined using femtosecond lasers, focused ion beams and chemical vapor deposition, and display excellent stability and sensing performance. However, these are hard to manufacture in batches due to the high cost and technical requirements. In addition to the above complex optical fiber structures, single polymer optical fibers have been demonstrated with a temperature sensitivity of ~10−<sup>3</sup> ◦C [27], where the temperature performance were revealed by the transmission power and the effect of relative and twist have been experimentally obtained [28,29]. Furthermore, their packaging size is hard to reduce further depending on the bending loss of the optical fiber [30], which will seriously limit their application in a narrow space; the latter ones are carried out as reflective structures, where the temperature sensitive cavity was constructed at the end of the optical fiber by laser or ion beam processing, chemical etching or film forming and special fiber splicing technologies [31–37]. Among them, femtosecond laser processing can machine a refractive index turning point with good repeatability in the optical fiber, which was used as a Fabry-Perot cavity and can work at high temperatures up to 1000 ◦C [31]; focused ion beams can etch an air cavity at the tip of an optical fiber, based which a Fabry-Perot temperature sensor with a sensitivity of −654 pm/ ◦C has been experimentally demonstrated [32]. However, the expensive and complex preparation processing, as well as the high technique requirements for engineers have become huge obstacles for commercial production [33]. The Fabry-Perot interferometer probe can be obtained conveniently and quickly by chemical etching or film forming technology [34], however, the fabrication repeatability is low, and the structural parameters are difficult to control accurately [35]. By using the special hollow-core photonic bandgap fiber (HC-PBF) or PCF, the temperature working range and sensitivity of cascaded splicing fiber based Fabry-Perot interferometer has been experimentally verified as high as 1200 ◦C and 17 nm/ ◦C, respectively, but their structures are relatively fragile [36,37].

Compared with conventional temperature sensors, the proposed Fabry-Perot interferometer temperature sensor costs less and is easier and faster to prepare. This compact Fabry-Perot temperature probe was proposed by encapsulating a microfiber and a single mode fiber (SMF) tip in a hollow core fiber (HCF), between which temperature sensitive polydimethylsiloxane (PDMS) was filled and cured. The microfiber was prepared by the one-step heating-stretching technique from a normal SMF. The microfiber and SMF can be easily aligned due to the comparable inner diameter of HCFs. The high transparency and low refractive index of PDMS causes little impact on the incident light. Furthermore, a sensitivity of higher than 11 nm/ ◦C has been experimentally demonstrated due to its high thermal expansion coefficient. This temperature sensor will be a promising candidate for monitoring temperature fluctuations in small spaces due to its high sensitivity and tiny scale (200 μm in diameter and <5 mm in length).

#### **2. Materials and Methods**

To fabricate the Fabry-Perot interferometer, a cut of transparent HCF was prepared firstly, as shown in Figure 1. The coating layer of a HCF (TSP150200, inner diameter: ~150 μm, outer diameter: ~200 μm, coating layer: polyimide, Polymicro Technologies, Inc., Phoenix, AZ, USA) was removed by a Bunsen burner (Dragon 200, fuel: butane, max-temperature 1300 ◦C, Rocker Scientific Co., Ltd., New Taipei, Taiwan), as shown in inset (a) of Figure 1. The cavity length of Fabry-Perot interferometer can be observed through its transparent wall. Both the microfiber and SMF can be inserted and aligned easily due to their small diameter difference. The microfiber was obtained from the SMF (Coating removed diameter: 125 μm, SMF-28, Corning Inc., Corning, NY, USA) using the scanning flame heating-stretching technique (inset (b) of Figure 1). Where, the diameter and length of microfiber were precisely controlled by optimizing the fabrication process of a fiber melting-drawing system (IPCS-5000-ST, Idealphotonics Inc., Hong Kong, China). This system uses the high-purity hydrogen and oxygen as the fuel to obtain a high heating temperature of up to 2500–3000 ◦C. When SMF reaches a melting state at high temperature, its two ends were fixed onto two motorized displacement platforms and stretched in opposite directions. By carefully adjusting the speed and scanning region of the flame, the microfiber with uniform diameter can be obtained in the heating zone. Different diameters were easily achieved by controlling the stretching velocity. Fabry-Perot interferometer was finally fabricated by assistance of a homemade micromanipulation system (inset (c) of Figure 1). A cut of transparent HCF was fixed on a slide glass substrate with UV glue. One end of SMF and microfiber were cut with a flat-face and acted as two reflecting surfaces of Fabry-Perot interferometer. The other tail-ends of SMF and microfiber were clamped by two fiber claps and fixed onto two three-dimensional (3-D) optical fiber adjusting frames (APFP-XYZ, adjusting precision <2 μm, Zolix Instruments Co., Ltd., Beijing, China).

**Figure 1.** Fabrication process of the microfiber and PDMS based Fabry-Perot temperature probe. I: Coating layer of HCF was removed to prepare the transparent HCF (inset (**a**)); II: MF taper was prepared by scanning flame stretching technique (inset (**b**)); III: Fabry-Perot temperature probe was fabricated by assistance of the micromanipulation method under a microscope (insets (**c**)&(**d**)).

In this case, the Fabry-Perot structure can be timely observed and measured by a microscope system (DMM-300C, Shanghai Caikon Optical Instrument Co., Ltd., Shanghai, China) on a computer and its cavity length was also timely precisely manipulated according to the reflected spectrum. The basic component and curing agent were mixed with a weight ratio of 10:1 to obtain the PDMS sol, which was filled into the HCF using a syringe (inset (d) of Figure 1) and cured in ~20 min. The experimental schematic was illustrated in Figure 2.

**Figure 2.** Experimental schematic of the microfiber and PDMS-based Fabry-Perot interferometer for sensing temperature. The light source, temperature sensor and spectrometer were contacted by a 1 × 2 coupler. The enlarged schematic and microscope picture of the temperature probe were illustrated.

An amplified stimulated emission (ASE, ASE-C light source, 1520–1610 nm, Shenzhen Golight Technology Co., Ltd., Shenzhen, China) was used as the light source. The light was launched into the Fabry-Perot interferometer temperature probe through a 1 × 2 coupler (with the splitter ratio of 50:50). The reflection optical signal was collected by an optical spectrum analyzer (OSA, AQ6370, 600–1700 nm, resolution 20 pm, Yokogawa Electric Corp., Tokyo, Japan). In this work, the optical polarization direction does not affect the sensing performance due to the circularly polarized light output of ASE source and the cylindrical structure of microfiber. The temperature probe was placed in a thermostat (25–250 ◦C, resolution 0.1 ◦C, Shanghai Boxun Medical Biological Instruments Co., Ltd., Shanghai, China). The inset is a micrograph of the proposed temperature probe. The microfiber has a uniform diameter of ~63 μm and a length of ~2 cm. This length should be carefully controlled depending on the cone angle of the microfiber taper. Too long and thin microfiber will be easily adsorbed on the inner surface of HCF because of van der Waals force. In this case, it will be difficult to parallel its end-face with the reflecting surface of SMF to construct the two reflectors of Fabry-Perot interferometer. The cavity length was finally determined as ~34 μm.

#### **3. Results**

In the experiment, the Fabry-Perot interferometer temperature probe was placed in a thermostat. The temperature was increased from room temperature to 100 ◦C with steps of 1 ◦C. The reflection spectra of the Fabry-Perot temperature probe were recorded by a spectrometer. The printed pictures of the spectrometer screen at different temperature (40 ◦C and 41 ◦C) are illustrated in Figure 3. The spectrum curve refers to the original spectrum reflected from the microfiber. The free spectral range (FSR) is ~21 nm, during which the wavelength values of the resonance dips were determined with demodulation equipment with the resolution of 1 pm. When temperature changed from 40 ◦C (Figure 3a) to 41 ◦C (Figure 3b), the resonance dip moved with a wavelength location shift of ~10.5 nm, indicating a resolution of lower than 10−<sup>4</sup> ◦C. Due to the limitation of one period of FSR, the ultra-sensitive temperature fluctuation monitoring can be achieved in the maximum range of 0 ◦C to ~2 ◦C (fluctuation level: ±1 ◦C). In order to achieve a commercial low-cost device, a photodiode can be used to monitor the change in intensity of a single wavelength to determine the direction and magnitude of temperature fluctuations.

**Figure 3.** Reflection spectra of the microfiber Fabry-Perot temperature probe at temperature of (**a**) 40 ◦C and (**b**) 41 ◦C, respectively.

The sensitivity of Fabry-Perot interferometer is dependent on the resonance shift as a function of temperature [38]:

$$s = \frac{\Delta\lambda\_T}{\Delta T} = \lambda\_m \alpha\_{FP} = \lambda\_m \left(\frac{L\_{PDMS}\alpha\_{PDMS,T} - L\_{f\,\text{here}}\alpha\_{f\,\text{here},T}}{L\_{PDMS}}\right) \tag{1}$$

where, α*FP* refers to the relative change for the cavity length of the Fabry-Perot interferometer. In this work, it is depended on both the thermal expansion of PDMS and silica fibers. The FSR can be expressed as:

$$FSR = \frac{\lambda^2}{2n\_{\rm PDMS}L\_{\rm PDMS}}\tag{2}$$

In addition to the incident wavelength, FSR is inversely proportional to the change in refractive index and length of PDMS, which are determined by its thermo-optic coefficient and thermal expansion coefficient, respectively. Here, the thermal expansion coefficient plays a dominant role in the temperature change process, since the bulk expansion of PDMS is limited by the HCF wall and transferred into a change in cavity length to improve the sensitivity of the sensor. For the proposed Fabry-Perot interferometer, FSR is inversely proportional to the spacing between the microfiber and SMF tips, which was demonstrated experimentally when we continuously moved the microfiber towards the SMF in the HCF.

To demonstrate the temperature sensing performance in a wider range, the wavelength movement of one resonance dip was marked and traced in the whole spectrum range of the light source from 1520 nm to 1610 nm, as shown in Figure 4. When the temperature increased from 43 ◦C to 50 ◦C with steps of 1 ◦C, a resonance dip was marked to trace its shift amount. The inset of Figure 4 illustrates eight reflection spectra recorded at the different temperature values, where the resonance dip is red-shifted for almost a half cycle of the FSR. This resonance dip was independently selected to clearly display the temperature sensing characteristics from 43 ◦C to 50 ◦C. The resonance dip red-shifted continuously from 1534.8 nm (43 ◦C) to 1607.3 nm (50 ◦C).

**Figure 4.** Reflection spectra of the microfiber Fabry-Perot temperature probe for the temperature increased from 43 ◦C to 50 ◦C with a step of 1 ◦C.

The corresponding temperature sensing characteristic curve was shown in Figure 5. Black (circle) and red (pentagon) experimental data points and corresponding linear fitting represent the temperature response results for heating and cooling process, respectively.

**Figure 5.** Location of resonance dip changed as a function of temperature with an excellent linear fitting during the increasing and decreasing process of 43–50 ◦C. The corresponding sensitivities were determined to be 10.37 nm/ ◦C and 10.67 nm/ ◦C, respectively.

During the heating process, a sensitivity of up to 10.37 nm/ ◦C for the temperature sensing was experimentally demonstrated with a linearity of 0.99965. To verify the recovery characteristics of the temperature sensor, the movement of resonance dip was recorded through the cooling process in the same temperature range (from 50 ◦C to 43 ◦C in steps of 1 ◦C). By linearly fitting the experimental data points, a sensitivity of up to 10.67 nm/ ◦C was obtained with a linearity of 0.99535. The performance curve illustrates the relationship between resonance dip and temperature, which will be stable for a temperature probe with fixed structure parameters. When a new sensor is used, the temperature can be determined by referring to calibration curve.

To reveal the repeatability and stability of the proposed temperature sensor, three-cycle experiments for a sensing probe with the cavity length of 31 μm and the microfiber diameter of 61 μm were performed, where the corresponding wavelength shift values depending on the temperature increasing/decreasing were recorded and illustrated in Figure 6. The highest sensitivity of 11.86 nm/ ◦C was experimentally demonstrated for the temperature increase process in the first round, which was higher than the probe in Figure 5 mainly due to the shorter cavity, which matches well with the theoretical analysis. Equation (1) indicates that the sensitivity is proportional to the relative change in cavity length. A shorter cavity will result in a more significant change than that of the longer one. Furthermore, its larger FSR enables high-precise temperature fluctuation monitoring in a wider range (see the analysis of Equation (2) and Figure 3).

**Figure 6.** Three-cycle experiments for the temperature-dependence curves for a Fabry-Perot interferometer with the cavity length of 31 μm and the microfiber diameter of 61 μm. Inset: Stability of the wavelength locations for the temperature changing between 45 ◦C and 46 ◦C.

The maximum wavelength backlash was determined as ~1.3 nm during the three-cycle measurement process. On the one hand, this is related to the thermal expansion relaxation time of PDMS; on the other hand, it is also limited by the temperature control accuracy of the oven, which is also indicated in the stability measurement of the proposed temperature probe (inset of Figure 6). When the temperature fluctuates between 45 ◦C and 46 ◦C, the positional fluctuation of the resonance wavelength was less than ~0.2 nm, and the corresponding response time (stabilization time) was ~3 min. The above fluctuations fall within the performance range of the thermostatic oven. In order to calibrate the sensing characteristics of this temperature probe in a larger working range, the specific resonance dips should be dynamically selected in different temperature ranges. Thereafter, the temperature sensing characteristic curve can be obtained by using the relative shift of the labeled resonance dips. In addition to the microfiber and SMF, the final working range of this temperature probe will be limited by the sensitive materials. PDMS in solid status has the stable physicochemical property in the

temperature range of −55–200 ◦C. Therefore, this temperature probe can work in a wider range, not limited to the results reported in this work.

#### **4. Discussion**

The optical fiber temperature sensor proposed in this work is compact and easily prepared. Its sensitivity is significantly higher than most of other fiber temperature sensors reported in recent years, as compared in Table 1.


**Table 1.** Sensing performance comparison for typical temperature probes based on optical fibers.

SMS: Single-muti-single mode fiber; NOA 73: Norland optical adhesive 73; RI: Refractive index; PMMA: poly(methyl methacrylate; LOCTITE 3493: Light cure adhesive 3493.

As can be seen from Table 1, the temperature sensitivity of FBGs is low, and the encapsulation technology and demodulation optical path are complex [16]. The dual-arms system of Mach-Zehnder interferometers are commonly built using special optical fibers (for example PCF [6,20] or microfibers [18,19,30]) or by splicing different optical fibers [18,21], where the sensitive liquid or polymer were introduced to create a temperature-sensitive probe [6,20,21]. In contrast, the Fabry-Perot fiber interferometer can be easily fabricated on a single fiber. It has a more compact structure for developing high-performance temperature microprobes. Femtosecond laser [31] or ion beam etching technology [32], as well as high-precision fiber-splicing technology [36,37], can improve its temperature detection limit to as high as 1200 ◦C, making it suitable for extreme high temperature environments; furthermore, sol coating [35] or temperature-sensitive polymer encapsulation technology [39] can be used for enhance normal temperature microprobes, which will be a promising candidate for implantable microsensors for health or environmental monitoring under 200 ◦C.

Compared with the polymer film reflector, in this work, the smooth end-faces of SMF and microfiber were used as the two reflectors of the Fabry-Perot interferometer. PDMS is used to fix the two reflectors and realize a highly sensitive response to temperature changing. The temperature response properties can be revealed by the contribution of the negative thermal-optics coefficient (α*to*: <sup>−</sup>450 <sup>×</sup> 10−6/ ◦C) and the thermal-expansion coefficient (α*te*: 960 <sup>×</sup> 10−6/ ◦C) effects of PDMS. When the temperature increases, a smaller effective refractive index and a longer cavity length will be obtained, respectively. In view of their contributions to the effective optical path between the two

reflectors of the Fabry-Perot interferometer, they have the opposite impact on the cavity length when the temperature changes.

#### **5. Conclusions**

In this paper, a compact and miniature Fabry-Perot interferometer based on a microfiber and SMF in a cut of HCF was proposed and experimentally demonstrated. The morphology parameters, such as microfiber diameter and cavity length, can be precisely controlled by the microfiber fabrication (scanning flame stretching technique) and micromanipulation processes (microscope- assised micromanipulation method), respectively. By filling PDMS into this Fabry-Perot interferometer with the microfiber diameter of ~63 μm and cavity length of ~34 μm, a temperature sensitivity of higher than 10 nm/ ◦C was experimentally obtained. When the cavity length was reduced to ~31 μm, a highest sensitivity of 11.86 nm/ ◦C has been experimentally demonstrated with an excellent repeatability and stability. Due to its high sensitivity and easily adjustable morphology, this Fabry-Perot temperature sensor has promising applications for precisely monitoring temperature fluctuations in biochemical reaction processes, industrial production and food storage.

**Author Contributions:** Z.L. and Y.Z. conceived and performed the experiments; they and C.R. analyzed the results and was involved in the paper writing; Z.S. was involved in the paper writing; J.L. was the supervisor and involved in the paper review and editing.

**Funding:** This research was funded by Fundamental Research Funds for the Central Universities (N170405003 and N170407005), Liaoning Province Natural Science Foundation (20180510015) and National Undergraduate Innovation and Entrepreneurship Program (190158).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Hybrid Plasmonic Fiber-Optic Sensors**

#### **Miao Qi 1,**†**, Nancy Meng Ying Zhang 1,**†**, Kaiwei Li 2, Swee Chuan Tjin 1,\* and Lei Wei 1,\***


Received: 20 April 2020; Accepted: 6 June 2020; Published: 8 June 2020

**Abstract:** With the increasing demand of achieving comprehensive perception in every aspect of life, optical fibers have shown great potential in various applications due to their highly-sensitive, highly-integrated, flexible and real-time sensing capabilities. Among various sensing mechanisms, plasmonics based fiber-optic sensors provide remarkable sensitivity benefiting from their outstanding plasmon–matter interaction. Therefore, surface plasmon resonance (SPR) and localized SPR (LSPR)-based hybrid fiber-optic sensors have captured intensive research attention. Conventionally, SPR- or LSPR-based hybrid fiber-optic sensors rely on the resonant electron oscillations of thin metallic films or metallic nanoparticles functionalized on fiber surfaces. Coupled with the new advances in functional nanomaterials as well as fiber structure design and fabrication in recent years, new solutions continue to emerge to further improve the fiber-optic plasmonic sensors' performances in terms of sensitivity, specificity and biocompatibility. For instance, 2D materials like graphene can enhance the surface plasmon intensity at the metallic film surface due to the plasmon–matter interaction. Two-dimensional (2D) morphology of transition metal oxides can be doped with abundant free electrons to facilitate intrinsic plasmonics in visible or near-infrared frequencies, realizing exceptional field confinement and high sensitivity detection of analyte molecules. Gold nanoparticles capped with macrocyclic supramolecules show excellent selectivity to target biomolecules and ultralow limits of detection. Moreover, specially designed microstructured optical fibers are able to achieve high birefringence that can suppress the output inaccuracy induced by polarization crosstalk and meanwhile deliver promising sensitivity. This review aims to reveal and explore the frontiers of such hybrid plasmonic fiber-optic platforms in various sensing applications.

**Keywords:** optical fibers; hybrid plasmonic sensors; surface plasmon resonance; localized surface plasmon resonance; 2D materials; graphene; transition metal oxides; gold nanoparticles; cyclodextrin

#### **1. Introduction**

Plasmonic fiber-optic sensors have captured intensive research attention in recent years due to their high degree of integration, high sensitivity, flexibility and remote-sensing capability [1–3]. Fiber-optic plasmonic sensors can be generally classified into two categories, surface plasmon resonance (SPR)-based sensors and localized surface plasmon resonance (LSPR)-based sensors. Conventionally, SPR- and LSPR-based hybrid fiber-optic sensors are realized by depositing thin metal films and metallic nanoparticles on various fiber structures (e.g., fiber gratings, side-polished fiber, microfiber, etc.) respectively to contribute strong plasmon–matter interaction. To further improve the measurement accuracy, sensitivity and selectivity to analyte molecules, specially designed fiber structures or functional materials are normally applied to strengthen the intensity of surface plasmon or the adsorption to target molecules. Specialty optical fibers like microstructured optical fibers (MOFs)

can achieve high-level integration so that the very small dimension waveguide and the microfluidic channels are able to be integrated within a single fiber with only micrometer-scale diameter, leading to effective plasmon–matter interaction. The recent breakthroughs in 2D materials such as graphene, transition metal dichalcogenides (MX2), transition metal oxides (TMOs), etc. reveal new opportunities in plasmon–matter enhancement by constructing 2D material/metal hybrid plasmonic structures or heavily doping-free carriers in 2D TMOs to realize intrinsic strong plasmonics in frequently used visible or near-infrared (NIR) optical windows. In addition, macrocyclic supramolecules have been recently proven to be excellent surface functionalization candidates for metallic nanoparticles, contributing to a simple functionalization process, selective target molecules recognition and improved biocompatibility.

In this review, the background and the state-of-the-art of SPR/LSPR fiber-optic sensors will be reviewed. More importantly, the aforementioned emerging hybrid fiber-optic plasmonic sensing solutions will be illustrated in detail. For instance, the exploration of how the highly birefringent MOF based SPR sensor can suppress polarization crosstalk and improve sensitivity in the meantime; how to integrate graphene-on-gold hybrid structure on the fiber-optic platform to strengthen the surface plasmon intensity and to effectively adsorb biomolecules; how to dope abundant free electrons in 2D MoO3 and achieve highly integrated microfiber based plasmonic sensing in NIR optical frequencies; how to synthesize cyclodextrin-capped gold nanoparticles (AuNPs) in a one-step process and realize microfiber based highly selective detection of cholesterol in human serum, etc. will be demonstrated.

#### **2. Fiber-Optic Surface Plasmon Resonance Sensors**

A surface plasmon polariton (SPP) is an electromagnetic wave that propagates in parallel at the interface between the metal film and dielectric medium. The SPP is TM-polarized. As illustrated in Figure 1a, the polarization direction of SPP is perpendicular to the metal–dielectric interface [4]. The SPP has the evanescent nature, which is strongest at the surface of the metal film and exponentially decays into the dielectric material. Conventionally, SPR is realized on a Kretschmann–Raether silica prism of which the base is coated with a nanometer-scale thin metal film (Figure 1b) [5]. The ambient medium of the thin metal film is dielectric and considered as semi-infinite, by solving the Maxwell equations, the propagation constant of SPP is given by:

$$K\_{SP} = \frac{\alpha}{c} \sqrt{\frac{\varepsilon\_{metal} \varepsilon\_{dielectric}}{\varepsilon\_{metal} + \varepsilon\_{dielectric}}} \tag{1}$$

where ω, *c* and ε are the frequency of the TM-polarized incident light, light velocity and dielectric constant, independently. The propagation constant of the evanescent wave at the interface is:

$$K\_{cv} = \frac{\alpha}{c} \sqrt{\varepsilon\_{prism}} \sin \theta \tag{2}$$

where θ is the angle of light incidence. SPR can be excited by the TM-polarized total reflected light at the silica-metal interface when the phase matches, that the propagation constant of reflected light equals to the propagation constant of SPP:

$$\frac{\omega}{c}\sqrt{\varepsilon\_{\text{prism}}}\sin(\theta\_{\text{res}}) = \frac{\omega}{c}\sqrt{\frac{\varepsilon\_{\text{metal}}\varepsilon\_{\text{dilicirctic}}}{\varepsilon\_{\text{metal}} + \varepsilon\_{\text{dilicirctic}}} + \Delta\beta} \tag{3}$$

The Δβ of the right expression denotes the effects of finite metal layer thickness and high refractive index of prism in real situation. In most fiber-optic SPR sensors, the wavelength interrogation method is employed. The refractive index sensitivity is defined as:

$$S\_n = \delta \lambda\_{\rm res} / \delta n\_{\rm s} \tag{4}$$

where δ*ns* is the change in the refractive index of the analyte and δλ*res* is the shift of resonant wavelength [6].

**Figure 1.** (**a**) The propagating surface plasmon polariton (SPP) at the metal–dielectric interface (Figure adapted with permission from reference [4]). (**b**) The schematic illustration of conventional Kretschmann-Raether prism configuration (Figure adapted with permission from reference [5]).

Along with the increasing demand for compact, highly-integrated, flexible and even in situ sensing devices, optical fiber-based SPR sensors receive more and more attention [7–11]. Various fiber-optic SPR configurations have been investigated to achieve highly sensitive SPR sensors. The key point in the design of the fiber-optic SPR sensor is to realize the phase-matching between the guided mode in fiber and the SPP at the metal-dielectric interface. Hence it is essential to coat the thin metal film at the surface of fiber structure where a strong evanescent field of guide mode can be exposed, leading to the strong SPP at the metal surface for effective light–matter interaction. Fiber gratings like long-period fiber grating (LPG) [12–14] and tilted fiber Bragg grating (TFBG) [15–17], tapered fiber [18–23], side-polished fiber [18,24], etc. have been demonstrated to be feasible for SPR sensing (Figure 2).

**Figure 2.** Fiber-optic surface plasmon resonance (SPR) sensor based on (**a**) long-period fiber grating (LPG) (Figure adapted with permission from reference [12]). (**b**) tilted fiber Bragg grating (TFBG) (Figure adapted with permission from reference [15]). (**c**) tapered fiber. (**d**) side-polished fiber (Figure (**c**) and (**d**) adapted with permission from reference [18]).

In recent years, MOFs are favored due to higher degree of integration, longer interaction distance and improved robustness, that the cladding air holes can function as microfluidic channels for liquid or gas analyte infiltration [25–27]. With the distinctive design of core dimension and cladding air holes arrangement, the thin metal films coated on the inner surface of air holes can effectively interact with the evanescent field of the core mode, which grants access to infiltrated analyte to the strong SPP. Numerous MOF structures have been proposed, including hexagonal MOFs [28], semicircular- channel MOFs [29], exposed-core MOFs [30], etc. In most MOF-based SPR designs, the prime consideration is

to facilitate simple analyte infiltration and large interaction area. Hence, birefringence commonly exists in MOF-based SPRs. Based on Equation (1), birefringence leads to the offset between SPR wavelengths corresponding to two orthogonal polarizations of core mode. When external perturbations such as fiber bending, twisting and pressure are applied on the fiber, the coupling from the desired mode polarization to undesired mode polarization will occur. Therefore, the overall SPR peak, which is the superposition of SPR of two orthogonal polarizations, will be unstable, leading to inaccurate sensing results.

To address the issue of birefringence induced measurement instability, polarization-maintaining MOF-based SPR sensor with high birefringence could be a promising solution. A large birefringence can be realized in a near-panda MOF with the two central air holes of the photonic-crystal arranged cladding holes enlarged (Figure 3a). The material of the MOF is fused silica. The enlarged two central holes can facilitate easier thin noble metal film deposition and analyte infiltration. Strong surface plasmons can be excited by the x-polarized fundamental core mode with the thin gold film deposited on the inner walls of central holes. As discussed earlier, SPP can only be excited by the TM-polarized incident light (i.e., the polarization perpendicular to the metal film surface), and y-polarized core mode corresponds to a much weaker SPP compared with that of x-polarized mode (Figure 3b). This indicates the SPR sensing output is predominated by the plasmonic behaviors of x-polarized code mode. For a low-birefringent MOF, of which the diameter of central holes (d2) is comparable to that of other cladding holes (d1) (e.g., d1/d2 = 0.95), both x- and y-polarized mode can excite relatively strong SPP. As a result, the existence of unwanted polarization could induce an offset of overall resonant wavelength as high as 0.67 nm from that of the desired polarization, which means the SPR sensing accuracy is considerably compromised (Figure 3c). On the contrary, even though a highly-birefringent MOF consists of two modal polarizations corresponding to even larger resonant wavelength difference, the immensely suppressed SPP of unwanted polarization has a bare influence on the overall resonant wavelength and the sensing accuracy. For instance, the wavelength offset of the proposed highly birefringent near-panda MOF with d1/d2 = 0.4 is as small as 0.06 nm (Figure 3d).

Based on the finite element method (FEM) simulation of photonic-crystal arranged MOFs with different d1/d2 ratios, the relation between phase birefringence and sensing inaccuracy can be deduced. As shown in Figure 3e, the resonant wavelength offset could increase to be as large as 18.89 nm when the phase birefringence increases from ~4 <sup>×</sup> <sup>10</sup>−<sup>5</sup> to ~1 <sup>×</sup> <sup>10</sup><sup>−</sup>4. When the phase birefringence exceeds beyond a threshold (~1 <sup>×</sup> 10−4), the wavelength offset effectively reduces and even tends toward 0 after 4 <sup>×</sup> <sup>10</sup>−<sup>4</sup> phase birefringence. The investigation indicates that small birefringence that commonly exists in MOF-based SPR sensors could induce non-negligible undesired resonant wavelength offset, which affects sensing accuracy. The proposed highly-birefringent MOF with intentionally introduced large phase birefringence ~4.2 <sup>×</sup> <sup>10</sup>−<sup>4</sup> can effectively suppress such impact of polarization crosstalk to be extremely small. In addition, more expanded central holes enhance the plasmon–matter interaction, thereby providing higher sensitivity. Figure 3f compares the sensitivities when d1/d2 = 0.4, 0.5, 0.6 and 1.0. It is clear that the sensitivity is improved when the central holes expand. At a high analyte refractive index range of 1.37–1.38, the proposed highly-birefringent MOF SPR sensor can achieve a sensitivity as high as 3000 nm/RIU.

Besides optimizing the design of fiber structure, integrating functional nanomaterials with a fiber-optic platform can also effectively promote the light-matter interaction. In the past decade, 2D materials have drawn extensive attention in various research fields including the highly integrated sensors. The extremely large surface-to-volume ratio, in situ plasmonic properties tunability and near field confinement are the great advantages of 2D materials in sensing applications [33–36]. The plasmonics of most common 2D materials such as graphene and MX2 fall in MIR or terahertz regions, which are not compatible with the well-developed optical communication window even though they can achieve superior plasmonic sensing performance [37,38]. Therefore, numerous research efforts focus on enhancing the plasmon–matter interaction by applying 2D material/metal film hybrid structures to SPR configurations. For instance, the thin gold film in conventional Kretschmann configuration has

been upgraded to graphene/gold [6,39–41], graphene oxide/gold [42–45], graphene-MoS2/gold [46], etc. hybrid film-like architectures (Figure 4). It is proven that the intensity of SPP on the gold film surface can be effectively strengthened by the seamlessly integrated graphene layer. When graphene and gold are in contact, the work function difference between the two materials (4.5 eV for graphene and 5.54 eV for gold) causes electrons to flow from graphene to gold to equilibrate the Fermi levels [47,48]. As a result, the electron density at the gold film surface increases as the graphene becomes p-type doped. Therefore, a stronger SPP so a higher sensitivity can be achieved.

**Figure 3.** (**a**) The configuration of proposed highly birefringent microstructured optical fiber (MOF). (**b**) x-polarized and y-polarized core mode pattern of the SPR sensor (d1/d2 = 0.4) (Figure adapted with permission from reference [31]). (**c**) Attenuation spectra of highly birefringent MOF when d1/d2 = 0.95 and (**d**) d1/d2 = 0.4. (**e**) The variation of wavelength offset along with phase birefringence. (Inset) The x-polarized core mode pattern. (**f**) The SPR sensitivities when d1/d2 = 1.0, 0.6, 0.5 and 0.4 respectively. (Inset) The x-polarized core mode (Figure adapted with permission from reference [32]).

**Figure 4.** The prism based SPR configurations with hybrid plasmonic structures of (**a**) single-layer graphene/gold (Figure adapted with permission from reference [39]); (**b**) multilayer graphene/Py/gold (Figure adapted with permission from reference [42]); (**c**) graphene-MoS2/gold (Figure adapted with permission from reference [46]).

Even though the 2D material/metal film hybrid structure had been widely proposed on the prism-based SPR configuration, systematic analysis and experimental demonstration of integrating such hybrid plasmonic structure with flexible waveguides such as optical fibers were rare. As a proof of concept, a graphene-on-gold hybrid structure is proposed to be seamlessly integrated with a side-polished optical fiber, purposing to demonstrate that the 2D material/metal hybrid structures could achieve enhanced plasmonic biosensing performance on flexible waveguide platforms. As illustrated in Figure 5a, the exposed evanescent field of guided core mode interacts with the graphene-on-gold structure deposited at the surface of polished facet of optical fiber, leading to strong SPP-biomolecules interaction. Meanwhile, the single graphene layer functions as excellent surface functionalization of the thin gold film. Since the SPP at the gold film surface exponentially decays with the penetration depth, the thickness of surface functionalization is a crucial factor that affects sensitivity. The graphene layer, as thin as 0.34 nm, could hardly compromise the SPR sensitivity [49]. Moreover, the carbon atoms of graphene arranged in honeycomb format can easily form π-stacking interaction with the aromatic rings commonly existed in biomolecules [50]. Hence it facilitates effective adsorption of target biomolecules such as ssDNA, providing high sensitivity and low limit of detection (LOD).

Simulation can verify the SPP enhancement capability of the additional graphene sheet on conventional gold film coated side-polished fiber. The inset of Figure 5b plots the whole electrical field distribution of guided core mode in the fiber as well as the SPP on the side-polished facet. The magnified field distribution of the SPP at the gold/graphene surface is shown in Figure 5b. As expected, introducing single or multiple graphene layers can effectively enhance the SPP intensity on the thin gold film surface, which benefited from the electrons transfer, as explained above. Another interesting finding in the simulation is that bilayer or multi-layer graphene slightly compromises the SPP intensity compared with the single-layer graphene. This is due to the electrons' energy loss induced by the increase of graphene layers [49]. Therefore, with the SPP intensity boosted by ~30.2%, a single graphene layer most enhances the plasmonic sensing behavior. The experimental results further verify that the graphene-on-gold hybrid structure can effectively improve the plasmonic sensing behavior. Figure 5c compares the resonant peaks of the conventional thin gold film coated side-polished fiber and the graphene-on-gold hybrid structure integrated side-polished fiber when both sensing configurations are immersed in deionized (DI) water. The inset of Figure 5c shows the microscopic view of the boundary of transferred graphene on the thin gold film coated side-polished fiber facet. The graphene-on-gold hybrid structure corresponds to a deeper resonant peak, indicating a stronger SPP intensity, which matches well with the simulation.

**Figure 5.** (**a**) The schematic illustration of proposed hybrid graphene-on-gold SPR sensor. The nucleobases of target ssDNA molecules can form stable π-stacking interaction with the honeycomb arrange carbon atoms of graphene. (**b**) The comparison of electric field intensities when the SPP is excited by bare thin gold film, single layer graphene/gold, 2-layer graphene/gold and 3-layer graphene/gold. (Inset) The electric field distribution over the entire fiber-optic graphene/gold hybrid structure. (**c**) The comparison of fiber transmission spectra with and without single graphene layer (Inset) The microscopic view of the single graphene layer transferred on the side-polished fiber. (**d**) The transmission spectra variation of proposed hybrid plasmonic sensor along with the increase of ssDNA concentration. (**e**) The comparison of sensitivities to ssDNA concentration of fiber-optic plasmonic sensors with and without graphene layer (Figure adapted with permission from reference [51]).

The biosensing capability of the proposed plasmonic hybrid SPR configuration can be validated by detecting ssDNA concentration. ssDNA quantization provides biomedical significance in gene expression, DNA sequencing and polymerase chain reaction (PCR) [52]. Figure 5d shows the magnified SPR peaks of the biosensing platform with the incrementing ssDNA concentration. This can be explained by Equation (1) that the surrounding refractive index of the plasmonic architecture is increased due to the efficient adsorption of ssDNA molecules on the graphene surface via π-stacking interaction. Also, the SPP evanescent field is scattered by the bonding of ssDNA molecules, which further induces transmission loss, thereby a deeper SPR peak. The LOD of the biosensor to ssDNA molecules is as small as 1 pM based on the distinguishable enhancement of the SPR peak (the red curve of Figure 5d). To experimentally verify that the biosensing performance is improved by the additional graphene layer, a conventional thin gold film based side-polished fiber-optic SPR sensor is prepared and applied to measure the same ssDNA solutions. The comparison of sensitivities corresponding to the two structures in Figure 5e can obviously indicate that the graphene-on-gold hybrid structure can effectively improve the sensitivity almost two-fold.

Wei et al. also compared the theoretically and experimentally performances of the fiber-optic SPR sensors with and without graphene in evaluating bovine serum albumin (BSA) concentration [6]. As shown in Figure 6a,b, the unclad portion of a plastic optical fiber is deposited with a gold film, and the graphene monolayer is transferred to the gold surface by PMMA. Figure 6d shows the variation of the reflection spectra of the graphene/Au fiber-optic SPR sensor with BSA concentration ranging from 0 to 2 mg/mL, which displays a 13.8-nm redshift compared with 6.1 nm for the Au fiber-optic SPR sensor (Figure 6c). After linearly fitting the resonant wavelengths and BSA concentration, the sensitivity of the graphene/Au hybrid sensor is 7.01 nm/(mg/mL), while the sensor without graphene is only 2.98 nm/(mg/mL). Additionally, regarding the full width half maximum (FWHM) of the two sensors, the graphene/Au hybrid sensor also possesses a more obvious variation tendency.

The finite element analysis (FEA) method based on COMSOL Multiphysics is established to clarify the improved sensing capability by graphene. The calculated electric field mode diagrams of fiber-optic SPR sensors with and without graphene are displayed in Figure 6e,f, respectively. On the sensing medium/Au interface, both sensors exhibit the similar confined electric field distributions, while the presence of graphene can strengthen the confined electric field with a maximum intensity of 6.4 <sup>×</sup> 104 <sup>V</sup>/m. Furthermore, in Figure 6g, the electric field intensity of the two structures perpendicular to the sensing interface (white dashed line) is extracted and compared. As can be seen, both electric field intensity exponentially decays along with the distance from Au film, and graphene/Au hybrid structure reveals a more considerable penetration depth of 256 nm, thus improving the sensitivity to the surrounding medium.

Based on the same mechanism, Wang et al. developed an SPR immunosensor employing graphene oxide (GO)-modified photonic crystal fiber (PCF) for human IgG detection [44]. PCF is a type of MOF consisting of a honeycomb structure with air holes, infiltrating liquid crystals into these air holes enables PCF tunable optical characteristics [53–60]. As shown in Figure 7a, the PCF with five layers of air holes is spliced between two multimode fibers (MMFs). After being deposited with Au film, the fiber is cleaned with piranha and then modified with Mercapto ethylamine (MEA) to enrich amine (-NH2) groups for further reaction with epoxy groups on GO. Subsequently, the EDC/NHS system is used to activate the carboxyl of GO, and anti-IgG is directionally linked by staphylococcal protein A (SPA) orientation. Finally, BSA is introduced to block the free SPA surface, and the sensor is ready for human IgG sensing.

**Figure 6.** (**a**) Fabrication of Au coated fiber and preparation of graphene monolayer. (**b**) The schematic illustration of cross-section view of the proposed graphene/Au fiber-optic SPR sensor. (**c**) Reflection spectra of Au (**c**) and graphene/Au (**d**) fiber-optic SPR sensor with varying bovine serum albumin (BSA) concentration. (**e**) Finite element analysis (FEA) simulation of electric field distribution of fiber-optic SPR sensors with (**e**) and without (**f**) graphene. (**g**) Electric field decaying along Y-direction (Figure adapted with permission from reference [6]).

**Figure 7.** (**a**) Fabrication of graphene oxide (GO)-modified SPR immunosensor. (**b**) Fitting curve of wavelength shift versus human IgG concentration. (**c**) Local enlarged drawing (Figure adapted with permission from reference [44]).

The immune reaction between anti-IgG and human IgG will cause wavelength redshift, as shown in Figure 7b, the wavelength shift and human IgG concentration can be fitted using the Langmuir equation. Compared with the Au-SPA sensor, Au/GO-SPA sensor exhibits a distinct redshift of 0.02 nm to 21.57 nm. After zooming in (Figure 7c), it can be observed that the LOD of Au/GO-SPA sensor (0.01 μg/mL) is 30 times lower than the Au-SPA sensor (0.3 μg/mL), which indicated GO significantly enhanced the immunosensor sensitivity.

Hu et al. also incorporate the graphene monolayer on a gold-coated TFBG (Figure 8a), the TBFG is further functionalized with ssDNAs by π–π stacking for dopamine detection. As shown in Figure 8d, there is an obvious differential amplitude increase when dopamine concentration raises from 10−<sup>14</sup> M to 10v13 M, and a quite linear correlation (R<sup>2</sup> = 99%) is observed over dopamine concentration from 10−<sup>13</sup> M to 10v8 M. TFBG enables the optic-fiber sensor with high RI sensitivity, narrow cladding and innate insensitivity to temperature and optical power fluctuations, which is feasible for biomedical sensing.

**Figure 8.** (**a**) The schematic illustration of proposed graphene/Au TFBG fiber-optic sensor (polarimetric sensing characteristic of TFBG and the energy distribution along fiber cross section). (**b**) Scanning electron microscope (SEM) image of graphene monolayer coated on Au surface. (**c**) photograph of the whole fiber-optic probe. (**d**) The differential amplitude output versus dopamine concentration. (**e**) The linear relationship between differential amplitude and dopamine concentration. (Figure adapted with permission from reference [41]).

Although the 2D material/metal hybrid structures facilitate remarkable light-matter interaction in plasmonic sensing, the intrinsic SPP of most common 2D materials (e.g., graphene and MX2) located at the MIR range is almost impossible for practical applications. Therefore, an alternative class of 2D plasmonic material, heavily doped ultrathin TMOs, have captured research attention in recent years aiming for manipulating the intrinsic plasmonics of 2D materials with exceptional field confinement and in situ plasmonic tunability in the frequently used visible and NIR optical window [61–64]. To realize SPP in visible or NIR frequencies, sufficient free carrier concentration must be achieved in 2D materials. The unique character of outer-d valence electrons enables TMOs to achieve sufficient free carrier doping via ionic intercalation. Taking the most representative TMOs, molybdenum trioxide (MoO3) and tungsten oxide (WO3), as examples, free electrons can be abundantly doped by introducing oxygen vacancies in the TMO lattice [65,66]. Therefore, the plasmonic behavior of 2D TMOs can be easily tuned by manipulating the oxygen vacancies. So far, the tunable plasmonics of heavily doped MoO3 nanoflakes in visible or NIR region has been most widely studied, yet the exploration on integrating such emerging 2D materials with optical devices especially the highly-integrated waveguide based sensing devices is very limited.

Driven by the purpose of investigating the potential of 2D TMOs on highly-integrated plasmonic devices, a biosensor based on a microfiber functionalized with α-MoO3 nanoflakes is developed and validated by BSA molecules detection. As shown in Figure 9a, few-layer α-MoO3 nanoflakes are synthesized by the liquid-phase exfoliation method [67] and then heavily doped with free electrons via an H<sup>+</sup> intercalation process [68]. After doping, a sub-stoichiometric <sup>α</sup>-MoO3−<sup>x</sup> nanoflakes solution with strong SPP at the NIR region is formed. Pristine MoO3 only introduces absorption at ultraviolet (UV) wavelengths, which is due to the large bandgap of 3.2 eV [69]. After electrons are increasingly doped, a distinct absorption peak appears and enhances at 700–800 nm range, in the meantime, undergoes a blueshift. This phenomenon can be explained by Drude model that the plasma frequency is inversely correlated to electron density [70,71].

**Figure 9.** (**a**) The schematic illustration of heavily-doped MoO3−<sup>x</sup> nanoflakes based hybrid fiber-optic plasmonic biosensor. (Inset 1) The crystal structure of α-MoO3 lattice. (Inset 2) Molecular structure of BSA. (**b**) The fluorescent microscopic view of MoO3−<sup>x</sup> nanoflakes functionalized microfibers coated with different concentrations of dye labelled BSA molecules. (**c**) The absorption spectra of MoO3−<sup>x</sup> nanoflakes solutions mixed with different BSA concentrations. (**d**) Transmission spectrum variation of proposed hybrid plasmonic biosensor along with increasing BSA concentration. (**e**) The linear increase of transmission minimum against log-scale BSA concentration (Figure adapted with permission fromreference [68]).

The MoO3−<sup>x</sup> nanoflakes can be stably immobilized on the microfiber surface via electrostatic interaction. Since MoO3−<sup>x</sup> is positively charged, the microfiber surface functionalized with evenly distributed negative charges (e.g., self-assembled poly(allylamine) (PAA)/poly(styrene sulfonate) (PSS) bilayer) applies strong attraction to the nanoflakes. Similarly, the immobilized positively charged MoO3−<sup>x</sup> nanoflakes on the microfiber surface can effectively attract negatively charged target molecules, such as BSA [67]. Dye-labeled BSA molecules are adopted to verify the effectiveness of electrostatic interaction-based target molecule adsorption as well as fiber surface functionalization. Figure 9b shows the fluorescent microscope views of four MoO3−<sup>x</sup> nanoflake-deposited microfibers after immersing in different concentrations of dye-labeled BSA solutions. It is evident that the fiber brightens as the BSA concentration increases. Also, the even brightness on the fiber surface implies the uniformity of adsorbed BSA molecules so as the MoO3−<sup>x</sup> nanoflakes.

The binding of negatively charged BSA molecules on the MoO3−<sup>x</sup> nanoflakes surface impacts the plasmonic behavior. When MoO3−<sup>x</sup> nanoflakes suspensions mix with different concentrations of BSA solution, the absorption peak of MoO3−<sup>x</sup> weakens as the BSA concentration increases (Figure 9c). This is due to the free electrons at the MoO3−<sup>x</sup> surface being repelled by the negatively charged BSA molecules, resulting in the reduced free electron density involved in the plasmonic resonance [61,64,72]. Therefore, a fiber-optic sensor based on MoO3−<sup>x</sup> nanoflakes shows a unique characteristic that the resonance peak on the fiber transmission spectrum gradually shallows along with the increasing concentration of target BSA (Figure 9d). Profited from the full utilization of a high aspect ratio of 2D MoO3−x, a LOD of BSA as low as 1 pg/mL is achieved. Moreover, the transmission minimum of the plasmonic resonance peak provides a linear response to the log-scale BSA concentration (Figure 9e).

With the vigorous development of material science, there is abundant research to introduce diverse materials into fiber-optic SPR sensors [73–77]. For instance, Santos et al. propose a refractive index sensor by combining Al2O3-Ag metamaterial film with D-type PCF fiber, in which the sensor performance can be adjusted by the thickness and component of metamaterial [78]. Semwal et al. wrapped Ag-coated optical fiber with the enzyme (ADH) and coenzyme (NAD)–containing hydrogel to establish an ethanol sensor. These will surely boost the advancement of fiber-optic fiber sensors [79].

#### **3. Fiber-Optic Localized Surface Plasmon Resonance Sensors**

By contrast with the propagating SPP at a thin metal film surface, the resonant electron oscillation induced by light interacting with a metallic nanoparticle is non-propagating due to the particle size restriction. Therefore, it is called localized SPR (LSPR). LSPR can be excited when the oscillation frequency of nanoparticle electron cloud matches with the frequency of incident light (Figure 10) [80,81]. A proper model for understanding how incident light is scattered and absorbed by a nanoparticle with a diameter much smaller than the wavelength is the Mie theory. The Mie theory constructs a model to deduce the extinction cross-section of nanoparticle based on the assumption that the nanoparticle is a homogeneous conducting sphere:

$$
\sigma\_{\rm ext} = 9 \left( \frac{\alpha}{\varepsilon} \right) (\varepsilon\_{\rm dielectric})^{\frac{3}{2}V} \frac{\varepsilon\_{\rm metal}^{\prime\prime}}{\left( \varepsilon\_{\rm metal}^{\prime} + 2 \varepsilon\_{\rm dielectric} \right)^{2} + \left( \varepsilon\_{\rm metal}^{\prime\prime} \right)^{2}} \tag{5}
$$

where *V* is the volume of nanoparticle, and ε*'metal* and ε*"metal* are the real and the imaginary parts of the metal-dielectric function, respectively, in the Drude model [82]:

$$
\varepsilon\_{\text{metal}}' = 1 - \frac{\omega\_p^2}{(\omega^2 + \chi^2)} \tag{6}
$$

$$
\varepsilon''\_{metal} = \frac{\omega\_p^2 \gamma}{(\omega^2 + \gamma^2)\omega} \tag{7}
$$

where γ is the damping of electron oscillation and ω*<sup>p</sup>* is the bulk plasma frequency. More detailed definitions of γ and ω*<sup>p</sup>* can be found in [83]. Since LSPR operation frequencies are generally within the visible and NIR optical windows where γω*p*, Equation (7) can be simplified as:

$$
\varepsilon'\_{metal} = 1 - \frac{\alpha\_p}{\alpha^2} \tag{8}
$$

Based on Equation (5), the resonance is satisfied (i.e., the extinction cross-section is maximum) when ε*'metal* = –2ε*dielectric*. The LSPR resonant frequency is thereby expressed as:

$$
\omega\_{LSPR} = \frac{\omega\_p}{\sqrt{2\varepsilon\_{dielectric} + 1}} \tag{9}
$$

Furthermore, for dielectric medium, ε*dielectric* = *n<sup>2</sup> dielectric*. Therefore, the refractive index of the ambient dielectric medium of the nanoparticle impacts the LSPR resonant wavelength:

$$
\lambda\_{LSPR} = \lambda\_p \sqrt{2r\_{dielectric}^2 + 1} \tag{10}
$$

**Figure 10.** The schematic diagram of localized surface plasmon resonance (Figure adapted with permission from Ref. [81]).

Similar to SPR-based fiber-optic sensing platforms, LSPR-based fiber-optic devices have also captured intensive research attention [84–89]. Various fiber structures such as microfiber, cascaded unclad fiber, fiber endface, etc. have been integrated with silver or gold nanoparticles and shown promising plasmonic sensing performance (Figure 11). To achieve efficient selectivity to analyte molecules, it is necessary to apply surface functionalization on metallic nanoparticles. Taking the most widely employed gold nanoparticle as an example, many effective functionalization strategies have been proven, such as biomolecule coating [90–92], ligand substitution [93], polymer deposition [94,95], etc. However, when sensitivity is the crucial factor of plasmonic sensors, it is critical to keep the surface functionalization as thin as possible. Due to the evanescent nature of surface plasmon, thick surface functionalization considerably compromises the plasmon–matter interaction. For instance, a study has compared the LODs and sensitivities of two functionalization strategies with thicknesses of 4.24 nm and 0.96 nm respectively and shown that surface functionalization thinner than 1 nm significantly improves the sensing performance [96]. In such cases, macrocyclic supramolecules have shown the potential to meet the challenges of achieving both sub-nanometer functionalization thickness and target molecules recognition.

**Figure 11.** The LSPR devices based on various optical fiber structures such as (**a**) cascaded unclad fiber (Figure adapted with permission from reference [97]); (**b**) microfiber (Figure adapted with permission from reference [98]); (**c**) optical fiber endface (Figure adapted with permission from reference [99]).

Benefiting from their macrocyclic cavities, macrocyclic supramolecules like cyclodextrins (CDs), cucurbiturils, pillararenes, calixarenes, etc. show excellent molecular recognition capability by the host-guest interaction. The host–guest interaction is noncovalent interaction between macrocyclic supramolecules and corresponding guest molecules to form inclusion complexations [100]. Encouragingly, it is proven that host–guest interaction is a more effective target molecule recognition and adsorption mechanism compared with the conventional biomolecule-ligand binding [101]. Another advantage of macrocyclic supramolecules being surface functionalization of metallic nanoparticles is the heights of their macrocyclic cavities being normally less than 1 nm [102–104], which facilitates sensitive molecular detection as discussed above. In addition, the macrocyclic supramolecules also eliminate the cytotoxicity of nanoparticles that is often favored in bio-medical sensing applications. In recent years, studies have been carried out to achieve functionalizing metallic nanoparticles with macrocyclic supramolecules during the synthesis of nanoparticles instead of through post-processing surface modification. In most of these attempts (Figure 12), however, harsh reducing reagents such as thiols, NaBH4, NaOH have to be introduced, which violates the purpose of achieving biocompatibility in many LSPR biosensing applications [105–107]. Therefore, inspired by the method proposed by Zhao et al. [108], where CDs act as both reducing and capping agent for AuNPs synthesis, a microfiber based LSPR biosensor is developed to comprehensively investigate the plasmonic sensing potential of one-step synthesized macrocyclic supramolecules decorated metallic nanoparticles.

Figure 13 illustrates the configuration of LSPR biosensor based on a microfiber integrated with one-step synthesized β-CD-capped AuNPs [109]. Functioning as both the reducing and capping agent, β-CD facilitates the AuNPs formation and forms biocompatible functionalization layer via the conjunction of carboxyl groups and gold surface. The evanescent field of guide mode in microfiber leaks out at the tapered portion and interacts with immobilized AuNPs at the fiber surface, leading to a strong LSPR peak on the transmission spectrum. Cholesterol, the guest molecule of β-CD [108,110], is employed to validate the biosensing performance of the proposed LSPR device. The sterol groups of cholesterol molecules can tightly fit into the β-CD macrocyclic cavity, meanwhile forming stable host–guest interaction through the hydrophobic associations [111,112].

**Figure 12.** (**a**) The synthesis formulas of (**a**) carboxylatopillar[5]arene capped gold nanoparticles (AuNPs) ((Figure adapted with permission from reference [105]). (**b**) Cyclodextrin (CD)-capped AuNPs (Figure adapted with permission from Ref. [106]). (**c**) CD capped silver nanoparticles, AuNPs and Agcore-Aushell/Aucore-Agshell bimetallic nanoparticles (Figure adapted with permission from reference [107]).

**Figure 13.** The schematic illustration of the β-CD-capped AuNPs based fiber-optic biosensor. (Inset1): The conjunction between β-CD molecule and AuNP surface. (Inset2): The molecular structure of cholesterol (Figure adapted with permission from reference [109]).

The as-synthesized β-CD-capped AuNPs show good uniformity in particle size with diameters range from ~18 nm to ~21 nm. The dynamic light scattering (DLS) characterization of the AuNPs further verifies the observation [109]. The resonance band of the absorption peak of β-CD-capped AuNPs solution with linewidth as narrow as 47 nm also indicates the monodispersity of the particles, which is comparable with that of conventionally synthesized AuNPs. Proton nuclear magnetic resonance ( 1H NMR) and Fourier transform infrared (FTIR) spectra are performed to further illustrate that the hydroxyl groups in β-CDs mainly contribute to reducing Au3<sup>+</sup> ions to Au<sup>0</sup> atoms [113].

The as-synthesized AuNPs are negatively charged. Hence it can be stably immobilized on positively charged microfiber surface (e.g., functionalize the fiber surface with homogeneous PAA layer) via electrostatic attraction. As shown in Figure 14a,b the prepared microfiber is 4 μm in diameter and decorated with evenly distributed AuNPs. The attached AuNPs induce a deep resonance band centered at 530.7 nm on the microfiber transmission spectrum. When the fiber-optic sensing device is sequentially immersed in cholesterol solutions with concentrations ranging from 5 aM to 0.5 μM, the LSPR resonance band gradually deepens along with the increasing cholesterol concentration meanwhile the resonant wavelength shifts from 530.7 nm to 531.4 nm (Figure 14c). Such an ultra-low LOD of 5 aM is profited from the highly efficient host–guest interaction between β-CD and cholesterol. The transmission minimum of the resonance band can be taken as the sensing parameter and provides a linear response to the log-scale cholesterol concentration (Figure 14d).

The selectivity of the proposed biosensor to cholesterol is validated by an interference study, where common interfering substances in human serum such as glutamic acid, cysteine, ascorbic acid, dopamine and human serum albumin (HSA) are introduced. Figure 14e shows the real-time average transmission intensity within 530–535 nm of microfiber when the interfering substances are introduced during the detection of cholesterol. It is clear that the β-CD-capped AuNPs based fiber-optic sensor only responses to cholesterol molecules but not interfered by other substances. To further validate the cholesterol recognition capability of the proposed sensor, recovery experiments are also carried out to evaluate the accuracy of detecting real human serum samples diluted by a factor of 1014 and spiked with different cholesterol concentrations. As summarized in Table 1, the measurement of cholesterol concentration in the unspiked human serum sample is 4.23 mM. The measurement of the same sample using commercial blood cholesterol monitor is 4.35 mM, which indicates the proposed fiber-optic biosensor is reliable. In addition, the recoveries of the spiked samples are 105.2–112.2%, which is also within a satisfactory range, further verifies the accuracy of the proposed sensor. Therefore, it indicates the tremendous plasmonic sensing potential of highly integrated fiber-optic sensors based on the macrocyclic supramolecules modified metallic nanoparticles.


**Table 1.** Recovery results of detecting cholesterol in human serum samples.

\* The values are mean of 4 independent experiments ± standard deviation (Reprinted with permission from reference [110]).

Another polysaccharide, chitosan, has also been used for AuNPs synthesis as a reducing and stabilizing agent [114]. Sadani et al. immobilize the synthesized chitosan-capped AuNPs (ChGnP) with a diameter of 20 nm on U-bent fiber for mercury (Hg(II)) detection [115]. The U-bent fiber is firstly incubating in (3-Aminopropyl)triethoxysilane (APTES) solution to enrich amine on the fiber surface, with glutaraldehyde crosslinking followed. Thereafter, BSA is linked to glutaraldehyde for further AuNPs' immobilization (Figure 15a,b)

**Figure 14.** (**a**) The SEM image of 4-μm-diameter microfiber. (**b**) The distribution of β-CD-capped AuNPs on the microfiber surface. (**c**) Transmission spectrum variation of microfiber based hybrid plasmonic biosensor along with increasing cholesterol concentration. (**d**) The linear decrease of transmission minimum against log-scale cholesterol concentration. (**e**) The real-time average transmission intensity within 530–535 nm of microfiber when the interfering substances are introduced during cholesterol detection (Figure adapted with permission from reference [109]).

**Figure 15.** (**a**) The schematic illustration of the detection system. (**b**) Functionalization of chitosan-capped AuNPs on U-bent fiber. (**c**) Selection of optimal receptor for Hg(II) detection. (**d**) The linear increase of absorbance against Hg(II) concentration. (**e**) Absorbance at 520 nm for 1 μM individual metal ions detection. (**f**) Absorbance increasement against time for 1 μM metal ions mixture detection (Figure adapted with permission from reference [115]).

The sensitivity and selectivity of four sensors: BSA attaching to citrate capped AuNPs (BSA on GnP), polyanionic poly(sodium 4-styrenesulfonate) (PSS) immobilized ChGnP (ChGnP on PSS), fluorescent BSA-Au nanoclusters (BSA-AuNC) and BSA immobilized ChGnP (ChGnP on BSA) are compared. As shown in Figure 15c, compared to ChGnP on the BSA system, the first three show more deficient absorbance at the same Hg(II) concentration, and only the BSA-AuNC exists insignificant selectivity. The proposed ChGnP on the BSA LSPR sensor shows a linear calibration curve from Hg(II) concentration 0.1 ppb to 540 ppb (Figure 15d). 1 μM of different metal ions are dissolved in DI water separately, the absorbance for Hg(II) is greater than 0.9 a.u. while all other control ions are less than 0.2 a.u. (Figure 15e). Also, as shown in Figure 15f, when 1 μM of diverse metal ions mixtures with and without Hg(II) are detected, the change of absorbance at 520 nm over time shows that only mixture with the presence of Hg (II) reveals significant enhancement, further proving the excellent selectivity. The chemisorbed of Hg(II) on lone pair electrons of N, O and S atoms in chitosan and BSA, the hydrophobic interaction and the Van-der-Waals interaction with thiol groups in BSA are hypothesized the dominant factors of sensitivity and selectivity towards Hg(II).

Lee et al. fabricate a fiber-optic LSPR sensor for the detection of ochratoxin A (OTA) utilizing aptamer-modified gold nanorods (GNRs) [116]. The GNRs are immobilized on the optical fiber by Au−S interaction, after being dipped into OTA solution, the LSPR spectrum is monitored exploiting the light reflection of a silver mirror at the end of the fiber. The aptamer's specific recognition of OTA induces an LSPR peak shift (Figure 16c), and OTA can be specifically and quantitatively detected with a LOD of 12.0 pM and excellent linear response. This fiber-optic LSPR sensor possesses superior simplicity, which only demands to dip into OTA solution. The methods and performances of the hybrid fiber-optic sensors referred to are summarized is Table 2.

**Figure 16.** (**a**) The schematic illustration of the fiber-optic LSPR aptasensor. (**b**) Localized magnification of the fiber surface with GNPs immbolized on. (**c**) LSPR shift after ochratoxin A (OTA) recognization (Figure adapted with permission from reference [116]).


**Table 2.** Summary of hybrid fiber-optic sensors in this review.

#### **4. Conclusions**

As discussed in this review, the proper design of MOFs with high birefringence provides wide possibilities in highly integrated microfluidic sensing devices with improved measurement accuracy and stability. Profiting from 2D material-based hybrid plasmonic structures, fiber-optic plasmonic sensors can deliver more promising sensing capability. The exceptional surface to volume ratio, near-field confinement and in situ plasmonic properties tunability of 2D materials facilitate the further enhancement of plasmon-matter interaction so as to enhance the sensitivity and LOD. In addition, the development of supramolecular chemistry brings new solutions in LSPR nanoparticles surface functionalization, leading to excellent target molecule selectivity via host–guest interaction. Given the numerous possibilities in optical design and hybrid plasmonic architectures construction, fiber-optic hybrid plasmonic sensors possess vast potential in various sensing scenarios with distinct advantages of high sensitivity, flexibility, miniaturization and high degree of integration. To further integrate the advances of specialty fibers and various nanomaterials, one major direction in the future is to achieve advanced functional fiber-based sensing by integrating multi-functional materials inside a single fiber or large-scale fabrics [117–135].

**Funding:** This work was supported in part by the Singapore Ministry of Education Academic Research Fund Tier 2 (MOE2019-T2-2-127), the Singapore Ministry of Education Academic Research Fund Tier 1 (MOE2019-T1-001-103 and MOE2019-T1-001-111) and the Singapore National Research Foundation Competitive Research Program (NRF-CRP18-2017-02). This work was also supported in part by Nanyang Technological University.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
