Recent advances in surface enhanced Raman spectroscopy (SERS): finite difference time domain (FDTD) method for SERS and sensing applications

There have been significant advancements in the field of surface-enhanced Raman spectroscopy (SERS). Despite being an ultra-sensitive analytical technique, challenges, such as how to get a proper match between the SERS substrate and light for better signal enhancement to obtain a stable, sensitive SERS substrate, prevent its widespread applications. Finite-difference timedomain (FDTD) method, a numerical tool for modeling computational electrodynamics, has recently been used to investigate SERS for understanding the underlying physics, and optimally design and fabricate SERS substrates for molecular analysis. In this review, we summarize the trend of using FDTD method in SERS studies by providing an introduction of fundamental principles, the studies of optical responses, electromagnetic (EM) field distribution, enhancement factor (EF) of SERS, the application in design and fabrication of SERS substrates, and SERS for biosensing and environmental analysis. Finally, the critical issues of using inherently approximate FDTD method and future improvement for solving EM problems and SERS applications are discussed.


Introduction
The basic concept of surface-enhanced Raman spectroscopy (SERS) is that the Raman scattering signal of molecules at or close to the surface could be enhanced to a factor of 10 10 -10 11 using large local field enhancements at metallic surfaces under the right conditions [1], [2]. SERS has been employed and widely used as a powerful spectroscopic tool for sensitive and specific detection of chemical [3], biological [4], and medical analytes [5]. In particular, it has progressed from studies of model systems on roughened electrodes to highly sophisticated studies, such as SERS was first reported over 40 years ago, and it has experienced the periods of highs and lows from the fundamental understanding of the phenomenon to the exploration of promising applications. The observation of the SERS spectra from single molecules led to a second wave of studies. It is currently a topic of great interest, with over 1000 papers published per year. This review has included only a few of these studies on the state-of-the art progress in SERS substrates and their applications in the FDTD-SERS field. In this review paper, we firstly provide a brief description of the theory behind the technique. Then, a systematic review is made on the FDTD simulations in SERS calculations by identifying different parameters such as EM field distribution and optical responses. Other issues regarding the method limitations, research challenges, and future trends are also discussed.

Theory of FDTD
The FDTD method has been widely used to solve Maxwell's equations in complex geometries in the fields of EM and photonics, and it contributes to the simulation of light scattering from metal particles. The basic mathematical and physics formalism behind the FDTD algorithm is discussed as an indispensable background in this section.

Physics basics
Each field component is solved on a discrete spatial and temporal grid cell named Yee Cell proposed in 1966, where an electric component is located on the edges of the box and the magnetic component is positioned on the faces. Moreover, FDTD is a time domain technique with E(t) and H(t). Results collected from the FDTD solver are automatically interpolated to the origin of each grid point. We also want to know the field as a function of wavelength, E(λ), or equivalently frequency, E(λ) [1].
FDTD method is used to solve Maxwell's equations in nonmagnetic materials: where H, E, and D are the magnetic, electric, and displacement fields, respectively. ε and µ0 are the complex relative dielectric constant and magnetic permeability coefficient, respectively.
Furthermore, in three dimensions, Maxwell equations have six EM field components. With the assumption that the structure is infinite in the z dimension and that the fields are independent of z, the Maxwell's equations are split into two independent groups of equations that can be solved in the x-y plane only, which results in the transverse electric (TE) and transverse magnetic (TM) equations. Then, we can use the components of Ex, Ey, and Hz to solve TE equations and those of Hx, Hy, and Ez to solve TM equations.
Considering that the function f(x,y,z,t) denotes the electric or magnetic field in the coordinate system, it can be discretized via the central difference approximation in both space and time: Benefiting from the Drude model developed and Fourier transform of the polarization used in the algorithm, FDTD method has been widely applied in SERS simulations [5], [7], [23].
Meanwhile, numerous commercial software have been implemented in the method, as they are simple and convenient to use, and can solve several types of problems. For example, FDTD simulations are used to measure the scattering enhancement of the arbitrarily complex geometries, with a linear simulation, thus making the calculation easier for setup and analysis, and giving broadband results.
The evolution of FDTD technique has progressed from one-dimensional (1D) to threedimensional (3D), coupled 3D unconditionally stable local 1D FDTD [24], finite-volume timedomain (FVTD) [25], etc. for more stable and less computational time, and to a high-order discontinuous Galerkin time domain (DGTD) which provides an easier way to handle elements of various types and shapes, irregular non-conforming meshes and varying polynomial degree with more flexibility [26], [27]. In addition, a generalized FDTD (G-FDTD) method has been developed for solving the time-dependent linear and nonlinear Schrödinger equation [28].
• Define the physical structures: focus on the key part that contributes to SERS, as under proper conditions, both the metallic nanoparticle (MNP) and the plasmonic surface of the substrate can produce LSPR and surface plasmon polariton (SPP). Moreover, the model for the optical constants of noble metals should be close to the real situation. • Define a simulation region and boundary conditions: set reasonable mesh size and adjust the mesh size first in the reasonable direction. Both reliable result and PC memory should be taken into account. Perfectly matched layer (PML) is used in most cases, while periodic boundary condition (BC) should be used when both the structures and EM fields are periodic. • Define a source of light: the incident light, a plane wave, usually propagated along the z direction with polarization in the x-direction. • Define monitors to record data for analysis: the middle section of the nanoparticle (NP) layer is often defined by the field monitor. • Define script commands: for optical response spectrum use.
As an example, Fig. 1 shows a schematic illustration of the FDTD model and additional simulation layout setting details. The spatial distribution of the near-field plasmonic E-field for various substrate configurations and AgNP loading densities was performed with a horizontal (x) E-polarization of the 633-nm excitation under normal incidence. The spatial parameters included a time step of 0.001 fs, a mesh size of 1 nm in all three directions with an orange box indicating the total mesh area, and a background index of 1.0 (air). It should be noted that a very small mesh can lead to large memory requirement and long simulation times. PML BC and periodic BC were applied along x-, y-, and z-boundaries of the unit cell, respectively. The indexes of Ag and substrates followed the data of optical constants of solids [29], [30].

Performance of FDTD method in SERS
From the theoretical description provided earlier, it is evident that the electrodynamic simulations for analyzing SERS can be performed using a number of variables, which results in different FDTD performance for SERS simulation with respect to optical response spectra, EM field distribution, and enhancement factor (EF). In this section, studies that investigate the effect of each performance in SERS are highlighted. In fact, several studies on SERS have been conducted, and this review cannot be exhaustive. Hence, the scope of the SERS field covered in this review article will be limited to recent SERS studies related to FDTD method.

Optical responses
Optical responses including extinction [31], absorption [32], scattering spectra [33], [34], transmittance [35], and reflectance [36], [37] could also be calculated by the FDTD method. The data of optical response versus wavelength are obtained. Among the response parameters, extinction is equal to absorption plus scattering. Table 1 lists the SERS substrates related to optical responses. We could use the FDTD method to obtain optical response spectra in complex structures. For example, Xu et al. used the Fabry-Pérot thin-film model to calculate the reflection coefficients r1 and r2 and the transmission coefficients t1 and t2 of the free-standing hole and the disk layers by FDTD simulations containing only the hole layer or only the disk layer.
Furthermore, the reflection and transmission coefficients, r and t, could be obtained for the whole structure, and the reflectance R and transmittance T can be then expressed as R = r 2 and T = t 2 [35].
It is worth noting that the maximum LSPR wavelength can be determined by the optical response. Modeling of hot spots mentioned earlier was attempted through calculation of the plasmonic near field using the FDTD method. It also required prior calculation of the far-field extinction spectra for the determination of the maximum LSPR wavelength [56]. This wavelength can help determine the part of the SERS substrate contributing to the resonance. The absorption spectrum was simulated to calculate the gap resonance of the proposed hole arrayed metal-insulator-metal (HA-MIM) structure. It showed an absorption peak at the wavelength of 757 nm while no peak for the single gold-layered structure, which revealed that the resonance in the HA-MIM structure was excited by the MIM configuration [32].
FDTD modeling can also be used to verify the experimental optical response. Experimental verification of the scattering spectra was carried out by Ye et al. [33] for each group with carbon (C) NP (CNP) as the Raman-active analyte as it is easily removed by the oxygen plasma [57]. The experimental observations were made based on the self-assembled monolayer (SAM) resulting from thiol-silver or gold bond interactions [58] with the substrate. The reflectivity showed the sharp minimum at 43.6° in air, which is the resonance angle of a propagating surface plasmon (PSP) excitation, and was in good agreement with 43.2° obtained by theoretical evaluation using Fresnel equations as well as FDTD calculations [36]. Besides, the dark-field scattering spectra of the Al nanovoids for deep-ultraviolet (UV) SERS showed the simulated and experimental data. The simulated spectra showed their first-order resonance at 282 nm, which matched well with the experimental data [34]. More importantly, the good consistency between both these data could be applied to verify the accuracy of substrate fabrication. For instance, contour plots of calculated reflectance as functions of different nanodisk pitches in x-or ydirections and the wavelength were consistent with the experimental results, which indicated that the double-resonance structure had been fabricated as designed with high accuracy [37].
Few studies have been performed to date to investigate the direct effect of optical response on SERS intensity. Most of them focus on the relationship between optical response and electric field. A typical example is the study of the effect of the optical standing waves (SWs) or the leaky mode resonance (LMR) in modulating the total effective E-field. In order to address the combined effect of SWs and LMR in a nanowire (NW) system, detailed FDTD calculations were performed on the 2.5D semiconductor nanostructures decorated with AgNPs. The calculated absorption spectrum showed resonant modes and the calculated absorption peak spectral positions for Si; the results show that LMR ensures higher optical absorption and the SWs provide a modulation of the E-field, which is the strongest at the node locations of the SWs [30]. Furthermore, Fig. 2 showed the extinction spectra of the aqueous colloid of 80-nm AuNPs, AuNP arrays on indium tin oxide (ITO)-coated glass, and AuNP aggregates on ITO-coated glass. The optical spectra of the arrays are modified by the strong coupling between AuNPs, as confirmed by the FDTD calculations, which showed complex light-field patterns in the NP array at the "blue" (around 550 nm) and the "red" (around 850 nm) regions by incorporating the geometrical details of the NP arrays and the 20-nm ITO/glass substrate [31].

Electromagnetic (EM) field distribution
3D-FDTD is an effective tool in simulating the EM field distribution around the illuminated substrate of arbitrary shape by numerical calculation of Maxwell's equations. Several SERS studies have showed the figures of local field distribution [38], electric field distribution [39], electric field intensity [40], and EM near-field enhancement [41] after simulations. In fact, they can be classified as EM field distribution, which is one of the most important FTDT applications in SERS, as shown in Table 1.
The hot spots/active sites of SERS substrates can be confirmed by EM field distribution. The electric field distribution simulations illustrated that the 3D biomimetic SERS substrate provided the hot spot area with high density within a detection volume for enormous SERS enhancement [39]. In addition, the simulations of electric field intensity were performed to demonstrate the increase in the plasmonic field around the nanohole arrays with refractive indexbased tuning [40]. Extremely strong EM fields, which contribute to the extraordinarily strong SERS sensitivity, are located across the roughened Ag nanoplate surfaces due to the FDTD simulations. However, in order to accelerate the simulation process, the density of the pits on the simulated model was greatly reduced relative to the experimentally prepared rough Ag nanoplates [51]. By reviewing several studies, an interesting aspect about the hot spots or active sites was observed. The EM field tends to concentrate at the tips of the Au-Gr-coated polystyrene beads where the hot spots formed [43]. According to the electric field distribution in the system containing the prism/silver film/silver NP, the electric field at the gap between the silver film and a silver NP increased by about 4000 times [48]. Farcau and Astilean computed EM field distribution on dimers of Au-HS on top of the polystyrene spheres [47]. For simulation, they found that the enhanced EM fields are located at the junction between the two Au HSs, using a Drude dielectric function for bulk gold and a refractive index of 1.55 for the underlying polystyrene spheres [47]. Hence, the tip, gap, and junction should be focused during SERS research, as strong EM fields are located at these particular sites. The EM field distribution could be an indicator of the qualities of various kinds of SERS substrates with different inter-particle distances, gap regions, nanopatterns, and fabrication methods. Hu et al. reported the model simulations that demonstrated the effect of the interparticle distance from 220 to 260 nm on the EM field distribution, although they cannot accurately predict the characteristics of real Au/Fe3O4 because of the calculation complexity caused by the inhomogeneous and random distribution of the large spheres (Fe3O4 NPs) and the small ones (AuNPs) [42]. Besides, FDTD simulations were carried out to analyze the EM field distributions for the inter-particle and particle-film gap region, and the SERS EF of gold nanostars (GNSs) on a metal film as a function of inter-particle separation follows a broken power law function as shown in Fig. 3. The EM field intensities are high in the regions of interparticle and particle-film gap. Moreover, [E]/[E0] increases with the increase in inter-particle separation for GNS/Ag, while it increases slightly in the case of GNS/glass [49]. In order to fully understand the electric field distribution of each gold-patterned SERS substrate, and how the 20nm fold film interacts with gold nanodots, Zheng et al. showed different EM simulations. Besides, they also found that nanopattern geometry had the primary effect on SERS EF over substrate, as, no matter the substrate was covered with gold or not, 55-nm gold dot arrays obtained the strongest EFs [44]. The electric field enhancements of the Au nanorod (AuNRs) assembly motifs in different orientations, side-by-side (SS), end-to-end (EE), and end-to-side (ES), were estimated using the FDTD method. By changing the laser light polarized along the assembled axis to it polarized perpendicularly to the assembled axis, Zhong et al. found that the highest EF in the simulated calculation is taken by the laser light polarized along the assembled axis with the sequence of EE [45].
The EM distribution could also be further applied to confirm EM wave direction and to verify the presence of other effects that contributed to SERS. For example, the local field distribution generated by ring-like assembled gold NRs (GNRs) showed the plane EM wave was polarized along the x-axis and propagated along the z-axis [38]. The EM distribution of anodic aluminum oxide (AAO)/Al-based Ag nanostructure arrays was used to demonstrate that the local EM field enhancement and the corresponding SERS are determined by both SPR and interference effects [54]. More importantly, the EF can be roughly calculated from the EM field distribution and has been discussed in section 3.3.
Furthermore, attempts have been made to increase the accuracy of the EM field distribution. The simulation volume needs to be defined and the FDTD simulation parameters could be chosen from the experimental results. For example, the size and shape of AuNRs were taken from the transmission electron microscopy (TEM) statistical data, while the NR orientations were taken into account from the scanning electron microscope (SEM) images [38]. Higher resolution certainly indicates higher computational requirements. A very unique example is the EM response of AgNP pair simulated by defining the spatial extent of the metal NPs from the highresolution TEM image. FDTD simulations, using a grid of 14 million points to discretize the spatial extent of the electric and magnetic fields with up to 2.5-nm resolution in the gap region and carrying out 200,000 time steps, were performed in multi-teraflop supercomputing facility that consists of 5472 CPUs connected by a high-performance, low-latency Myrinet network [46].

Enhancement factor (EF) simulations
The EF is one of the most important aspect for characterizing the SERS effect and practical applications, as the first concern is generally to know the magnitude of the EF that can be achieved. However, it is difficult to make a single general definition of the EF because of the complex mechanism of EM and chemical enhancement, and the diversity of situations, such as single molecule, multiple molecules, unclear molecule numbers, spatial distribution, and orientations of the probe on the surface [59] This situation will be changed, since a standardization approach has been proposed during the International Conference on Raman Spectroscopy (ICORS) in 2012 in Bangalore, India.
Due to the lack of rigorous definitions, there are three major representative definitions that include the single-molecule EF and the SERS EFs each from the substrate and the analytical chemistry point of view [59] Among these, most studies of SERS EF have indeed focused on the average SERS enhancements across different substrates. The most widely used definition used for the average SERS EF is: where ISERS is the Raman signal intensity at a specific vibration mode of the probe molecules attached on the substrate; Ibulk is the normal Raman signal of the probe molecules; and Nads and Nbulk are the number of probe molecules adsorbed on the substrate and in the bulk exposed to the laser light, respectively. However, as it is difficult to control the number of molecules on each NP and the number of NPs within the focal volume, the number of molecules contributing to an SERS-enhanced signal is generally unknown and is usually ascertained through statistical analysis [46].
We can also use EM field distribution to roughly calculate the EF as mentioned earlier, as the EF is usually defined as ([E]/[E0]) [4] where E is the local maximum electric field and E0 is the input source electric field in a linear simulation [39]. For example, as shown in Fig. 3 [43]. Moreover, Theiss et al. reported that the maximum electric field intensity, lying in the gap between the NPs, had a value 82,400 times that of the incident field intensity. Therefore, the SERS enhancement EF, for this NP pair at the most intense point, is given by the square of this electric field intensity EF, giving a value of 6.9 × 10 9 [46].
There is a significant relationship between the EF, the excitation wavelength, the analytes, and the composition, size, and geometry of the substrates. The EFs are different based on the association between different parameters, as shown in Table 1. When AuNPs are functionalized at the ends of AuNRs, the enhancement could be increased by 2-3 orders of magnitude, which could significantly improve the signal for target thrombin detection [41]. Besides, the EF values of 10 6 and 10 8 were obtained by converting the electric field intensity of 10 3 and 10 4 in the case of large and small overlaps, respectively [47]. For a desirable SERS substrate, for example, in the study of metal-dielectric-graphene sandwich for SERS, the EF of 6.3 × 10 5 could be obtained by 3D-FDTD simulations, with perfectly matched BC applied in the z-direction while periodic BC set in the x-and y-directions [55]. The accuracy of substrate fabrication has been ascertained by verifying the EF by the FDTD method. A high EF of 10 11 is ascribed to the significant electric field enhancement at the cavities of nanostructures and nanogaps between them, which is supported by FDTD simulations [53].

FDTD method for SERS substrate design, fabrication, and its application
Molecules with very low Raman efficiency, samples at very low concentration, and small quantities or volume of samples are the main limitations of the classical Raman spectroscopy.
However, the Raman scattering intensity, arising from molecules located in the vicinity of the nanostructured metallic surface excited by the visible light, can be significantly enhanced because of the excitation of the LSPR [60]. Many probe molecules have been tested [61], and Table 1 shows that Rhodamine 6G (R6G) is the most commonly used probe in SERS. More importantly, SERS is highly dependent upon the substrate. Furthermore, FDTD method has gained much attention for substrate design and fabrication due to the advent of convenient commercial software, which also further promotes the development of SERS techniques.

FDTD method in SERS substrate design and fabrication
In general, there are two main SERS systems. One is the colloid systems that include dispersed optical resonant colloids in solutions, and the other is the substrates with constructing or patterning optical resonant structures [44]. In terms of fabrication method, Sharma et al. presented three representative examples of SERS substrates that include the substrates prepared by bottom-up methods, namely silica-coated clusters of gold NPs, and top-down fabricated immobilized NR assembly (INRA) substrates and tip-enhanced Raman spectroscopy (TERS) tips [20], while Fan et al. classified three general types of substrates, including MNPs immobilized in planar solid supports and metallic nanostructures fabricated using nanolithographic methods and template techniques [19]. Computational modeling tools, such as FDTD, may provide a convenient way for "simulation-based engineering and design optimization" in the SERS substrate fabrication, which enables rapid exploration of multiple design configurations without the need for extensive fabrication and physical testing. The "virtual prototyping" approach for SERS substrate design, rather than conventional trial-anderror-based experimental methods, will benefit the nanofabrication method development and experimental testing in terms of saving fabrication time and cost.
The desirable SERS substrates should possess high SERS sensitivity, good spot-to-spot uniformity (reproducibility and reversibility), and long-term stability of enhanced Raman signals [44]. High SERS sensitivity means large EF. Based on the understanding of the effects of inter-particle gap separation on the particle-film plasmon couplings, Lee et al. fabricated optimally designed SERS substrates with tailored surface plasmons that can detect 2,4dinitrotoluene on the scale of an attomole [49]. Good spot-to-spot uniformity usually originates from the different areas of a substrate, regardless of geometrical robustness, where there is almost no SERS signal deviation. Lee et al. studied closely spaced gold nanosphere chains on Pt mirrors exhibiting strong plasmon coupling between both horizontally and vertically oriented modes relative to the chain with the FDTD simulation. The broadband enhancement features together with effective light concentration make their structure suitable for the fabrication of reproducible and integrable SERS substrates [62]. Long-term stability requires the long shelf life of high SERS sensitivity. Zheng et al. reported that Raman measurements on an SERS substrate based on large-area well-defined gold NP arrays after 1 year showed almost the same magnitude of Raman signal as that of the original measurement [44].
FDTD method can be used to direct SERS substrate fabrication before experiment, which is a convenient and time-saving compared compared to the trial-and-error approach. In order to match the grating constant of the nanodisk array to excite surface plasmon polaritons (SPPs) in water and to tune the diameter of Au nanodisk for controlling the LSP resonance wavelength, Shioi et al. first calculated the electric field for a model representing double-resonance substrates, based on Au nanodisk arrays placed in water, by the FDTD method. From the calculation results, they fabricated the desirable double-resonance substrates using electron beam lithography [37]. FDTD calculation was used to explain the size-dependent SERS activity. The SERS-active substrate, being a simple model, was simulated with a fixed inter-particle distance of 2 nm, an excitation laser of 632.8 nm, a Yee cell of 1 × 1 × 1 nm 3 , and the time steps of 30000. After finding the optimal size, Fang et al. utilized AuNPs as the core to synthesize Au-core-Pdshell (Au@Pd)NPs to greatly enhance the SERS activity [52]. Furthermore, for the situation that, even with state-of-the-art equipment, high resolution, high crystallinity, and low sidewall roughness cannot be achieved simultaneously, Gao et al. reported that the laser shock imprinting, being a cost-effective direct nanoshaping method, could be used for high-throughput fabrication of smooth 3D crystalline nanostructures at ambient conditions, with FDTD simulations confirming the field enhancement and showing reflectance and absorbance spectra of the simulated structure. Further details are shown in Fig. 4 [63]. It is necessary to conclude a discussion of future directions for substrate fabrication. Deep-UV SERS would be highly useful for identifying biological molecules, including protein residues and DNA bases which have electronic resonances in this wavelength range, resulting in both electronic resonance and SERS enhancements. In particular, Al could replace Au and Ag as an alternative plasmonic material for deep-UV SERS, since the strongest enhancements for Au and Ag are found in the visible or near infrared (NIR) region [34]. The promising combination of SERS with spatially offset Raman spectroscopy, that is, SESORS, could contribute to both in vitro on tissue sections and in vivo in small animals in the realm of cancer diagnostics, making it possible to record Raman signals from tissues at depths ranging from millimeter to centimeter, which is a significant step forward in the ability to adapt this approach in a clinical setting for disease diagnosis [21], [64]. The surface-enhanced femtosecond stimulated Raman spectroscopy (SE-FSRS) that combines plasmonics and ultrafast science not only has enabled the study of molecule-plasmon coupling and dynamics, but also can facilitate time-resolved studies of molecular dynamics occurring on or near plasmonic substrates [20]. Therefore, studies on the applicability of desirable substrates for UV-SERS [34], SERORS [65], and SE-FSRS [66] are necessary.
The versatility and flexibility of the FDTD method indicate that it could be used in the study of LSPR, the optical property spectra, and the EF, of different kinds of substrate structures. also In addition, it can also be used investigate the influence of the SERS substrate designed with respect to different material parameters and different structure types, such as the material's refractive index, size, and the shape of the nanostructures, on the position of the LSPR in comparison with the position of the maximum SERS intensity. Thus, the FDTD method could play an important role in the design and the fabrication of substrates.

The applications of new substrates in analytical science
Currently, SERS has become one of the most sensitive analytical techniques [67], [68], [69], and its applications extend from the investigation of the surface adsorption at the electrode surfaces to analytical studies in the biomedical and environmental areas [19]. Two typical and exciting examples are that three meningitis pathogens were successfully quantified by SERS in a multiplexed test with calculated limits of detection in the picomolar range [70], and Cr(VI) in aqueous solutions was successfully detected using the alizarin red S (ARS)-TiO2 complexes [71]. Furthermore, we conclude some recent typical examples for SERS applications of biosensing and environmental analysis, as shown in Table 2. For example, Wang et al. [78] proposed a kind of quantitative and accurate detection of multiple biomarkers that would allow for the rapid diagnosis and treatment of diseases induced by pathogens with a limit of detection of 1 ng/L. As shown in Table 1, Table 2 of this review, based on the new substrates associated with FDTD method, SERS could be applied for single-molecule detection, biomedical diagnostic and target detection, environmental analysis, and food safety screening. It has become a sensitive and specific tool due to label-free detection of molecules at very low concentrations and identification of molecules based on their vibrational fingerprints [101].
Arrays of plasmonic NPs with separations on the order of 1 nm could be used in devices used for chemical detection at the single-molecule level, with FDTD simulations predicting an EM SERS EF of 10 9 -10 10 [46]. A molecular sentinel (MS) probe-modified metal film over nanosphere (MFON) substrate would be applied for point-of-care high pathogenicity influenza (HPAI) virus detection, with FDTD method showing the hot spots at the tips of the Au-Gr-coated polystyrene beads (PS) [43]. The smart liquid SERS substrates based on Fe3O4/AuNPs improved the quantitative detection of analytes and might be a significant step in employing SERS for environmental analysis, with FDTD model verifying the combined effect of the quasi-photonic crystal and the Au NPs on the EM field penetration [42]. A Au/HfO2/grapheme sandwich with a high-k dielectric layer could promote the design of SERS devices for food safety inspection, with FDTD calculations determining optimized thickness of Au and HfO2 [55].
Based on the application of FDTD method, we have highlighted some of the potential applications of the new substrates. Furthermore, with the advent of UV-SERS, SERORS, and SE-FSRS techniques in the near future, as shown in the concept graphs in Fig. 5, the applications could be extended to material imaging, deep Raman signal sensing, and probing of chemical reaction dynamics with new SERS-augmented scientific techniques. It is hoped that this could also improve the development of SERS quantitation-driven sensing for quantitative analysis, which is still the biggest challenge in SERS field. [ [34], [65], [66]].

Conclusions and outlook
We hope this review can help readers who are interested in both theoretical and experimental studies in the areas of SERS or modeling computational electrodynamics, to better understand the FDTD method in SERS research and sensing applications. FDTD method, being a numerical analysis technique to evaluate Maxwell's differential equations, has become an important tool to calculate the parameters related to SERS intensity. Owing to the versatility and flexibility of the FDTD, it could be used to study the EM field distribution, the optical property spectra, and the EFs of different kinds of substrate structures. It also could be used to indicate the influence of different parameters and different structure types on SERS intensity, which will further direct SERS substrate design, fabrication, and its applications.
It should also be noted that the FDTD method is inherently approximate; hence, the results obtained from the FDTD method would be approximate regardless of the infinite numeric precision provided by the powerful computers. With this in mind, when we use the FDTD for solving EM problems, for example, the optical response or EM field distribution in SERS substrates, the results can be obtained using a computer if there is no catastrophic failure during the implementation of a solution; however it might not be capable of generating sufficiently accurate results. Fortunately, the development of the improved FDTD method with a finer discretization, such as the abovementioned 3-D FDTD, FVTD, DGTD, and G-FDTD, will give more ways to cross-test the implementation and the solution in EM field, and thus the better results in terms of operation stability, reproducibility, and accuracy. It can be expected that, in the future, FDTD method will be used more often to understand the EM mechanisms, to design and fabricate SERS substrates for high enhancement factor, excellent reproducibility, and longterm stability for highly sensitive, quantitative molecular analysis in real-world applications.