For most single-pass optical devices, such as waveguide and optical interferometers, light interacts with the analyte molecules or proteins only once. Thus to enhance the interaction between the light and the sensing target usually one must increase the physical length of the sensor. However, in a resonator-based sensor where light is confined in a micron scale sensor element and where the light waves interfere constructively, we can increase the effective interaction length by increasing the optical quality factor (Q) of the resonant system. The quality factor (Q), which is used to quantify the temporal confinement of the electromagnetic energy is defined as

Figure 1

Prominent microsystem biosensing technologies developed for label-free detection down to single molecules. From left to right: optical resonator, plasmon resonance biosensor, nanomechanical resonator and nanowire sensor. Reproduced with permission from [1–5].

where U(t) is the total energy of the confined light field and -(dU(t)/dt)/ω₀ is proportional to the energy that is lost for each electromagnetic oscillation. From this it follows that the energy of the charged cavity will decay over time with U(t)=U₀ exp(-ω₀t/Q) once the light source has been shut off. The ringdown time τ measured with a photodetector placed in close proximity to the microcavity, is τ=Q/ω₀. The exponentially decay of the energy U(t) is characteristic of a resonant system. The energy U(t) is proportional to the electromagnetic field strength squared U(t)∞E(t)², and it follows that the complex field evolves in time as: E(t)=exp(-ω₀t/2Q)exp(-iω|t|). Time and frequency domains are linked by Fourier transforms:

and we therefore expect a spectral response that exhibits a Lorentzian line:

where ω₀ is the resonance wavelength and Δω=ω₀/Q is the linewidth (full width at half maximum, FWHM). The (total) intrinsic Q-factor of an optical mode is calculated from 1/Q_total ≈ 1/Q_absorption + 1/Q_scattering + 1/Q_radiation, where 1/Q_absorption denotes losses of photons by material absorption, 1/Q_scattering represents scattering losses due to surface inhomogeneities or surface defects, and 1/Q_radiation is set by the radiative losses [33]. For optical resonators with negligible amount of losses from radiation such as for example ∼100 μm silica microsphere, the ultimate material limited Q factor [33, 34] is close to

The reported optical attenuation coefficient of α_glass∼7 db/km (where ∼5 dB is bulk Rayleigh scattering and ∼2 dB is absorption). Low dissipation systems with high Q factors are particularly suitable for monitoring perturbations of the resonance signal [35] and thus for detecting biomolecules [10]. For example, in an optical microsphere-resonator (Figure 2) with a Q-factor of 10⁸ a photon (with a wavelength of 600 nm) has a lifetime of ∼30 ns and can travel in the cavity for ∼10 meters before it is lost. If the round trip of the photon in the cavity is 100 μm, the photon will interact with a target molecule more than 10⁵ times. This creates a “signal build-up process” inside the resonator that gives rise to significant signal amplification and allows ultra-sensitive detection of biomolecules.

High-Q optical resonators have been investigated intensively over the past decade as a promising technology for ultra-sensitive and label-free biosensing [10, 36]. One of the most promising designs is the whispering-gallery-mode (WGM) resonator [37, 38], in which light is trapped in circular orbits by total internal reflections from the boundaries of the resonator (Figure 2). WGM resonators exhibit the highest Q factor [39], up to three orders of magnitude higher than other types of resonators. Other types of high-Q WGM-type microcavity biosensors are ring resonators [40], microtoroids [16], and glass capillaries – so called liquidcore optical-ring resonators (LCORR) [41]. The ultra-high-Q resonators possess many interesting properties, not available in other photonic devices, such as ultra-low loss, ultra-long photon lifetime and ultra-high intracavity power and intensity. Due to these unique attributes, the minimum detectable concentration of a target molecule, i.e., the detection limit, is significantly lower in the ultra-high-Q resonators than in other types of sensors. In fact, it has been demonstrated that a single virus or nanoparticle can be detected by a WGM resonator [3, 30].

As evidenced in Figure 3, few components are needed for an experimental realization of a WGM biosensor. In one implementation, a continuous-wave tunable distributed feedback laser (DFB), for example operating in the telecom band at 1.3 μm wavelength, excites high-Q optical resonances in a microsphere via a tapered optical fiber. The resonances are excited by evanescent coupling from a tapered section of a single mode optical fiber [43, 44], a method that transfers a high power fraction from the fiber to the microcavity. In fact it has been shown that even critical coupling (no transmission at the output of the waveguide on resonance) can be achieved if the coupling distance and the taper diameter is carefully controlled [45]. Moreover, tapers can be fabricated in automated setups [46]. Once coupled, the microsphere spectrum is recorded by sweeping the wavelength of the laser, typically over a narrow spectral range of ∼0.3 nm. It is worth noting that the Q factor is now further modified by the coupling Q factor [47].

Figure 2

Left: total internal reflection of a whispering-gallery-mode (WGM) in a glass microsphere. Right: corresponding wave optics.

More recent biosensing demonstrations utilize other evanescent coupling schemes such as those based on prism coupling [48, 49] and coupling from integrated bus waveguides [24, 50–53]. Both approaches provide for better mechanical stability [54], improved signal-to-noise-ratio [52] and potential for multiplexed excitation, as well as referenced readout of several microcavities in parallel [49]. For a robust biosensor application the integration of such sensor components in a mechanically stable platform is essential. Ultimately, the laser source could be fully integrated on-chip [55], but for now different external lasers, such as DFB as well as external cavity lasers, must be used [13, 30, 56].

Although the principle applies to any optical microcavity geometry, we will review the detection mechanism for a WGM microsphere sensor, where analytical results are most instructive. Like a nanomechanical resonator biosensor [4], the microsphere optical resonator detects the binding of analyte molecules as changes in the resonance frequency. A WGM exhibits a quite large sensitivity to such perturbations because the light field is confined close to the surface where the evanescent field interacts strongly with the surrounding medium.

The binding of a biomolecule will shift the WGM resonance frequency by a miniscule amount (Figure 4). The shift to longer resonance wavelength occurs because the bound biomolecule will effectively “pull” part of the optical field to the outside of the microsphere, thereby increasing the roundtrip path length by 2πΔl. This increase in path length produces the shift (Δω) to lower frequency.

The binding of a streptavidin molecule to a microsphere provides an example [13]. Note that a streptavidin molecule (∼4 nm) is small with respect to the radial extension of the evanescent field associated with a WGM. Once bound at the surface where the evanescent field strength E(r) is high, the molecule will become polarized at the optical frequency ω. The overall induced dipole moment P is calculated as P = α_exE, where α_ex is the excess polarizability of streptavidin, in excess of that of the water it displaces [12, 13]. The energy that is needed to polarize the molecule and induce this dipole moment is ½ α_ex |E(r₀)|², where E(r₀) is the electric field strength at the streptavidin binding site r₀, and

Figure 3

Probing high Q resonances of different optical resonators in biosensor applications. (A–D) Reproduced with permission from [13, 24, 30, 42]. (reprinted with permission from [23]. Copyright (2002), American Institute of Physics. Reprinted from [42]. Copyright (2010), with permission from Elsevier.).

α _{ex,streptavidin} ∼4πε₀×3.3×10^-21 cm³. By first order perturbation theory, we can now estimate the frequency shift by comparing the energy that is needed to polarize the biomolecule to the total electromagnetic energy stored in the unperturbed resonator [12]:

where ε is the permittivity of the medium. Equation (2), the reactive biosensing principle, allows one to quantify the frequency shift of any optical resonator in response to molecule or nanoparticle binding events.

A large Q factor is necessary in order to resolve the fractional frequency shift Δω/ω predicted by Eq. (2). In practice one monitors the resonance wavelength shift Δλ, which is Δλ/λ= -Δω/ω. The limit of detection, i.e., the smallest detectable wavelength shift Δλ_min is not simply the linewidth Δλ_FWHM. A shift of only a fraction F of Δλ_FWHM can be sensed [38], so that Δλ_min/λ=F* Δλ_FWHM/λ=F/Q, where the figure of merit F is typically [57] 1/50–1/100 and determined by noise sources such as thermal-refractive noise [58].

Importantly, the magnitude of the wavelength shift Δλ itself is inversely proportional to the mode volume V_mode given by the denominator in Eq. (2), which for a WGM in a microsphere is [59]

Reducing the size (modal volume) of the optical resonator thus boosts sensitivity. For the microsphere, the optimal resonator diameter can be calculated, because radiation loss limits the Q factor in very small microspheres as the diameter approaches the wavelength of the confined light [59].

An alternative to the reactive biosensing principle is the modesplitting effect, i.e., one resonant mode splitting into two resonances due to interaction of light with nanoscale objects, such as nanoparticles, in the mode volume. Because light can propagate clockwise (CW) or counterclockwise (CCW) in a WGM resonator with a circular boundary, the optical mode in such a resonator possesses a two-fold degeneracy. The degenerate modes share the same resonant frequency ω_c and field distributions but propagate in opposite directions; a

Figure 4

(A) Illustration of a Whispering Gallery Mode (WGM) optical resonance in a glass microsphere. The binding of a protein to the microsphere surface increases the WGM path length by Δl, which is detected as a resonance frequency shift Δω. (B) The reactive sensing principle. A molecule binding to the microsphere surface is polarized within the evanescent field (yellow ring) of a WGM. The energy that is needed to polarize the molecule causes the resonance frequency to shift.

single resonance appears in the transmission spectrum of the resonator.

When a WGM field encounters a light scattering center, such as a molecule or particle, on its optical path, the light is scattered elastically from the particle/molecule. A portion of light is scattered and lost to the environment, while some is scattered back into the mode volume and takes the optical path of the counterpropagating frequency-degenerate mode, coupling between the CW and CCW modes. This lifts the degeneracy of the two modes. Two standing wave modes are formed in the cavity, and this is reflected in the splitting of the single resonance into a doublet structure, i.e., two resonances, in the transmission spectra of the resonator (see Figure 5). Because the two split modes reside in the same resonator, they share many noises sources, such as temperature variation, and since they are compared to one another, this is a self-referencing sensing technique [30].

The mode-splitting effect in WGM resonators was first reported as an interesting phenomenon in high-Q microspheres [60, 61]. It has since been demonstrated in various WGM resonators with different geometries, including microspheres, microtoroids and microdisks [30, 62–64]. Many factors, such as surface defects, material inhomogenities, and nanoparticles, all of which can induce light scattering in the mode volume of a WGM, may cause mode splitting. Assuming a subwavelength scattering center (Rayleigh scatterer), and that the electric field of the coupled WGMs induces a dipole in the scatter, the interaction between the WGM and the scatterer can be modeled using a dipole approximation [30, 61, 62]. Polarizing the scatterer induces a coupling between the two counter-propagating WGMs. The coupling coefficient g depends strongly on the polarizability α of the scatterer and the location of the scatterer in the mode volume V_c with respect to the mode distribution. We can derive g as:

where ω_c is the resonance frequency, f(r) accounts for the cavity mode distribution of the WGM at location r, α represents the polarizability of the scatterer which for a spherical scatterer of radius R can be expressed as α=4πR³(ε_p-ε_s)/(ε_p+2ε_s) in which ε_p and ε_s denote the electric permittivities of the particle (scatterer) and the surrounding medium, respectively.

Through the interaction of the scatterer with the WGM field, a portion of the field is lost to the environment, adding to the damping. Using a Weisskopf-Wigner approximation, the damping rates due to coupling to the reservoir via Rayleigh scattering is:

Figure 5

Mode splitting in a WGM resonator. Left: WGM resonators supports degenerate counter-propagating modes: clockwise (CW) and counterclockwise (CCW). Light scattering from a scattering center introduces additional damping to the optical modes and couples the initially degenerate CW and CCW modes, lifting the mode degeneracy. Right: Experimentally obtained mode-splitting transmission spectrum (blue) after the deposition of a single nanoparticle in the mode volume of a microtoroid WGM resonator and the fitted curve (red) adapted from [30].

where $v=c/\sqrt{{\epsilon }_m}$ is the velocity of light in the surrounding environment and c is the speed of light in vacuum.

The polarizability of a nanoparticle can be calculated from the mode-splitting transmission spectrum of WGM resonator after a nanoparticle is adsorbed on its surface. The coupling strength g is quantified by the doublet splitting g=πδ, where δ is the frequency detuning of the two split modes. The additional linewidth broadening is ${\Gamma }_R=\pi \left|{\gamma }_1-{\gamma }_2\right|$, where γ₁ and γ₂ represent the linewidths of the split modes. The ratio of (γ₁-γ₂)/δ carries information about the polarizability of the particle, and hence on its size and permittivity (i.e., its refractive index). Comparing Eqs. (4) and (5) it is clear that the polarizability of the nanoparticle is then: α=4πR³ (ε_π-2ε_σ)/(ε_π+2ε_σ)=-(3λ³/8π²)(Γ_R/g). It is important to note here that the measurement of Γ_R/g=(γ₁-γ₂)/δ is independent of the position of the particle on the resonator. The position-independence of the value of Γ_R/g gives the mode-splitting technique advantages over schemes using single-resonance spectral shift, which are affected by the particle position.

Once mode-splitting is introduced by the first scatterer, two orthogonal standing-wave modes are formed by the superposition of the split modes in the resonator. The next scatterer deposited/adsorbed scatterer in the mode volume interacts with the two orthogonal modes differently, and subsequently the two standing waves re-distribute themselves to maximize the amount of mode splitting. Given this physical mechanism, we can derive the polarizability of the nth light scatterer by analyzing the relative changes in the linewidths and frequencies of the two split modes in two consecutive mode-splitting spectra. We do not need to record the whole history of the changes in the mode-splitting spectra to measure the size of the nanoparticle in the mode volume [65]. In fact Influenza A virions deposited on a microtoroid resonator have been measured one by one and the derived size distribution of the Influenza A virions agreed well with the size range of such virions as reported in the literature [66, 67].

The detection limit of the mode-splitting technique is set by the condition that the two split modes need to be resolved in the transmission spectra, which requires that the amount of splitting quantified as $\left|2g\right|$ should be greater than the sum of the frequency linewidth ω_c/Q of the WGM and the additional scattering-related damping rate Γ_R[68], i.e.,

Otherwise, the two split modes will appear as a single resonance in the transmission spectrum. For small particles, Γ_R is usually much smaller than ω_c/Q, and therefore the detection limit of the mode splitting technique is mainly set by the Q-factor, which determines the smallest resolvable mode splitting. Once the mode splitting is observed and measured, polarizability and hence the size or refractive index of the scatterer can estimated.

Usually the mode-splitting spectrum is measured by scanning the wavelength of a tunable laser through the frequencies and, once the mode splits, recording the transmission spectrum. An interferometric setup introduced in Ref. [69] allows mode splitting to be detected even if $\left|2g\right|$ is smaller than the linewidth of the modes. It also provides an alternative way to detect mode-splitting signals without requiring a tunable laser. Recently an optical gain medium, which helped narrow down the linewidths of the optical modes, was also used to recover mode-splitting from what initially appeared as a broad, single resonance [70].

The first experimental demonstration of size measurement of single nanoparticles utilized a high-Q microtoroid WGM resonator [30]. In this work, single potassium chloride (KCl) and polystyrene nanoparticle as small as 30 nm in radius were detected and measured by analyzing the mode-splitting spectra. Consecutive size measurement of nanoparticles deposited one by one into the mode volume of a high-Q resonator was also achieved by anaylzing the frequencies and linewidths of the resonances in the mode-splitting spectra right before and after the deposition of each particle [65].

When a resonator is in liquid solution, the smaller difference between the refractive indices of the resonator and surrounding medium degradates Q factor because it leads to increased radiation loss and decreased particle polarizability. Absorption loss in the surrounding medium is also greater, which further decreases the Q factor of resonant modes. As Eq. (6) suggests, sensing applications in an aquatic environement are more difficult than those in air. To address this issue, a resonator with a Q factor high enough to tolerate the Q degradation must be used [71, 72]. One strategy is to use a thermall-stabilized reference interferometer to reduce the frequency noise in the measurement of mode-splitting. By means of such a scheme, a high-Q microtoroid has been used to detect polystyrene nanoparticles with radii down to 12.5 nm and Influenza A virions in aquatic environment [67].

The detection capabilities of a spherical optical resonator biosensor can be predicted by analytic equations [12, 73–79]. For a single nanoparticle binding on the microsphere excited in the fundamental WGM mode Y_ll one predicts a resonance frequency shift of [3, 12, 38]

where r_p is the particle binding site, L is the evanescent field length, R is the microsphere radius, n_s and n_m are the refractive indices of the sphere and the medium, respectively. For a biomolecule that is small compared to the wavelength the exponential factor is negligible. For N molecules binding at random locations to the microsphere surface with surface density dN/dΩ=σ4πR²/4π (Ω is the solid angle) Eq. (7) becomes [12]

where σ is the surface density of bound biomolecules (typically reported in units of cm^-2). The excess polarizability of protein or DNA biopolymers is determined from measurements of the refractive index increment dn/dc of the pure biopolymer solution. The excess polarizability is then related to the refractive-index increment by

where ε₀ is the vacuum permittivity, n_m is the solvent (water) refractive index, m is the molecular weight of the biomolecule (streptavidin: ∼1×10^-19 g) and the dn/dc of protein [11] is ∼0.183×10^-9 mm³ pg^-1, dn/dc of DNA ∼0.166×10^-9 mm³pg^-1. With this Eq. (9) in hand, we can calculate the mass loading (mass.l.) associated with the resonance frequency shift and plot the sensor response (sensogram) in units of picograms per millimieter squared sensor area [10]:

A similar analysis has been done for LCORRs where calculations for resonance wavelength shifts are calculated as [80]

where n₂ refractive index of medium, n₃ is the refractive index of the LCORR, and bulk index sensitivity S=δλ/δn₃. For details of this derivation we refer to [80]. For other resonator geometries such as toroids, this analysis has do be done by simulation [81], because no analytic equations that describe the confined light fields are available. In practice, the sensor response is often experimentally determined (calibrated) for different solution concentration levels [51, 82–84]. Examples for this are shown in Figure 6. In surface-based biosensing the most physical and accurate figure of merit is the limit of detection (LOD) in units of pg/mm² mass loading, see Eq. (10). Very often, researchers also report the sensitivity to bulk index change in units of nm wavelength shift per refractive index unit (RIU): nm/RIU. Although the field has accepted measurements of bulk index change as a form of biosensing we would like to point out that a high sensitivity to bulk index change is not always directly correlated with a high sensitivity to mass loading at the microcavity surface. We also emphasize that label-free and specific detection by molecular recognition has thus far only been achieved in sensitive surface-based detection schemes.

The mass loading reached at a given time depends on the concentration of the analyte (Figure 6). By fitting of the time response of the sensor to a kinetic binding model, such as the Langmuir model, one can determine on (k_on) and off rates (k_off) as well as the dissociation constant K_d=k_off/k_on which characterizes the affinity of the analyte towards its binding site on the resonator surface [11, 23, 84, 87–89]. More complicated models can account for conformational change or denaturation that may occur after analyte binding at the resonator surface [87]. Alternatively, the dissociation constant K_d can be determined from an equilibrium plot of mass loading or percentage surface coverage versus solution concentration level. Such a plot is called Langmuir isotherm and the dissociation constant K_d equals the concentration level at which surface coverage is 50% [11]. In practice one often plots the mass loading divided by analyte concentration versus mass loading. The slope of this linear plot is just -1/K_d[90]. Although binding kinetics are typically measured by stepwise titration of analyte concentrations, researchers recently reported a method for determining kinetic rates and equilibrium binding affinities in a single experiment during which analyte concentration increased in a nonlinear gradient [91].

Once the sensor response for different solution concentrations has been experimentally determined or analytically calculated, the sensor can then be used to determine unknown concentrations of an analyte. To ensure only the molecules of interest are detected, the resonator surface is modified (“functionalized”) with elements that recognize the analyte, such as antibodies [82, 84, 92], DNA oligonucleotides [57, 89, 93–96], aptamers [97] or other receptors [13, 98–101]. A few examples are shown in Figure 6 and Table 2 gives an overview of the recognition elements and the sensitivity levels that have been obtained with each element. Opto-fluidic ring resonators (OFRR, LCORR) in particular have been used extensively in various biosensing tasks, such as the detection of organophosphorus pesticide [109], label-free DNA methylation analysis [100], detection of breast cancer biomarker CA15-3 in serum samples [110], and detection of the HER2 breast cancer biomarker [42]. Also silicon, silicon nitrate and polymer ring resonators have been used to detect: antigen (PSA), alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), tumor necrosis factor-alpha (TNF-alpha), and interleukin-8 [51]; single nucleotide polymorphisms [90], DNA and mRNA [88, 96] as well as tmRNA [105], C-reactive protein (CRP) [84], cytokines interleukin-2 (IL-2), interleukin-4 (IL-4), interleukin-5 (IL-5), and tumor necrosis factor alpha (TNF alpha) [83]; IL-2 [82]; avidin [53, 111]; cadaverine [52] and antibodies [92]. In combination with nanoparticles, different kinetic modalities for sensing have been demonstrated in toroidal cavities [112] and enhancement of sensitivity in a nanoparticle-sandwich assay was demonstrated on silicon ring resonators [84].

Crucial to achieving specificity is the chemical modification of resonator surfaces, which strongly depends on their material properties. For example, for the modification of glass (and the thermal oxide layer on silicon) one can use silanization of the glass-silanol groups [13, 16] or physisorption of proteins and hydrogels [57, 107]. In multiplexed arrays, each sensor element can be individually functionlized, by, for example fluids controlled with microfluidics [23, 24], or ink-jet printing [113] or patterning with silicon etchants [114]. Also very recently, one and two-dimensional photonic crystal resonators have been explored to detect biomolecules, including interleukins 4, 6 and 8 [115] and bovine serum albumin [116] as well as immunoglobulins [106]. Table 2 gives an overview on

Figure 6

Examples of biorecognition schemes utilized in microresonator biosensing applications: (A) antibody-based detection of a cancer biomarker carcinoembryonic antigen (CEA) with silicon rings in serum (adapted with permission from [85], copyright (2009) American Chemical Society), (B) detection of DNA oligonucleotides by hybridization to dextran-functionalized microspheres (reprinted from [57], copyright (2003), with permission from Elsevier), (C) breast cancer biomarker HER2 detection in LCORRs functionalized with antibody (reprinted from [42], copyright (2010), with permission from Elsevier), (D) silane-based bioconjugation techniques that have been explored for achieving high surface densities in DNA detection [86].

molecules that have been detected with optical microcavities, as well as the reported detection limits. Detection limits in the 10 pM range rival those of conventional enzyme-linkedimmunosorbent assays (ELISAs).

Perhaps the simplest example for an optical resonator is a ∼100 μm-diameter silica microsphere, where light remains confined due to total internal reflection [150], see Figure 7. Although glass has low dissipation, light confined inside the microsphere is eventually lost due to absorption and scattering. The high-Q microspheres are simple to fabricate. They can be made by melting spheres at the end of high quality optical fiber with a hydrogen torch [33] or a CO₂ laser beam [34]. Microspheres fabricated from a standard single mode optical fiber and immersed in a liquid have been used in many biosensor experiments [13, 48, 57, 87, 97, 101, 117, 152], including the first biosensor demonstration in 2002 [13]. Placing the sphere in a liquid limits Q-factor to ∼10^7–8[71, 153], however, because the liquid absorbs light.

Among the resonances of the high-Q micro-spheres, the modes localized along the equator are of particular interest [34]. These WGMs combine a good spatial confinement of the field with the very high-Q factors (Q), two features which are of great importance for sensitive biodetection. Figure 7 illustrates an equatorial mode in an erbium-doped microsphere excited by light transferred from a tapered fiber.

The transverse electric and transverse magnetic fields confined in a microsphere are described by

where Y_nm(θ,φ) are the vector spherical harmonics, and j_n(nkr) represent the spherical Bessel functions, with k=2π/λ. These resonances can also be viewed as quasi-bound states of a Schroedinger-like equation that describes a complex angularmomentum particle in an effective potential well that occurs due to the refractive index contrast between the higher index of refraction of the sphere and lower index of refraction of the surrounding medium. The Bohr atom thus seems a reasonable analogy, and microsphere resonances are often called “photonic atom modes” [154, 155]. For the Bohr as well as the photonic atom, wave functions are indexed by three quantum numbers n,l,m, where n represents the number of nodes in radial directions, l is the angular-mode number proportional to the number of wavelengths in one orbit, and m is the azimuthal mode number. Modes with the same l-number but different m values are degenerate in a perfect sphere, however, for spheres made by melting of a fiber tip this degeneracy is typically removed and instead they exhibit a spectrum rich with modes of different m numbers. By observing different TE and TM modes in a microsphere it is possible to detect the orientation of molecules bound at the microsphere surface [104, 156]. Furthermore, wavelength multiplexing of several WGMs simultaneously can provide information on the thickness of an adsorbed layer [157].

Microspheres are particularly suitable for table-top experiments since they are easily fabricated [33, 34, 120] and their mode structure is amenable to rigorous theoretical analysis [73, 76–78, 158, 159], as well as simulations based on Mie theory [147, 160]. Recent demonstrations of high-Q microspheres include studies of protein adsorption [87], thrombin detection [97] and refractive index sensing [48, 117]. Very high sensitivity allows the detection of single particles such as polystyrene nanospheres and virus particles [3, 32]. Recent developments include efforts to further improve the sensitivity [58, 81, 147, 152, 161], multiplexing capability [49] and functionalization strategies [16, 54].

Figure 7

Equatorial WGM. The green ring, which traces the trajectory of the fundamental equatorial WGM, is due to the luminescence of the erbium ions with which the microsphere is doped image taken from [151].

Ring and disk resonator geometries were among the first to be proposed for biosensing applications, and they have since become the most prominent resonators in biosensing. They are fabricated on-chip with photolithographic techniques in different materials systems and in large sensor arrays [24, 40, 50, 52, 53, 89, 130, 131, 162–168]. For example, ring, racetrack and disk resonators can be fabricated in silica, silicon [24, 50, 89, 111, 166, 168] and organic polymers [52, 165, 167, 169, 170]. Early biosensing demonstrations proposed a high-finesse (low loss) WGM disk resonator [130] and utilized a vertically coupled glass microring resonator with Q of about 12000 [111] as well an integrated silicon nitride (Si_xN_y/SiO₂) ring resonator [53]. Within arrays, individual sensors can be used as internal reference, thereby eliminating noise due to thermal fluctuations and nonspecific molecular binding events [24]. Often compatible with CMOS techniques, ring and disk resonators are ideal candidates for integration into portable electronic devices and commercial bench-top systems [92] and indeed silicon ring resonator biosensors are already on the market. Microfabricated ring/disk resonator arrays are also quite easily incorporated with microfluidics because they are fabricated on planar substrates.

One interesting approach to integration is an electrical tracing-assisted dual-microring resonator-based optical sensor system fabricated on silicon-on-insulator (SOI) wafer. The resonance wavelength shift of a sensing microring is measured by the electric power needed for a second, tracing microring to match the resonance shift. The electrically tunable tracing microring thus eliminates the need for wavelength scanning of a laser source [171]. Nanoporous polymer ring resonators reportedly increase device sensitivity by 40% because bioanalyte enter pores and interact with electromagnetic energy in the core of the ring, not just its surface [169]. Cascaded microrings that exploit the Vernier effect have also recently been proposed to increase sensitivity [172, 173]. In addition, it was found that for certain sensing schemes extending the ring cavity length of SOI microring sensors increases the resonator quality factor, and can thereby enhance the sensor resolution and minimum level of detection [103]. Vertically stacked microracetrack made from the photoresist SU8 have low noise and are exceptionally stable, although prone to temperature fluctuations [52].

In the meantime, the applications for these sensors have also been expanding. A commercial silicon ring resonator biosensor has been used for the first time to detect a biomarker (the cancer biomarker carcinoembryonic antigen) in the complex media of undiluted serum [85]. The same group used silicon ring resonators to detect single-nucleotide polymorphisms via kinetic DNA dissociation measurements [89]. They also used silicon ring resonators to simultaneously measure several protein biomarkers secreted from cells in cell culture media [82]. SOI ring resonator arrays with one output port that rely on wavelength division multiplexing have shown excellent sensitivity (0.3 pg/mm²) monitoring multiple molecular-binding events [168].

Silicon nitride waveguides have also found applications in biomolecular detection [53]. Because silicon nitride is resistant to the diffusion of moisture and sodium ions, it can maintain a stable refractive index when operating in biological liquids. A packaged array of optical-slot-waveguide siliconnitride ring-resonator sensors demonstrated high sensitivity of 0.9 pg/mm [174]. Another approach using slot-waveguidebased ring resonators on SOI demonstrated a refractive-index sensitivity of 2×10^-5 RIU within a small footprint [122].

Microring resonators made of high index glass have been developed as a platform for detecting whole bacterial cells, proteins and nucleic acids [123]. Finally, ring resonators made from silicon nanowire optical waveguide (SNOW) have also shown promise for biosensing [175].

WGM resonances are also sensitive to biomolecules binding to the interior of thin-walled microcapillaries, and this is the basis for liquid-core optical ring resonator (LCORR) [56, 94, 176, 177]. The capillaries are fabricated by pulling the capillary while softening the glass with a CO₂ laser [56, 178]. The hollow-core cylindrical devices are easily integrated with microfluidics; indeed one of the main advantages of this approach is the ease of fluid handling as shown in Figure 3C. Theory and experiment put the protein detection capability of the LCORR in the sub-pg/mm² range [80]. LCORRs have been used in various sensing applications including the detection of DNA, viruses, proteins and biomarkers of cancer [80, 94, 179, 180]. Theoretical analysis shows that the thickness of capillary wall can be optimized to minimize or eliminate thermal drift, which could help stabilize the resonant modes in the presence of temperature fluctuations [181]. Recently a fluorescence based refractometric sensing was demonstrated in a glass capillary coated on the interior with fluorescent silicon nanocrystals [182].

LCORRs can be arranged in parallel on integrated waveguides to create a sensor arrays already integrated with microfluidics [183]. By positioning multiple waveguides along a single capillary, it is possible to interrogate the region of the capillary near each waveguide independently. This linear configuration is well suited to such applications such as capillary electrophoresis [184].

Although LCORRs have found diverse applications in biosensing, they have not yet demonstrated single particle/molecule resolution. Recent analysis has showed, however, that optofluidic ring resonators (OFRRs), which resemble LCORRs but which confine the WGM in three rather than two dimensions (by creating a bubble in capillary) are capable of detecting nanoparticles <20 nm in radius [185].

Silica toroidal resonators are fabricated from layers of silica (glass) grown on crystalline silicon substrates by thermal oxidation. The silica is etched into discs with hydrofluoric acid, and the discs are then undercut with a xenon difluoride etch to form silica discs on silicon posts [186]. The disc are transformed into toroids (donuts) by heating the silica with a CO₂ laser, a self-terminating process. The modal volumes of toroidal resonators are smaller than those of microspheres of comparable diameter, and the resonators also have ultrahigh Q factors [186] because of the almost atomically smooth surface produced by reflowing glass. Thus, the toroid geometry boasts a sensitivity comparable to that of the microsphere and is a good candidate for achieving single molecule detection [30, 66, 67, 70, 81]. Toroids could potentially be produced in arrays like those described for disks and rings. However, integrating the resonators with waveguides or optical fibers requires secondary fabrication steps and remains a challenge. This challenge may be addressed by avoiding the secondary fabrication step yet obtaining high-Q microcavities of up to 875 million for on-chip applications, for diameters larger than 500 μm [187]. Toroids are increasingly used in biosensor demonstrations [16, 65, 102, 112].

Researchers have recently used microcavity ring down spectroscopy to track the Q factor and wavelength shift during a biodetection event in order to make the resonator less susceptible to noise [98]. Computational schemes can also be employed to calculate the toroid response to detection events in order to achieve higher sensitivitys [81].

An interesting variant of the toroidal microcavity is the goblet resonator, made of a thermoplastic polymer that assumes a cupped shape on reflowing [188]. Finally, another modification of the fabrication scheme produces high-Q toroids with beveled edges, called wedge resonators [187].

Mie resonances in microspheres or capillaries doped with fluorescent dyes can be identified by epifluorescence microscopy and spectroscopy. These are lower-Q sensors, but they still provide a reasonably small modal volume for biosensing [101, 189, 190]. Fluorescent beads a few μm in diameter are capable of detecting proteins [126, 127, 191, 192]. They can also be used to detect the forces during endocytosis because microbead deformation induces azimuthal mode splitting [128]. In addition, a new detection scheme using clusters of fluorescent microresonators was introduced recently [191]. Each clusters of microresonators has a specific WGM spectrum that serves as its “fingerprint” and would allow clusters on a chip to be read out properly without knowing their exact locations.

Another recent innovation is the use of flexible split-wall fluorescent microtube resonator sensors to detect mouse fibroblast cells [133]. The same authors then discussed the integration of this device with others to constitute a lab-on-a-tube [193]. Quantum dots embedded in polystyrene microspheres can also serve as fluorescent sensors; the wavelengths of their resonant modes shifts as the bulk index of refraction is changed in ethanol-water mixtures [194]. Another approach combines fluorescence imaging with WGM resonators for multiplexed detection of biomolecules [49].

Optical fiber nanowires can be coiled-up to form microcoil resonators where a microfluidic channel runs through the inside of the coil for biosensing applications [144–146]. Recently a novel refractometric sensor based on a coated all-coupling optical-fiber-nanowire microcoil resonator with device sensitivity as high as 700 nm/RIU was reported [146, 195]. An entire review is dedicated to the subject of microcoil resonators and we refer the interested reader to [144]. Also, controlling the melting of a capillary microbubble resonator has been shown to have potential in biosensing by filling the bubble with analyte solution [141–143]. Another fiber-based high-Q cavity with potential applications in biosensing is the bottleneck resonator fabricated by melting a standard single mode fiber [132].

A very promising planar platform for biosensing applications is based on two-dimensional silicon photonic crystals where light can be confined by Bragg reflection and within the photonic bandgap [115, 116, 196, 197]. Photonic crystal cavities can offer ultra-low modal volumes combined with high-Q factors which continue to improve as the design files and the nanofabrication recipes advance [198, 199]. The highest sensitivity for detection with a Q∼9300 photonic crystal microcavity has recently been demonstrated at mass loading sensitivity of 0.8 pg/mm², and with a concentration of only 0.1 μg/ml protein in the pico-molar range [106]. Also recently, a one-dimensional photonic crystal resonator array was introduced for significantly enhancing the light-matter interaction. It was shown that spatially localized optical fields down to mode volumes of a wavelength cubed improve the detection limits of previous designs [115].

Extended photonic crystal membranes without cavities have also been proposed as biosensing elements for detecting particles by “dark field” imaging [200]. In another fundamental study it has been shown that by disordering of the photonic crystal lattice, very high-Q cavities with various modal volumes can be introduced, for example along the length of a photonic crystal waveguide [201]. The random cavities can be potentially used in multiplexed detection and are more easily to fabricate as compared to highly-engineered defect cavities.

We also note that photonic crystal structures that effectively form 1D gratings have been used extensively in biosensing [135]. Because of the low Q, large modal volume and resulting lower sensitivity for detecting a limited number of molecules, such reflectivity filters fall outside the scope of this review and we refer the reader to [202, 203].

An entirely new approach to enhancing the sensitivity of WGM biosensors utilizes the fact that the frequency shift

Figure 9

Optical gain to enhance the sensing performance of optical resonators. (A) Typical transmission spectra of an active resonator (resonator doped with optical gain medium) when the excitation pump for the gain medium is off (left: a broad resonance) and on (right: two distinct narrow resonance [70]). (B) Detection of nanoparticles using mode-splitting in a microcavity laser. A beatnote is generated by photomixing the split lasing modes in a photoreceiver. The changes in beat frequency with time reflect the scatterer induced frequency splitting in the microlaser which indicates a scatterer entering the laser mode volume. If a multimode microlaser is used, each lasing mode undergoes splitting with the splitting amount depending on the overlap of the scatterer with the volume of the lasing mode. Using a multimode laser reduces the possibility of missing a binding scatterer since one splitting signal could capture a binding event that might have been missed by the other one (lower right panel) adapted from [66].

signal produced by a protein or nanoparticle binding to the microcavity is in proportion to the intensity ∼E²(r₀) encountered at the binding site r₀[147], see Eq (2). Any mechanism that can amplify the field intensity at the binding site while maintaining high Q factor will therefore produce a boost in the frequency shift signal, dramatically increasing the sensitivity in single molecule detection. Hot spots of high field intensities can be generated by evanescent coupling of the microcavity resonance to a plasmonic nanoantenna, Figure 10. WGM evanescent excitation of plasmon resonances, for example in a gold nanoparticle, can produce very large field enhancements. Many nanoantenna geometries can be explored for this purpose and their designs are similar to nanostructures that are investigated in surface enhanced Raman spectroscopy (SERS) [208]. Examples for nanoantennas commonly investigated in SERS are nanoparticle or nanoshell dimers [209] and bowtie antennas [210]. Different from SERS, which seeks to enhance the far-field scattering signal, WGM biosensing relies entirely on enhancing the near-field intensity. In fact it is necessary to minimize the scattering loss to maintain high-Q factor which is required for sensitive biomolecular detection. Through careful choice of WGM wavelength it is indeed possible to enhance the near field intensity at the nanoantenna site while minimizing scattering loss. Simulations, theory and experiments indicate that by detuning the microcavity resonance from the plasmon resonance an overall boost in sensitivity by several orders of magnitude is attainable [147, 148, 211, 212]. Indeed, first experimental reports indicate such dramatic sensitivity enhancements [108, 147–149].

The first demonstrations of this plasmon-enhanced microcavity detection scheme utilized a random nanoparticle layer to generate hot spots of high field intensity after coupling to a WGM microsphere cavity [147, 148]. High near-field enhancements were obtained by excitation of a WGM at optimal wavelength. It can be shown that the optimal WGM resonance wavelength is close to the plasmon resonance but significantly detuned from it to limit scattering loss. In theory, optimized plasmon coupling can produce near field enhancements up to three orders in magnitude, increasing the frequency shift signal for detecting a protein molecule and

Figure 10

Plasmonic nanoantennas coupled to microcavities enhance sensitivity in biodetection. (A) A molecule binding to a plasmonic nanoantenna coupled to a WGM microcavity experiences field strengths that are enhanced in proportion to $E^2/E_0{}^2$, boosting the frequency shift upon binding [147]. (B) Up to three orders of sensitivity enhancement has been predicted for protein detection with a nanorod coupled to a toroidal cavity, reprinted with permission from [58], copyright (2011), American Institute of Physics.

bringing label-free single molecule detection within reach [58, 213]. For example, a nanorod immobilized on a microtoroid will boost the frequency shift signal obtained for binding of a single BSA molecule into the MHz-range, which is more than an order of magnitude larger than the expected measurement noise [58]. Other demonstrations for plasmon enhanced frequency shifts have utilized plasmon nanoshells optically trapped with the WGM and delivered towards another gold nanoparticle for several-fold more sensitive detection of the nanoshell [149]. Recently it has also been shown that sensitivity for virus detection can be enhanced with a plasmoncoupled microcavity [108].

The large near field enhancements in a random nanoparticle layer were proposed to be capable of also optically trapping single protein molecules, producing femto-molar sensitivity levels in μl-scale sample volumes [148]. In other demonstrations, the high field strengths of WGMs [214, 215] as well as WGMs coupled to nanoparticles were used to enhance the far-field SERS scattering signal [216], without any utility in boosting the actual WGM frequency shift signal, however. Furthermore, enhanced evanescent coupling to a WGM with the help of a plasmonic nanoparticle has been reported [217]. Interestingly, hybrid photonic plasmonic microcavity structures promise ultra high sensitivity not only in biosensing but also in cavity-QED experiments [213].

It should be noted that microcavities have also been used for the excitation of surface plasmon polaritions (SPP) in a continuous metal film coated on a micodisc-shaped cavity [218]. Such high-Q SPP-WGM has also been predicted in a metal coated WGM resonator [219]. Here, an exterior surface mode possesses high quality factors at room temperature, and can be efficiently excited by a tapered fiber. Furthermore, metal films have been shown to enhance prism-based coupling to WGMs [220]. It was also shown theoretically that a nanosized scatterer immobilized on the microresonator surface enables the efficient coupling to WGMs from a free space beam [221]. Another configuration for realizing hybrid photonic plasmonic microcavities are microring silicon resonators optically coupled to a metal microdisk located inside the ring. Here, optical energy is greatly concentrated in the narrow gap between the metal and the silicon ring [222]. Also a microring resonator with a metal strip has been shown to efficiently couple to a plasmonic nanostrip [223].

Other non-plasmonic enhancements of microcavity biosensing rely for example on confining light in dielectric slotted ring resonators [224] and slotted photonic crystals where high field amplitudes at the slot are accessible for molecules. Slotted photonic crystal cavities were for example used to detect bound molecules at low surface mass densities [197]. Also, slotted ring resonators have been utilized for improved detection limits, for example within a 100 nm slot [122]. In a packaged slot-waveguide ring resonator sensor array, a surface mass density detection limit down to 0.9 pg/mm² has been demonstrated [174]. Enhanced sensitivity of a spherical microcavity biosensor was demonstrated by confining the WGM near the surface using a subwavelength high refractive index coating [152].

Multiple-hole defects have been used to improve the sensitivity in silicon slab photonic crystal cavities [196] by increasing the surface area available for label-free detection without degrading the quality factor. Compared to photonic crystals with L3 defects, adding multiple hole defects into photonic crystal cavities enabled a 44% increase in detection sensitivity. Refractive index sensitivity of guided resonances in photonic crystal slabs could be enhanced by three-fold for suspended slab designs, in contrast to designs with a slab resting over a substrate. Spectral sensitivity values are additionally shown to be unaffected by quality factor reductions, which are common to fabricated photonic crystal nanostructures [225].

A cluster of evanescently coupled 2D WGM optical microcavities (termed “photonic molecules”) were introduced for biosensing with enhanced sensitivity [211]. Here, photonic molecules supporting modes with narrow linewidths and wide mode spacing show enhanced sensitivity to the changes in the dielectric constant of their environment and to the presence of individual subwavelength-sized nano-particles in the evanescent-field region.