Label-Free Biosensors Based onto Monolithically Integrated onto Silicon Optical Transducers

Abstract

The article reviews the current status of label-free integrated optical biosensors focusing on the evolution over the years of their analytical performance. At first, a short introduction to the evanescent wave optics is provided followed by detailed description of the main categories of label-free optical biosensors, including sensors based on surface plasmon resonance (SPR), grating couplers, photonic crystals, ring resonators, and interferometric transducers. For each type of biosensor, the detection principle is first provided followed by description of the different transducer configurations so far developed and their performance as biosensors. Finally, a short discussion about the current limitations and future perspectives of integrated label-free optical biosensors is provided.

1. Introduction

Starting more than 50 years ago from the report of Clark and Lyons [1], biosensors have been the subject of intensive research efforts, but also of numerous attempts of commercial exploitation aiming to overcome the limitations of classical analytical systems and provide solutions for on-site determinations. The research effort has generated over the years an enormous variety of biosensors considering the multiple variations of the more popular ones, such potentiometric or the surface plasmon resonance based sensors [2]. Nonetheless, with respect to the transduction principle, most of the biosensors reported in the literature can fall into one of the following categories: Mass-sensitive, electrochemical, or optical sensors.

Regardless of the transduction principle implemented, biosensors can also be divided into two categories depending on the implementation or not of labels for the analyte determination. Biosensors employing labels are in general more sensitive than the label-free ones, reaching, in some cases, the ultimate detection limit of a single analyte molecule [3]. They are also, generally, more suited for laboratory applications due to the size and complexity of the relative instrumentation. Even so, nowadays, taking advantage of the advances in microfluidic miniaturization and cell phone technology (especially of cell phones capabilities for illumination, imaging, and/or data transfer), portable instruments based on sensing systems employing labels have started to emerge [4,5]. Portable systems are no longer restricted to electrochemical transducers, for which miniaturization of both the sensing element and the instrumentation is more straightforward compared to other transducers, but has been extended to optical ones, the miniaturization of which is hindered by the need to use external light sources and detectors. Despite these advances, the portable systems so far that have been developed based on optical transduction principles are lacking in analytical sensitivity compared to the respective laboratory systems, a fact that restricts to a great extent the potential applications.

On the other hand, label-free optical biosensing systems have also evolved throughout the years, especially concerning their analytical performance, demonstrating, in some cases, detection sensitivities similar to those offered by optical biosensors employing labels [6,7]. This is mainly the result of the progress made the last two decades in the field of nanotechnology, which has given a significant boost in label-free biosensors since it provided viable solutions to long standing bottlenecks, such as the amplification of minute signal changes as well as the realization of compact devices encompassing the transducer along with all the necessary electrical and/or optical connections onto the same chip [6,7,8]. Apart from the quest for outstanding analytical performance, effort has also been devoted to the realization of multiplexed determinations due to the inherent advantage of optical biosensors of reduced cross-over signals as opposed to electrochemical ones [6,7]. This has led to efforts towards applying them for sensing of multiple targets simultaneously (multiplexed sensing), which can help their spread in diverse fields, including the commercial market [6,7,8].

Despite their versatility, the majority of the label-free optical biosensors rely on the same detection principle, namely the evanescent field optics. Therefore, prior to analysing the main current trends regarding label-free optical biosensors, a simplified description of the evanescent field optics principle will be provided. Then, after a short reference to surface plasmon resonance transduction approaches, the review will focus on the integrated optical sensors, especially those based on (a) grating couplers, (b) photonic crystals, (c) ring resonators, and (d) interferometers, including Mach–Zehnder, Young, and bi-modal interferometers. In the last section, monolithically integrated broad-band interferometric transducers will be presented in more detail, and the review will conclude with a comparison of the presented sensors and a brief discussion about the future of integrated optical transducers.

2. Evanescent Field Optics

The evanescent field is generated by the electromagnetic field of light as it transverses a waveguide by means of total internal reflection. The main characteristic of the evanescent field (Figure 1) is that its effect decays exponentially in the medium above the transducer surface. Usually, the evanescent field effect extends to a depth ranging from a few tens to a few hundreds of nanometers depending on the material and the waveguide geometry. Nonetheless, this field is very sensitive to alterations in the refractive index on the transducer surface and can thus “sense” the interaction of a binder immobilized onto the transducer surface with its counterpart molecule/analyte present on the medium over the transducer. More specifically, the binding of the analyte to the immobilized binder molecules induces a change in the effective refractive index on top of the optical transducer, which affects the light guiding properties of the transducer, resulting in some change of the output beam in terms of intensity, polarization, phase, etc. Thus, it is feasible to monitor binding interactions without the need for labels taking advantage of the evanescent field sensitivity to effective refractive index alterations.

Transducers based on evanescent wave field optics have acquired over the years various formats. Nonetheless, the major categories of these sensors include sensor plasmon resonance (SPR), fiber optic, grating coupler, photonic crystal, ring resonator, and interferometric sensors.

SPR is the most widely explored label-free detection principle [8]. Surface plasmons are generated when an electromagnetic wave propagates at the interface between a noble metal thin layer (usually gold) and a dielectric with oppositely signed optical constants. This electromagnetic field creates evanescent waves that penetrate into both adjacent layers. Plasmons are usually excited by means of a light beam guided to the metal/dielectric through a coupling prism, a grating coupler, or a dielectric waveguide (Figure 2a). By changing the angle or the wavelength of the incident light, resonance is achieved when the propagation constant of the evanescent filed matches that of the surface plasmon wave. This angle is influenced by changes in the refractive index of the medium over the dielectric surface. Thus, by monitoring the change of resonance angle or wavelength when binding reactions are taking place onto the dielectric layer, the course of the reaction can be monitored in real-time. Nonetheless, the need for input and output light coupling elements limits considerably the ability for miniaturization and multi-analyte determinations of the classical SPR set-ups. Although multiplexity has to some extent been fulfilled by the development of imaging SPR, miniaturization remains a great challenge.

To this end, sensing approaches based on localized SPR (LSPR) might provide the desired solution [5]. In LSPR, instead of a continuous noble element layer, metallic nanostructures of a sub-wavelength size (in the form of nanospheres, nanorods, or nanodisks) are created on the dielectric layer. Thus, when the light beam strikes at these nanostructures, the free electrons oscillate collectively, scattering light of a specific wavelength range for which resonance is achieved. A schematic of the LSPR detection is provided in Figure 2b. As in the classical SPR, binding reactions can be monitored in real-time, since the associated changes in the refractive index of medium over the nanostructures causes a shift at the extinction peak wavelength [6]. Due to the fact that each nanostructure is potentially an individual sensing element, the multiplexing capabilities of LSPR sensors become apparent. Even so, LSPR sensors still lack in sensitivity compared to classical SPR [9,10,11] that has demonstrated sensitivity resolution in the range between 10⁻⁵ to 10⁻⁷ of refractive index units (RIU), while LSPR sensors has so far demonstrated a sensitivity resolution in the 10⁻⁴ to 10⁻⁶ range [12], regarding bulk refractive index changes detection. Despite that, reports show that LSPR can demonstrate the same detection sensitivity as classical SPR in certain biosensing applications [13,14,15], especially at low analyte concentrations, a finding that might be related to the strong confinement of the LSPR field in the proximity of the nanostructures. Thus, while in classical SPR the evanescent wave field expands a few hundreds of nanometers into the sensor surrounding medium; in the case of LSPR, the electromagnetic field is limited to only a few nanometers from the surface [16]. As a result, SPR might be more sensitive to bulk refractive index changes, but LSPR might be equally efficient in monitoring changes that occur in the close proximity of the surface [16]. As LSPR is a sensing technique widely explored currently, the full capabilities and limitations of this technique are still to be witnessed in the future.

Another configuration of sensors based on the SPR phenomenon relies on metal layers with nanohole arrays [17,18,19]. Although this configuration did not provide for higher detection resolution as compared to standard SPR, it offers increased potential for multiplexing and miniaturization [18,19]. In addition, it facilitates fluidic integration since the nanoholes can act as nanochannels or a “nano-drain”, allowing the flow-through of the reagents and the concentration of biomolecules in the nanoholes, thus improving detection sensitivity [20,21,22]. Another advantage of these sensors is the fact that they can be fabricated with mass-production compatible processes, such as nanoimprint lithography [19,20,21,22], leading to highly reproducible structures that fulfil the requirements of a commercial product. Regarding their application as biosensors, plasmonic nanohole arrays have been used for the determination of the binding kinetics of cholera toxin b to sensors coated with a biomimetic supported lipid membrane containing ganglioside receptors using toxin concentrations as low as 8.4 nM [19], as well as for the direct detection of different types of live viruses (vesicular stomatitis virus, PT-Ebola, and Vaccinia) using sensors functionalized with specific antiviral antibodies [23] (see Figure 3). The latter sensor provided a dynamic range over three orders of magnitude (10⁶–10⁹ PFU/mL) and a projected detection limit of 10⁵ PFU/mL in biologically relevant media, proving the feasibility of its application for detection of viruses directly in human blood.

Looking at the other categories of evanescent field based sensors, i.e., grating coupler, photonic crystal, ring resonator, and interferometric sensors, one can deduce that those based on integrated optics can be more easily down-sized and integrated with the necessary external optical components and fluidics. To the advantages of these sensors, one could count the compatibility of the materials used (silicon, glass, quartz, polymers, etc.) with well-established microfabrication procedures that allow the creation of arrays of sensors with repeatable operational characteristics suitable for multiplex analyses without excessive increase of the complexity and cost of the fabrication.

In the following section, the main categories of integrated optical sensors i.e., grating-couplers, photonic crystals, microring resonators, and interferometers, will be discussed, focusing on the progress and the achievements reported in the last years rather than the particulars of their detection principles. Thus, the improvements of the analytical performance for each sensor category will be presented along with the challenges faced in their application in real-life scenarios as well as the capabilities of each approach for multiplexed analysis and evolution to analytical platforms for on-site determinations. Due to the fact that in each report, a different binding reaction might be used to evaluate the analytical performance of the presented sensor, the detection limit expressed as RIU or as surface mass density (g/mm²) will be used as comparison criterion whenever possible.

3. Integrated Optical Sensors

3.1. Grating-Coupled Waveguide Sensors

Grating coupler sensors are based on planar waveguides with a grating on their surface, i.e., a periodic pattern, that enables the excitation by the incident light of a mode guided in the waveguide when the following condition is satisfied:

$$n_{eff}=n_{air}sin\alpha +l\left(\frac{\Lambda }{\lambda }\right),$$

(1)

where n_eff is the effective refractive index of the waveguide, n_air is the refractive index of air, α is the angle of the incident light, l is the diffraction order, Λ is the grating period, and λ the incident light wavelength. According to this equation, the angle for light in-coupling depends on the effective refractive index of the medium over the waveguide surface (n_eff). Thus, one can monitor binding reactions that take place onto the waveguide surface by monitoring the changes in the incident light in-coupling angle. In an analogous manner, the light out-coupling angle can also be used to monitor reactions that occur at the waveguide surface.

The coupling of the incident light to the waveguide is achieved by guiding polarized laser beams to the gratings at different angles. This process requires precise mechanical movements to achieve alignment of the sensor with both the light source and the photodetector when the in-coupling light angle is monitored [24,25,26]. However, there is no need for precision alignment when the light out-coupling angle is monitored, leading to a simpler experimental set-up that can be used for determinations beyond the laboratory limits [27,28]. The basic configurations of a grating-coupled waveguide sensor are provided in Figure 4.

Grating-coupled waveguide sensors are amongst the first label-free optical biosensor configurations explored. In fact, the first grating coupling set-ups developed during the early ‘90s [25] exhibited sensitivities in the range of 10⁻⁶ RIU as defined through the implementation of model binding assays, such as IgG/anti-IgG or biotin/streptavidin reaction [25,26,27,28]. Soon afterwards, in an attempt to avoid moving optical components, a set-up based on reflection-mode operation was developed, exhibiting similar limits of detection (LOD) in terms of RIU with the previous ones [29]. The analytical performance of the specific set-up was evaluated, implementing both competitive immunoassays [30] and DNA hybridization reactions [31]. To improve the analytical performance of grating coupler sensors, different microfabrication techniques (like imprint lithography) have been employed to create two-dimensional grating structures, resulting in two-fold enhancement of detection sensitivity as compared to one-dimensional grating sensors [32].

Another interesting configuration developed was the so-called wavelength interrogated optical sensor (WIOS), shown in Figure 5a, according to which two gratings were fabricated onto a single mode waveguide, one for light in-coupling and the other for out-coupling [33,34]. The two gratings were designed so that the in- and out-coupling angles differ. Thus, for a fixed angle of incident light, the changes in the refractive index over the waveguide caused by the binding reactions could be monitored by scanning the resonance peak using a tunable laser diode while a multimode fiber was used to collect the emitted light and transfer it to the photodiodes. This configuration allowed for multi-analyte determination by combining four channels, leading to up to 24 sensing areas.

The analytical evaluation of this multiplexed device was performed employing both low molecular weight analytes, e.g., biotin, as well as large biomolecules, demonstrating its ability to reach detection limits as low as 0.3 pg/mm². The device was also implemented for the detection of sulphonamides, a family of widely used veterinary antibiotics, following a competitive immunoassay format according to which the antigen was covalently linked to the sensor surface by means of reaction with a photopolymerized dextran layer [35]. By employing a secondary antibody for signal enhancement, antigen concentrations as low as 0.5 ng/mL could be detected, covering the requirements for detection of these antibiotics in milk samples. Following the same detection principle, a multiplexed sensor for the simultaneous detection of four different classes of antibiotics (34 in total) in milk samples with detection limits similar to that of the sulphonamides sensor was also demonstrated [36]. Furthermore, through a redesign of the microfluidics cartridge [37], an attempt to develop a lab-on-chip device for on-site determinations was reported using the same sensor configuration, which was then applied to the semi-quantitative detection of three families of veterinary antibiotics. The same sensor has been also applied for the multiplexed detection of three cytokines in cell cultures [38].

These reports set the borderline to what has been so far achieved with grating-coupled waveguide biosensors to the direction of multiplexed on-site determinations, since there is not any further significant development reported for most of the last decade, with the exception of a biosensor system that used a micro-electro-mechanical system (MEMS) micro-mirror to interrogate the grating regions of a waveguide by scanning the angle of the incident coherent light (Figure 6). This angle-tunable MEMS sensor allows measurements at various wavelengths and optical powers, thus enabling multiplexed detection in a compact format [39], as the so-called angle interrogated optical sensor (ARGOS) system. Moreover, the sensor demonstrated an LOD of 2 × 10⁻⁷ in terms of refractive index and ability to detect the binding of both high and low molecular weight analytes.

In addition to the several research set-ups, the company, Microvacuum Ltd. [40], commercializes instruments based on the grating coupler principle called Optical Waveguide Light-mode Spectroscopy (OWLS). OWLS immunosensors have been employed for the detection of pesticides, mycotoxins, and bacteria, while using cells as binders, and the detection of antibiotics and pesticides has also been reported [41].

3.2. Microring Resonators

Ring-resonators are amongst the label-free optical transducers with the higher detection sensitivity and increased potential for multiplexed analysis. The basic configuration of a ring resonator is depicted in Figure 7a. As shown, the light propagating to the input waveguide is coupled through the evanescent field to a circular one, on which is propagates in the form of whispering-gallery modes. Changes in the effective refractive index in the vicinity of the ring surface, via, for example, a binding event, affects the spectral position of the whispering-gallery modes and resonance is achieved when the incident light wavelength, λ, is:

$$\lambda =\frac{2\pi n_{eff}r}{m} ,$$

(2)

where m is an integer describing the whispering-gallery modes’ angular momentum, r is the radius of the ring, and n_eff is the effective refractive index. Thus, binding reactions occurring at the ring surface could be monitored by recording the shift in the resonance either through scanning of the output spectrum or by measuring the output light intensity at a fixed wavelength.

Contrary to linear waveguides, in ring waveguides, the interaction of the waveguided light with the surrounding medium is independent of the waveguide length and is determined by the number of revolutions within the ring, which is expressed by the resonator quality factor (Q factor). High Q values (in the order of 10⁶) mean low optical losses and long photon lifetimes, which in turn signify narrow line widths and high peak resolutions, thus leading to high sensitivity. This superior performance with respect to linear waveguides is indicated by the fact that the aforementioned Q values can be provided by ring resonators with a light propagation path ranging between 50 and 200 μm, while for linear waveguides, a length of several cm will be required to achieve a similar performance. This practically means that in the case of ring resonators, high detection sensitivities can be achieved employing devices of a smaller size as compared to linear waveguides, with the concomitant advantages of the realization of resonator arrays in a relatively small surface area.

To take advantage of the inherent advantages of the ring resonators sensors, several configurations have been realized and evaluated as biosensors over the years, including both planar resonators in the form of microdisks [42,43,44] or microrings (Figure 7b) [45,46] as well as 3D resonators in the form of microtoroids (Figure 7c) [47,48,49,50]. Although such structures have been made of glass [51], Si_xN_y-SiO₂ [52] or polymers [53], ring-resonators based on silicon-on-insulator (SOI) are the most widely explored [54].

To this end, SOI microring resonators, with a bulk refractive index LOD of 10⁻⁵ RIU [45], have been reported to provide detection limits around 10 ng/mL in terms of avidin binding to a biotinylated surface [45,54], reaching detection limits comparable to those achieved by other label-free biosensors, such as SPR. An advanced version of this sensor included an array composed of three series of four rings, each series connected to a single input and output waveguide [55]. To discriminate between the four resonators of the same series, each one of them had a different circumference ratio and thus a resonance spectrum that did not overlap with those of the rest of the three resonators. To couple the in-coming and out-coming light, a grating was employed, while the output signal of the 12 resonators was imaged by an infrared camera. To realize multiplexed determinations, the 12-ring resonators’ chip was combined with a suitable microfluidic module [55]. The detection sensitivity achieved in terms of surface mass LOD was calculated at 3.4 pg/mm² and the array was implemented for the multiplexed detection of antibodies against human immunoglobulins (IgG), human serum albumin, and bovine serum albumin.

In another attempt for multiplexed determination, 32 ring resonators were integrated onto the same chip all of them accessed via a bus waveguide; the cladding layer was removed from 24 out of the 32 resonators to create windows for interaction with the sample, while the rest of the eight resonators that remained fully covered with the cladding layer were used as reference in order to compensate for signal drifts caused by temperature variations [46]. A microfluidic module with all the necessary fluid inlets and outlets, channels and reservoirs, as well as gaskets for fluidic isolation of each individual sensor, was fabricated to provide for fluid delivery to the 32 sensors of the array, and consequently allowing for the simultaneous monitoring of 24 reactions taking place onto the respective sensing elements. The device was evaluated at first implementing model binding assays, such as anti-IgG/IgG and biotin/streptavidin reactions, as well as real-time multiplexed analysis of DNA hybridization reactions, demonstrating a bulk refractive index LOD of 7.6 × 10⁻⁷ RIU, which corresponded to a mass LOD of 1.5 pg/mm² [46]. Then, the device was exploited for the multiplexed detection of cytokines through modification of the different sensors of the assay with the respective specific antibodies. Using a secondary antibody for signal enhancement detection limits lower than 0.1 ng/mL in the assay buffer has been reported [56].

The same sensor configuration has also been applied for the single-analyte detection of the cancer marker, a carcinoembryonic antigen (CEA) in serum with an LOD of 25 ng/mL [57], and the multiplexed detection of five markers, namely of prostate-specific antigen (PSA), α-fetoprotein (AFP) CEA, α-tumor necrosis factor (TNF-α), and interleukin-8 (IL-8) [58], with LODs appropriate for detection of these markers in clinical samples. Finally, multiplexed detection of four micro-RNAs was achieved by modification of the microrings with the respective single-strand DNA probes. The detection was completed in about 10 min with LODs in the sub-nM range [59]. Multiplexed label-free detection of human immunoglobulin E (IgE) and human thrombin has been also presented using DNA aptamer modified silicon microring resonators. The surface sensitivity and the limit of detection of the sensor in water were found to be 0.24 nm/ng mm⁻² and 38 pg/mm², respectively, while the detection limit for human IgE and thrombin were 33 pM and 1.4, respectively [60].

A slightly different biosensor based on a silicon dual-microring resonator that comprises one resonator as a sensing element and the other integrated with an electrical controller as a tracing element has also been developed [61] (Figure 8a,b). This configuration allows the detection of resonance wavelength shifts of the sensing microring due to refractive index changes caused by the binding reaction to be performed by the voltage applied to the tracing microring. The advantage of this sensor is that the laser light source can be replaced by a broad-band one, leading to less expensive and compact instruments. The sensor was combined with isothermal solid-phase DNA amplification for the rapid detection of Mycobacterium tuberculosis in less than 20 min using the device shown in Figure 8c [62].

In addition to microring resonators, toroidal-shaped microcavities based on the same detection principle have been developed and exploited as biosensors. Toroids, as planar ring resonators, could be integrated onto arrays while exhibiting much higher Q factors (in the range of 10⁸) than the planar ones [63,64], and were therefore expected to be endowed with a higher detection sensitivity. Thus, toroids fabricated on silicon wafer by standard lithography techniques have been used for the detection of cytokines through immobilization of specific anti-cytokine antibodies onto a protein G functionalized surface [47]. The LODs achieved using cytokine solutions prepared in buffer were in the range of 5 aM (5 × 10⁻¹⁸ M), whereas the working range extended up to 12 orders of magnitude. This impressive performance led to single-molecule [46] or single particle detection [65]. At the same time, the detection of the bacteria, Helicobacter hepaticus, at a density of 1 × 10⁴ cells/mL within 750 s has been demonstrated using an optical microcavity-based sensor [66]. Toroidal structures have also been fabricated on low-loss polymer [40] or polymer-silica hybrid materials [67]. These devices are expected to have lower Q factors (10⁵–10⁷) as compared to silicon based ones, but still their sensitivity should be higher compared to planar ring resonators. This, however, is still to be proved.

3.3. Photonic Crystal Waveguides

Photonic crystals are nanostructures with a periodicity of the order of a wavelength in one, two, or three dimensions (1D, 2D, or 3D photonic crystals, respectively). This periodicity creates photonic bandgaps where light cannot propagate in the crystal. Defects in the periodic structure induce a “defect mode” within the bandgap, which has as a result the propagation of the incident light in the photonic crystal when resonance with the defect mode is achieved. The spectral position of the defect mode is influenced by the refractive-index changes in the vicinity of the defect, thus, offering a way to exploit the photonic crystal as a biosensor. This is, in practice, achieved through careful design of the nanostructure to define the photonic bandgap and the defect mode wavelength, and thus, define the sensitivity of the transducer to refractive index changes [60]. The structures more usually implemented are linear or 2D gratings on which discrete or line defects are introduced to create the transduction sites (Figure 9). Due to their operation principle, photonic crystal sensors can have a very small size, since the light should be confined in the periodic pattern in the proximity of the defects.

The first photonic crystal biosensors [68,69,70,71] were based in 1D polymer gratings covered with a layer of TiO₂, and the binding reactions could be monitored through the recording of the shift of light reflected from the sensor surface when probed perpendicularly with white light. Multiplexed detection is easily achieved since different areas of the transducer can be interrogated at the same time without any cross-talk effects [72]. Practically, this is achieved through integration of the transducer to microtiter wells or microarray slides [72], or through combination with microfluidics. The latter approach provides a shorter analysis time and higher detection sensitivity as compared to static assays [73]. These sensors have been applied to cell-based assays, protein and virus detection, drug screening, and protein-protein interactions [74]. Moreover, a commercially available system based on this technology has been launched by SRU Biosystems [75]. (www.srubiosystems.com).

In addition to gratings, waveguides with 1D and 2D arrays of holes are also popular planar photonic crystal structures, on which the guided wave is generated by missing holes on the pattern, as shown in Figure 10a [76]. Thus, hexagonal lattices of holes with a single missing row line defect and point cavities in different arrangements were fabricated on silicon-on-insulator (SOI) photonic-crystal waveguides, demonstrating a bulk refractive index LOD of 10⁻³ RIU, a mass sensitivity of 24.7 nm/pg, and a LOD of 500 pg/mm² for direct adsorption of bovine serum albumin (BSA) or avidin onto the sensor surface [77]. Another configuration consisted of a photonic crystal waveguide with a length of 20 μm, lattice constant of 390 nm, and holes radius of 111 nm [78,79]. This transducer was combined with an appropriate microfluidic module to monitor the binding of an anti-BSA antibody to BSA immobilized onto the transducer surface, demonstrating a mass LOD of 2.1 pg/mm² [78] as well as hybridization reactions with LODs in the nM range [79].

Another research group exploited the fact that the resonant wavelength can be tuned by changing the defect cavity spacing, and created the nanoscale optofluidic sensor array (NOSA), a microcavity structure evanescently coupled to an adjacent single-mode silicon waveguide [80]. Through adjustment of the cavity spacing, resonator arrays of different Q factors were realized on a single waveguide, allowing for multiplexed detection when combined with an appropriate microfluidic module, and providing a bulk refractive index LOD of 7 × 10⁻⁵ RIU [80]. The particular transducer has been applied to hybridization experiments, as well as for the multiplexed determination of interleukins 4, 6, and 8 [81].

A two-dimensional photonic crystal-based transducer was also fabricated on SOI (Figure 10b) and evaluated using the biotin-streptavidin binding reaction, providing a mass LOD of about 2.5 fg [82]. An alternative design of SOI-photonic crystal sensors employed the creation of the defect line adjacent to the photonic microcavities, resulting in a multichannel sensor [74] with a bulk refractive index LOD of about 10⁻² RIU. Regarding the detection of biomolecular interactions, an LOD of 67 nM was determined for an anti-IgG/IgG reaction [83].

Another interesting design combined a photonic crystal with a ridge waveguide, allowing the detection of anti-biotin antibodies binding to sensor-immobilized biotin-BSA with an LOD of 20 pM, corresponding to a mass sensitivity of about 4.5 fg [84]. A photonic crystal sensor based on total-internal-reflection was employed for the label-free detection of the cardiac biomarker, Troponin I, in human plasma samples with a detection limit of 0.1 ng/ mL [85], whereas the shifts in the wavelength of photonic crystal nanolasers created on GaInAsP (Figure 10c) have been implemented to monitor the Limulus amebocyte lysate reaction to detect endotoxin at a concentration of 0.0001 EU/mL within 33 min [86].

In conclusion, photonic crystal sensors are characterized in general by lower detection sensitivities compared to other types of integrated optical sensors. To improve their analytical performance, instead of bulk fictionalisation with binding molecules, their confinement onto the holes where the light confinement is also maximum has been exploited as a mean to increase the resonance shifts to a binding event. This way, the detection of single particles, virus, or pathogen cells has been achieved [87,88].

3.4. Integrated Interferometers

Interferometric sensors are the most abundant class of integrated optical sensors with a variety of configurations employed in biosensing applications. The most popular amongst the different interfererometric traducers are: Mach-Zehnder (MZI; Figure 11a), Young (YI; Figure 11b), and bi-modal interferometers (Figure 11c) [89].

An integrated MZI is essentially an integrated waveguide (made almost exclusively from silicon nitride or silicon oxynitride) that at some point splits into two arms, the sensing and the reference, which recombine again after a certain distance (Figure 11a). The reference arm is either covered by a cladding layer or appropriately passivated to minimize non-specific binding of the analyte, while the sensing arm is exposed to the analyte solution and modified with binder molecules specific for the analyte. Thus, in the sensing arm, the waveguided light could interact through the evanescent field with the adlayer formed during the binding reaction, which induces a change in the effective refractive index, resulting in a phase difference between the light propagating in the sensing and the reference arm. For the integrated MZI to be used as a sensor, the waveguides should be monomodal, otherwise, due to the differences in the propagation velocities of multiple modes, the output signal will be impossible to correlate to effective reactive index changes. In monomodal waveguides, when the two arms recombine, the output light intensity, I, and the phase difference, ΔΦ, are provided by Equations (3) and (4), respectively:

$$I=\frac{I_0}{2\left[E_{S^2}+ E_{R^2}+2E_S{E}_{R}cos\Delta \mathsf{\Phi }\right]} ,$$

(3)

$$\Delta \mathsf{\Phi }=\frac{2\pi L}{\lambda }\left[n_{eff,S}-n_{eff,R}\right],$$

(4)

where I₀ and I are the light intensity at the input and output, respectively, E is the electric field propagating along the waveguide, n_eff is the effective refractive index, Φ is the phase, L is the interaction length in the sensing arm, λ is the light wavelength, and S and R refer to the sensing and the reference arm, respectively. According to Equation (3), the output light intensity has a cosine dependence with the phase difference, which means that the sensitivity to effective refractive index changes would be maximum at the quadrature points and minimum in the vicinity of the extrema. In addition, according to Equation (4), it is expected that an increase of the sensing arm interaction length would result also in a sensitivity increase.

Additional parameters that regulate the MZI performance as a refractive index sensor are the geometrical characteristics of the waveguide and the difference in the refractive index of the materials used for the waveguide and the cladding layer. To this end, the waveguides were, almost from the first MZI sensor, made of glass, SiO₂, Si₃N₄, and, in some cases, of polymers, whereas the cladding layer is usually SiO₂ [90,91,92,93,94].

Integrated MZIs have been evaluated as biosensors from the first reports [90,91], in which antibody-functionalized transducers have been used to detect human chorionic gonadotropin at concentrations as low as 50 pM within 20 min. The LOD was reduced to 30 nM when an MZI sensor based on thinner waveguides was fabricated [92]. Another research group developed an MZI sensor based on rib SiON waveguides and applied them to the determination of 2,4-dichlorophenoxyacetic acid (2,4-D) with an LOD of 1 mg/mL following a competitive immunoassay format [93]. In another, design, a 3 × 3 coupler was integrated to the MZI output, improving the signal-to-noise ratio by a factor of 10, and allowing the detection of the pesticide, simazine, with an LOD of 0.1 ng/mL [94]. Silicon MZIs based on total internal reflection [95] or anti-resonant reflecting optical waveguide MZIs (ARROW-based MZI) [96] have also been reported by the same group. The first configuration provided a bulk refractive index LOD of 8 × 10⁻⁶ RIU, which corresponded to a surface sensitivity of around 2 × 10⁻⁴ nm⁻¹, and an LOD of 0.06 pg/mm², when evaluated through DNA hybridization experiments. On the other hand, ARROW-based MZI provided a refractive-index detection limit of 2.5 × 10⁻⁶ RIU when applied for the detection of the insecticide, carbaryl, following a competitive immunoassay format [96]. In an attempt to minimize the optical transmission losses and optimize the MZI performance, the divisors that split and recombine the two arms were redesigned and the sensor was applied to the detection of L. monocytogenes at a concentration of 2.8 × 10⁵ colony forming units (CFU)/mL, which is lower that the infective dosage for humans, and a dynamic range of five orders of magnitude [97].

In all the above reports, the MZIs were symmetric, as shown in Figure 11a, i.e., the sensing and the reference arms had equal lengths; nonetheless, asymmetric MZIs with small path differences between the two arms have been also fabricated. Such an MZI has been employed for the immunochemical detection of aflatoxin M₁ (AFM₁) in milk samples [98]. The sensor LOD for AFM1 was 3 mg/L, which is orders of magnitude higher than the maximum allowable limit in the EU (0.05 mg/mL), thus requiring a sample pre-concentration.

Another MZI sensor design that was exploited for biosensing was based on a silicon nitride slot waveguide, in which the sensing arm of the sensor consists of a slot waveguide while the reference arm consists of a strip waveguide. A slot waveguide consists of two slabs of high refractive index material separated by a nanometer-scale low refractive index slot region and surrounded by low refractive index cladding materials [99] (Figure 12a). A bulk refractive index sensitivity of 1864π/RIU was recorded, which translated to an LOD of 18.9 fM in terms of streptavidin in a biotin/streptavidin model binding reaction experiment [100]. The same sensor was applied for the quantification of the methylation of the death-associated protein kinase (DAPK) gene, a biomarker for human cancers, at a concentration as low as 1 nM [99]. An array of such MZIs was also fabricated and applied for the detection of multiple microRNAs in urine samples. It was shown that the sensor could discriminate single nucleotide polymorphism of the let-7 family of miRNAs from synthetic and cell line samples in an assay time of 15 min [101]. Moreover, an instrument was built around the slot MZIs array, as shown in Figure 12b, which could be evolved to a point-of-care device.

Although the majority of integrated MZI sensors depend on devices fabricated on silicon materials, polymers have also been implemented due to their lower cost compared with silicon-based ones. Thus, a polyimide-based MZI modified with streptavidin to couple a biotinylated anti-human IgG antibody was applied for the determination of human IgG with an LOD of 3.1 nM for direct binding, which was reduced to 30 pM when a second anti-human IgG antibody was employed [102]. Considering the LODs achieved so far, polymer-based MZIs could evolve to low-cost, disposable immunosensors.

In addition to the various research efforts, there is an MZI-based system commercialized by Optiqua Technologies Pte Ltd (http://www.optiqua.com), indicating the potential of MZI-based sensing systems for real-world applications.

Young interferometers (YI) are also based on waveguides that split into two arms through a Y-junction; nonetheless, the two arms do not recombine, but the respective light beams interfere in free space, as in Young’s famous two-slit experiment (Figure 11b). Thus, the output signal consists of an interferogram that can be depicted onto a CCD camera. As in MZIs, changes in the effective refractive index over the sensing arm due to binding reactions result in a phase difference between the two interfering beams that is recorded as a spatial shift of the interference fringes according to Equation (5):

$$I(\mathsf{\lambda })=I_1+I_2+2\sqrt{I_1{I}_2}\cos [\frac{2\mathsf{\pi }b}{\mathsf{\lambda }L}x+\Delta \phi ]=I_1+I_2+2\sqrt{I_1{I}_2}\cos [\frac{bk}{L}x+\Delta \phi ],$$

(5)

where I₁ and I₂ are the light intensities of the two beams, b is the distance between the two arms, L is the distance between the waveguide end and the readout camera, and x is the position of the fringe pattern on the camera. The phase difference, Δφ, between the two arms is provided by the same equation with the MZIs (Equation (4)).

Despite the fact that YIs have evolved almost in parallel with MZIs [103], their development was not so widespread, although they are characterized by a simpler read-out as compared to MZIs, and are less sensitive to temperature and wavelength drifts. Amongst the first YIs developed, the same group presented two devices, one based on silicon nitride waveguides [104] and the second on Ta₂O₅ waveguides [105]. Both sensors were at first evaluated with model binding assays, and the LODs achieved in terms of refractive index changes were 9 × 10⁻⁸ and 9 × 10⁻⁹, respectively. Then, the second device was modified with antibodies and used to detect proteins and pharmaceutical substances (methotrexate) in serum [106] to monitor the production of a recombinant protein in cell lysates [107], and to detect tuberculosis-specific antibodies in serum samples from infected patients [108]. It should be noticed that the YI-based sensor was found to be more sensitive than an SPR, a grating coupler, and a commercially available reflectometric interference spectroscopy (RIfS) sensor.

Another research group developed a four-channel integrated YI that allowed for independent multiplexed detections (Figure 13a), through combination with an appropriately designed microfluidic module (Figure 13b), while using one of the sensors as a reference [109]. The devise demonstrated a bulk refractive index LOD of 8.5 × 10⁻⁸ RIU, which is one of the lowest LODs in terms of mass coverage resolution (20 fg/mm²) [110]. In addition, the same configuration was employed in the immunochemical detection of herpes simplex virus type 1 (HSV-1), demonstrating an LOD of 850 particles/mL in buffer, which was a bit more elevated in serum [111].

In addition to silicon-based YIs, polymeric integrated YI sensor chips, with roll-to-roll mass-manufactured waveguides, have also been reported, which were able to detect a refractive index difference of 6.4 × 10⁻⁶ RIU [112]. The polymeric YI were functionalized with a molecularly imprinted polymer (MIP) as the binder and used for the detection of melamine, whereas the multi-analyte capabilities have been demonstrated through the detection of C-reactive protein (CRP) and human chorionic gonadotropin (hCG) using YIs functionalized with specific antibodies [113]. Nonetheless, the evaluation of the analytical performance of the specific sensors was not complete.

Bimodal waveguide interferometers are most recently developed, and consist of a single waveguide with two different zones; a first one with single-mode behaviour and a second one supporting two modes (zero- and first-order modes). These two modes propagate at different velocities depending on the refractive index of the cladding layer, and thus, when the refractive index changes as a result of a binding reactions occurring at the waveguide surface, the interference pattern at the waveguide output also changes (Figure 14A). The sensor exhibited a bulk refractive index LOD of about 5 × 10⁻⁷ RIU, which is similar to those achieved using other interferometric configurations [114].

Bi-modal interferometric sensors have been used for the immunochemical detection of clinically relevant protein analytes, such as human thyroid stimulating hormone (hTSH) [115] and human growth hormone (hGH), in undiluted urine matrix [116] at the picomolar range; of Bacillus Cereus Escherichia Coli to densities down to 12 CFU/mL and 4 CFU/mL, respectively [117]; of micro-RNAs related to bladder cancer at attomolar concentrations [118]; as well as of the pollutant Irgarol 1051 in seawater with an LOD of 3 ng/L that is below the maximum allowable concentration in a seawater set by the EU (16 ng/L) [119].

In general, the analytical performance of bi-modal interferometers is comparable to that of other interferometric transducers and in particular to single-wavelength MZIs. Moreover, the sensor chip (Figure 14B) was combined with a valve-based microfluidic module appropriate of automating complex fluid handling (Figure 14C). The module was fabricated on a poly(dimethylsiloxane) (PDMS) following an injection moulding technique that enabled the creation of integrated pneumatically actuated elastomeric valves in a mass production compatible way. The module was designed so as to achieve the movement of the reagents required for the performance of competitive immunoassays, and the device was applied for the detection of the antibiotic, tetracycline, as a proof-of-concept assay [120].

4. Monolithically Integrated Broad-Band Optoelectronic Transducers

The integrated optical transducers described so far employed in their majority monochromatic light sources, i.e., lasers. The replacement of the bulky and energy consuming lasers by broad-band light sources was explored in an attempt to realize sensors suitable for on-site determinations, but also as a solution to the phase ambiguity that limited the MZI performance. Therefore, the first efforts to replace the laser light source by a polychromatic one were performed with MZIs.

In the first reports [121,122], a broad-band light source was coupled to an MZI, and the output light was recorded by an external optical spectrum analyzer in the 1200–1700 nm spectral window. The detection sensitivities achieved were rather moderate, but the potential to overcome the phase ambiguity issue was demonstrated. In an effort to create a low-cost integrated nanophotonic lab-on-a-chip platform, a silicon chip that included multiple Mach–Zehnder interferometers along with an on-chip optical spectral analyser consisting of an arrayed-waveguide grating as well as a grating for the coupling of broad-band light was fabricated [123]. The sensor exhibited a refractive index LOD of 6 × 10⁻⁶ RIU that is comparable to those of LODs achieved with standard MZIs. In addition, the bioanalytical capabilities of the device have been demonstrated through the direct detection of an antigen related to tuberculosis in urine samples as well as the detection of the inflammation marker, C-reactive protein, in buffer.

Similarly to MZIs, it was expected that a broad-band source will enhance the analytical capabilities of the Young interferometers. Thus, it was proven through simulations that since the evanescent field is smaller for shorter wavelengths and more expanded for longer ones, it is possible, through the implementation of broad-band sources to discriminate between biomolecules and bulk RI changes or larger particles, thus offering the ability to distinguish specific binding from non-specific binding and bulk RI changes. The predicted LOD was about 10⁻⁶ RIU; nonetheless, these set-ups have not yet been evaluated with binding reactions [124].

Despite the aforementioned efforts for the integration of optical components onto the same chip with the transducer, the light source was not integrated and, even with the more compact broad-band sources, the device size remained substantial. Moreover, the light coupled to the device was just a little fraction of the incident light [125]. The only reliable solution provided so far to this problem was the implementation of silicon light emitting diodes (LED) integrated onto the same silicon chip with planar silicon nitride waveguides [126] as light sources. The LEDs were silicon avalanche diodes that emitted polychromatic light covering all the visible and near IR spectrum when biased beyond their breakdown point [127]. To take full advantage of the integrated light source and due to the low intensity of the emitted light as compared to laser sources, the coupling efficiency had to be maximized and the propagating light losses to be minimized. This was achieved by careful design of the waveguide bending from the horizontal to the vertical direction towards both the light source and the detector, and the alignment of the up-going waveguide segment with the integrated LED. The smooth bending was achieved through the creation of SiO₂ spacers with a curvature radius much larger than the waveguide thickness, as shown in Figure 15, where the main steps for the fabrication of LED are presented. The alignment of the LED to the up going segment of the waveguide was realized using the waveguide as a mask during the procedure for the creation of the LED. In detail, the steps followed to realize the integrated LED included: (i) A 2-μm thick thermally grown oxide layer was deposited on the silicon substrate; (ii) the base (N+) side of the avalanche junction was formed by phosphorus implantation in lithographically defined windows in the thermally grown oxide; (iii) a 2-μm silicon dioxide layer was created over the thermal oxide by tetraethyl orthosilicate (TEOS) deposition followed by lithographic patterning, and anisotropic reactive ion etching in CHF₃, to end-up with an overall oxide thickness of 2.5 μm; (iv) a nitride film was deposited by low pressure chemical vapour deposition (LPCVD) in a mixture of NH₃ and SiH₂Cl₂, and lithographically patterned and etched in CHF₃ anisotropic reactive ion plasma to create the strip waveguides, the one end of which was just above the avalanche diode base area. The thickness of waveguides ranged between 100–150 nm, aiming to increase the sensitivity due to multiple interactions of the evanescent field with the analyte; (v) boron was implanted through the nitride film to form the avalanche junction by phosphorus compensation followed by rapid thermal annealing of the boron implant; and (vi) the cladding layer was formed by silicon oxide deposition and etched away from the middle of the waveguides to create the sensing window. Following this procedure, the misalignment of the avalanche diode emitter to the base was restricted to less than 1 micron, resulting in a light coupling efficiency of about 40% [118]. In addition, the LED was found to be very stable in long-term use, facilitating the implementation of the integrated transducers for biosensing applications.

The first editions of the fully integrated devices described above included arrays of linear waveguides on which signal transduction was based on the modulation of the optical coupling efficiency due to the binding from the immobilized binder molecules of counterpart ones labelled with light absorbing (e.g., fluorescent molecules) or scattering moieties (e.g., gold nanoparticles) [126]. Label-free detection was attempted with the particular transducer through engineering of the waveguide surface with gold nanoparticles [128] or through the creation of a gradient of binder molecules onto the waveguide surface [129]. In the first approach, signal transduction was based on surface plasmon resonance of the surface-immobilized metal nanoparticles. On the second one, transduction was based on the creation through photolithography and plasma etching of one-dimensional pattern of alternating lines of protein-adsorbing and non-adsorbing areas. Binding reactions with the immobilized onto the pattern protein resulted in the filtering-out of resonant photons, providing a mean to monitor them in real-time without using labels.

Nonetheless, the label-free detection capabilities of the integrated transducer were fully exploited, then the integrated waveguide was patterned as a Mach-Zehnder interferometer [130,131,132]. The polychromatic nature of the integrated LED led to the establishment of a new detection principle defined as Frequency-Resolved Mach-Zehnder interferometry [131]. It was shown that the new detection principle was able to surpass the phase ambiguity and signal fading limitations of standard single wavelength MZIs. This resulted from the fact that the phase change is different for each wavelength, and thus the effect of the effective refractive index change due to the binding reactions taking place on the sensing arm of the MZI can be easily deduced from the full transmission spectrum as opposed to a single wavelength. Regarding the fabrication of the integrated MZIs, the LED and the silicon oxide spacers were the same as in the linear transducer; nonetheless, a mode filter was introduced prior to the waveguide to ensure that the light transmitted was monomodal.

Two transducer configurations have been realized; the first described as “fully-integrated” (Figure 16a) and the second as “semi-integrated” (Figure 16b). The fully integrated version incorporated on a single-chip the emitting source, the MZI, and the detector in the form of a photodiode, and provided an LOD of 1 × 10⁻⁵ RIU in terms of the refractive index, while using model binding reactions, the detection of 1 nM of streptavidin and 10 nM of anti-mouse IgG was demonstrated [133]. In the semi-integrated version, the whole output spectrum was recorded, employing an external spectrophotometer, and the spectral shifts over the entire LED spectrum were monitored in real-time. In addition, the two arms of the MZI have been engineered so their phase difference has a linear dependence on the wavelength [132]. Under this condition, the transmission spectrum contains two characteristic frequencies, one for the TE and one for the TM, which can be deconvoluted through signal splitting in the Fourier transform domain, offering the ability to discriminate between cover refractive index changes and binding events occurring at the MZI sensing arm. Regarding the analytical performance of the semi-integrated device, a bulk refractive index LOD of 5 × 10⁻⁶ was determined, while for the model binding assays of biotinylated bovine serum albumin-streptavidin and anti-mouse IgG-mouse IgG, LODs of 5 pM and 32 pM, respectively, were demonstrated, which were considerably lower than that achieved with the fully-integrated device [133].

Due to the superior performance of the semi integrated configuration, the sensor was applied for single- and multi-analyte immunoassays. Thus, the detection of bovine milk in goat milk at levels as low as 0.04% (v/v) and with a dynamic range that varied from 0.1 to 1.0% (v/v) for application to goat milk adulteration was reported [134]. The device was also applied for the detection of mycotoxin ochratoxin A in beer samples with a detection limit of 2.0 ng/mL and a dynamic range of 4.0–100 ng/mL, demonstrating the potential for applications of the device in food analysis [135]. Another single analyte application concerned the determination of CRP, a widely used inflammation marker, in human serum samples with an LOD of 2.1 ng/mL and a quantification limit of 4.2 ng/mL [136]. The CRP concentrations determined in human serum samples were in agreement with those determined for the same samples by a standard clinical laboratory method run in a clinical analyser, further supporting the accuracy of the measurements performed with the developed device.

In addition to single analyte determinations, multiplexed determinations were also performed through appropriate functionalization of the different MZIs of a single chip with the different biomolecules. Thus, the multiplexed detection of mycotoxins, aflatoxin B₁, fumonisin B₁, and deoxynivalenol in beer samples was presented following a competitive immunoassay format with LODs of 0.8, 5.6, and 24 ng/mL for aflatoxin B₁, fumonisin B₁, and deoxynivalenol, respectively, and an overall assay duration of 12 min [137]. Through the analysis of beers of different types and origin, the sensor performance was compared with those provided by established instrumental laboratory methods [138]. In addition, the simultaneous, label-free immunochemical detection of four allergens, bovine milk protein, peanut protein, soy protein, and gliadin, in 6.5 min, with LODs of 0.04, 1.0, 0.80, and 0.10 μg/mL, respectively, has been reported. The sensor was evaluated by analysing samples from the cleaning in the place system of a dairy industry and the results obtained were in good agreement with those received by commercial Enzyme-Linked ImmunoSorbent Assays (ELISAs). Apart from the excellent analytical performance, the sensor due to the small chip size (a 10-MZI array is accommodated in a chip of 4.25 × 8.0 cm²) is suitable for the development of a portable instrument for accurate point-of-care applications.

In addition to polychromatic MZIs, the broad-band interferometry principle was also explored in polychromatic YI [139]. It was demonstrated that the use of a broadband source led to an interferogram consisting of two distinct fringe packets, one for each polarization, making feasible the independent determination of the phase signal for the two polarizations over a range of wavelengths. The dual polarization detection ability makes possible the deconvolution of the effect due to cover medium changes from that due to a binding reaction. In the first experimental set-up, an interference filter was employed and only the selected region of the wide emission spectrum of the integrated LED was monitored [140]. Using this set-up, the simultaneous monitoring of biotin-streptavidin and rabbit IgG-anti-rabbit IgG binding reactions was demonstrated. Concentrations of 1 nM streptavidin and 10 nM anti-rabbit IgG resulted in peak shifts of more than one period, suggesting that LODs down to the sub-nM range are feasible, making broad-band YI a competitive sensing approach to other label-free transduction principles.

5. Conclusions and Outlook

In

Table 1. Comparison of label-free integrated optical biosensors.

Sensor Type	Detection Resolution (RIU)	Multiplexing Capability	Portability	Overall Cost
SPR	10−5 to 10−7	+	+	+++
LSPR	10−4 to 10−6	+++	++	++
Nanohole arrays	10−4 to 10−6	+++	++	++
Grating couplers	10−6 to 10−7	++	++	++
Microring resonators	10−6 to 10−8	+++	++	++
Photonic crystals	10−5 to 10−7	+++	++	++
MZIs	10−5 to 10−7	+++	++	++
YIs	10−8 to 10−9	+++	++	++
Bimodal Interferometers	5 × 10−7	+++	++	++
BB-MZIs (semi-integrated)	5 × 10−6	+++	+++	+

+ low, ++ medium, +++ high.

, the types of biosensors reviewed are compared in terms of detection resolution, expressed in RIU, multiplexing capability, portability, and overall cost of transducer fabrication and instrumentation.

From the information provided in Table 1, as well as at the previous sections it is obvious that amongst the biosensors developed so far, those based on integrated optical transducers demonstrate a high detection resolution that it is translated to excellent analytical performance in different application fields ranging from biodiagnostics to environmental monitoring and food safety. This is to a great extent the result of the great versatility in transducer design enabled by the flexibility of the available principles of operation. Moreover, integrated optical transducers seem to be the only viable solution towards the development of portable platforms that could be used for analysis outside the analytical laboratory, i.e., at the points of need. To this end, an additional asset is the fact that these transducers are fabricated with techniques compatible with large-scale production at a reasonable cost. Despite these advantages, for the time-being, only a few commercial instruments are currently available and the majority of the research efforts has resulted in prototypes that work efficiently only in a laboratory environment. The limiting factor is no longer the integration of all components in a compact device, but the standardization of all the biosensor components as well as of the on-chip assay procedures, the handling of the sample and liquids in general, in a manner that will retain the small size and the portability benefits without affecting the sensor performance. Provided that these challenges will be addressed, this versatile class of optical biosensors could be proved to be the ideal solution for a wide range applications where on-site analytical determinations of high accuracy at short times are required. These applications are expected to expand beyond the already identified ones in food and beverage safety assessment, environmental monitoring, and quantification of disease markers in biological samples. Taking into account the continuing research effort as well as the companies that show an interest in developing or commercializing relevant systems, the label-free biosensors based on integrated optical transducers seem to have not only an interesting past, but a very promising future too.