1. INTRODUCTION

Chip-scale integrated photonic sensing systems have been rapidly developed and have begun to play essential roles in point-of-care (POC) diagnosis [1]. In the past decade, with the increasing demands in early detection of fatal diseases such as cancer, it is highly desirable to develop chip-scale integrated photonic sensors with ultrahigh sensitivity [1,2]. Whispering-gallery-mode- (WGM-) based resonators that have high $Q$ factors and are capable of integration with optical communication networks have thus been intensively studied. In 2003, Vollmer et al. reported the detection of protein through wavelength shift of a silica microsphere [3]. While the optical path changes are negligibly small by the protein, the circulating waves within the high-$Q$ ($Q\sim 7.4\times {10}^5$) resonators can drastically enhance the detection sensitivity. Soon after, single molecule detection was experimentally realized in a higher-$Q$ microsphere ($Q>{10}^8$) [4,5]. With the progress in nanofabrication technology and sensing techniques, high-$Q$ microtoroids have been employed to detect proteins, viruses, and nanoparticles, making the highly sensitive detecting systems compatible with silicon systems [6–12].

Motivated by the miniaturization of sensor systems and the prospect for integration with signal processing/communication, the research interests in highly sensitive photonic sensors have gradually turned to high-$Q$ silicon resonators [13]. By optimizing the nanofabrication process, the $Q$ factors of silicon photonics have been improved to around ${10}^6$ in the past decade [14,15]. Very recently, such a kind of high-$Q$ silicon microdisk has been massively fabricated with standard photolithography for the first time [16]. Such progress has circumvented the barriers between silicon microdisks and highly sensitive detection, and thus triggered intensive research attention on high-$Q$ silicon microdisk-based photonic sensors. Washburn et al. have experimentally verified the capability of silicon microdisks ($Q\sim {10}^5$) in detecting 2 ng/mL carcinoembryonic antigen (CEA) without additional labeling [17]. This detection limit has been further improved to below 1 pg/mL by combining the silicon resonator with plasmonic nanostructures [18]. Now, the sensitivity of silicon microdisk sensors is approaching the requirements of cancer diagnosis in very early stages. However, the practical applications of microdisk-based optical sensors are still facing severe challenges in portability, cost, and integration.

In WGM sensors, the light is evanescently coupled into the microresonators [3–18]. Typically, the tapered fiber is usually too sensitive to the ambient vibration and not applicable to portable devices [3–11,14,16]. Although the silicon waveguide is fixed on the silicon-on-insulator (SOI) wafer and is immune to the vibration, the waveguide width and waveguide–microdisk separation distance must be precisely controlled by high-precision nanofabrication, such as electron-beam lithography, to realize high-efficiency coupling (see Supplement 1) [12,13,15,17,18]. Consequently, the portability and costs of conventional WGM sensors are strongly restricted, and a breakthrough in a high-efficiency coupling mechanism is highly desirable for applications in POC devices.

2. RESULTS AND DISCUSSION

A. Working Principle

To simplify the structures and reduce the cost, we have dropped the evanescent coupling configuration and proposed a new coupling mechanism. As schematically depicted in Fig. 1(a), the new design is a circular Si microdisk connected by a Si waveguide in the radial direction. Intuitively, due to the normal incident angle, one may consider that the incident light from the waveguide will mostly transmit out. The truth is quite different due to the presence of resonances. Basically, the constructive interference between the incident light and the resonant modes can significantly improve the coupling coefficient. This injection process is more straightforward, considering its time-reversal process. In a waveguide-connected microdisk (WCM) microlaser, there are two main decay channels, i.e., the direct tunneling and the leakage to the channeling waveguide. Due to the broken total internal reflection (TIR) at the waveguide–cavity joint position, the latter one is orders of magnitude larger and thus dominates the far-field emission [see Fig. 1(b)]. These kinds of directional outputs and their near-unity collection efficiencies have been experimentally verified in microdisk lasers [19,20]. In principle, as the above process does not involve nonlinear and time-dependent parameters, the unidirectional emission from WCM must be time reversible, and the light on resonance will be efficiently injected into the WCM via the channeling waveguide.

 

Fig. 1. Design for end-fire injection. (a) Schematic picture of the WCM configuration. (b) Unidirectional emission along the channeling waveguide (see inset) and the dependence of collection efficiency on waveguide width ($w$). (c) Reflection spectrum at channeling waveguide. The inset is the numerically calculated field pattern, which is a WG-like resonance with minimal field distribution at the waveguide–disk joint position. (d) Input coupling and reflection coefficients as a function of waveguide width $w$ and tilt angle $\theta$ [the mode marked as red circle in (c)]. 1° equals 87.3-nm position shift. Compared with the evanescent coupling, this new configuration is robust to the waveguide width and thus is independent of the highly precise nanofabrication technologies.

Download Full Size | PDF

The above process is termed as “end-fire injection” [21,22]. Figure 1(c) shows the transverse electric (TE)-polarized reflection spectrum at the channeling waveguide with three-dimensional finite-difference time-domain simulation. Here, the radius $r$ and the width of waveguide $w$ are 5 μm and 550 nm, respectively. The refractive indices of silicon and silica are set at 3.45 and 1.46, respectively. Different from the typical add–drop microdisk transmission (see Supplement 1) and reflection spectra at a curved surface, a series of narrow peaks appears in the spectrum. The corresponding field patterns reveal that these peaks relate to the resonances in the silicon microdisk [see an example in the inset of Fig. 1(c) and details in Supplement 1]. From the transmission spectrum, the coefficient of end-fire injection [the mode marked as a red circle in Fig. 1(c)] is determined as 52%. In the case of the reflection spectrum, the light experiences an injection process and a rejection process. Thus, the reflection spectrum can be considered as dropped signals, and the reflection coefficient is around 27%, which is the square value of injection efficiency and consistent with the transmission spectrum well.

The high-efficiency end-fire injection is also found to be quite robust to the fabrication deviations. Figure 1(d) shows the dependence of input efficiency on the width of the waveguide. With the increase of the waveguide width from 450 nm to more than 700 nm, the input and output efficiencies are well preserved with small variations. Similarly, the slight dependence of coupling efficiencies on the waveguide tilting angle has also been numerically confirmed. Therefore, both of the theoretical analyses and numerical calculations confirm that the light can be effectively injected into the silicon microdisk by simply connecting a waveguide to it. In contrast to the complex waveguide–cavity evanescent coupling, here the coupling efficiency is quite robust to the fabrication deviations, and thus the new configuration with $d>500\mathrm{nm}$ is achievable with standard photolithography [23]. Note that the injection efficiency is slightly lower than the unidirectional output. This is because the mode in the waveguide here is fixed at the fundamental mode, which is much easier to be realized in experiment (see below) but is different from the rejection waves in the channeling waveguide [see inset in Fig. 1(b)]. This kind of difference varies with resonant wavelength and waveguide width.

B. Experimental Realization of End-Fire Injection

Based on the above analysis and numerical calculation, we have fabricated silicon microdisks to experimentally verify the proposed end-fire injection. The silicon WCM was fabricated onto a SOI wafer with a 220 nm top silicon film and a 3 μm ${\mathrm{SiO}}_2$ insulating layer. Figure 2(a) shows the top-view scanning electron microscope (SEM) image of the fabricated WCM. It is a circular microdisk that is connected to one end of a straight waveguide. The other end of the waveguide is a Y-splitter, whose two arms are patterned with grating couplers. Figure 2(b) shows the enlarged top-view SEM image of WCM. Similar to the design in Fig. 1, here the radius of the microdisk is 5 μm, and the width of the waveguide is 550 nm, respectively. The structural information of the Y-splitter and grating couplers are shown in Supplement 1.

 

Fig. 2. Experimental results of WCM. (a) and (b) are the low- and high-resolution top-view SEM images of the WCM. A Y-splitter has been employed to record the reflection spectrum, and the radius of the microdisk is 5 μm. (c) Experimentally measured reflection spectrum. (d) Magnified view of the resonant mode of WCM with 15 μm in radius, giving a $Q$ factor around $2\times {10}^5$. The inset in (d) is the top-view SEM image of the large WCM.

Download Full Size | PDF

To test the coupling efficiency, a TE-polarized tunable laser (Yenista T100S-HP/SCL) is coupled into the waveguide via a grating coupler on one arm of the Y-splitter. The coupling in and coupling out losses are both around 8 dB (see the optical setup and detailed experimental results in Supplement 1), and the propagating loss is negligibly small. The input light can efficiently pass the junction, whereas the reflected waves in the channeling waveguide were divided by the Y-splitter (see Supplement 1), and half of the reflected waves are detected from the other grating coupler. By taking account of the coupling efficiencies of two reference couplers and the splitting ratio of the Y-junction, we normalized the reflection spectrum from the silicon microdisk, as shown in Fig. 2(c). Similar to the numerical calculations, a large number of resonances can be observed around the telecommunication wavelength. The highest reflection intensity is around 26.5%. Considering the coupling in and out process, the total coefficient of a single-pass injection or rejection is thus around 51.5%. Both values are well consistent with the numerical calculations.

Besides the coupling efficiencies, we have also compared the numerically calculated and the experimentally recorded wavelengths (see Supplement 1). Interestingly, while slight deviations that are caused by the different material dispersion can be seen, the experimental peaks are almost a one-to-one correspondence with the numerical calculations. Consequently, the experimentally recorded peaks can be identified. The modes 1–10 are the quasi-WGMs. Basically, these modes are formed by the coupling between two initial WGMs, and their field distributions around the waveguide–cavity joint position are destructively reduced [see the inset in Fig. 1(c)] [24–26]. Then the quasi-WGMs are less affected by the broken TIR and have relatively high $Q$ factors. We note that the $Q$ factors can be further improved in the large cavity, where the influence of the channeling waveguide is further minimized. One example is shown in Fig. 2(d). When the radius of WCM is increased to 15 μm and all the other parameters are kept, the $Q$ factors of WG-like modes are increased to around $2\times {10}^5$, which is comparable to the values of conventional micro-resonator-based biosensors [17,18]. The end-fire injection coefficient of such a high-$Q$ mode is around 40%. Thus, we can confirm that the light on resonance can be efficiently injected into high-$Q$ resonances of a WCM on SOI substrate.

Similar to the numerical calculations, the dependence of end-fire injection on fabrication deviations has also been checked by fabricating WCMs with different channeling waveguides and characterizing their reflection spectra. In this experiment, the size and shape parameters of the circular microdisk are the same as Fig. 2(c). All of the experimental results are summarized in Fig. 3. The detailed reflection spectra and the corresponding SEM images can be found in Supplement 1. For a resonance at 1550 nm, the coefficient of end-fire injection are kept above 40% when the width of the channeling waveguide is increased from 450 nm to 700 nm [Fig. 3(a)]. Similarly, the dependence of the reflection peak on the tilt angle of the channeling waveguide has also been experimentally explored. As shown in Fig. 3(b), the reflection coefficients also remained at very large values and show very tiny variations with the increase of tilt angle from 0° to 5.74° [see inset in Fig. 3(b)]. All of these experimental results, associated with the numerical results in Fig. 1, have clearly demonstrated the simplicity, high-efficiency, and robustness of the end-fire injection.

 

Fig. 3. Robustness of end-fire injection. (a) and (b) show the dependence of coupling efficiency as a function of waveguide width $w$ and tilt angle $\theta$ of the channeling waveguide.

Download Full Size | PDF

The numerically calculated results and the experimentally recorded spectra both include a series of dips within some of the reflection peaks. Interestingly, the linewidths of these dips are much narrower and thus can give higher $Q$ factors. Different from the regular reflection peak, the dips cannot be explained with a single quasi-WGM. To understand the formation of these dips, we have numerically studied the resonant modes by introducing a local refractive index perturbation $\mathrm{\Delta }n$ (see Supplement 1). When $\mathrm{\Delta }n$ increases from $-0.1$ to 0.1, we can see that a sharp peak and a broad peak cross one another in wavelength. When two peaks overlap, the narrow dip is formed and the $Q$ factor is enhanced (see Supplement 1). Thus, it is easy to know that the dip is formed by the coupling of two quasi-WG modes, similar to the optical analogue of electromagnetically induced transparency (EIT) [27–30]. The dip in the reflection spectrum can be understood with the destructive interference of reflected waves of two quasi-WGMs. As the EIT dips are formed by the interference between quasi-WGMs, their linewidths are also dependent on the cavity size. One example is shown in Fig. 4. When the radius increased to 15 μm, the linewidth of the EIT dip improved further to about 2 pm, giving $Q$ factors of more than $7\times {10}^5$ and coupling efficiency around 57%. This value is even comparable to the record $Q$ factors in Si-based circular microdisks [14–16] or ${\mathrm{SiO}}_2$ microspheres [3] with similar sizes. In the case of evanescent coupling, by precisely controlling the fiber diameter and the fiber–cavity gap distance, the nearly critical coupling has also been realized. However, a clear advantage of WCM is the ability to insert light into the microcavity without the need for careful alignment of a tapered fiber. Meanwhile, considering the high $Q$ factor and the large nonlinearity of silicon, the designed WCM requires only a few milliwatts of injected laser to generate the frequency comb [31].

 

Fig. 4. EIT in WCM. (a) Top-view SEM image of the silicon WCM. (b) High-resolution reflection spectrum of one resonant dip. A $Q$ factor around $7\times {10}^5$ can be calculated.

Download Full Size | PDF

C. End-Fire-Injection-Based Optical Sensors

Based on their high $Q$ factors and high coupling efficiencies, the potential of WCMs in optical sensors has been experimentally examined. Taking a WCM with $r=15\mathrm{\mu m}$ and $w=550\mathrm{nm}$ as an example, we have studied its applications in temperature sensing. The sample was placed onto an electric heater to control the substrate temperature. As shown in Fig. 5(a), an EIT dip with $Q$ factor around $4.9\times {10}^5$ has been obtained at 1549.621 nm. With the increase of ambient temperature, we can see that the resonant dip gradually shifts to a longer wavelength due to the increase of refractive index of silicon. Figure 5(b) shows the wavelength of the resonant dip as a function of the ambient temperature. The fitting result shows that the sensitivity of the WCM sensor is around 487 nm/RIU and the corresponding figure of merit (FOM) is as large as ${10}^5$. Both values are also comparable to or even better than the results of a conventional silicon microdisk on SOI substrate.

 

Fig. 5. WCM-based optical sensors. (a) Resonant spectrum as a function of ambient temperature. (b) Dependence of resonant wavelength on the ambient temperature. (c) Resonant spectrum of a WCM without (solid line) and with (dashed line) a nanoparticle. Inset shows the high-resolution SEM image of the nanoparticle on the microdisk. (d) Shift of resonant wavelength as a function of nanoparticle number. Here, the radius of the microdisk and width of the waveguide are kept at $r=15\mathrm{\mu m}$ and $w=550\mathrm{nm}$, respectively. Inset is the SEM image of the WCM with attached nanoparticles.

Download Full Size | PDF

In addition to the detection of temperature, a high-$Q$ microcavity has also been frequently used in label-free detection, which has been widely accepted as important for practical applications, such as environmental monitoring, homeland security, and sequence-specific DNA detection. Here, we demonstrate the capability of WCM in a single nanoparticle detector. As the solid line shows in Fig. 5(c), a resonant dip can also be observed at $\sim 1549.62\mathrm{nm}$. According to its narrow linewidth, the calculated $Q$ factor is about $4\times {10}^5$. Once a nanoparticle is transferred by a tapered fiber (see Supplement 1) onto the top surface of the microdisk [see inset in Fig. 5(c)], the optical path length of the quasi-WG mode is increased. Consequently, the resonant wavelength of the dip shifts to a longer wavelength. One example is illustrated as the dashed line in Fig. 5(c). After the attachment of a nanoparticle with radius $\sim 30\mathrm{nm}$, the resonant wavelength shifted about 30 pm, which is large enough to be resolved in the reflection spectrum. Note that the typical wavelength shift can be disturbed by the temperature changes. Here, the influences of environmental temperature can be minimized by using a parallel reference microdisk on the same integrated wafer. Thus, the detectivity of the on-chip WCM sensor can be much better than 30 nm, which is close to the detection limit by a mode shift mechanism [10]. Interestingly, as the sizes of exosome are around 40 nm to 100 nm, this new coupling configuration will have important impacts on the early detection of cancers.

Moreover, the high-$Q$ WCMs can also be used to detect multiple nanoparticles. For convenience, we have increased the nanoparticle to $r=250\mathrm{nm}$ and studied the corresponding reflection spectra. As depicted in Fig. 5(d), the wavelength of the resonant mode in one WCM shifts from 1540 nm to 1540.2 nm, 1540.3 nm, and 1540.35 nm. The corresponding SEM images show that the wavelength shift closely relates to the attachments of nanoparticles, from 0 to 3, clearly demonstrating the potential of silicon WCM in the detection of multi-nanoparticles. We note that the wavelength shift in Fig. 5(d) is not perfectly linear, because the nanoparticles are attached to different radial locations on the microdisk and thus affect the resonant modes differently. This linearity of wavelength shift can be improved by replacing the microdisk with a microring [17,18].

3. CONCLUSIONS

In summary, we have theoretically proposed and experimentally demonstrated a new and counterintuitive configuration to simply couple light into high-$Q$ silicon microdisks. When a waveguide is directly connected to a circular microdisk, the coupling efficiency can be as high as 57% and the recorded $Q$ factors are around $2-7\times {10}^5$. Both the coupling efficiency and the $Q$ factors are comparable to or even better than those of conventional silicon microdisks. As all of the size parameters are possibly larger than 500 nm, the new configuration can be fabricated without a high-cost nanofabrication technique. For example, high-$Q$ silicon WCMs have been successfully fabricated with a simple shadow mask method (see Supplement 1) [32]. Overall, the designed WCM scheme can have much better fabrication tolerance, lower fabrication cost, more robust coupling efficiency, and even comparable $Q$ factors. Considering the capability of WCMs in detecting nanoparticles and tiny temperature changes, this new configuration will be extremely important for the practical applications of silicon microdisks and microrings, especially for the low-cost, portable, and highly sensitive POC devices.