1. Introduction

Future lab-on-a-chip (LOC) refractive index (RI) sensors would need to offer ultra-high sensitivity, multi-channel operation and CMOS compatible manufacturability for high volume and low-cost production when targeting disposable tests. Silicon photonics and plasmonics have attracted immense research efforts towards powerful RI sensors partially addressing those requirements. Photonic integrated RI sensors typically employ low-loss slot and strip-based waveguides as sensing transducers to quantize ambient RI variations in MZI configurations [1–3], microring resonators [4–6], photonic crystal cavities (PhC) [7,8] and subwavelength waveguides [9,10], showcasing moderate bulk-sensitivities up to 1538 nm per refractive-index-unit (RIU) [8]. On the contrary, plasmonic counterparts based on the principle of SPP-resonance modes, demonstrate superior sensitivities [11,12], yet at the expense of increased optical loss and limited miniaturization capabilities due to their bulky excitation scheme. To address those challenges while meeting all requirements in a single CMOS device, selective co-integration of plasmonics with photonics is envisioned in order to synergize the benefits of both technologies [13–18], considering at the same time the deployment of plasmonics via CMOS materials and CMOS compatible fabrication processes [15,16,18]. However, the majority of experimental demonstrations of planar plasmo-photonic sensors so far, have been limited to the exploitation of noble metal-based SPP waveguides, having demonstrated rather moderate bulk sensitivities of up to 2000nm/RIU [19], hindering their volume production due to their CMOS incompatible material platform. In this work, we experimentally demonstrate a record-high bulk sensitivity of 4764 nm/RIU using, for the first time, an aluminum stripe waveguide, integrated in the sensing arm of a Si₃N₄ MZI, offering in this way a plasmo-photonic interferometric sensor based on a CMOS compatible material platform, achieving the highest sensitivity among all the planarly integrated plasmo-photonic sensors reported so far [19–26]. Additionally, our experimentally validated model shows that the MZI design parameters can be properly engineered to scale the bulk sensitivity values of the sensor.

2. Principle of operation and design analysis

A top-view of the proposed refractometric sensor is illustrated in Fig. 1(a). The Si₃N₄-based MZI sensor is composed of a reference (upper) arm and a sensing (lower) arm. In the sensing arm, a 70 µm long, open cladded aluminum-based metal stripe is butt-coupled with Si₃N₄ waveguides, serving as the sensing transducer. Aluminum is inherently protected by a thin layer (3 nm) of alumina (Al₂O₃) which is formed rapidly in ambient conditions on the metal surface, providing corrosion resistance [27,28]. In addition, simulation results in [16] have dictated that the alumina layer has a slight impact, on the SPP waveguide performance. More specifically, the presence of the oxide layer can lead to a slight improvement of the interface loss equal to 0.7 dB per interface while the L_spp value, on the other hand, is slightly decreased about 3 µm. The performance of the aluminum waveguide transducer was experimentally evaluated in [29] as an individual structure, revealing an L_spp value of 65.8 µm in aqueous environment at 1550 nm, matching the simulated L_spp value of 64 µm. In addition, the Si₃N₄-to-Al interface loss in water was found equal to 2.8 dB at 1550 nm, being also in agreement with the 2.5 dB of interface loss derived from the 3D FDTD simulations. The reference arm comprises a thermo-optic phase shifter followed by an MZI-based Variable Optical Attenuator (VOA). Multi-Mode-Interference (MMI) couplers with a power transmission and a splitting ratio of 48% and 50%, respectively, were incorporated in the MZI while fiber-to-chip coupling relied on Si₃N₄ grating couplers optimized for Transverse-Magnetic (TM) polarization [30].

 

Fig. 1. (a) Top view representation of the plasmo-photonic MZI sensor. (b) Sideview impression of the Si₃N₄-to-Al interface. (c), (d) Waveguide cross section geometries on the sensing and reference arms, presenting also the respective numerically simulated electric field distributions in the bottom of the two figures. (e) Simulated transfer function of the VOA at 1550 nm with respect to the applied thermal power. (f) Assembled sensor chip on the evaluation board.

Download Full Size | PDF

The sensing mechanism is based on detecting any RI change, occurring in the liquid cladding of the sensing arm. This RI change will induce a phase change on the propagating SPP mode. The interferometer translates this phase change, on the sensing arm, into a spectral shift of the interferometer resonance. In order to maximize the spectral extinction ratio of our sensor, we deployed a VOA in the reference arm, as the means to counter-balance the losses of the SPP waveguide, resulting in equal amplitude field interference at the sensor’s output. A side-view of the sensing transducer is illustrated in Fig. 1(b). An 80 nm × 7 µm aluminum stripe is recessed in a cavity formed between two Si₃N₄ tapered waveguides, exploiting a butt-coupled interface. Vertical and longitudinal offsets equal to 400 nm and 500 nm were applied between the waveguide facets in order to optimize the Al-to-Si₃N₄ coupling efficiency. Figure 1(c) depicts the cross-sectional dimensions and the respective quasi-TM mode profiles for the Si₃N₄ strip (I), Si₃N₄ tapered (II) and Al SPP (III) waveguides. The structural parameters considering the butt-coupled interface and the respective waveguide geometries were derived from the optimization process described in our previous work [31].

For designing the thermo-optic phase shifter and the VOA we deployed two parallel Ti-based metallic wires on top of the oxide cladding with symmetric displacement to each other and away from the 360 nm × 800 nm Si₃N₄ waveguide core, as illustrated in Fig. 1(d). This heater configuration was employed to minimize the overlap of the photonic TM mode with the titanium wires, while maintaining a strong thermo-optic effect. A nested MZI with an equivalent 500 µm long phase shifter was deployed in the upper arm of the sensor as a variable attenuation stage. The VOA allows for dynamic control of the attenuation introduced in the reference arm, by adjusting the electrical power applied to the phase shifter of the nested MZI. In this way, power balancing between the two arms of the sensor MZI can be achieved. The simulated optical response of the VOA at 1550 nm, with respect to the applied thermal power on the VOA phase shifter, is illustrated in Fig. 1(e).

To investigate the performance of the proposed sensor and to evaluate its potential for improvement we conducted circuit-level simulations assessing the spectral response of the complete layout. The sensor circuit design was based on simulations in frequency domain performed using a commercial-grade circuit simulator for photonic integrated circuits [32]. Fundamental components were initially modelled individually taking into account dispersive material and waveguide properties. Frequency-dependent properties (effective index, group index, etc.) of the photonic and plasmonic waveguides were obtained by conducting 2D numerical simulations using the eigenmode solver [33]. Photonic tapers, MMI couplers and plasmo-photonic interfaces were modeled by their S-parameters which were derived by performing 3D broadband simulations using the FDTD-based simulator [34]. Heater operation was modeled by exploiting an optical phase modulator component which causes a phase shift depending on the input thermal power. Eventually, the aforementioned building blocks were inserted in the circuit simulator and connected properly in order to form the proposed MZI sensor circuit, as illustrated in Fig. 1(a). Additionally, the refractive index, used for the aluminum plasmonic waveguides in our simulations, was based on experimentally verified data reported in the literature [35]. Simulation results are given in comparison with the equivalent experimental results in Figs. 3(b) and 3(e) of section 4.

3. Fabrication

The fabrication of the biosensor chips (Fig. 2) started with a 6” silicon wafer on which a 2.2 µm thick SiO₂ layer was thermally grown. On top of that a 360 nm thick Si₃N₄ layer was deposited via low pressure chemical vapor deposition (LPCVD). In this layer the waveguide structures have been patterned using an i-line stepper tool for definition and a CHF₃ based reactive ion etching (RIE) process for transfer. On top of the waveguide layer, a 1 µm thick LPCVD SiO₂ cladding was deposited and annealed at 1000°C for several hours. The annealing process improves the optical properties of the oxide, which have a significant impact on the waveguide propagation losses. A cross-section SEM image of the fabricated photonic waveguides is shown in Fig. 2(b). The same procedure was used to fabricate the MMI couplers (Fig. 2(d)). The active elements, Ti heater lines with Al contact pads, were fabricated on top of the cladding layer. First, the 100 nm thick Ti heater lines were deposited via a sputtering process and patterned using i-line stepper lithography and a BCl₃ based RIE process. Then, the 200 nm thick Al structures were again deposited via sputtering and defined using the i-line stepper but transferred using a wet etch solution with a very high selectivity towards Ti to ensure that the heater lines are not damaged. As the Al structures have large lateral dimensions the anisotropy of the wet etch step was negligible. SEM images of the heater structures are depicted in Figs. 2(e) and 2(f). After the fabrication of the metal structures the plasmonic cavity was defined via i-line stepper and etched 1.36 µm deep using the CHF₃ based RIE process. The goal was to etch through the LPCVD SiO₂ cladding, the Si₃N₄ waveguide layer and 400 nm deep into the thermal SiO₂. The final and most critical step was the fabrication of the plasmonic Al stripes inside the cavity. For this only a lift-off process was suitable as it prevents sidewall coverage of the cavity, which would disturb the photonic-plasmonic coupling. In the prototyping phase, electron beam lithography was used for this step. A double layer resist stack of polymethyl methacrylate (PMMA) was exposed using an EBPG 5200 tool with an acceleration voltage of 100 kV. In the subsequent development step, the lower PMMA layer was dissolved faster due to its lower molecular mass thereby creating a significant undercut. After development, 80 nm of Al were deposited via electron gun evaporation with a rate of 0.3 nm/s. The lift-off was performed in heated dimethyl sulfoxide (DMSO) which prevents the re-deposition of already lifted metal flakes. After the lift-off the wafers were rinsed in isopropyl alcohol and dried with nitrogen. In order to make the process flow suitable for higher volume fabrication a comparable lift-off process for i-line stepper lithography was developed. Here a double layer resist stack consisting of a negative tone image reversal photoresist and an additional lift-off resist (LOR) was used. The LOR is strongly dissolved by the developer for the photoresist and therefore creates an additional undercut that improves the lift-off. With the stepper lift-off process very similar plasmonic Al stripes were fabricated but within a much shorter time frame.

 

Fig. 2. (a) Top-view image of the fabricated sensor chip, hosting multiple sensors, with the micro-fluidic gasket on top. (b) Cross section SEM image of the Si3N4 photonic waveguide. (c) Top-view SEM image of the Al stripe lying within a cavity formed between the Si₃N₄ tapered waveguides. (d) Top-view SEM image of the fabricated MMI coupler. (e), (f) Top-view SEM images of the Ti-based heaters lying on top and symmetrically of the photonic waveguide.

Download Full Size | PDF

4. Experimental results

For the characterization of the proposed sensor device, we performed broadband Fiber-to-Fiber (FtF) optical measurements by sweeping the wavelength of a Tunable Laser Source (TLS) from 1500 to 1580 nm, while a Polarization Maintaining (PM) fiber was used at the input of the chip to ensure TM polarized light injection into the chip. An output Single Mode (SM) fiber was placed on the sensor output and fed to an optical power meter. An electrical Printed Circuit Board, mounted on the metal base, was employed to connect all the titanium heaters to an external voltage source via aluminum-based electrical pads. A silicone-based microfluidic gasket was attached on the chip’s surface (Fig. 2(a)) as the means to transfer the fluids on top of the sensing areas, exploiting open chambers directly located over the sensing arm of each sensor. A PMMA cap, mounted on the metal base, was employed to seal the microfluidic gasket on the chip and to connect each of the microfluidic chambers to an external tubing. Figure 1(f) shows the test board where the sensor chip was assembled including the microfluidic cartridge and the spring-loaded PCB for electrical control. The optical fiber interfaces are located between the PCB and the microfluidic cartridge on the sensor chip. A two-step process was followed for the characterization of the proposed sensor device. In a first step, we calibrated the sensor by sweeping the applied voltage on the VOA to maximize the spectral extinction ratio, hence improving the sensing resolution of the sensor. In parallel, the resonance wavelength of the MZI sensor was tuned in our laser spectral window (1500 - 1580 nm), utilizing the integrated phase shifter. Following the successful sensor calibration, the sensitivity of the proposed device with respect to bulk RI changes was quantified. For this reason, we infiltrated aqueous solutions of 10% and 100% Phosphate-Buffered Saline (PBS) as well as a NaCl solution over the aluminum waveguide surface, using the microfluidic channels. Buffer solutions with pH levels around 7 were used in order to avoid pH changes that may harm the surface of the aluminum transducer [36–38], also mimicking the pH of blood (pH= ∼7.4) making the proposed sensor suitable for future biosensing applications. The different sample fluids were delivered to the sensing area of each sensor using a syringe pump (neMESYS 290N) via Polytetrafluoroethylene tubes while software controlled fluidic valves were utilized to easily select which fluid is injected in the sensor.

Between measurements the sensing area was rinsed with DI water as the means to remove any residues. The refractive indices of the fluids were defined using a commercially available refractometer (Mettler Toledo 30PX), resulting in RI values equal to 1.3324 for DI water, 1.3328 for 10% PBS solution, 1.3343 for 150 mM NaCl solution and 1.3389 for 100% PBS solution. In our initial experiments, two different sensor prototypes, exploiting EBL-fabricated aluminum SPP waveguides, have been tested with Free-Spectral-Range (FSR) values of 43 nm (sensor₁) and 50 nm (sensor₂) and E.R. values of 25 dB (sensor₁) and 38 dB (sensor₂), respectively. Figures 3(a) and 3(d), depict the measured sensor spectra when each of the analytes flows over the surface of the sensing arm of each sensor, revealing a clear red-shift of the resonance dips with increasing RI values. Figures 3(b) and 3(e), present the corresponding simulation curves. Bulk RI sensitivity was determined by performing least squares linear fit on the wavelength shifts with the RI change. Using this method, we calculated the bulk RI sensitivity for both sensors, demonstrating sensitivities up to 4764 nm/RIU, revealing strong agreement with the corresponding simulation results. Subsequently, plasmo-photonic sensors, fabricated solely with i-line stepper lithography, were also evaluated, following the same procedure, demonstrating sensitivities up to 3000 nm/RIU, validating the successful transfer from rapid prototyping to wafer scale fabrication of CMOS compatible plasmo-photonic sensors within a shorter time frame. As can be observed in Figs. 3(c) and 3(f), sensor₂ (FSR = 50 nm), exhibits a higher sensitivity than sensor₁ (FSR = 43 nm), indicating a sensitivity scaling with the FSR increment although both devices involved an identical 70 µm length aluminum-based plasmonic transducer. We performed further investigation on this sensitivity scaling, utilizing our experimentally validated simulation model and concluded that the proposed sensor scheme can, in principle, achieve extraordinary bulk sensitivity values by going through some redesigns of the sensor configuration involving the appropriate adjustment of the optical paths that lead to larger FSR values, following the methodology and theoretical background described in [39].

 

Fig. 3. (a), (d) Measured sensors’ spectra for various RIs. (b), (e) Simulated transmission wavelength response of the sensor under test. (c), (f) Resonance shift with respect to the refractive index change of the tested fluids and least-squares linear fits for both experimental and simulation data.

Download Full Size | PDF

Figure 4 illustrates the bulk sensitivity of the proposed sensor with respect to the sensor FSR value. Circuit-level simulations dictated bulk sensitivities equal to 3880, 4924, 6756, 14930, 34960, 63720 nm/RIU for FSR values of 40, 50, 125, 215, 326, 540 nm, respectively, verifying the sensitivity dependence on the FSR of the interferometer. It should be noted that the plasmonic waveguide length remained unchanged and equal to 70 µm in all simulated devices with different FSR values, suggesting that the sensitivity of the layout increases with FSR without increasing the absolute value of the accumulated phase shift at the plasmonic transducer. The higher sensitivity values originate from the higher fraction of the plasmonic phase shift within the total differential phase shift that ultimately defines the interferometer’s FSR, implying that even a rather short plasmonic waveguide of 70 µm can allow for ultra-high sensitivity values without necessitating additional plasmonic transducer sections and as such higher insertion losses.

 

Fig. 4. Sensitivity as a function of the sensor Free Spectral Range (FSR).

Download Full Size | PDF

5. Discussion and conclusion

To give a perspective on the features and benefits of the proposed device, a detailed comparison between all plasmo-photonic integrated RI sensors employing silicon photonics and plasmonic sensing transducers in various configurations that can be found in the literature is presented in Table 1. To highlight our record and provide an intuitive way to represent the trade-off between compactness and induced sensitivity, we introduce a new Figure-of-Merit (FoM) defined as the sensitivity [S] (nm/RIU) over the length of the sensing transducer (µm). The proposed sensor exhibits the highest sensitivity and the highest FoM among all the plasmonic-photonic RI sensors, yet exploiting a CMOS-compatible material platform based on aluminum plasmonic transducers and Si₃N₄ photonic waveguides.

Table 1. Experimental results for state-of-the-art plasmo-photonic integrated RI sensors employing silicon photonics and plasmonic sensing transducers.^a

View Table