1. Introduction

The unique combination of versatility and high specificity achievable with optical spectroscopy is the driving force motivating the development of miniaturized spectrometers. Spectroscopy is instrumental in many scientific disciplines, including physics, chemical and biological sensing, medical research as well as telecommunication [1,2]. Photonic integrated circuits (PICs) have enabled many innovative on-chip spectrometers. These offer low cost, portability, robustness and sensitivities that are comparable to their bulky and expensive free-space counterparts [3]. Moreover, they can be implemented using batch fabrication and are suitable for in-situ measurements. However, most of them targets one application or are sensitive only to a narrow portion of the optical spectrum.

Thus, in a significant number of cases, to obtain a complete picture of the chemical composition of a solution under study, multiple optical spectra must be acquired with different spectrometers [4–8], since the signatures that uniquely identifies active molecules falls in different bands [9,10]. For instance, in biological applications, the therapeutic window covers visible and near-IR wavelengths (750-930 nm) [11–14]. Many molecules can be detected in this window, such as oxyhemoglobin (HbO₂), deoxyhemoglobin (Hb), and oxidized cytochrome c oxidase [15]. Nevertheless, it is also beneficial to detect other components while having absorption spectra at other wavelength bands. For example, glucose, which plays a vital role in numerous biological processes, has an absorption peak at wavelengths in the C-band [16]. In pharmaceutical applications, isopropyl alcohol can be detected with a signal at 785 nm, whereas ethanol requires measurements at 1450 nm [11]. In point-of-care testing, it could be useful to measure Na-to-creatinine (NCR) in urine at 1400 nm and tuberculosis antigen in the very near IR [17,18]. Covering such widely different wavelength bands with a single device is challenging with current integrated spectrometers.

Nowadays, integrated spectrometer architectures involve either static or tunable filters and/or reconfigurable components. In static spectroscopic systems, three major architectures are used. The first configuration relies on cascaded integrated optical filters with high Q-factors in conjunction with an array of output waveguides coupled to on-chip or off-chip detectors. These include Bragg grating filters [14,19], series of Fabry-Perot resonators [20], cascaded Mach-Zehnder interferometers [4] and ring resonators (RR) arrays [21,22]. These systems offer high resolution and high sensitivity due to their sharp resonant peaks. However, these techniques can support broadband operation at the expense of employing a combination of different building blocks or a large number of filters, which increases the footprint, requires complex readout systems, and degrades the signal-to-noise ratio (SNR). For instance, a spectrometer based on an array of 84 RRs provides a fine resolution of 0.6 nm but covers an operating bandwidth of only 50 nm [23]. Moreover, RRs are prone to fabrication variations. Thermal tuning can be used to compensate wavelength shifts, but it may lead to thermo-optic non-linearity, waveguide thermal expansion and dispersions [24].

On the other hand, on-chip multi-channel filters based on a single diffractive component, such as photonic crystal super prisms [25], echelle gratings (EG) or arrayed waveguide gratings (AWG), have spurred significant interest in spectroscopic sensing systems [18,26]. The group in [27] developed a spectrometer that covers the wavelength range from 1521.40 nm to 1611.35 nm. The microspectrometer designed with an elliptical Bragg mirror developed by Pottier et al. covers a band of 30 nm around 1550 nm [28]. Two other spectrometers designed to operate in the C-band with a wide free spectral range (FSR) of 115 nm from 1500 nm to 1600 nm were reported. The first is a 1 ${\times}$ 30 dense spectrometer based on an echelle grating with facets made of distributed Bragg reflectors and a channel spacing of 3.2 nm. The second is a 1 ${\times}$ 100 two stigmatic grating demultiplexer with a Δ$\lambda $ of 1 nm [29]. Dispersive spectrometers can provide high resolution and an array of output channels, along with acceptable channel crosstalk (i.e., < -20 dB) [8,9,26,30–32]. However, in most instances they can cover at most one or two neighboring bands. In the case of [8], the design works only in the visible and the NIR. The group in [9] had two different devices based on the same configuration, the first operates in the visible wavelength range and the second in the NIR.

An alternative method to static devices is to use high-speed tunable optical filters. These filters are compact, eliminate the need for an array of detectors, and some enable real time signal acquisition of the spectrum. Different designs were demonstrated, such as tunable diffraction gratings [33,34], Fabry-Perot interferometers with a tuning range of 60 nm at a center wavelength of 1517 nm [35], fiber Bragg gratings with a tuning range of 45 nm [36] or over 90 nm from 1544 to 1634 nm [37], and a thin-film tunable filter consisting of an assembly of individual tunable bandpass filters that is capable to scan a small bandwidth range from 50 nm to 90 nm. The tuning is obtained by rotating the angle of incidence from 0 to 60 degrees to select the required wavelength. The tuning is done in the visible range and spans four different wavelengths range from 398.4 to 710 nm [38].

Fourier transform spectrometers (FTS) are particularly appealing due to their broad operational wavelength range and the high spectral resolution they provide compared to other approaches [2,39,40]. The group in [18] demonstrated a co-propagative stationary Fourier-transform (FT) spectrometers with a wide operational range of ∼100 nm at a wavelength of 850 nm that can provide a resolution of 6 nm. An on-chip spatial heterodyne Fourier-transform spectrometer (SHSFT) with a multi-aperture input was shown to achieve a resolution as low as 49 pm and a bandwidth of 340 nm in the NIR [41]. Other implementations, such as active scanning FTS including on-chip silicon micro-electro-mechanical systems (MEMS) [42] or using thermal tuning [40], can provide sharp resolution in the pm range. However, this comes at the price of an increase in device size or acquiring a broad operational bandwidth with moderate resolution offered. Very high resolutions can be achieved with arrays of Mach-Zehnder interferometers (MZIs). However, these spectrometers have very narrow bandwidths. A resolution of 40 pm was obtained with an array of 32 MZIs but the device had a bandwidth of only 0.75 nm [5]. A spatial heterodyne spectrometer (SHS) necessitates up to 200 MZIs to acquire high resolution spectra with a resolution of 25 pm and a range of 2.5 nm [4]. On the other hand, stationary wave-integrated Fourier transform spectrometers (SWIFTS) and Si-FTS have a smaller footprint, and a high resolving power of R = 40 in the visible range in the case of SWIFTS, but only over a small wavelength range (i.e., < 100 nm) for a given detector pitch [31,40,43].

All these integrated sensors are designed to cover at most two wavelength bands, which is significantly less than their free-space counterpart. Therefore, in order to cover many bands, multiple spectrometers are needed on a chip, which can occupy a large area. To solve this issue and demonstrate that multiple wide spectra can be acquired with a single integrated diffractive device, we propose a sensing spectrometer design capable of operating over four different wavelength bands that are each at least 95 nm wide and that provides a 3-dB resolution R = Δλ of 1.51 nm or less. This integrated optical four bands spectrometer (IOFBS) consists of an echelle grating with 39 outputs and 4 inputs built on a silicon nitride waveguide platform compatible with MEMS. Thus, MEMS actuators could be used to create a switch to select which input of the echelle grating is activated. Note however that the bandwidth of the light source used for sensing must match or be less than the one of the selected input. In other words, the four bands cannot be used simultaneously. The proposed design results in a compact yet versatile integrated spectrometer that could find applications in a wide range of areas.

The article is structured as follows: first, we discuss the advantages of using silicon nitride waveguides in integrated spectrometers. Second, we describe the operating principle of the proposed spectrometer. Then, we explain the approach used to design the grating, input/output waveguides and tapers to operate at 850 nm, 1310 nm, 1450 nm and 1570 nm. Next, we present simulation results with a detailed analysis of the different performance metrics demonstrating the potential of the IOFBS. Finally, we discuss potential applications for the proposed spectrometer.

2. Silicon photonics technology: silicon nitride platform for on-chip sensing spectrometers

PICs based on silicon-on-insulator (SOI) are a mature technology that has been widely adopted. This material platform can be compatible with CMOS fabrication processes and offers many advantages including mass production, relative affordability at high volumes, and small footprints [44,45]. The high refractive contrast of this platform allows for tight bends and enables high-density integration of multiple and novel functionalities [46,47]. Despite all these advantages, this material platform has some limitations. These include: a transparency window limited to a minimum wavelength of 1.1 µm, high scattering losses due to waveguide roughness, and a high sensitivity to fabrication variations.

Another appealing material platform that is gaining wide adoption is silicon nitride (SiN). Like SOI, silicon nitride can also be compatible with CMOS fabrication processes, but it is transparent at shorter wavelengths (down to less than 0.5 µm), and offers a moderate effective refractive contrast of ∼0.5 with SiO₂ [31,48]. Thus, it can be employed for emerging system-on-chip, data communication, and sensing for therapeutic and spectroscopic applications operating at shorter wavelengths [18,26]. In contrast to SOI, the moderate index contrast offers a relaxed sensitivity to fabrication variations, while maintaining a relatively compact footprint. Another attractive feature is that SiN waveguides provide scattering losses an order of magnitude lower than SOI. Furthermore, silicon nitride can be deposited by plasma-enhanced vapor deposition (PECVD) or low-pressure vapor deposition (LPCVD), which offers more flexibility for fabrication and allows to adjust the refractive index in the range of 1.9 to 2.15 [8,31,48]. All these characteristics make this material platform an excellent choice for on-chip spectrometers. The relatively moderate refractive index and higher tolerance to fabrication variations allows for lower phase noise and losses induced by scattering resulting from waveguide roughness. This also help reduces the insertion loss, which makes the device more power efficient. Also, they provide more accurate wavelength discrimination in demultiplexers, resulting in lower channel crosstalk and increased signal-to-noise (SNR) ratio [31]. For sensing spectrometers, crosstalk better than -30 dB is needed and thus silicon nitride is a promising platform. As stated earlier, SiN waveguides exhibit lower propagation losses when compared to SOI. This allows to increase the interaction length and enhances the sensitivity of the spectrometer. Finally, the refractive index of silicon nitride is 5 times more thermally stable than that of silicon [31]. Therefore, the silicon nitride material platform is a viable solution to implement the multiband spectrometer we are proposing.

3. Spectrometer design

3.1 Multiband spectrometer operating principle

The spectrometer is designed to operate over four bands, namely the near infrared, O-band, E-band, and L-band. However, to avoid crosstalk, the bandwidth of the light source must be within the wavelength band covered by input selected with the MEMS switch. Therefore, only one band can be used at time. The proposed design can measure spectra with a resolution as low as 1.7 nm. It is based on a SiN planar concave grating that can be integrated with a translational MEMS actuator with a bi-axial motion, which is used to switch the input beam to the intended input waveguides (IWGs) depending on the operational wavelength (i.e., operating as a MEMS optical switch). More information about the MEMS optical switch can be found in [49,50]. By taking advantage of the design flexibility offered by SiN, it is possible to create waveguides with a square core of 435 nm ${\times} $ 435 nm that are single mode over the four wavelength bands. The proposed spectrometer design is depicted in Fig. 1 along with an enlarged cross-sectional view of the grating facets coated with a thin layer of parylene. In order to couple light in and out of the chip, tapered waveguides are preferred, since no grating couplers design can be tailored to work efficiently with these 4 bands simultaneously. The input tapered edge coupler is connected to a 1 cm long waveguide spiral that serves as the interacting window with the analyte under test. However, this section could be removed if the interaction is taking place outside of the chip. The other end of the spiral waveguide is connected to a 1 × 4 switch, such as the MEMS switch presented in [50], to select one of the input waveguide (IWG) of the spectrometer. These strip waveguides are routed on the chip and tapered to rib waveguides using an adiabatic taper. Each input waveguide angular position is optimized for a specific central operational wavelength, namely 850 nm, 1310 nm, 1450 nm, and 1570 nm for IWG1, IWG2, IWG3, and IWG4, respectively. Light guided in the input waveguides enters a slab region where it diverges, gets reflected and spatially separated at the same time by the grating, then focused into the 39 channels of the output waveguide array. The chirped and blazed planar concave grating is designed based on a classical Rowland mounting [51,52]. The angle of the blazing and chirping of the grating facets is tailored for the 4 input angular positions, as described below. All entrance and output apertures connected to the free propagation region (FPR) of the grating are shallowly etched (i.e., 0.218 µm deep) and are 1.3 µm wide. This is done to reduce the insertion losses and minimize reflections. As for photodetector array at the output, we envision two possible solutions. For applications where low cost is a priority over sensitivity, germanium photodiodes could be used whereas for the ones with more stringent sensitivity requirements, high performance InP photodiodes could be coupled to the spectrometer. As for the integration of the photodiodes array with the spectrometer, we can take advantage of the predefined cavities built in the silicon-on-insulator wafer used to fabricate the MEMS switch to create openings where the array can be mounted and butt coupled to the output waveguides [50]. By switching between the input waveguides, we are combining 4 devices into one. Thus, reducing the chip-size and prices significantly, output especially since only one array of output detectors and one sensing spiral is required.

 

Fig. 1. Schematic of the multiband spectrometer with a MEMS optical switch, and a sensing spiral. Top left: enlarged view of the grating facet with a metal reflector and a quarter wavelength thick layer of parylene. Bottom right: enlarged top view of the taper edge coupler.

Download Full Size | PDF

3.2 Planar concave grating design

The core of the IOFBS is a planar concave grating also known as an echelle grating. A detailed description of the operating principles of echelle gratings is provided in [52]. This design follows an Eagle mount configuration to maximize diffraction efficiency and minimize aberrations for each input. The SiN layer thickness is 435 nm and is cladded at the top and bottom by 4 µm of SiO₂. This nitride thickness provides a moderately confined slab mode. The simulated TE mode has a vertical size of 0.3 and 0.51 µm and an effective index (n_eff) of 1.96 and 1.826, at a wavelength of 850 nm and 1.57 µm, respectively. The input signal is guided to the slab waveguide, called the free propagation region (FPR), with a single mode rib waveguide where the light diverges, illuminates the grating facets, diffracts, and then each wavelength converges to a specific output along the focal plane [46,51]. The 39 output waveguides are centered around the diffraction angle β equal to 50.1°. In the design we propose, there are four input waveguides (IWG), each with an angular position ${\theta _{in}}$ optimized for a different operational bandwidth centered at λ₁= 850 nm, λ₂= 1310 nm, λ₃= 1450 nm and λ₄= 1570 nm. These angles are listed in Table 1 along with the value of the other critical design parameters. The Rowland radius is 900 µm and the grating period (d) is 4.207 µm. The choice of d involves a trade-off between the capacity of the fabrication process, including its minimum feature size and minimizing the impact of corner rounding, and using a small diffraction order to maximize the operational range defined by the free spectral range (FSR). The grating teeth are aligned on a curved line that is 1350 µm in length, which results in 321 periods. The grating length is chosen to capture the entirety of the incident light from the different inputs. Moreover, the large number of periods increases the resolution of the spectrometer. The spectrometer diffraction order m and the corresponding linear dispersion for each band are presented in Table 1. The blazing angle and chirping of the grating facets are chosen as a compromise between the angular positions of the 4 input channels. The blazing angle was chosen to be ${\raise0.7ex\hbox{${({\theta _{in}} + \beta )}$} \!\mathord{\left/ {\vphantom {{({\theta_{in}} + \beta )} 4}} \right.}\!\lower0.7ex\hbox{$4$}}$, where ${\theta _{in}}$ is set to 49.5°. This is the average value of the angle spaned by all input positions.

Table 1. Summary of the parameters and characteristics of the grating optimized for four operational wavelengths

View Table | View all tables in this article

To mitigate the Fresnel losses resulting from non-vertical facets and improve the reflectivity, a 0.45 µm thick layer of parylene with a refractive index of 1.6 covered by a 0.3 µm thick layer of aluminum can be deposited on the grating facets [45]. The thickness of the parylene layer is chosen as a compromise between the different bands and to avoid deposition variation effects, as shown in Fig. 2. The 0.45 µm coating thick offers more robustness to fabrication variations since there is no sharp drop in reflectivity in this region as it is the case below 0.1 µm. According to 3D-FDTD simulations conducted with software from ANSYS Lumerical (Vancouver, BC, Canada) [53], this increases the reflectivity from 92% to better than 97%.

 

Fig. 2. 3D-FDTD simulations showing the reflectivity of aluminum deposited on the grating facets for different parylene interlayer thickness for the four central wavelengths of each band. Two thicknesses are shown to enhance reflectivity, highlighting that 0.45 µm is more robust to fabrication variations.

Download Full Size | PDF

3.3 Input and output waveguide apertures design and optical crosstalk

To achieve a compact design with small bending radii and low bending and propagation losses, strip waveguides are used to route the light to and from the grating. As depicted in Fig. 3, the waveguide will be single mode for widths smaller than 0.45 µm at a wavelength of 800 nm (Fig. 3(a)) and smaller than 1.2 µm at 1620 nm (Fig. 3(b)). These wavelengths are at the extremes of the bands supported by the spectrometer. Thus, a square waveguide with a side of 0.435 µm ensures that the single mode condition is fulfilled at all wavelengths. The effective index (n_eff) variation over the wavelength range from 800 nm to 1650 nm is presented in Fig. 3(c). The n_eff of the square waveguide decreases from 1.83 at 800 nm to 1.49 at 1650 nm. The TE and TM modes are symmetrical and have an effective index equal to 1.83 and 1.52 for 800 nm and 1620 nm, respectively. The TE and TM modes for both 800 nm and 1620 nm are displayed in Fig. 3(d-g).

 

Fig. 3. Simulations of the routing waveguides modal behavior at the limits of the operational wavelength range. Single-mode condition for different waveguide widths at wavelengths of (a) 800 nm and (b) 1620 nm for a core thickness of 435 nm. Strip waveguides are single mode at wavelengths of 800 nm and 1620 nm for widths below 0.45 µm and 1.2 µm, respectively. Thus, a square waveguide with a cross-section of 0.435 µm × 0.435 µm is single mode at both extreme wavelengths. (c) Effective index value of the square strip waveguide for wavelengths between 800 nm to 1650 nm. TE and TM mode profiles of the square waveguide at 800 nm (d & e) and 1620 nm (f & g).

Download Full Size | PDF

A double linear adiabatic taper is used to connect the single mode fully etched strip waveguides to the shallow etched apertures of the FPR. A schematic of the rib-to-strip converter and cross-sections showing the geometry of the structure at the beginning, middle and end of the taper are illustrated in Fig. 4(a). The taper allows to expand the fundamental mode from the 435 nm wide fully etched strip waveguide to the 1.3 µm rib waveguide. The TE mode profiles at 3 points (X₁, X₂, X₃) along the taper for wavelengths of 800 nm and 1620 nm_, are shown in Fig. 4(b-g).

 

Fig. 4. (a) Illustration of the transition between the routing and shallow etched waveguides using a double linear taper as a rib-to-strip converter. Electric field distribution of the fundamental TE mode at (b) X₁, (c) X₂, (d) X₃ at 800 nm and (e) X₁, (f) X₂, (g) X₃ at 1620 nm.

Download Full Size | PDF

Although the shallow etched waveguides are multimodal, the adiabatic tapering avoids the excitation of higher order modes. Thus, only the fundamental mode carries energy in the rib waveguide. In order to ensure an adiabatic transition of the fundamental mode at all four operational bandwidths, eigen mode expansion (EME) simulations using MODE Solutions from Ansys Lumerical (Vancouver, BC, Canada) [53] were performed to optimize the length of the taper. As shown in Fig. 5(a), a taper length as short as 6 µm is sufficient to ensure a 98% transmission in the fundamental mode for the wavelengths at the extremes of the bands. However, a taper length of 29 µm is used to ensure a compact and low loss device. The transmission of this taper over the full spectrum covered by the spectrometer is shown in Fig. 5(b).

 

Fig. 5. (a) Simulated transmitted power in the fundamental mode as a function of taper length for both 800 nm and 1620 nm wavelengths. A taper length of more than 20 µm provide a transmission of more than 99% of the fundamental mode. (b) Normalized transmission over the full wavelength range for a taper length of 29 µm.

Download Full Size | PDF

In order to avoid coupling between adjacent access rib and strip waveguides, EME simulations were performed to find the minimal pitch required. The simulations presented in Fig. 6 show that a separation of 4.1 µm between rib waveguides is sufficient to ensure that there is no crosstalk at both extremes of the operational bandwidths. Thus, a pitch of 4.1 µm between access rib waveguides is considered in the design. Figures 6(a) and 7(a) show the optical field of the fundamental mode in the rib and strip waveguides respectively, at both boundary bands has decayed sufficiently to ensure that coupling between the waveguides will be negligible. Figures 6(b) and 7(b) demonstrate that light can propagate over 550 µm in both access and routing waveguides without crosstalk. This ensures that no coupling will occur over the shorter distances rib waveguides considered in the design.

 

Fig. 6. (a) Cross-section view of the TE mode profile in rib waveguides separated by 4.1 µm at wavelengths of 800 nm and 1620 nm. The electrical field has decayed sufficiently proving no coupling will occur. (b) Top view of the electric field intensity propagating over 550 µm at a wavelength of 1620 nm showing that no coupling will occur in the access waveguides

Download Full Size | PDF

 

Fig. 7. (a) Cross-section view of the TE mode profile in strip waveguides at separated by 4.1 µm at wavelengths of 800 nm and 1620 nm. The electrical field has decayed sufficiently proving no coupling will occur. (b) Top view of the electrical field intensity propagating over 550 µm at a wavelength 1620 nm showing that no coupling will occur over the access waveguides.

Download Full Size | PDF

4. Simulation results

To demonstrate the potential of the multiband spectrometer, we performed three-dimensional fully vectorial simulations based on the Huygen-Fresnel principle using the software EPIPPROP from Photon Design (Oxford, UK) [54]. The simulations were conducted to model and optimize the grating geometry, and the inputs and outputs angular positions. To ensure that the simulations model devices with realistic features, we consider in all our simulations grating facets with corners rounded with a radius of 1.1 µm and a vertical tilt of 8°. The simulated optical spectra obtained for the 39 output channels using the 4 different inputs are presented and discussed below. The performance of the spectrometer is characterized by different metrics, which are illustrated in Fig. 8. These metrics vary when switching between the inputs, i.e., operating in different band. This is attributed to the change in operational wavelength, angular position, and the deviation from the nominal angle at which incident light is hitting the grating facets, i.e., the blaze angle.

 

Fig. 8. Representation of the performance metrics used to characterize the response of the spectrometer.

Download Full Size | PDF

The device spans the following wavelength ranges: from 802.6 to 897.6 nm (95 nm), 1 262 to 1 357 nm (95 nm), 1 402 to 1 498 nm (96 nm), and 1 522 to 1 617 nm (95 nm), corresponding to input IWG1 IWG2, IWG3 and IWG4, respectively. The simulated spectra for each band are presented in Fig. 9. The resolving power (RP) defined as λ/∂λ, where ∂λ is determined using the 3-dB bandwidth (BW). The simulated resolution (3-dB BW) is 0.4 nm, 0.9 nm, 1.06 nm, and 0.92 nm, for central wavelengths ${\lambda _1}$, ${\lambda _2}$, ${\lambda _3}$, and ${\lambda _4}$, respectively. This corresponds to a RP of 2125, 1455.5, 1368, and 17,065, respectively. These results agree well with the analytical values. These RP values are comparable to spectrometers reported in the literature which operates only over a single band [8,9,31]. The simulated mean value of the 3-dB BW for the 39 channels varies from 0.37 nm for the 850 nm band (IWG1) to 1.51 nm for the L-band (IWG4). The aspect ratio defined as the ratio of the 1-dB BW to the 10-dB BW has a mean average value of 0.505, which shows that the tunable spectrometer has a Gaussian passband spectral response over the 4 bands.

 

Fig. 9. Simulated spectral response of the 39 output channels (a) for the near IR using IWG1, (b) for the O-band using IWG2, (c) for the E- and S-bands using IWG3, and (d) for the L-band using IWG4.

Download Full Size | PDF

The spectrum from IWG1 has the highest insertion loss (IL) non-uniformity of the four bands, which is 2.08 dB compared to 0.61, 0.28 and 0.42 dB for IWG2, IWG3 and IWG4, respectively. The IL of the central channels is around -0.96 dB and drops to a maximal value of -3.2 dB, for IWG1. However, the worst insertion loss value is better than -1.92 dB for the other 3 bands. The roll-off in insertion loss for IWG1 is mainly attributed to the excitation of higher order modes in the slab waveguide due to the slanted grating facets considered in the simulation. For other bands, the echelle grating exhibits less non-uniformity since the channels are distributed over a larger fraction of the FSR compared to the three other bands. For IWG1, the operational range is 95 nm for a FSR of 122.3 nm. For IWG2, the channels occupy a bandwidth of 95 nm across a FSR of 275 nm. For IWG3, it is 96 nm over a FSR of 243 nm, and for IWG4, it is 89.7 nm over a FSR of 332 nm. This is a common drawback in spectrometers based on diffractive optics [31,32,55,56], where non-uniformity across channel increases when they occupy a larger portion of the FSR. Nevertheless, the channel non-uniformity is comparable to other spectrometers that operate across one band with less output channels [31,32,55,56].

The simulated inter-channel crosstalk (X_t-int), also known as neighbour crosstalk, is better than -35 dB for IWG1. On the other hand, the phase error crosstalk (X_t) increases from -43 dB at the shortest wavelength to -35 dB for $\lambda $ = 898.5 nm. This variation is expected because for larger angles of diffraction, stronger side lobes appear due to increased residual aberrations. When the spectrometer is tuned to operate at the three other central wavelengths, the device has a better performance in terms of X_t-int and X_t. The worst neighbour crosstalk simulated, considering the fabrication limitations described above, is –27.3 dB. As for the X_t, it is as low as -32 dB, with a maximal variation of 3.4 dB and 2.1 dB across the 39 output channels for the O (IWG2) and S (IWG3) bands, respectively. Lastly, in the L-band (IWG4), the spectrometer provides a higher uniformity across the spectra at the price of higher phase-error crosstalk ($< $-30 dB).

The highest resolution and lowest side lobes are obtained for IWG1, thanks to the higher linear dispersion and mode confinement in the waveguide. This leads to a smaller divergence of the beam in the FPR, thus less facets away from the pole are illuminated, which limits aberrations. The performance is comparable to other echelle grating spectrometers reported previously but the proposed device has more output channels and denser wavelength filtering [55,57,58]. The characteristics of the filtered spectra for the four bands are summarized in Table 2 below.

Table 2. Simulated Performance Metrics of the Multiband Spectrometer

View Table | View all tables in this article

4.1 Potential applications

The spectrometer described above can be leveraged for many applications, and thus, can reduce the need for multiple devices. For instance, it could be used for label-free biosensing of analytes in aqueous solutions at 850 nm and 1310 nm since in these wavelength ranges the water absorption is 10 times less than at longer wavelengths. It could also be used in the near infrared for tuberculosis antigen detection in urine. Another potential application in the S-band is to control food quality by detecting alcohol in beverages. When using IWG3 as the input, the device could be used as an absorbance sensor to measure the concentration of ethanol in solutions. By targeting four wavelengths, 1435, 1443, 1451 and 1459 nm, different ethanol concentration could be detected. Lastly, this on-chip spectrometer could be used as a glucose sensor to obtain concentration dependent spectra in the solution to be tested. The device provides the resolution needed (3-6 nm), crosstalk better than -30 dB, and covers the wavelengths in the first overtone band required to measure light attenuation caused by the fundamental vibrations of glucose molecules. These are only a few examples to illustrate the versatility of the proposed approach.

5. Conclusion

We presented the design and simulation of an integrated optical spectrometer operating over four wavelength bands that can be fabricated with silicon nitride waveguides. The operational bandwidth can be chosen by switching between the four inputs. A double adiabatic taper was designed at the interface between strip waveguides and the free propagating region to avoid coupling to higher order modes. The reflectivity of the grating across the four wavelength bands was maximized by including a parylene coating with a thickness of 450 nm between the waveguide facet and the aluminum layer, and by optimizing the blazed angle. The device exhibits a slightly different resolutions for the different bands, but it is at most 1.51 nm. The channel spacing is less than 2.6 nm. The spectrometer provides a phase crosstalk of approximately −32 dB, which is sufficient for most spectral analysis applications, and it has low insertion losses $({ \le \textrm{ - 3}\textrm{.2 dB}} )$. The simulated spectra consider fabrication imperfections but nevertheless show a good uniformity across the 39 channels. The results demonstrate the potential of the device as a low-cost, compact solution for fully integrated multi-bandwidth on-chip spectrometers that can be leveraged for many spectral analysis applications and eliminate the need for multiple devices.