1. Introduction

The temperature-dependent behavior of integrated photonic devices is a fundamental issue that affects almost every aspect of circuit and device design and every technological platform. This is particularly true for silicon-on-insulator (SOI) and other semiconductor-based waveguides that exhibit thermo-optic coefficients almost ten times larger than that in silica. This results in large wavelength shifts of the order of several tens of picometers per degree in the optical communication bands and causes severe distortions of the device responses [1,2]. In order to avoid or limit the use of energy-hungry active temperature control, several solutions have been proposed in the literature to mitigate the effect of temperature fluctuations by designing athermal devices. For wavelength filtering and routing, arrayed waveguide grating (AWG), lattice filter and echelle grating are the main device options, each with its advantages and shortcomings [3–6]. For AWGs and lattice filters in which the phase accumulation takes place in wire or ridge waveguides, athermal behavior may be achieved by using a material with a negative thermo-optic coefficient as the waveguide top cladding to counterbalance the positive coefficient of the core and substrates. Exploited materials include resins [7], sol-gel materials [8], polymers [9, 10], and amorphous titanium dioxide [11]. In echelle gratings where the phase accumalation is in the slab free-propagation region, a similair method has been applied to devices based on III–V semiconductors [12]. However, this method is ineffective in echelles based on a submicron silicon platform particularly for devices operating in the transverse electric (TE) polarization since the modal overlap with the cladding material is very small. Additionally, the use of these materials requires a substantial change to the standard SOI photonics fabrication processes, which may hinder their adoption outside of research scenarios. On the other hand, it is possible to utilize the dimensional dependece of the waveguide temperature sensitivity to engineer the overall device thermo-optic behavior [13,14], and an athermal lattice filter has been reported exploiting this technique [15].

In this work we propose a way to design a SOI athermal echelle grating filter operating in the TE polarization. We achieved the desired goal without using materials with negative thermo-optic coefficient and relying exclusively on a standard fabrication process for SOI waveguides. Inspired by the concept of wavelength-synchronized interferometers for flat-top transmission [16,17], we exploit a temperature-synchronized (T-synced) Mach-Zehnder interferometer (MZI) as input to the echelle to match and compensate its focal position shift with temperature, thereby achieving a large reduction in the temperature dependence of the overall transmission spectrum. Unlike for the case of wavelength-synchronized interferometers, the proposed MZI is designed to have a very broad free-spectral-range (FSR) to obtain a T-synced input that has a neglegible variation with wavelength of the generated field profiles. This modified input is similar to that proposed by Kamei et al. [18] but there it was only used to compensate for the small second-order temperature dependence of silica-based AWGs in combination with resin-filled grooves. To the best of our knowledge, this is the first proposal of an athermal echelle grating in TE for the high-index-contrast SOI platform. The designed device shows a residual zero-mean wavelength fluctuation of the output spectrum and a maximum shift of less than 45 pm over a temperature range of 20 K, compared to a 1.6-nm shift for the same grating with a conventional waveguide input. The increase of the channel insertion loss due to the use of the T-synced input is limited to about 0.5 dB. Moreover, the proposed design shows a good tolerance to fabrication uncertainties, with limited degradation of the performance for waveguide width uncertainty of ±10nm.

Design concepts for the Mach-Zehnder interferometer and the interface with the echelle filter are presented in sections 2.1 and 2.2, respectively. Section 3 shows the results of the overall device simulations, the analysis of the performance and sensitivity to uncertainties.

2. Principle and design of the device

In an echelle grating filter, the incoming and outgoing angle of the light beams ϑ and φ_k are governed by the following relation [19]:

The proposed temperature-insensitive (athermal) echelle grating filter is schematically shown in Fig. 1. The temperature synchronized Mach-Zehnder interferometer (MZI) placed at the entrance to the slab area is made of two 3-dB directional couplers (marked as first and interface couplers in the figure) and two unbalanced branches with lengths L₁ and L₂ and widths w₁ and w₂, respectively. Different waveguide widths allow varying the effective thermo-optic coefficients of the two waveguide branches, thus precisely controlling the dependence of the MZI behaviour on temperature [13, 15]. As depicted in Fig. 1, the coupled waveguides of the interface 3-dB coupler form the input to the echelle filter and, driven by the temperature-dependent phase delays in the MZI, generate a field distribution at the interface that shifts with temperature. The shift of field peak position is designed to match the thermal dispersion of the echelle grating over a selected temperature range ΔT and thus ensures that the field reflected by the grating at a given wavelength is imaged on the same output waveguide for all temperatures. To achieve a temperature-independent behavior for all the wavelength channels of the echelle filter, the MZI is designed to have a large FSR, minimizing the roll-off of its transfer function over the full operational bandwidth and hence ensuring that the field generated at the entrance of the slab area is the same for all the channels.

 

Fig. 1 Schematic of the SOI athermal echelle grating with a temperature-synchronized Mach-Zehnder input for temperature-insensitive operation. The inset shows the designed changes of the optical field with temperature at the entrance of the slab area.

Download Full Size | PDF

As a proof-of-concept, we consider a standard SOI technology with a 220-nm silicon core thickness and silicon dioxide upper and lower claddings. The echelle grating is designed to have four channels with 800 GHz spacing centered around λ₀ = 1550 nm and 1-dB bandwidth of about 140 GHz. The grating is in a Rowland circle configuration with a radius of 250 μm, a diffraction order of 20, and a mounting angle of 45 degree. The choice of a small Rowland circle radius helps reducing the impact of the silicon thickness inhomogeneity as pointed out in [3,6]. As will be discussed in later sections, small dimensions are also advantageous for the compensation of the thermo-optic effects. The angular dimension of the grating is 15 degree in order to capture the entire field of the T-synced input. The reflectance of the grating facets is assumed to be 100% in the simulations. High reflectance can be achieved in practice exploiting Bragg reflectors [6]. At the Rowland circle waveguides have a width of 2 μm and are then tapered to 500 nm to ensure single-mode propagation. With the employed 220-nm silicon core with oxide cladding, the temperature-induced angular shift of the echelle grating is 0.0076 deg/K (see Eq. (1)). The corresponding spacial shift is approximately 950 nm over a 20 K temperature range.

Key considerations for the design of the complete device are athermal operating temperature range, overall device size, excess loss and tolerance to fabrication uncertainties. The design of the Mach-Zehnder interferometer is presented in Sec. 2.1 while Sec. 2.2 discusses in detail the design of the coupler at the interface with the slab area. The electromagnetic simulations presented in the next sections for the Mach-Zehnder were obtained with a commercial finite difference mode solver and a 3D mode expansion propagation solver [20]. The design and simulation of the echelle grating were performed with a dedicated 2D semi-vectorial in-house tool based on the Huygens-Kirchhoff diffraction theory and the effective index approximation. The software can handle input fields of arbitrary and strongly dispersive shapes as could be possibly generated by some T-synced input configurations.

2.1. Design of the Mach-Zehnder interferometer

During the design of the MZI, we considered for the first 3-dB coupler a width of 500 nm for both waveguides and a gap of 300 nm, obtaining a 3-dB length L_f = 46 μm. Interconnecting waveguide sections were assumed to have the same length in the two MZI branches and a width of 500 nm, thus not contributing to the overall transfer function. The design was performed by defining for each branch a section of length L₁ and L₂ and width w₁ and w₂, respectively, to simultaneously maximize the FSR and completely switch the Mach-Zehnder output from bar to cross condition (or vice-versa) within the given temperature variation ΔT to match the echelle focal spot shift with temperature.

Assuming a linear frequency dispersion around the central wavelength λ₀, the FSR of the Mach-Zehnder (in wavelength) is defined as

Equations (3) and (4) allow to calculate the lengths L₁ and L₂ for any chosen combination of w₁ and w₂ in order to simultaneously maximize the FSR and match the shift of the grating focal point over the temperature range ΔT. To this purpose, the effective index of the fundamental TE-like mode of an SOI waveguide was simulated as function of waveguide width, temperature and wavelength. To model silicon properties, we considered at λ = 1550 nm and T = 300 K a refractive index of 3.479, a thermo-optic coefficient of 1.87 ×10⁻⁴ K⁻¹ and a spectral dispersion of −0.083 μm⁻¹.The same parameters for silica were chosen as 1.444, 0.85×10⁻⁵ K⁻¹ and −0.012 μm⁻¹, respectively [21]. From these simulations, the dependence of the group index n_g and the (effective) thermo-optic coefficient ∂n_eff/∂T on waveguide width were calculated. Figure 2 shows the obtained results for λ₀ = 1550 nm and T₀ = 300 K.

There are several considerations for choosing w₁ and w₂. An MZI with w₁ < 350 nm and w₂ > 500 nm (or vice-versa) maximizes the difference between the two thermo-optic coefficients and the two group indices, which together contributes in defining the required lengths of the branches through Eqs. (3) and (4). Consequently this would reduce L₁ and L₂ and the overall device size. On the other hand, this choice would make the device very sensitive to fabrication uncertainties because of the steep dependence of both n_g and ∂n_eff/∂T on waveguide width for w < 350 nm. In the proposed design we hence chose w₁ = 380 nm and w₂ = 800 as a viable compromise between robustness and device compactness (dashed lines in Fig. 2). In this case ∂n_eff,1/∂T = 1.981×10⁻⁴ K⁻¹, ∂n_eff,2/∂T = 1.936×10⁻⁴ K⁻¹, n_g,1 = 4.426, and n_g,2 = 3.909 for T₀ = 300 K and λ₀ = 1550 nm. For an athermal operating temperature range ΔT = 20 K, the corresponding lengths of the two branches obtained with Eqs. (3) and (4) are L₁ = 1707 μm and L₂ = 1933 μm. The choice of the temperature range is discussed in Sec. 2.2.

 

Fig. 2 Effective thermo-optic coefficient (TOC) and group index for the fundamental TE-like mode of an SOI waveguide as function of the waveguide width. The dashed lines mark the chosen widths w₁ = 380 nm and w₂ = 800 nm.

Download Full Size | PDF

2.2. Design of the coupler-slab interface

The design of the 3-dB directional coupler at the interface with the echelle slab area is particularly important for minimizing the excess loss generated by replacing a simple input waveguide with the proposed T-synced input. It should be noted that the modal mismatch between the coupler and the slab area also generates back-reflections that might be critical for certain applications and should hence be minimized. A schematic of the directional coupler and interface section is shown in Fig. 3(a). Here the design parameters are the waveguide distance d and the waveguide width w_c. The goal is to match the focal point shift with temperature in the echelle for as large a temperature range as possible, while maximizing the modal overlap between the field generated by the T-synced input (and imaged by the grating to the echelle outputs) and the fundamental mode of the 2-μm-wide output waveguides.

 

Fig. 3 (a) Schematic of the interface 3-dB coupler. (b–d) Field profile |E_x|² for λ = 1550 nm at the entrance to the echelle slab area for temperature of 290 K, 300 K and 310 K, respectively.

Download Full Size | PDF

To this purpose, the distance d must be equal to the shift produced by the echelle slab over the selected temperature range ΔT. A small value of d is desirable to both improve the overlap with the output mode and reduce the length of the 3-dB interface coupler. Since the echelle angular shift with temperature is determined by the thermo-optic properties of the slab waveguide according to Eq. (1), a small d can be achieved by reducing the Rowland radius. On the other hand, considerations for example on channel cross-talk, grating truncation loss and fabrication issues limit the minimum radius usable. As a balanced choice, we selected a Rowland radius R = 250 μm that for a temperature range ΔT = 20 K centered at 300 K is expected to generate a focal point shift of 950 nm, which is hence chosen as value for d.

Overlap losses and back-reflections depend on w_c in two opposite ways. Narrow waveguides are required to expand the field profile in the horizontal direction and improve the in-plane overlap with the fundamental mode of the echelle output. On the other hand, narrow waveguides delocalize the field also in the vertical direction, decreasing the out-of-plane overlap and increasing the back-reflections. For our design, w_c = 255 nm represents a good compromise, giving an overlap loss with the fundamental mode of a 2-μm waveguide at T₀ = 300 K of about 0.7 dB and back-reflections smaller than −14 dB. The increase of the overlap losses at T = 290 K and T = 300 is smaller than 0.4 dB. Two 35-μm-long linearly tapered sections were used to adiabatically adjust the width from the initial 500 nm to w_c = 255 nm with negligible losses. A total coupler length L_O = 36 μm (including tapers) ensures a 3-dB operation at λ₀ = 1550 nm and T₀ = 300 K.

As an example, Figs. 3(b)–3(d) shows the simulated field profile |E_x|² generated at the entrance to the slab area by the interface coupler at temperature 290 K, 300 K and 310 K, respectively. For these simulations, we assumed the fundamental quasi-TE modes being excited at the beginning of the two 500 nm wide waveguides with the same power and a phase difference of −π/2 rad, 0 rad and π/2 rad respectively, as generated by the Mach-Zehnder designed in section 2.1 for the three considered temperatures. As expected, the center of the spot shifts about 950 nm with a temperature variation of 20 K. The field shape also changes with temperature, causing a slight variation of the shape of the echelle pass-bands as described in the next section.

3. Athermal echelle simulations and performance

For the evaluation of the entire device we considered the parameters optimized in the previous sections: ΔT = 20 K centered at T₀ = 300 K; central wavelength λ₀ = 1550 nm; for the first 3-dB coupler, L_f = 46 μm, waveguide width 500 nm and gap 300 nm; for the MZI branches w₁ = 380 nm, w₂ = 800 nm, L₁ = 1707 μm, and L₂ = 1933 μm; for the interface 3-dB coupler d = 950 nm, w_c = 255 nm and length L_O = 36 μm; a waveguide width of 2 μm for the echelle output waveguides. The simulation of the entire T-synced input was performed exploiting a transfer matrix approach, describing its transmission as a function of wavelength and temperature as [17]

The simulation results are presented in Fig. 4. As a reference, Fig. 4(a) shows the transmission spectra obtained with a conventional simple 2-μm-wide input waveguide for the four channels of the echelle grating at 290 K, 300 K and 310 K, respectively. At 300 K, the four channels are located at λ₁ = 1540.56 nm, λ₂ = 1546.92 nm, λ₃ = 1553.33 nm, and λ₄ = 1559.80 nm. Channel 3-dB bandwidth is about 250 GHz. The minimum insertion loss is 0.4 dB, the power roll-off is about 0.3 dB and channel cross-talk is smaller than −35 dB. By using the proposed T-synced input, the spectral shift is significantly reduced, as shown in Fig. 4(b). This design modification introduces a slight dependence of the pass-band on temperature due to the variation of the corresponding field distribution generated at the entrance to the slab area (see Figs. 3(b)–3(d)). Channel cross-talk slightly increases only at the edges of the design temperature range.

 

Fig. 4 Simulations of the transfer functions of the four echelle filter channels with (a) a conventional 2-μm waveguide input and (b)) using the proposed temperature-synchronized input. The numerals denote the channel number. (c) Channel wavelength shift as a function of temperature in the case of waveguide input (black solid line - the same for all the channels) and T-synced input (red diamonds and blue crosses for channels 1 and 2, respectively). (d) Insertion loss for channels 1 and 2 calculated at the channel central wavelength at 300 K.

Download Full Size | PDF

Figure 4(c) shows the shift of the central wavelength of the channels. For the conventional waveguide input, a linear shift with temperature of approximately 1.6 nm over the 20-K temperature range (or 80 pm/K, see the black solid line) is observed. For the T-synced input case, red diamonds and blue crosses report the residual wavelength shift for channels 1 and 2. Channels 3 and 4 have similar properties and are omitted here for clarity. As can be seen, there is a residual zero-mean fluctuation of the channel spectral position with temperature. This is due to the fact that the “center-of-gravity” of the field distribution generated by the MZI depends on temperature through a squared sine function. The maximum of the average residual fluctuation is about ±45 pm at 295 K and 305 K.

Lastly, Fig. 4(d) reports the insertion loss for channels 1 and 2 at their central wavelengths λ₁ = 1540.56 nm and λ₂ = 1546.92 nm, respectively, for both the waveguide and T-synced input cases. Results are similar for channels 3 and 4. With a conventional waveguide input, central channels (2 and 3) have an insertion loss of about 0.4 dB at 300 K that quickly increases to more than 2 dB with temperature variations of ±10 K due to the shift of the pass-band. The use of the T-synced input increases the insertion loss at 300 K for these channels to about 0.9 dB due to the additional overlap losses with the fundamental mode of the 2-μm output waveguide. On the other hand, this insertion loss has a much smaller dependence on temperature and reaches only 1.3 dB at 290 K and 310 K. In this case the additional losses mainly come from a slight residual wavelength dependence of the Mach-Zehnder transfer function and hence of the field generated at the entrance of the slab area. For both types of input designs, the outer channels (1 and 4) have behavior similar to the central channels but with an additional loss of about 0.3 dB due to the power roll-off of the echelle grating.

The careful selection of the waveguide width for the two Mach-Zehnder branches presented in Sec. 2.1 and the balanced design of the interferometer provide a good robustness of the proposed device to fabrication errors. In order to test this sensitivity we repeated the simulations assuming a variation Δw = ± 10 nm of the waveguide width in the MZI, including both 3-dB couplers and the two branches. This range of dimensional changes well represents the expected linewidth uniformity with 193 nm deep-UV lithography [22]. The dimensions of the echelle grating are kept unchanged as variations on this scale on the grating facet positions do not significantly change the filter performance, causing at most a slight increase of the channel cross-talk. Figure 5 reports the simulation results along with the ideal response (Δw = 0, same data shown in Fig. 4(b)) for channel 1 at 290 K, 300 K, and 310 K, respectively. Since the entire MZI design is performed at the central wavelength 1550 nm, the outer channels 1 and 4 are expected to suffer more spectral distortion compared to the central channels 2 and 3. As can be seen, the device exhibits a good tolerance to fabrication uncertainties. The effect of the waveguide width variation on the pass-band shape and central wavelength are negligible at the center of the considered temperature range (300 K). At the edges (290 K and 310 K) the compensation of the thermal shift of the transfer function is still effective, preserving the functionality of the device, with some minor effects on the shape of the pass-band. Similar results were obtained for channel 4 while channel 2 and 3 showed an expected higher tolerance to waveguide width deviations, with negligible effects also at 290 K and 310 K. The same device tolerance is predicted also considering waveguide width deviations between the two branches that are uncorrelated but within the same range.

 

Fig. 5 Worst-case fabrication tolerance analysis: pass-band of outer channel 1 (λ₁ = 1540.56 nm) at 290 K, 300 K, and 310 K, respectively, with a waveguide width deviation of ±10 nm in the T-synced input.

Download Full Size | PDF

4. Conclusions

We have proposed a design for an athermal echelle grating filter operating in TE polarization that relies exclusively on standard silicon-on-insulator photonics fabrication processes without using any additional material with negative thermo-optic coefficient. The device exploits a Mach-Zehnder interferometer with a very large free spectral range as a temperature-synchronized input of the echelle grating to match the shift of the grating focal point with temperature. In this way, a large reduction of the sensitivity of the overall transfer function with temperature has been achieved, with a residual zero-mean fluctuation of about ±45 pm over a temperature range of 20 K compared to a 1.6-nm shift obtained with a simple waveguide input. The use of the Mach-Zehnder at the input slightly increased the device insertion loss by about 0.5 dB. The device functionality has been numerically demonstrated taking into account also a waveguide width fabrication uncertainty of ±10 nm. The covered temperature range could be increased by further minimizing the radius of the echelle grating and by exploiting different types of interface couplers (e.g. multi-mode interference couplers).