1. Introduction

Photonic integration brings various functionalities from multiple chips into a single chip while greatly reducing the number of optical interfaces, optical packages, and thermo-electric coolers (TECs) associated with the multiple discrete chips. Among many photonic integration platforms including GaAs, InP, silica, and other material platforms, silicon photonics has rapidly emerged as a viable and attractive integration platform [1,2] due to the availability of high-quality oxides that enable low-loss and compact waveguides while exploiting widely available CMOS fabrication facilities. However, practical deployment of silicon photonics exploiting multi-wavelength parallelism is seriously suffering from its strong temperature dependent characteristics. Silicon has a relatively large thermo-optical coefficient (TOC) of 1.84 × 10⁻⁴/°C, which is more than an order of magnitude greater than that of silica. As a result, typical silicon photonic devices suffer from ~0.1 nm/°C wavelength dependence with temperature variations, and they require optical packages with TECs for practical deployment. Since athermal silica photonic integrated circuits are already widely deployed in communication networks in simple packages without TECs, the cost advantages and reliability of silicon photonics suffer from their strong temperature dependence.

There has been in large, three main methods to overcome this challenge. The first method has been to introduce integrated heaters onto the silicon photonic devices to actively monitor and control the temperature using feedback control circuits, but this adds complexity and additional power consumption [3–5]. The second method has been to integrate a ring resonator with a Mach-Zehnder interferometer which compensates the thermal dependence of the ring [6,7], but this solution increases the total footprint and typically shows significant wavelength dependence. The third method has been to introduce an upper cladding made of a material with negative TOC to compensate for the positive TOC of silicon. Early demonstrations have utilized polymers with negative TOC as an uppercladding to realize athermal silicon photonics [8–10]. Recently, CMOS-compatible fabrication process involving titanium dioxide (TiO₂) has been demonstrated and investigated [11–14] for operation in the 1.550 nm wavelength range. TiO₂ has a relatively strong negative TOC of approximately - 1 × 10⁻⁴/°C ~- 2 × 10⁻⁴/°C around 1550 nm due to presence of a soft electronic band [15] and is already adopted in CMOS processes due to superior reliability. On the other hand, athermal operation of silicon photonic devices in the 1310 nm wavelength range is very important because semiconductor lasers have higher T₀ values and optical fibers have lower dispersion at this range compared with the 1550 nm wavelength range [16]. In this paper, we investigate athermal operation of silicon ring resonators overclad with TiO₂ cladding in the 1310 nm wavelength range. The CMOS compatible fabrication led to TiO₂-clad silicon ring resonators that showed near complete cancelation of the first order thermo-optical dependence and exhibited residual second order thermo-optical effects in the 1310 nm wavelength range.

2. Design of athermal Si-TiO₂ waveguides

An athermal silicon photonic waveguide consists of silicon core, buried oxide (BOX) undercladding and TiO₂ overcladding. Figure 1(a) shows the schematic cross-section of the athermal waveguide. The equivalent thermo-optic coefficient ($\frac{d{n}_{eff}}{dT}$) of the TiO₂-overclad silicon waveguide (Si-TiO₂) can be expressed as

 

Fig. 1 (a) A cross-sectional schematic of athermal Si-TiO₂ waveguide. (b) Calculated dn_eff/dT as a function of the waveguide width for different refractive index values of TiO₂ cladding for 1310 nm operation.

Download Full Size | PDF

3. Device fabrication

Figure 2 illustrates the fabrication process. The device fabrication utilized a 150mm silicon-on-insulator (SOI) wafer with 220nm thick top silicon and 3µm thick buried oxide layers. The top silicon layer is thinned down to 200nm by thermal oxidation. A silicon nitride hard mask of 30nm is deposited by low-pressure chemical vapor deposition (LPCVD). The ring resonator structures are patterned with 248nm projection lithography followed by reactive ion etching (RIE) with CF₄ for nitride and HBr for silicon. Then the ring waveguide structures are thermally oxidized with the hard mask still on top in order to smooth the sidewalls and to reduce the waveguide width below the lithography resolution of 250 nm dimension. The oxide and hard mask are then stripped in hydrogen fluoride and hot phosphoric acid, respectively. Figure 3 shows the scanning electronic microscopy (SEM) images of the fabricated device before TiO₂ cladding deposition. The ring resonator has a radius of 25 µm and a gap spacing of 550nm between ring and straight waveguides. The waveguide core has a width of 200nm and a slab thickness of 30nm. At the two end points of the coupling waveguides at the facet are inverse tapers with tip widths of 200nm to facilitate low-loss optical coupling.

 

Fig. 2 Fabrication process steps: (a) Initial SOI wafer. (b) SiN hard mask deposition. (c) Waveguide layer patterning. (d) Thermal oxidation to reduce waveguide width. (e) SiN and thermal SiO₂ strip. (f) SiO₂ layer deposition. (g) Trench opening for TiO₂ cladding. (h) Deep etch for waveguide edge coupler. (i) TiO₂ cladding deposition.

Download Full Size | PDF

 

Fig. 3 Waveguide geometries. (a) Top-view SEM image of a ring resonator before TiO₂ cladding deposition; (b) Ring-to-bus waveguide coupling region; (c) Cross-sectional SEM image of a waveguide before TiO₂ cladding deposition; (d) Top-view SEM image of an inverse taper.

Download Full Size | PDF

The patterned devices are then deposited with 1µm thick SiO₂ by LPCVD. The trench for TiO₂ cladding is opened by CF₄ RIE etching followed by wet etching. The waveguide facet is deep reactive-ion etched (DRIE) by 100µm to accommodate fiber edge coupling. Finally TiO₂ cladding of 1.3 µm thickness is deposited by magnetron-sputtering for athermal operation. The relatively thick (1.3 µm) TiO₂ cladding was chosen to suppress optical scattering from the air-TiO₂ interface, and the optimal sputtering condition of 360W RF power and 12% oxygen content [11] was used to realize low loss TiO₂ cladding.

4. Device characterization

Transmission spectra measurements utilize a tunable laser and a photodiode with a device in the interferometer of the optical vector network analyzer (OVNA). The OVNA interferometer utilizes a pair of single-mode lensed fiber couplers with 2.5µm diameter spot sizes (1/e²) to interface with the inverse tapers of the Si-TiO₂ waveguide at input and output facets to launch light into and to collect light from the athermal Si-TiO₂ ring resonator. The silicon photonic chip containing the Si-TiO₂ ring resonators is mounted on a stage with a thermo-electric cooler (TEC) in order to control the device temperature with better than 0.1°C accuracy. We measured the transmission spectra of the Si-TiO₂ ring resonators as well as control devices (Si-SiO₂ ring resonators) overclad with SiO₂ instead of TiO₂ prepared on the same silicon photonic wafer. Figure 4 shows the measured transmission spectra of the two types of devices: Si-SiO₂ ring resonators and Si-TiO₂ ring resonators. The device with SiO₂ cladding shows a significant resonant wavelength redshift of 0.86 nm over 15°C variations. The resonance extinction ratio also changes when temperature increases, which may result from the temperature-induced change of coupling coefficient between ring and bus waveguides. The device with TiO₂ cladding shows less than 0.1 nm over the same temperature range, which is below the full width at the half maximum (FWHM) linewidth of 0.25 nm for the resonator. We extract the propagation losses of the waveguides by fitting the transmission spectra of the ring resonators. The propagation losses of 200nm waveguide core width with SiO₂ and TiO₂ cladding are 7 dB/cm and 20 dB/cm, respectively. Thus the extra loss for TiO₂ cladding as compared to SiO₂ cladding is 13 dB/cm, which is comparable to other existing results in 1550nm wavelength range [11,13].

 

Fig. 4 (a) – (b) Measured transmission spectra at various temperature of the devices with (a) SiO₂ and (b) TiO₂ over cladding. (c) Resonance wavelength λ_r as a function of temperature for the devices of 200nm width with SiO₂ and TiO₂ cladding.

Download Full Size | PDF

Figure 4(c) shows the measured resonance wavelength λ_r as a function of temperature for the devices of 200nm waveguide core width with SiO₂ and TiO₂ cladding. The curve-fit parameter to this measured data shows a wavelength shift of 50 pm/°C for the ring resonator with SiO₂ cladding. The curve-fit parameter to the data obtained for the ring resonator with TiO₂ cladding shows nearly complete cancellation to first order (< 6 pm/°C) while exhibiting the second order dependence at 0.8 pm/°C² centered at the design temperature of 35°C. The net result was less than 0.2 nm variation in the 20-50 °C temperature range.

In order to explain the quadratic dependency of wavelength upon temperature, second-order material TOCs as well as the confinement factor dependence on the first order material TOCs have to be taken into consideration. Material TOC can be expressed as:

Here, ${\Gamma }_k$ is the confinement factor of each constituent part. Again, we assume that the $d\Gamma /dT$ induced second-order effect is negligible and proceed with the approximation in Eq. (5) similarly as in Eq. (1). Plug Eq. (4) into Eq. (3) and we can get $d{\lambda }_r/dT$ as a first-order polynomial function of temperature T which can explain the quadratic relationship between wavelength and temperature. The TOC of silicon at 1310 nm is reported as$\text{1}.\text{834}\times {10}^{-4}+\text{4}.\text{887}\times {10}^{-7}\cdot T$ [17], where T is in unit of °C. The${\beta }_{Si{O}_2}=1\times {10}^{-5}\text{/}°\text{C}$ and${\gamma }_{Si{O}_2}$is assumed to be 0. The TOC of TiO₂ is calculated to be $-3.07\times {10}^{-4}+4.45\times {10}^{-6}\cdot T$ based on fitting the measurement data.

Our simulation for the waveguide with a core width of 200nm and TiO₂ overcladding indicates that the $d\Gamma /dT$ induced second-order coefficient $d^2{\lambda }_r/d{T}^2=2.3\times {10}^{-5}\text{nm}/{}^o\text{C}{}^{\text{2}}$, which qualitatively agrees with the result in [10]. This second-order coefficient is much smaller than the one observed in our measurements (8 × 10⁻⁴nm/°C²), which verifies the approximation of neglecting$d\Gamma /dT$in Eqs. (1) and (5).

The temperature-dependent wavelength shift can be varied at different spectral ranges. We measured the transmission spectra from 1270 nm to 1340 nm over the 20-50°C temperature range. Figures 5(a)-5(c) show the measured resonance wavelength shift as a function of temperature at various spectral ranges. We extracted the slope of temperature-dependent wavelength shift dλ_r/dT at T = 35°C. Figure 5(d) shows the dλ_r/dT as a function of wavelength. At 1270nm wavelength, resonance redshifts when temperate increases and dλ_r/dT is ~8pm/°C. At 1310nm wavelength, resonance wavelength first blueshifts and then redshifts as temperature rises, and dλ_r/dT at T = 35°C is close to zero. At 1340nm wavelength, resonance wavelength blueshifts as temperature rises and dλ_r/dT at T = 35°C is approximately −8pm/°C. For a fixed waveguide geometry, when the wavelength becomes longer, mode confinement increases in TiO₂ cladding, which contributes to the resonance blueshift and athermalization.

 

Fig. 5 (a) – (c) Measured (black circle) and quadratically fitted (red line) resonance wavelength shift as a function of temperature. The device with a waveguide width of 200nm is measured at various spectral ranges around (a) 1270nm, (b) 1310nm and (c) 1340nm. (d) Temperature-dependent wavelength shift dλ_r/dT at T = 35°C as a function of wavelength. Black square: extracted dλ_r/dT from the measurement result. Red line: linear fitted dλ_r/dT at T = 35°C as a function of wavelength.

Download Full Size | PDF

The temperature-dependent wavelength shift is also varied for different waveguide core widths. For a wavelength division multiplexing (WDM) application, the athermal operation of silicon ring resonators at different wavelengths can be achieved by appropriately designing the waveguide core width. Figure 6 shows the measured transmission spectra at various temperatures of the ring resonators with different core widths. The 200nm and 220nm wide devices achieve athermal operation at around 1315nm and 1336nm, respectively.

 

Fig. 6 (a) Measured transmission spectra at various temperature of the devices with 200nm and 220nm widths. (b) Measured resonance wavelength λ_r as a function of temperature for the devices with 200nm and 220nm widths.

Download Full Size | PDF

5. Conclusions

We demonstrate an athermal silicon ring resonator clad with TiO₂ cladding and measure temperature-dependent resonant wavelength shifts less than 6 pm/°C, exhibiting second order effects near 1310 nm. In order to athermalize silicon photonic devices over a wide temperature range, the waveguide geometries, first and second order TOC values of the materials, and operating wavelengths have to be carefully considered. In particular, the second order effects of TOC of TiO₂ need to be included in the design and operation near the athermal condition.