1. Introduction

Silicon photonics is expected to provide an exceptional platform for chip-level interconnect technology, due to its electronics integration capability, proven manufacturing record and price volume curve [1–3]. For chip-level interconnect applications, the power budget is projected to sub-pJ per bit for an optical link [4,5]. Another constraint for the optical components is the need for compactness to conserve silicon chip area. For a wavelength division multiplexed (WDM) optical link, electro-optic modulators and multiplexing filters are two critical components. Although silicon modulators have been demonstrated using Mach-Zehnder interferometers, they usually have energy consumption in excess of 5 pJ per bit [6,7]. On the other hand, silicon microcavity modulators, including microrings [8,9] and mircodisks [10], are expected to be promising candidates due to their smaller sizes and lower power consumption. For instance, an average switching energy of a few fJ per bit is claimed in Ref [11]. For WDM filters, cascaded microcavities were demonstrated to have advantages in terms of compactness and reconfigurability on both wavelengths and channel spacing [12–14]. However, the silicon waveguide structures commonly used in these devices are very sensitive to fabrication error, with a sensitivity figure on the order of 100 GHz/nm for a waveguide geometry of ~0.5 µm by ~0.25 µm (i.e. 1 nm of dimensional error of waveguide cross section results in a wavelength shift of 100 GHz for a resonator) [5,15]. In addition, silicon has a relatively high thermo-optic coefficient (1.86x10⁻⁴ /°C), which results in a resonance shift of about 0.1 nm/°C. In order to solve these implementation challenges, thermally tunable microcavities have been developed, but usually with relatively large tuning powers of 20 mW-100 mW to achieve one free spectral range [14,16–21]. For a transmission data rate of 25 Gbps, this thermal tuning power will add up to 1-4 pJ per bit in an optical link. Significant reduction of the tuning power is the key to practical low power optical interconnects using silicon microcavities.

In this paper, we report thermally tunable racetrack resonators with an ultralow tuning power of 2.4 mW per FSR, facilitated using free-standing structures with undercuts beneath the resonators. This tuning power is almost one order of magnitude lower than those of previously reported similar devices without undercuts. In addition, the 10%-90% switching time is about 170 µs. The demonstrated device is particularly important in applications such as wavelength-tunable microcavity modulators (to realize wavelength tunability rather than high speed modulation) and WDM filters where the tuning time is not required to be very fast but the tuning power must be minimized.

2. Device structure and fabrication

In our previous paper [14], we demonstrated thermally tunable microrings with moderate tuning power (21 mW per FSR) using air trenches beside the silicon waveguides. As shown in Fig. 1(a) , we employ metal micro-heaters on top of the resonator. The silicon waveguide has a cross section of 0.45 µm x 0.25 µm, the buried oxide thickness is 3 µm and the top cladding oxide thickness is 1.2 µm. The heater metal is Ti with a thickness of 100 nm. The heater width is 1 µm. A second layer of oxide with a thickness of 0.5 µm is deposited on the top of the Ti heater. Air trenches beside two sides of the ring waveguide are defined to improve heating efficiency. The distance between the trench edge and waveguide edge is 2 µm, far enough for the trenches not to disturb the optical mode. In the current design, we etch away the silicon underneath by isotropic dry etch after the air trenches are formed, as shown in Fig. 1(b). This undercut structure can significantly increase the tuning efficiency as the substrate silicon is the major heat sink. Previously demonstrated similar waveguide structures in Mach-Zehnder interferometers have proven that submilliwatt power can be achieved for a π-phase shifter [22], owing to the fact that the air gaps provide excellent thermal isolation between waveguides and the underlying silicon. Other related work with rib-waveguide based large 100 µm diameter rings were demonstrated in [23] and achieved approximately 4 mW per FSR with backside etching.

 

Fig. 1 (a) Cross section of the resonator waveguide with air trenches. (b) Cross section of the resonator waveguide with undercuts beneath the waveguides.

Download Full Size | PDF

Device fabrication up to air trench formation has been described in Ref [14]. The steps include sequential silicon waveguide etching, oxide cladding, heater/metal traces fabrication, and air trench anisotropic etching. We then apply an SF₆ isotropic dry etch to achieve the undercut structures. Figure 2 shows the tilted top-view scanning electron microscopy (SEM) images of two fully fabricated devices with different resonator sizes. In these devices, the resonators have a racetrack shape with straight coupling lengths of 11 µm and bending radii of 4 µm [Fig. 2(a)] and 10 µm [Fig. 2(b)]. The suspended membranes are supported by oxide beams to avoid bending. The trench widths, as shown in Fig. 2, are 6.5 µm. In general, the thicker the suspended membranes, the more stable the structures in terms of mechanical stability. Our suspended membranes have a total thickness of about 5 µm, and very few devices are found broken or bent after full fabrication. In addition, one can always reduce the trench areas to increase the dimensions of supporting oxide beams to increase the mechanical stability.

 

Fig. 2 Tilted top-view SEM for two fully fabricated free-standing racetrack resonators with a 4 µm bend radius (a) and a 10 µm bend radius (b).

Download Full Size | PDF

3. Test results

We test these thermally tunable micro-resonators using an optical detector and a tunable laser source together with a voltage-current source-meter. Spectra of the through port with different heating powers were collected for the two resonators and are shown in Fig. 3(a) and Fig. 4(b) . The spectra demonstrate FSRs of 11.5 nm for the device with 4 µm bend radius and 6.4 nm for the device with 10 µm bend radius, respectively. From these spectra, we determine the resonant wavelength shifts of each resonator with different heater powers and the results are presented in Fig. 3(b) and Fig. 4(b). The tuning power for an entire FSR is 2.4 mW for both the resonators with different FSRs. A tuning range of one FSR can compensate the resonance variations regardless of fabrication tolerance and thermal variations. In addition, as tuning efficiency (wavelength shift per unit power) highly depends on the FSR or the size of the resonator, it is more reasonable to compare the tuning power per FSR shift in order to compare the heater efficiency in different configurations. Shown here for two resonators with different sizes, the tuning powers per FSR are almost identical, indicating that this power have little dependence on the resonator size for the same heating structure (this is generally true if the ring radius is larger than a few µm). Our previous devices without undercut but with same waveguide geometries and heaters were demonstrated to have tuning powers of 27 mW and 21 mW for one FSR tuning without and with trenches, respectively. Therefore, the undercut configuration reduces the tuning power by almost one order of magnitude. The tuning efficiency in terms of wavelength shift per unit power for the devices demonstrated here is estimated to be 4.8 nm/mW and 2.7 nm/mW for the racetrack rings with 4 µm bend radius and 10 µm bend radius, respectively. They can be further increased if the resonator size is reduced. Further improvement can be made by increasing the area of the suspended membranes, which can be achieved by using larger trench width or longer silicon isotropic etching time. Reducing the size of the supporting oxide beams may help further reduce the tuning power.

 

Fig. 3 (a) Through port spectra with various heating powers for the device in Fig. 2(a). (b) Resonance shift as a function of tuning power.

Download Full Size | PDF

 

Fig. 4 (a) Through port spectra with various heating powers for the device shown in Fig. 2(b). (b) Resonance shift as a function of tuning power.

Download Full Size | PDF

In practical applications, it is important to keep high optical performance during large-range tuning. It is seen that the extinction ratios vary during thermal tuning in Fig. 3(a) and Fig. 4(a). This variation may be attributed to non-uniform heating over the resonators [24]. A more uniform heating design has been reported in Ref [24]. to keep high extinction ratios over a large tuning range. In our devices, the quality factors are approximately 8,000, nevertheless, these low quality factors are mainly due to the coupling to drop waveguides rather than resonator waveguide scattering loss.

For the demonstrated free-standing resonators, the tuning efficiency is enhanced at the expense of reduced tuning speed. We measured the tuning speed for the device shown in Fig. 2(a) by driving the heater with a 0.5 kHz square-wave voltage signal. The input wavelength is set at the resonance of the resonator at 0 V. The 10%-90% switching time is measured as ~170 µs and ~150 µs for the optical rise edge and the fall edge, respectively (Fig. 5 ). This switching time is about one order magnitude longer than for the device without undercuts [14], and comparable with those for MZI devices with undercuts [22]. Table 1 summarizes tuning powers per FSR and tuning speed for some of the reported thermally tunable microcavities we found in the literature. There are mainly three groups: (1) metal heaters on top of silicon waveguide cladding without undercuts [12,14,16], (2) silicon waveguide heaters fabricated by doping and annealing [20], and (3) metal heaters with undercuts (our current devices). Both devices from group 1 and 2 exhibit an optimal tuning power ~20 mW per FSR, however, the speed for metal heaters in the first group is about 10 times slower than for silicon waveguide heaters. Undercut structures can significantly reduce the tuning power, but with slower tuning speed (~170 µs). In order to choose a proper heater for a particular application, one may need to consider the required speed and power together with fabrication complexity. Metal heaters, compared with silicon waveguide heaters, have the advantage including simple fabrication without ion implantation and annealing, and negligible excess loss from heaters.

 

Fig. 5 Temporal response of the resonator in Fig. 2(a). The green line represents the electrical drive signal with a voltage swing of 0.5 V, and the blue line indicates the optical transmission at the through port. 10%-90% switching times were measured as ~170 µs for the rise time and ~150 µs for the fall time.

Download Full Size | PDF

Table 1. Tuning power and speed comparison of some of previously demonstrated thermally tuned silicon microcavities and the device presented in this work

View Table

4. Discussion and conclusion

The wavelength tunability required for silicon microcavity modulators and WDM filters needs to have low power consumption, but not necessarily high speed. In practical applications, the environmental temperature may change over a time period far longer than 1 ms. In this case, the ultralow power micro-resonators demonstrated here with response times of ~170 µs may be fast enough. Considering the achieved low tuning power of 2.4 mW per FSR, tunability would add a maximum of ~100 fJ/bit for a data rate of 25 Gbps in an optical link. This low power/energy consumption further validates the use of silicon microcavities to realize chip-level optical interconnects [4,5]. Beside the power consumption, another concern regards the thermal crosstalk between adjacent microcavities. In our current single microcavity structure, no information on thermal crosstalk can be extracted. However, it is expected that undercuts would increase thermal crosstalk, compared with trenches-only devices in Ref. [14], since the undercuts block the heating flux to substrates and force the heating flux to oxide membranes. However, thermal crosstalk can be reduced if two adjacent microcavities are separated by air trenches rather than oxides. In on-chip optical networks, each microcavity filter may be close to its own receiver or transmitter sites, so that they are far away from each other. In this case, thermal crosstalk is not an issue. Another concern for microcavities with undercuts is how to implement this configuration for silicon microcavity modulators, which usually require a slab layer to realize carrier injection or extraction. If the slab layer stops before the trenches, the silicon waveguide and metal in the modulator can be well protected during the etching to achieve undercuts. Therefore, it is not difficult to implement this heating configuration with undercuts for active devices such as modulators.