Compact silicon-on-insulator (SOI) waveguide thermo-optically tunable Fabry-Perot microcavities with silicon/air Bragg mirrors are demonstrated. Quality factors of Q=4,584 are measured with finesse F=82. Tuning is achieved by flowing current directly through the silicon cavity resulting in efficient thermo-optic tuning over 2 nm for less than 50 mW applied electrical power. The high-Q cavities enable fast switching (1.9 μs rise time) at low drive power (<10 mW). By overdriving the device, rise times of 640 ns are obtained. Various device improvements are discussed.
©2007 Optical Society of America
Silicon-on-insulator (SOI) is a promising substrate material for high-speed electronics and increasingly finds application in optoelectronics. Thus, we can anticipate an eventual merging of the two fields on a common material platform. Application areas include high-speed microprocessors with optical interconnects, and optoelectronics for wavelength division multiplexing (WDM) with on-chip control circuits.
Silicon’s thermo-optic effect  is significantly larger than its electro-optic effect  and enables low-power switching via temperature dependent modulation of the refractive index. Therefore, the thermo-optic effect lends itself well to wavelength-division multiplexing (WDM) devices and components. A number of different silicon-based thermo-optic devices have been reported to date. Mach-Zehnder interferometer (MZI) switches with rise-/fall times of a few microseconds have been demonstrated with heating powers in the range of 100 mW [3–4]. By utilizing a multi-step voltage circuit to overdrive a thermo-optic switch, response times <1 μs have been achieved . Sub-μs speeds with sub-mW switching power were demonstrated by differentially heating the two arms of a ≈100 μm long silicon strip waveguide MZI . Finally, microring , Fabry-Perot , and photonic crystal cavity  resonators enable high-speed, low power switching due to their potential for high quality factor. However, such resonant thermo-optic switches have so far been limited in terms of speed or power consumption due to their large thermal mass  or moderate Q-factor [8, 9].
In this paper we present an SOI waveguide thermo-optic switch. Fast response time and low power are demonstrated by utilizing a high Q-factor Fabry-Perot resonator that is thermo-optically tuned by passing current directly through the silicon microcavity. The high-Q enables long effective optical interaction lengths in a compact device. In contrast, an MZI generally requires a long device for similar optical interaction length. Furthermore, passing current directly through the cavity enables rapid heating at low electrical powers, resulting in efficient and low-power tuning and high speed switching. We model and experimentally characterize our device and discuss alternate designs that may enable an order of magnitude improvement in terms of speed and power consumption.
The Fabry-Perot cavities consist of silicon-on-insulator (SOI) rib waveguides with deep-etched silicon/air distributed Bragg reflector (DBR) mirrors. Due to the large index contrast between silicon/air (Δn∼2.5 at λ=1550 nm), high reflectance R>99% can be achieved with only a few mirror periods. The DBR’s consist of λ/4≈390 nm wide air trenches with silicon slabs of width W Silicon≈550 nm and thickness t Silicon≈4.7 μm. Fabrication consists of electron-beam lithography and cryogenic dry etching in SF6/O2 plasma using an ICP/RIE, resulting in smooth and vertical DBR sidewalls. Additional fabrication details can be found in refs. [10, 11]. The present devices are similar to previous static Fabry-Perot cavities , but feature aluminum contacts and a 1 μm wide etched isolation trench. The charging visible in the scanning electron micrograph in Fig. 1(a) attests to the good electrical and thermal isolation of the cavity from the rest of the silicon device layer. The measured resonances in Fig. 1(b) (E-field perpendicular to wafer plane) for a device with LC=12 μm long cavity and W=6 μm wide rib waveguide indicate high quality factor (Q=λ0/FWHM=4,584) and high finesse (F=FSR/FWHM=82), where λ0 is the resonant wavelength, FSR is the free-spectral range, and FWHM is the full-width-at-half-maximum or 3 dB bandwidth.
In Fig. 2 we show infrared (IR) images of the optical cavity. The device was tested by coupling light into and out of the rib waveguide while sweeping the wavelength using a tunable laser. The sample had a LC=12 μm long cavity with W=5 μm wide rib (Q≈4,600). An IR camera (Sensors Unlimited) and IR objective (20x) oriented perpendicularly to the sample were used to image the cavity while sweeping the wavelength. The images clearly illustrate resonance [λ=λ0=1614.5 nm, Fig. 2(a)], as indicated by the high intensity inside the cavity.
At off-resonance [λ=1615.5 nm, Fig. 2(b)] little scattered light is seen with no light at the output.
In order to tune the devices, current flows directly through the optical cavity [Fig. 1(a)] similar to the MZI heating approach in Ref. . Current flow results in Joule heating of the silicon cavity and thermo-optic resonance tuning. Our approach enables more efficient heating and tuning compared to using a thin-film heater on top of the cavity surface. This is significant for achieving fast and low power switching in devices with thick silicon layers (here t Silicon=4.7 μm), where there may be a large temperature gradient between the heater and the bottom of the silicon device layer. Low-power index tuning (Δn) and high-Q cavities enable fast switching for appropriate probe wavelengths, as will be shown.
In addition to the thermo-optic effect, our device may also enable tuning/switching via electro-optic effects, although these are known to be small in silicon. Soref et al. calculate an index change Δn≈-10-4 for an electric field E=106 V/cm (Kerr effect) . The refractive index can also be modulated by free carriers, resulting in Δn≈-10-4 for a carrier injection ΔN≈1017 . Although this latter effect can be significant, the expected free carrier injection in our measurements is ΔN<1015. Finally, the effect of free-carrier losses due to current injection is judged by observing the change in Q-factor during tuning, where any losses will lower the Q. In our measurements we did not observe any change in Q, as we will show.
3. Experimental results
For all measurements we set the polarization with the E-field perpendicular to the substrate. Lensed fibers are used to couple light to the waveguide and electrical probes supply the heating current to the device via the contact pads. Using the lensed fibers, we measured a fiber-waveguide coupling loss of αInsertion≈5 dB/facet. A typical Fabry-Perot cavity introduces an additional 5-6 dB of loss compared to an identical test waveguide with no cavity .
3.1 Thermo-optic tuning
A thermo-optic tuning measurement is shown in Fig. 3(a), and the extracted resonant peak and change in cavity index (Δn) vs. electrical power is shown in Fig. 3(b), indicating Δλ=2 nm tuning. We previously demonstrated Δλ=15 nm tuning in a low-Q cavity (Q≈700, I=2.2 mA heater current) . For our LC=12 μm long cavity, we measured a resistance R= 125 kΩ. The cavity index increases with electrical power, in agreement with silicon’s positive thermo-optic coefficient. The temperature increase is linear with ΔT=23.5 K at 42 mW power.
Silicon electro-optic effects generally decrease the cavity index with blueshift tuning [2, 13], in contrast to the observed redshift and positive Δn in Fig. 3. We observed little change in the full-width-half-maximum (FWHM) and Q-factor. Therefore, we conclude that current injected free carrier absorption losses are minimal for our measurements.
3.2 Thermo-optic attenuation
Next, we demonstrate thermo-optic attenuation. The resonances at λ= 1575.73 nm and 1604.32 nm are shown in Fig. 4(a) along with the location of suitable probe wavelengths. We measured the cavity transmittance vs. applied electrical power [Fig. 4(b)] for two probe wavelengths: probe 1=1604.32 nm (Q=4,584) and 1575.73 nm (Q=2,626). For probe 1=1604.32 nm, 10 dB attenuation was achieved for 8.4 mW electrical power, and 25 dB attenuation was achieved at 40 mW, indicating that low power, high rejection ratio switching is possible. The 25 dB extinction is limited by the Fabry-Perot cavity rejection ratio, similar to previous devices . As expected, higher Q-factor results in larger attenuation.
3.3 Thermo-optic switching
For switching, we set the wavelength to probe 1 or probe 2 and modulate the electrical power. Probe 1 is set to the resonant peak and probe 2 is set to the wavelength at which the optical power drops to -10 dB [Fig. 4(a)]. The choice of probe 1 or probe 2 affects the phase of the optical response to thermo-optic tuning. A current-voltage source (Keithley model 2400) was used along with a custom switching circuit that enabled voltage (current) pulses with 20 ns rise-/fall times. The electrical response time is thus much faster than any thermo-optic effect.
Figure 5 shows a typical switching measurement for probe 1=1604.31 nm and probe 2=1604.91 nm [defined in Fig. 4(a)], measured using a DC-125 MHz photodetector. Due to the thermo-optic redshift of the resonance, probe 2 is in phase with the applied electrical signal, while probe 1 is out of phase. The applied electrical power was an 11.5 mW square wave with a frequency f=1 kHz and 10 % duty cycle. For probe 1 the rise time (10–90 %) was t RISE, 1=1.9 μs and the fall time (90–10 %) was t FALL, 1=9.0 μs. Probe 2 resulted in t RISE, 2=7.3 μs and t FALL, 2=3.5 μs. Probe 1 exhibits a faster rise, whereas probe 2 has a faster fall time. This is the result of the different optical response with temperature for the two probes. The exponential response of the thermal time constant and the optical response for probes 1 and 2 results in the discrepancy in switching times for the probes.
The switching measurements indicate high speed and large electrical bandwidth. The high Q, however, limits the optical bandwidth. One simple method to extend the optical bandwidth is to lower the Q-factor by decreasing the DBR reflectance. Another method for increasing the optical bandwidth is to tune the cavity resonance as required; low power tuning has already been achieved (see Fig. 3), and even larger tunability is possible . Although devices with large optical bandwidth can be realized, e.g. using photonic crystal waveguides , these devices require large heating and temperature shifts for switching and therefore do not present a significant advantage over resonant devices in terms of speed and power consumption.
We model our cavity temporal response as follows. Silicon’s thermo-optic coefficient is Δn/ΔT≈+1.86×10-4/K  resulting in a redshift tuning with temperature increase (ΔT):
where n C=3.48 is the silicon cavity refractive index. The cavity temporal response is obtained by noting that Δλ is a function of the time-dependent temperature increase, ΔT:
Here T 0 is the maximum temperature increase and τ is the thermal time constant given by:
where ρSi=2.33 g/cm3 is the silicon cavity mass density, c Si=0.7 J/g-K is the specific heat, κSi=1.3 J/cm-sec-K is the thermal conductivity, and W C=50 μm is the cavity width [Fig. 1(a)].
The thermal-temporal filter response is obtained from the time dependence of ΔT(t) and the Lorentzian lineshape of the Fabry-Perot cavity transmittance:
where P 0 is the peak optical power, λProbe is the probe wavelength, FWHM is the resonance 3 dB bandwidth, and λ0(ΔT) is the tuned resonance. For a calculated thermal time constant τ=3.2 μs we find excellent agreement between measurement and experiment (Fig. 5). Although probes 1 and 2 exhibit different rise/fall times, the thermal time constant, τ, remains constant since it is determined by device geometry and not probe wavelengths.
One method to improve the device response is by overdriving (Fig. 6). We apply an electrical pulse with power significantly larger than the 8.4 mW required for -10 dB switching in Fig. 4(b), resulting in t RISE=640 ns for P=58.4 mW power (probe 1=1604.30 nm). The fall time, however, is increased to t FALL=22.4 μs compared to Fig. 5, since over-driving results in increased heating and hence a longer cooling time. Our thermal model with time constant τ=3.2 μs still gives accurate results for the rise time, but underestimates the fall time. This is likely the result of contact pad heating at high electrical powers, which our model does not take into account. Although the decreased rise time comes at the expense of an increased fall time, a two-step actuation can be used, as in Ref. . A short high-voltage pulse followed by a low-voltage signal enables fast rise time [Fig. 6(a)], without increasing the fall time.
A number of other improvements can be made to the present devices. Higher Q-factors enable switching at lower powers with higher extinction. For example, in Fig. 4(b) the electrical power required to obtain -10 dB attenuation (switching) is 8.4 mW for probe 1=1604.32 nm (Q=4,584) and increases to 17.7 mW for probe 1=1575.73 nm (Q=2,626). A second approach makes use of a two-voltage step actuation to increase switching speed, as previously mentioned. Another design change is to reduce the cavity thermal time constant, τ=ρSi c Si W C 2/(κπ2), by reducing the cavity width, W C. For example, by shortening W C from the present 50 μm to 25 μm (10 μm) we decrease the thermal time constant to τ=800 ns (130 ns). Therefore, MHz-range thermal modulation appears possible. Finally, our high-Q cavities may also enable fast electro-optic (EO) switching  approaching GHz-range speeds. Although silicon EO effects are small , high-Q cavities enable long effective optical interaction lengths and large phase shifts in compact devices.
We demonstrated a compact thermo-optically tunable SOI waveguide Fabry-Perot microcavity. Thermo-optic tuning results in Δλ>2 nm (<50 mW electrical power). By appropriate choice of probe wavelength, thermo-optic attenuation (-25 dB) and optical switching was demonstrated with high speed (640 ns) and low power (<10 mW). Modeling enables accurate prediction of the temporal behavior. Design improvements were also discussed, potentially enabling MHz-range thermo-optic switching at mW-level powers.
The authors thank D. Park, D. King, J. W. Baldwin and K. Bussmann for assistance with device processing. The devices were fabricated in the Nanoscience Institute (NSI) cleanroom facility at NRL. This research was performed while M. Pruessner held a National Research Council (NRC) Research Associateship Award at the Naval Research Laboratory (NRL).
References and links
1. G. Cocorullo and I. Rendina, “Thermo-optical modulation at 1.5 μm in silicon etalon,” Electron. Lett. 28, 83–85 (1992). [CrossRef]
2. R. A. Soref and B. R. Bennett, “Electro optical effects in silicon,” IEEE J. Quantum. Electron. QE-23, 123–129 (1987). [CrossRef]
3. R. L. Espinola, M.-C. Tsai, James T. Yardley, and R. M. Osgood, “Fast and low-power thermo optic switch on thin silicon-on-insulator,” IEEE Photon. Technol. Lett. 15, 1366–1368 (2003). [CrossRef]
4. Y. Li, J. Yu, and S. Chen, “Rearrangeable nonblocking SOI waveguide thermo optic 4×4 switch matrix with low insertion loss and fast response,” IEEE Photon. Technol. Lett. 17, 1641–1643 (2005). [CrossRef]
5. M. Harjanne, M. Kapulainen, T. Aalto, and P. Heimala, “Sub-μs switching time in silicon-on-insulator Mach-Zehnder thermo optic switch,” IEEE Photon. Technol. Lett. 16, 2039–2041 (2004). [CrossRef]
6. M. W. Geis, S. J. Spector, R. C. Williamson, and T. M. Lyszczarz, “Submicrosecond submilliwatt silicon-on-insulator thermooptic switch,” IEEE Photon. Technol. Lett. 16, 2514–2516 (2004). [CrossRef]
7. I. Kiyat, A. Aydinli, and N. Dagli, “Low-power thermo optical tuning of SOI resonator switch,” IEEE Photon. Technol. Lett. 18, 364–366 (2006). [CrossRef]
8. C. A. Barrios, V. R. Almeida, R. R. Panepucci, B. S. Schmidt, and M. Lipson, “Compact silicon tunable Fabry-Pérot resonator with low power consumption,” IEEE Photon. Technol. Lett. 16, 506–508 (2004). [CrossRef]
9. H. M. Chong and R. De La Rue, “Tuning of photonic crystal waveguide microcavity by thermooptic effect,” IEEE Photon. Technol. Lett. 16, 1528–1530 (2004). [CrossRef]
10. M. W. Pruessner, W. S. Rabinovich, T. H. Stievater, D. Park, and J. W. Baldwin, “Cryogenic etch process development for profile control of high aspect-ratio sub-micron silicon trenches,” J. Vac. Sci. Technol. B 25, 21–28 (2007). [CrossRef]
12. M. W. Pruessner, T. H. Stievater, and W. S. Rabinovich, “Tunable Fabry-Perot waveguide microcavities with high index contrast mirrors, in Conference on Lasers and Electro-Optics/Quantum Electronics and Laser Science and Photonic Applications Systems Technologies, Technical Digest (CD) (Optical Society of America, 2007), paper CThP3.
13. B. Schmidt, Q. Xu, J. Shakya, S. Manipatruni, and M. Lipson, “Compact electro-optic modulator on silicon-on-insulator substrates using cavities with ultra-small modal volumes,” Opt. Express 15, 3140–3148 (2007). [CrossRef] [PubMed]
14. M. Tinker and J. -B. Lee, “Thermal and optical simulation of a photonic crystal light modulator based on the thermo-optic shift of the cut-off frequency,” Opt. Express 13, 7174–7188 (2005). [CrossRef] [PubMed]