Silicon waveguides are now widely used to guide radiation in the near-infrared, mainly in the wavelength range of 1.1 – 2.2 µm. While low-loss waveguides at longer wavelengths in silicon have been proposed, experimental realization has been elusive. Here we show that single-mode integrated silicon-on-sapphire waveguides can be used at mid-infrared wavelengths. We demonstrate waveguiding at 4.5 µm, or 2222.2 cm−1, with losses of 4.3 ± 0.6 dB/cm. This result represents the first practical integrated waveguide system for the mid-infrared in silicon, and enables a range of new applications.
©2010 Optical Society of America
It has recently been shown that silicon waveguides can be used to construct all of the components of a photonic data transmission system on a single chip [1,2]. These components can be integrated together with CMOS electronics to create complex electronic-photonic integrated circuits . High field confinement silicon waveguides enable exciting new applications, from chip-scale nonlinear optics  to biosensors  and light-force activated devices . To date, most of the experiments in silicon waveguides have been at wavelengths in the near-infrared, ranging from 1.1 to 2.2 µm [7,8]. Here we show that single-mode integrated silicon-on-sapphire waveguides can be used at mid-infrared wavelengths, in particular at 4.5 µm, or 2222.2 cm−1, with losses of 4.3 ± 0.6 dB/cm. This idea has been proposed in theoretical literature, but experimental realization has been elusive . This result represents the first practical integrated waveguide system for the mid-infrared in silicon, and enables a range of new applications.
When building telecommunications systems, it is vital to operate at near-infrared wavelengths in order to maintain compatibility with existing systems. Many other types of optical systems operate in the near-infrared as a matter of convenience: A wide variety of commercial optical components are readily available at 1310, 1480 and 1550 nm, largely as a spinoff of the extensive commercial work in telecommunications. But these wavelengths, while convenient for telecommunications, are obviously not suitable for all applications. It is often necessary to manipulate light with wavelengths in the 2-20 µm (mid-infrared) regime, for a variety of applications: A few prominent examples consist of thermal imaging (2.5 µm to 15 µm wavelengths) , chemical bond spectroscopy  (which often spans from the visible to 20 µm and beyond), astronomy , gas sensing , and military applications such as missile countermeasures .
Historically, the mid-infrared has been a problematic region for photonics. Coherent sources have been bulky and expensive , or have required cryogenic cooling , as did common mid-infrared detectors . Of course, the lack of integrated optical waveguides at these long wavelengths meant that mid-infrared systems were implemented using free-space optics. Recently, the landscape has begun to change dramatically. Inexpensive, reliable, single-mode quantum cascade lasers are now available commercially all the way to 9 µm wavelengths, with powers from 10 to 100 mW, offering near room-temperature operation in most cases [18,19]. Single-mode fibers are now available at wavelengths out to 6 µm [20,21], as are mid-infrared photodetectors with bandwidths over 1 GHz . As a result, building a single-mode optical system in the mid-infrared is now within the financial and technical reach of any modestly well-funded research group or small company. In fact, it is possible to build a full test system with decent performance for well under $100,000, which is comparable to the cost of many swept-wavelength near-infrared systems. This combination of new capabilities strongly suggests that mid-infrared photonics is poised to take off as a field in the immediate future.
But there are many missing pieces. High bandwidth modulators do not exist for these wavelengths. Neither are there low-loss splitters, tunable filters, or any of the rest of the building blocks of complex fiber-coupled photonic systems. Silicon waveguides provide an ideal platform for us to build all of these components. Furthermore, the silicon photonic platform offers the opportunity to integrate all of these components along with control electronics on the same substrate – opening the possibility of building, for instance, FTIR systems-on-a-chip with multi-centimeter path lengths but nanoliter sample volume. The first step on this path is to construct low-loss silicon waveguides.
2. Silicon-on-sapphire material system and waveguide design
We chose to build waveguides using the silicon-on-sapphire (SOS)  materials system, which is used in the electronics industry as an alternative to silicon-on-insulator (SOI). SOS is particularly desirable for this application because of the lack of a high index substrate, which eliminates the issue of substrate leakage. The resistivity of the silicon for the wafers used was specified by the manufacturer to be 100 Ω-cm, suggesting that optical loss due to free-carriers will be minimal . Mid-infrared guides at longer wavelengths could also be built using free-standing silicon guides  or germanium-on-silicon. SOS, in particular, has the advantage of offering the ability to build high-confinement, fully etched waveguides from 1.1 µm all the way to around 6.2 µm  – over two octaves of bandwidth, including the telecommunications region, while maintaining electronics compatibility. While SOI has recently been used to build waveguides at 2.2 µm , substrate leakage will become an increasingly large problem at longer wavelengths. Another problem is that silica becomes extremely lossy at wavelengths longer than 4 µm.
The SOS waveguides were simulated using a Yee-grid based eigensolver . It was found that a 1.8x0.6 µm ridge waveguide offered a small mode (around 1.1 µm2), shown in Fig. 1 , a tight bend radius (FDTD simulations predicted 10 µm with negligible losses), and allowed the silicon etch to stop on the sapphire, simplifying fabrication. The mode size is around 1/19 of a square wavelength in free space at 4.5 µm. At this wavelength, the TM0 mode was expected to either not guide or be weakly confined, since it is near the theoretically predicted cutoff. Higher order modes are predicted to not be supported. The devices are terminated in waveguides with widths of 8 µm, which were coupled into the 1.8 µm waveguides with a taper. This configuration was used to increase coupling efficiency. Predicted coupling losses to the 1.8 µm wide guide from the single-mode mid-infrared fibers that we used were on the order of 18 dB, including the loss from the taper.
We designed a chip with a series of waveguide bends. Each waveguide started and stopped with an 8 µm wide region, which extended for 5 mm on the chip. These two regions were at 90 degrees, for the purpose of minimizing the noise due to light that might scatter directly from the source to the detector. Figure 2 shows a more detailed picture of the layout of the chip, including the two planes on which the chip was cleaved. Both cleave planes were cleaved by hand orthogonal to the waveguides. The distance from the cleave to the start of the taper was approximately 2.5 mm. Since it was identical for all waveguides, the insertion loss due to this region should be nearly identical for all measurements.
Some of the devices present on the chip were not used for waveguide loss measurements; the ones that were used are marked in Fig. 2. It was possible to identify which device was guiding light by observing the location of the fiber. The bend radii used were 40 µm, well in excess of the minimum bend radius of 10 µm suggested by FDTD simulations.
The waveguides were fabricated using standard semiconductor processing techniques. Epitaxial silicon-on-sapphire wafers (100 mm diameter) were used as starting material. Patterning was accomplished using an electron beam lithography system and HSQ resist on a wafer fragment. The resist was developed and the chip was etched using a CF4 plasma in an Oxford Instruments parallel-plate RIE. Resist was first removed using just a wet resist remover, although some resist residue appeared to remain on the surface of the waveguides due to incomplete stripping. Further processing then involved cleaning the chip with a piranha etch, which appeared to improve waveguide loss dramatically, as shown in the following section. It is likely that the piranha etch removed some remnants of the resist. The chip was finally manually cleaved through the 8 µm wide waveguide segment, leaving an optical quality edge.
The devices were tested using an Ekspla PL2241 Nd:YAG laser that drove an Ekspla PG501/DFG optical parameteric generator/difference frequency generator (OPG/DFG). The OPG/DFG is capable of producing linearly polarized light from 2 to 9 µm, with 4.5 µm radiation used in our testing. We restricted our testing to this wavelength as the fiber cut off at longer wavelengths, and we were unable to achieve as great a dynamic range at other wavelengths, due to decreased emission power and beam stability from the OPG/DFG. We intend to explore the behavior of our waveguides over a range of wavelengths in future work. The laser was coupled through a polarizer into ZnSe lens with 12.7 mm focal length, and into a 9/125 µm single-mode optical fiber S009S17 from IR Photonics. The fiber, chip and detector were held on stages with piezoelectric actuated micrometers (Newport PZA12). The output of the chip was coupled directly to free space and then into a PVI-5 detector from Vigo Systems. Using a fiber for input allowed us to easily determine visually which device we were coupling radiation into, as both the fiber position and the device locations could be identified with an inspection scope.
Because the OPG/DFG provided 30 ps pulses of IR light at a repetition rate of 50 Hz with pulse energies around 150 µJ, we used a boxcar amplifier (Stanford Research SR250) to reject the signal during the times when the laser was not providing output. The signal to noise ratio was further enhanced by mechanically chopping the laser at 2 Hz, and using a Signal Recovery 7265 lock-in amplifier to detect the 2 Hz modulated signal. This resulted in an overall signal to noise ratio of 85 dB. Coupling from the free-space mode into the single-mode fiber was achieved with around 12 dB of insertion loss, leaving adequate dynamic range to perform measurements on the SOS waveguides.
In the process of testing the chip, we mechanically swept a single-mode fiber across one of the faces of one chip. We found a transmission pattern that closely matched the lithographically defined waveguide pattern, as shown in Fig. 3 . By measuring a series of devices with different waveguide lengths on a chip, it was possible to determine the overall waveguide loss, and to separate this from the loss of the 8 µm waveguide regions and the insertion loss in getting the light onto and off of the chip. Devices that were tested ranged in total length from 1 mm to 1.4 cm, not including the input tapered regions, which were identical on all devices. The waveguide losses on the best chip were found to be 4.3 ± 0.6 dB/cm. A bend radius of 40 µm was used for the devices. We also included two control structures with waveguide widths of 1.2 µm (as opposed to the 1.8 µm devices, which were shown to guide), identical to other devices in all other respects. As predicted theoretically, the narrower devices did not show any transmission at all.
The light emerging from the OPG/DFG was linearly polarized with a TE polarization, that is, the optical field was parallel to the plane of the chip. To ensure that the polarization of the pulses was constant during measurements, a linear polarizer was placed in the optical path directly before the lens and the fiber, as shown in Fig. 3. Unfortunately, it was not possible to perfectly control the input polarization to the waveguides; a non-polarization maintaining fiber was used, due to the lack of a commercially available polarization-maintaining fiber at these wavelengths. We measured the polarization of the light emerging from the fiber, and found it to be TE polarized; it is likely that this was the polarization state at the output of the fiber when the waveguides were tested. However, because the fiber was not polarization maintaining, it is impossible to be sure that this polarization state was uniformly maintained. It is likely that only the TE0 mode guides a meaningful amount of optical power, in light of the fact that the TM0 mode is so close to cutoff, and the relatively consistent results seen with initially TE polarized light entering the fiber. However, the possibility of some power propagating in the TM0 mode cannot be excluded completely; we would observe that in any case, single-mode waveguides are typically defined as supporting the TE0 and TM0 modes [28,29]. We show the normalized optical power as a function of waveguide length for a series of devices measured on several different chips in Fig. 4 .
For each device, generally eight measurements were taken at around five-second intervals. These measurements were averaged to produce the final value, while the uncertainty was determined by observing the root mean squared deviation from the mean. Typically, the standard deviation was on the order of 5%. This estimation of error is shown on the final plots of the insertion loss versus waveguide length in Fig. 4. The position of the detector was also optimized for each device measured, in order to maximize the amount of light captured from the output of the waveguide. It was found that the optimal position for the detector moved as expected as waveguides on different portions of the chip were tested.
We tested the best-performing chip multiple times, in one instance testing the shortest devices first, and in another testing the longer devices first; this was an attempt to expose any systemic errors that our test setup might have had. We found that the waveguide losses were in agreement, with the second measurement predicting losses of 4.9 ± 0.6 dB/cm. A second chip that had been piranha cleaned also exhibited waveguide losses of 4.7 ± 0.6 dB/cm that were in statistical agreement with the best chip. A chip that was not piranha etched exhibited substantially higher losses of 9.6 ± 1.2 dB/cm.
As is not surprising, the uncertainty for each waveguide loss measurement exceeded what would be expected solely on the basis of individual device measurement uncertainties. This is due to the fact that there are additional uncertainties beyond the fluctuation on OPG/DFG power; these would include the coupling loss for each endfacet, the random fabrication defects on each device, and a number of other factors. Crucially, however, the final uncertainties due to the linear regression, estimated based on the RMS error between the data points and the fit, are easily small enough for the data to be considered statistically meaningful. It is worth noting here that the linear dependence of the insertion loss on device length suggests that there cannot be a large level of nonlinear loss in the waveguides, as might be caused by a multi-photon absorption mechanism.
To determine the noise floor, the fiber was moved laterally out of alignment with a waveguide, until the power observed on the detector was flat as the fiber moved; the average value for the power reading was then noted. This value turned out to be around 20 dB lower than the typical peak transmission observed for the shortest waveguide bend on the best-performing chips. This should be very adequate for a conclusive measurement of the low level of loss that we obtained. This dynamic range is also evident in Fig. 3. The pulse energy at the detector for the peak power level was approximately 100 pJ after the device insertion losses. In the case of the chip with higher losses, shown in Fig. 4 (d), the shortest device exhibits optical power only 14 dB from the noise floor. As a result, the longest waveguide bend measurement is close to the noise floor, and has a larger uncertainty. But for the other chips shown, more than 10 dB remain between the response of even the longest device and the noise floor. Based on this, the total insertion loss from the fiber to the detector for the shortest devices is 50 dB. The theoretical loss that would be present for coupling from the fiber to a waveguide, and back to a fiber is around 36 dB. The likely discrepancy is due to higher losses coupling to a detector than a fiber, and additional losses from imperfections on the device endfacets. Small alignment errors also likely play a role.
Lower waveguide losses should be possible with improved processing. Based on the material losses of silicon and sapphire, an ultimate loss of around 0.1 dB/cm should be possible at 4.5 µm. We suspect the presently observed losses are due to surface roughness, though it is also possible that surface states could participate in the loss mechanisms .
With low-loss waveguides, it will be possible to build integrated mid-infrared lasers and detectors using techniques such as wafer bonding  and selective-area growth, and to construct a wide variety of further devices within the silicon platform. It should also prove possible to build high-confinement integrated nonlinear optical devices, such as integrated OPOs and difference frequency generators, because these wavelengths are so long that more than several photons are required to reach the silicon bandgap energy. Another interesting opportunity emerges from the limits of lithography: A 20 nm trench  represents a small fraction of a wavelength at 1.55 µm, but represents a significantly smaller fraction for light at 4.5 µm; mid-infrared waveguides may be an ideal ‘playground’ for exploring ideas in ultra-subwavelength photonics. In the long run, we anticipate integrated mid-infrared optical systems, which will be much smaller and cheaper than contemporary systems.
The authors would like to acknowledge use of the facilities at the Cornell Nanoscale Facility for the device fabrication portion of this work. In addition, they would like to thank Gernot Pomrenke for his support through the AFOSR YIP program, the NSF STC MDITR Center at the University of Washington, Tektronix Corporation, and the Murdock Foundation. M. Hochberg would like to thank Marko Loncar and Richard Soref for valuable discussions, and Phil Reid for his assistance.
References and links
1. A. Liu, L. Liao, D. Rubin, J. Basak, Y. Chetrit, H. Nguyen, R. Cohen, N. Izhaky, and M. Paniccia, “Recent development in a high-speed silicon optical modulator based on reverse-biased pn diode in a silicon waveguide,” Semicond. Sci. Technol. 23(6), 064001 (2008). [CrossRef]
3. A. Huang, C. Gunn, G.-L. Li, Y. Liang, S. Mirsaidi, A. Narasimha, and T. Pinguet, “A 10Gb/s Photonic Modulator and WDM MUX/DEMUX Integrated with Electronics in 0.13μm SOI CMOS,” in International Solid-State Circuits Conference (IEEE, 2006), pp. 24–25.
4. C. Koos, P. Vorreau, T. Vallaitis, P. Dumon, W. Bogaerts, R. Baets, B. Esembeson, I. Biaggio, T. Michinobu, F. Diederich, W. Freude, and J. Leuthold, “All-optical high-speed signal processing with silicon–organic hybrid slot waveguides,” Nat. Photonics 3(4), 216–219 (2009). [CrossRef]
5. A. Ramachandran, S. Wang, J. Clarke, S. J. Ja, D. Goad, L. Wald, E. M. Flood, E. Knobbe, J. V. Hryniewicz, S. T. Chu, D. Gill, W. Chen, O. King, and B. E. Little, “A universal biosensing platform based on optical micro-ring resonators,” Biosens. Bioelectron. 23(7), 939–944 (2008). [CrossRef]
7. M. Lipson, “Guiding, Modulating and Emitting Light on Silicon – Challenges and Opportunities,” IEEE J. Lightwave Tech. 23(12), 4222–4238 (2005). [CrossRef]
8. X. Liu, R. M. Osgood, Y. A. Vlasov, and W. M. J. Green, “Broadband mid-infrared parametric amplification, net off-chip gain, and cascaded four-wave mixing in silicon photonic wires,” in Group IV Photonics (IEEE, 2009).
9. R. A. Soref, S. J. Emelett, and W. R. Buchwald, “Silicon waveguided components for the long-wave infrared region,” J. Opt. A, Pure Appl. Opt. 8(10), 840–848 (2006). [CrossRef]
10. G. C. Holst, and S. W. McHugh, “Review of thermal imaging system performance,” in Proceedings of SPIE (SPIE 1992), pp. 78–84.
11. H. B. Gray, Chemical Bonds: An Introduction to Atomic and Molecular Structure (University Science Books, 1994).
13. P. Werle, F. Slemr, K. Maurer, R. Kormann, R. Mucke, and B. Janker, “Near- and mid-infrared laser-optical sensors for gas analysis,” Opt. Lasers Eng. 37(2-3), 101–114 (2002). [CrossRef]
14. The Federation of American Scientists, “AN/AAQ-24 Directional Infrared Countermeasures (DIRCM),” http://www.fas.org/man/dod-101/sys/ac/equip/an-aaq-24.htm.
15. F. Rotermund, V. Petrov, and F. Noack, “Difference-frequency generation of intense femtosecond pulses in the mid-IR using HgGa2S4 and AgGaS2,” Opt. Commun. 185(1-3), 177–183 (2000). [CrossRef]
16. M. Tacke, “Lead-salt lasers,” Philosophical Transactions: Mathematical, Physical and Engineering Sciences 359(1780), 547–566 (2001). [CrossRef]
17. J. Piotrowski and A. Rogalski, “New generation of infrared photodetectors,”, Sensors Act. A 67(1-3), 146–152 (1998). [CrossRef]
18. Alpes Lasers, “Lasers In Stock,” http://www.alpeslasers.com/lasers-on-stock/index.html.
19. A. Lyakh, C. Pflugl, L. Diehl, Q. J. Wang, F. Capasso, X. J. Wang, J. Y. Fan, T. Tanbun-Ek, R. Maulini, A. Tsekoun, R. Go, and C. K. N. Patel, “1.6 W high wall plug efficiency, continuous-wave room temperature quantum cascade laser emitting at 4.6 µm,” Appl. Phys. Lett. 92, 111110 (2008). [CrossRef]
20. IRPhotonics, “Single Mode Infrared Fiber,” http://www.iguide-irphotonics.com/en/products/single-mode-infrared-fiber.html.
21. E. M. Dianov, “Single-Mode As-S Glass Fibers,” Inorg. Mater. 39(6), 627–630 (2003). [CrossRef]
22. Boston Electronics, “Infrared Detectors,” http://www.boselec.com/products/detir.html.
23. R. A. Johnson, P. R. de la Houssaye, C. E. Chang, P. F. Chen, M. E. Wood, G. A. Garcia, I. Lagnado, and P. M. Asbeck, “Advanced Thin-Film Silicon-on-Sapphire Technology: Microwave Circuit Applications,” IEEE Trans. Electron. Dev. 45(5), 1047–1054 (1998). [CrossRef]
24. R. Soref and B. Bennett, “Electrooptical Effects in Silicon,” IEEE J. Quantum Electron. 23(1), 123–129 (1987). [CrossRef]
25. P. Y. Yang, S. Stankovic, J. Crnjanski, E. J. Teo, D. Thomson, A. A. Bettiol, M. B. H. Breese, W. Headley, C. Giusca, G. T. Reed, and G. Z. Mashanovich, “Silicon photonic waveguides for mid- and long-wave infrared region,” J. Mater. Sci. Mater. Electron. 20(S1), 159–163 (2009). [CrossRef]
26. E. D. Palik, Handbook of Optical Constants of Solids (Elsevier, 1998).
27. A. Taflove, and S. C. Hagness, Computational Electrodynamics: The Finite-Difference Time-Domain Method (Artech House, 2005).
28. R. Hu, D. Dai, and S. He, “A Small Polymeric Ridge Waveguide With a High Index Contrast,” IEEE J. Lightwave Tech. 26(13), 1964–1968 (2008). [CrossRef]
29. R. A. Soref, J. Schmidtchen, and K. Petermann, “Large Single-Mode Rib Waveguides in GeSi-Si and Si-on-SiO2,” IEEE J. Quantum Electron. 27(8), 1971–1974 (1991). [CrossRef]
31. A. W. Fang, M. N. Sysak, B. R. Koch, R. Jones, E. Lively, Y. H. Kuo, D. Liang, O. Raday, and J. E. Bowers, “Single-Wavelength Silicon Evanescent Lasers,” IEEE J. Sel. Top. Quantum Electron. 15, 535–544 (2009). [CrossRef]