Ultrafast and selective reduction of sidewall roughness in silicon waveguides using self-perfection by liquefaction

Transcription

1 IOP PUBLISHING Nanotechnology 20 (2009) (5pp) NANOTECHNOLOGY doi: / /20/34/ Ultrafast and selective reduction of sidewall roughness in silicon waveguides using self-perfection by liquefaction Qiangfei Xia, Patrick F Murphy, He Gao and Stephen Y Chou NanoStructure Laboratory, Department of Electrical Engineering, Princeton University, Princeton, NJ 08544, USA chou@princeton.edu Received 13 May 2009, in final form 9 July 2009 Published 4 August 2009 Online at stacks.iop.org/nano/20/ Abstract We use a novel technique, self-perfection by liquefaction (SPEL), to smooth the rough sidewalls of Si waveguides. An XeCl excimer laser with 308 nm wavelength and 20 ns pulse duration is used to selectively melt the surface layer of the waveguide. This molten layer flows under surface tension and this results in smooth sidewalls upon resolidification. Our experimental results show that this technique reduces the average sidewall roughness (1σ )from13to3nm. Our calculations show that the waveguide transmission loss due to sidewall roughness in these waveguides would be reduced from 53 to 3 db cm 1, an improvement with light transmission five orders of magnitude greater. Due to a low viscosity of molten Si (below water), SPEL can be achieved on a Si surface within 100 ns. This short time, together with SPEL s material selectivity, makes it possible to repair defective components on a chip without damaging surrounding components and materials, making SPEL a promising candidate for defect repair in integrated optics and nanophotonics. (Some figures in this article are in colour only in the electronic version) 1. Introduction Integrating optical components on a chip with high packing density using existing Si fabrication technology is currently under extensive research [1]. Silicon waveguides are attractive for guiding optical signals for several reasons. First, the high-index difference between Si and surrounding materials (e.g., n is approximately 2 for Si/SiO 2 ) results in low losses at waveguide bends, and thus allows for denser packing of integrated optical components. Second, current Si planar technology makes fabrication of nanoscale waveguides (and thus nanoscale optical systems) easy to achieve. However, as device feature sizes shrink, geometric defects introduced in the fabrication process will become increasingly prominent and will seriously degrade light propagation properties. Sidewall roughness of a waveguide is one of the typical defects caused by fabrication. Roughness can greatly increase light scattering, leading to significant light transmission loss. Such scattering becomes more severe in narrow waveguide, since a significant portion of propagating wave leaks out of the waveguide. The fabrication caused surface roughness is inevitable since it is due to the intrinsic noise in the lithography and etching processes. Many efforts have been devoted to reduce the sidewall roughness to improve the performance of optical components. These methods include anisotropic wet etching [2, 3], thermal oxidation [4], and thermal oxidation in combination with wet etching [3, 5]. However, these methods are either limited to certain crystalline facets of semiconductor materials, or often involve harsh processing conditions, high temperatures or complicated steps. For example, anisotropic wet etch only applies to certain crystalline semiconductor materials, particularly (100) and (110) silicon; thermal oxidation is incompatible with materials which cannot withstand high temperatures. More importantly, these roughness reduction methods are not selective, which means that other components and/or materials on the same chip that are not supposed to be processed at the same time will be affected. As integrated circuit technology advances, more components and materials that serve different functions will be integrated into one single chip, each requiring different, /09/ $ IOP Publishing Ltd Printed in the UK

2 perhaps incompatible, processing procedures. This requires a fabrication method with high selectivity to prevent damaging surrounding components. Previously, CO 2 lasers have been used to smoothing optical components such as toroid-shaped microresonators [6] and a silicon rich oxide waveguide [7], however, the relatively long processing time and the long wavelength (10.6 μm) will introduce adverse thermal effects on underneath materials and/or surrounding components. In this paper, we report a novel method that can repair the sidewall roughness of Si waveguides. The method, coined self-perfection by liquefaction (SPEL) [8], repairs the defects of a structure by selectively and rapidly heating the structure into a molten state, at which point the material flows under surface tension to remove the defects. Using SPEL, we reduced the average sidewall roughness of a silicon waveguide from 13 to3nm(1σ ), leading to five orders of magnitude increase in light transmission for a 1 cm long waveguide. This process is finished within 100 ns, and only the surface layer is melted, while other parts of the chip remains at a low temperature without being affected. 2. Experiment Both microscale waveguides and nanoscale lines were fabricated. For microscale waveguides fabrication, it began with a bonded silicon on insulator (SOI) wafer which has a 1.5 μm thick silicon device layer on a 1 μm thick buried oxide layer. First, a thin layer of thermal SiO 2 was grown on the Si surface. HMDS and AZ5214 photoresist were spin-coated on the wafer at a spinning speed of 4000 rpm, followed by a soft bake on a 95 C hot plate for 1 min. Waveguides with a width ranging from 4 to 10 μm were defined on the photoresist using a photomask and an optical contact lithography aligner. After development, a 20 nm thick Cr layer was evaporated using an electron beam evaporator, followed by liftoff in warm acetone. The top SiO 2 layer was etched through using fluorine based reactive ion etch (RIE), while the silicon layer was etched using chlorine base RIE. After RIE, the wafer was briefly dipped in diluted HF solution to remove the top oxide layer and to lift off the Cr. The fabricated waveguides were cleaned at 70 C in a solution of 1 part of NH 4 OH, 1 part of H 2 O 2 and 5 parts of deionized H 2 O for 15 min (RCA #1). For nanoscale lines, the fabrication procedures were similar to those for microscale waveguides except that nanoimprint lithography [9] was used instead of photolithography for patterning, and the substrate used was a SIMOX (separation by implantation of oxygen) SOI wafer with 190 nm Si device layer and 380 nm thick buried oxide layer. SPEL was carried out using an XeCl excimer laser with 308 nm wavelength and 20 ns pulse duration. The laser spot is about 3 mm 3 mm, and the laser pulse energy can be changed by adjusting a variable attenuator. A continuous wave HeNe laser (λ = 633 nm) used for alignment was incident with an angle on the sample at the spot which is going to be exposed to the excimer laser. In order to measure the molten time during which the smoothing took place, the reflected HeNe laser signal from the sample was collected by a high speed Si pin photodetector (Newport, 818-BB-21A) and sent to an Figure 1. (a) A 4 μm wide Si waveguide on SiO 2 with rough sidewalls was smoothed, (b) after exposure to 20 laser pulses of 900 mj cm 2 at a repetition rate of 1 Hz. oscilloscope (Tektronix, TDS-220) for analysis. Before and after SPEL, the Si waveguides were imaged using a scanning electron microscope (SEM). And the SEM images were used for line edge roughness analysis. 3. Results For microscale waveguide, we use multiple pulses for best results. For example, the rough side wall of a 4 μm wide waveguide (figure 1(a)) was smoothed out after exposure to 20 pulses (900 mj cm 2 for each pulse) with a repetition rate of 1 Hz (figure 1(b)). For nanoscale lines, a single laser pulse is enough to smooth the sidewall. Figure 2(a) is the as-etched 250 nm wide Si lines on SiO 2. After SPEL using a single pulse of 490 mj cm 2, the sidewalls are smoothed (figure 2(b)). It should be noticed that in both figures 1 and 2, the cross-section of the Si waveguides are changed from square to hemisphere. This might be helpful in optical coupling between the Si waveguides and round optical fibers. However, in the case when the square profile needs to be preserved, we have succeeded in fabricating flat-top Si waveguides using capped SPEL (C-SPEL) by preventing the laser from directly heating the waveguide top surface or by holding the molten top surface using a blank plate [8]. The roughness was quantified using a digitized analysis from SEM images using digital image processing software. We obtained high magnification top-view SEM images of several processed and unprocessed waveguides. We then analyzed 2

3 Figure 3. Roughness profile along the length direction of a 4 μm wide Si waveguide on SiO 2 before (a) and after (b) self-perfection (SPEL). The 1σ line edge roughness in (a) and (b) are 13 and 3 nm, respectively. Figure nm wide Si gratings on SiO 2 with rough sidewall (a) were smoothed upon a single laser pulse exposure (b) at a laser fluence of 490 mj cm 2. The gratings were made on an SOI wafer with a 190 nm thick Si layer on a 380 nm thick SiO 2 layer. these digital images with a simple edge detection algorithm implemented in Matlab. For a 4 μm wide Si waveguide, the sidewall profiles are shown in figure 3. It is found that the 1σ roughness was reduced from 13 to 3 nm. In order to estimate the effect of roughness on the light propagation, we calculated the light transmission loss in a 500 nm wide, 200 nm high Si waveguide as a function of different sidewall roughness and roughness autocorrelation lengths. We use a well-known model [10] for calculating the effect of the sidewall roughness characteristics on scattering loss that assumes that the roughness has an exponential autocorrelation function, an assumption that is claimed to be justifiable theoretically and experimentally [3, 11]. Figure 4(a) shows that the waveguide transmission loss reduces with smaller autocorrelation length and smoother surface. Figure 4(b) plots the loss as a function of waveguide sidewall roughness for a 200 nm high Si waveguide with a width of half and quarter wavelength used in telecommunication (i.e., 1.55 μm), assuming a roughness autocorrelation length of 5 nm. It indicates that for a 220 nm wide waveguide, a reduction of roughness from 13 to 3 nm can result in a decrease of propagation loss from 53 to 3 db cm 1. This means the transmitted power could increase by 5 orders of magnitude. To study the time that Si is in a molten state ( molten time ) for a successful SPEL, we measured the Si molten time as the function of laser fluence, using a time-resolved reflectivity technique (figure 5). This technique is based on the fact the molten Si has a higher reflectivity than its solid phase [12], and thus the signal of a second laser beam (HeNe is used here) reflected from the Si is greater when the Si is molten. As shown in figure 5, the molten time is a linear function of laser fluence, and is ranging from 15 to 220 ns in our experiments depending on the substrate. Comparing the smoothing results in figure 2, it shows that 50 ns molten time is sufficient for smoothing out typical edge roughness in a nanoscale waveguide. Although the laser spot is only 3 mm 3mminsize, larger coverage area has been achieved by a step and repeat exposure system [13]. In that study, an overlap of about 200 μm was used between each pulse for full area coverage. Our experimental results showed that smoothing results in the overlapping area has no difference with those which have experienced only a single pulse [13]. This fact suggests that no rigorous alignment is required in SPEL for large area and makes the process simple to implement. Also, as long as the laser fluence is high enough to melt the surface layer, there is a fluence window of about 25% for the nanoscale Si gratings. This finding indicates that SPEL has a high tolerance for energy fluctuation from pulse to pulse which is usually a problem for gas lasers. 4. Discussion Comparing our waveguide smoothing technique with others, it can be found that there are several advantages of SPEL. First, this process is ultrafast, usually takes less than 200 ns to finish for Si. Second, this process is highly selective. Both are essential in order to repair the sidewall roughness without 3

4 (a) (b) Figure 5. Measured molten time as a function of laser fluence for Si and SOI. The SOI wafer has a buried oxide layer of 380 nm and Si device layer of 190 nm. Figure 4. (a) Calculated waveguide loss contours for a Si waveguide of 200 nm high and 500 nm wide. (b) Calculated waveguide transmission loss versus the roughness for Si waveguides with a width of half (220 nm) and quarter (110 nm) wavelength. In this calculation, the waveguides are 200 nm high with an autocorrelation length of 5 nm. The free space wavelength used is 1.55 μm. damaging or destroying other elements or materials on a chip. Third, this technique can be used for large area by a step and repeat exposing system. The ultrafast speed of SPEL (<200 ns) is due to the fact that molten Si has a low viscosity and reasonably high surface tension. SPEL is achieved by the surface tension which drives the liquid flow to minimize the surface energy, resulting in a smooth surface. According to a simplified 2D model for rough surface leveling [14], the smoothing time is proportional to the surface tension of the liquid, while is inversely proportional to the viscosity. Molten Si has a surface tension about 780 mn m 1 [15], and a viscosity about 0.57 mpa s [16], while surface tension for a polymer (e.g., PMMA melt) is about 30 mn m 1 [17] and viscosity is on the order of several hundreds to thousands Pa s, depending on the molecular weight and temperature [18]. These result in a smoothing time about five orders of magnitude shorter for SPEL of Si than that for isothermal smoothing of polymeric materials. In addition to shortening the processing time, this means less damage to the materials underneath since the absorption to such a short laser pulse will be in the surface layer. The selectivity of SPEL is twofold. First, it is materials selective. The heating in SPEL is performed by a pulsed laser, which is absorbed by only a thin surface layer of the materials with a band gap smaller than the photon energy (in this case, Si) [19], so that the other materials on the surface or underneath will be kept at much lower temperature during SPEL (SiO 2 and materials beneath it). Unlike smoothing techniques that use thermal oxidation, SPEL avoids a global isothermal high temperature process. Second, this process is area selective. The laser spot is about 3 mm 3 mm and it can be adjusted within a certain range. This allows us to selectively expose the area with defects while leave other components intact. The advantages of SPEL will be more and more prominent as waveguide sizes shrink to allow for only transmission of a single optical mode. As the waveguide width and thickness are further reduced below a half of wavelength in order to increase the waveguide packing density, the effect of sidewall roughness in increasing transmission loss becomes greater, and SPEL will become increasingly important to achieve high performance integrated optical circuits. In addition, as chips become ever more densely integrated and hybridized with optical, electronic, and other functions, chip fabrication technology must keep pace. SPEL is an excellent candidate for highly selective fabrication of ultrasmooth surfaces. 5. Summary We present the use of SPEL to smooth the sidewall roughness of waveguides. Our experimental results showed that with this technique, waveguide with rough sidewall could be smoothed, and the standard deviation of the surface roughness 4

5 was reduced from 13 to 3 nm. With this reduction in the sidewall roughness, the Si waveguide transmission loss was decreased from 53 to 3 db cm 1 (an improvement of 5 orders of magnitude) according to our calculation. SPEL has demonstrated as an ultrafast, highly selective method to smooth the sidewall roughness of a silicon waveguide. These advantages will be increasingly important when the size of silicon waveguides shrink for higher density nanophotonics. Acknowledgments This work is supported in part by the US Defense Advanced Research Program Agency (DARPA) and the Office of Naval Research (ONR). References [1] Kimerling L C et al 2004 Silicon Photonics (Topics in Applied Physics) vol 94, ed L Pavesi and D J Lockwood (Berlin: Springer) p 89 [2] Tsumita N, Melngailis J, Hawryluk A M and Smith H I 1981 J. Vac. Sci. Technol [3] Lee K K, Lim D R, Kimerling L C, Shin J and Cerrina F 2001 Opt. Lett [4] Arentoft J, Sondergaard T, Kristensen M, Boltasseva A, Thorhauge M and Frandsen L 2002 Electron. Lett [5] Juan W H and Pang S W 1996 J. Vac. Sci. Technol. B [6] Armani D K, Kippenberg T J, Spillane S M and Vahala K J 2003 Nature [7] den Hertog M I 2005 Master Thesis FOM-Institute AMOLF (Amsterdam) [8] Chou S Y and Xia Q F 2008 Nat. Nanotechnol [9] Chou S Y, Krauss P R and Renstrom P J 1996 Science [10] Payne F P and Lacey J P R 1994 Opt. Quantum Electron [11] Ladouceur F and Love J D 1995 IEE Proc. Optoelectron [12] Chou S Y, Keimel C and Gu J 2002 Nature [13] Xia Q F and Chou S Y 2008 Nanotechnology [14] Orchard S E 1962 Appl. Sci. Res. A [15] Przyborowski M, Hibiya T, Eguchi M and Egry I 1995 J. Cryst. Growth [16] Sato Y, Kameda Y, Nagasawa T, Sakamoto T, Moriguchi S, Yamamura T and Waseda Y 2003 J. Cryst. Growth [17] Brandrup J and Immergut E H (ed) 1989 Polymer Handbook 3rd edn (New York: Wiley) [18] Heyderman L J, Schift H, David C, Gobrecht J and Schweizer T 2000 Microelectron. Eng [19] Silvesrelli P L, Alavi A, Parrinello M and Frenkel D 1996 Phys. Rev. Lett