1. Introduction

Recently silicon-on-insulator (SOI) has emerged as an attractive materials system for photonic integrated circuits (PICs). SOI offers the advantages for potentially being integrated with standard silicon electronics. In addition, because of its compatibility with silicon planar processing, CMOS fabrication tools and techniques can be utilized. Furthermore, the large refractive-index contrast of SOI allows very tight modal confinement, which leads to device miniaturization. These advantages have already led to several reports of ultrasmall passive optical structures such as 90° bends, and T- or Y-branches with excellent performance [1]. For example, sharp bend radii of the order of 1 µm with low losses have been fabricated [2]; this particular capability shows clearly the advantages of SOI for deeply scaled down PICs.

One important challenge in realizing greater use of this materials technology is to make a set of active devices based on SOI. Since silicon is centrosymmetric, it does not exhibit a linear electro-optic effect. However, in the last two years high-performance optical modulators and switches based on thermooptic and free-carrier dispersion effects have been demonstrated [3, 4]. Achieving on-chip light emission is an even more important capability and has been a major component of research on active devices. For example, erbium doping has been carefully explored but the results have been, thus far, limited [5]. Recently Claps et al. have proposed and demonstrated the use of the Raman effect in silicon to achieve an on-chip amplifier [6]. The large Raman gain coefficient of silicon (~10⁴ larger than that of silica) can, in principle, be used to achieve practical levels of gain with a diode laser pump. In this connection, Claps et al. have demonstrated spontaneous Raman emission [6], Raman amplification [7], and coherent anti-Stokes Raman scattering (CARS) [8]. In these studies, they used a rib waveguide structure that yielded a modal area of 5.4 µm², from which signal amplification of 0.25 dB was observed with 1.6W of coupled pump power [7].

Because the Raman effect is a nonlinear optical process, tighter optical confinement can lead to an increase of the efficiency of the process. Hence, from the viewpoints of practical SOI-device integration, further reduction in the transverse waveguide dimensions is a necessity. For this reduction to be realized, however, two experimental issues must be solved. First, the sidewall roughness of the waveguide must be lowered to reduce the high propagation loss. Second, the input and output coupling from fiber to waveguide must be efficient.

In this paper, we employ low-loss, ultrasmall-core SOI waveguides to demonstrate stimulated Raman amplification at 1550 nm using a 1435 nm diode pump. We observe On-Off gains of up to 0.7 dB for small pump powers of <30 mW. Our experiments make use of SOI strip waveguide devices with a cross-section of 0.098 µm².

2. Fabrication and experimental setup

The devices were patterned on 200 mm SOI Unibond wafers (SOITEC) with a 220 nm-thick, lightly p-doped silicon top layer on a 1 µm SiO₂ layer. A 50 nm-thick oxide was deposited via low pressure chemical vapor deposition (LPCVD) as a hard mask for the etching process. The patterns were defined by electron beam lithography using a Leica VB6-HR commercial 100 keV system. The exposed wafers were then etched in a standard 200 mm CMOS line at IBM Watson Research Center [9]. The resist pattern was transferred to the oxide mask using a CF₄/CHF₃/Ar etch chemistry. After resist removal, the oxide mask was transferred to the top silicon layer with an HBr-based etch. A second lithography step defined the polymer (n~1.58) used, in conjunction, with an inverted waveguide taper as the spot-size converter-couplers. The samples were then cleaved on each side to enable edge-coupling. Figure 1 shows the optical and scanning electron microscopy (SEM) micrographs of the fabricated devices. The sidewall angles were ~90° and the roughness values were 5 nm. The final devices were 4.6 mm long. The polymer spot-size converter-couplers were 3 µm wide, 2 µm thick, and 200 µm long, with a tapered-tip size of 75 nm. The single-mode strip waveguides were 445 nm wide, 220 nm thick, 4.2 mm long, and support only the TE polarization.

 

Fig. 1. Optical and SEM micrographs of fabricated SOI devices. (a)Waveguides of varying lengths. (b) Zoom of polymer taper spot-size converter. (c) SEM of polymer with inverted silicon taper tip

Download Full Size | PDF

The schematic of our experimental setup for measuring Raman gain is shown in Fig. 2. A Corning Lasertron diode, operating at 1435 nm, was used as our pump and a JDS Fitel broadband noise source with power ~1 mW, bandwidth ~40 nm and centered at λ=1550 nm was our signal source. The counter-propagating configuration for the pump and signal beams was used to minimize other nonlinear optical processes, e.g., Four-Wave Mixing (FWM), that may be phase-matched along the forward direction. The pump beam was sent to a pump-signal combiner and was in-coupled into the input facet of the waveguide through a tapered polarizationmaintaining (PM) fiber with a spot size of ~2.5 µm, as depicted by Fig. 2. The pump beam was then out-coupled into a receiving tapered fiber and demultiplexed into the pump channel of a similar pump-signal combiner. Conversely, our broadband signal was coupled into the waveguide via the combiner and the tapered fiber in the opposite direction. The power of the broadband radiation was several orders of magnitude larger than the power of the spontaneous emission, centered at λ=1550.7 nm, generated in both forward and backward directions by the pump beam. Finally the counter-propagating broadband signal was coupled to the signal line of the input pump-signal combiner and subsequently monitored by an optical spectrum analyzer (OSA) with a 2 nm resolution. We used a high-bandwidth OSA detection setting in order to improve the signal-to-noise ratio (SNR) of our device. However, in order to obtain the correct linewidth of the gain spectrum, we used a higher-resolution, lower-bandwidth setting of 0.5 nm, as discussed below.

 

Fig. 2. Experimental setup for measuring stimulated Raman gain in SOI waveguides.

Download Full Size | PDF

3. Results and discussions

The propagation loss of our waveguides was measured using the cutback method and found to be 3.6±0.1 dB/cm at 1550 nm [2]. Input and output coupling losses were ~1.5–2 dB/coupler. All results were for the TE polarization.

We measured the On-Off gain of the device, defined as, G=10logR, where R is the output power while the pump is on divided by the output power while the pump is off. Figure 3(a) shows the measured Raman gain spectrum of the ultrasmall SOI waveguides. The input pump power was 20.5 mW with an On-Off gain of 0.4 dB. The data exhibits a gain maximum at λ=1550.7 nm, which corresponds to the predicted Δν=15.6 THz (521 cm ^-1) Raman shift in silicon [6]. We measured accurately the Raman linewidth of Δλ~1 nm using the higher-resolution OSA setting; this value agrees with the convolution of the pump beam linewidth and the silicon Raman linewidth. We also measured the spontaneous Raman spectrum, i.e., the pump in the absence of the signal, for the same pump power using the same experimental parameters; this is shown in Fig. 3(b) for comparison. Clearly, the stimulated Raman data agree well with the spontaneous Raman peak position and linewidth.

In order to examine the power dependence of the gain, we measured the On-Off gain versus the input pump power (i.e., power entering the waveguide) as shown in Fig. 4. The maximum gain was G~0.7 dB (15%) for a pump power of P_R~29 mW and for a waveguide length of L=4.2 mm. The Raman amplifier figure-of-merit (FOM), which we define as FOM=G/(P_RL)=57.47 dB/cm/W, is approximately 3 orders of magnitude greater than previously reported for SOI-based Raman amplifiers [7]; this increase in FOM shows that the gain scales approximately linearly with modal area. The data in Fig. 4 is approximately linear with a slope of 0.029 dB/mW. We compared this data to a numerical solution of the stimulated Raman differential equations [10], viz.

 

Fig. 3. Stimulated (a) and spontaneous (b) Raman emission spectra with high-bandwidth OSA detection for SOI waveguides.

Download Full Size | PDF

where P_P is the pump power, P_R is the Raman power, ν_P and ν_R are the pump and signal frequencies, α=3.6 dB/cm is the propagation loss of the waveguide, β=0.44 cm/GW is the Two Photon Absorption (TPA) coefficient [7], A _eff=0.059 µm² is the effective modal area, and g_R=29 cm/GW is the stimulated Raman scattering (SRS) coefficient. The solution takes into account the effective pump power due to the finite pump bandwidth, i.e. $P_{\mathrm{eff}}=P_P⁄\left(1+\frac{\Delta {\nu }_P}{\Delta {\nu }_{R,\mathrm{Si}}}\right)$, where Δν_P=160 GHz and Δν_R=105 GHz [7, 11]. The effect of pump depletion, i.e., the coupling term in equation (1), is also accounted for in the calculation and found to be negligible. The calculated data, shown as a dashed line in Fig. 4, has a slope that matches the experimental data to within ±10%. From this comparison, we estimate the SRS coefficient to be ~29±4 cm/GW.

Our On-Off gain and SRS coefficient agree well with the results of Ref. 7, but there is clearly a discrepancy between the calculated and the experimental data since there is an offset of ~0.2 dB in the lower-power extrapolation of the linear gain. We believe this offset is attributed to a thermally-induced change of the tapered fiber tip at higher pump powers; this was a reproducible effect, which caused fiber misalignment. It is possible to limit this effect by adding a temperature bias on the system and using piezoelectric XYZ actuators for more accurate and stable positioning. At present, our maximum gain is limited by the available pump power. Either a higher-power pump diode laser or lower losses within the optical components of our experimental setup, e.g., beam combiners, connectors, and tapered fibers, would increase gain. Because of the tight confinement of our waveguides, other nonlinear processes, such as TPA and Stimulated Brillouin Scattering, can be present as well [7]. However, our calculations indicate that these effects are negligible because of the low pump powers used. The effect of TPA-induced free-carrier absorptionwas recently proposed as a limitation on the achievable Raman gain in SOI [12]. However, this effect is negligible in our experiments because the linearity of the power dependence on the spontaneous emission data indicates the absence of free-carrier induced loss. Furthermore, our deeply-scaled down waveguide cross-section reduces the transit time of the carriers. Hence, the effective recombination lifetime has a calculated upper bound of 0.77 ns. According to Claps et al., a lifetime value below 1 ns would render the free-carrier absorption negligible [13].

 

Fig. 4. On-Off gain versus input pump power. The maximum gain is 0.7 dB (17%) with a pump power of ~29 mW. A linear fit with a slope of 0.029 dB/mW corresponds to an SRS coefficient, gR~29 cm/GW.

Download Full Size | PDF

4. Conclusion

In conclusion, we have obtained significant Raman On-Off gain of 0.7 dB from 4.2 mm long submicron-cross-section SOI waveguides using low CW pump powers from a laser diode. The Raman amplifier had a FOM of ~57 dB/cm/W, approximately 10³ greater than obtained in large-area Si waveguides and consistent with the low loss and small cross-section of our waveguide system. Further work in SOI waveguide fabrication using optimized CMOS processing technology can lead to even lower propagation losses, thereby allowing longer device lengths and higher Raman gains.