21% los consumidores que son

A second sample underwent identical processing as the sample just described, includ- ing the deposition of a silicon nitride cap, lithography, dry-etching, ZEP removal, and 3×Piranha/HF treatment. As expected, the best measured linewidth of 0.7 pm was very similar to the previous sample and is shown in Fig. 5.18. However, this time a 10 nm TOX layer was grown on the Si surface at 1000◦C for 3.1 minutes in an attempt to form a good Si interface with a hydrogen free material. After switching off the O2, the sample was allowed to cool slowly under an N2 ambient for approximately 5

minutes before retesting. The best linewidth after thermal oxidation was 3.5 pm, a result similar to the silicon nitride cap. However, the same 40 min FGA had virtually no impact on the sample with the thermal oxide cap. A 4.5 hr high-temperature anneal in an N2 ambient was found to significantly improve the losses, where the

best linewidth was measured to be 1.6 pm. The high-temperature anneal consisted of holding the furnace at 1000◦C for 3 hours and then letting the temperature slowly ramp down to 400◦C over the course of the remaining 1.5 hours. Assuming that the high-temperature anneal successfully healed the Si interface and bulk TOX, a 40 min FGA was conducted on the sample. The FGA was found to slightly reduce the quality factor of the best resonance, where the linewidth was measured to be 1.8 pm.

δλ

(pm)

0

Successive Processing Steps 3x Pir / HF 40 min FGA

@400o_C 1 2 3 10 nm TOX @1000o_C 4.5 hr N₂ @1000o_C 4.5 hr N₂ @1000o_C 40 min FGA @400o_C

Figure 5.18: Summary of best linewidths after selected processing steps for 5−10 μm radii disks fabricated with a thermal oxide encapsulation layer along with various annealing trials.

A second 4.5 hr high-temperature anneal drove out the hydrogen from the sample, resulting in aδλ= 1.2 pm on the best resonance. This set of four anneals showed that the thermal oxide needed time at high temperature to remove material and surface defects. Also, the FGA had a marginally degrading effect on the optical losses.

A third sample was similarly fabricated but did not have an initial silicon nitride cap prior to lithography and dry-etching. After HF undercutting and 3×Piranha/HF treatments, the best linewidths were measured to be 1.0 and 0.6 pm, respectively (shown in Fig. 5.19(a)). The marginal improvement in this case was attributed to a simplified single material dry-etch and a less damaged top Si surface. The latter was confirmed after an identical 10 nm TOX cap with 5 minute cool-down showed a best linewidth of 2.0 pm, much better than the second sample’s 3.5 pm linewidth after oxidation. Omitting any FGA step, a final 4.5 hour high-temperature anneal completely healed the Si-interface and bulk silica cap, showing identical linewidths prior to oxidation. Figure 5.19(b) shows a high-resolution transmission spectrum of the 1444.2 nm resonance on a 7.5 μm radius disk after the final high-temperature anneal, along with a doublet model fit. As described in Section 6.3.5, totally oxidized and annealed microdisks had Q >3×106_{. As these disks were most likely surface-}

scattering limited due to their design, quality factors this high indicate that the encapsulating oxide is of extreme quality after high-temperature anneals. Further described in Section 6.3.5, thermal oxide encapsulated and non-undercut microrings were observed to have quality factors ∼5×106.

δλ

(pm)

0

Successive Processing Steps

HF Undercut 10 nm TOX @1000o_C 1 2 3 3x Pir / HF 4.5 hr N₂ @1000o_C (a) (b) -6 -4 -2 0 2 4 6 0.92 0.93 0.94 0.95 0.96 0.97 0.98 0.99 1.00 Wavelength (pm) Normalized Transmission Qc= 2.6 x 106 Qs= 2.0 x 106 Q_β= 1.2 x 106 Qe= 6.1 x 107 δλc= 0.57 pm δλs= 0.74 pm Δλ = 1.20 pm δλe= 0.02 pm

Figure 5.19: (a) Summary of best linewidths after selected processing steps for 5−10μm radii disks fabricated without an initial protective cap. This sample also had a thermal oxide encapsulation layer but no FGA. (b) High-resolution transmission spectrum of 1444.2 nm resonance on a 7.5 μm radius disk after the final high-temperature anneal.

The results shown in the previous figure represent the successful encapsulation of the once delicate Si-surfaces, as 10 nm of thermal oxide will completely prevent native or chemical oxide growth during any subsequent fabrication steps. While silicon nitride would have been slightly preferable in terms of chemical resistance, the high-quality thermal oxide is a simple and effective method of sealing the disks from environmental contamination while still allowing optical access to the mode’s near-field. Having also demonstrated this technology on a planar resonator represents a significant milestone in ultra-low-loss silicon photonics technology. Any number of functionalized materials can now be placed as claddings over the resonators, allowing for the technologically viable development of biological/chemical sensors, microlasers, and highly nonlinear devices integrated with electronic circuitry on a silicon chip.

Chapter 6

Silicon-Based Lasers

6.1

Introduction

Despite the recent successes in silicon microphotonics, a practical silicon-based laser remains an elusive goal. Commercial ventures, such as Xponent Photonics and Lux- tera, require the use of flip-chip integration of III-V lasers on top of silica or silicon photonic circuits to provide coherent light sources. These companies have made sig- nificant progress in reducing the costs associated with this technology, and they have begun to create real markets for silicon-based photonics. However, this type of pick and place technology lacks the scalability to provide hundreds of laser sources for optical interconnect applications [6]. One common idea is to wafer bond a III-V material to the top surface of an SOI wafer. Hattori et al. [112] demonstrated the heterogeneous integration of an InP microdisk laser, an SOI strip waveguide, and an InP detector. The optically pumped 2.5 μm radius InP microdisk exhibited good coupling effiencies. Unfortunately, the large 500 μW thresholds revealed that signif- icant obstacles still exist for this type of technology. Silicon nanocrystals embedded in gate oxide have also been used to produce field-effect electroluminescence [113]. Unfortunately, quantum confinement always raises the radiative transition energies. Thus, all nanocrystals, quantum dots, etc., produced from silicon emit above the bandgap where bulk silicon becomes highly absorbing. However, nanopatterning of crystalline silicon was used to create 1278 nm stimulated emission just below sili- con’s bandgap [114]. In this work, it was hypothesized that a defect state, located 0.17 eV below the conduction band edge of silicon, allowed direct recombination be-

photovoltaic technology, has also shown that relatively efficient (∼ 1%) one- and two-photon-assisted sub-bandgap emission from silicon is possible [115, 116]. The ap- proximately hundred-fold improvement in electroluminescence efficiencies were made possible in large part by the surface passivation techniques developed for solar cells. Two more approaches to achieve silicon-based lasers are the Raman effect and rare- earth-doped glasses. This chapter will describe these two common technologies in detail as well as provide our experimental progress in regards to achieving silicon lasers. Raman based silicon lasers offer the appeal of an “all-silicon” material sys- tem, but suffer from an extremely large drawback: the useful gain bandwidth is less than a nanometer wide. This narrowband gain and the need for coherent excitation relegates the Raman laser to performing wavelength conversion. By contrast, the erbium-doped glasses presented in Section 6.3 have broadband gain and are routinely used for DWDM amplification applications. This material was deposited over high-Q silicon-based microresonators to demonstrate highly efficient single-mode laser sources with sub-μW thresholds.