1. Introduction

Silicon Photonics is quickly proving to be suitable interconnect technology for meeting the future goals of on-chip bandwidth and low power requirements [1]. However, it is not clear how silicon photonics will be integrated into CMOS chips, in particular, microprocessors. With increasing device density on a single layer, and process/device incompatibilities, the design of future generations of interconnect networks will be a challenge. Consequently, the issue of where to integrate photonic circuits into electronic IC fabrication processes to achieve maximum flexibility and minimize complexity and cost is an important one. In addition, in order to best use chip real estate, it will certainly be advantageous to integrate in three-dimensions [2,3].

Crystalline silicon photonic devices have been integrated into three vertically stacked layers using multiple implantation and epitaxial growth steps [4]. However, this process requires a high thermal budget, which significantly limits integration flexibility. In contrast, non-crystalline forms of silicon such as polycrystalline and amorphous silicon can be arbitrarily deposited yielding low-cost and flexible fabrication. Three-dimensional integration of polycrystalline silicon has been achieved [5]. However, polysilicon has higher optical loss from scattering at the grain boundaries and backend compatibility will be a challenge due to the high thermal budget required to crystallize polysilicon [5–7]. While laser annealing can considerably reduce the thermal budget, it is not clear if waveguides can be realized with it [8]. On the other hand, amorphous silicon (a-Si) could be used but it inherently has a large density of point defects and dangling bonds which significantly increase loss. However, with the addition of hydrogen to amorphous silicon, the dangling bonds can be significantly passivated [9]. This results in a very low optical loss of less than 2-3 dB/cm, more than adequate for realizing waveguides on centimeter scale chips [10–12]. Hydrogenated amorphous silicon also has the distinct advantage over other forms of silicon in that it can be deposited using low temperature (~200-400 °C) plasma-enhanced chemical vapor deposition (PECVD) which automatically passivates the material with hydrogen. Low-temperature integration yields maximum flexibility, enabling photonics to be placed into an integrated circuit at any point in the fabrication process, particularly at the backend where the device density is lower and consequently where the relatively large size of photonics waveguides (in comparison to electronic devices) can be accommodated.

However, a significant challenge with a-Si:H is the realization of fast, compact, and low powered devices. The difficulty is that silicon lacks a large electro-optic effect other than free-carrier plasma dispersion effect (FCPD). While free-carriers have successfully been used to modulate light in crystalline silicon and polycrystalline silicon [13–15], the significantly lower free-carrier mobility in a-Si:H could significantly limit device speed [9]. This was seen in amorphous silicon filled microstructured optical fibers where a modulation rate of only 1.4 MHz was achieved [16]. However, here we show that by leveraging nanosized waveguides, fast sub- nanosecond surface recombination of carriers is possible. In addition, the low free-carrier mobility of a-Si:H actually has a significant benefit in that it dramatically enhances free-carrier absorption due to the strong scattering of photons with slowly moving carriers. Consequently, here we realize large modulation without the use of a resonant cavity, in just a waveguide, unlike other approaches which require interference effects [13–15].

2. Hydrogenated-amorphous silicon (a-Si:H) resonators

We fabricated the devices by first creating an a-Si:H-on-insulator substrate. A 3-µm thick wet oxide was grown on top of a p-doped <100> silicon substrate by thermal oxidation. Hydrogenated amorphous silicon was then deposited using a plasma-enhanced chemical vapor deposition (PECVD) technique to form an a-Si:H-on-insulator substrate. The deposition parameters are given in Table 1 . The thickness of the film was determined to be ~251 nm using an ellipsometer and its refractive index was measured 3.48 at 1550 nm.

Table 1. PECVD deposition parameters for a-Si:H

View Table

In order to demonstrate the low loss of our film, we fabricated passive a-Si:H microdisc resonators with a radius of 5 µm using electron beam lithography followed by chlorine inductively coupled-plasma etch. The devices were then covered with a 2 µm thick SiO₂ deposited using PECVD to serve as cladding-layer. The coupling gap between the microcavities and bus waveguides (~460 nm wide) was ~250 nm. Quasi-TE polarized (electric field parallel to the substrate) light was launched into the input waveguides from a tunable continuous wave laser source through a polarization controller using a tapered lens fiber. Tapered adiabatic couplers were used to couple light into and collect light from the device and tapered lens fibers [17]. From the measured transmission of the resonances in Fig. 1 , the estimated Q of the disk resonator is Q ~92,000 with an extinction ratio of ~12 dB. From these measurements, we determined that a-Si:H material loss is less than 3.5 dB/cm, which can be considered low enough to realize a platform of high quality, low loss optical devices on a chip.

 

Fig. 1 Measured response of 5 µm radius microdisc resonator fabricated on a-Si:H. The measured quality factor is Q ~92000.

Download Full Size | PDF

3. All-optical modulation in amorphous-silicon (a-Si:H)

In order to show that a-Si:H is a viable platform for high performance active nanophotonic devices, here we demonstrate broadband and high speed all-optical modulation in a compact device. Modulation is achieved by using a short optical pump pulse to induce an absorption change in an a-Si:H waveguide (460 nm wide x 251 nm tall) as shown in Fig. 2 . A mode-locked Ti-Sapphire laser tuned to 810 nm provides 100 fs pump pulses at a 80 MHz repetition rate that are frequency doubled using a beta-barium borate (BBO) crystal to 405 nm. The frequency doubled pump pulses are then coupled into a SMF-28 fiber using a fiber collimator (optimized for λ = 405 nm) and are incident on top of an amorphous silicon waveguide. We estimate that at least 90% of the incident pump pulse is absorbed in the a-Si:H within a ~15 µm diameter spot. The free-carriers generated by the pump are used to modulate a continuous-wave tunable laser diode at λ=1550 nm. The probe (polarized to be quasi TE or E-field parallel to the substrate) is coupled into the waveguide using a tapered lens fiber. The modulated probe output is detected using a 20 GHz photodetector and the temporal response is observed with a 40 GHz digital sampling oscilloscope.

 

Fig. 2 Experimental set-up to demonstrate amplitude modulation in a-Si:H waveguides. The 405 nm pump pulses are incident on an a-Si:H waveguide to modulate the 1550 nm cw probe light. PC: Polarization Controller

Download Full Size | PDF

4. Results and analysis

We see that in Fig. 3 , the probe amplitude is quickly modulated by each pump pulse. We attribute this modulation to free-carrier absorption induced by the photo-excited carriers generated through linear absorption of the pump pulse (the quasi-bandgap of a-Si:H is ~1.7 eV). In the figure shown, the probe undergoes a ~3 dB modulation when the incident pump pulse energy is E_pump = 75 pJ. Due to the poor overlap of the pump light with the a-Si:H waveguide, we estimate that only 1.5 pJ of the pump energy is actually absorbed in the waveguide. Consequently, the modulation depth could be improved with better overlap. We should note here that no fabry-perot oscillations from the end facets were observed in the transmission measurements. We see in Fig. 3 that there are two distinct time-scales in the modulation of the probe - a fast and a slow component. The fast transition occurs when the pump photons are absorbed. Since the photon energy is much larger than the quasi-bandgap energy, the electrons are initially excited high into the conduction band, which dramatically increases the free-carrier absorption. The carriers then rapidly thermalize into the extended band tail states in the conduction band [18,19]. We could not determine the exact time-scale over which this occurs because this transition was limited by the time-response of our detector/oscilloscope but, in other works, it was determined that the carriers thermalize over a time-scale of less than 5 picoseconds [19]. There is then a slow return (~400 ps) of the signal which we attribute to surface recombination of free carriers with the etched sidewalls of the waveguide. We note that this recombination time is comparable to what is seen in similarly sized crystalline waveguides [13,15]. We also note that the slow temporal components’ maximum absorption is less than that of the faster temporal component. We attribute this to a density-of-state dependent free-carrier absorption coefficient as previously observed and discussed in more detail in [18,19].

 

Fig. 3 All-optical modulation in a-Si:H using a pump-probe scheme. The carriers undergo a rapid thermalization before undergoing non-radiative recombination.

Download Full Size | PDF

In order to verify the enhancement of free-carrier absorption in a-Si:H, we performed an identical experiment in crystalline silicon waveguide and did not observe any modulation. This enhancement can be understood by considering the Drude model, where free-carrier scattering introduces optical absorption. The applicability of the Drude model to an amorphous material has been shown to give good agreement with experimental results [19]. The free carrier absorption cross-section σ, describing the change in the absorption due to change in free carrier concentration is given by [20]

Consequently, free-carrier absorption in a-Si:H is at least an order of magnitude higher than in crystalline silicon (σ = 1.45∙10⁻¹⁷ cm²) [21]. From the degree of modulation seen in Fig. 3, we estimate that a free-carrier density of N ~1.98∙10¹⁸ cm⁻³ was injected into the waveguide. This can be confirmed from calculating the carrier density induced by an absorbed pulse energy of 1.5 pJ, yielding a carrier density of N ~1.52∙10¹⁸ cm⁻³, which is in very good agreement with our estimate from the Drude model.

5. Conclusion

We have demonstrated broadband all-optical modulation of light in amorphous silicon waveguides (a-Si:H) based on the free carrier absorption effect using a pump-probe technique. The carrier recombination time in a-Si: H nanosized waveguides is also comparable to that in crystalline silicon. We envisage that the enhanced free-carrier absorption in a-Si:H could enable compact, broadband electro-absorption modulators. Ultrafast response in a-Si:H due to free-carriers and nonlinear effects have also been recently reported [22–24]. Consequently, hydrogenated-amorphous silicon can be an enabling platform for backend integration of optical devices in a hybrid electro-optic network-on-chip architecture.