1. Introduction

Silicon is an ideal material for dense on-chip integrated photonic circuits due to its high refractive index, transparency at the 1.55 µm telecommunication wavelength, and its well established fabrication process technology of the CMOS industry. In future optical networks, all-optical switching devices, in which the modulation of one light beam is controlled by another beam, will be essential components for signal processing at highest operation speeds. The relevance of all-optical switching on silicon has been recently underlined by the presentation of an all-optical logic based on a silicon microring resonator [1].

Silicon does not exhibit second order nonlinear χ⁽²⁾-effects, including a lack of the Pockels effect. Thus, switching devices in silicon-on-insulator (SOI) technology are based either on third-order optical χ⁽³⁾-nonlinearities or on the plasma dispersion effect. All-optical switching approaches based on third order nonlinear processes, i.e. the optical Kerr effect, are regarded to offer high potential for ultrafast sub-picosecond switching applications [2-4], as they do not rely on the lifetime and transport dynamics of charge carriers. However, these approaches often utilize hybrid material systems and require long interaction lengths on mm-scale due to the relatively weak third order nonlinearity of silicon, which is a drawback in terms of CMOS compatibility and costs. On the other hand, charge carrier based photonic switching facilitates truly monolithic and densely integrated photonic devices, although they are limited in speed by the charge carrier lifetime and transport.

SOI microring resonators have proven to be well suited switching devices due to high achievable quality factors [5]. Electro-optic modulators based on carrier injection in SOI microring resonators via lateral p-i-n diode structures have been realized [6, 7]. Recently an advanced design for a diode structure for switching times up to 25 ps has been proposed [8]. As electrical injection of electron-hole pairs may be undesirable in certain applications, photoexcitation via absorption of optical pulses is an attractive alternative, i.e. for all-optical switching [9-15].

On-off modulation speed in carrier based switching devices is limited by the effective lifetime of the generated charge carrier densities inside the microring resonator structure. Carrier lifetimes between some hundred picoseconds and several nanoseconds have been reported for bare submicrometer-sized silicon-on-insulator waveguides [14, 17]. With the use of p-i-n diode structures effective lifetimes down to 50 ps at 10 V reverse bias voltage could be achieved by electrical carrier sweep out [9]. However, this concept and especially the recently presented advanced diode structure approach [8] suffer from a high technological effort.

An alternative method to significantly reduce free carrier lifetimes in silicon is the ion implantation induced introduction of artificial defect states, which act as fast trapping, i.e. recombination, centers. This approach is considerably less complex from a technological point of view. Reducing the carrier lifetime down to 1.9 ns via Helium ion implantation of micrometer-sized waveguides has been successfully applied to increase Raman gain [18]. A reduction of the carrier lifetime in silicon down to 70 ps by Argon ion implantation has been exploited for fast all-optical switching [10].

Oxygen ion implantation is a third alternative material approach to efficiently reduce the free carrier lifetime in silicon [19]. We have recently presented the feasibility of realizing high-speed all-optical switching in oxygen ion implanted silicon microring resonators with a switching time of about 100 ps at an extinction ratio of 9.5 dB [20]. As ion implantation introduces additional light propagation losses due to amorphization of the silicon waveguides a trade-off between short carrier lifetimes and considerable losses has to be made. In this work we carefully investigate the influence of different implantation doses on switching performance and accompanied additional propagation losses in SOI microring resonators. Compared to the state-of-the art, we achieve a fourfold reduction of the switching time down to 25 ps at an extinction ratio of 9 dB.

2. Device fabrication and oxygen ion implantation

Photonic devices under investigation are SOI microring resonators. In general, these devices rely on the interference of input light with an amount of light coupled from the ring resonator at specific resonant wavelengths. Figure 1(a) shows a SEM image of one investigated symmetrically coupled microring resonator with a ring radius of 5 µm and a coupling gap of 150 nm. Waveguide widths and heights are 400 and 340 nm, respectively.

 

Fig. 1. (a) SEM image of one investigated symmetrically coupled microring resonator with a ring radius of 5 µm and a coupling gap of 150 nm. (b) Schematic cross sectional view of the SOI waveguide.

Download Full Size | PDF

The devices are fabricated on SOI substrates consisting of a 2 µm thick buried oxide layer and a 340 nm thick device layer. Definition of the structures is carried out by electron-beam lithography (EBL) using hydrogen silsesquioxane (HSQ) as an EBL resist. Pattern transfer into the silicon is achieved by HBr inductive coupled reactive ion etching. Figure 1 (b) shows a schematic view of the waveguide cross-section after patterning. On top of the silicon waveguides a 100 nm thick layer of HSQ was left intentionally to act as a scattering layer for homogeneous defect state profiles in the following implantation step.

To avoid implantation induced losses, which can arise from both scattering at lattice imperfections and absorption at electronic defect states, in mm-long access waveguides, the whole sample except for an 80×80 µm² quadratic area centered at the microring resonator was protected by a 1.6 µm thick photo resist in a conventional photolithographic step. After ion implantation, this mask was removed completely in a solution of sulphuric acid (H₂SO₄) and hydrogen peroxide (H₂O₂).

To effectively reduce the charge carrier lifetime for high-speed switching applications the vertical distribution of implantation induced trapping centers is required to match the optical mode inside the silicon waveguide. Simulations of the optical waveguide mode and the implantation profile were carried out to optimize the implantation energy with respect to this overlap. Figure 2(a) shows the two-dimensional profile of the dominant electric field component of the TM mode inside the waveguide simulated by use of the finite element simulation tool HFSS. Figure 2(b) depicts the depth profile of defect states introduced into the SOI waveguide by oxygen ion implantation at 160 keV together with the normalized one-dimensional TM mode profile along the vertical direction at the center of the waveguide. The implantation profiles were calculated by a Monte-Carlo based SRIM (Stopping and Range of Ions in Matter) simulation. The shown implantation voltage of 160 keV was found to maximize the overlap of optical mode and implantation profile and was used for device fabrication within this study.

 

Fig.2. (a) Simulated mode profile of the SOI waveguide for TM polarization, shown is the absolute value of the dominant component of the electric field |E_y(x,y)|. (b) Simulated axial oxygen ion distribution inside the silicon waveguide at an implantation voltage of 160 keV and calculated TM mode profile along axial direction at the center of the waveguide.

Download Full Size | PDF

To analyze the influence of oxygen ion implantation on the propagation properties of the ring resonator, transmission characteristics of investigated microrings were measured before and after implantation in a conventional continous-wave (cw) endfire setup with a spectral resolution of 0.01 nm. Measurements are carried out with implantation doses ranging from 3×10¹¹ to 7×10¹² cm^-2. Exemplarily, the result of the microring implanted with the highest dose is presented first.

 

Fig. 3. Measured normalized transmission characteristics of through and drop channel of one resonance of the investigated microring resonator. (a) Optical transmission before ion implantation. (b) Optical transmission after 7×10¹² cm^-2 oxygen ion implantation.

Download Full Size | PDF

Figure 3(a) illustrates the transmission spectra of one resonance of the unimplanted microring resonator for TM polarization with an on-resonance on chip insertion loss of about 15 dB at the through port. Figure 3(b) depicts the same resonance after 160 keV oxygen ion implantation at an implantation dose of 7×10¹² cm^-2. The through port data were normalized to the maximum transmission out of resonance and the drop port data were normalized to the maximum transmission on resonance. An implantation induced shift of the absolute position of the resonance itself could not be clearly identified, mainly because the measurements were not carried out in a temperature controlled setup.

An important figure of merit of a microring resonator is its quality factor Q, which is defined as the resonance wavelength λ₀ divided by the 3 dB bandwidth Δλ_3dB of the resonance. Obviously, implantation results in a broadening of the resonances and a decrease in modulation depths at drop and through ports. The broadening corresponds to a decrease of the quality factor from 2100 to 1300. By applying a fit of the transmission data to an analytical transfer function as described in the literature [19, 20], intrinsic losses of 34 dB/cm and additional implantation induced losses of 68 dB/cm have been found [5, 20, 21]. However, these rather high losses lead to an additional transmission drop of only 2.5 dB at the drop port resonance wavelength. At the through port the transmission out of resonance is mainly degraded by the implanted comparably short part of the access waveguide (0.5 dB), while on resonance the modulation depth of about 10 dB is still well suited for switching application. Nevertheless these additional losses degrade the device performance and are in general the limiting factor for the maximum feasible implantation dose, as will be discussed later.

3. Experimental setup

For a detailed analysis of the spectral and temporal switching characteristics the implanted microring resonators are investigated in a femtosecond pump probe setup, as schematically shown in Fig. 4. Pump pulses of 120 fs duration at a repetition rate of 82 MHz are derived from a Ti:sapphire laser at 800 nm wavelength and focused onto the microring to generate electron-hole pairs by linear absorption. The temporal pulse spacing is 12.8 ns which compared to the carrier lifetime inside the microring is large enough to negelect carrier accumulation. The spectral response of the microring resonator is probed by 220 fs pulses from a synchronously pumped optical parametric oscillator (Δλ_FWHM=15 nm). These probe pulses are coupled to the SOI sample via a polarization maintaining lensed fiber and an adiabatic taper structure of the silicon bus waveguide with estimated coupling losses of 13 dB per coupler. A motorized translation stage provides an adjustable time delay between pump and probe pulses. By discrete variation of this time delay a time-resolved investigation of the transmission spectra at drop and through channels is possible by outcoupling the light from the respective SOI waveguides over a standard multimode fiber to a sub-nanometer resolution optical spectrum analyzer (OSA).

 

Fig. 4. Schematic view of the experimental setup used for the time resolved measurements of the implanted microring resonator structure.

Download Full Size | PDF

4. Time-resolved characterization and discussion

The temporal evolution of the drop channel resonance shown in Fig. 3(b) of the microring implanted with a 7×10¹² cm^-2 oxygen ion dose is displayed in Fig. 5. We define Δt=0 as the point of maximum wavelength shift and the time delay is varied in steps of 1 ps. The energy and spot diameter of each vertically incident pump pulse are 0.84 nJ and 26 µm, respectively, corresponding to a laser fluence of about 0.16 mJ/cm². The resolution of the optical spectrum analyzer is set to 0.2 nm and the transmission spectra are measured using a step size of 4 pm. It was carefully checked that this lower resolution, compared to the cw measurement setup, did not distort the spectral shape of the investigated resonance.

At zero time delay, the resonance wavelength is blue shifted due to the plasma dispersion effect and concomitantly the resonance intensity is decreased by free carrier absorption. At positive time delays the resonance shifts back to its equilibrium on a time scale determined by the free carrier lifetime inside the silicon waveguide.

 

Fig. 5. Intensity plot of the time-resolved spectral response of the microring resonator at the drop channel after optical excitation. The intensity is color-scaled in linear arbitrary units. The oxygen ion implantation dose of this device is 7×10¹²cm^-2.

Download Full Size | PDF

The photogenerated carrier density inside the ring resonator can be extracted from the maximum observable shift of the center wavelength Δλ_max. Considering an effective refractice index of the waveguide mode n_eff=2.39, the maximum wavelength shift of Δλ_max=−1.1 nm corresponds to an effective refractive index change Δn_eff=−1.69×10⁻³ and to a change of the silicon refractive index Δn_Si=−1.41×10⁻³.

Using Eq. (1), in which ΔNe,h present the electron/hole densities in units of cm^-3, this refractive index change results from a maximum free carrier density of ΔN_e,h=4.1×10¹⁷ cm^-3 [22], which corresponds to an absorbed energy of 0.43 pJ per pulse. For comparison, we have estimated the amount of laser energy absorbed in the microring by taking into account the spot size, microring dimensions, and the silicon absorption coefficient of α_Si=1500 cm^-1 at 800 nm wavelength. For the adjusted laser fluence an amount of absorbed energy per pump pulse of 1.0 pJ is estimated, which is in good agreement with the value calculated from the resonance shift.

Figure 6(a) depicts the center wavelength of the microring resonance as function of time delay. The time-dependent wavelength shift is a direct measure for the free carrier density in the implanted microring structure. Applying an exponential fit to the wavelength shift recovery (shown as red solid line), results in an exponential decay constant, i.e. a free carrier lifetime, of 15 ps for the chosen implantation dose. To our knowledge, this lifetime is about two times shorter than carrier lifetimes reported so far in silicon waveguides realized either by carrier sweep out or by ion implantation [9, 10].

From Fig. 6(a) a small gap between the center wavelengths at positive and negative time delays of about 50 pm can be identified. Laser induced heating of the resonator can be excluded to induce this gap as the thermo-optic coefficient of silicon at infrared wavelengths would cause a red-shift of the resonance. Hence, we attribute the observed gap to the presence of residual charge carriers in the resonator waveguide even at longer time delays of 100 ps. The slightly unmatched overlap of the implantation and mode profiles may result in a small volume fraction of charge carriers that exhibit longer lifetimes within the waveguide.

Figure 6(b) illustrates the drop port intensity modulation at a fixed wavelength of λ=1554.36 nm. The maximum modulation depth of the device at zero time delay is 10.7 dB.

 

Fig. 6. (a) Time-resolved center wavelength of the drop channel (black) and corresponding exponential fit (red line). (b) Time-resolved modulation for an operating wavelength of 1554.364 nm. Both data are extracted from Fig. 5.

Download Full Size | PDF

A switching time, which we define by the level of 15 % of the maximum modulation depth, of 25.4 ps is achieved. This represents the shortest experimentally demonstrated switching time of a free carrier injection based SOI microring resonator. Furthermore this switching speed is even competitive with the recently proposed sophisticated diode design for a high-speed electro-optic microring based modulator [8].

 

Fig. 7. (a)-(c) Time-dependent center wavelength shifts Δλ (black) and corresponding exponential fits (red) for mircroring resonators implanted with different implantation doses of (a) 5×10¹²cm^-2, (b) 1×10¹²cm^-2 and (c) 3×10¹¹cm^-2. Extracted carrier lifetimes τ are noted. (d) Normalized wavelength shifts for all four implanted microring resonators.

Download Full Size | PDF

To analyze the influence of different implantation doses on carrier lifetime and additional waveguide loss further microring resonators have been implanted by oxygen ions at implantation doses of 3×10¹¹ cm^-2, 1×10¹² cm^-2 [20] and 5×10¹² cm^-2.

Figure 7 shows, in analogy to Fig. 6(a), temporal evolutions of the center resonance wavelengths of these microring resonators after optical excitation. Excitation powers were chosen to shift the microring resonances by at least one FWHM (full width half maximum) in each experiment, thereby taking into account decreasing quality factors with increasing implantation doses that result in enhanced waveguide losses (see also Fig. 9). Carrier lifetimes of 190 ps, 50 ps, and 33 ps were found by exponentially fitting the wavelength shift recovery after optical excitation of electron-hole pairs (red solid lines). As expected, these carrier lifetimes decrease with increasing implantation doses. As all resonances have been shifted by at least their corresponding FWHM, different absolute wavelength shifts of (a) Δλ=−0.7 nm, (b) Δλ=−0.75 nm, and (c) Δλ=−0.28 nm are obtained. Figure 7(d) illustrates the decrease of resulting carrier lifetimes with increasing implantation doses by plotting the normalized wavelength shifts of all four implanted ring resonators.

The corresponding time-dependent modulation amplitudes measured at the center wavelengths of the lower implanted microring resonators are shown in Fig. 8(a)-(c). In Fig. 8(d) the normalized modulation of all four implanted microring resonators are summarized. In analogy to Fig. 7, an improvement in switching performance with increasing implantation doses is clearly visible and switching times of 25 ps, 60 ps, 85 ps and 280 ps are obtained – again determined by the intensity levels at which the signals reach 15 % of their maximum modulation depth (black dashed line Fig. 8(d)).

 

Fig. 8. Time-dependent modulation characteristics for mircroring resonators implanted with different implantation doses of (a) 5×10¹²cm^-2, (b) 1×10¹²cm^-2 and (c) 3×10¹¹cm^-2. (d) Normalized modulation of all four implanted microring resonators.

Download Full Size | PDF

Carrier lifetimes extracted from the wavelength shifts for different implantation doses are shown in Fig. 9(a), together with carrier lifetimes measured in unpatterned oxygen-ion implanted SOI substrates. The latter data were obtained from time-resolved measurements in a standard pump-probe setup in reflection geometry similar to the one described in Ref. 19. Both sets of measurement data exhibit an inversely proportional dependence of implantation doses on the free carrier lifetimes, which is in agreement with results reported earlier [19] (s. Fig. 9(a)).

 

Fig. 9. (a) Measured inverse carrier lifetime over implantation dose for investigated microring resonators (red circles) and unstructured SOI samples (black squares), the dashed line is only a guide for the eyes. (b) Additional by oxygen implantation introduced losses for the different implantation doses σ. The black solid line represents a σ^0.63 fit of the measured data and the inset shows the same graph on a linear scaling of the x-axis.

Download Full Size | PDF

Finally, the influence of ion implantation on the additional waveguide propagation losses is studied. To this end, ring resonators have been investigated before and after implantation as reported in chapter 2. The additional propagation losses in the microrings versus implantation dose σ are plotted in Fig. 9(b) (red circles). The additional propagation losses increase with increasing implantation dose, and they are found to be in good agreement with a numerical fit that assumes a σ^0.63 dependence (black solid line). We compare this dependence with an earlier investigation of implantation induced densities of defect states [23, 24]. For silicon, this density generally scales with σ^0.63 for different implantation materials including oxygen ions. As the additional waveguide losses are expected to scale linearly with the defect state density we can unambigiously identify the increased propagation losses to be aroused by the implantation induced defects. In order to point out this root-function like dependency the inset of Fig. 9(b) shows the same data for a linear scaling of the x-axis.

Figure 9 clearly shows the trade-off between short carrier lifetimes and additional propagation losses, which has to be taken into account when choosing an ion implantation dose. Ring resonators implanted at higher implantation doses of 1×10¹³ cm^-2 have also been investigated within our study. However, due to the high additional losses no viable resonances were observed in a continuous-wave inspection of the investigated devices. Thus, we believe that the presented oxygen-ion implantation dose of 7×10¹² cm^-2 is close to the highest feasible value with respect to the additional introduced losses and a reasonable modulation depth.

On the other hand, it has been shown recently that a post annealing step of implanted photonic waveguide structures can significantly reduce introduced losses at a moderate increase in carrier lifetime [10, 18]. These results have been reported for helium and argon as implantation materials. Although carrier lifetimes of below 1 ps have been measured in argon implanted bulk silicon [25], the highest reported switching speed in argon implanted and post-annealed silicon waveguides was restricted to 70 ps at a feasible optical performance.

In order to evaluate the potential of a similar process for oxygen implanted waveguides, we have investigated the influence of different annealing processes on the carrier lifetime in unpatterned oxygen-implanted SOI samples. The samples were implanted with oxygen ion doses ranging from 5×10¹² cm^-2 to 1×10¹⁴ cm^-2 and were annealed in a rapid thermal annealing process for 30 s at temperatures ranging from 700°C to 1000°C. We found, that the O+ implantation induced short carrier lifetimes were completely annihilated by the different annealing steps. Compared to the rather inert implantation materials helium or argon, oxygen reacts rather sensitively with silicon in the annealing process. Thus, we believe that the development of a possible post-implantation annealing step for oxygen implanted waveguides which improves device performance would need further extensive studies which are beyond the scope of this work.

5. Conclusion

We have presented high-speed switching in silicon-on-insulator microring resonators with a switching speed of 25 ps at an extinction ration of 9 dB. This switching is achieved by reducing the carrier lifetime via oxygen ion implantation to only 15 ps. In addition we have pointed out the trade-off between implantation induced losses and short carrier lifetimes by investigating implantation doses between 3×10¹¹ and 7×10¹² cm^-2. A carrier lifetime of 15 ps is closely to the minimum achievable at considerable losses and device performance without further technological treatment.