1. Introduction

Silicon photonics offers the prospect of combining multiple optical and electrical functions on a single chip to achieve minimal size and maximum cost-effectiveness in device fabrication. Every essential optical function has now been demonstrated either in silicon or in silicon-based hybrids, including efficient fiber-chip couplers [1,2], lasers [3], detectors [4], variable optical attenuators [5], and high-speed modulators [6,7].

Silicon-based detection at standard telecommunications wavelengths (i.e. 1530 nm – 1570 nm) has generally been pursued for high-speed terminal receivers using either germanium-on-silicon [8], hybrid (III-V) silicon waveguide photodetectors [9] or defect-mediated sub-band detection [10]. Although grating-coupled silicon waveguide photodetection has been demonstrated for wavelengths around 850 nm [11], it has yet to be implemented for defect-mediated detection at sub-band wavelengths. This approach offers the advantage of being remarkably easy to implement, requiring only two additional back-end process steps to transform a silicon waveguide-based diode into a waveguide-based photodiode operable at 1550 nm [12]. A logical progression in the use of such detectors is their integration with other optical functions and with silicon electronics so as to take full advantage of the inherent versatility and compactness of silicon photonics.

One such application is the dynamic optical channel equalizer. Power variations versus wavelength can occur in telecommunication links due to slope in the amplifier gain spectrum (and are worsened by subsequent amplifier stages). This can be compensated either by using multiple amplifiers with complementary gain spectra [13] or attenuating with an appropriately tuned filter [14]. However, fluctuations can also occur due to laser or amplifier drift and link damage. In these cases a dynamic channel equalizer operating at standard telecommunication wavelengths can be used after wavelength demultiplexing to mitigate power fluctuations at the receiver on an individual channel-by-channel basis.

Such dynamic channel equalizers have been demonstrated using both MEMS-based devices [15] and planar waveguide devices in silica [16]. More recently a dynamic channel equalizer has been demonstrated that employs a defect-mediated photodetector in silicon and an external electronic VOA [17].

In the current work we report the monolithic integration of a variable optical attenuator (VOA) with a defect-mediated photodetector such that the detector signal can be used to control the VOA output without tapping any optical power from the output path. We further demonstrate the performance of this device as a “black-box” optical channel leveler over wavelengths ranging from 1530 nm to 1570 nm using an external digital controller that mimics a simple analog feedback loop. This is the first demonstration of a defect-mediated photodetector monolithically integrated with a VOA in silicon for monitoring purposes, and also the first demonstration of such a device used with a feedback loop for channel leveling.

2. Device design

2.1 Concept

The device incorporates a thermo-optically tuned Mach-Zehnder Interferometer (MZI) as the VOA. The photodiode is situated on the cross-state output while the output fiber is grating-coupled to the bar-state output as shown in Fig. 1 . With the MZI biased such that output power is several dB below peak output, the VOA can compensate for input power fluctuations (as monitored by the photodiode) by allocating power between the cross-state and bar-state paths so as to maintain constant bar-state output. A feedback loop between the photodiode and the VOA therefore enables channel leveling functionality.

 

Fig. 1 Schematic of the silicon photonic channel leveler showing grating-coupled input and output (bar-coupled path), MZI with tuner, and defect-mediated photodiode (cross-coupled path). The overall length of the MZI is 2.5 mm.

Download Full Size | PDF

2.2 MZI considerations

Transmission (bar-state) through the MZI is described by Eq. (1):

From Eq. (1) it is clear that for θ an odd multiple of π, T will be maximized and equal to 1, whereas for θ an even multiple of π, T will be minimized and equal to 1 – 4k(1-k). Thus the maximum cross-state (minimum bar-state) output is limited by deviation of the directional couplers from 50% coupling (i.e. k = 0.5) such that there is a trade-off between extinction ratio in the bar-state and excess loss in the cross-state. The cross configuration (i.e. fiber coupled to the cross-state output) guarantees high extinction at the expense of peak transmission, while the bar configuration (i.e. fiber coupled to the bar-state output) guarantees maximum throughput at the expense of extinction ratio. We demonstrate the device in the bar configuration to maximize throughput rather than extinction ratio.

2.3 Monitor considerations

With the photodiode located on the fiber output path, photocurrent will generally be proportional to output power such that a feedback loop that lowers output power with increasing photocurrent is straightforward. However this approach necessarily involves the drawback that power must be tapped from the output to the photodiode [17]; this power is over and above the power already sacrificed in the VOA.

There will necessarily be a trade-off, therefore, between the photodiode signal-to-noise ratio and the optical throughput with the monitor situated on the fiber output path. This trade-off can be circumvented by positioning the photodiode on the opposite (in this case cross-coupled) output path as shown in Fig. 1. With this architecture the power sacrificed in the VOA is routed to the photodiode (which can then be designed to tap as much power as possible from the cross-coupled output) such that sufficient monitor signal for leveler operation can be achieved with lower input power. The disadvantage to this approach is complexity: the relation between photocurrent and output power is no longer straightforward since increased photocurrent could imply either increased input power and hence excess output power, or increased coupling to the photodiode and hence a deficit of output power. This ambiguity can be resolved by taking bias voltage into account in the feedback loop as described below.

2.4 Leveling algorithm

For operation in the linear region near quadrature, output power and photocurrent are approximately proportional to tuner V = V_applied – V_quadrature according to Eq. (2) and Eq. (3):

Setting P = P_ref for arbitrary input power α yields a simple iterative solution for tuner bias voltage according to Eq. (4):

Initial measurements of dP/dV and di/dV, and i_ref at quadrature for a reference output power can therefore be treated as constants for a feedback loop which applies tuner voltage according to Eq. (5):

3. Fabrication

The devices were manufactured using the ePIXfab platform offered via LETI [19]. This platform uses standard silicon processing of <100> 220 nm silicon on insulator (SOI) wafers for the fabrication of silicon photonic waveguides. The optical rib waveguides were 500 nm wide and etched to a depth of 170 nm to leave a thin 50 nm slab for electrical connection. Diodes were fabricated by low energy boron and phosphorus implantation on either side of the waveguide followed by an activation anneal. After contact metallization mid-gap electrical states were introduced by means of a final implantation step, specifically 350 keV boron to a dose of 1×10¹³ cm⁻², using a thick resist to mask the chip. The energy of the boron implantation positioned the dopant below the silicon overlayer, and hence this final process step merely introduces defects into the waveguide for responsivity enhancement. There is no effect on the device due to the boron dopant.

4. Results and discussion

4.1 Loss and responsivity

Fiber-to-fiber loss of straight waveguides which did not receive the final defect inducing implantation was measured to be 17 ± 1 dB for unpolarized 1550 nm SMF input. The gratings couple TE polarized light to the waveguide almost exclusively [20], so for the unpolarized input used in this work there is 3 dB polarization dependent loss at the input grating and 0.24 dB/mm propagation loss [21] for the 3.8 mm between grating-couplers, so coupling loss at the gratings is 6.5 ± 0.5 dB. This is comparable to the 6 dB coupling loss reported by Laere et al [20] for similar grating couplers.

Responsivity of the photodetectors on test straight waveguides was determined to be 18 ± 2 mA/W where responsivity is here defined as photocurrent (0.57 ± 0.01 μA at 10 V reverse bias) divided by optical power incident on the photodiode (calculated as input optical power of −5 dBm minus input losses comprised of 3 dB polarization dependent loss, 6.5 ± 0.5 dB grating coupling loss, and 0.6 dB propagation loss (the diode is situated 2.6 mm from the input coupler). Excess loss for the photodiode was measured to be 16 ± 1 dB. Quantum efficiency is therefore 0.02. Dark current for these devices was consistently less than 1 nA and can be neglected for this application. A device with the same geometry which did not receive the final defect implant was measured for comparison and found to have responsivity of 0.37 ± 0.05 mA/W. The implantation step therefore improves responsivity by a factor of 50. Both the responsivity enhancement and the excess loss are attributed to optical excitation of carriers from defect states within the bandgap.

The responsivity observed here is somewhat higher than that reported by Bradley et al. [4,22] but considerably lower than the value of 0.8 A/W reported by Geiss et al. [10] due to the lower defect density produced by a boron versus silicon implant at comparable doses. Integration of functionality, rather than maximum responsivity was the goal of this work, however it is encouraging that the channel leveler operates successfully with these parameters while higher photocurrents are achievable if necessary.

4.2 VOA operation

Figure 2 shows the output (externally measured optical power and photocurrent) for a typical device in the bar-coupled configuration with the photodiode on the cross-coupled output. Tuning efficiency is 63 mW/π while optical extinction is 13 dB at 73 mW tuning on the MZI. The photocurrent shows an extinction ratio of 6 (versus optical extinction of at least 20). Since the minimum photocurrent is much larger than dark current we attribute the extinction degradation to scattered radiation from the gratings or other features on the chip.

 

Fig. 2 Optical output and photocurrent versus thermal tuning for silicon photonic monitored VOA with 1550 nm unpolarized input.

Download Full Size | PDF

The VOA easily tunes through a 6π MZI phase shift with some variation in maximum T evident particularly at 250 mW thermal dissipation and above. We attribute this to heating of the directional couplers, thus altering k and therefore T as described in section 2.2. The capability to monitor MZI output with an on-chip photodiode is useful in its own right since it makes wholly electrical control of the optical output possible for applications such as 2x2 digital switching without need for a separate optical tap. The VOA has a 3 dB bandwidth of 110 KHz which would limit overall device speed in an integrated circuit configuration.

4.3 Channel leveler operation

The device was operated as a channel leveler using the setup and virtual circuit feedback loop shown in Fig. 3 . The feedback loop adjusts bias according to the photocurrent and Eq. (5) for which the constants c_1, c₂, and c₃ have been calculated from i_ref, dP/dV, and di/dV measured in advance at quadrature (44 mW or 2.6 V for the device characteristic plotted in Fig. 2). Results over 7-10 dB of dynamic range at wavelengths from 1530 nm to 1570 nm are shown in Fig. 4 (the above-mentioned constants are re-measured for each wavelength).

 

Fig. 3 Test setup for channel leveler (left) using virtual circuit (right) used for feedback with constants C₁, C₂, and C₃ given by Eq. (4) and Eq. (5). ECL=external cavity laser, PS=polarization scrambler, DUT=device under test, Amm=ammeter.

Download Full Size | PDF

 

Fig. 4 Output power variation vs. input power variation for wavelengths from 1530 to 1570 nm across a 7 - 10 dB dynamic range. The inset shows a close-up for 1530 nm – 1550 nm with error bars denoting standard deviation over the course of 90 feedback loop iterations.

Download Full Size | PDF

The leveler holds output power within ±1 dB across a 7 dB dynamic range for the full 40 nm wavelength range tested, and across a 10 dB dynamic range for wavelengths from 1530 nm to 1550 nm as shown in Fig. 4 or within ±0.3 dB for input power variation of ±2 dB for 1530 nm to 1550 nm as shown in the inset to Fig. 4.

The extent to which the leveler can compensate near the top of the dynamic range is dictated by VOA extinction ratio, which in turn is dictated by the MZI directional coupler coefficients as described in section 2.2. Since these coupling coefficients are wavelength dependent the channel leveler’s performance at high input power improves as the couplers approach 50%. Figure 5 shows the correlation between VOA extinction ratio and the leveler circuit’s dynamic range for ± 0.5 dB tolerance as the extinction ratio varies with wavelength. This relationship explains the degradation in leveler performance at longer wavelengths for high input power. The dynamic range of the leveler can therefore be improved by tailoring the directional couplers in the MZI to k = 0.5.

 

Fig. 5 Leveler dynamic range (using +/− 0.5 dB cutoff) and VOA extinction ratio vs. wavelength. Note that the dynamic ranges shown for 1530 nm and 1540 nm are conservative since the leveler output is flat at the highest input power tested.

Download Full Size | PDF

The signal-to-noise ratio (SNR) of the photodiode determines the channel leveler’s performance at the lower end of the dynamic range since power is redirected away from the cross-coupled path to the bar-coupled path leaving very little photodiode signal and invalidating the linear approximation in Eq. (3). SNR can be improved by adding a spatial filter (e.g. metal features on each side of the waveguide) near the photodiode to reduce signal produced by scattered light.

Cascading several channel levelers in parallel can be used to level the gain for a range of demultiplexed channels across the C-band, while cascading devices in series offers the possibility of improved extinction ratio and dynamic range. Since each leveler can be integrated on-chip, additional excess loss is determined largely by the dynamic range required at each step (e.g. to compensate for ± 3 dB power variations the VOA must be biased at least 3 dB below peak transmission). In the current configuration (i.e. using a thermal MZI) heat dissipation poses problems for such architectures, however, since each leveler consumes 50 mW of thermal power. This may be mitigated through the use of VOAs which consume less power per channel (e.g. by designing the MZI such that quadrature falls at or near zero tuning bias).

5. Conclusion

An optical channel leveler integrating a VOA and a photodiode has been demonstrated in silicon using an external feedback loop between a monolithically integrated waveguide photodiode and VOA. The feedback loop mimics a simple circuit that can be incorporated on the chip, and has the advantage of monitoring excess power in the VOA rather than tapping power from the output path, thus minimizing optical loss. The device is entirely compatible with standard silicon processing, has a size of 6 mm x 0.1 mm per channel, and successfully holds output power to within ± 1 dB across a 7-10 dB dynamic range for wavelengths from 1530 nm to 1570 nm.