1. Introduction

Coherent technology is widely deployed for long-haul and now metropolitan networks due to its improved performances: better spectral efficiency, higher sensitivity, capability to compensate chromatic dispersion (CD) and polarization mode dispersion (PMD) [1,2]. Coherent receiver technologies are mainly based on photonic integrated circuits made on InP material [3,4], silicon photonics [5] or PLC using hybrid integration [6,7]. Most coherent systems operates at 28-32 Gbaud which allows a data rate transmission at 100-200 Gbit/s using respectively DP-QPSK and DP-16 QAM modulation scheme.

With the increase of the bandwidth demand in short and medium reach links, the capability of coherent technology to compensate CD and PMD and its increased sensitivity compared to direct detection technology become attractive. To increase the reach of unamplified links or the splitting ratio in access networks, optically preamplified coherent receiver is a promising option. Indeed, even if the local oscillator boost the receiver sensitivity, there is a practical limit its output power (≈17 dBm currently) and boosting significantly it will also impact strongly their power consumption. Assessment of SOA preamplification in coherent receiver (LO power 16 dBm) demonstrates 3 dB sensitivity improvement using a SOA with 8.2 dB noise Fig [8]. which allow to increase the maximum transmission reach by 15 km or to double the number of subscribers in an access network. Another trend in coherent communication is the increase of the baud rate to lower transmission cost and first demonstration of high speed coherent receiver with 100 Gbaud capability has been published recently [9]. This will increase photodiode bandwidth requirements above the ≈25 GHz used for standard coherent receiver.

A few demonstrations of coherent receivers with integrated SOA and tunable lasers has been published [10–12] but no significant improvement of responsivity compared to state of the art coherent receiver (0.18 A/W in [10] versus 0.14 A/W in [1]) have been demonstrated so far. In addition, photodiodes bandwidth are still low (10 GHz [10,11] and 35 GHz [12]).

In this contribution, leveraging our SOA-PIN integration platform [13,14] we developed a coherent receiver with an integrated SOA and high speed photodiodes to boost simultaneously receiver sensitivity and speed for >56 Gbaud baud rate operation. In our integrated coherent receiver, we achieve for the first time responsivity above 4 A/W and 40 GHz bandwidth. To assess coherent detection capability, detection of QPSK signal up to 32 Gbaud with error free operation was also demonstrated.

2. Receiver design and fabrication

Figure 1 shows the layout view of our integrated coherent receiver. It comprises two spot-size converters (for the signal input and the local oscillator input respectively), an SOA on the signal path, a 2 × 4 MMI for the optical 90° hybrid and four UTC photodiodes. The overall chip size is 1.2 mm × 4 mm (4.8 mm²) which is comparable to InP coherent receiver without SOA (1.6 × 4.1 mm² in [1] and ≈0.8 × 4.4 mm² in [15]). SOA and passive waveguides are realized in buried technology [13, 14]. The SOA is 500 µm long, 1.1 µm wide and the active structure comprises a 120 nm InGaAs thick tensile bulk layer sandwiched between InGaAsP layers as seen in Fig. 1 [13]. Passive waveguides used for ligth routing and 90° hybrid are made with a thick InGaAsP layer (λ_G = 1.3 µm). A low confinement fiber waveguide made with a thin InGaAsP layer allows low coupling losses with an optical fiber (≈1.5 dB). Relaxed geometry of 5 × 25 µm² are used on our deep ridge UTC photodiode to increase fabrication yield without compromising speed response (>40 GHz).Their vertical structure comprises a 200 nm InGaAs absorption layer and a 300 nm InGaAsP collector layer.

 

Fig. 1 Top-schematic view and cross section of our preamplified integrated coherent receiver (1.2 × 4 mm²).

Download Full Size | PDF

The fabrication process starts with GSMBE (gas-source molecular beam epitaxy) growth of SOA active waveguide on a semi-insulating InP substrate. InGaAsP passive waveguide and UTC photodiodes are successively regrown using butt joint technique made by MBE. The SOA waveguide, the passive waveguide and the fiber waveguide are defined by RIE plasma etching and buried by P-MOVPE regrowth followed by H + implantation to define the P-N junction area [13]. Passive waveguide are hydrogenated to neutralize Zn dopant in P-InP which reduces intervalence band absorption (IVBA) losses. The photodiode junctions are etched using dry and wet etching and passivated by a SiN_x CVD layer. We implement coherent receiver with and without SOA to assess its influence on the receiver performances.

3. Results

We present UTC photodiode speed response measurements to assess their intrinsic properties. Measurements were made using a lensed fiber (beam diameter 4.2 µm) to couple the light into the photodiode and a responsivity of 0.37 A/W with a low PDL (<0.7 dB) was demonstrated (without AR coating). This corresponds to a responsivity of 0.5 A/W with an AR coating, comprising coupling losses with the optical fiber and the photodiode internal quantum efficiency.

We performed S11 measurements up to 65 GHz using a vector network analyser and a photodiode capacitance and resistance of respectively 58 fF and 16Ω are extracted which result in a RC cut-off of 41 GHz. Figure 2 represents the frequency response (measured with an heterodyne setup) of a single 5 × 25 µm² photodiode and compare it with the frequency response of the coherent receiver. A 3-dB bandwidth above 40 GHz for both test diode and coherent receiver, compatible with 56 Gbaud applications, is demonstrated, in accordance with equivalent circuit extraction. The slightly higher roll off observed for the coherent receiver is associated to its longer transmission line which were affected by process deviation during gold deposition and which can easily solved in a next fabrication run.

 

Fig. 2 Left: Frequency response of a 5 × 25 µm² photodiode (I_ph = 1 mA, R = 0.5 A/W); Right: frequency response of the 4 photodiodes of a coherent receiver.

Download Full Size | PDF

The characterization of coherent receivers with and without integrated SOA is made to assess the intrinsic characteristics of our receiver and the impact of the integrated SOA. Diced receiver chips were tested directly on a test support regulated at 20°C. Figure 3 shows the responsivity and imbalance of our receiver without SOA in the optimal polarization state. A low imbalance <1 dB between 1530 and 1600 nm is achieved for both LO and signal inputs. In the LO input, a very low imbalance around 0.2 dB is demonstrated over the full C-band. The responsivity is close to 0.04 A/W for the LO input and is up to 0.045 A/W on the signal path. The difference between the two inputs is probably due to the bends which are more pronounced on the LO path. Compared to an ideal coherent receiver (0.31 A/W), this represent 8.9 dB excess losses. As single photodiode responsivity is 0.5 A/W, 4-dB losses can be associated to fiber coupling losses and limited quantum efficiency of the photodiode and the additional 4.9 dB to MMI and waveguides excess losses. Waveguides losses are probably due to insufficient neutralization of Zn dopant by H + hydrogenation.

 

Fig. 3 Responsivity and imbalance of a coherent receiver without SOA preamplifier (left: LO input, right: signal input).

Download Full Size | PDF

Figure 4 shows responsivity of a coherent receiver with integrated SOA. No AR coating was applied on the receiver input waveguides. A record responsivity of 4 to 5 A/W were demonstrated at a SOA bias current of 140 mA (power consumption: 240 mW) and an input power of −20 dBm which corresponds to 5.5 to 6.8 A/W with AR coating. Therefore, the SOA gain is above 20 dB for a drive current of 140 mA. This high performance compared to results published in [10] can be attributed to the following points: decrease of coupling losses due to sport size converters, lower propagation losses due to a higher bandgap waveguide (λ_g = 1.3 µm vs 1.4 µm) and reduction of IVBA losses by hydrogenation and lower efficiency of their offset quantum well structure (low confinement in the QW) for SOA and PD. The imbalance is below 1 dB between 1540 nm and 1595 nm. The higher imbalance on the LO input compared to the component without SOA is associated to small deviation on MMI width during fabrication. On the LO path, the responsivity is comprised between 0.008 and 0.01 A/W (>0.011 A/W with an AR coating). The degradation (≈6 dB) compared to coherent receiver without SOA is due to an unwanted proton implantation in the LO waveguide.

 

Fig. 4 Responsivity and imbalance of a coherent receiver with integrated SOA (left: LO input, right: signal input).

Download Full Size | PDF

Figure 5(a) shows the influence of the SOA drive current on the photodiode responsivity for 2 different input powers. The optimal responsivity is reached for a SOA drive current of 140 mA. The degradation of the responsivity above 140 mA is due to thermal effects. As thermal dissipation of diced chips mounted directly on test support is significantly less efficient than thermal dissipation of Chip on Carrier (CoC) or CoC into module, higher responsivity could be reached in a packaged receiver. At a very low (60 mA) drive current (80 mW power consumption) and −20 dBm input power, the responsivity is slightly above 1.25 A/W (without AR coating) corresponding to > 6 dB gain compared to ideal coherent receiver responsivity and > 9 dB gain compared to state of the art 100 Gbit/s coherent receiver [1]. As high speed/high gain TIA usually consumes more than 200 mW [16] (2 TIA required per polarization resulting in>400 mW power consumption for a single polarization coherent receiver), it can be interested to realize a first optical amplification stage to lower the required TIA gain and therefore their power consumption. Figure 5(b) shows the variation of coherent receiver responsivity versus the input optical power. The 3-dB saturation input power is above −4 dBm (−5.5 dBm assuming ideal AR coating). This high saturation power should prevent the receiver from non linear distortions of the optical signal in the SOA.

 

Fig. 5 Responsivity versus SOA drive current (a); Responsivity versus optical input power (b).

Download Full Size | PDF

As coherent receiver use a polarization beam splitter (PBS), usually followed by a polarization rotator on one output of the PBS, coherent receivers can work in a single polarization. Therefore, the power consumption of our SOA can be further optimized by using compressively strained MQW to amplify a single polarization and which would be more energy efficient. Furthermore, this paves the way for the integration of the local oscillator laser as it can share its active layers with the SOA.

Finally, system measurements using QPSK modulation were performed to demonstrate the capabilities of our integrated coherent receivers (Fig. 6). The SOA drive current is fixed at 140 mA in our preamplified receiver and the temperature is regulated at 20°C. The QPSK signal was generated by a high speed DAC which modulate a commercial PDM-IQ modulator. Two external cavity lasers are used for the signal and the local oscillator. The modulated signal and the local oscillator are launched into our coherent receiver which is connected to a real time oscilloscope (40 Gsample/s, 16 GHz bandwidth). Due to the absence of TIA between the integrated coherent receiver and the real time oscilloscope, we are limited by the output signal of the receiver so we can’t measure sensitivity of the receiver in practical environment. Due to this limitation, the signal input power is 8 dBm on the receiver without SOA and −10 dBm on the receiver with an SOA. Figure 6 shows that our receiver is able to detect 17 Gbaud and 32 Gbaud constellations, (both with SOA and without SOA). The optimal Q factor is similar for both receivers with only a very low penalty (<0.7 dB) due to the SOA noise. With a Q factor above 17 dB at 17 Gbaud and above 14 dB at 32 Gbaud, we are below classical FEC threshold (typically 11.5 dB corresponding to BER = 10⁻⁴). We observe a degradation of the constellation at 32 Gbaud, probably due to the limited bandwidth of the real time oscilloscope. With 40 GHz photodiodes bandwidth, our receiver should be compatible with 56 Gbaud applications.

 

Fig. 6 a) QPSK constellation of coherent receiver without SOA; b) QPSK constellation of coherent receiver with SOA (140 mA drive current).

Download Full Size | PDF

In [8], an improvement of 3 dB of receiver sensitivity was demonstrated using a SOA with 8.2 dB noise figure which is the typical noise figure of our SOAs [13,14]. Therefore, compared to standard coherent receiver, our new chip will allow an increase of the transmission reach of unamplified links by 15 km or to double the number of customer in point to multipoint networks like access networks.

4. Conclusion

We demonstrate a preamplified coherent receiver integrated on InP substrate with a record responsivity of 5.5 to 6.8 A/W (with AR coating). An improvement of 12.5 dB and >15.5 dB are respectively achieved compared to ideal coherent receiver and state of the art coherent receiver. The power consumption is moderate (240 mW) and can be optimized by an improvement of the active structure. The high bandwidth of our photodiodes (40 GHz) make it compatible with 56 Gbaud applications and can be improved for 100 Gbaud by reducing photodiode sizes. A high input saturation power (−4 dBm) is demonstrated and our preamplified receiver detects a 32 Gbaud QPSK modulation with Q>14 dB, limited by the measurement setup. This work pave the way for highly efficient preamplified detector for short to medium reach links and for the integration of the local oscillator to further reduce cost and complexity of coherent receivers.