Open Access
Add to BibSonomyAbstract
We present an avalanche photodiode (APD) with high-speed, high-responsivity and high-linearity operation to cope with higher order modulation format, such as pulse-amplitude modulation (PAM). A hybrid absorber configuration with thin depleted region which we newly employed successfully eliminates the space charge effect in the APD while maintaining high responsivity and operating speed. The fabricated APD shows an improved optical-input-electrical-output linearity for an optical input power over −8 dBm, and an optical receiver with this APD achieves both an error-free operation with a KP4 FEC and a high sensitivity of −17 dBm against a 28-Gbaud PAM4 signal.
© 2015 Optical Society of America
1. Introduction
The huge growth of intra/inter-data-center IP traffic due to cloud and mobile services requires rapid progress in Ethernet systems, such as 100 Gbit/s Ethernet (100GbE) [1]. Although 400GbE is now extensively discussed [2], Ethernets with even larger capacity are also envisioned. A bit rate of 25 or 28 Gbit/s per a wavelength employed in 100GbE is not sufficient for such a situation, thereby higher bit rates are needed in order to cope with the capacity demands in Ethernet systems. Improvements in both the performance of receiver components and optimization of modulation formats have contributed to increasing the bit rate per wavelength. For example, 50-Gbaud optical transmitters and optical receivers have been demonstrated [3–5]. Regarding the modulation format, some higher order modulation (HOM) formats, such as pulse-amplitude modulation (PAM) or discrete multi-tone (DMT), have been proposed [6–8]. Such HOM formats are strongly expected to be applicable to the next generation of Ethernet systems because they will make it possible to increase the bit rate by using optical devices with relatively low bandwidth performance.
Another requirement for future Ethernet systems is extension of the transmission distance while maintaining low levels of power consumption and small package sizes. As transmission speed increases in optical fiber communications systems, the transmission distance generally becomes short because of the distortion of optical signals in the fiber and limitation of launch power in the transmitter. This issue can be resolved by using fiber amplifiers or semiconductor optical amplifiers (SOAs) to boost the power of optical signals. The drawback is that these amplifiers consume large power. Furthermore, integrating an SOA into an optical receiver will require a large area in optical transceivers. Thus, although the use of the optical amplifiers will extend the transmission distance, they cannot meet the requirements of low power consumption and small transceiver size in Ethernet systems. Avalanche photodiodes (APDs) are attractive alternatives because of their higher responsivity than that of conventional photodiodes (PDs) due to their internal avalanche gain, and because of their small size and low power consumption, which are comparable to those of PDs.
The requirements for APDs in 400GbE and beyond and those in the standardized 10GbE, 40GbE and 100GbE differ in that the former must have high linearity, as well as high responsivity and speed in order to employ HOM formats. Generally, APDs have poor linearity compared with conventional PIN-PDs or uni-travelling carrier (UTC)-PDs [9, 10]. This is because the gain of APDs degrades at higher optical input powers. This behavior can be explained as follows. Figure 1(a) and 1(b) respectively show schematic band diagrams around the avalanche layer when the optical input power is low and high. When the optical input is low, the photo-generated holes in the absorption layer and the holes multiplied in the avalanche layer run through the whole absorption layer with hole-drift saturation velocity. In this situation, the APD shows good linearity comparable to that of a conventional PIN-PD. On the other hand, when the optical input is high, the holes around the p-doped field control layer, in which the holes are far from the p-type contact layer, accumulate. These accumulated holes cancel out the charges of acceptors in the depleted p-doped field control layer at the operating condition. This results in lowering the electric field of the avalanche layer and thus in degrading the gain of the APD. For the reasons explained above, the linearity of the APD deteriorates.
The degradation of linearity in APD at a high input power was experimentally investigated in a previous study [11]. However, as far as we know, methodology to avoid such degradation has not been discussed so far. Although using a UTC-type absorber or very thin depleted absorption layer can resolve such a problem, these solutions will affect the responsivity.
In this paper, we present a high-linearity APD in which we achieved high linearity while maintaining high responsivity and high-speed operation by using a hybrid absorber consisting of p-doped and undoped InGaAs absorbers. The fabricated APD exhibits a small degradation of less than 1 dB for optical input power of up to −8 dBm with a 3-dB bandwidth (f3dB) of 21 GHz and responsivity of 0.69 A/W at a gain of unity. Furthermore, the optical receiver module made with the APD successfully demonstrated receiver sensitivity of −17 dBm for a bit-error rate (BER) of 2.3 × 10−4 against 28-Gbaud PAM4 optical signals. These results indicate that our high-linearity APDs are applicable for future large-capacity Ethernet systems with high-baud-rate HOM formats.
2. Device design and structure
Figure 2(a) shows a schematic cross-sectional view of the fabricated APD. The epitaxial structure was grown on a semi-insulating InP substrate by using the MOCVD method. It consists of a p-type contact, p-type InGaAs absorption, undoped InGaAs absorption, p-type field control, 90-nm avalanche, n-type field control, edge-field buffer, and n-type contact layers. The total thickness of whole absorber is 600 nm. The doping concentrations of p-type and n-type contact layers are 1 × 1019 and 2 × 1019 cm−3, respectively. The p-doped absorber has graded doping profile; in the side of p-type contact layer, the doping concentration is 2 × 1018 cm−3, and it decreases toward the undoped absorber. The doping concentrations of p- and n-type field control layers are carefully adjusted so that the device can operate at a certain bias condition to avoid unexpected breakdown in absorber or edge-filed buffer layer. The configuration of the device is based on the inverted p-down triple-mesa structure [12]. The top mesa formed with the n-type contact layer, which defines the active area of the device, has a diameter of 20 μm. The key feature in the structure to simultaneously obtain high linearity, high speed, and high responsivity is a hybrid absorber consisting of p-doped and undoped InGaAs. A schematic band diagram of our APD is shown in Fig. 2(b). In the p-doped part in the hybrid absorber, only the photo-generated electrons, which are minority carriers, affect the carrier transport characteristics. In the p-doped part, the travelling delay time of the photo-generated holes and the multiplied holes from the avalanche layer is independent of the carrier transit time or CR time constant, and comparable to the dielectric relaxation time. This is because the holes are majority carriers in the p-doped layer; thus, the travelling delay time in it can be negligible. On the other hand, in the depleted absorber, the photo-generated holes and the multiplied holes from the avalanche layer dominantly affect the transport characteristics. It means that when we employ thin depleted absorber in the hybrid absorber, the carrier-transport characteristics can be improved. The distance with which the holes have to travel is just equal to the thickness of the depleted absorber, although the responsivity of the APD is determined by the whole thickness of the hybrid absorber composed of p-doped and undoped parts. Thus, our employed hybrid absorber relaxes the hole accumulation in the depleted absorber while ensuring high responsivity.
Fig. 2 (a) Schematic cross-sectional view of the proposed APD. (b) Schematic band diagram of proposed APD.
We determined the thickness of the undoped absorber so as to make the effective thickness of a depleted region in the absorber to be 400 nm at an operating bias condition. According to mechanism of degradation of linearity in APD, it is expected that the thinner depleted absorber provide further improvement of linearity, however, it will increase the carrier transit time in p-doped absorber. To avoid sacrificing bandwidth, we designed the depleted absorber thickness as above.
3. Device characteristics
First, we characterized the fabricated APD by measuring I-V, gain, and output linearity characteristics in the on-wafer condition. Figure 3 shows the obtained I-V and gain-voltage characteristics: the I-V characteristics are smooth without any unexpected breakdown, such as edge breakdown, even though the breakdown voltage is as large as 30 V, and there is a continuous increase of the gain to over 35. The gain reaches 10 at a voltage of 27.4 V.
To characterize the output linearity, we measured the photocurrent of the APD for various voltages and optical input powers. Figure 4 shows the optical input power dependence of the experimental photocurrent compression against the ideal photocurrent linearity. When the gain of the APD is low, the measured photocurrent maintains low compression almost in the whole measured range of optical input power. As the gain rises, the compression becomes significant even in the lower optical input power range. This behavior is different from that of conventional PIN-PDs. In general, conventional PIN-PDs show degradation of photocurrent only against high-power input because of band bending (namely, lowering of electric field) in the absorption layer caused by hole accumulation (space charge effect). Thus, the linearity can be improved if we increase the PD bias to increase the electric field in the absorption layer. In contrast, according to the measurement results shown in Fig. 4, the degradation of the linearity of the APD becomes pronounced as the gain (namely, the APD bias) increases.
Fig. 4 Optical input power dependence of linearity compression of fabricated APD for various gain conditions.
To examine the impact of the thin depleted absorber on relaxing the space charge effect, we compared the photocurrent compression of two APDs. One has a structure described above with a thinner (400-nm-thick) depleted absorber, and the other has a thicker depleted absorber. The latter has a total thickness of absorber of 1 μm, and the thickness of the depleted absorber is set to be twice of the thinner one. The bias voltages to each APD were set to have the same gains of 10. The thinner one maintains the compression of less than −1 dB up to the output photocurrent of 0.9 mA, while the thicker one shows the compression of −1 dB at output photocurrent of less than 0.46 mA. The significant improvement of the linearity is achieved by reducing the thickness of the depleted absorber against the whole thickness of the hybrid absorber. Note that the condition of 0.9 mA for thinner depleted absorber is corresponding to the input power of −8 dBm. Thus the proposed structure with thin depleted absorber obtains the 1-dB compression of photocurrent for optical input power over −8 dBm. See Fig. 5.
Fig. 5 Output photocurrent dependence of compression of the linearity of two APDs with different thicknesses of depleted absorber.
4. Application to HOM format transmission
4.1 Linearity characterization of optical receiver using the APD
We applied the fabricated high-linearity APD to the transmission test with an HOM format. An optical receiver module was assembled together with the fabricated APD and a trans-impedance amplifier (TIA). We chose a CAN-type package with LC-receptacle optical input and polyimide-film flexible printed circuit (FPC) based electrical output. The TIA is a commercially available linear-type one with a typical bandwidth of 23 GHz. Figure 6 shows the frequency characteristics of the APD in the on-wafer condition and those of the optical receiver module. The gain was set to 10 for each characterization. Although the frequency characteristics of the optical receiver module show abrupt roll-off above 18 GHz, the fabricated optical receiver module shows a marginal penalty on f3dB compared with that in the on-wafer condition. The obtained f3dB is sufficient for HOM format operation with a baud rate of 28 Gbaud.
Figure 7(a) and 7(b) show the frequency characteristics of the fabricated APD and the optical receiver module made with it for various optical input powers. The optical input power ranged from −20 to −10 dBm at a gain of 10. As shown in Fig. 7(a), the frequency characteristics show good linearity for the whole range of optical input power at on-wafer condition. In contrast, those of the optical receiver module show saturation for optical input power above −14 dBm. This behavior is considered to be due to the input saturation of the TIA.
Fig. 7 Frequency characteristics of (a) fabricated APD and (b) its optical receiver module for various optical input powers.
4.2 BER evaluation of APD optical receiver for 28-Gbaud PAM4
We evaluated the BER characteristics of the fabricated APD optical receiver module for HOM format signal. A 28-Gbaud (namely, 28-GSymbols/s) PAM4 signal was chosen for the evaluation because this modulation format, which requires linearity of the receiver, is a possible format candidate for future Ethernet systems. Figure 8 shows the experimental setup. The electrical 28-Gbaud PAM4 signal was generated by combining two binary pseudo-random bit sequences (PRBSs) with lengths of 215-1, one of which was attenuated by 6 dB. We used a 1.3-μm electro-absorption modulator integrated with a DFB laser (EML) featuring an electro-optic 3-dB bandwidth of over 30 GHz [13]. The power of the optical signal launched from the EML was + 3.15 dBm. The single-mode fiber was inserted between the EML and the variable optical attenuator (VOA), and the receiver sensitivities were defined by the VOA. The electrical signal output from the APD receiver module was sampled and digitized using an 80-GSample/s real-time storage oscilloscope with an analog bandwidth of 33 GHz. We demodulated the received signal by offline digital signal processing, in which we used an adaptive feed-forward equalizer (FFE). The eye diagram of the optical signals output from the EML with an optical power of + 2 dBm is shown in the inset of Fig. 8. The eye is clear and open.
Fig. 8 Experimental setup of BER measurement for 28-Gbaud PAM4 with high-linearity APD optical receiver module.
We measured the BER under two different conditions. One was constant APD bias, and the bias was adjusted to be the gain of 10 at the optical input power of –16 dBm (condition I). In this condition, the BER will degrade when the input optical power exceeds −14 dBm because the linearity of the optical receiver module will degrade as shown in Fig. 7(b). In the other condition, we adaptively changed the bias of the APD to control the gain in order to suppress the linearity degradation (condition II).
The measured BER characteristics are shown in Fig. 9. In condition I, the BER improved with increasing input optical power from −18 dBm, and it showed the best value at −14 dBm. Then the BER rapidly degraded with further increases of the optical input power; that is, the BER in condition I shows a poor dynamic range, although the receiver sensitivity is very high. In condition II, the BER improved up to the optical input power of −10 dBm. The BER penalty at lower optical input power was not observed. Thus, it was shown that adaptive changing the APD bias was an effective way to improve the dynamic range of the optical receiver without sacrificing receiver sensitivity. The difference in the dynamic range between the two conditions can be explained by the performance of the TIA: linearity or input saturation. According to the optical input power dependence of the OE response of the APD receiver module shown in Fig. 7(b), we attribute the degraded dynamic range to the input saturation of the TIA.
Fig. 9 BER characteristics of fabricated APD optical receiver module for 28-Gbaud PAM4 with different APD bias conditions.
The BER of less than 2.3 × 10−4, which corresponds to error-free operation with a KP4 FEC [14,15], was obtained at the received power of −17 dBm under both conditions. The loss budget reached as large as 20.15 dBm by considering the launch power of EML of + 3.15 dBm. This loss budget allows over 40-km transmission when we assume the fiber loss of 0.5 dB/km at a wavelength of 1.3 μm. The power consumption of the APD receiver is about 270 mW with over 98% of it consumed by the TIA. Thus, the developed APD offers a good solution for higher bit-rate transmission systems employing HOM formats without significantly increasing the power consumption.
5. Conclusion
We presented structural design technology to improve the linearity of APDs without scarifying responsivity and speed. The APD features a hybrid absorption layer configuration to suppress the space charge effect caused by hole accumulation and to help improve the linearity. The fabricated APD exhibits the linearity of −1dB compression for output photocurrent of 0.9 mA (corresponds to the optical input power of −8 dBm) at a gain of 10. We also investigated the BER characteristics in 28-Gbaud PAM4 transmission, which requires high linearity for the optical receivers. The fabricated optical receiver module made with the APD successfully demonstrates a BER of 2.3 × 10−4, which provides error-free operation if we assume KP4 FEC, with a receiver sensitivity of −17 dBm. Furthermore, the extension of the dynamic range is obtained when we adaptively control the bias voltage to the APD. These results indicate that our high-linearity APD is feasible for the future large-capacity Ethernet systems with HOM formats.
Acknowledgments
The authors thank S. Kanazawa, Y. Nakanishi, W. Kobayashi, Y. Doi, T. Ohyama, T. Ohno, K. Takahata for valuable discussions, and K. Murata and S. Kodama for their continuous encouragement.
References and links
1. http://www.ieee802.org/3/bs/.
2. http://www.ieee802.org/3/400GSG/
3. K. Nakahara, Y. Wakayama, T. Kitatani, T. Taniguchi, T. Fukamachi, Y. Sakuma, and S. Tanaka, “56-Gb/s Direct modulation in InGaAlAs BH-DFC lasers at 55 °C,” in Proceedings of OFC2014, Th3A.1 (2014).
4. W. Kobayashi, T. Ito, T. Yamanaka, T. Fujisawa, Y. Shibata, T. Kurosaki, M. Kohtoku, T. Tadokoro, and H. Sanjo, “50-Gb/s direct modulation of a 1.3-um InGaAlAs-based DFB laser with a ridge waveguide structure,” IEEE J. Sel. Top. Quantum Electron. 19(4), 1500908 (2013). [CrossRef]
5. M. Nada, H. Yokoyama, Y. Muramoto, T. Ishibashi, and H. Matsuzaki, “50-Gbit/s vertical illumination avalanche photodiode for 400-Gbit/s Ethernet systems,” Opt. Express 22(12), 14681–14687 (2014). [CrossRef] [PubMed]
6. K. Zhong, W. Chen, Q. Sui, J. Man, A. Lau, C. Lu, and L. Zeng, “Experimental demonstration of 500Gbit/s short reach transmission employing PAM4 signal and direct detection with 25Gbps device,” in Proceedings of OFC2015, Th3A.3 (2015). [CrossRef]
7. M. Traverso, M. Mazzini, M. Webster, C. Muzio, S. Anderson, P.-C. Sun, D. Siadat, D. Conti, A. Cervasio, S. Pfnuer, J. Stayt, C. Togami, T. Daugherty, and K. Yanushefski, 25Gbaud PAM-4 error free transmission over both single mode fiber and multimode fiber in a QSFP form factor based on silicon photonics,” in Proceedings of OFC2015, Th5B.3 (2015).
8. T. Takahara, T. Tanaka, M. Nishihara, Y. Kai, L. Li, Z. Tao, and J. Rasmussen, “Discrete multi-tone for 100 Gb/s optical access networks,” in Proceedings of OFC2014, M2I.1 (2014). [CrossRef]
9. A. Beling, H. Pan, H. Chen, and J. C. Campbell, “Measurement and modeling of a high-linearity modified uni-traveling carrier photodiode,” IEEE Photonics Technol. Lett. 20(14), 1219–1221 (2008). [CrossRef]
10. W. Sun, Y. Fu, Z. Lu, and J. C. Campbell, “Study of bandwidth enhancement and non-linear behavior in avalanche photodiodes under high power condition,” J. Appl. Phys. 113(4), 044509 (2013). [CrossRef]
11. T. Kagawa, Y. Kawamura, and H. Iwamura, “Saturation of multiplication factor in InGaAsP/InAlAs superlattice avalanche photodiodes,” Appl. Phys. Lett. 62(10), 1122 (1993). [CrossRef]
12. M. Nada, Y. Muramoto, H. Yokoyama, T. Ishibashi, and H. Matsuzaki, “Triple-mesa avalanche photodiode with inverted p-down structure for reliability and stability,” J. Lightwave Technol. 32(8), 1543–1548 (2014). [CrossRef]
13. W. Kobayashi, T. Fujisawa, T. Ito, S Kanazawa, Y. Ueda, and H. Sanjoh, “Advantages of EADFB laser for 25 Gbaud/s 4-PAM (50 Gbit/s) modulation and 10 km single-mode fiber transmission,” Electron. Lett. 50, 683 (2014). [CrossRef]
14. J. Man, W. Chen, X. Song, and L. Zeng, “A low-cost 100GE optical transceiver module for 2km SMF interconnect with PAM4 modulation,” in Proceedings in OFC2014, M2E.7 (2014).
15. http://www.ieee802.org/3/bs/public/14_11/parthasarathy_3bs_01a_1114.pdf.
