Silicon photonics traveling wave photodiode with integrated star coupler for high-linearity mm-wave applications

Next-generation wireless communication will require increasingly faster data links. To achieve those higher data rates, the shift towards mmWave frequencies and smaller cell sizes will play a major role. Radio-over-Fiber has been proposed as a possible architecture to allow for this shift but is nowadays typically implemented digitally, as CPRI (Common Public Radio Interface). Centralization will be important to keep next-generation architectures cost-effective and therefore shared optical amplification at the central office could be preferable. Unfortunately, limited power handling capabilities of photodetectors still hinder the shift towards centralized optical amplification. Traveling wave photodetectors (TWPDs) have been devised to allow for high-linearity, high-speed opto-electronic conversion. In this paper, an architecture is discussed consisting of such a TWPD implemented on the iSiPP25G silicon photonics platform. A monolithically integrated star coupler is added in the design to provide compact power distribution while preserving the high linearity of the TWPD. The traveling wave structure (using 16 photodetectors) has a measured 3 dB bandwidth of 27.5 GHz and a fairly flat S21 up to 50 GHz (less than 1 dB extra loss). Furthermore, the output referred third-order intercept point at 28 GHz, is improved from -1.79 dBm for a single Ge photodiode to 20.98 dBm by adopting the traveling wave design. © 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
Future wireless communication links will require increasingly higher data rates to accommodate next generation applications [1].Densification of communication cells will play a major role in the shift towards faster wireless data rates.To keep this technique economically and ecologically viable, a centralized approach needs to be pursued.A second key enabler for higher data rates is the usage of different parts of the spectrum.Millimeter wave (mmWave) frequencies are of great interest since they offer larger bandwidths and they are significantly less congested.Different schemes have been devised to enable centralization for mmWave communication.For instance Analog Radio-over-Fiber (ARoF) is a straightforward implementation where the RF signal is generated at the central office (CO), modulated on an optical carrier, and subsequently transmitted to the desired remote antenna unit (RAU) [2].After transmitting the signal from the CO to the RAU, an opto-electronic conversion is performed by a photodetector.This RF signal should then be amplified such that it is sufficiently strong to overcome path losses in the wireless channel between the RAU and the end user.This is typically done by adding a transimpedance amplifier (TIA) followed by a power amplifier [3,4].The aforementioned architecture can be altered to a more centralized topology by moving the amplifiers to the optical domain, where the amplification can be done for multiple RAUs at once, resulting in a significant decrease in RAU complexity and power consumption.However, the linearity of a high-speed photodetector (PD) is typically inadequate to allow for high RF output power [5].Improving the power handling capabilities of the photodetectors eventually allows for the omission of electrical amplification in the DL-RAU (downlink RAU, i.e. from central office to mobile end user), paving the road for passive DL-RAUs [6].Implementing such high-power-handling photodiodes on a silicon photonics platform enables the low-cost manufacturing of such devices in high volume.Moreover, it also allows the integration of other optical functionality (e.g. an optical beam forming network) on the same circuit.To improve the power handling capabilities of an integrated high-speed p-i-n photodetector one can opt to use different materials and detector principles (e.g.III-V uni-travelling carrier photodetectors integrated on silicon [7][8][9][10][11][12][13]) or more complex photodetector configurations.While UTC photodetectors are often being used as high-power, high-speed photodetectors, they are not CMOS compatible because of their complex layer stack based on III-V materials (typically InGaAs-InP).To make optimal use of the aforementioned benefits of silicon photonics, a high-power variant of the existing Si-integrated Ge PiN photodetector needs to be constructed without altering the technology stack of the iSiPP25G platform or requiring heterogeneous integration.The traveling wave photodiode (TWPD) structure is the most popular configuration to realize high power handling while relying on high-speed p-i-n photodetectors [14].In this paper, a highpower-handling traveling wave photodetector (TWPD) structure integrated on a silicon photonics waveguide platform will be discussed.The traveling wave structure (using 16 photodetectors) has a measured 3 dB bandwidth of 27.5 GHz and has a fairly flat S 21 up to 50 GHz (less than 1 dB extra loss).Furthermore, the output referred IP3 linearity at 28 GHz is improved from −1.79 dBm for a single Ge photodiode to 20.98 dBm by adopting a traveling wave design with dual-fed photodetectors.

Ge traveling wave photodetectors on a Si photonics platform
In this work, the iSiPP25G silicon photonics platform of imec is used to realize the traveling wave photodetector structure.The germanium photodiodes available on the platform have a responsivity of 0.8 A/W and a bandwidth of more than 50 GHz.However, their power handling capabilities are limited, as will be discussed below.A clever rearrangement of the photodetector structure and splitting the power over multiple PDs can significantly increase the linearity.

Dual fed photodetectors
While the standard mode of operation for the Ge photodiode on the iSiPP25G platform is to use a single optical input waveguide, one of the features available in the platform is that the Ge PDs can be optically fed from two sides.This immediately allows for a linearity improvement by increasing the optical input power for which compression occurs in the optoelectronic conversion.By adding a splitter in front of the PD, the linearity can be improved at the cost of a slight increase in insertion loss (due to the excess loss of the splitter, which is specified to be below 0.2 dB).This enhanced power handling capability is caused by the increased portion of the absorption layer that is used for the opto-electronic conversion [14].

Increasing the number of photodetectors
Another solution to improve the power handling capability of the optoelectronic conversion is to increase the number of photodetectors per RAU.The power handled per individual PD drops proportional with the amount of PDs, enhancing the power limit of the RAU significantly.However, in the electrical domain the RF signals will need to be recombined constructively.

F
Combinin linearity impr increased cap to recombine high-linearity embedded in

Propose
In Fig. 2, the with a 16-PD of three majo electronic con of each Ge P individual Ge optical and el closer to the l the transmissi sure that the simultaneousl Typical imple this paper, a m the power ha photon absorp become even photodiode is eventually ena       ctures is given fabricated devic ions on the te be expected w d.The most i ce is that eve ed when the evels in the f nt sunk into th le power at th otodetector rem photodetectors, ent per photod backwards pro xpected when e losses will re d its resistance

Star coup
The star coup integrated 16 platform.Con photon absorp in the on-chip variant of the between the d input power w

Bandwidt
In this sectio compared in t 1, will be com the effect of standalone Ge

Linearity
In this sectio discussed.For

DC linea
The first linea power (at 15 measurement EDFA used in h this reference be determined han subtracted Fig. 6.
rent devices ar , the curves are n coupling losse ent.Important indeed provid power handlin since these str the fiber prob losses from t typically betw coupled to th which is typic that the value used.Additio main excess l MMIs needed optical delay other hand re not the same grating coupl multi-PD stru For the TW as discussed absence of lig proportional optical input responsivity).linearity of th opto-electroni 7. Setup used for : source measure u to notice whe des a higher lin ng capabilities ructures are st be).The availa the fiber prob ween 3 and 7 d e chip is betw cally between es given in Tab onally, the expe losses present i d for the dual lines (approxim sistive losses i for the differe ler, it is difficu uctures. WPD with term in Fig. 8 (illu ght at the inpu to the applied power is gen The former co he opto-electron ic contribution DC linearity mea unit).en observing t nearity and tha at DC.For the ill linear at ma able power at be to the outp dB depending o ween 125 mW 100 mA and 2 ble 1 are at the ected current r in the system a fed operation mately 0.26 to in the transmis ent DUTs due ult to make a     can see a 7 dB increase in power handling capabilities when dual feeding the PD.Additionally, it is clear that multi-PD structures are capable of handling higher powers.Adding multiple photodetectors in parallel results in RC low pass filtering as discussed before.Taking into account the frequency dependent current-to-power conversion, displayed in Fig. 4, of the parallel combination of 16 PDs, an improvement of approximately 9.5 dB over the dual fed, standalone germanium photodetector is expected.This value agrees reasonably well to the measured linearity improvement, namely 10.6 dB.For the TWPD with and without dummy termination improvements of respectively 14.2 dB and 14.6 dB are to be expected over the standalone, dual fed Ge PD based on Fig. 4. The measured improvements in linearity are respectively 15.8 dB and 13.2 dB.Simulations indicated that the slightly higher power handling capabilities at 28 GHz would be achieved without dummy termination while the measurements show that more power can be obtained by adding the resistor.A possible explanation for the discrepancy is a deviation in the equivalent network of the Ge PD.

Conclusion
Centralization is key in making future wireless network architectures feasible.Typical implementations of an ARoF link comprise of electrical amplification at the antenna.To push towards an increase in centralization and therefore a drop in cost and power consumption, amplification can be done for multiple RAUs simultaneously.Optical amplification at the CO will however require opto-electronic conversion with high power handling capabilities.In case of mmWave communication, a traveling wave photodiode with a monolithically integrated star coupler implemented on a silicon photonics platform is proposed in this paper.
The TWPD combines high linearity with high bandwidth while the star coupler allows for a compact solution that preserves the high linearity obtained from adopting a multi-PD architecture.In this paper, a TWPD is described that has a 3 dB bandwidth of 27.5 GHz with a gentle roll-off off at higher frequencies allowing for mmWave operation.Additionally, the OIP3 of the TWPD is found to be nearly 21 dBm which is a significant improvement over the standard single fed Ge PD where the OIP3 is −1.8 dBm.

Fig.
Fig. 1.Lumped pa Fig. 2 fabric length 2.4 Traveling In this sectio electronic con This load imp transfer from the optical an lines, as discu line should b TWPD and t termination r backwards pr to the forward Fig. 3 segme fabric A first ste model for the model is show simplified des unloaded tran (C j is the junc the substrate) parameter me in our travelin photodetector found: R s ≅ 3 depicted in F and those guid Eq. (4), the 1550 nm for t the iSiPP25G photodetector biasing.Takin (imposed by d by design rule of 5 μm and Subsequently For the dem obtained.Th dimensioning attributed to t desired 190 μ is shown in Fi Fig. 4 indiviThe simul These simulat the TWPD h TWPD show device is ada difference be terminated T erroneous.Th TWPD.When expected to s following a q same.Since frequency out kept constant waves in a tr termination to power gain.T 4. Simulated outpu idually generates 2 lated frequency tions are perfo has a character that only a 0 apted such tha tween the fab WPD some m his is due to n a multi-PD s scale with the quadratic scalin the implemen tput power sh t.Additionally ransmission lin o the TWPD.These substrate enotes the spee de waveguides ing from Eq. ( designed TWP nt these param nd the heating nar waveguide 4.2 μm) and ere matched b ice, a 35.4 O was obtained b n relative to th he PD spacing ned TWPD seg ut power in a 50 O 2 mA peak-to-peak y dependent be rmed with the ristic impedan 0.8 GHz increa at the correct ricated and th minor interfere the non-perfe structure is use number of PD ng with the PD nted multi-PD ould increase , since a termi ne, a 6 dB dr In practice, ho e losses are mo ed of light and s used in the de 1), with a targe PD has a 35 fF meters and the r of the traces) shown in Fig. periodically l by designing an Ohm characteri by re-simulati he targeted 50 O g ΔLe was cho gment is depic Ohm load for the k current (2V rever ehavior of the dimensions ad nce of 35.4 O ase in 3-dB b dimensions a he ideally dime ence effects c ectly matched ed, the low fre Ds and therefo D count if the c structures co with 24 dB w ination resistor rop at low freq owever, additi odeled by R p i n g is the group elay lineμm inste d) and the entir res when each PD hm termination).

Fig. 5
Fig. 5 Zehnd The setup measurements point and to c reference PD Fig. 7 SMU: Fig. 9 ElectrNon-linea intermodulati presented as F