In the format provided by the authors and unedited. DOI: 10.1038/NPHOTON.2016.233 A monolithic integrated photonic microwave filter Javier S. Fandiño 1, Pascual Muñoz 1,2, David Doménech 2 & José Capmany 1 1 ITEAM Research Institute, Universitat Politècnica de València, Valencia, 46022, Spain. 2 VLC Photonics, Valencia, 46022, Spain. Building block characterization Tunable laser (TL 1): This is a wavelength-tunable distributed Bragg reflector (DBR) laser, which is made up of four distinct building blocks: 2x tunable DBR mirrors of different length (with maximum reflectivities of 6% and 97.34%, respectively) to provide feedback into the laser cavity; 1x semiconductor optical amplifier (SOA) to introduce gain; and 1x current-injection phase shifter (PS) for fine wavelength control. Measured figures of merit for this laser at an operation temperature of 25ºC and 60 ma of bias current include: a mode spacing of 0.399 nm with more than 40 db of side mode suppression ratio (SMSR), fibercoupled output power higher than -8 dbm, 15 ma of threshold current, about 5 nm of wavelength tunability through the use of the longer DBR mirror, and continuous fine wavelength control within the mode spacing bandwidth by means of the intra-cavity PS. Photodetectors: The dark currents of the different on-chip photodiodes were characterized with a sourcemeter (Keithley 2401), measuring the current flowing through the p-n junction while applying a fixed reverse voltage when no optical power was injected at their input. The dark currents were measured as a function of reverse voltage and for different operation temperatures. Typical values are between 1 na and 3 na for a typical reverse voltage of -2 V at 25ºC. The responsivity is estimated to be about 0.85 A/W at 1550 nm, and the 3 db bandwidth is expected to be higher than 10 GHz. Thermo-optic heaters: Multiple thermo-optic heaters were placed on the RAMZI filter so as to be able to tune its optical response. All of them have a length of 200 µm, with measured DC resistances between 18.4 Ω and 28.5 Ω. An important parameter for any thermo-optic heater is the Pπ. That is, what is the electrical power (in mw) required to induce a 180º optical phase shift. The Pπ of these heaters was characterized thanks to an auxiliary 2x3 MZI (see Fig. 1b in the main text). Its three ouputs are connected to three DC photodiodes, and it also includes a heater on one of its arms. First, TL 1 was swtiched on (60 ma bias current), and then the photocurrents in the three DC diodes (PD 1, PD 2, PD 3) located at the output of the 2x3 MZI were measured for different currents injected into the heater. Afterwards, a post-processing algorithm was applied 42. The measured Pπ is 43.8 mw. Polarization alignment procedure This procedure was devised to ensure a proper measurement of the RAMZI filter when using optical fibers connected to the PIC package. As it has been said in the main text, the auxiliary Input and Output waveguides are coupled to a pair of single-mode fibers that are soldered to the chip package, which provide a convenient way of connecting the PIC to external equipment. However, these are normal single-mode fibers, which do not preserve the state of polarization. As the RAMZI filter critically depends with the polarization state of the input signal, it is compulsory to devise a method to ensure that only the right polarization (TE, in this case) is being measured. The alignment procedure requires a manual polarization controller, an in-fiber polarizer and a power meter to be connected at the PIC output fiber; and a tunable laser and another manual polarization controller to be connected at the input one, as shown in Fig. 1a. It is based on two key assumptions. First, on-chip lasers only emit TE polarized light. Second, integrated waveguides do not (ideally) exchange power between the TE and TM modes as the light propagates within the PIC. If these two assumptions are true, then we might proceed as follows. In step I, an on-chip laser (TL 1 in this case) is switched on. If the two previous assumptions are true, then we know that the polarization coming out of the PIC is horizontally polarized (TE). However, the optical fiber connected to the auxiliary Output waveguide will randomly rotate this state of polarization, so that an unknown mixture of both TE and TM will be finally injected into the in-fiber polarizer. In step II, we proceed to adjust the manual polarization controller at the output, until a maximum power level is measured in the power meter. At this point, it is ensured that only TE light coming out of the chip will be transmitted through the in-fiber polarizer. If there were any TM power radiated by the PIC it would be immediately eliminated, as it would be orthogonal to its transmission axis. This process is equivalent to having a free-space polarizer coupled to the output facet and aligned with the horizontal axis (TE) of the chip. In step III, we switch off the on-chip laser and switch on instead an external tunable laser source. Again, as the singlemode fiber that connects the external laser with the PIC input scrambles the polarization, we do not know how much power is being coupled to the TE and TM modes of the integrated, auxiliary Input waveguide. However, we know that the PIC should (ideally) preserve power in the TE and TM modes. As a consequence, any power coupled to the TE (TM) mode at the Input waveguide will also appear as TE (TM) at the chip Output waveguide. However, we already adjusted the output polarization controller in steps I and II. This means that any TM component at the chip output will be blocked by the in-fiber polarizer, and only the TE component, for which the chip was designed, will be measured. We only need now to maximize power by using the input polarization controller, which is finally done in step IV. Page 1 of 5 NATURE PHOTONICS www.nature.com/naturephotonics 1
Once the power is maximized, we know that only TE is being excited at the chip Input waveguide. Finally, note that under these conditions controlling the input polarization is not actually critical. If the input polarization state were changed due to mechanical vibrations or other external factors, then a fraction of the incoming power would be coupled to the TM mode instead. However, as the TM power is being filtered at the output, the measured transfer function of the RAMZI filter (in db) would still preserve its shape. The only difference is that it will get attenuated by a certain factor, which is nothing but a constant power offset when measured in logarithmic units. RAMZI filter characterization In order to measure the tunable RAMZI filter, the polarization alignment procedure described in the previous section was first performed. Afterwards, the power meter was replaced with an optical spectrum analyzer (OSA, Ando AQ6317C) synchronized with the external laser source (Ando AQ4321D). As it has been said, the polarization alignment is based on the key assumption that integrated optical waveguides do not exchange power between the TE and TM modes, an effect also known as polarization rotation. If that is the case, then it is ensured that only the TE response of the PIC is measured, even if TM is also being excited. As a consequence, changing the polarization state at the input should only change the amplitude of the measured transfer function, which relates to the amount of power coupled to the TE mode, but not its shape. However, this did not happen during the measurements. Figures 2a and 2b show a couple of measured transmission spectra when changing the input polarization after the alignment. Since they are completely different, this pointed out to the fact that polarization was actually rotating inside the chip, contrary to the original assumption. To check this hypothesis, we switched on the on-chip TL 1 laser and measured the polarization state of light at the output fiber using an optical polarization analyzer (Agilent 8509C). If there were no polarization rotation, changing the wavelength of the laser should induce a change of output power, as each wavelength is affected differently by the transfer function of the RAMZI filter. However, the output polarization should remain constant. Figure 2c shows that the polarization is actually changing when a current between 0 and 5 ma is pumped into the rear DBR mirror of TL 1, which changes the wavelength of the laser. This confirms the earlier observations. A similar reasoning can be done for the thermo-optic heaters of the RAMZI filter. If there were no polarization rotation, changing the relative optical phase shifts of any of the ring resonators would change the transfer function of the filter. If the laser is now being kept at a fixed wavelength, only a change in output power should be observed, but polarization should remain constant. Figure 2d shows the measured polarization change when one of the thermooptic heaters of the filter was pumped with a current between 0 and 80 ma, with TL 1 now being kept at a constant wavelength. These measurements confirmed that both polarizations are actually exchanging power inside the PIC, as the polarization completely rotated over the Poincaré sphere. Fig. 1 Polarization alignment procedure. Diagram illustrating the procedure to correctly align the polarization when measuring the RAMZI filter with external equipment. Ext. TL: External tunable laser. PC: Polarization controller. Pol.: In-line fiber polarizer. PM: Power meter. Polarization rotation is a well-known effect in integrated optical waveguides 43,44, which can be described as follows. In straight sections, single-mode waveguides normally support both quasi-te and quasi-tm modes, whose horizontal/vertical electric field components are much greater than their vertical/horizontal ones, respectively. That is, E x TE /E y TE and E y TM /E x TM >> 1. However, if one solves for the modes of the same cross-section inside a tight bend, it will observe that the aforementioned condition does not longer hold as the bending radius gets smaller. The modes start to get hybridized, and the negligible field components (E y TE, E x TM ) become stronger relative to the dominating ones (E x TE, E y TM ). Unfortunately, mode hybridization can get significantly enhanced in waveguide cross-sections that are not symmetrical in Page 2 of 5 NATURE PHOTONICS www.nature.com/naturephotonics 2
the vertical direction 44. Even though this asymmetry might be already present due to the way waveguides are defined in a specific photonic platform, it will be nevertheless produced naturally during lithography and etching, as real manufactured waveguides always feature slanted sidewalls with a small angle (θ). As an example, the cross-section of the waveguides in the technology employed to manufacture the chip is schematically shown in Fig. 3a. Fig. 2 Experimental evidence of on-chip polarization rotation. a,b, Measured transmission spectra of the RAMZI filter for two different settings of the input polarization controller, once the alignment procedure was completed. c, Measured state of polarization at the output fiber when the frequency of TL 1 was changed by pumping current into its rear DBR mirror. d, Idem, when the frequency of TL 1 was kept fixed and current was injected into one of the thermo-optic heaters of the RAMZI filter. Polarization rotation is known to occur mainly at straight/bend interfaces, as shown in Fig. 3b. Here, we have assumed a simple case for illustration purposes, where a couple of perpendicular straight waveguides are connected with a 90º bend. If a pure TE mode is launched at the input straight section, mode hybridization will excite both TE and TM modes in the bent waveguide. Neglecting radiation losses and higher-order modes, we can understand this situation as if there were an optical coupler between orthogonal polarizations, where the coupling constant (κ 1 ) specifies how much of the input power of the straight TE mode is coupled to the TM mode of the bent section. Afterwards, each polarization travels along the bend with different effective indices, which will yield different relative phase shifts (Φ TE and Φ TM ) before the output interface. Here, the TE and TM modes of the straight and bent sections will exchange power again (κ 2 ). As illustrated in Fig. 3c, the whole process can be viewed as an MZI, whose two input/output ports represent the orthogonal modes before and after the bend, respectively. The input/output couplers, parametrized by κ 1 and κ 2, represent the inter-modal coupling at the straight/bent interfaces, while Φ TE and Φ TM represent the distinct optical phase shifts as seen by each orthogonal polarization due to their different effective indices. Depending on the radius (R) and arc length (L) of the bend, a different amount of polarization crosstalk (defined as the ratio between TE and TM powers at the output of the bend, in db) will be observed at the output waveguide, which will also change with operation wavelength. Numerical simulations performed for the employed waveguide cross-section of Fig. 3a are shown in Figs. 3d and 3e. Figure 3d shows the expected TE and TM powers at the output of a bend with a radius (R) of 150 µm as a function of its arc length (L), for a value of θ of just 5º. Note the periodic behaviour of the response, just like what would be expected for an MZI. Figure 3e represents the expected polarization crosstalk as a function of the slanted sidewall angle θ. This simulation considered a 90º bend with the same 150 µm radius that was employed for all the ring resonators and routing waveguides of the PIC. It shows how θ critically impacts on the polarization crosstalk, going from a maximum of about 70 db for a perfectly vertical waveguide down to 30 db when the angle is just increased by 5º. Finally, it must be noted that polarization rotation is a coherent process. Thus, even if relatively low values of crosstalk are observed at a single bend, rotation might build up as the light propagates within a large structure with multiple bends, causing significant crosstalk at the output. Optical cavities, such as ring resonators, are also known to enhance this effect, as the light effectively sees a longer optical path near the resonance 44. To sum up, we believe that the significant polarization rotation observed in the PIC is due to a combination of two effects. First, a higher than expected value of the sidewall angle in the fabricated waveguides. Second, a coherent enhancement of the polarization rotation effect due to the resonant nature of the rings in the RAMZI filter. NATURE PHOTONICS www.nature.com/naturephotonics 3
Fig. 3 Polarization rotation simulations. a, Cross-section of the InP waveguides. b,c, Modelling of polarization rotation in a 90º bend. d, TE and TM powers at the output of a bend with a 150 µm radius, for different arc lengths (L) and a waveguide sidewall angle (θ) of 5º. e, Polarization crosstalk (TE divided by TM, in db) for a 90º bend with a 150 µm radius, and different values of θ. Electrical to electrical (E/E) response of the MWP filter This experiment was similar to that described in the Tunable MWP filter section of the main text. However, this time the signal going out of the RAMZI filter was not collected at the output fiber, but was detected internally using an RF photodetector (PD 4), as shown in Fig. 4a. For that purpose, PD 4 was connected to another bias tee and reverse biased at -2 V, while its RF output was directly connected to the VNA. The E/E response of the system was then measured, but in this case the laser frequency was kept constant. Figure 4b shows the measured E/E response when the current set #1 was applied in the thermo-optic heaters of the RAMZI filter. As it can be seen, the response is quite different from those of Fig. 3b (current set #1) in the main text. In fact, the measured response did not change much even when TL 1 was switched off, as only a small drop for frequencies lower than 1 GHz was appreciated (Fig. 4c). The explanation for this unexpected behaviour lies in the existence of a significant internal RF crosstalk. The device packaging was subject to strict rules relative to the position of the optical and electrical ports in order to be manufactured and packaged. This means that the electrical ports had a predefined position in the layout in order to be able to fit into the standard package developed in the project. The electrical RF lines and pads are based in microstrip lines with predefined RF pad positions spaced 250 microns so in the layout the ports were placed at the edge of the die and with the predefined pitch. The small distance between the RF pads for the MZM and the photodetector suggested that the interference was due to the internal coupling. To confirm this, we carried simulations of the RF microstrip line structure, which predicted a crosstalk level around -25 db in agreement with the results shown in figure 4.c. PD 4 was placed very close to the MZM electrodes inside the chip, with no further electrical isolation measures. As a consequence, the input RF signals in both MZM arms are contaminating the measured photocurrent, which is orders of magnitude smaller due to the inherent losses of the MWP filter. Only for very low frequency values the RF crosstalk becomes negligible, and a small difference can be appreciated. Unfortunately, this unwanted effect ultimately prevents the E/E operation of this PIC, for which it was originally designed. The main reason for this limitations stems from the fact that the package is NOT designed for transceiver housing (both transmission and reception as our chip) but only for standalone transmitter or receiver operation. A reduction of the internal onchip crosstalk would require increasing the spacing between the RF ports and / or employing different RF lines like Coplanar Waveguides (CPW) which could provide a better isolation among RF ports. NATURE PHOTONICS www.nature.com/naturephotonics 4
Fig. 4 Experimental results for the E/E response of the tunable MWP filter. a, Diagram of the experimental setup for the E/E measurement. b, Measured response if TL 1 is switched on and the current set #1 is applied in the thermo-optic heaters of the RAMZI filter. c, Idem, when TL 1 is switched off. NATURE PHOTONICS www.nature.com/naturephotonics 5