1. Introduction

With the number of multimedia services, wireless access, internet devices and mobile users constantly growing, signal processing techniques such as time division multiplexing [1] (TDM) and wavelength division multiplexing [2] (WDM) have been developed to extend the operational bandwidth of existing optical networks. As a result, 10 Gb/s and 40 Gb/s optical networks are now standard, whereas 100 Gb/s network are being tested and implemented [3]. However, those methods have limitations and one solution toward increasing the bandwidth of the networks is to accelerate the data transmission speed. This solution requires the development of a number of components, one of which is the modulator. Ultrahigh speed modulators are key components in the development of optical fiber networks as they set the transmission capacity from the electrical to the optical domain. A variety of modulators operating into the millimeter-wave (mmW) range have previously been developed for the telecommunication market [4,5]. In addition, interest in mmW imaging has also contributed to the support and development of mmW modulators [6]. Electro-optic (EO) polymer modulators [7] and electroabsorption (EA) modulators [8] have shown the ability to operate in the mmW region, but lithium niobate (LiNbO₃) EO modulators possess several advantages over the others. They can operate at very high bandwidths, offer very small frequency chirp, handle high optical power, have low optical loss, require relatively low drive voltages because of a high EO coefficient and are very stable over time. Moreover, the LiNbO₃ EO modulator is a mature technology that is widely used in the current optical network infrastructure and it has shown strong optical response up to 110 GHz [9,10]. In this paper, we present a LiNbO₃ EO phase modulator that operates over the entire mmW region, with sidebands up to 300 GHz demonstrated.

2. Device design and fabrication

In LiNbO₃ EO phase modulator design, the modulating radio frequency (RF) signal interacts with the optical signal to create sidebands on the optical carrier. To maximize this interaction, we designed a ridged coplanar waveguide (CPW) electrode on top of a Ti-diffused optical waveguide to support the RF signal [11,12]. To efficiently convert the electrical energy into optical energy and create sidebands, five main criteria need to be optimized. First, the mmW effective index has to be reduced from ~6 down to the optical effective index of 2.19 in order to maximize the copropagating interaction of the modulation and optical signal [13]. Such index matching is reached by combining together a ridged CPW structure, thick electrodes and a silicon dioxide (SiO₂) buffer layer between the electrodes and the LiNbO₃ surface. The ridge structure and the buffer layer substitutes a high index material, LiNbO₃, with low index materials, respectively air and SiO₂ of index 1.46, whereas a thick electrodes CPW structure pulls the electric field into the air. Second, the dielectric and conduction losses need to be minimized. The dielectric loss depends greatly on the LiNbO₃ substrate and the buffer layer properties, whereas the conduction loss is mostly determined by the CPW geometry and material [14]. Third, the CPW impedance must be matched to limit reflection losses when coupled to a standard 50Ω transmission line. Fourth, overlap between the RF and optical mode must be maximized. The mode overlap can be maximized by reducing, as much as possible, the gap between the RF electrode and the ground electrodes of the CPW. Finally, coupling of RF energy into substrate modes must be eliminated. The modes supported by the substrate are directly related to its thickness [15,16]. As the frequency of operation increases, the RF wavelength decreases to the point where the RF modes are supported by the substrate, causing RF energy to leak out of the CPW structure into the substrate. If the substrate mode and the CPW mode propagate down the modulator at the same speed, those modes strongly interact with each other and deteriorate the electrical propagation properties of the CPW mode at that particular RF frequency. Therefore, the substrate needs to be thinned-down to prevent substrate modes at higher frequencies [17–19]. The transmission parameter S₂₁ has been measured for a modulator at two different substrate thicknesses to show the effect of the substrate modes on the propagating RF modes (Fig. 1 ). By thinning the LiNbO₃ substrate down to 65 µm, the RF modes only start coupling into the substrate at 180 GHz, as opposed to about 70 GHz for a 500 µm thick substrate.

 

Fig. 1 Effect of substrate mode coupling on mmW electrical propagation properties. The transmission parameter S₂₁ shows the importance of reducing the substrate’s thickness to eliminate the coupling of the RF signal into substrate modes and therefore enhance the transmission. With a substrate thickness reduced to 65 µm, substrate mode coupling is observed starting at 180 GHz.

Download Full Size | PDF

Ultimately, a trade-off between those five main criteria is necessary to achieve optimal design for a given application. A numerical simulator based on the finite element method (FEM) is used to optimize the mode-overlapping between the RF signal and the optical signal as well as the impedance matching [20]. The design of the modulator cross-section resulting from this analysis is represented in Fig. 2 . The CPW thickness T, the central electrode width S, the electrode length L, the gap G, the buffer oxide thickness B, the ridge height H and the ridge width R are respectively 25 µm, 8 µm, 2 cm, 25 µm, 0.9 µm, 3.6 µm and 10.5 µm with a substrate thickness D equal or less than 39 µm [17], which corresponds to the cutoff thickness of the substrate modes over the entire mmW region.

 

Fig. 2 Modulator cross-section. A ridged CPW structure is built on top of a Ti in-diffused waveguide. High aspect-ratio electrodes combined with a SiO₂ buffer layer and a ridge structure allows effective index matching between optical and RF signals represented by E. A small gap G between the CPW electrodes leads to a strong mode-overlapping. A thinned LiNbO₃ substrate underneath the CPW structure eliminates substrate modes over the full mmW range.

Download Full Size | PDF

Following the design presented in Fig. 2, a set of sixty modulators has been fabricated on a 500 µm thick z-cut LiNbO₃ wafer substrate using the following fabrication process. First, a titanium strip is diffused into the LiNbO₃ substrate at around 1000°C for 10 hours to form the optical waveguide structure [21]. Second, the ridge structure is fabricated by dry etching the LiNbO₃ material surrounding the optical waveguide using inductively-coupled plasma (ICP) reactive ion etching (RIE) technology in a chlorine environment [22]. Third, a silicon dioxide buffer layer is deposited on the substrate using Plasma Enhanced Chemical Vapor Deposition (PECVD) technique, followed by a 6-hour annealing process at 600°C. Fourth, the high aspect-ratio CPW structure is defined by lithography using SU8-2015 photoresist from MicroChem, before electroplating the open surface using a gold solution to build up the CPW electrodes until the desired thickness. Fifth, the modulators are diced into small groups and each group's end faces are polished. Sixth, each modulator is diced into a single chip and individually thinned down to the desired thickness. For this thinning process, a 400 µm wide groove is machined underneath the signal electrode over the entire length of the modulator. The two modulators reported later in this paper have both been thinned down below the 39 µm substrate mode cutoff thickness, one to 30 µm and another one to 20 µm. Finally, polarization maintained optical fibers are bonded to both end faces of a modulator using UV curable epoxy.

3. Experimental setup

The S-parameters, the optical sidebands and the DC-V_π are the three different measurements used to characterize the modulator. Four sets of equipment are necessary to perform those measurements over the 300 GHz bandwidth, which is divided into four bandwidths, 0-110 GHz, 110-170 GHz, 170-220 GHz and 220-300 GHz. An Agilent E8361C Programmable Network Analyzer (PNA) is used as the main RF source. In the 0-110 GHz range, we use Agilent N5260 T/R modules, 1mm cable waveguides and corresponding GGB Industries probes. The RF power is measured using an Agilent E4418-B power meter. In the 110-300 GHz range, to each bandwidth corresponds a set of VNA extension modules from OML, Inc., Millitech rectangular waveguides and GGB Industries probes. An Erickson PM4 power meter is used to measure the RF power delivered in those regions. An EM4 laser centered at 1557.3 nm is used as the optical source. The optical sidebands are observed by connecting the modulator's output fiber to a Yokogawa AQ6319 Optical Spectrum Analyzer (OSA). An external Mach-Zehnder interferometer configuration setup around the modulator is used to perform the half-wave voltage DC-V_π measurement, which is determined by examining the interference pattern resulting from an applied voltage and measuring the voltage corresponding to a 180° optical phase shift.

4. Device characterization

Scanning Electron Microscope (SEM) pictures of the end face of a 30 µm thick modulator, before fiber bonding, in the CPW launching area are shown in Fig. 3 . In the launching area, the width of the signal electrode is 25 µm and the gap between the signal electrode and the ground electrodes is 75 µm.

 

Fig. 3 SEM pictures of the modulator fabricated. (a) CPW gold plated structure with LiNbO₃ etched on each side. (b) Overall view of the end face of the modulator before optical fiber bonding. The LiNbO₃ substrate has been thinned to 30 µm by micromachining a groove under the CPW structure to eliminate the substrate modes over the entire mmW region.

Download Full Size | PDF

The S₂₁ parameter of the 30 µm thick modulator has been measured over a 280 GHz bandwidth and showed no sign of substrate mode coupling (Fig. 4 ).

 

Fig. 4 Measured transmission parameter S₂₁ over the 280 GHz bandwidth. The transmission parameter S₂₁ confirms that the substrate modes have been suppressed for a substrate thickness of 30 µm.

Download Full Size | PDF

The electrical properties of another modulator, 20 µm thick, have been extracted from a full S-parameter characterization using a transmission line ABCD matrix curve fitting technique [23]. The electrical parameters of the modulator, consisting of the conduction losses α_m, dielectric losses α_d, CPW input impedance Z_in and effective index n_RF, are first approximated to generate an ABCD matrix from which the S-parameters can be backed up and compared to the measured S-parameters. Using a curve fitting algorithm, the electrical parameters are adjusted until the S-parameters calculated and the S-parameters measured are matched. The electrical properties α_m, α_d, Z_in and n_RF have been estimated to be 0.28 dB/(cm·GHz^1/2), 0.01 dB/(cm·GHz), 47 Ω and 2.19, respectively. An optical insertion loss of 3.7 dB has been obtained. The DC-V_π has been directly measured at 8.6 V.

As a result of good index matching, low propagation losses, good input impedance and very thin substrate, first order optical sidebands for the 20 µm thick modulator have been observed up to 300 GHz on the OSA. The modulator's response has been measured in 1 GHz increments. However, to better illustrate the sidebands, a modulation spectrum with a 5 GHz spacing between each RF frequency is represented in Fig. 5 . In the figure, each pair of sidebands, upper and lower, corresponds to the modulator's optical response normalized to the RF input power feeding the modulator at the corresponding RF frequency.

 

Fig. 5 300 GHz optical modulation spectrum. Each pair of sidebands centered around the optical carrier represents the mmW energy for a given mmW frequency upconverted to optical energy using the EO effect of the LiNbO₃.

Download Full Size | PDF

The probe and feed losses, harmonic generation in the RF source, and power meter limitations are the factors accounted for in the optical sidebands normalization process. The probe's insertion loss are provided by the vendor, whereas the feed losses were determined by backing out the return losses from S₁₁ parameter measurements. The harmonic generation in the RF sources was determined directly by measuring the corresponding optical sidebands. However, limits in the available power of the RF source above 170 GHz and the measurement capabilities of power meters at these frequencies inhibited the accurate characterization of the modulator beyond 170 GHz. The power meter measurements at the bandwidth transitions were adjusted, as necessary, to obtain a consistent and continuous modulation characterization.

The V_π of the modulator can be extracted from the sidebands measurements through the relation $V_{\pi }\left(f\right)=\pi \sqrt{Z_{in}\left(f\right)/\left(2P_{sb}\left(f\right)\right)}$ [24], where Z_in(f) is the CPW characteristic impedance and P_sb(f) is the power in Watts of the normalized optical sideband. The V_π measured using the normalized optical sidebands and the V_π calculated using the S-parameters measurements are represented in Fig. 6 in the 0-170 GHz range, which corresponds to the bandwidth where the equipment allows accurate sidebands normalization.

 

Fig. 6 Measured and calculated modulator half-wave voltage V_π. The measured V_π extracted from the sidebands measurements is in agreement with the V_π calculated using the S-parameters.

Download Full Size | PDF

The good agreement in Fig. 6 between the V_π measured from the normalized sidebands and the V_π calculated from the S-parameters shows that the normalization process used to characterize the modulator is correct and that the modulation spectrum shown in Fig. 5 is accurate in the 0-170 GHz range. A DC-V_π on the order of 8.6 V is obtained from extrapolating down to the DC the measured V_π represented in Fig. 6, which was also verified experimentally.

5. Conclusions

In conclusion, a LiNbO₃ EO phase modulator operating over the entire mmW bandwidth has been experimentally demonstrated, showing continuous sidebands up to 300 GHz. The design of the modulator has been presented and its main design criteria discussed. A precise index matching at 2.19 combined with low propagation loss, good impedance matching and strong mode overlap proved to be critical but not sufficient to allow the modulation over the entire mmW domain. Thinning the LiNbO₃ substrate to a few tens of microns to eliminate substrate modes is key to extending the modulation bandwidth to 300 GHz. A driving voltage V_π of 8.6 V is obtained at DC. We believe the integration of such a ultrahigh speed modulator in the existing WDM optical fiber network will considerably increase the capacity of the current network at minimum cost.