1. Introduction

In leveraging the CMOS manufacturing infrastructure to fabricate optical components and integrated circuits, one has immediate access to repeatable processes, reduced cost via economy-of-scale, ease of large-scale integration, improved component reliability, as well as advanced design, simulation and characterization tools. While these benefits are compelling on many levels, for some optical devices silicon is a sub-optimal material, which can impose significant limitations on device performance.

For example, silicon-based modulators typically operate with a bandwidth of 30 GHz [1] and have a theoretical limit below 60 GHz [2]. While such modulators may be adequate for some applications that need data transmission rates below 40 Gb/s, there is a range of emerging applications in RF photonics that require modulation rates well in excess of 100 GHz. In these cases, pure silicon-based modulators are no longer a viable option and, therefore, CMOS-compatible integration of silicon with other materials becomes necessary to address the pressing need for high-frequency performance. Thus, the integration of organic electro-optic materials with silicon has been demonstrated to yield modulators with low power consumption [3] and large bandwidths [4]. However, the organic nature of these materials makes them challenging to process and brings into question their long-term stability [5]. Similarly, the integration of compound semiconductors with silicon for high-speed and high-power modulation introduces a challenge of undesirable third-order nonlinearities, which leads to detrimental 4-wave mixing, two-photon absorption and degraded spur-free dynamic range. In this context, thin-film LiNbO₃ has been receiving increased interest [6–15] due to its attractive nonlinear properties (χ⁽²⁾ is an order of magnitude larger and χ⁽³⁾ is two orders of magnitude smaller than that in III-V’s) and wide-spread use in electro-optic applications, most notably, in full-spectrum millimeter-wave modulators [16]. In addition, the integration of thin-film LiNbO₃ on a quartz handle to realize a traveling-wave modulator operating from DC to 40 GHz offered a glimpse of the potential for heterogeneous integration [17]. The present paper demonstrates a record-breaking 110 GHz modulation frequency and describes techniques used to enable such a CMOS-compatible device on a silicon platform.

A traveling-wave EO modulator requires an exquisite matching of the RF index with the optical index in order to operate efficiently over a broad range of frequencies. This poses a problem for lithium-niobate-based devices since the material is highly dispersive over that broad range of frequencies: whereas the refractive index at optical wavelengths is about 2.2, the RF index can be as high as 9.2. To bridge this disparity, a number of approaches have been implemented that rely on pulling the RF field out of lithium niobate into the surrounding low-index medium and thereby reduce the effective index of the RF traveling wave. To this end, tall electrodes have been used, ridges etched in lithium niobate, and SiO₂ buffer layers implemented [16]. While these techniques have been successful in reducing the RF index [16] and matching it more closely with the optical index, they tend to also reduce the overlap between the optical and RF modes and hence compromise modulation efficiency. An alternative approach to reducing the RF index relies on aggressively thinning the entire LiNbO₃ device layer [18]. A variation of this method is presented here with the additional benefit of the ability to ‘tune’ the device iteratively, since the tuning is the last step of the fabrication process and can be repeated as needed to optimize modulator’s performance. The tuning process is possible thanks to the use of the inverted ridge optical waveguide structure [19]. The inverted ridge structure as well as the tuning process are discussed in greater detail in the following sections. Both are made possible through the use of mechanical thinning, rather than the more commonly used crystal ion slicing (CIS) [6–15,17] technique, to attain the thin lithium niobate sample. Whereas CIS has been successfully employed to repeatably produce single-crystal films with prescise sub-micron thicknesses, the ion implantation leads to crystal-structure damage, which is repaired using high-temperature annealing. Furthermore, ion implantation requires specialized, high-cost equipment. In contrast, mechanical thinning is a low-temperature, low-energy process that relies on more widely available lab equipment.

2. Fabrication

The CMOS-compatible lithium-niobate modulator is fabricated by first preparing the composite substrate [19]. To this end, a ridge that provides lateral confinement for light propagation is etched in an X-cut lithium niobate substrate, which is then affixed to an undoped silicon wafer via low-index adhesive such as UV15. The UV15 adhesive layer is ~7 µm thick and varies no more than 2 µm across the 3 inch wafer. The exposed facet of lithium niobate is then ground and polished to an optical quality surface on a wafer scale to a thickness of 4 µm with a thickness variation no greater than 2 µm across the wafer. The lithium-niobate thinning step is a low-temperature, low-energy process and as such is compatible with CMOS. Figure 1(b) shows a cross-section of the resulting LiNbO₃ and adhesion layers. Note that the upper cladding is only used for protection during the end-facet polishing to obtain the diagnostic cross-sectional view of Fig. 1(b) and is absent from the functional device as in Fig. 1(a).

 

Fig. 1 SEM images of the fabricated device. (a) Tuned CPW structure, the dotted yellow lines are where the inverted ridge waveguide exists. (b) Cross sectional view of the inverted ridge waveguide. The thick polymer layer on the top surface in (b) is required to protect the device layer during end facet polishing and is not present in (a).

Download Full Size | PDF

The prepared substrate is then used to fabricate the modulator. Electrodes of the coplanar waveguide (CPW) are patterned using photo-lithography followed by electroplating up to a thickness of 6 µm, Figs. 2(a) and 2(b). In this work gold was used for the CPW electrodes. However, other highly conductive metals such as aluminium or copper could be substituted for gold without significant degradation in device performance. The length, width and gap between the electrodes are 1 cm, 45 µm and 15 µm, respectively. After building up the electrodes, 600 nm of chromium is sputtered on the entire structure, Fig. 2(c), to protect the soft gold electrodes and act as a hard mask in subsequent etching steps. The resist and the electroplating seed layers are then stripped, and the end result is an Au CPW covered with a Cr hard mask, as shown in Fig. 2(d).

 

Fig. 2 Fabrication and tuning LiNOS modulators. (a) Seed layers are deposited and photoresist is patterned; (b) 6 µm of Au is electroplated; (c) 600 nm Cr hardmask is sputtered; (d) resist and seed layers are stripped, (e, f) LiNbO3 is partially etched to reduce RF index and match it to the optical index.

Download Full Size | PDF

At this stage, scattering parameters are collected for the devices, and with a transmission line ABCD matrix curve fitting technique [14] the electrical properties are extracted for subsequent tuning. Partially etching lithium-niobate between CPW electrodes, Figs. 2(e), 2(f) and 1(a), replaces the high-dielectric-constant LiNbO₃ with low-dielectric-constant air and thereby reduces the effective index of the RF waveguide; the more material that is removed, the lower the effective index becomes. This way, fine-tuning of the RF waveguide parameters is achieved and velocity matching realized. A directional, highly anisotropic lithium-niobate etch is obtained in an inductively-coupled CF₄(6 sccm)/N₂(28 sccm)/O₂(0.5 sccm) plasma with previously deposited chromium acting as an etch mask that is self-aligned to the CPW electrodes. The etch is time multiplexed. The number of cycles determines etch depth and each cycle consists of 1 minute etching in a 600 W plasma under 400 W bias. The etch rate of X-cut LiNbO₃ is ~27 nm a minute and the selectivity between LiNbO₃ and Cr is ~5.4:1. The lithography and etch used here are steps routinely used in the conventional CMOS process. It can be seen below that exposing the CPWs to the inductively-coupled plasma tuning etch does not introduce excessive roughness to the electrode sidewalls, nor does it diminish the device’s scattering parameters.

3. Characterization

The removal of 1.5 µm of LiNbO₃ in one of the tested devices reduced the effective RF index from 2.54 to 2.33 and increased the waveguide impedance from 39.85 Ω to 44.95 Ω. This brings the RF waveguide closer in line with both the optical-waveguide index 2.2, and the standard 50 Ω RF impedance. Also, notably, the transmission characteristics of the CPW improved following the tuning etch: the reflection (S₁₁) fell consistently and the transmission (S₁₂) increased at frequencies above 90 GHz, Fig. 3(a). During RF characterization the transmission lines are terminated with 50 Ω loads.

 

Fig. 3 (a) Measured transmission S₂₁ and reflection S₁₁ parameters of a device before and after tuning etch. (b) Measured modulator half-wave voltage Vπ (values extracted from data shown in Fig. 4).

Download Full Size | PDF

The optical portion of the modulator was characterized by coupling polarization-maintaining single-mode fibers to the end facets of the inverted-ridge lithium-niobate waveguide. For TE launched light, FDTD simulations indicate that with a ridge etch depth of 1.3 µm, so long as the etched slab thickness is greater than 2.4 µm and less than 4.5 µm, the waveguide is capable of single mode operation; our experimental results agree with the simulated device data. The waveguide propagation loss of ~1 dB/cm was found by the cutback method. This value appears to be unaffected by the tuning process. The overall insertion loss of about 15 dB is dominated by the coupling loss between a single-mode fiber with mode size ~10.5 µm and the end facets of the device with mode diameter ~5 µm. The coupling loss may be reduced significantly by introducing a taper [20] or by using a high numerical aperture fiber [21]. Future devices will ideally utilize interlayer coupling via either grating couplers [8,13], or adiabatic tapers [10].

The completed modulator was characterized by launching TE-polarized light with a wavelength 1550 nm into the optical waveguide. An RF signal with frequencies ranging from DC to 110 GHz was launched into the RF coplanar waveguide and the resulting modulating spectrum was observed at the optical output, Fig. 4. CPW transmission lines were not terminated for measurements observing modulation. The modulation of the input wavelength manifests as the appearance of sidebands that can be easily discerned using an optical spectrum analyser [22]. Figure 4 shows a comparison of the spectra obtained with a modulator that was not etch-tuned, Fig. 4(a), with one that was, Fig. 4(b). In both devices, DC-V_π is about 9.4 V. However, the high-frequency response of the etch-tuned modulator is considerably improved over its un-tuned counterpart, which shows a significant roll-off in the modulation spectrum. Values for half-wave voltage plotted in Fig. 3(b) were extracted from data collected in Fig. 4, and indicate that the 3 dB Vπ(𝑓) of the tuned device is ~40 GHz. The tuned modulator shows a frequency response up to the measured 110 GHz with a roll off of 7.85 dB.

 

Fig. 4 Optical spectra of a carrier modulated with RF up to 110 GHz in (a) untuned and (b) tuned device.

Download Full Size | PDF

4. Conclusion

Presented in this paper is, to the best of our knowledge, the first instance of a traveling wave LiNbO₃ modulator integrated onto a silicon substrate using a CMOS-compatible process having a measured frequency response up to 110 GHz. Etch-tuning the device brings the RF index down to match the optical index and thereby increases the frequency response beyond 100 GHz. The combination of the modulator’s exceptional broadband performance and its CMOS compatibility opens unprecedented opportunities in electronic-photonic integration. With the existing wafer-bonding, alignment and thinning technologies, used routinely by the industry to fabricate 3D integrated circuits [23,24], the integration of thin-film LiNbO₃ with electronic circuits will likely yield a cost effective approach to the realization of high-performance 3D electronic-photonic chips. Ultimately, hybrid, CMOS-compatible, thin-film 3D integration of III-V laser sources and photodetectors, silicon electronics, and broadband LiNbO₃ modulators will enable cost-effective digital and analogue data processing and communication at rates hitherto unattainable.