Magneto-optical non-reciprocal devices in silicon photonics

Epitaxial growth of a magneto-optical garnet on silicon is challenging because of the large mismatch in physical properties between the garnet and silicon. For example, the lattice constant of yttrium iron garnet Y₃Fe₅O₁₂ (YIG), which is the most popular magneto-optical garnet, is 1.238 nm, whereas that of silicon is 0.543 nm. Furthermore, the linear thermal expansion coefficients of YIG and silicon are 10.4 × 10⁻⁶ and 2.33 × 10⁻⁶ K⁻¹, respectively.

Although a single-crystalline magneto-optical garnet has not been grown on silicon, progress in the deposition of magneto-optical garnets has been made. Stadler et al demonstrated the deposition of YIG poly-crystalline thin films on MgO and quartz substrates through a novel reactive sputtering method with rapid thermal annealing [19, 20]. Körner et al [21] deposited a poly-crystalline Bi₃Fe₅O₁₂ layer through pulsed laser deposition (PLD) on SiO₂ with an annealed YIG buffer layer.

Bi et al [22] demonstrated the growth of a magneto-optical garnet on silicon through PLD. They first deposited a 20 nm-thick YIG seed layer on an oxidized silicon substrate at 550 °C. The deposited layer was subjected to rapid thermal annealing at 850 °C for achieving crystallization in the YIG phase. Subsequently, bismuth-substituted YIG (Bi_1.8Y_1.2)Fe₅O₁₂ (Bi:YIG) or cerium-substituted YIG (Ce₁Y₂)Fe₅O₁₂ (Ce:YIG) was deposited on the YIG seed layer through PLD at 650 °C. The grown Bi:YIG and Ce:YIG layer showed poly-crystalline garnet phases with saturation Faraday rotations of −838 and −830° cm⁻¹, respectively, at a wavelength of 1550 nm.

It is desirable to prepare magneto-optical garnets on silicon using deposition techniques from the viewpoint of the device fabrication process. The Faraday rotation available in the magneto-optical garnet prepared by deposition is much less than that of an epitaxially grown single-crystalline garnet, i.e. −830° cm⁻¹ in PLD Ce:YIG [22] and −4500° cm⁻¹ in sputter-epitaxially grown Ce:YIG [23] at a wavelength of 1550 nm. However, in order to reduce the required magneto-optical interaction length, a greater improvement in the Faraday rotation of the deposited garnet layer is required.

Compared to the deposition techniques, bonding techniques are advantageous because single-crystalline layers that are epitaxially grown in a separate process can be used. When integrating a magneto-optical garnet with other materials for optical waveguide applications, a tight, uniform contact between the materials is required for obtaining a sufficient interaction between the light waves and magneto-optical garnet. A surface activated direct bonding technique is a powerful tool to realize a tight contact between dissimilar crystals [24–26]. We developed a surface activated direct bonding technique for integrating a magneto-optical garnet on a silicon waveguide [14]. The surfaces of the target wafers are activated in a vacuum chamber. Subsequently, the activated surfaces are brought into contact at room temperature. While applying a constant pressure, the contacted sample is annealed to obtain firm bonding.

There are several ways to achieve surface activation; they include argon ion beam and argon plasma irradiation. We found that exposing wafers to oxygen or nitrogen plasma was effective for a wafer combination of silicon and magneto-optical garnet. The surface smoothness is an important factor for successful bonding. We examined the effects of pre-cleaning and surface activation processes on the roughness of the wafer surfaces [13]. A 500 nm-thick cerium-substituted yttrium iron garnet (YCe)₃Fe₅O₁₂ (Ce:YIG) single-crystalline layer was grown on a (111)-oriented substituted gadolinium gallium garnet (GdCa)₃(GaMgZr)₅O₁₂ (SGGG) substrate through a sputter epitaxial growth technique. A (100)-oriented SOI wafer, in which the silicon layer was 220 nm thick, was used for examining the effects of pre-cleaning and surface activation processes on the roughness of the wafer surfaces. The details of pre-cleaning processes are described in [13].

Oxygen, nitrogen or argon gas was supplied at a flow rate of 100 sccm to a plasma reactor chamber with a pressure of 120 Pa. To generate the plasma, an RF power of 400 W at 13.56 MHz was applied to an electrode of 4 inch diameter. Ce:YIG and SOI wafers were cut into 20 × 20 mm² pieces and placed on the electrode. The roughness of the wafer surfaces was measured using an atomic force microscope (AFM). The surface roughness, which is defined by the average of rms values obtained in the AFM measurement, is shown in figure 1 as a function of the plasma irradiation time. The measurement was performed on three areas of each wafer. As shown in figure 1, compared with the initial pre-cleaned state, the oxygen plasma irradiation reduces the surface roughness of both Ce:YIG and SOI for irradiation times between 10 and 30 s. In the case of nitrogen plasma irradiation, slight increases occurred in the surface roughness of both Ce:YIG and SOI for irradiation times <30 s. In contrast, the argon plasma irradiation increases the surface roughness of Ce:YIG monotonically with the increase in irradiation time. From these observations, it can be concluded that an oxygen or nitrogen plasma irradiation between 10 and 30 s is suitable for reducing the roughness of Ce:YIG and SOI surfaces.

Zoom In Zoom Out Reset image size

After the surface activation process through the oxygen plasma irradiation for 30 and 10 s, respectively, for Ce:YIG and SOI, the wafers were brought into contact inside a vacuum chamber. The wafers were pressed together with a pressure of 5 MPa for 8 h. During this process, the sample was annealed at 200 and 250 °C for Ce:YIG and SOI, respectively. The Ce:YIG–SOI bonding was successfully achieved. The bonding strength measured through a tensile test was <0.02 MPa for the sample activated with oxygen plasma using an RF power of 400 W.

The bonding strength was improved to >1.8 MPa by exposing the wafers to oxygen plasma generated using a higher RF power of 500 W [13]. Such a high bonding strength can be attributed to the increase in surface activation energy. There was no remarkable degradation in the surface roughness of the wafers after the plasma irradiation generated using an RF power of 500 W. In addition, a bonding strength of 5.3 MPa was obtained by applying the surface activation process of nitrogen plasma generated with an RF power of 500 W for 10 s. In this process, the annealing was performed at 200 °C at a pressure of 6 MPa for 30 min.

The structure of an SOI waveguide optical isolator based on an MZI is shown in figure 2. The MZI is composed of 3 × 2 couplers. The magneto-optical phase shifters installed in the waveguide arms of the interferometer have a Ce:YIG cladding layer directly bonded on a silicon waveguide core. A magnetostatic field is applied in the transverse direction of the light propagation in the film plane of Ce:YIG to saturate its magnetization. By virtue of the first-order magneto-optical effects, the propagation constant of TM modes propagating in the magnetized waveguide changes from that propagating in a non-magnetized waveguide. The change in the propagation constant varies depending on the propagation direction and the direction of the applied magnetostatic field. Since the external magnetic fields are applied in anti-parallel directions in two arms of the interferometer, different magneto-optical phase changes are induced in the two arms. That is, the light wave propagating from left to right experiences a phase change of Φ₋ and Φ₊ in the left and right arm, respectively. When the propagation direction is reversed, the light wave experiences a phase change of Φ₊ and Φ₋ in the left and right arm, respectively.

Zoom In Zoom Out Reset image size

Because of the magneto-optical effect known as a non-reciprocal phase shift, a phase difference of Φ₋–Φ₊ occurs for the light wave propagating in the magneto-optical phase shifters from left to right. The phase difference can be set to −π/2 by adjusting the magneto-optical effect or length of magneto-optical phase shifter. This phase difference is canceled by the π/2 phase bias installed in the left arm. The phase bias can be realized by adjusting the optical path difference between the two waveguide arms. Hence, the light wave propagating in the waveguide arms becomes in phase and interferes constructively in the output 3 × 2 coupler. The light wave launched into the central input port of the left 3 × 2 coupler emerges at the central output port of the right 3 × 2 coupler. This corresponds to the forward direction.

For the backward direction, the phase difference produced by the magneto-optical effect changes its sign, i.e. Φ₊ − Φ₋ = π/2. Because the remaining phase bias yields a π/2 phase difference in the left arm, a total phase difference of π is introduced between the two interferometer arms. Destructive interference occurs at the left coupler. The light wave does not come out at the initial input port, but instead is radiated out from the side waveguides of the 3 × 2 coupler.

The non-reciprocal phase shift Φ₊–Φ₋ is calculated for a waveguide composed of a Ce:YIG cladding layer, silicon waveguide core and buried oxide (SiO₂) under the cladding layer. The under cladding layer with a refractive index of 1.444 should be sufficiently thick so that the effects of the silicon substrate can be neglected. The upper cladding layer Ce:YIG is assumed to have a saturation Faraday rotation of θ_F = −4500° cm⁻¹ and a refractive index of 2.20 at a calculated wavelength of 1550 nm. Figure 3 shows the calculated non-reciprocal phase shift as a function of the silicon waveguide core thickness. If we compare the non-reciprocal phase shift between a Ce:YIG/Si/SiO₂ and Ce:YIG/GaInAsP/InP slab waveguide, the Ce:YIG/Si/SiO₂ waveguide provides a much larger non-reciprocal phase shift than the Ce:YIG/GaInAsP/InP waveguide [36]. This can be understood by considering that the under cladding layer SiO₂ has a much lower refractive index than InP (n = 3.17 at λ = 1550 nm). This results in a large penetration of the optical field into the upper cladding magneto-optical layer and consequently a larger magneto-optical effect. A non-reciprocal phase shift of 7.12 mm⁻¹ can be obtained for the Ce:YIG/Si/SiO₂ slab waveguide composed of a 200 nm-thick silicon waveguide core at a wavelength of 1550 nm.

Zoom In Zoom Out Reset image size

For a practical device, a three-dimensional waveguide such as a rib or rectangular core waveguide is used for the lateral optical field confinement. The non-reciprocal phase shift Φ₊–Φ₋ is calculated in a silicon rectangular waveguide using (1), which was derived from perturbation theory [37, 38].

$$\begin{equation} {\Phi _ + } - {\Phi _ - } = \frac{{\int\!{{\varepsilon _0}\gamma \frac{\partial }{{\partial x}}\frac{{{{\left| {{H_y}} \right|}^2}}}{{{n^4}}}{\rm{d}}S} }}{{\int {\frac{1}{{{n^2}}}{{\left| {{H_y}} \right|}^2}{\rm{d}}S} }}, \end{equation} \tag{ 1 }$$

where n and ε₀ are the refractive index of Ce:YIG and vacuum permittivity, respectively. H_y denotes the distribution of the transverse magnetic field component of the TM mode in a waveguide. The integral is computed over the entire cross section of waveguide. γ denotes the off-diagonal component of the Ce:YIG permittivity tensor, which is related to the saturation Faraday rotation θ_F by

$$\begin{equation} \gamma = \frac{{n\lambda }}{\pi }{\theta _{\rm{F}}}, \end{equation} \tag{ 2 }$$

where λ is the wavelength of light.

The waveguide structure assumed in the calculation is shown in figure 4. It consists of a silicon rectangular waveguide core. The over and under cladding layers are Ce:YIG and SiO₂, respectively. The side regions of the silicon waveguide core are assumed to be air.

Zoom In Zoom Out Reset image size

The calculated non-reciprocal phase shift that the fundamental TM guided mode experiences at a wavelength of 1550 nm is plotted as a function of the waveguide core thickness in figure 3 for various core widths W. The optical field distribution is calculated through a full vector finite element method. Compared to the slab waveguide, the non-reciprocal phase shift decreases in rectangular waveguides because the transverse field component in a Ce:YIG cladding layer is smaller. This tendency becomes prominent for narrower waveguides. The maximum non-reciprocal phase shift is obtained in a 200 nm-thick silicon slab waveguide. In the case of silicon rectangular waveguides with widths of 450 and 550 nm, the maximum non-reciprocal phase shift is obtained at the core thicknesses of 240 and 220 nm, respectively. Considering the specifications of commercially available SOI wafers, the device is designed in an SOI wafer with a 220 nm-thick top silicon layer. The length of the non-reciprocal phase shifter was calculated to be 430 and 327 μm for silicon waveguides having widths of 450 and 550 nm, respectively, for a 220 nm-thick silicon waveguide.

Non-reciprocal phase shifts are provided only for TM modes in vertically asymmetric waveguides as shown in figure 4. Non-reciprocal phase shifts are obtainable for TE modes when the magneto-optical material is placed in a horizontally asymmetric manner as is shown in figure 5. The MZI isolator that works for a TE mode can be realized using non-reciprocal phase shifters with the configuration shown in figure 5. Moreover, the polarization independent operation of a MZI isolator can be obtained by introducing both in-plane and out-of-plane asymmetric structures together with the proper magnetization of magneto-optical material [39]. In addition, the polarization independent isolator can be realized by adopting a proper design of an interferometer waveguide [40]. It is worth noting that the MZI based optical isolator can be extended to an optical circulator by replacing the 3 × 2 couplers with 3 dB directional couplers. This will be addressed in section 5.

Zoom In Zoom Out Reset image size

An SOI waveguide optical isolator based on an MZI was fabricated through the process shown in figure 6. An SOI wafer having a 220 nm-thick top silicon layer and a 3 μm-thick buried oxide layer is used in this device. Here, the thickness of the buried oxide layer was so chosen that the silicon substrate did not affect a fundamental TM mode propagating in the top silicon layer. The silicon waveguide pattern was drawn using an electron beam lithography system. Using a reactive ion etching (RIE) technique with CF₄, the waveguide mask was transferred into a 300 nm-thick SiO₂ layer deposited on the silicon guiding layer. The silicon rectangular waveguide of height 220 nm and width 450 nm was formed through SF₆ RIE.

Zoom In Zoom Out Reset image size

A single crystalline Ce:YIG layer was grown on a (111)-oriented SGGG substrate. The Ce:YIG layer was 500 nm-thick; this was sufficient for an evanescent field to be unaffected by the SGGG substrate. A 1.5 × 1.5 mm² Ce:YIG die was directly bonded as an upper cladding layer using a surface activated direct bonding technique. Nitrogen plasma generated using an RF power of 500 W was applied for 10 s as a surface activation process. After bringing the two wafers into contact in a vacuum chamber, a pressure of 6 MPa was applied at 200 °C for 30 min [41].

In order to apply anti-parallel magnetic fields inside the two arms of the interferometer, a three pole magnet system composed of a pair of compact permanent magnets was used. A magnetic field of 16 kA m⁻¹ was sufficient to saturate the magnetization of the Ce:YIG layer in the in-plane direction. The distance between the two arms of interferometer was fixed as 500 μm to accommodate the magnet. Because of the strong optical field confinement of the silicon waveguide in the lateral direction, the curved sections were designed with a curvature radius of 10 μm. By virtue of this, the device footprint was greatly reduced, as shown in figure 7, compared to the optical isolator fabricated with a silicon rib waveguide, which was 4 mm in length [15].

Zoom In Zoom Out Reset image size

The fabricated isolator was characterized using the experimental setup shown in figure 8. A fundamental TM mode emitted from a broadband light source was introduced into the isolator through a polarization maintaining fiber (PMF). The transmitted light wave was probed using another PMF. Its spectrum was measured using an optical spectrum analyzer. A 2 × 2 optical switch was used to reverse the light propagation direction through the device/fiber test setup for measuring the forward and backward transmittance.

Zoom In Zoom Out Reset image size

Figure 9 shows the measured fiber-to-fiber transmittance under a fixed external magnetic field. The transmittance was observed to be remarkably different depending on the light propagation direction. The interference was also reversed when the magnetostatic field direction was reversed. This implies that the sign of the magneto-optical phase shift was reversed. The maximum isolation that was defined by the transmittance ratio of the forward to backward directions was measured to be 30 dB at a wavelength of 1548 nm. Compared to the transmittance of the reference waveguide without a Ce:YIG cladding layer, which included the propagation loss of a straight Si waveguide and 17 dB fiber-to-waveguide coupling losses at the input and output interfaces, the excess loss of the isolator was measured to be 13 dB. One of the causes of the insertion loss is the losses caused by the reflection and scattering at the interfaces between the air and Ce:YIG cladding regions. The loss is calculated to be 3.7 dB/interface at a wavelength of 1550 nm using an analysis based on the mode expansion algorithm. The losses can be reduced by depositing an upper cladding material having a refractive index of 2.20, which is close to that of Ce:YIG. The remaining loss of 5.6 dB can be attributed to the absorption of Ce:YIG and the bending losses of the curved waveguides included to form the MZI.

Zoom In Zoom Out Reset image size

The isolator exhibited a bandwidth of 1 nm, in which an optical isolation >20 dB was obtained. Because the phase bias was fixed at 100.5π, which is rather large compared to an ideal value of 0.5π, the free spectral range of the interferometer was greatly reduced. Hence, the isolation bandwidth was limited to 1 nm. When a phase bias of 0.5π is employed, the bandwidth of the optical isolations >20 dB is calculated to be 39 nm [42]. It is shown that the isolation bandwidth can be increased further by canceling the wavelength dependence of the magneto-optical phase shift with that of the phase bias [43].

Ghosh et al [44] demonstrated a silicon MZI optical isolator with an isolation of 25 dB. The advantage of this device is that it operates in a unidirectional magnetostatic field. They bonded a Ce:YIG die on a silicon MZI waveguide through an adhesive benzocyclobutene (BCB) bonding technique. A Ce:YIG layer was grown on an SGGG substrate through a sputter epitaxial growth technique. It should be noted that since the evanescent field penetrating the Ce:YIG layer is responsible for the magneto-optical phase shift, the thickness of intermediate layer between the silicon and garnet must be stringently controlled in order to obtain the desired magneto-optical phase shift [45].