1. Introduction

The first-order magneto-optical (MO) effect provides lightwave with a unique feature of gyrotropic polarization rotation so-called Faraday effect and polar/longitudinal/transverse Kerr effect. Optical nonreciprocity, which breaks time reversal symmetry of light, is necessary to realize optical isolator. Magneto-optical garnet such as yttrium iron garnet (Y₃Fe₅O₁₂, YIG) and terbium iron garnet (TIG) has large MO effect as well as low optical absorption in a telecommunication wavelength range [1, 2]. Cerium-substituted YIG (CeY₂Fe₅O₁₂, Ce:YIG) has the highest figure of merit (FOM) defined by the ratio of MO effect to optical absorption [3–6]. The FOM of Ce:YIG is 90 °/dB for a Faraday rotation of −4500 °/cm and an optical absorption of ~5 dB/mm. Although MO isolators using ferromagnetic metals [7, 8] and magneto-optical compound semiconductor [9–11] have been investigated, they suffer from higher loss. The drawback of MO garnet lies in difficulty in the epitaxial growth on semiconductor substrate. Bi, et al. achieved growth of poly-crystalline Ce:YIG with a YIG seed layer on a Si substrate, in which YIG is relatively easily to crystallize itself without inheriting the lattice from a substrate [12, 13]. However, the high process temperature up to 800 °C should be compromised with a CMOS process for silicon photonic integrated circuits (PICs).

A direct bonding technique is a promising approach to integrate a single-crystalline garnet with Si PICs. Plasma assisted or surface activated process can effectively bond MO garnet on a silicon-on-insulator (SOI) substrate at low process temperature. We have demonstrated Mach-Zehnder interferometer (MZI)-based MO isolators and circulators with high isolation and wide operation bandwidth [14, 15]. Also, we investigated temperature dependence of optical properties of Ce:YIG and demonstrated temperature insensitive isolator operation for the backward propagation [16]. Huang, Pintus, et al. demonstrated MO isolators and circulators based on a ring resonator (RR) and an MZI with an electromagnet to control a magnetic field and device temperature [17–20]. A (Ca, Mg, Zr)-substituted gadolinium gallium garnet (SGGG) substrate used for Ce:YIG growth was thinned from the backside to a few micrometer by mechanical polishing. Small footprint, high isolation, and widely tunable bandwidth were demonstrated.

By controlling the magnetization direction of MO material, MO effect changes its sign. This can be used to construct an optical switch with a 2 × 2 port configuration employing MZI or RR. The magnetization is flipped by an external magnetic field induced with electric current flow or permanent magnet. An MO switch was demonstrated with an amorphous silicon (a-Si:H) waveguide formed on Ce:YIG in an MZI configuration [21]. Fast temporal response was confirmed by 2.5-Gbps-modulated current in the RR-based MO circulator [19].

In this article, we firstly review the MO nonreciprocal devices especially based on an MZI configuration. We then discuss the insertion loss which is the critical factor for practical applications. We describe numerical analysis of absorption and junction losses at the MO cladding region. Next, we present an MO switch having latching operation, which means a switch state is maintained without any power supply. It contributes to lowering the power consumption of optical circuit switching with relatively low switching frequency as well as realizing optical logic circuits.

2. MO nonreciprocal devices fabricated by direct bonding

The MZI has two phase shifters between two arms as shown in Fig. 1. The nonreciprocal phase shifter provides phase difference of ${\text{Φ}}_{NPS}=\text{Δ}\beta \cdot L=\mp \pi /2$ derived from MO phase shift $\left|\text{Δ}\beta \right|=\left|{\beta }_{+}-{\beta }_{-}\right|$ induced by the transverse MO Kerr effect. The sign depends on the propagation direction for the magnetization in anti-parallel direction fixed by external magnetostatic field. The reciprocal phase shifter provides phase bias of ${\text{Φ}}_{RPS}=\beta \cdot \text{Δ}L=\pi /2+2m\pi$ (m: integer) caused by an optical path length difference $\text{Δ}L$ between two arms. For the forward direction, light waves propagating in two arms is set to be in-phase. Input light induces constructive interference and transmits to the output port. For the backward direction, light waves propagating in two arms become out-of-phase. Light coming back from the output port interferes destructively and is not coupled to the input port. Therefore, optical isolation is achieved. In the case of MO circulator, the 3 dB coulperes are replaced by 2 × 2 couplers. Light transmission to the cross or bar direction is then determined by the in-phase or out-of-phase interferences, respectively. Input and output directions among the four ports are connected as 1→2→3→4→1.

 

Fig. 1 Structure of the MZI-based waveguide optical isolator [14].

Download Full Size | PDF

The above operation is designed at a specific wavelength. The wavelength dependences of MO effect and refractive index of the materials change the ${\text{Φ}}_{NPS}$ and ${\text{Φ}}_{RPS}$ for different wavelengths. The operation bandwidth is mainly determined by ${\text{Φ}}_{RPS}$ where shorter $\text{Δ}L$ results in wider free-spectral range (FSR) of MZI interference. Figure 2(a) shows the measured spectra of an MZI-based isolator with $\text{Δ}L$ = 3.7 µm fabricated by our direct bonding process [15]. The blue and red lines show the forward and backward transmission of isolator for the transverse magnetic (TM) mode transmission. Over 20 dB isolation was obtained for 8-nm bandwidth.

 

Fig. 2 (a) Measured spectra and (b) Microscopic image of fabricated MZI-based MO isolator [15].

Download Full Size | PDF

A 1.5-mm-square die of Ce:YIG/SGGG was bonded on the Si waveguides as shown in Fig. 2(b). Straight reference waveguides were formed adjacent to the isolator beneath and outside the Ce:YIG cladding region. The measured spectra of reference waveguides with and without the Ce:YIG cladding layer are shown in the green and orange lines, respectively, in Fig. 2(a). The results indicate that propagation loss of Si waveguides and coupling loss between the fiber and waveguide were up to 20 dB. The insertion loss of the device excluding the coupling loss was estimated to be 13.5 dB, which includes a loss of 11 dB due to the Ce:YIG clad and an excess loss of 2.5 dB.

Let us discuss the loss due to the Ce:YIG clad. It is inherent for this device and should be reduced for practical uses. There are scattering and reflection losses owing to mode mismatch at the boundary between the air and Ce:YIG upper cladding regions, which is called “junction” hereafter. We simulated the transmittance at the waveguide junction using an eigen-mode expansion method with a parameter of Si height as shown in Fig. 3(a). The waveguide width was fixed at 450 nm. In the following discussion, the wavelength was assumed to be 1550 nm. At the input waveguide with an air clad, the mode field distributes more in a lower cladding layer. On the other hand, at the output waveguide with a Ce:YIG cladding layer, the mode field distributes more in an upper cladding layer than in a lower cladding layer. This mode mismatch causes the transmission loss due to reflection and scattering. The junction loss is 4.3 dB for the TM mode at a Si height of 220 nm. Since two junction exist in the actual device, it is significant for the insertion loss.

 

Fig. 3 (a) Simulated transmittance of the waveguide junction for the TE and TM modes at a wavelength of 1550 nm. Upper cladding materials of input waveguide are air and SiO₂ for the filled and open circles, respectively. (b) Simulated propagation loss in the Ce:YIG/Si/SiO₂ waveguide with the width of 450 nm and height of 220 nm at a wavelength of 1550 nm.

Download Full Size | PDF

In addition, the propagation loss is induced by the optical absorption of ~5 dB/mm of the Ce:YIG layer. As show in Fig. 3(b), the TM mode shows higher propagation loss than the transverse electric (TE) mode because of larger field distribution in the Ce:YIG layer. The Si height strongly affects to the amount of MO phase shift $\left|\text{Δ}\beta \right|$, which is maximized at around 220 nm. Although one can see that an increase in Si height can reduce the loss, it results in larger device size as well as being incompatible with other Si PICs typically optimized for 220-nm height.

The junction loss for TM mode can be reduced by covering the silicon waveguide with an upper cladding of SiO₂. Furthermore, the TE mode propagation shows lower junction loss, while the MO phase shift is not obtained for the TE mode with the waveguide structure fabricated with a bonded MO upper cladding layer. Fabrication of horizontally asymmetric structure to obtain MO phase shift for the TE mode [22] is still challenging on an SOI platform. A straightforward approach for loss reduction is installation of TE-TM mode conversion at the input and output ends of the nonreciprocal phase shifter operating for the TM mode. Waveguide TE-TM mode converters are widely studied in Si-PICs [23, 24]. MO isolators integrated with these converter have been demonstrated for TE mode input light [25, 26]. Figure 4 shows the conceptual image of the low-loss MO isolator compared with the current configuration. Since light propagates as the TE mode at the junctions of air and Ce:YIG cladding regions, junction loss can be reduced. The Ce:YIG cladding region is minimized to reduce the absorption loss. The length of a TE-TM mode converter is expected to be ~0.2 mm including a taper. The TE mode experiences the absorption loss of Ce:YIG for ~0.5-mm-long propagation distance. The breakdown of simulated loss is shown in Table 1. The total loss can be about 2.5 dB when the excess loss of the TE-TM mode converter is negligible.

 

Fig. 4 Schematic of (a) current and (b) low-loss configurations. A Ce:YIG upper cladding layer exists in the green regions.

Download Full Size | PDF

Table 1. Loss breakdown for waveguide with the width of 450 nm and height of 220 nm.

View Table

3. MO switch with latching operation

Optical output intensity of an MO device is controlled by the MO effect which changes its sign when the magnetization is flipped. Thus, an MO MZI with 2 × 2 ports can function as an optical switch with cross-bar connections. A magnetostatic field is applied by electric current flow or permanent magnet, while the magnetization can be self-holding using magnetic nonvolatility. We then proposed an MO switch with latching operation by integrating a thin-film magnet [27].

Figure 5 shows the device structure and cross section of the MO phase shifter with a thin-film magnet. A-Si:H waveguide was formed on a Ce:YIG/SGGG substrate with an MZI configuration having two phase shifters similar to the MZI nonreciprocal device mentioned above, where the magnetization direction of Ce:YIG is dynamically controlled. FeCoB is a soft magnetic material, which has strong residual magnetization and a relatively low coercive force. The magnetization of FeCoB is aligned by a current-induced magnetic field as shown in Fig. 5(b). The optical switching state is determined by the magnetization direction of the Ce:YIG layer where a magnetic field is applied by the FeCoB layer as shown in Fig. 5(c). A 800-nm-thick SiO₂ cladding layer is deposited to separate FeCoB from a-Si:H waveguides so that the optical absorption of FeCoB can be negligible. The dimension and structure of FeCoB magnet is designed to generate magnetic field of 50 Oe at the Ce:YIG layer enough to saturate the magnetization. In order to direct the magnetization transverse to the light propagation direction, magnet strips with the shape anisotropy in the transverse direction are arrayed along the waveguide of MO phase shifter.

 

Fig. 5 MO switch with latching operation. (a) Device structure of MZI configuration. (b) A current changes the magnetization direction of FeCoB thin-film magnet, which changes a switch state. (c) The switch state is maintained because of non-volatility of FeCoB.

Download Full Size | PDF

We fabricated and characterized the MO switch. TM mode light at a wavelength of 1562 nm was launched to the device under test. The electric current was injected to the silver electrode using micro probe. A pulsed-current of 1-µs width and a repetition rate of 10 kHz was fed by a function generator. The peak current flowing the electrode was measured to be +/−200 mA. The temporal response of optical output at the cross port was measured using a photodiode and an oscilloscope. Figure 6 shows the measured results. We observed that the optical output power had two steady-state levels when the electric current was zero. The optical output level was changed by the pulsed current with different direction and maintained without current after the switching. Therefore, the latching switch operation was successfully demonstrated. Because of the impedance mismatch of the electrode, slight overshoot and undershoot of the applied voltage occurred. The spike response of the optical output was caused by the magnetic field generated by the current flowing in the electrode, which also affected the magnetization of the Ce:YIG layer. The slow tail of ~100 µs after the spike was due to the electrical response of the gain amplifier in the photodiode used to compensate for weak optical output.

 

Fig. 6 Measured temporal response of MO switch to a pulsed current [27].

Download Full Size | PDF

4. Conclusion

We presented recent progress of waveguide MO devices with an MZI configuration. We discuss the insertion loss caused by the integration of MO garnet. The prospect of minimum loss down to 2.5 dB is shown by introducing the TE-TM mode conversion. We also presented an MO latching switch with a thin-film magnet. An experimental demonstration of changing and maintaining the switch state was shown with 1-µs pulsed-current driving.

The unique feature of MO effect is useful for photonic functional devices. Further improvement of both device and material including spintronics are expected.