1. Introduction

New classes of applications can be enabled by monolithically integrating optical signal processing with semiconductor device technology [1]. For a fully integrated optical integrated circuit, optical components including lasers [2], photodetectors [3], filters, modulators and waveguides have been fabricated on a silicon platform using compatible processes. One major roadblock in the integration of these components is the reflection of optical and near-infrared pulses from other components of the circuit back to the gain region of the laser due to geometric differences, refractive index mismatch, roughness, and scattering, leading to signal instabilities. In order to prevent back-reflection of optical pulses, optical isolators [4–7] need to be fabricated alongside lasers, requiring magneto-optical (MO) materials to obtain the necessary non-reciprocal light transmission. This integration has been a major processing problem because of differences between crystal structures, lattice parameters and thermal expansion coefficients of MO thin films and silicon. Additionally, even when MO films are grown, their performance is typically poorer than that of the bulk MO material, and their growth requires high thermal budgets [8]. An improved integration method for MO films on Si-based substrates including silica and silicon nitride is therefore necessary for enabling on-chip optical isolation and modulation.

The most commonly used MO materials are iron garnets based on yttrium iron garnet (YIG, Y₃Fe₅O₁₂), in particular cerium- or bismuth-substituted YIG (Ce:YIG or Bi:YIG, (Ce or Bi)_xY_3-xFe₅O₁₂ with x usually ~1) has been demonstrated as a candidate material for optical isolators [4,8]. YIG has a large unit cell with lattice parameter a = 12.376Å and space group Ia_{$\overline{з}$}d) [9], much larger than that for any of the common photonic substrates including Si (a = 5.431Å), GaAs (a = 5.6533Å) and InP (a = 5.8696Å) [9–11]. The thermal expansion coefficient for garnet films is much larger than that of non-garnet substrates (YIG: 10.4 x 10⁻⁶ K⁻¹, Si: 3 x 10⁻⁶ K⁻¹, GaAs and InP: 5 x 10⁻⁶ K⁻¹) [12]. The garnet phases also require a large thermal budget for growth on photonic substrates, and for some garnet compositions, the processing route must avoid the formation of unwanted phases such as ceria in the growth of Ce:YIG.

Previously, YIG and substituted garnet films have been obtained on non-garnet substrates using wafer bonding [13], metal-organic chemical vapor deposition (MOCVD) [14], rf sputtering [15–24] and pulsed laser deposition (PLD) [4,8,25–29]. Wafer-bonding of bulk garnet crystals to non-garnet substrates produces devices with excellent MO properties but this approach is not easily scalable and issues with alignment and contamination need to be addressed. MOCVD growth requires surface reactions in order to form stoichiometric oxides, narrowing the processing window for single phase growth. Garnet films have been grown by sputtering or PLD on a range of non-garnet substrates including Si, ceria, MgO and quartz, but the growth requires high deposition or annealing temperature which can lead to cracking [16,23]. Despite this, sputtered Bi:YIG films with magnetization (M_s) of 135 emu cm⁻³ and coercivity (H_c) of 80 Oe on Si substrates [17], Ce:YIG with M_s = 120 emu cm⁻³ and H_c = 15 Oe on silica and Si [22] and YIG films with M_s = 100 emu cm⁻³ and H_c = 35 Oe on MgO and quartz [18] have been grown.

We have previously presented a three-step fabrication process consisting of growth of a YIG seedlayer, a rapid thermal anneal, then growth of Ce:YIG or Bi:YIG by pulsed laser deposition [4,8,29] to give well-crystallized Ce:YIG or Bi:YIG films on silica, silicon on insulator (SOI) or Si. The polycrystalline YIG seed layer acts as as a virtual substrate, templating the growth of the Ce:YIG. In [8], we investigated the effect of anneal temperature and film thickness on the structure, magnetic and magneto-optical properties of YIG and Ce:YIG/YIG films. That study focused on reducing the thermal budget as far as possible while retaining a useful fraction of the bulk magnetization and magneto-optical properties of YIG and Ce:YIG. The results indicated the need for a sufficiently thick and good-quality YIG template for subsequent growth of Ce:YIG. Most samples had M_s values < 140 emu cm⁻³ and Θ_F(max) ~1100 ° cm⁻¹.

The present article demonstrates the integration of YIG and Ce:YIG on silicon nitride, including the effect of growth temperature and deposition rate of the YIG layer, to achieve bulk or near bulk magnetization and Faraday rotation at 1550 nm wavelength. Silicon nitride is commonly used in Si CMOS (complementary metal oxide semiconductor) and photonic circuits and the growth of magneto-optical Ce:YIG on nitride is an important contribution to fully integrated photonic circuits. Silicon nitride is less stable than silicon oxide or Si, because the silicon nitride can oxidize (i.e. become SiON_x) during the high-temperature YIG deposition under an oxygen atmosphere. This can affect the YIG stoichiometry requiring a separate process development for growing YIG on silicon nitride films. This study shows how the quality of YIG template can be optimized for the silicon nitride layer by changing the growth temperature and deposition rate. The structural and magnetic quality of YIG has significant impact on the quality of Ce:YIG grown subsequently. The process presented here will be useful in making structures for the investigation of nonreciprocal photonic phenomena and applications on silicon [30] including optical isolators, circulators, modulators, displays, and sensors [31].

2. Ce:YIG growth on YIG virtual substrate

Samples consisted of (001) Si with nitride of 760 nm thickness deposited by plasma enhanced chemical vapor deposition. Before YIG and Ce:YIG deposition, each sample was cleaned using sonication in acetone, isopropanol and trichloroethylene solvents for 20 minutes each. Pulsed laser deposition [28] was carried out using a 248 nm wavelength excimer laser and oxide targets as described previously [8]. During PLD, the base and oxygen pressure used for YIG growth were 5 x 10⁻⁶ Torr and 5 mTorr, respectively. YIG films of 40 ± 5 nm thicknesses were first deposited at a slow laser pulse rate of 2 Hz at 500, 560 and 640 °C substrate temperature for the first sample batch and at 10 Hz pulse rate at 500, 560 and 640 °C for the second sample batch. The YIG deposition rates were 5 Å min⁻¹ and 25 Å min⁻¹ for 2 Hz and 10 Hz respectively. Next, the YIG films were rapid thermal annealed (RTA) at 850 °C for 300 s. After RTA, Ce:YIG films were grown on top of the annealed YIG at 640 °C substrate temperature, 160 ± 10 nm thickness, at 10 Hz. Each deposition and recrystallization step was carried out in an oxygen ambient. Control of oxygen pressure and substrate temperature can preserve the stoichiometry and prevent the formation of secondary phases such as ceria [8,23,32]. After Ce:YIG deposition, no further RTA step was necessary, because Ce:YIG grew as a large grained polycrystalline phase on the lattice-matched polycrystalline YIG layer.

3. Structural, magnetic and magneto-optical characterization of Ce:YIG

A properly crystallized and single phase YIG template is desired for forming good quality Ce:YIG on top of YIG layer. In this section, we investigate the effect of growth temperature and growth rate on the phase quality and purity of the garnet films on silicon nitride. The structural quality and phases present in Ce:YIG 160 nm/YIG 40 nm films were characterized using x-ray diffraction (XRD), transmission electron microscopy (TEM) and atomic force microscopy (AFM). Ce:YIG and YIG lattice parameters are very similar (lattice parameters for Ce:YIG and YIG are about 12.41 Å and 12.38 Å, respectively) [33,34], so within the resolution of the x-ray diffractometer (PANalytical X’Pert PRO MPD in ω-2θ mode with Cu K_α radiation source at a wavelength of 0.1541 nm), the Ce:YIG and YIG phases were not distinguishable.

The ω-2θ plots in Fig. 1(a) indicate the Ce:YIG/YIG samples in which the YIG was deposited at 500, 560 and 640°C substrate temperature and pulse rate of 2 Hz. In the measurement of the ω-2θ scan, the samples were tilted by 1° to avoid strong peaks from the Si substrate. In addition, a zero-background holder was used to avoid spurious signals. Si₃N₄ peaks are absent since the nitride layer is amorphous. Peaks for YIG (211), (400), (420), (422) and (611) planes were present with no other detectable phases. Peak locations for Fe₂O₃ are also indicated to demonstrate that the Ce:YIG layers did not contain detectable iron oxide phases. The garnet peaks became stronger and narrower with higher deposition temperature. This transition indicates that the crystal size and quality improve with higher temperature deposition.

 

Fig. 1 XRD patterns of Ce:YIG (160 nm)/YIG (40 nm)/Si₃N₄/Si films grown at substrate temperatures of 500°C, 560°C, and 640°C when YIG was deposited at (a) 2 Hz pulse rate and (b) 10 Hz pulse rate. Dashed lines indicate the 2θ positions of the corresponding peaks of YIG. The peaks at 2θ = 38° and 2θ = 44° correspond to Y₂O₃ (024) and Y₂O₃ (015), respectively.

Download Full Size | PDF

Figure 1(b) shows the ω-2θ scan for the Ce:YIG films grown on YIG which was grown at 500, 560 and 640°C substrate temperature and pulse rate of 10 Hz. This higher deposition rate of YIG led to dramatically different growth. At 10 Hz, the YIG did not crystallize at any of the temperatures investigated, and iron oxide and yttrium oxide phases appear in each Ce:YIG film. The peaks are identified on both Figs. 1(a) and 1(b), as well as on Fig. 1 caption. Based on this observation, a slow growth rate (2 Hz pulse rate) is necessary for obtaining sufficient surface diffusion to crystallize the YIG needed to template the Ce:YIG.

Figures 2(a)–2(d) and 2(e) show the TEM images and AFM surface profile of the Ce:YIG (160 nm)/YIG (40nm, 2 Hz)/Si₃N₄ layers, respectively. Figure 2(b) confirms that the silicon nitride is amorphous, and YIG is crystalline with the expected lattice parameter. The nitride-YIG interface, marked 1, is diffuse whereas the YIG-Ce:YIG interface is sharp. The Ce:YIG shown in Fig. 2(c) was not crystallized, but the magnetic data (described below) is characteristic of crystalline Ce:YIG and we therefore assume that the focussed ion beam preparation of the TEM sample led to amorphization. Figure 2(d) shows the Ce:YIG-Au interface (the Au is added during sample preparation).

 

Fig. 2 TEM and AFM scan of the surface topography of Ce:YIG (160 nm)/YIG (40 nm)/Si₃N₄/Si films in which the YIG was grown at a substrate temperature of 640°C and at 2 Hz pulse rate. Grain sizes are around 8 μm in diameter, and rms roughness is 2.7 nm. (a) TEM image of Ce:YIG/YIG/Si₃N₄ layers (scale bar: 50 nm). high resolution TEM image of the interfaces between (b) silicon nitride and YIG, labelled 1, (c) YIG and Ce:YIG, labelled 2, (d) Ce:YIG and Au, labelled 3. Scale bars for (b) – (d): 5 nm, (e) AFM surface profile over 10 µm x 10 µm area.

Download Full Size | PDF

Figure 2(e) is a representative surface profile of a 160 nm Ce:YIG layer on YIG (40 nm, 2 Hz) grown at 625°C. The image is notable for the large ~8 μm diameter grains with radial patterns and occasional 1-2 µm wide voids assumed to represent the early stages of dewetting, or from resputtering of the film. Based on their depth (up to 15 nm), these voids are present only at the top of the Ce:YIG layer. This suggests that the Ce:YIG growth conditions favor significant surface diffusion, and a lower growth rate or temperature for this layer may improve the surface morphology (all Ce:YIG was grown at 640°C and 10 Hz).

The grain size of the Ce:YIG films increased with the growth temperature of the YIG, and the root-mean-square roughness of the Ce:YIG was 9.6 nm, 3.6 nm and 2.7 nm, for YIG 2 Hz grown at 500, 560 and 625°C respectively. Thus in these experiments the YIG grown at lower deposition rate and higher temperature produced Ce:YIG with the best crystallinity, but the Ce:YIG deposition conditions need to be further improved to reduce voiding and roughness.

Figure 3 shows the elemental mapping of Fe, O, Ce, Y, N and Au from the scanning transmission electron microscope sample together with a bright field TEM image. Ce is detected inside the Ce:YIG layer, and Y and Fe ions in the YIG and Ce:YIG layers. Comparing the depth of the Y-rich and Fe-rich regions with that of the O-rich region, it appears that oxygen is present in the upper ~10 – 15 nm of the nitride substrate. This is consistent with the diffuse interface observed in Fig. 2(b) and suggests partial oxidation of the nitride during the YIG growth.

 

Fig. 3 STEM-EDX elemental mappings for Fe, O, Ce, Y, N, Au ions. The scale bar (shown below TEM image) is 600 nm. Cerium has not diffused into YIG or nitride layers, but there is oxygen interdiffusion in the top surface of the nitride.

Download Full Size | PDF

Magnetic properties of Ce:YIG films were measured with magnetic fields applied perpendicular to the film plane (out-of-plane, OP, configuration, along Si [001] direction) or in the film plane (IP, along Si [010] direction) at room temperature (RT) using a vibrating sample magnetometer (VSM), and measured at 50 K and 200 K using a superconducting quantum interference device (SQUID) magnetometer. IP magnetization hysteresis loops for Ce:YIG/YIG are shown in Fig. 4(a). Figure 4(b) shows a narrower magnetic field range to more clearly indicate the remanent magnetization and coercivity of the samples. For Ce:YIG 160 nm/YIG 40 nm in which the YIG was deposited at 2 Hz, the M_s was 125, 134 and 142 emu cm⁻³ (with ~5% error estimate) and coercivity H_c was 76, 81, and 76 Oe for YIG films grown at 500, 560 and 625°C, respectively. For YIG deposited at 10 Hz, the M_s was 13, 19, and 25 emu cm⁻³ and H_c was 33, 53, and 119 Oe for YIG films grown at 500, 560 and 625°C, respectively. The M_s presented is the average over both garnet layers, and these M_s values were calculated by first subtracting the linear diamagnetic substrate contribution and then dividing the signal by the film volume (0.85 cm x 0.85 cm x 200 nm). The RT bulk saturation moment of YIG is about 140 emu cm⁻³ [34] and for Ce:YIG it is similar, 120 emu cm⁻³ [35]. The coercivity of single layer YIG is typically a few Oe, so the measured coercivities of up to 80 Oe are attributed to the magnetically harder Ce:YIG layers. We therefore see that the YIG growth rate profoundly affects the magnetic properties as well as the microstructure of the Ce:YIG. These results differ from those of YIG/Ce:YIG grown on Si in [8], where 10 Hz grown YIG was able to provide a seed layer for Ce:YIG with high magnetization.

 

Fig. 4 Room temperature hysteresis loops of Ce:YIG/YIG/Si₃N₄/Si with magnetic field applied (a,b) in plane, (c,d) perpendicular to the film plane. Growth temperatures and PLD pulse repetition rates for the YIG layer for each film are shown in the legend. All Ce:YIG layers had the same deposition conditions, 640°C and 10 Hz.

Download Full Size | PDF

Figures 4(c) and 4(d) show the corresponding OP magnetic hysteresis loops. The IP and OP measurements of M_s for each sample were within 5% of each other. The OP coercivities were 45, 37, and 100 Oe for 2 Hz deposited YIG and 23, 23, and 120 Oe for 10 Hz deposited YIG grown at 500, 560 and 625°C, respectively. All of the films had their easy axes in-plane, regardless of growth temperature and YIG deposition rate. Out of plane saturation fields of around 2 kOe for the samples with 2 Hz grown YIG are consistent with a shape anisotropy of 4πM_s indicating shape anisotropy is the major contributor to the film anisotropy.

The temperature dependence of magnetization of a Ce:YIG film on YIG (2 Hz, 500°C) was measured using a SQUID magnetometer, Fig. 5. Figure 5(a) shows the temperature dependence of M_s for this film (after substrate subtraction) when cooled at 10 kOe magnetic field applied in the film plane. The M_s was 125 emu cm⁻³ at RT and increased to 135 and 165 emu cm⁻³ at 200 K and 50 K, respectively. The substrate contribution to the magnetization was found by fitting a linear diamagnetic signal to the hysteresis loop at 200 K and evaluating its magnitude at 10 kOe. The same linear diamagnetic contribution was subtracted for all temperatures.

 

Fig. 5 Temperature dependence of magnetic properties of Ce:YIG film grown on YIG grown at 500°C and at 2 Hz. (a) Temperature dependence of saturation magnetization when the film is cooled under 10 kOe in plane bias field. (b) In plane magnetic hysteresis loops of the same film at 50, 200 and 292K. Inset shows the remanent magnetization and coercivities at 50, 200 and 292K.

Download Full Size | PDF

Figure 5(b) shows the IP magnetic hysteresis loops at 50, 200 and 292 K, with the diamagnetic Si signal subtracted. The coercivity was 76 Oe at RT, and increased to 107 Oe and 246 Oe at 200 K and 50 K respectively. The films are ferrimagnetic without any phase transitions at low temperatures. Previous measurements of temperature dependence of magnetism in YIG [36] gave M_s at 50K, 200K and 292K of 189.4 emu cm⁻³, 166 emu cm⁻³ and 140 emu cm⁻³, respectively, which are close to the experimental values measured here for Ce:YIG on YIG (2 Hz) films.

In comparison to these results for YIG/Ce:YIG, in single layer Ce:YIG films grown directly on nitride, the Ce:YIG did not crystallize and M_s never exceeded 3 emu cm⁻³. A lattice-matched YIG virtual substrate for Ce:YIG is therefore necessary to achieve the garnet structure with bulk saturation magnetization. Given the difference in structural and magnetic properties of Ce:YIG on YIG grown at 2 Hz and 10 Hz, growing YIG at low growth rates is also important.

Magneto-optical and optical properties of the films were measured using a custom-built Faraday setup and a spectrophotometer in transmission mode. Faraday rotation of the films was measured for near-infrared light (λ₀ = 1550 nm) passing perpendicularly through the film, while sweeping the magnetic field perpendicular to the film plane from −10 kOe to 10 kOe in a full loop. Figure 6(a) shows the total Faraday rotation (FR) loop of Ce:YIG (160 nm)/YIG (40 nm) with 2 Hz rate and 625°C growth temperature for the YIG. The total Θ_F was −2100 ° cm⁻¹. We extract the saturation Θ_F (Ce:YIG) by subtracting the contribution of YIG based on the thicknesses of the layers, i.e. Θ_F (total) = {160 × Θ_F (Ce:YIG) + 40 × Θ_F (YIG)}/200. Since Θ_F (YIG) in our samples is about + 100 °cm⁻¹ [29], Θ_F (Ce:YIG) = −2650 ± 150 °cm⁻¹. The Θ_F (Ce:YIG) obtained in films on silicon nitride was 20% lower than a value of Θ_F reported for bulk Ce:YIG, which was around −3300°cm⁻¹ [37]. The Θ_F for Ce:YIG/YIG/nitride obtained is consistent with the Θ_F obtained for Ce:YIG on silica substrates [22]. The noise of ± 150 °/cm is due to mechanical vibrations of the Faraday rotation measurement setup and backside scattering from the sample because the back side of the substrate was not polished.

 

Fig. 6 (a) Faraday rotation of Ce:YIG (160 nm)/YIG (40 nm)/Si₃N₄/Si films grown at 640°C show around −2100 °cm⁻¹ total Faraday rotation. Since YIG has opposite sign of Faraday rotation ( + 100 ° cm⁻¹, measured but not shown here), the Ce:YIG layer has about −2650 ± 150 ° cm⁻¹ Faraday rotation. The solid line shows smoothed data. All Faraday rotation loops were measured at 1550 nm wavelength. Backsides of Si substrates were not polished, which reduced the overall transmission of the samples. (b) Near-infrared transmission spectra of the Ce:YIG/YIG films grown at different YIG substrate temperatures. The fringes in the transmission spectra occur due to Fresnel reflections between the top and bottom surfaces of the films.

Download Full Size | PDF

Faraday rotation values for Ce:YIG/YIG (2 Hz) are −1400 ± 150°cm⁻¹, −1800 ± 25°cm⁻¹, and −2100 ± 150°cm⁻¹ for YIG grown at 500°C, 560°C and 640°C respectively. Corresponding Faraday rotation values for Ce:YIG/YIG (10 Hz) grown at 500, 560 and 640°C are −340 ± 100°cm⁻¹, −650 ± 200 °cm⁻¹, and −980 ± 250 °cm⁻¹, respectively. The Faraday rotation increased with saturation magnetization.

Optical transmission measurements, as shown in Fig. 6(b), indicate that total optical propagation loss was similar for 760 nm thick Si₃N₄ coated silicon wafers with and without Ce:YIG/YIG (except for the shift in the thickness fringe positions). The optical loss was therefore dominated by the silicon nitride, not the garnet films, and the loss of the garnet bilayer could not be determined by this method.

7. Conclusion

In this study, we demonstrated a growth method for integrating onto silicon nitride a transparent magnetooptical oxide, Ce:YIG, with bulk or near-bulk M_s and Faraday rotation. Growth on silicon nitride differed from that on SiO₂ shown in [8], and the same deposition parameters as used in [8] did not yield pure YIG phases. The difference is attributed to the greater reactivity of the nitride surface which oxidized during the growth process, affecting the stoichiometry or microstructure of the growing YIG. By using a low deposition rate for the YIG followed by its recrystallization using RTA, a well crystallized 40 nm thick YIG layer was made which served as a seed layer to grow crystalline Ce:YIG. The magnetic and magnetooptical properties of the Ce:YIG, specifically the saturation magnetization of 140 emu cm⁻³ and the Faraday rotation of up to −2650 °cm⁻¹, were equal to or close to reported bulk values, and the coercivity of the films was on the order of 80 Oe.

The integration of MO oxides on silicon-based substrates opens up avenues for non-reciprocal photonic devices [38,39], for example fabrication of asymmetric photonic band structures or hybrid CMOS-integrated photonic devices. Magnetic garnet films are also attractive for spin-wave filters, spin transistors and other spintronic devices based on spin-wave (magnon) transport.