Compact Brillouin devices through hybrid integration on silicon

1. INTRODUCTION

Stimulated Brillouin Scattering (SBS) has recently emerged as a flexible tool for optical processing and radiofrequency (RF) photonics [1]. SBS is one of the strongest nonlinearities known to optics, hundreds of times larger than Raman scattering in SMF-28 fiber [2], and is capable of providing exponential gain over narrow bandwidths of the order of tens of megahertz. This narrowband amplitude response is accompanied by a strong dispersive response, capable of tailoring the phase or group delay of a counterpropagating optical signal. In light of these effects, a rich body of applications have been explored such as slow light [3], stored light [4], narrowband RF photonic filters [5–7], dynamic optical gratings [8,9], narrowband spectrometers [10], optical amplifiers [11,12] and RF sources [13] among others. When pumped in a resonator configuration, a narrow linewidth, spectrally pure SBS laser can be generated [14–16]. Highly coherent lasers are used in optical communication and LIDAR, and in the production of pure microwave sources [17] among other applications. While the majority of previous works have traditionally utilized SBS in optical fiber, a number of these applications have been demonstrated in integrated form factors [1,18–21]. Most recently, the demonstration of 52 dB Brillouin gain [22] in centimeter-scale ${\mathrm{As}}_2{\mathrm{S}}_3$ rib waveguides proves that performance equivalent to kilometers of optical fiber is achievable in integrated devices.

The capability to embed SBS as a functional component in active photonic circuits will enable the creation of a new class of optoelectronic devices, in particular for integrated microwave photonics [23]. The desire to harness SBS optical processing in CMOS-compatible platforms has recently culminated in demonstrations of SBS in various silicon on insulator (SOI) device architectures [24–27]. Underetching of different waveguide geometries is performed to create guided acoustic modes, generating strong SBS from the high optoacoustic overlap. Initial works have demonstrated large Brillouin gain coefficients [24–26] in excess of ${10}^3{\mathrm{m}}^{-1}{\mathrm{W}}^{-1}$, ${10}^4$ times higher than single-mode optical fiber, made possible due to boundary forces which exist in these subwavelength structures [28]. More recent work has focused on reducing propagation losses to improve amplification factors to more than 5 dB [27]. But in general, higher gains in SOI devices have been prevented due to nonlinear losses in silicon [29,30] and linewidth broadening due to small-dimension fluctuations introduced during device fabrication [31].

In this work, we introduce a hybrid integration approach to generate large Brillouin gain in a silicon-based device, free from nonlinear losses. We embed a compact 5.8 cm ${\mathrm{As}}_2{\mathrm{S}}_3$ spiral waveguide into a silicon circuit, enabling record Brillouin gain of 22.5 dB (18.5 dB net gain) on a silicon-based chip. Traditional silicon grating couplers are used for coupling in and out of the chip, with silicon tapers providing low-loss transitions between the Si and ${\mathrm{As}}_2{\mathrm{S}}_3$ sections of the circuit. To further explore the flexibility of this approach, we fabricate precisely designed ${\mathrm{As}}_2{\mathrm{S}}_3$ ring resonators, enabling the first demonstration, to our knowledge, of Brillouin lasing in a planar integrated circuit. This work marks a significant step towards the realization of fully integrated active SBS devices, such as integrated optoelectronic oscillators [32], lossless microwave photonic filters [33], and compact optical gyroscopes [34], in the near future.

2. SILICON-INTERFACED ${\mathsf{As}}_{\mathsf{2}}{\mathsf{S}}_{\mathsf{3}}$ SPIRAL WAVEGUIDE

Figure 1(a) shows the schematic of the fabricated hybrid circuit. The circuit consists of a base silicon section and an ${\mathrm{As}}_2{\mathrm{S}}_3$ SBS active section. Silicon grating couplers are used for chip coupling [35], followed by a 2 mm long silicon waveguide. The nanowire waveguide ($450\times 220\mathrm{nm}$ cross section) then linearly tapers, over a length of 100 μm, to a width of 150 nm and ends in a open silicon region of $0.1\mathrm{mm}\times 4\mathrm{mm}$. Amorphous ${\mathrm{As}}_2{\mathrm{S}}_3$ was deposited in the open region with a thickness of 680 nm, completely covering the silicon tapers. Overlay waveguides were processed over the silicon taper before proceeding to the rest of the circuit; a scanning electron microscopy (SEM) of the end of the taper region, before cladding deposition, is shown in Fig. 1(b). Optical propagation simulations, discussed in Supplement 1, indicate a total insertion loss on the order of 0.1 dB for transmission into the fundamental mode of the ${\mathrm{As}}_2{\mathrm{S}}_3$ waveguide. Silica cladding of 1 μm thickness was sputtered over the sample after ${\mathrm{As}}_2{\mathrm{S}}_3$ etching, with care taken to keep processing temperatures suitable for optimum losses [36]. A schematic of a typical waveguide geometry is shown in Fig. 1(c), along with a cross-sectional SEM [Fig. 1(d)] and optical mode simulation of the fundamental mode of a 1.9 μm wide waveguide [Fig. 1(f)]. Further details of the fabrication process are provided in Supplement 1. The ${\mathrm{As}}_2{\mathrm{S}}_3$ region of the circuit is confined to within a small region of $0.4{\mathrm{mm}}^2$, requiring significant design optimization to achieve high performance.

 

Fig. 1. ${\mathrm{As}}_2{\mathrm{S}}_3$ silicon hybrid circuit. (a) Schematic of the hybrid circuit with a number of components indicated: 1. Silicon grating couplers with tapers to $450\mathrm{nm}\times 220\mathrm{nm}$ nanowires; 2. Silicon nanowire taper region with ${\mathrm{As}}_2{\mathrm{S}}_3$ overlay waveguide; 3. ${\mathrm{As}}_2{\mathrm{S}}_3$ waveguide lead into the hybrid structure; 4. Spiral waveguide formed out of ${\mathrm{As}}_2{\mathrm{S}}_3$; 5. Alignment markers formed in the silicon layer for patterning the ${\mathrm{As}}_2{\mathrm{S}}_3$ structures; 6. Reference silicon structures existing on the same chip. (b) SEM image of the end of the silicon taper before cladding deposition. (c) Schematic cross section of waveguide in chalcogenide-only region. (d) SEM image of chalcogenide region cross section with silica cladding. (e) Calculated effective indices for ten waveguide modes with increasing waveguide width. Waveguide widths used throughout the work, 1.9 μm in the spiral, 2.6 μm in the resonator, and 0.85 μm in the coupler, are indicated with dashed vertical lines. (f) Optical mode simulation of fundamental TE mode of 1.9 μm wide ${\mathrm{As}}_2{\mathrm{S}}_3$ waveguide.

Download Full Size | PPT Slide | PDF

To maximize the physical waveguide length in the available device area ($0.1\mathrm{mm}\times 4\mathrm{mm}$), we employ a folded spiral design with a rectangular shape and identical bends for each loop. As we will explain in further detail, the device geometry was chosen to give the highest Brillouin amplification in this confined area. The expected gain of a weak probe, $P_{\mathrm{o}}$, for a coupled pump power, $P_P$, in the small signal gain regime of backwards SBS is given by

To explore these tradeoffs more quantitatively, propagation losses for a number of waveguide widths were measured, and simulations calculating $G_{\mathrm{SBS}}$ were performed for corresponding geometries. The effective lengths assuming a 6 cm long device and the peak gains are shown in Fig. 2(a). The resulting $L_{\mathrm{eff}}\times G_{\mathrm{SBS}}$ is plotted in Fig. 2(b); while widths greater than 2 μm provide further improvement, the required bend radii prevents use in the compact spiral. From this comparison, we determined an optimum waveguide width of 1.9 μm, with effective bend radii of 16.5 μm calculated through FDTD simulations; further data is provided in Supplement 1. A total device length of 5.8 cm consisted of eight loops (36 bends including external connections), with a very compact structure achieved through a small waveguide spacing of 1.4 μm. We measured a total propagation loss of 4 dB through the spiral when correcting for the coupling losses from the grating couplers and $\mathrm{Si}-{\mathrm{As}}_2{\mathrm{S}}_3$ transitions. An estimated propagation loss of 0.7 dB/cm resulted in an $L_{\mathrm{eff}}$ of 3.9 cm for the nonlinear interaction. The overall form factor of this spiral represents orders-of-magnitude reduction compared to that of previous ${\mathrm{As}}_2{\mathrm{S}}_3$ waveguides used for SBS [22]. The typical half-etch rib geometries with multi-micrometer widths, used for the low losses $<0.5\mathrm{dB}{\mathrm{cm}}^{-1}$, require bend radii of more than 100 μm and are incapable of high density due to the significant crosstalk introduced from the partial waveguide etch. Similarly, underetched devices require an appropriate spacing between adjacent waveguides to prevent acoustic interactions and maintain structural support; $\sim 20\mathrm{\mu m}$ width was used for a single underetched membrane structure [27].

 

Fig. 2. (a) Peak SBS gain coefficient and effective lengths for varying waveguide width. (b) Corresponding $G_{\mathrm{SBS}}\times L_{\mathrm{eff}}$ values.

Download Full Size | PPT Slide | PDF

3. BACKWARDS SBS IN ${\mathsf{As}}_{\mathsf{2}}{\mathsf{S}}_{\mathsf{3}}$ SPIRAL WAVEGUIDE

To experimentally investigate the behavior of different devices, we performed two sets of pump-probe SBS measurements, a coarse measurement using an optical spectrum analyzer (OSA), and a high-resolution setup with an electrical vector network analyzer (VNA). In the OSA measurement, a high-resolution OSA (0.8 pm) was used to measure the transmission of a weak probe while an amplified pump laser was counterpropagated through the sample. A schematic of the setup is provided in Supplement 1. This measurement allowed for a rough estimate of the Brillouin frequency shift in the device and enabled simultaneous monitoring of the gain and loss response. An on–off gain of $>10\mathrm{dB}$ was observed at 80 mW coupled power, as shown in Fig. 3(a). A Brillouin shift of $\sim 7.6\mathrm{GHz}$ was measured relative to the residual back-reflected pump (centered at 1551.18 nm), and symmetric gain and loss spectra were measured.

 

Fig. 3. Backwards SBS in ${\mathrm{As}}_2{\mathrm{S}}_3$ spiral waveguide. (a) Optical spectrum measurement of SBS gain and loss. (b) Setup schematic for high-resolution pump probe. (c) High-resolution SBS spectrum for various pump powers. (d) Peak gain values up to 180 mW coupled pump power with fit. (e) Nonlinear loss comparison of this work, silicon nanowire, and silicon membrane.

Download Full Size | PPT Slide | PDF

To measure the SBS response in further detail, we implemented a high-resolution ($<1\mathrm{MHz}$) pump-probe experiment through the use of a radiofrequency VNA [40]. A laser of frequency ${\omega }_{\mathrm{c}}$ was split into two arms to create the pump and the probe wave. The pump was upshifted in frequency by ${\omega }_{\mathrm{o}}$ from the carrier through the use of a Mach–Zehnder intensity modulator and optical bandpass filter. The pump was then amplified with a high-power erbium-doped fiber amplifier, passed through ports $1\to 2$ of an optical circulator, and was coupled with TE polarization silicon-grating couplers into the hybrid circuit. In the probe arm, the laser underwent single-sideband with carrier modulation to produce a weak probe upshifted by frequency ${\omega }_{\mathrm{RF}}$. The carrier and probe were both coupled to the device and passed through the hybrid circuit. After coupling at the output, the transmitted waves were routed from ports $2\to 3$ of an optical circulator, and then a bandpass filter was used to remove any residual back-reflected pump. The remaining optical waves beat on a high-speed photodetector, and the change in RF power at frequency ${\omega }_{\mathrm{RF}}$ was measured by the VNA. In the frequency region of the Brillouin shift from the pump $({\omega }_{\mathrm{RF}}\approx {\omega }_{\mathrm{o}}-{\mathrm{\Omega }}_{\mathrm{SBS}})$, the modulated sideband will experience Brillouin amplification, as shown in Fig. 3(c). Further details of the measurement process can be found in Supplement 1. We measured the frequency spectrum for increasing pump powers up to 180 mW coupled power, as shown in Fig. 3(d). Net amplification was achieved above 25 mW on-chip power, overcoming the 4 dB of propagation losses, with a maximum on–off gain of 22.5 dB and a net gain of 18.5 dB. This represents a $>20\times$ improvement of net gain compared to recent demonstrations for forward [27] and intermodal SBS [41] in suspended silicon membrane waveguides. Fitting the slope of the measured peak gain data from Fig. 3(d) results in a Brillouin gain coefficient of $G_{\mathrm{SBS}}=750\pm 50{\mathrm{m}}^{-1}{\mathrm{W}}^{-1}$, a 50% increase over previous chalcogenide waveguides [22]. This increase is primarily due to the reduction of $A_{\mathrm{eff}}$ compared to the previous partially etched rib structures.

Here, we compare the effects of pump attenuation through nonlinear losses in the devices in this work with simulated silicon geometries [Fig. 3(e)]. Nonlinear losses are a key limiting factor in integrated silicon waveguide devices at telecommunications wavelengths [42,43]. For silicon-based Brillouin systems, the effect is two-fold: a direct reduction in pump power from two-photon absorption and free-carrier absorption, leading to a power-dependent $L_{\mathrm{eff}}$ for the pump and also the direct attenuation of the probe wave through cross-photon absorption and free carriers, which are generated by the pump. We experimentally measured the transmission through a 2 mm reference hybrid waveguide, maximizing the optical power through the silicon leads of 3 mm on either side of the hybrid structure. This is a worst-case scenario, with negligible linear losses in the hybrid region and 6 mm of silicon waveguide contributing to nonlinear losses. Even so, only 0.5 dB was measured at high coupled powers of 150 mW. This is in stark contrast to pure silicon structures, with close to 4 dB of pump attenuation expected for a simulated silicon membrane and almost 6 dB for a nanowire geometry, effectively saturating the input pump power and preventing any further gain [25]. Careful theoretical analysis [29,30] has indicated that reducing linear losses and increasing device lengths may enable higher Brillouin amplification at low pump powers, but this can be hampered by dimensional broadening as explored in the following paragraph.

We explore dimensional broadening in ${\mathrm{As}}_2{\mathrm{S}}_3$ waveguides by measuring the SBS response of a number of different waveguide widths and lengths, including straight and spiral structures. Dimensional broadening has been identified as a key issue which reduces the expected gain, particularly in nanoscale waveguides reliant on transverse acoustic waves such as forward SBS structures [26,31]. The effect manifests in changing Brillouin lineshapes; as device lengths are modified, measured mechanical quality factors are reduced by almost half when moving from millimeter to centimeter scales in suspended membrane structures [27]. In this work, we found that the natural linewidth did not vary with device length, and only minor fluctuations were measured for lengths of 1, 2, 4, and 40 mm; results are provided in Supplement 1. This indicates that, for these geometries, the linewidth is primarily governed by the deposited material properties of ${\mathrm{As}}_2{\mathrm{S}}_3$, allowing for scaling to large device lengths.

The significant increase in available gain and negligible effects of nonlinear losses will enable a number of new applications beyond the limited Brillouin signal processing currently demonstrated in silicon [21], including Brillouin lasing, as explored in the next section.

4. COMPACT RING RESONATOR

Brillouin lasers are capable of spectrally narrowing laser sources and, if cascaded, can produce pure microwave frequencies. Achieving Brillouin lasing in microresonators is challenging due to the requirement for the cavity free spectral range (FSR) to closely match the Brillouin shift. Initial demonstrations used highly overmoded resonators, such that two resonances between different mode families were aligned [44,45]. More recently, precise matching of the cavity FSR and the SBS shift was achieved in lithographically processed silica wedge resonators [46]. These previous devices have extremely low losses, enabling low-threshold oscillation, but require external coupling via tapered optical fibers or free-space optics. To show a further application of the combined ${\mathrm{As}}_2{\mathrm{S}}_3$ and Si platform, we fabricate high-$Q$ ring resonators designed for Brillouin lasing and achieve the first demonstration, to our knowledge, of Brillouin lasing in a planar integrated circuit.

A schematic of the ring design is shown in Fig. 4(a). To achieve low-threshold lasing in the sample, we must satisfy three competing challenges: the FSR must match the Brillouin shift, the loss throughout the cavity must be minimal, and the whole structure must be as compact as possible. The SBS shift scales (${\mathrm{\Omega }}_{\mathrm{SBS}}$) with effective index $(n_{\mathrm{eff}})$ of the optical mode and acoustic velocity of the acoustic mode $(v_{\mathrm{ac}})$, such that ${\mathrm{\Omega }}_{\mathrm{SBS}}=2n_{\mathrm{eff}}{v}_{\mathrm{ac}}/{\lambda }_p$. The FSR of a resonator depends upon the total roundtrip time of the cavity, and is related to the length $(L)$ and group index $(n_{\mathrm{g}})$ such that $\mathrm{FSR}=c/(L\times n_{\mathrm{g}})$. Thus, we need to take into account the change of $n_{\mathrm{eff}}$ and $n_{\mathrm{g}}$ with waveguide width when determining the appropriate length of the resonator, as represented in Fig. 4(b). The threshold for a Brillouin laser in a resonator with an FSR matching the Brillouin shift is given by [47]

 

Fig. 4. Brillouin lasing in planar ${\mathrm{As}}_2{\mathrm{S}}_3$ resonator. (a) Schematic of the hybrid ring resonator structure. (b) Concept figure for the lasing conditions. The cavity free spectral range needs to precisely match the Brillouin shift. (c) Typical optical transmission of ring resonator. (d) Setup used for measuring the laser and resonator. (e) Lasing signal measured on OSA. The Brillouin lasing signal is observed in blue solid trace. The tunable laser is shifted slightly, and the lasing no longer occurs. A number of peaks due to the modes of the laser are observable in the orange-dashed trace. (f) RF beat of the back-reflected pump and lasing signal. The measured linewidth was less than 5 MHz, significantly narrower than the natural lifetime of 40 MHz, confirming that we are above the lasing threshold. (g) Brillouin lasing while monitoring the resonance position. Both the pump and generated Stokes are aligned to cavity resonances.

Download Full Size | PPT Slide | PDF

Optical transmission measurements were performed on the fabricated device with the same high-resolution OSA used in the pump-probe measurements [Fig. 4(c)]. From these measurements, we observed a FSR of 7.62 GHz at 1553 nm and an extinction ratio of around 0.65 dB or, equivalently, a transmission of 85%, which corresponded to $K=0.04$. The measured resonance linewidth was 4.5 pm, corresponding to a $Q_{\mathrm{tot}}$ of $4\times {10}^5$. The measured $Q$ factor was limited by the propagation losses in the 2.6 μm waveguide, estimated as 0.5 dB/cm, and losses due to the ring coupler on the order of 0.2 dB. This value compares favorably to previous demonstrations of planar centimeter length scale ${\mathrm{As}}_2{\mathrm{S}}_3$ ring resonators with $3.5\times {10}^5$ in ${\mathrm{As}}_2{\mathrm{S}}_3$ on lithium niobate [49] and $1.5\times {10}^5$ for directly written ${\mathrm{As}}_2{\mathrm{S}}_3$ waveguides [50]. From Eq. (2), we determine an expected threshold of 80 mW for the fabricated sample, assuming optimum matching of the SBS shift to the cavity FSR.

5. BRILLOUIN LASING

To demonstrate Brillouin lasing in the fabricated resonator, we seamlessly tuned a pump onto the resonance for a range of power levels, while monitoring the back-reflected optical waves [Fig. 4(d)]. The pump light source was an external cavity laser capable of fine-resolution tuning with a continuous step size of 10 MHz, allowing for accurate alignment to the center of the resonance. The pump was amplified before being passed through a circulator (ports $1\to 2$) and coupled to the chip through silicon grating couplers. Back-scattered light from SBS and the back-reflected pump wave then passed through the circulator (ports $2\to 3$), and was monitored on a high-resolution OSA while the RF beat was measured on an electrical spectrum analyzer (ESA). A weak reference output of the OSA was also used as a probe to measure the transmission of the resonator when desired.

For coupled powers above 50 mW, Brillouin lasing was observed on the OSA and ESA. Figure 4(e) shows a Brillouin lasing signal on the OSA, along with a reference signal at the same power level with the pump shifted just past the resonance. A single lasing signal is observed at a power level in the range of $-30\mathrm{dBm}$. A strong signal close to 0 dBm, at the pump wavelength, was also measured due to the $\sim 1\%$ back-reflection of pump from the chip grating couplers. A number of cavity side modes from the back-reflected pump were also observed; these were around 50 dB below the pump signal, in line with the pump laser specifications. To confirm that the measured signal was not due to spontaneous scattering, we measured the electrical beatnote above threshold on the ESA [Fig. 4(f)]. The measured beatnote was significantly narrower than the SBS natural linewidth of 40 MHz [22], plotted with a dashed line in Fig. 4(f). The frequency of the beatnote was at $7.60\pm 0.005\mathrm{GHz}$, and slow drifts on the order of 5 MHz were observable on the ESA over minute time scales. The lack of active locking of the pump to the resonator prevented the measurement of a slope efficiency of the Brillouin laser above the threshold level. For the weak coupling case in which we have ($K=0.04$), a low slope efficiency of 4% is expected [47]. Improving the coupling to the ring would drastically improve this and also reduce the lasing threshold. Finally, to confirm that the Brillouin lasing is indeed occurring on the resonances of the ring, we performed an OSA measurement while sweeping a weak probe signal to measure the resonator transmission. In this case, the coupled pump power was 75 mW. Figure 4(g) shows that the lasing signal and pump are both aligned to cavity resonances.

6. DISCUSSION

To provide further details on how the ${\mathrm{As}}_2{\mathrm{S}}_3-\mathrm{Si}$ hybrid circuit results compared with previous demonstrations of SBS in integrated waveguides, we prepared a comparison summary in Table 1. Initial silicon devices focused on achieving the highest gain coefficients possible using highly sub-wavelength structures which harnessed radiation pressure [24–26]. Issues arising from high scattering losses, dimensional broadening, and nonlinear losses resulted in low net gain values of below 1 dB. More recent work has shifted to larger device geometries, resulting in interactions produced almost entirely through electrostriction. High sensitivity to local wafer conditions prevented the membrane structure from being folded, resulting in reduced compactness with straight waveguides up to 3 cm long. However, the reduced propagation losses enabled significantly higher net gain, up to 5 dB [27], than previous silicon demonstrations. In comparison it is clear that, being free from nonlinear losses, the ${\mathrm{As}}_2{\mathrm{S}}_3$ devices are capable of significantly higher on–off gain (greater than 50 dB) compared to full silicon devices. In this work, we address the compactness limitations of previous ${\mathrm{As}}_2{\mathrm{S}}_3$ demonstrations through the high-density and tight bends of fully etched structures while maintaining large net gain. Finally, to understand the relative efficiency of different devices, we introduce a simple figure of merit (FOM), $G_{\mathrm{SBS}}\times L_{\mathrm{eff}}$, which is from the exponential term in Eq. (1). To achieve 20 dB of gain, which is sufficient for many microwave photonics applications [1], with 50 mW coupled pump power requires an FOM $\sim 100$. None of the currently demonstrated devices have approached this regime, which is equivalent to half a km of SMF optical fiber, but further improvements to the $G_{\mathrm{SBS}}$ and $L_{\mathrm{eff}}$ are expected to accomplish this goal in the near future.

Table 1. Comparison of SBS Performance in Different Integrated Devices^a

View Table

One of the most desirable characteristics of Brillouin lasers is a significant linewidth narrowing of lasing Stokes lines. The key requirement for entering this regime is for the optical damping to be less than the acoustic damping or, in terms of linewidths, the cavity linewidth to be narrower than the natural linewidth of the acoustic mode [14]. If this regime is achieved, then the Stokes spectrum will narrow and the full width at half maximum will be given by

7. CONCLUSION

In this work, we have introduced a hybrid integration approach to generating large Brillouin gain in a silicon-based device. We embedded a compact 5.8 cm ${\mathrm{As}}_2{\mathrm{S}}_3$ spiral waveguide into a silicon circuit, enabling a record Brillouin gain of 22.5 dB (18.5 dB net gain) on a silicon-based chip. Fabrication of a compact ring resonator enabled the first demonstration of Brillouin lasing, to our knowledge, in a planar integrated circuit. Combining active photonic devices, such as modulators and detectors, with the work shown here will enable the creation of compact, high-performance devices with capabilities beyond traditional RF systems.