1. Introduction

Silicon photonics has seen a dramatic growth in the last decade, enabling a wide range of applications in telecommunication to sensors including spectrometers, light detection and ranging (LIDAR) and gyroscopes [1–4]. Such an intense rise in both research and commercialization of silicon photonics has been facilitated by the high refractive index contrast b/w Si, SiO₂ and air in combination with the large-scale availability of high-quality silicon-on-insulator (SOI) wafers and a mature CMOS processing ecosystem. The high contrast between silicon and SiO₂ has not only allowed the realization of heavily miniaturized low loss and uniform submicron scale waveguides as well as a diverse range of nanophotonic components such as modulators, optical filters [5], photodetectors [6], and arrayed waveguide gratings [7], but also enabled a platform for the design of photonic integrated circuits. The applicability of the mature silicon CMOS processing on the other hand, has not only further reduced the costs of fabrication and improved scalability but also led to the development of a common integrated platform combining both photonics and electronics on a chip leading to increased data transfer rates from 100 Gbps to Tbps while intensely lowering the power consumption requirements [8].

However, such a rapid drop in the size of the waveguides has led to an increasing modal size mismatch between nanoscale on-chip waveguides and standard optical fiber used in data communications today. As a result, one of the key challenges currently facing the telecom industry is efficiently coupling light between waveguides and external light sources such as fibers and free space optics in a repeatable manner, without additional fabrication, post-processing, and complex optical alignment. A diverse range of coupling platforms have been explored in the last decade, which can be broadly categorized into in- plane grating couplers and edge couplers. Edge couplers typically offer insertion losses of < 0.5 dB as well as polarization and wavelength diversity [9]. However, they are not suitable for high-volume manufacturing and wafer-level testing and demand a highly sensitive optical alignment step [10] . On the other hand, grating couplers provide the freedom to be placed anywhere on the chip and allow for wafer-scale automated testing and have relatively looser alignment tolerances.

A wide range of Grating Couplers (GC) have been realized in the last decade with an aim to significantly improve the coupling efficiency (CE) by enhancing directionality and increasing the modal overlap [11,12]. The improvement in directionality has been achieved by either introducing an overlay on the grating or incorporating a metallic reflection plane below the grating layer [13]. However, this improvement has come at the cost of a set of complex fabrication steps and the use of materials that are CMOS incompatible. Another technique that has been extensively used to achieve improved CE is tailoring the optical scattering by each grating element, which involves varying the etch depth and fill factor of each grating period [14]. Although this technique is CMOS compatible, the involvement of multiple alignment steps has led to a significant increase in fabrication complexity. Apodised GC is another design that has been widely adopted in which either the etch depth or the duty cycle of the gratings are varied gradually and linearly to achieve maximum coupling [15–17]. While the etch depth can be gradually varied using lag effects in ICP-RIE etching, the duty cycle variation can be realized in a single etch step. However, the design of apodised grating couplers is computationally intensive and time-consuming, which requires an exploration of the vast parameter space leading to the calculation of the apodization curve giving maximized CE [9].

While a vast majority of GC designs have been passive, exhibiting a fixed CE after fabrication, an active design is highly desired in which the CE or the band of frequencies in which the device displays a high CE can be dynamically tuned. For instance, the operational window in the packaged device often drifts slightly away from the designed configuration mainly due to fabrication errors, fiber misalignment, incident angle variation, and cladding material change during the manufacturing and assembly process. In such scenarios a reconfigurable GC having the capability to compensate for the shift in the spectrum is very advantageous. Similarly, a GC with tunable CE can be extremely useful as a low-footprint variable optical attenuator to achieve dynamic channel equalization of multichannel optical receivers replacing regular ridge waveguide or MZI-based attenuators having comparatively significantly larger footprints. Thus, a diverse range of active designs have been put forward in which the CE can be modulated by introducing a volatile material whose optical properties can be dynamically tuned with an external stimulus such as MEMS, thermo-optic and carrier modulation-based tunable GC’s. These have been mainly exploited using volatile tunable platform such as Si and its derivatives, whose optical properties can be reversibly altered in response to an external stimulus such as temperature change, electrical bias or thermo-mechanical deformation. As a result, tuneability is achieved at the expense of higher power consumption [18–21]. Thus, an active GC with non-volatile switching capability allows for the realization of multifunctional and adaptive photonic components with built-in memory functionality in silicon photonics that also unlocks post-fabrication trimming for photonic circuits. Furthermore, the non-volatile nature of the platform is highly desirable in enabling lower-power consumption as the device does not need power to maintain a particular state of operation. Non-volatile switching functionality can be found most prominently in alloys of sulfur, selenium, and tellurium known as chalcogenide semiconductors that present a wide range of stoichiometrically tunable material properties and are flexible for applications from the visible to the mid-infrared portions of the optical spectrum. Most importantly, chalcogenides exhibit non-volatile switching in the form of phase change and photo-ionic effects [22,23]. The former involves switching between an amorphous and crystalline state using optical, electrical, or thermal stimuli [23–26] and has been widely exploited in optical disk (Blurays and DVD’s) and phase change random access memory (PCRAM) technology for decades. The amorphous-to-crystalline transition in chalcogenides is an annealing process that can be initiated by increasing the ambient temperature or locally by laser- or electrical current- induced heating to a temperature above the alloy’s glass-transition point, T_g (∼160 °C for GST), but below its melting point, Tm (∼600 °C). The reverse transition —a melt-quenching process—may be driven by shorter, higher energy pulsed excitation that momentarily brings the material to a temperature above T_m.”

In this work, we propose a reconfigurable chalcogenide-on-insulator grating coupler using the phase change alloy Germanium antimony telluride (Ge₂Sb₂Te₅ or GST) with ultrahigh modulation contrast. This concept holds the potential to replace silicon-on-insulator (SOI) based grating couplers in conventional silicon photonic circuits, offering a wide range of reconfigurable functionalities. To achieve this, we employed an inverse electromagnetic design technique, considering ease of fabrication using conventional lithographic techniques like electron beam lithography (EBL). During the optimization process, a fabrication constraint of 100 nm was incorporated to ensure practical manufacturability. The resulting design features a circular grating pattern with linear apodization and a compact, optimized taper. In the amorphous phase of the chalcogenide material (GST), our optimized design enables efficient coupling of over 50% from a standard single-mode fiber with a core diameter of 10µm and cladding of 127µm to a silicon waveguide with dimensions of h = 220 nm and w = 450 nm.

When the grating coupler (GC) transitions to the crystalline phase, there is a significant reduction in coupling efficiency approaching zero allowing our design to reach an ultra-high modulation contrast of 70 dB. More importantly, this modulation contrast can be achieved by controlling the position and size of the crystalline spot on the amorphous GC, thus allowing the realization of miniaturized switching component to achieve nonvolatile, ultrafast, and reconfigurable coupling of light to and from silicon waveguides for efficient inter-chip communication. Similar modulation contrasts can be achieved across S, C and L telecommunication bands. We show that by simply changing the thickness of the film or the etch profiles different devices can be manufactured targeting a variety of wavelengths across the aforementioned telecom bands.

2. Design optimizations

In a typical optimization technique, the first goal is to define a figure of merit (FOM) of the design which best describes its efficiency, followed by its minimization or maximization by varying a set of geometrical parameters. It is finding this minima or maximum that can be time consuming with traditional genetic or particle swarm optimization techniques as they rely mainly on parameter sweeps or random perturbations to find the optimized device parameters for a minimized FOM. Thus, as a starting point gradient-based methods were used which can quickly find the optimal parameters with a relatively small number of FOM evaluations. However, parametrization of the geometry and a choice of the initial geometry also plays a major role in rapid convergence towards an optimized design [27]. To evaluate the gradient, an adjoint-based technique was employed using Lumerical’s LumOpt [28] package which exploits the properties of linear partial differential equations to evaluate the gradient using forward and adjoint simulations, independently of the number of optimization parameters.

2.1 Theoretical design and geometrical optimization

A conventional GC design mainly exploits the use of a periodic grating structure which is usually etched on the surface of the waveguide. The grating consists of a series of parallel grooves or ridges with a specific periodicity$\; \Lambda $. When light propagates along the waveguide, it encounters the grating, and a portion of the light is diffracted at specific angles determined by the grating period and the incident wavelength ${\lambda _c}$. The grating period is fixed and is calculated using the phase matching condition [9,11,14,16,29] as shown in [Eq. (1)].

The effective index of the grating, ${n_{eff}}$ in the previous equation is estimated using [Eq. (2)], which is a function of the fill factor of the gratings, ${F_G}$, effective index of the unetched teeth, also a high index region (${n_{tooth}})$ and etched trenches or low index region(${n_{trench}}$) and can be expressed as [14,16]:

The fill factor, ${F_G}$ is the ratio of the thickness of a tooth to the grating period, $\Lambda $.

Using a series of simulations using the FDTD method a set of 4 different design parameters namely ${n_{tooth}}$, ${n_{trench}}$, $\Lambda $, and ${F_G}$ are adjusted to optimize the CE. The efficiency of coupling in this basic grating coupler is limited by two factors. Firstly, the light emitted from the waveguide is coupled both upwards towards the fiber and downwards towards the substrate. Secondly, there is a mismatch between the field produced by a uniform grating and the mode of the fiber. Wide range of techniques have been explored to significantly enhance the CE of these GC designs such as the introduction of back reflectors embedded in the substrate [10,30–32], and the use of polysilicon overlayers [33,34,15] which involve extra fabrication steps. On the other hand, CE can also be enhanced without any extra fabrication steps by linear apodization of the fill fraction of the GC along the grating. Thus, to achieve an initial starting design with enhanced CE, a linear apodization [9,35] approach was applied to the amorphous GST based grating resulting in a spatial variation of the fill factor as shown in [Eq. (3)]:

The apodization function can be defined using [Eq. (3)] and [Eq. (4)] and is shown by equation [Eq. (5)].

Using the MODE FDE solver of Lumerical, the slab modes in the tooth and trench regions were calculated with an initial angle of incidence of 5° and the effective indices namely ${n_{tooth}} = 2.848847613$, and ${n_{trench}} = 2.480869063$ were extracted at an initial operating wavelength of ${\lambda _c} = 1580nm$. The calculation to determine these effective indices in the etched and the unetched regions of the GC design involved an amorphous GST slab of height 220 nm patterned on top of a 2 µm thick oxide layer on a silicon (Si) substrate having n_Si= 3.47668. The etch thickness was set to be at 80 nm and the refractive index of both amorphous and crystalline GST used here was measured using variable ellipsometry and shown in Fig. 1(a) for a broad range of wavelengths from 400 nm to 1700nm [36]. An initial value of R and $F_G^0$ were chosen to be 0.03µm^-1 and 0.95 to determine the pitch of the initial GC design. The initial fill factor $F_G^0$ in ideal situation would be 1 as one starts from a solid waveguide, but a value of 0.95 was chosen to avoid extremely narrow trenches. Similarly, the apodization factor depends on material, geometry and polarization of incident light. A series of different trial sweeps were performed to determine the best initial value of apodization factor. These values were determined through parameter sweeps of the design.

 

Fig. 1. Reconfigurable phase change chalcogenide-on-insulator grating couplers. (a) Shows the variation of refractive index (n) and (inset) extinction coefficient (κ) across telecommunication bands. (b) Schematic representation of the optimized GC design. (c). Transmission of a typical optimized GC for amorphous and crystalline phases of the GST (D = 100 nm, T = 220 nm). (Inset) Close-up transmission for grating coupler with GST in its crystalline phase (d). Spectral dispersion of modulation contrast for a typical optimized GC between amorphous and crystalline phases.

Download Full Size | PDF

Using this linearly apodised initial design of GC, a parameter sweep of a series of fiber positions is performed in which light in the form of TE polarized Gaussian beam was launched from a single mode fiber (SMF) having core diameter of 10um and an outer diameter of 127µm onto the GC at an initial angle of 5°, using Lumerical 2D FDTD. This determines the optimized start position of the GC, which was around 18µm, displaying a maximum transmission of around 39%. A background cladding of silica was assumed throughout the optimization process.

During the FDTD simulations, we fixed the timestep's stability factor at 0.99, which equates to a timestep of 0.046704fs. Furthermore, we adopted Lumerical's standard auto-shutoff threshold of 10⁻⁶ to ensure the residual energy in the simulation domain was negligible. For the simulations, the PML boundary condition was applied along the x and y directions for 2D and extended to the z direction for 3D models. We selected an auto-non uniform FDTD mesh type with a mesh accuracy of 3 and utilized a minimum mesh step of 0.00025um.

This apodised grating comprising of a spatially varying period of the grating was then used as a starting point for a 2D optimization step in which a shape function describing the grating pattern as a polygon was used to parametrize a figure of merit while enforcing a feature constraint of 100 nm. The optimization FOM (oFOM) was defined as the coupling efficiency (CE) for a specific shape function calculated using Lumerical's 2D FDTD solver which was optimized using a gradient-based optimization technique. The optical power coupled in the waveguide by the GC was determined by placing a frequency domain power monitor in the waveguide at around 1µm away from the GC. This is basically the transmission, which is also the CE of the GC design, as shown in [Eq. (6)], which relates oFOM to the CE (${{\boldsymbol \gamma }_{{\boldsymbol CE}}})$ which is ratio of power coupled to the waveguide to the input power.

The variation of the FOM with iterations showed a steady improvement in the FOM from 0.52 to 0.57 after which no further improvement could be seen. Similarly changes in the optimized parameter values become less intense after a few iterations since the FOM gets closer to the optimum value. The changes in FOM gradient (based on changes in the parameter values) also decreases on average as the iteration continues.

Thus, the optimization process is highly efficient, with close to 170 iterations an optimal shape can be achieved. The final 3D design, incorporating the initial parameters of grating start position and taper angle, underwent further refinement. Utilizing gradient-based optimization coupled with Lumerical’s 3D FDTD solver, we achieved the final optimized circular grating pattern, characterized by its ellipticity as depicted in Fig. 1(b). Apart from this the shape of the connector connecting the first tooth of the GC design to the waveguide was also optimized from a straight line into a polynomial shape which further helped to increase the CE. Thus, a total of 130 parameters were simultaneously optimized. The final optimized design could be exported to GDS file which can be directly used in subsequent lithographic steps (Figure S1).

2.2 Tuneability of the optimized GC design

The optimal distance determined between the fibre and the GC was 0.3µm. It’s ideal position was gauged to be 18µm from the commencement of the GC, and the taper angle was identified as 29.7°. Figure 1(b) shows the optimized geometry of the chalcogenide-on-insulator grating coupler design. When the GST is in amorphous phase, the grating coupler shows a high coupling efficiency of 57% at λ = 1580 nm, and in crystalline phase the CE drops significantly to near zero value as shown in Fig. 1(c). The calculated modulation contrast using 10log(A/C), where A and C are the CE corresponding to amorphous and crystalline GST respectively, shows an ultrahigh value of around 70 dB as shown in Fig. 1(d). The normalized electric field intensities at λ = 1580 nm are in the top and cross-sectional view of the GC in Fig. 2(a) and 2(c) which clearly show a strong coupling in the amorphous phase as compared to close to no observable coupling in the crystalline phase shown in Fig. 2(b) and 2(d).

 

Fig. 2. Spatial distribution of electric-field across both operational modes. Shows the normalized electric field distribution at the operational wavelength of λ = 1580 nm for an optimized GC design for (a,b) amorphous and (c, d) crystalline phases from birds-eye (a, c) and cross-section (b, d) perspective.

Download Full Size | PDF

This approach and design yield a highly tunable platform that can be configured to operate across the entire telecom band. As an example, we show that by varying the thickness of the GC in the optimized design, the operation window can be tuned across the E to L telecom bands. In Fig. 3(a), it can be observed that the variation in the thickness of the GC offers a higher tunability as compared to the variation in the etch depth shown in Fig. 3(c)-(d). When the thickness is varied between 150 nm to 250 nm, the operational peak shifts from λ = 1475 nm to λ = 1675 nm while maintaining a high modulation contrast (MC) and moderate CE as shown in Fig. 3(a)-(b), whereas when the etch depth is varied between 57 nm to 220 nm, the operation window showed a comparatively smaller tuneability from λ = 1475 nm to λ = 1625 nm with a considerable drop in the CE as shown in Fig. 3(c)-(d).

 

Fig. 3. Tunability of operational window across telecommunications bands. (a) Spectral dispersion of transmission for the amorphous phase GC for various thickness (t) of GST. (b). Variation of the modulation contrast for varying thicknesses. (c). Transmission of the amorphous phase GC for varying etch depths (D) of the GC. (d). Corresponding modulation contrast of the GC design for varying etch depths.

Download Full Size | PDF

3. Impact of crystallization spot size and position on the optical response

To achieve reconfigurability, such a component with an overall dimension of around 30 µm ${\times} $ 30 µm requires an integrated switching infrastructure consisting of an electrically driven nanoheater design to switch phases. However, this design consumes significant power to heat up the entire footprint of the component to achieve phase transitions, which adversely affects the overall coupling efficiency (CE) of the device. Consequently, it is important to identify the minimum required phase change volume to achieve a desired modulation contrast in such grating couplers.

3.1 Variation in the size of the crystallization area

Using the previously optimized design of the GC, the center position where the fiber was illuminated was chosen, and a crystallized spot of different dimensions was simulated to study the effect of crystallization volume/area on the CE. Figure 4(a) shows a schematic representation of the model depicting the growth of the crystallized spot, starting from fully amorphous GC all the way to fully crystalline GC depicted by Fig. 4(b). It can be observed in the transmission plot at various crystallization spot sizes in Fig. 4(c) that as the spot size grows, the transmission drops. This drop in transmission is accompanied by an exponential increase in the MC, as shown in Fig. 4(d). Notably, a small footprint crystalline spot can create a large modulation contrast, for example at 22% coverage, a 9.5 dB change in coupling efficiency is observed. Thus, this demonstrates that the entire GC does not need to be crystallized, and instead, the switching of a small precisely positioned spot is enough to provide the desired MC. The switching infrastructure needed to achieve a reversible phase transition for this smaller volume requires significantly reduced power and will exhibit enhanced switching speeds.

 

Fig. 4. Size of switching volume. (a, b) Schematic representation of the growth of crystalline spot from fiber illumination origin outwards. Plot of transmission for varying crystallization spot sizes. (d). Transmission and modulation contrast dependence on crystalline area at λ = 1580 nm.

Download Full Size | PDF

3.2 Position of the crystallized spot

The position of the crystalline spot plays a crucial role in realizing high modulation contrasts. To demonstrate this, various elliptical crystalline spots were selected. The semi-minor axis length was set to half that of the semi-major axis, while the radius of the semi-major axis ranged from 250 nm to 5 µm. These spots were placed at different positions starting from the origin of the amorphous GST-based GC. The origin is precisely defined as the point of convergence between the GC taper region and the waveguide. To faithfully replicate the characteristics of a real-world crystalline spot generated using a diode laser with an elliptical beam shape resulting from higher beam divergence along one axis compared to the other, an elliptical shape was deliberately chosen for the crystalline spot. Subsequently, the spot's orientation was adjusted so that its major axis was perpendicular to the waveguide, and its center coincided with the axis passing through the center of the waveguide. The spot was then systematically moved along in close proximity to the waveguide, measured 250 nm. However, with an increase in spot semi-major axis length(s), the closest possible position, p of the spot relative to the waveguide, allowing the spot to fit, necessitated a shift away from the waveguide (Fig. 5(a)). Consequently, the start position had to be adjusted away from the origin to accommodate the larger spot semi-major axis length. After this, the CE efficiency of the amorphous phase GC was calculated for each crystalline spot dimension at their respective positions.

 

Fig. 5. Position of switching volume (a). Shows a schematic of a series of optical simulations sweeping across the position, p and spot size, s of a crystalline spot on an optimized amorphous GC design. (b). Variation of transmission and MC with p for 5 different spot sizes with s ranging in 0.25µm ≤ s ≤ 5µm (c). Variation of the FOM (plotted on a logarithmic scale) as a function of p for 5 different spot sizes with s ranging in 0.25µm ≤ s ≤ 5µm.

Download Full Size | PDF

Figure 5(b) shows the variation in the transmission through the waveguide and the corresponding MC for various crystallization spot sizes as a function of the position of the spot with respect to the origin. It can be observed that as the spot moves away from the origin, the transmission increases and MC decreases. However, as the spot sizes increase, a noticeable decrease in overall transmission is observed, leading to an increase in MC.

Specifically, for spot sizes ranging from 1µm up to 5µm, a significant reduction in transmission (near zero) is observed when the spot is positioned close to the origin (p = 0µm). As the distance from the origin increases, the modulation contrast decreases rapidly. The primary objective of this exercise is to determine the relative position, p, of the smallest crystalline spot size from the start of the GC. Analyzing the modulation contrast of various spot sizes located near the origin reveals a steady increase in MC as spot size increases as shown in Fig. 5(b). For instance, the MC of spots closest to the origin shows an incremental rise from 8 dB for a 250 nm spot to 23.5 dB for a 1µm crystalline spot. However, beyond the 1µm spot size, the rate of increase in MC diminishes significantly. The MC slightly increases to 28 dB for a spot size of 2.5µm and only increases by 1 dB when the spot size is further enlarged to 5µm.

These findings demonstrate that achieving high modulation contrast is attainable with a smallest spot size of 2.5µm when positioned at a distance of 4.6µm from the waveguide. This has been shown more clearly using a plot of figure-of-merit, FOM shown in Fig. 5(c) defined by [Eq. (7)] as follows.

The findings suggest that while larger spot sizes exhibit higher FOM values up to 2.5µm, further enlarging the spot size does not yield significant improvements in FOM. This saturation phenomenon indicates a critical limit in the FOM enhancement achievable through spot size adjustments beyond 2.5µm.

These dimensions can easily be realized using conventional electrical switching techniques, which are heavily used in phase change memory (PCRAM) technology using CMOS-compatible fabrication processes. This technique would mainly require a top and bottom electrode interfacing the amorphous GST-based GC design at the desired spot. The switching can also be realized using an electrothermal switching approach, by proximally positioning a nano/micro heater with comparable dimensions of the spot as an adjacent layer in the device architecture. Passing current through the heater would also favour a more spatially uniform switching volume.

Furthermore, fabricating chalcogenide-based GC structures on standard Si photonic circuits necessitates the use of well-established deposition and lithographic methods. While various techniques, including sputtering, evaporation, and specific low-temperature CVD processes such as PECVD [37] and MOCVD [38–41]are used for chalcogenide deposition, patterning can be achieved through standard photolithography or electron beam lithography (EBL) [22]. These methodologies have demonstrated superior effectiveness in large-scale production. These processes can then be combined with either lift-off or dry etching techniques, facilitating the smooth transfer of patterns from the resist directly onto the device layer. A detailed fabrication process flow has been proposed in figure S7 in the Supplement 1.

4. Conclusion

We show the realization of a high modulation contrast reconfigurable grating coupler with a view to fabrication and switching considerations. The chalcogenide-based grating coupler design optimized using inverse design techniques allows the realization of a non-volatile, CMOS compatible reconfigurable grating coupler platform with ultra-high coupling efficiency up to 50% for integration with Si waveguides.

The optimized design offers ultrahigh modulation contrast of up to 70 dB (an order of magnitude higher than competing phase change-based integrated switches and modulators reported to date (Table S1). Chalcogenide gating couplers can be engineered, such that the operational window can be tuned across the entire telecom band by choosing the appropriate grating thickness and etch depth. The high modulation contrast in our design together with the potentially minute switching volumes makes such grating couplers highly desirable for high contrast, fast switching applications in integrated photonic circuits for interposers as well as classical and quantum telecommunication and computing as well as emerging neuromorphic photonic accelerator platforms. Furthermore, the highly efficient optimization routine detailed here allows fast optimization of any reconfigurable GC design for any desired CE at any given wavelength across the telecommunication band.