ITO-Based Microheaters for Reversible Multi-Stage Switching of Phase-Change Materials: Towards Miniaturized Beyond-Binary Reconfigurable Integrated Photonics

Inducing a large refractive-index change is the holy grail of reconfigurable photonic structures, a goal that has long been the driving force behind the discovery of new optical material platforms. Recently, the unprecedentedly large refractive-index contrast between the amorphous and crystalline states of Ge-Sb-Te (GST)-based phase-change materials (PCMs) has attracted tremendous attention for reconfigurable integrated nanophotonics. Here, we introduce a microheater platform that employs optically transparent and electrically conductive indium-tin-oxide (ITO) bridges for the fast and reversible electrical switching of the GST phase between crystalline and amorphous states. By the proper assignment of electrical pulses applied to the ITO microheater, we show that our platform allows for the registration of virtually any intermediate crystalline state into the GST film integrated on the top of the designed microheaters. More importantly, we demonstrate the full reversibility of the GST phase between amorphous and crystalline states. To show the feasibility of using this hybrid GST/ITO platform for miniaturized integrated nanophotonic structures, we integrate our designed microheaters into the arms of a Mach-Zehnder interferometer to realize electrically reconfigurable optical phase shifters with orders of magnitude smaller footprints compared to existing integrated photonic architectures. We show that the phase of optical signals can be gradually shifted in multiple intermediate states using a structure that can potentially be smaller than a single wavelength. We believe that our study showcases the possibility of forming a whole new class of miniaturized reconfigurable integrated nanophotonics using beyond-binary reconfiguration of optical functionalities in hybrid PCM-photonic devices.


Introduction
Manipulation of the refractive index forms the backbone of reconfigurable integrated photonics, a notion that constantly seeks alternative material platforms and modulation schemes for achieving new functionalities and better performance measures. [1][2][3][4] The latest trends show a growing interest in hybrid photonic devices based on phase-change materials (PCMs) with a specific focus on germanium-antimony-telluride (Ge-Sb-Te or shortly GST) alloys for operation in the infrared wavelengths. This increasing attention stems from the unprecedented large, reversible, and non-volatile change in the real (n) and imaginary (k) parts of GST refractive index upon the phase transition between amorphous and crystalline states. For instance, at the telecommunication wavelength of 1550 nm, GST provides an index change of ∆n > 2, 5 which is orders of magnitude larger than what can be obtained via alternative methods such as the thermooptic effect in silicon with ∆n /∆T ~ 1.86×10 -4 K -1 , 6 corresponding to ∆n ≈ 0.054 at ∆T = 300 o C.
Access to such a large ∆n budget enables the design of phase shifters, modulators, and switches with ultra-compact footprints, potentially down to subwavelength scales. [7][8][9][10][11] In addition, the induced index-change is nonvolatile, fast (a few ns), and reversible for many overwriting cycles (> 10 15 ), a collection of features that ensures the low-power, high-speed, and long-term reconfiguration of GST-based photonic devices. 1,4,12,13 These features have made GST alloys very appealing for reconfigurable photonic platforms such as all-optical memories, 14,15 active metasurfaces, [16][17][18] neuromorphic computing, 7,19,20 and optical switching. 10,21,22 The success of reconfigurable PCM-based photonic platforms is entangled with the reliable control of the PCM phase, which can be reversibly switched between amorphous and crystalline states through a controlled heating process. Heating GST at a moderate temperature (150-250 o C) induces transition from amorphous to crystalline, and the melting of GST at temperatures above 600 o C followed by a fast cooling (>1 o C/ns) results in the reverse conversion from crystalline to amorphous, a process that is widely referred to as the melt quenching. [23][24][25] The required heat for phase switching can be supplied directly through heating or indirectly by optical and/or electrical stimuli. The use of direct heating, however, has been limited to preliminary proof-of-concept demonstrations as it only allows for the one-way amorphous-to-crystalline conversion. On the other hand, the reversible switching of the GST phase 26,27 as well as the selective conversion of individual GST inclusions have been demonstrated using the optical stimulation. 16,28 On the downside, however, optical stimulation necessitates bulky external laser sources or complex nonemonolithic integration schemes, which hinder the implementation of fully integrated device platforms for practical applications. Moreover, because of the low optical absorption and small absorption cross-section of amorphous GST nanostructures, recrystallization cycles require highpower optical pulses. 29 Among different options, electrical heating, also known as Joule heating, holds a great promise for the ultimate miniaturization of GST-based active and reconfigurable photonic devices, primarily because the miniaturized heaters can be integrated into the device platform. In its simplest form, the Joule heating can be implemented by flowing an electrical current (I) directly through the GST film. 30 In this scheme, however, the large variation of GST resistance (RGST) between amorphous and crystalline states requires a large variation of input current to generate enough heat for the phase conversion (heat varies linearly with RGSTI 2 ). 31 This issue can be addressed by decoupling GST from the current path, potentially through the integration of GST on top of a microheater, so that the Joule heating can be controlled in the microheater (independent of the GST phase). Designing a microheater that confines the heating volume for efficient GST heating, provides a short thermal-time-constant, and does not add optical loss is challenging. The use of metals with large optical loss should be refrained in most photonic applications. As a solution, a recent study proposed the use of two-dimensional graphene for designing microheaters with low optical loss. 32 However, because of abundant defects and grain boundaries, graphene is prone to fast oxidation in the ambient, 33 which makes its long-term operation under recurring heat cycles unreliable. In addition, the integration of graphene with photonics devices requires a mechanical transfer procedure that is not compatible with mainstream fabrication processes. Other reported methods such as doped-silicon microheaters [34][35][36] are not holistic and cannot be adopted for alternative material systems such as silicon nitride (SiN) and silicon carbide (SiC) as promising material platforms for integrated nanophotonics. Thus, designing a compact, low-loss, and CMOScompatible microheater is a major bottleneck for the practical deployment of integrated photonic solutions that utilize PCMs for reconfiguration.
Here, we report the realization of a microheater platform that employs optically transparent (i.e., low-loss) and electrically conductive indium-tin-oxide (ITO) bridges for the fast and reversible switching of the GST phase between crystalline and amorphous states. By the proper assignment of width and amplitude of electrical pulses applied to the ITO bridge, we show that our platform allows for the registration of virtually any intermediate crystalline state into a GST film integrated on the top of the designed microheater. Our estimation of energy-per-pulse consumed for the amorphous-to-crystalline phase conversion in GST is ~ 6.5 nJ in our test devices with the possibility of sub-nJ operation in an optimal device configuration. Granted by the low optical loss of the ITO films, we further show that our designed microheater platform can be directly integrated on SiN-based photonic devices for the realization of electrically reconfigurable optical functionalities. In a representative demonstration, we integrate our designed ITO-based microheaters into the arms of a Mach-Zehnder interferometer (MZI) to realize electrically reconfigurable optical phase shifters. By applying optimized electrical pulses to microheaters, the Under these conditions, we obtain a deposition rate of ~ 1.7 nm/min ( Figure 1€) and a thickness uniformity of better than 10% across a two-inch wafer. Such a uniformity leads to a negligible thickness variation across our small device footprints (Figure 1(f)).
The electrical conductivity of the ITO bridge is an important design parameter that directly affects the efficiency of the Joule heating. The sputtering deposition of ITO typically yields a thin film with relatively small crystalline domains, leading to a suboptimal electrical conductivity in as-deposited films. However, the post-deposition annealing is shown to be effective for enhancing the electrical conductivity of ITO thin films, primarily via increasing crystalline-domain sizes and activating Sn donors. 37,38 Thus, we perform a rapid thermal annealing (RTA; 15 mins @ 450 o C) on our sputtered ITO, which reduces the sheet resistance of the film from ~3000 Ω/sq (before RTA) to ~100 Ω/sq (after RTA). Additionally, our RTA process is conducted under a mild flow of oxygen, which, according to a previous study, 39 reduces the optical loss of ITO thin films, a critical feature for the direct integration of our microheater with photonic devices without the addition of extra optical loss. Furthermore, the RTA process immunes the ITO bridge from random resistance variations during the heat cycles generated for the phase conversion of GST patches.
Before testing, we covered the surface of the fabricated device in Figure 1 with a 200 nmthick silicon dioxide (SiO2) capping layer. The role of the capping layer is threefold: (i) protecting the GST patch from gradual oxidation in the ambient (see Figure S1, Supporting Information (SI)), (ii) preventing the decomposition of GST during the heating cycles, and (iii) precluding the failure of the microheater caused by the electric breakdown of the air at the sharp corners of the device. Accordingly, with the SiO2 capping layer, (i) samples can be kept in the ambient for an arbitrarily long period of time, (ii) the GST patch can be successfully maintained during the full cycle of the amorphous-to-crystalline conversion, and (iii) the applied voltage can be increased to generate sufficient heat for the phase conversion of the GST patch integrated on the microheater.
For optical characterization, we primarily rely on the change of refractive index (n) and extinction coefficient (k) of GST as we electrically derive a gradual transition from the amorphous to crystalline phase via Joule heating of the GST patch. Thus, we use optical reflection spectroscopy to demonstrate the electrical control of the GST phase from the amorphous to the full-crystalline state with multiple intermediate states. As shown in Figure 2(a), after each Joule heating event, the reflection of a broadband light source from the center of the GST patch was collected via a 20X objective lens (numerical aperture, NA ≈ 0.42) and normalized to that obtained from the ITO region without GST. The measured reflection spectra contain multiple peaks and dips, which are generated by the light interference inside the stack of layers constituting the device.
The position and amplitude of these reflection features are dictated by the optical constants of various layers in the stack, among which only those of the GST layer alter upon the Joule heating events. Therefore, the modulation of peaks and dips in the reflection spectrum of a device can be used for the back-calculation of the optical properties of the GST layer and to estimate its crystalline phase after each Joule heating event.
To extract the degree of crystallization from the measured reflection spectra, we need to benchmark the estimated n and k values of GST against a reference dataset. Thus, following previous reports, 31 we vacuum-annealed an amorphous GST layer at ~ 250 o C for 5 mins, which provides a reference sample for the crystalline state of GST. Subsequently, we used ellipsometry measurements with a Tauc-Lorentz dispersion fitting 5 to directly obtain n and k of reference crystalline/amorphous GST films as depicted with solid lines in Figures 2(b) and 2(c), respectively.
To obtain similar references for the intermediate states of GST, we used an effective-medium approximation in which an α fraction of the GST volume is assumed to be in the crystalline state and the (1-α) fraction of it in the amorphous state (0 < α < 1). Thus, using the measured n and k values of amorphous (i.e., α = 0) and crystalline (i.e., α = 1) GST films, we can extrapolate n and k values for any arbitrary intermediate GST state using the Lorentz-Lorenz equation: 40 In Eq. (1), the α-dependent dielectric constant of the effective medium ( To estimate the crystallization level after each Joule heating event, we employ the transfermatrix method to regenerate the experimental reflection spectra shown in Figure 2(a). In these calculations, we model each material layer in Figure 1(a) by its thickness and optical (n and k) constants. We use (α) and (α) as the optical constants of the GST layer, and, following a brute-force approach, we sweep the α parameter from 0 to 1 and look for the best fit to the experimental results. Our calculations show that the reflection spectra displayed in Figure 2(a) can be successfully fitted with α = 0, 0.3 ± 0.1, 0.6 ± 0.1, and 0.9 ± 0.1. We also note that, the optical spectra in Figure 2 This observation suggests that the crystallization of GST is primarily a temperature-driven phenomenon and obtaining higher crystallization levels mandates the temperature increase via the application of an electrical pulse with a larger amplitude.
The GST phase can be switched back to the amorphous state by melt-quenching, a process that needs the high-temperature melting and fast cooling of crystalized GST. Thus, we increase the amplitude of the voltage pulse to ~9 V, for melting, and narrow the pulse width down to 50 nsec for the fast cooling of the GST patch. As indicated by an arrow in Figure 3(b), the application of this short pulse to the microheater successfully resets the GST resistance back to the amorphous level, which verifies the feasibility of reversible switching in our developed ITO-based microheater platform. Interestingly, the re-crystallization of the melt-quenched GST (i.e., the response in Figure 3(b) after pulse index 125) displays a sudden drop in RGST, which is then followed by the recovery of the stepwise resistance drop in the subsequent crystallization steps. In other words, the number of achievable intermediate crystalline states in the melt-quenched GST is less than that in the as-deposited amorphous film. Previous studies suggest that such a difference stems from the residual crystalline domains in the melt-quenched GST film, which bypasses the nucleation step that is required for the crystallization of as-deposited GST film. 30 Therefore, compared to as-deposited films, an accelerated crystallization is anticipated in melt-quenched GST films. Nevertheless, our platform offers reversible switching of the GST phase with multiple intermediate crystalline states in a repeatable way. We repeated this experiment several times and noticed that the subsequent crystallization cycles (not shown here) follow consistently the trend observed in the second cycle, hinting that a pre-conditioning step is required prior to the long-term operation of the device. To the best of our knowledge, this is the first demonstration of an optically transparent microheater architecture for reversible multi-stage GST phase change with an electrical stimulus. Figure 3(b) shows one order-of-magnitude change in RGST upon GST phase transition from amorphous to crystalline. This is noticeably smaller than the expected two-orders-of-magnitude change, as reported elsewhere. 11 We attribute this discrepancy to the heat-sinking effect of the large metallic (Au) electrodes (see Figure 3(a)) used for the readout of RGST. The overlap of these Au electrodes with the ITO bridge disturbs the otherwise-uniform heat profile of the microheater and creates two locally cold regions close to the left and right metallic electrodes in Figure 3(a), which leaves behind two amorphous segments on the sides of the central crystalline GST segment (see Figure S2 in SI). In other words, the measured resistance (i.e., RGST) in Figure 3 We further study the spatial and temporal temperature profiles of the device in Figure 1(a) in response to the voltage pulses applied to the ITO microheaters by solving electro-thermal equations in the COMSOL Multiphysics simulation package for the experimental condition explained in Figure 3 (see SI for details). As shown in Figure 4(a), despite some expected temperature drop across the ITO width, the temperature profile is spatially uniform along the ITO bridge, suggesting that a GST patch located at the center of the microheater can be uniformly switched between amorphous and crystalline states. In addition to the spatial temperature uniformity, our designed platform ensures the fast thermal dynamics needed for the reamorphization of crystalline GST. As shown in Figure 4   We believe that the extra loss of the device is primarily due to the material loss of the GST in the crystalline phase. Considering the existing reports for GST with considerably lower values of k at the 1550 nm wavelength, we think that further optimization of our deposition process can reduce the overall loss of the device. In addition, at the cost of making the structure a little larger, the loss can be reduced via mode engineering in the hybrid SiN/ITO/GST waveguide to shift a portion of the mode energy from GST to SiN upon conversion from the amorphous to the crystalline phase. Finally, the use of alternative GST-like alloys with lower k values (e.g., Ge2Sb2Se4Te1) 2 can further reduce the loss at the expense of relatively lower operation speeds.
We note that our developed microheater can be used for almost all PCMs, and the change of material will not cause any major change in our proposed fabrication process.
A key advantage of our phase-shifter structure in Figure 5 is that it achieves the phase shift of 50 degrees using a SiN waveguide segment with a length < 2 m. To achieve the same phase shift using a conventional Si waveguide with the free-carrier-plasma-dispersion effect at the 1550 nm wavelength using similar voltages to our structure, a waveguide length of ~ 1mm is needed.
Achieving the same phase shift with similar voltages in lithium niobate (LiNbO3) requires a waveguide length of ~ 2 mm. Thus, our platform offers 3 orders-of-magnitude reduction in the device size at the same operation voltage. This can revolutionize the field of reconfigurable integrated nanophotonics for applications where reconfiguration times of 10's of ns are sufficient.
We note that phase shifters using the thermo-optic effect in Si can have one order-of-magnitude smaller size than other conventional mechanisms (e.g., ~ 0.2 mm in this case). However, their excessive power consumption (due to the volatility of the mechanism and the need for constant heating) and slow response hinder their practical use. While the demonstrated results in Figure 5 only show the proof-of-principle for the feasibility of such integrated photonic devices, this unique advantage combined with the nonvolatility (and thus, potential ultra-low-power operation) can enable a reliable platform for a large range of reconfigurable devices. We believe through optimizing the waveguide device architecture and the material deposition and fabrication processes, our hybrid material platform with the electrically stimulated microheater architecture can serve as a reliable platform for miniaturizing reconfigurable integrated photonic structures.
To obtain a better understanding of the field profile of the structure in Figure 5,