1. Introduction

The use of optics (photonics) for low-loss, long distance (beyond 10 km) information transfer and microelectronics for information processing and storage are among the great technological achievements of the past century. In recent decades, the benefits of using photonics rather than electronics for short-distance (e.g., rack-to-rack, board-to-board, interchip, and intrachip) information transfer have been outlined [1]. Silicon photonics, boosted by the silicon processing expertise resulting from a mature microelectronics industry, has emerged as a leading platform for on-chip photonic information transfer. Although other material platforms (e.g., InP, GaAs) are being explored, this review is limited to silicon-based integrated photonic devices. The demonstration of low-loss optical transmission in nanoscale silicon waveguides [2] has spurred the development of optical switches for photonic routing and electro-optic modulators and photodetectors to connect the photonic and electronic domains in optoelectronic systems. More recently, the extent to which photonic technologies can entirely replace their electronic counterparts for information processing [3] and storage [4, 5] has been discussed, leading to increased pursuit of all-optical devices. Although silicon is suitable for low-loss optical transmission, its relatively modest range of optical properties has driven interest in hybrid silicon photonic devices in which optical functionality is provided by a secondary material (e.g., two-dimensional materials [6–10], metals [11], electro-optic polymers [12, 13], germanium [14–16], and optical phase change materials [O-PCMs]) integrated in a silicon photonic structure. In particular, O-PCMs offer a promising opportunity for realization of broadband, small footprint photonic devices due to their large changes in optical properties and compatibility with silicon processing. In this review, we outline recent theoretical and experimental work integrating O-PCMs in thermo-optic, electro-optic, and all-optical silicon photonic devices for a variety of functionalities, in addition to discussing potential Si/O-PCM next-generation photonic devices. However, we do not discuss O-PCMs integrated into silicon photonic structures for all-optical memory, since there is a recently published review [4].

2. Optical phase change materials

Optical phase-change materials (O-PCMs) have emerged as a unique class of materials that exhibit large changes in optical properties (Δn > 1, Δκ ~order of magnitude) in response to an external stimulus (e.g., temperature, applied voltage, ultrafast optical excitation). The most widely studied O-PCMs are transition metal oxides and chalcogen-based alloys; these materials have been exploited for a variety of tunable photonic applications ranging from smart windows [17] to tunable metamaterials [18–20]). Transition metal oxide O-PCMs, such as those listed in Fig. 1(a), generally undergo crystalline-crystalline transitions while chalcogen-based O-PCMs [Fig. 1(b)] undergo amorphous-crystalline transitions [21, 22]. There continues to be extensive, fundamental investigation into the nature of these transitions, and for detailed theoretical analysis, we refer the reader to [23] and [24] for transition metal oxide and chalcogen-based O-PCMs, respectively. To date, the most widely studied transition metal oxide and chalcogen-based O-PCMs have been vanadium dioxide (VO₂) and Ge₂Sb₂Te₅ (GST), respectively. The focus of this review article is integration of VO₂ and GST, in addition to other chalcogen-based O-PCMs (GeSe, Ge₂Sb₂Se₄Te₁), into silicon-based photonic devices. We first present a detailed comparison of VO₂ and GST, focusing on their optical properties and providing a qualitative overview of the mechanisms that allow access to these properties.

 

Fig. 1 (a) Selected transition metal oxide O-PCMs and the temperatures at which they demonstrate a change in their optical properties. Figure reprinted with permission from [21]. © 2011 Annual Reviews. (b) Ternary phase diagram for Te, Ge, and Sb, showing selected chalcogen-based O-PCMs. Figure reprinted with permission from [22]. © 2008 Nature Publishing Group.

Download Full Size | PDF

2.1 VO₂ and GST as prototypical O-PCMs

Vanadium dioxide, a canonical transition metal oxide O-PCM, undergoes both insulator-to-metal and crystalline-crystalline transitions, from an optically transmissive, electrically resistive, monoclinic state (VO₂:M) to an optically opaque, electrically conductive, rutile state (VO₂:R). The VO₂:R state is achieved via breaking of the V-V dimers in VO₂:M and the straightening of the V atoms along the c axis, shrinking the unit cell twofold. This transition can be accessed thermally (T > 68°C) [25], electrically [26], or with ultrafast optical excitation [27]. The VO₂:M state exists at room temperature and ambient pressure, thus making VO₂ volatile, a characteristic of transition metal oxide O-PCMs. This volatility provides a challenge for applications like non-volatile optical memory but makes transition metal oxide O-PCMs potentially well suited for volatile applications such as modulation. GST, a prototypical chalcogen-based O-PCM, undergoes an amorphous-crystalline transition from an optically transmissive, electrically resistive, amorphous state (GST:A) to an optically opaque, electrically conductive, crystalline state (GST:C). In chalcogen-based phase change materials like GST, the large change in optical properties is widely attributed to resonant bonding in the crystalline state [28]. Generally, GST:C is accessed by heating GST:A to achieve crystallization, whereas GST:A is accessed by heating GST:C above the melting temperature of GST (~600°C) followed by rapid quenching. The thermal environments required to form GST:A and GST:C are often realized with optical [29, 30] or electrical pulses [31]. Both states (GST:A and GST:C) are non-volatile, a characteristic of chalcogen-based O-PCMs, making them well suited for non-volatile optical memory but challenging for volatile applications like low-power signal modulation, because power input is required to access both states.

Figure 2 presents schematics of the atomic structure of the two states of VO₂ and GST [32, 33] along with their respective optical properties from 1450 nm to 1600 nm [28, 34]. The changes in optical properties across the phase transitions of VO₂ and GST shown in the figure are achieved with thermal excitation, and the transitions are fully driven, giving the maximum change in optical properties of the materials. If the transitions are not fully driven, regardless of whether the phase transition is accessed by thermal, optical, or electrical means, only a fraction of the change in optical as well as electrical properties will occur. Ultrafast pump-probe experiments on O-PCM thin films demonstrate that the optically-induced change in optical properties of VO₂ [27, 32, 35–40] and GST [41] can be accessed on ~femtosecond (“on”) and ~1-10 picosecond (“off”) times and provide evidence of the fractional changes in optical properties that can occur if the transition is not fully driven. In addition, femtosecond time-resolved photoelectron spectroscopy [39] and simultaneous electron-diffraction and optical transmission measurements [38, 41] have revealed the decoupling of atomic structural changes and the change in optical properties. These studies suggest that integrated, all-optical devices that incorporate VO₂ or GST must operate in fluence regimes where changes in atomic structure are not initiated in order to access the fastest “off” times. While ultrafast dynamics of VO₂ and GST have been measured following optical excitation, the minimum temporal response to an electrical impulse remains unknown. Throughout the text, for brevity, any change in the optical properties of an O-PCM, regardless of a corresponding structural change, is referred to as an optical phase change (OPC).

 

Fig. 2 Atomic structures and optical properties of VO₂ and GST. (a) Three-dimensional schematics of the low temperature (T < 68°C), monoclinic (left) and high temperature (T > 68°C), rutile (right) crystal structures of VO₂. Vanadium atoms are shown in light blue. The orange shadows highlight the V-V dimers exhibited in the monoclinic structure. Oxygen atoms are not shown. The monoclinic and rutile states of VO₂ are labeled VO₂:M and VO₂:R, respectively. Figures adapted and reprinted with permission from [32] © 2012 American Physical Society. (b) Two-dimensional schematics (Te atoms in blue; Ge and Sb atoms in gold) of the amorphous (left) and crystalline (right) states of GST. The amorphous and crystalline states of GST are labeled GST:A and GST:C, respectively. Figures adapted and reprinted with permission from [33] © 2015 Nature Publishing Group. (c) Refractive indices of VO₂ and GST. (d) Extinction coefficient of VO₂ and GST. For (c) and (d), optical properties were taken and replotted from [34] and [28] for VO₂ and GST, respectively. In both cases, the change in optical properties was thermally induced.

Download Full Size | PDF

3. Si/O-PCM hybrid photonic devices

In this section, we describe resonant and non-resonant silicon photonic devices, both in simulation and experiment, which integrate O-PCMs to achieve active functionality. These devices are classified by the mechanism (thermal, electrical, or optical) used to induce the OPC. Devices using continuous-wave photothermal heating are classified under thermo-optic functionality (Section 3.1) while those which use optical pulses to induce an OPC are classified under all-optical functionality (Section 3.3). Devices for which current passes through the O-PCM are classified under electro-optic functionality (Section 3.2), while those which use integrated electrical heaters adjacent to the O-PCM are classified under thermo-optic functionality (Section 3.1). Simulation-based devices are categorized under what is deemed to be the most plausible experimentally realized mechanism to actuate the OPC.

3.1 Thermo-optic functionality

Due to the high temperature (~600°C) required to access GST:A, active tunability of a Si/GST photonic device has not been demonstrated with simple thermal heating (i.e., without the use of optical pulses). However, due to the relatively low temperature (~68°C) required to initiate the OPC of VO₂, active tunability of hybrid Si/VO₂ waveguide and resonant-based devices has been experimentally demonstrated using various thermal mechanisms including external substrate heating [42, 43], photothermal heating [44], and integrated electrical resistive heating [45]. In [42], the authors integrated a 2 µm long VO₂ patch on a 400 µm diameter silicon ring resonator [Fig. 3(a)]. Using an external substrate heater, the temperature-dependent optical response was measured [Fig. 3(b)] and used to extract the induced absorptive optical loss resulting from the OPC of VO₂. For a 2 µm long, 65 nm thick VO₂ patch atop a linear ridge waveguide, the authors calculate losses of 2 and 9 dB for VO₂:M and VO₂:R, giving insertion loss and extinction ratio of 2 dB and 7 dB, respectively.

 

Fig. 3 (a) Schematic and scanning electron microscopy (SEM) images of VO₂ coated silicon ring resonator. (b) Temperature-dependent transmission of Si/VO₂ ring resonator in (a), demonstrating the change in optical response as VO₂ undergoes its OPC. Figures in (a) and (b) reprinted with permission from [42] © 2010 The Optical Society. (c) Optical transmission of 1.5 µm radius Si/VO₂ ring resonator (SEM inset top left with VO₂ false colored maroon). At the selected wavelength (dashed line), optical transmission is low with no laser-induced photothermal heating (“laser off” inset) while transmission is high with laser induced photothermal heating (“laser on” inset) due to the resonance shift induced by the OPC of VO₂. Small scale bar in SEM image inset is 250 nm. Figures adapted and reprinted with permission from [44] © 2012 The Optical Society. (d) Proposed 2 × 2 Si/VO₂ microring switch. Figure reprinted with permission from [46] © 2016 Institute of Electrical and Electronics Engineers. (e) SEM images of design showing VO₂ (false colored green) embedded within a silicon waveguide. Each bifurcated silicon waveguide (false colored navy) splits into a control waveguide (blue box) and VO₂ embedded waveguide (orange box). The left side of the figure shows tilted (top) and normal incidence (bottom) SEM images of the VO₂ embedded waveguide. The integrated heaters are false colored gold. Figure reprinted with permission from [45] © 2017 The Optical Society. (f) Schematics of proposed pass polarizer using VO₂ on a silicon waveguide (blue). Purple and grey blocks represent VO₂:M and VO₂:R, respectively. Quasi TE and TM light are represented by blue and red arrows, respectively. Figure reprinted with permission from [47] © 2015 The Optical Society.

Download Full Size | PDF

Subsequent work substantially reduced the device footprint by integrating VO₂ on ultracompact silicon ring resonators [44]. For a 500 nm long VO₂ patch on a quasi TE mode ring resonator of radius of 1.5 µm, the authors demonstrated extinction ratio in excess of 10 dB by photothermally inducing the OPC of VO₂. In this experiment, the refractive index change accompanying the OPC of VO₂ led to a change in the optical path length of the ring waveguide and subsequently a change in the resonance wavelength of the ring resonator [(Fig. 3(c)]. Although these thermo-optic experiments demonstrated modulation on inherently slow time scales (i.e., seconds to minutes) due to the selected methods of thermal actuation of the OPC of VO₂, they did realize a platform for high extinction ratio (~10 dB), low loss (~1-2 dB) operation of hybrid Si/VO₂ devices. With the demonstrated ultrafast dynamics of VO₂ under optical excitation [27, 32, 35–40], it is expected that Si/VO₂ hybrid ring resonators can be leveraged to realize high-speed optical modulation with out-of-plane ultrafast optical excitation. Where slower routing speeds are acceptable, [46] suggests a plausible design for a 2 × 2 optical switch based on a Si/VO₂ hybrid ring resonator with two bus waveguides, as in Fig. 3(d).

More recent work has explored linear, non-resonant designs to further increase the optical interaction of guided modes in silicon waveguides with an O-PCM. Figure 3(e) shows silicon waveguides with embedded patches of VO₂ [45]. Using resistive heating from an integrated electrical heater adjacent to the waveguide, approximately 10 dB broadband extinction was demonstrated using a silicon waveguide with a 500 nm long VO₂ embedded segment. While insertion loss of this structure was 6.5 dB, simulations suggest 13.8 dB extinction ratio with 2.2 dB insertion loss is possible with optimized processing conditions. The most promising implementation of this design is in an all-optical modulator with in-waveguide excitation and will be discussed further in Section 4.2. Pass polarizers are another integrated photonic application that could benefit from using O-PCMs. Moreover, pass polarizers do not require high-speed implementation and could therefore be implemented with thermal actuation of the OPC. Ref [47]. reports designs of Si/VO₂ structures that are suitable for pass polarizer applications [Fig. 3(f)]. For example, using a 300 nm wide silicon waveguide with VO₂ on the sides of the waveguide, ~15 dB extinction of TE light is predicted for VO₂:R while input light with TM polarization is expected to experience only ~3 dB extinction. Similarly, to primarily pass TE polarized light, actuating the OPC of a patch of VO₂ on top of a 400 nm wide silicon waveguide with a 20 nm silica spacer is simulated to give greater than 15 dB extinction of TM light and ~2.8 dB extinction of TE light. We note that successful implementation of this design will require extremely careful fabrication to place the desired VO₂ patches on the sidewalls of the waveguides. In addition, if implemented thermally, the design considerations must ensure that in a geometry as proposed in Fig. 3(f), individual VO₂ segments can be selectively actuated.

3.2 Electro-optic functionality

Optical phase change materials have been proposed and integrated into electro-optic silicon-based photonic devices. Experimentally, electrical actuation of the OPC of a VO₂ patch coating a silicon waveguide has been demonstrated [48–50]. In both linear-waveguide [Fig. 4(a)] and ring-resonator geometries [48, 50], the size of the active VO₂ patch (controlled by either limiting the volume fraction of VO₂ that is switched or depositing smaller patches) has been shown to affect simultaneously the extinction ratio and response time. Generally, switching larger active VO₂ patches on silicon photonic structures leads to larger extinction ratios and longer VO₂:R to VO₂:M transition times. For example, a hybrid Si/VO₂ electro-optic waveguide modulator [Fig. 4(a)] exhibited transmission “on” and “off” times commensurate with a 10 nanosecond electrical square pulse but with only 1 dB extinction ratio; applying longer electrical pulses allowed a larger volume fraction of the VO₂ patch to switch, resulting in larger extinction ratios but at the expense of longer “off” times [48]. To date, the fastest VO₂:M to VO₂:R and VO₂:R to VO₂:M transitions of VO₂ have been shown to be no more than 2 [48, 51] and 3 nanoseconds [48], respectively, in purely electrical measurements, although there remains some potential for improved temporal response.

 

Fig. 4 (a) SEM image of Si/VO₂ electro-optic waveguide device. VO₂ and Au are false colored purple and gold, respectively. Figure reprinted with permission from [48] © 2015 American Chemical Society. (b) Optical microscope image of Si/VO₂ electro-optic waveguide device which delocalizes the optical mode to increase interaction with VO₂:R. Figure reprinted with permission from [49] © 2015 The Optical Society. (c) Proposed Si/VO₂ electro-optic modulator design based on directional coupler theory. Figure reprinted with permission from [57] © 2014 The Optical Society. (d) Proposed Si/VO₂ electro-optic design including a vertically embedded VO₂ section within the silicon waveguide. Figure reprinted with permission from [58] © 2017 Institute of Electrical and Electronics Engineers. (e) Proposed Si/Au/VO₂ electro-optic modulator design based on near field plasmonic coupling. Figure adapted and reprinted with permission from [59] © 2015 The Optical Society. (f) Proposed Si/GST electro-optic device whereby a thin ribbon of GST embedded within a silicon waveguide is electrically actuated. Figure reprinted with permission from [60] © 2015 Institute of Electrical and Electronics Engineers.

Download Full Size | PDF

While it is generally agreed that a purely thermal mechanism does not drive the OPC of VO₂ in electrical measurements [52], inquiries continue to determine whether the mechanism is purely field-based [52, 53] or is instead mediated by carrier injection effects, (i.e., Poole-Frenkel emission) [48, 54, 55]. For a detailed analysis of electrical induction of the OPC of VO₂, we refer the reader to [21] and [56]. From a practical point of view, it is clear that minimizing the duration of the VO₂:M to VO₂:R and VO₂:R to VO₂:M transitions is critical for the operation of electro-optic modulators and is discussed in Section 4.1.1. Modifying the geometry of hybrid Si/VO₂ photonic structures may hold the key to realizing electro-optic modulators that have both ultrafast response time and large extinction ratios. In Ref [49], the authors shrank the silicon waveguide width to 300 nm to delocalize the mode and thereby force increased modal interaction with VO₂:R. This optimized design [Fig. 4(b)] yielded increased optical contrast, giving 12 dB extinction with a 1 µm long VO₂ patch, but response times greater than 1 µs were measured. This device was also characterized as a photodetector, demonstrating a responsivity exceeding 10 A/W.

Several interesting electro-optic device geometries that embed an O-PCM within a silicon waveguide have been proposed, and structures with VO₂ embedded within silicon waveguides are described in Refs [57, 58, 61]. Ref [57]. describes a device geometry where optical transmission is controlled by electrically actuating a thin film of VO₂ in a Si/SiO₂/VO₂/SiO₂/Si waveguide adjacent to the input silicon waveguide. The proposed device geometry is shown in Fig. 4(c) and suggests that implementation as a modulator will produced an extinction ratio greater than 3 dB with ~1 dB insertion loss over the entire C-band (1.53 - 1.565 µm). In Ref [58], a vertical slot of VO₂ is located in the center of a heavily doped silicon waveguide [Fig. 4(d)]. The optimal device geometry for maximum changes in effective index (n_eff) and propagation losses (α) was determined by varying the widths of the silicon waveguide and VO₂ sections. As an electro-absorption modulator, simulations for the optimized design suggest very large extinction ratio (21 dB) and modest insertion loss (3 dB) are possible for a device length of 1 µm. Similarly, using the engineered Δn_eff, the authors also proposed a DC 1 × 2 optical switch for efficient routing. The feasibility of this modulator and switch will require improved fabrication leading to successful deposition of VO₂ in high aspect-ratio nanoscale gaps. For successful implementation of either design as a modulator, it must be confirmed that high-speed electrical performance is not compromised when using silicon for electrical actuation of the OPC of VO₂ since the electrical conductivity of highly p-doped silicon (10²⁰ cm⁻³ dopant concentration) is three orders of magnitude less than that of gold at room temperature [62, 63]. In addition to these embedded designs, Si/VO₂/metal hybrid photonic-plasmonic designs have been described [59, 64]. One example shown in Fig. 4(e) is a Si/VO₂/Au electro-optic modulator based on near field plasmonic coupling that is expected to demonstrate ~9 dB/µm extinction ratio/length [59].

Designs for electro-optic devices with embedded chalcogen-based O-PCMs have also been proposed, although experimental demonstrations have yet to be reported. Both GST [59, 65] and GeSe [66], which exhibits less optical absorption than GST in both its crystalline and amorphous states, have been considered for integration in electro-optic devices. For example, integration of a thin 10 nm film of GST or GeSe at the midlevel of a doped silicon waveguide [Fig. 4(f)] has been proposed for optical routing in MZIs and directional couplers, considering responses to both quasi TE and quasi TM inputs [59, 65, 66]. In these geometries, the doped silicon waveguide sections above and below the O-PCM film supply the necessary electrical actuation of the OPC. In addition, the authors of Ref [59]. proposed using this geometry as a variable optical attenuator and digital modulator, suggesting that extinction ratios exceeding 10 dB with less than 1 dB insertion loss can be achieved over a broad wavelength range. As with [57, 58], successful implementation of these devices requires that electrical performance is not compromised when doped silicon is used for electrical activation of the OPC.

3.3 All-optical functionality

Silicon photonic devices with all-optical functionality have been proposed and realized using optical activation of the OPC in both VO₂ and GST. Following the work described in Section 3.1 in which VO₂ was integrated on an ultrasmall silicon ring resonator coupled to a silicon waveguide [44], transient functionality was demonstrated using a 25 nanosecond (FWHM) optical pulse to initiate the OPC of VO₂ [67]. Figure 5(a) shows the Si/VO₂ ring resonator in addition to the transient optical transmission as a function of incident fluence. At lower fluences, the optical response approaches the temporal envelope of the excitation pulse. With increasing fluence, the maximum transmission intensity increases as a result of accessing the full OPC of VO₂; however, the response time increases as well. For the highest fluences shown, longer device response times are associated with heating the device beyond what is required solely to access the OPC of the VO₂. Thus, for the highest fluences, heat dissipation timescales strongly affect the overall response time of the Si/VO₂ ring resonator.

 

Fig. 5 (a) Transient response of Si/VO₂ ring resonator as a function of increasing pump fluence from 0.45 to 4.74 mJ/cm² (blue to red). SEM image of Si/VO₂ ring resonator in top right (VO₂ is colored maroon). Small scale bar in SEM image inset is 250 nm. Figure adapted and reprinted with permission from [44, 67] © 2012, 2013 The Optical Society. (b) Schematic and optical microscope image of Si/GST ring resonator. Figure reprinted with permission from [68]. © 2013 American Institute of Physics. (c) Schematic of Si/GST multimode waveguide device. Figure reprinted with permission from [69] © 2012 The Optical Society. (d) Schematic (top) and SEM images (bottom) of a 2 µm long GST patch embedded inside of a silicon waveguide. Bottom left SEM shows device cross section (A-B) perpendicular to the direction of propagation. Bottom right SEM shows device cross section (C-D) parallel to the direction of propagation. Out-of-plane optical pulses (660 nm, 89 mW peak power, 500 nanoseconds) crystallize the GST, and the change in optical propagation through the silicon waveguide is measured. Figure reprinted with permission from [70] © 2010 Institute of Engineering and Technology. (e) Schematic for proposed 2 × 2 switch implementing a chalcogen-based O-PCM with low optical loss (GSST) as the active component. Figure reprinted with permission from [73] © 2018 The Optical Society. (f) Schematic for proposed Si/Au/VO₂ all-optical modulator. Figure reprinted with permission from [74] © 2018 Institute of Electrical and Electronics Engineers.

Download Full Size | PDF

Nanosecond optical pulses have also been used to activate the OPC of GST, demonstrating optical tunability of silicon photonic structures including GST-coated ring resonators [68] [Fig. 5(b)] and multimode waveguides [69] [Fig. 5(c)] in addition to silicon waveguides embedded with GST [70, 71] [Fig. 5(d)]. In these examples, transient functionality was not examined. The nanosecond optical pulses simply served to access the GST:A and GST:C states. In simulation, the design of a 2 × 2 optical switch was proposed using a GST waveguide between two silicon waveguides whereby the state of the GST determines transmission through the bar and cross ports [72]. The authors in Ref [73]. explored a similar geometry [Fig. 5(e)] but integrated Ge₂Sb₂Se₄Te₁ (GSST) which, compared to GST, exhibits superior optical transparency in both its amorphous and crystalline states. Additionally, recent work has proposed the use of a Au/VO₂ hybrid pattern on a silicon waveguide for in-waveguide, all-optical modulation [74] [Fig. 5(f)]. Although this design suggests large extinction ratio/length (24 dB/µm), it exhibits large insertion loss (of order 7 dB) and is expected to require relatively complex fabrication.

4. Outlook for O-PCMs in silicon electro-optic and all-optical modulators

While a general overview of Si/O-PCM photonic devices for a variety of potential applications was presented in Section 3, here we focus on the implementation of Si/O-PCM hybrid photonic structures as electro-optic or all-optical modulators. Considering the performance metrics for optical modulators (modulation speed, extinction ratio, optical bandwidth, insertion loss, device footprint, and energy consumption) [75], O-PCMs demonstrate significant promise with respect to device footprint, optical bandwidth, and extinction ratio as a result of their high contrast and broadband optical properties. However, practical implementation of Si/O-PCM electro-optic and all-optical modulators requires demonstrated improvement with regards to modulation speed, insertion loss, and potentially energy consumption. Below, with these three metrics in mind, experimental progress and promising avenues for further development of O-PCMs are discussed for both electro-optic and all-optical modulators.

4.1 Potential of Si/O-PCM electro-optic modulators

Although a Si/O-PCM electro-optic device with Gbps operation capability has yet to be demonstrated, there is experimental evidence suggesting this is a reasonable possibility. Below is a discussion of the potential of transition metal oxide (e.g., VO₂) and chalcogen-based (e.g., GST) O-PCMs to be utilized for these devices.

4.1.1 VO₂ and other transition metal oxide O-PCMs

As mentioned in Section 3.2, the ultimate temporal dynamics of the electrically induced OPC of VO₂ remain to be determined. With respect to the VO₂:M to VO₂:R transition, in Ref [51], the authors measured the voltage-induced electrical response of VO₂ in the device geometry shown in Fig. 6(a). The authors define the VO₂:M to VO₂:R switching time (τ) as the time it takes for VO₂ to change its electrical resistivity by 90% in log scale, measured to be about 1.9 nanoseconds. However, as shown in Fig. 6(b), the current density appears to increase linearly from the start of the voltage pulse until τ, demonstrating that the VO₂:M to VO₂:R transition begins taking place on time scales faster than 1.9 nanoseconds. Assuming the OPC occurs simultaneously with the change in electrical conductivity, this suggests electrically induced optical changes in VO₂ can occur faster than 1.9 nanoseconds, although likely with lower magnitude optical contrast since the transition is not fully driven. However, it is expected that operation in this regime of fractional optical and electrical change in VO₂ will provide a faster pathway to VO₂:M, giving faster VO₂:R to VO₂:M transition times.

 

Fig. 6 (a) Schematic of device used for probing voltage-induced electrical dynamics of VO₂. (b) Current density response of device in (a), demonstrating the increase in current in response to a voltage pulse. Figures reprinted with permission from [51] © 2013 Institute of Electrical and Electronics Engineers.

Download Full Size | PDF

In addition, we note that fast (below 1 ns) metallization transitions have been observed in other transition metal oxide O-PCMs (e.g., V₂O₃) [76]. Effective utilization of the fastest dynamics of the OPC of VO₂ will require an electro-optic design that utilizes VO₂ only between the electrical contacts where the electric field intensity is strongest (not achieved in [48, 50]). Such a design could be achieved by embedding VO₂ within a silicon waveguide to increase the modal interaction, potentially using designs proposed in Ref [59], Ref [57]. [Fig. 4(c)], Ref [58]. [Fig. 4(d)], or a modified version of the design used in Ref [48] [Fig. 4(a)] in which VO₂ is embedded in the silicon waveguide to increase the light-material interaction. Assuming 1 - 100 GHz operation, for a waveguide-based modulator with VO₂ on top of the waveguide, [49] reported calculations that suggest 0.35 - 35 fJ/bit energy consumption is attainable, which is within the desired modulator performance [75]. For Si/VO₂ modulators, insertion loss must be carefully considered due to the non-zero κ of VO₂:M, resulting in optical absorption in the passive state of the modulator. Further design considerations to minimize optical interaction with VO₂:M but maximize interaction with VO₂:R will be beneficial to achieve low insertion loss while maximizing the extinction ratio. Moreover, designs for electro-optic devices must incorporate features to reduce optical losses from metal contacts or sections of doped silicon.

4.1.2 GST and other chalcogen-based O-PCMs

Since the proposal of GST for use as “universal memory” [77], two primary limitations have been highlighted: (i) relatively slow “set” (GST:A to GST:C) speed and (ii) high “reset” (GST:C to GST:A) energy. While overcoming these limitations is crucial for realizing high-speed, low-energy memory applications, improvements could also pave the way for use of GST and other chalcogen-based O-PCMs in electro-optic modulators. One approach to increase the temporal dynamics of the OPC of GST reported in [78] is to apply a constant low voltage to modify the crystallization kinetics while initiating the “set” and “reset” electrical pulses. Using this approach, both crystallization (“set”) and amorphization (“reset”) speeds of 500 ps were achieved. Reductions in both “set” time and “reset” energy may also be achieved through the use of alternative materials. For example, devices using Ti_0.4Sb₂Te₃ [79] and GeTe [80] exhibited lower minimum switching energy and faster switching times compared to identical respective devices employing GST; however, further improvements are still required to realize competitive switching performance metrics. Reducing the cell size of the O-PCM is another approach to reduce both time dynamics [31] and switching energy [79]. Given the progress to date, further exploration of chalcogen-based O-PCM materials is warranted and could result in their integration within silicon photonic devices for high-speed, energy-efficient, high extinction ratio electro-optic modulation. For operation with low insertion loss, design considerations must account for placement of the O-PCM within the modulator if the O-PCM of choice has a non-zero κ across the operating wavelength range. In addition, as mentioned above for VO₂, design considerations must be evaluated to minimize optical losses resulting from interaction with electrical contacts or sections of doped silicon used to initiate the OPC.

4.2 Potential of Si/O-PCM all-optical modulators

Although optical modulation of an in-waveguide signal has been achieved using an out-of-plane pump signal (see Section 3.3), in this section, we consider only integrated photonic all-optical modulators where both the pump and probe signals are in the same plane of propagation. Although first demonstrated in a silicon photonic crystal nanocavity [81], in an effort to further increase operation speeds and reduce energy consumption, all-optical modulation in these integrated photonic structures has often relied on methods to access non-linear optical properties not available in silicon. This has included, for example, use of other substrates (e.g., GaAs, InGaAsP) [82, 83] or integration of additional materials (e.g., nonlinear polymers, silicon nanocrystals embedded in SiO₂) in silicon-based platforms [12, 84]. However, these devices have either suffered from large footprints [12] or required the use of a resonant cavity [82–84]. Optical phase change materials provide an interesting alternative in these hybrid geometries, especially with recent demonstration of ultrafast dynamics (discussed in Section 2.1) in thin film samples. Geometries where the O-PCM is embedded within a silicon waveguide [45, 70, 71] [Fig. 3(c) and Fig. 5(d)] may provide a platform where those same ultrafast dynamics can be accessed within a silicon photonic all-optical modulator. In these geometries, the OPC can be actuated by pump light within the silicon waveguide. While not yet experimentally achieved due to fabrication limitations of sputtering deposition in high aspect ratio trenches, a 200 nm long embedded VO₂ section in a silicon waveguide is expected to give 13.8 dB and 2.2 dB extinction and insertion loss, respectively, for quasi TE optical input [45]. Conformal deposition techniques (e.g., atomic layer deposition [85, 86]) may be a promising way to realize these geometries in practice, but this has not yet been demonstrated. Considering the band-edge threshold fluence of VO₂ at 0.75 eV [87] and the cross-sectional area of the silicon waveguide, it is expected that a pump-pulse energy of 880 fJ is required to access the OPC of VO₂ [88]. Energy consumption could be reduced by optimizing the waveguide geometry, integrating plasmonic features [89], or utilizing photonic cavities (although cavities would compromise broadband operation). While it has been shown that the OPC of VO₂ can be activated with near-infrared radiation [87], to our knowledge, this has not been demonstrated in all O-PCMs. Therefore, in particular for silicon photonic platforms, consideration of other O-PCMs in this geometry requires verification that the OPC can be induced using wavelengths where silicon waveguides are essentially transparent.

The disparity between simulated and experimental performance of Si/VO₂ devices leads naturally to the question of how optical scattering in the polycrystalline VO₂ films deposited on or embedded in hybrid devices affects device performance. The only examples of such devices actually produced thus far have incorporated polycrystalline VO₂ films produced by sputtering or pulsed laser deposition. Both techniques produce dense, robust films with grain sizes of order 100 nm, and rms surface roughness of order 5-15 nm. However, there are as yet no systematic studies that would permit some correlation to be drawn between scattering and device performance. Nor are there any systematic studies that would make it possible to ascertain whether GST – a non-crystalline solid – can be also integrated into silicon structures to produce photonic devices with appropriate scattering properties. For now, it must simply be said that this is an area for near-term investigation.

5. Conclusion

In summary, we presented an overview of silicon-based integrated photonic structures that incorporate optical phase change materials (O-PCMs) for their active functionality. Specifically, the promise of O-PCMs is demonstrated through theoretically proposed and experimentally realized thermo-optic, electro-optic, and all-optical Si/O-PCM devices. There is no question that the unique optical properties of O-PCMs exhibit the potential to realize integrated silicon photonic devices with high optical contrast, low footprint, and large optical bandwidth. While further improvement of switching speed, insertion loss, and potentially energy consumption will be required to realize commercially viable integrated Si/O-PCM modulators, this review suggests that these metrics can indeed be further improved with intelligent waveguide designs that consider the most desirable modal interaction with the O-PCMs, and by tuning the elemental composition of chalcogen-based O-PCMs to reduce optical loss and switching energy.