VO₂ nanophotonics Open Access

Nanophotonics has enabled tremendous advances in the understanding of light–matter interaction and opened up new ways to control light at the nanoscale.^1–3 Through the arrangement of designed nanostructured materials, a myriad of possibilities has recently emerged for applications in datacom, quantum optics, displays, bio-sensing, or wavefront shaping. In particular, recent progress in dielectric metasurfaces made possible the design and fabrication of flat optics devices that hold promise to replace conventional bulk optics.^4,5 At optical frequencies ranging from the UV to the mid-infrared (MIR) regions, nanophotonic systems are patterned at the micro- and nano-scale, resulting in building blocks whose geometries and arrangement are definitively set after fabrication. Such low-dimensional devices are static, which makes the dynamic variation of their physical properties not exempt of challenges. Most (if not all) nanophotonic applications would strongly benefit from tunable and reconfigurable properties, and finding solutions to overcome such challenges is currently an intense field of research in which many different strategies are explored. A non-exhaustive list would include electro-mechanical systems, liquid crystals, thermal modulations, non-linear optics, and piezoelectric effects. More recently, phase-change materials (PCMs) have become a popular method of optical tunability without any moving parts. These materials are indeed very promising to enable dynamic modification to the physical properties of devices at the micro- and nano-scale. PCMs are a class of materials with unique physical properties: their structural arrangement can be controllably modified back and forth on a fast timescale using a thermal, electrical, or optical excitation.^6,7 For some of these materials, the crystallographic re-arrangement translates into a large refractive index modification (Δn ≥ 1). Such a large and fast refractive index modulation is a long-sought effect for photonics: an enabling technology to control and tune in real-time the optical properties of devices at the nanoscale.

Among them, vanadium dioxide (VO₂) is a prototypical example of functional materials showing large modifications in their physical properties upon specific external excitation. The complex physics of VO₂—which we will briefly describe later in this Perspective —ignited discussions among researchers on whether this material should belong to the class of PCMs. Regardless of this debate, we feel that the term PCM is particularly well-suited to group together materials whose optical properties can be dynamically modified via a change in their atomic structure, a definition that works for both the tunable chalcogenide materials (e.g., GST and GeTe) and oxides (VO₂).

The insulator-to-metal transition (IMT) partly governing the drastic physical changes in VO₂ was first discovered by Morin in 1959 during his investigations on the temperature dependence of electrical conductivity in several transition-metal oxides.⁸ A few years later, another team of researchers from the same laboratory (Bell Labs) published a complete experimental dataset of the optical properties of VO₂ above and below the transition temperature, which were found largely tunable for both bulk crystals and thin films.^9,10

Decades later, in the late 2000s, after years of intense fundamental research on this fascinating material, VO₂ started to be integrated in photonic devices, driven by the need to dynamically modify the response of optical metamaterials.^11–13 Since then, VO₂ thin films have been exploited in a plethora of devices, concepts, and systems in the field of nanophotonics. In this Perspective review, our motivation is therefore to summarize this decade of research exploiting the IMT of VO₂ to dynamically modulate nanophotonic devices and systems.

A few reviews on the physics and properties of VO₂ can already be found in the literature.^14–16 They are, in most cases, either dedicated to understanding the complex physics of this material or to review their multiple possible applications. The objective of the present Perspective is not to add another layer of extensive review of all research studies conducted on VO₂. Our goal here is to give an in-depth and specific look at the recent results obtained on VO₂ applied to nanophotonics in wavelength ranges from the UV to the mid-infrared. We will see that this topic alone has already garnered so much interest that a review paper appears to be necessary. From there, we want the reader to get an overall panorama of current research on the field and on future promising perspectives.

In Sec. II, we briefly describe the intriguing physics of VO₂ and explain the origin of the transition observed in this material. In Sec. III, we then describe in more detail the optical properties of VO₂ and give recommendations to study and analyze them. Section IV is dedicated to review recent works on VO₂ for integrated guided wave optics, and Sec. V is focused on VO₂-based metasurfaces for free-space photonics. In Sec. VI, we describe a few interesting directions to explore that we find particularly promising for future works on the field, namely, the electrical control, the absorption mitigation, the spatial addressability combined with the memory effect, and the ultrafast switching of VO₂.

VO₂ is a particularly complex material with extremely rich physics that have been actively investigated in the past decades. It is a prototypical example of correlated oxides and Mott insulators. Describing the physics of Mott insulators is outside the scope of this Perspective paper, and we refer the readers to several excellent reviews on the topic.^17,18 However, the basic physics behind this phenomenon can be understood as follows. Band theory has proven to be a very powerful tool to predict and understand the physical properties of materials with known elements arranged in a specific crystallographic structure. This has helped classify materials in “boxes” according to their physical properties, linked to their band diagram. For example, insulators have their Fermi level in a bandgap, while metals have their Fermi level inside a partially filled band. The distribution of energy and carriers, as calculated through the energy band diagrams, therefore dictates the physics of the materials. However, for some materials, this band theory fails, at least partially. Indeed, the physical properties of some transition-metal oxides do not follow band theory. VO₂ is one of the most well-known examples of such materials: it has a partially filled d-electron band and, according to band theory, should therefore behave as a metal but surprisingly shows insulating properties at room temperature. This unexpected effect is due to the strong electron–electron Coulombic repulsion that exists in this material and is not taken into account in band theory. Such a correlation between electrons “freezes” them in their sites and prevents electrical conduction: this is the basic principle of Mott insulators and strongly correlated materials such as VO₂. The IMT is a natural consequence of this strong electron–electron correlation: an appropriate stimulus will break this equilibrium (e.g., heat, above 70 °C) and electrons will start to behave as free carriers, similarly as in a regular metal.

Simultaneous to this IMT behavior, VO₂ undergoes a structural change from a monoclinic phase M1 with space group of P2₁/c at room temperature to a rutile phase with space group of P4₂/mnm at temperatures above 70 °C.⁸ Under strain¹⁹ or doping with, e.g., Al or Cr,^20–22 another type of monoclinic phase M2 with space group of C2/m and triclinic phase T with space group of P1 were stabilized. These monoclinic phases M1 and M2 and triclinic phases possess different patterns in V–V bond length and volume change with respect to the rutile phase. A thermodynamic phase diagram has been established experimentally by Park et al., which illustrates the stability of these monoclinic phases in various strain-temperature windows.¹⁹ 

This crystallographic transition does not simplify the analysis and understanding of the overall physical properties. Peierls suggested that the structural change induces a lattice deformation, which modifies the periodic ionic potential in the material, resulting in a band structure change.²³ Consequently, we have at least two possible explanations as to what could be the driving force and the physical mechanism behind the transition of VO₂. On the one hand, we have the evidence of strong electron–electron correlation, suggesting a Mott–Hubbard scenario with strong Coulomb interaction between electrons, triggering the insulator–metal transition.²⁴ On the other hand, a structural transformation with strong dimerization, as predicted by Peierls, is also possible, wherein modification of the band structure is caused by the lattice distortion. Critical theoretical works in the last couple of decades by Eyert,^25,26 Biermann et al.,²⁷ He and Millis,²⁸ and van Veenendaal,²⁹ among others, have described the electronic structure as well as the magnetic ground state of vanadium dioxides.¹⁴ Using first principles studies, Eyert reproduced the basic features of rutile and monoclinic phases with density functional theory and local-density approximations.²⁵ With hybrid functional corrections, better agreement with experiments such as bandgaps and antiferromagnetic ordering has been obtained.²⁶ Based on cluster dynamical mean field theory within density functionals, Biermann et al.²⁷ suggested that correlation assisted dynamical V–V singlet pairs play a critical role in the transition of VO₂. Recent theoretical studies on the transition and band structure suggested an intermediate theory combining both Peierls and Mott, described in terms of Mott-assisted Peierls transition.³⁰ This theory seems also consistent with the recent reports of a photoinduced transient “metal-like” state of VO₂ that is produced without modifying the monoclinic phase, hence indicating a purely electronic transition.^31,32 The ultrafast THz phase diagram reported by Cocker et al.³³ shows that by increasing the incident laser fluence, the rutile phase can be nucleated and stabilized. Below 180 K, a transient metallic monoclinic state emerges prior to the nucleation of the rutile phase. This photoinduced ultrafast dynamics of VO₂ is consistent with the theoretical models proposed by He and Millis²⁸ and van Veenendaal²⁹ in critical fluency, coherent structural motion, and metastable M1 metal phases. The difficulty in experimentally untangling the structural and electronic aspects of this transition^34,35 due to the transient nature poses further challenges to this field. Further technique developments to probe local dynamics of phase evolution under external fields will be important as this class of materials becomes more widely studied and is implemented in device technologies.

Regardless of the intricate physical mechanisms governing the change of phase in VO₂, one of its main advantages is its multifunctional and multistimuli character. Indeed, the IMT of VO₂ can be induced using thermal heating (beyond T_c ∼ 70 °C at standard pressure), applied electric fields (E = 10⁵ V cm⁻¹),^36,37 injected carrier densities (n_e = 10¹⁸ cm⁻³),³⁸ and optical^39,40 and terahertz pulses.⁴¹ Conversely, this material and its IMT can then be used as an electrical switch, a thermal modulator, a thermochromic window, and an optical attenuator, to name just a few. For more details on all the possible applications of VO₂, we refer the readers to recent reviews.^14,16

The optical properties of VO₂ thin films can be affected by a variety of factors, including deposition conditions, post-treatments, lattice matching to substrates, microstructures, and impurity concentrations. Understanding the optical behaviors of each specific sample at insulating and metallic phases and their optical evolution through the IMT process is therefore not only crucial for further utilization but will empower the optimization of VO₂ performance as a tunable medium in various optical and optoelectronic devices. Describing the various growth methods and related fabrication strategies is outside the scope of the present Perspective paper, and we redirect the interested readers to a recent review focused on that aspect.¹⁵ 

In this section, we describe the optical properties of VO₂ and their relation to the thin film’s structure and pinpoint some of the difficulties in analyzing these complex and tunable optical properties.

Spectroscopic ellipsometry has been the standard method to investigate the optical properties of thin VO₂ films. Their refractive indices across the characteristic IMT evolution have been reported from ultra-violet to infrared frequency ranges.^42–45 The most pronounced changes occur at infrared wavelengths (Fig. 1), where the refractive index value strongly decreases [Fig. 1(a)], and there is simultaneously a large increase in the extinction coefficient [Fig. 1(b)]. Around the transition temperature and above, Drude losses associated with the metallic delocalized electrons take the place of Lorentzian absorptions in the imaginary dielectric function ε₂, and the permittivity ε₁ gradually goes to negative, indicating the transition from the insulating to metallic phase [see Fig. 1(c)]. VO₂ therefore presents broad absorptions even in its insulating phase. Such a lack of transparency is known to pose great challenges to optically determine the film thickness due to the commonly observed correlations between the thickness and the refractive index for absorbing materials. The simplest method to prevent these correlations is to measure the film’s thickness via alternative methods such as profilometry, AFM, or x-ray diffraction, hence avoiding the need to let the thickness be a free fit parameter. However, these methods are not necessarily relevant for very thin films having a non-negligible surface roughness. Another common strategy to break such correlations is to simultaneously analyze ellipsometric data acquired from films with different thicknesses. Assuming that the set of films have identical dielectric functions,⁴⁶ the thickness was regarded as the sole cause in varying ellipsometric data, thus enabling the independent determination of the thickness and refractive index. The main concern associated with this multi-sample approach is the assumption validity when applied to thin VO₂ films. Many findings contradicted this assumption, reporting optical properties that depend on the thickness,^44,47 due to several factors such as surface roughness and vertical gradients in microstructures and in stoichiometry.

Another approach that we find particularly suited to break the correlations is through analyzing IMT dynamic data using a tunable optical model while maintaining a constant thickness to fit all states of VO₂.⁴³ This method focuses on the dominant effect on ellipsometric data originated from the evolution of dielectric functions during the phase transition process. By constraining the absorption resonance energies and linking oscillator parameters during the IMT, the correct thickness is expected to fit the large amount of spectroscopic data at various temperatures and different phases. The thickness can be confirmed through the uniqueness test—i.e., only one thickness value can fit all states of VO₂—and better sensitivity in the metallic phase was reported. Note that this method is only possible because the IMT of VO₂ does not modify the thin film thickness.

Other new methods could be found to break such correlations, but we emphasize that these correlations between thicknesses and complex refractive indices are very important, as one can easily (and unfortunately) overlook them during the analysis and thus obtain incorrect dispersion datasets.

The imaginary dielectric functions, ε₂, of the insulating films deposited on silicon or silicon oxide substrates typically consist of two broad absorptions in the visible and near-infrared (NIR) spectrum, with center energy varying around 3.5 eV and 1 eV, respectively. As film grows thicker (>100 nm), a peak at ∼2.1 eV becomes more distinguishable. For films on a sapphire substrate, the presence of this absorption peak has been unearthed at much lower thickness (∼30 nm), which could be related to the improved film quality on sapphire even at the early stage of deposition. Each of these absorption peaks finds their physical roots in inter-band transitions between different orbitals (for more details, see, e.g., Ref. 14). Lorentz oscillators have been commonly applied to describe these absorptions.^10,43,49 The energies, amplitude, and broadening of these oscillators are adjusted across the IMT to accommodate optical changes induced by the phase transition. The Lorentz peak at ∼1 eV red-shifts noticeably as temperature rises. It is eventually replaced by a Drude oscillator during the IMT to better represent the metallic absorptions. By monitoring these oscillator parameters through the thermal ramping process, Kakiuchida et al.⁴² provided insights into band structural transition and bandgap changes through the IMT process.

During the IMT of VO₂, interesting regimes of intermediate phases occur, in which the medium becomes a combination of metallic and insulating regions, as shown via scanning near-field infrared microscopy [Fig. 1(e)].⁴⁸ Effective medium theory (EMT) has been applied to derive the optical properties of thin VO₂ films at these intermediate states when dynamic monitoring is not feasible. The method assumes the coexistence of insulating and metallic domains and approximates the refractive index based on the domain volume fractions at intermediate temperatures during the IMT process. The EMT method is potentially valuable in describing gradual transitions, such as VO₂ films on silicon substrates.⁴⁴ However, sharp optical changes in thin films with better quality or long-range order will greatly challenge the effectiveness of EMT models. The demand of drastic changes in oscillator parameters near the percolation transition could lead to failure in regression analysis or non-physical outcomes.

The film microstructure has been shown to strongly impact the optical properties of VO₂ films. Important information on the dynamic structural transformation of VO₂ can be obtained via various techniques such as x-ray diffraction, transmission electron microscopy, pump–probe electron diffraction, and Raman spectroscopy.^31,50–53 Using in situ real-time spectroscopic ellipsometry (RTSE), Motyka et al.⁴⁷ reported changing optical properties and structural variation during film growth and post-deposition treatments. A two-layer model revealed a more disordered/amorphous film with Lorentzian characteristics at an early stage of the deposition or at the film surface. As the film grows thicker, the bulk materials become better ordered with larger grain sizes and stronger metallic conduction, leading to the potential shift of the bandgap to lower energy.

This structural and compositional dependence can also be found in the transition profile that can vary from abrupt to gradual and in the phase transition temperature that can be shifted depending on film quality, grain size, stoichiometry, degree of strain, and impurity concentrations.^43,45 Figure 1(f) shows the phase transition hysteresis loop and the transition temperatures optically determined from the reversible IMT thermal process on two different VO₂ samples. Thin films deposited on sapphire substrates present a sharp and narrow transition in comparison to the large hysteresis measured on films prepared on silicon.⁴³ These findings are consistent with resistivity measurement,^44,47 both resulting from better film quality on c-plane sapphire likely due to a smaller lattice mismatch at the film/substrate interface.

Wan et al.⁴⁴ studied the optical properties of VO₂ at both insulating and metallic phases from the visible to the far-infrared regions, as shown in Figs. 1(g) and 1(h). The complex refractive indices from 2 μm to 11 μm were reported to be less sensitive to the deposition process and film thickness. In that region, Drude losses dominate the metallic phase, while Lorentzian absorptions tapered down toward longer wavelengths at room temperature. Their measurements therefore indicate a spectral window (2 μm–11 μm) in which there are no, or negligible, microstructure- or substrate-dependence in the optical properties of VO₂ [see Figs. 1(g) and 1(h)]. On the other hand, they also noticed a couple of substrate-dependent features: in the metallic state, Lorentz absorption around ∼10 μm was found to be absent for films on sapphire but present for films on silicon, likely due to other polymorphs of vanadium oxide. Moreover, strong vibrational resonances with distinctive substrate-dependence were observed between 17 μm and 25 μm at insulating states. Further investigation on the cause of these absorptions is necessary in tuning VO₂ films for infrared applications.

The crystalline structure of VO₂ theoretically present anisotropy. In most cases though, an isotropic optical model will sufficiently describe VO₂ thin films due to the following reasons. First, the films may consist of polycrystalline domains of different orientations, or different polymorphs of VO_x, resulting in isotropic behaviors on a macroscopic scale. This is typically what happens when VO₂ is deposited on Si or SiO₂.^15,54,55 Second, the optical measurement is generally less sensitive to the out-of-plane refractive index for a very thin absorbing film. Finally, as we have seen before, in some cases, correlations between the thickness and the refractive index for thin absorbing films add uncertainties to the accuracy and reliability of anisotropic analysis. More confidence in identifying anisotropy would be expected in thicker films with better crystallinity and stoichiometry control.

As we have seen in this part, the optical properties of VO₂ as well as the transition behavior are affected by a number of factors mainly linked to the substrate, thickness, and microstructure of the VO₂ thin film. It is therefore crucial to carefully analyze the optical properties of each VO₂ sample to adapt future devices’ design to its specificity.

Photonic integrated circuits (PICs) allow us to exploit the benefits of light for communication and data processing at the micro- and nano-scale. To this end, silicon has become the mainstream technology for novel developments in a wide variety of applications (datacom, telecom, sensing, high-performance computing, etc.).^56–58 However, silicon is not a particularly well-suited material for enabling active functionalities due to its intrinsic properties. In particular, the control of the optical phase and/or optical amplitude based on the plasma dispersion effect usually yields trade-offs among speed of operation, energy consumption, insertion losses, or footprint.⁵⁶ Silicon active devices with switching speeds in the picosecond timescale are possible. However, active lengths in the millimeter range are typically required for enabling large extinction ratios. More compact devices can be achieved by using resonant structures but at the expense of significantly narrowing the optical operation bandwidth. In this context, the hybrid integration of silicon with CMOS-compatible materials featuring unique properties has opened a path to achieve ultra-compact, broadband, and highly efficient guided photonic devices.⁵⁹ Such types of devices are desired to develop advanced PICs with complex functionalities and large-scale integration. As has been mentioned, VO₂ stands out for the ultra-large change in its complex refractive index between the insulating and metallic states. This feature enables hybrid tunable VO₂ waveguides with lengths down to hundreds of nanometers. Moreover, these kinds of waveguides are broadband since the changes in the VO₂ refractive index span a spectral range from the visible to the mid-infrared wavelength regions. The scheme of a commonly reported hybrid VO₂/Si waveguide is depicted in Fig. 2. The latter comprises a silicon waveguide with a patch of VO₂ atop. By triggering the IMT of VO₂ with an external excitation (heat, electric field, or light), the guided mode could experience a change in both optical phase and amplitude.

In the following, we describe recently reported results, organized in three different parts, according to the methods used to trigger the IMT of VO₂.

Several hybrid VO₂/Si waveguide devices, intended to function as amplitude modulators, have been proposed and demonstrated by thermally triggering the IMT.^54,60–67 To this end, the temperature of the chip can be controlled using a Peltier device or local heat into the VO₂ patch can be applied using Joule heating with metallic microheaters. One of the first hybrid VO₂/Si devices [Fig. 3(a)] was demonstrated by Briggs et al.⁵⁴ They showed the capabilities of VO₂ for optical switching at telecom wavelengths with an active length of only 2 μm but with a moderate extinction ratio of 6.5 dB and insertion losses of around 2 dB. For a given polarization of the guided optical mode, there is always a trade-off between the extinction ratio and the insertion losses, and most groups try to enhance the former and reduce the latter by engineering the hybrid VO₂/Si waveguide. Extinction ratios of 16 dB and insertion losses of 3.8 dB have been demonstrated with an optimized 3-μm-long hybrid VO₂/Si waveguide [Fig. 3(b)].⁶⁰ Ultra-short hybrid VO₂/Si waveguides with a length of only 500 nm have also been demonstrated by embedding VO₂ within the waveguide instead of placing it on top [Fig. 3(c)].⁶² In this configuration, extinction ratios with almost 10 dB but relatively high insertion losses of around 6.5 dB have been reported. The optical switching performance can be further improved by engineering the morphology of the VO₂ layer [Fig. 3(d)]. In such a way, insertion losses below 1 dB and extinction ratios above 20 dB with switching times in the microsecond range were demonstrated for a 20-μm-long hybrid VO₂/Si waveguide.⁶³ On the other hand, the integration of VO₂ on a silicon waveguide could also open alternative applications such as tunable polarizers. Sánchez et al. demonstrated a 20-μm-long transverse-electric (TE) pass polarizer exploiting the polarization dependence loss of the hybrid waveguide [Fig. 3(e)].⁶⁴ They showed a rejection of 19 dB for the transverse-magnetic (TM) polarization in the active state together with switching times of few microseconds for a wavelength range between 1540 nm and 1570 nm. The integration of hybrid VO₂/Si waveguides in add-drop ring resonators could also enable 2 × 2 optical switches.⁶¹ Recent experimental results have shown an extinction ratio up to 25 dB with an insertion loss of 1.4 dB by placing a 19-μm-long VO₂/Si waveguide in a ring resonator [Fig. 3(f)].⁶⁶ 

The control of the VO₂ phase transition in a hybrid waveguide has also been demonstrated by applying an electric field between two separated metallic contacts, as shown in Figs. 3(g) and 3(h). Such a scheme is more interesting for future integrated applications and should enable faster switching times compared to the purely thermal heating counterpart as the applied heat can be localized and confined only to the VO₂ patch on the chip. Markov et al. used this approach to investigate the electro-optical switching dynamics in an ultra-short VO₂/Si hybrid waveguide [Fig. 3(g)].⁶⁸ They showed switching times of less than 2 ns for the IMT. The relaxation time to fall back to the insulating state involved a thermal dissipation process making the recovery slower. Nonetheless, by limiting the current to reduce Joule heating, they predicted switching time of the relaxation down to 3 ns but at the expense of a lower extinction ratio. Joushaghani et al. demonstrated a similar electro-optic VO₂/Si switch [Fig. 3(h)].⁶⁹ They achieved a high extinction ratio of 12 dB in a 1-μm-long device with insertion losses of 5 dB. Moreover, by biasing VO₂ near the onset of the IMT, they tested the capabilities of the device as a photodetector and achieved a responsivity in excess of 10 A/W with optical powers lower than 1 μW. Several proposals have also been made for developing hybrid plasmonic modulators.^70–74 However, the experimental demonstration of such devices is still lacking. On the other hand, the overall power consumption highly depends on the VO₂ patch size and the external resistance required to limit the maximum current flowing in the metallic state and avoid damaging the electrical contacts.⁷⁵ Thus, the only reasonable solution to reduce the power consumption of electrically controlled devices with respect to thermal heating approaches is to design very short hybrid waveguides, what could restrict the maximum achievable extinction ratio.

Finally, all-optical switching schemes to control the state of hybrid VO₂/Si waveguides could be the most promising approach. Notably, the switching timescale of the IMT, triggered by optical excitation, has been demonstrated down to the femtosecond and would allow ultra-fast speed with a seamless integration in PICs.^{39,55,76–80} However, such a type of ultra-fast device has not been demonstrated yet. Ryckman et al. demonstrated the first all-optical hybrid VO₂/Si device.⁷⁶ They integrated the hybrid VO₂/Si waveguide in a small ring resonator and induced the IMT of VO₂ by pumping the patch out-of-plane with a pulsed laser in the visible. No switching times were reported, and the optical switching of VO₂ was attributed to be photothermal. Afterward, the same authors investigated the timescale of such devices [Fig. 3(i)].⁵⁵ In this case, they used a pulsed laser of a few nanoseconds to excite the IMT of VO₂. On the one hand, a fluence dependence in the completion of the IMT was observed. For this device, pump fluences above 12.7 pJ/μm² drove VO₂ to its metallic state. On the other hand, the measured IMT timescale was found similar to the pump-pulse width (∼25 ns) with a minimum influence of the fluence. However, the switching time from the metallic to insulating phase (relaxation time) showed a high dependence with the fluence and the VO₂ patch size. By increasing the pump fluence above the threshold up to ∼70 pJ/μm² and enlarging the VO₂ patch from 500 nm to 1 μm, the relaxation time increased from ∼30 ns to ∼3 µs. Thus, the relaxation dynamics were found to depend on the thermal diffusion of the monoclinic phase. Haglund et al. investigated the timescale of hybrid VO₂/Si waveguides [Fig. 3(c)] by illuminating out-of-plane the device with a femtosecond laser acting as a pump and using an in-plane probe laser to record any change in VO₂.⁷⁹ For this case, a 900-nm-long VO₂ patch reportedly showed switching speeds lower than 2 ps for fluences between 50 pJ/μm² and 100 pJ/μm². More recently, sub-ps switching times have also been demonstrated by optimizing the fluences and reducing the volume of VO₂ in the hybrid waveguide.⁸¹ 

While these are promising results, the out-of-plane excitation is not the best approach for integration in PICs, and future all-optical schemes with both the pump and the probe guided within the waveguide would be more desirable. All-optical switching with hybrid waveguides using an in-plane approach has been recently demonstrated.⁸⁰ In this case, SiN instead of Si was used for the hybrid waveguide to handle the pump between 700 nm and 1000 nm and the probe at 1550 nm [Fig. 3(j)]. An extinction ratio of 10 dB was achieved for a 5-μm-long hybrid waveguide with a switching energy as low as 6.4 pJ. However, the timescale was not reported. Parra et al. have recently addressed this question using a similar in-plane pump–probe technique in the telecom wavelength region with a hybrid VO₂/Si waveguide.⁸² Their temporal results suggest a thermal dynamics in which the phase change of VO₂ is thermally triggered and therefore limited to the nano/microsecond range.

In view of this recent work, a remaining important challenge is to find ways to reach the femtosecond timescale of the IMT in integrated devices. Promising directions could be to precisely adjust the excitation source to only trigger the electronic IMT of VO₂ without introducing parasitic heat generation.^83–85 In addition, engineering the thermal environment of devices to efficiently dissipate heat would also greatly help minimizing the relaxation time of the IMT.

Recent years have seen the emergence of optical metasurfaces.^4,5 In these devices, an abrupt phase/amplitude shift is printed on a surface through engineered nano-elements. By spatially arranging such meta-atoms on a substrate, one can design metasurfaces tailored for specific optical functionalities such as lenses, polarizers, retroreflectors, holograms, and perfect absorbers, among many others.^4,86–88 Some of these flat optics devices already surpass the performances of conventional diffractive optics components. As this field is becoming mature with conventional “passive” materials, researchers are now actively seeking means to dynamically modify the properties of these nano-elements to demonstrate actively reconfigurable metasurfaces, what may revolutionize the field of integrated optics. We review in the following different approaches and concepts that exploit VO₂ to dynamically tune the free-space optical response of flat optics devices.

A thin-film material with adjustable complex permittivity can be considered as the simplest form of a metasurface, especially if it presents spatial variations of permittivity. As described in Sec. III, across the transition of VO₂, intermediate states are produced, in which metallic and insulating phases coexist at the nanoscale. These different mixed states can be regarded as naturally disordered metamaterials with tunable optical properties, and in the following, we analyze the different features of these appealing states.

Perfect absorption was achieved by Kats et al. in a system where a thin layer of VO₂ (∼180 nm) was grown on a sapphire substrate [Fig. 4(a)].⁸⁹ At temperatures close to the IMT [Fig. 4(b), ∼343 K], the absorption losses equal the radiative losses and the so-called critical coupling conditions are reached, producing an absorption of 99.75% at a wavelength of λ = 11.6 μm. Given that this near-perfect absorption is dynamically tunable, over the transition temperature range, the reflectivity at λ = 11.6 μm can be largely modulated from 80% to 0.25%. Similarly, Butakov et al. reported broadband tunable reflection and transmission in a tri-layer system (Ge/VO₂/Al₂O₃) in the mid-infrared and demonstrated electrical tuning of such a system.⁹⁰ Furthermore, Rensberg et al. reported that the suppression of reflection can be engineered by depositing an ultrathin layer of VO₂ on epsilon-near-zero substrates such as aluminum-doped zinc oxide (AZO), SiO₂, and ZnO [Figs. 4(e)–4(g)], and tuned by temperature.⁹¹ A minimum of reflectance is found close to the plasma resonance of AZO [Fig. 4(e)] and the reststrahlen band of SiO₂, while the local minimum is absent on ZnO. Conversely, as stated by Kirchhoff’s law, the absorption is closely related to the thermal emission, implying that a good absorber should also be a good thermal emitter. As shown in Figs. 4(c) and 4(d), a 150-nm-thick VO₂ film deposited on sapphire exhibits “perfect” blackbody-like emissivity ∼1 in the vicinity of the IMT (∼74.5 °C) over a wavelength range of 40 cm⁻¹.⁹² 

Utilizing the distinct optical feature of VO₂ between metallic and insulating phases, several optical and radiative thermal devices have been proposed based on un-patterned thin films. A limiting optical diode, in which the phase transition of VO₂ is triggered asymmetrically depending on the direction of incident light [Figs. 4(h) and 4(i)], was designed using a stack comprising a semi-transparent metallic layer, a VO₂ layer, and a transparent substrate.⁹³ In such a device, a backward illumination triggers the IMT, hence resulting in a reduced transmission in that direction, while a forward illumination with the same intensity leaves the VO₂ state in its insulating phase, leading to high transmission. A similar yet different concept imagined exploiting VO₂ to control heat fluxes: a radiative thermal transistor, capable of modulating and amplifying radiative heat transfer in the far-field, was proposed by Joulain et al. [Figs. 4(j) and 4(k)].⁹⁴ By placing VO₂ between two blackbodies having different temperatures, a radiative flux amplification factor α larger than 1 can be achieved in the transition region from 341 K to 345 K driven by the emissivity variations of VO₂ at metallic and dielectric phases.

Engineering the geometric structure or spatial variation of phase through controlling the growth condition and introducing defects is another interesting method to fabricate metasurfaces without relying on etching processes. Using ion irradiation through masks [Figs. 4(l) and 4(m)], Rensberg et al. introduced defects into designated regions of VO₂ and locally changed the transition temperature.^95,96 As a result, a metasurface composed of metallic and insulating phases of VO₂ forms upon heating (T ∼ 60 °C) and vanishes at temperatures away from it (e.g., 30 °C and 80 °C), the reflectance of which then shows engineered switchability and polarization dependence across the transition region between 25 °C and 90 °C. Another original method of creating “natural” metasurfaces can be obtained by controlling the structure and texture of VO₂ through specific strain governed by the substrate nature and orientation, as shown in Figs. 4(o) and 4(p). By using an a-cut (11$2¯$0) sapphire substrate that supports anisotropy growth, Ligmajer et al. grew a layer of self-structured VO₂ nanobeams having widths in the range of 100 nm–200 nm and lengths of 1000 nm–2000 nm.⁹⁷ The measured extinction spectra exhibited a broadband strong polarization dependence in both metallic and insulating phases. Such devices may be used for large-scale modulators with polarization control.

Given the large modifications in the local dielectric environment during the IMT, VO₂ can also be utilized as a tunable substrate to form heterostructures via direct contact with other photonic materials. Folland et al. reported a tunable hyperbolic metasurface device [Fig. 4(q)] by transferring a natural hyperbolic material medium—an isotopically enriched hexagonal boron nitride (hBN)—on top of a VO₂ crystal.⁹⁸ In their study, they demonstrated that the insulating and metallic domains of VO₂ can reflect, transmit, and launch hyperbolic phonon polaritons (HPhPs) at domain boundaries and showed a reconfigurable control of in-plane HPhP propagation. Modulation of the wavelength of HPhPs by a factor of 1.6 was achieved across these domains.

Since the pioneering work of Purcell, we know that the physics of spontaneous light emission is governed both by the quantum mechanical electronic transitions of the emitter and by the optical environment, also known as Local Density of Optical States (LDOS). As VO₂ presents very large modulations of its complex permittivity upon the IMT, it can be exploited to dynamically modify the LDOS and therefore open up interesting means to control the spontaneous emission of quantum emitters in integrated devices. This has been experimentally demonstrated in a multilayer stack comprising thin films of quantum emitters (Er³⁺:Y₂O₃) and VO₂ [see Fig. 5(a)].⁹⁹ The device, comprising a quarter-wavelength phase-change layer located between an emitter layer and a metal mirror, was specifically designed such that the VO₂ IMT can be externally switched optically while also having optimized influence on the LDOS of the emitter layer. Upon switching the VO₂ layer, there is a π phase shift in the effective optical path length, which maximizes the influence of the change in refractive index on the surrounding LDOS. Using this device, combined with the symmetry difference in the polarization of electric dipole (ED) and magnetic dipole (MD) transitions of erbium ions, one can dynamically switch between spectrally distinct ED-dominant and MD-dominant emission by simply changing the state of VO₂. With this concept, a broadband all-optical direct modulation of 1.5 μm emission from erbium ions was shown. Interestingly, such a dynamic optical modulation scheme was experimentally demonstrated to be more than three orders of magnitude faster than the excited state lifetime of the erbium emitters [see Fig. 5(b)], hence enabling fast direct modulation even for long lifetime quantum emitters.⁹⁹ 

Other interesting ideas were further developed from that concept, as described in the following. Very recently, Jha et al. have used a similar configuration, in which VO₂ thin films are integrated in proximity to quantum emitters (hBN) as a means to modulate their LDOS. This tunable LDOS is then used to modulate the emission rate of quantum emitters, which, in turn, enabled resolving both the spatial position of the quantum emitter and its three-dimensional dipole orientation.¹⁰⁰ Szilard et al. have calculated the enhancement of the spontaneous rate of ED and MD emitters in the vicinity of VO₂ layers [see Figs. 5(f) and 5(g)].¹²⁷ They have shown that both ED and MD transition rates can be strongly enhanced, especially in the mid-infrared range at some specific stages of the IMT (in between insulator and metal). They further suggest that the IMT hysteresis could be used as another degree of freedom for dynamic control of the spontaneous emission with a memory effect. Other works studied the potential of VO₂ nanolayers associated with plasmonic antennas [see Figs. 5(c)–5(e)^,101] or hyperbolic metamaterials [Figs. 5(h) and 5(i) ¹⁰²] for further controlling the enhancement of spontaneous emission of quantum emitters.

In the last decade, VO₂ has been largely used as a means to control the resonance of plasmonic scatterers. These devices are usually based on plasmonic resonances in metallic nanostructures, fabricated on top of a VO₂ thin film.^{13,41,89,103,104,128} Some of the earlier works were based on split-ring resonator devices patterned above VO₂ thin films.^13,105 By changing the phase of VO₂, the authors were able to controllably modify the resonant wavelength of the system in the MIR and visible range. Many different shapes and sizes of metallic nano-antennas were reported and experimentally demonstrated to be tuned by the underlying VO₂ thin film, spanning all wavelength regions from the visible to the terahertz range [see some selected examples in Figs. 6(a)–6(h)]. For the vast majority of these works, the VO₂ layer produces an amplitude modulation of the free-space reflection or transmission through the devices. Later on, numerous studies made use of similar plasmonic nano-element (disk, antenna, and slot) arrays for alternative applications such as optically triggered memory devices,¹⁰⁶ tunable color generation,^107,108 optical phase-arrays,¹⁰⁹ switchable polarization rotation,¹¹⁰ and active directional switching of surface plasmon polaritons.¹¹¹ As VO₂ transitions from an insulator to a metal, Butakov et al. exploited this IMT to demonstrate switchable dielectric–plasmonic resonators using directly patterned VO₂ scatterers.¹¹² Alternatively, Muskens et al. exploited the spatially confined hotspots of antennas to locally switch VO₂ thin films. This method helped both reducing the energy consumption and the recovery time of VO₂, hence enabling reversible switching at over two million cycles per second, i.e., much faster than for a VO₂ thin film alone.¹¹³ 

We have seen that VO₂ thin films have been used for about a decade to dynamically tune and switch nanophotonic devices, both for integrated guided wave optics and for free-space optics. In the majority of cases, VO₂ was used as a simple on–off optical switch, driven thermally or optically. Given its natural functionalities, we feel that this material is so far underused and its salient features such as the electrical control, the large hysteresis, and the multilevel intermediate states deserve to be better exploited in nanophotonic devices. We review in this section different works that explore these directions that we find particularly promising.

As described previously, the IMT of VO₂ can be triggered via a large number of different stimuli. Among them, the electrical control of the state of VO₂ appears as the most useful implementation toward real world applications. Indeed, although it is much easier to demonstrate proof-of-concepts in a laboratory using controlled hot plates or lasers, many of the future applications should not rely on such external means of switching. Electrical control, on the other hand, is a widespread technique that is ubiquitous in modern devices. It is however challenging to create functional electrically controlled VO₂ devices for different reasons. The first one is mainly technological: it is not straightforward to integrate electrodes with VO₂, or even to nanopattern it altogether, because this material combines a high-reactivity with most of the wet-etchant chemicals used to process metals and a good resistance to a common dry-etching process. This implies that the technological processes for this platform are not yet mature and require internal developments for each laboratory. The second and most important reason is that a design that would work for thermal or optical switching has little chance to be readily adapted for electrical excitation. This is due to the presence of electrodes that obviously introduce differences in the overall optical properties. The architectures of devices should therefore be completely redesigned to properly take into account the influence of electrodes. This should not be seen as a limitation of the technology but rather as another layer of complexity that has to be thought through. In this section, we therefore review interesting advances in experimental demonstrations of electrically controlled VO₂-based devices for nanophotonics.

One of the first works on the electrical control of VO₂ to tune an integrated optical device was reported by Driscoll et al. in 2009.¹² In this study, the authors demonstrated a frequency-agile metamaterial based on split-ring resonators operating in the THz range [see Fig. 7(a)]. Using a simple planar electrode architecture at controlled temperatures, they were able to electrically induce persistent tuning of the metamaterial’s resonance. Following this pioneering demonstration, a few works reported the electrical control of the VO₂ state on top of silicon-based waveguides, as described in Sec. IV.^68,69 More recently, three different groups reported experimental demonstrations of electrically controlled VO₂-based nanophotonic devices. Liu et al. proposed, in 2016,¹¹⁴ a metal–insulator–metal configuration, comprising a VO₂ thin film sandwiched in between metal antennas and a dielectric spacer, as displayed in Figs. 7(d) and 7(e). Interestingly, the top metallic cross-shaped antennas are patterned as an array that connects them together, hence enabling electrical current to flow through the structure and to electrically trigger the IMT of VO₂. The authors then demonstrated a very large optical reflectance modulation from ∼0% to ∼80% at a wavelength of 3 μm upon electrically switching VO₂. This report was followed a year later (2017) by the experimental demonstration of a near-infrared spectral tuning of metadevices comprising VO₂ nano-elements placed at the feed gap of bow-tie antenna suspended membranes.¹¹⁵ As shown in Fig. 7(b), VO₂ unit cells, about ∼30 nm-wide, are all electrically connected via metallic lines and the spectral tuning is driven by Joule heating. This configuration presents the advantage of minimizing the volume of VO₂, hence reducing thermal mass, energy consumption, and switching times (reported to be in the millisecond range). These two demonstrations were followed in 2019 by the report of an electrically triggered modulation of optical phase in a one-dimensional metasurface array of metal–VO₂–insulator–metal waveguides [see Figs. 7(f) and 7(g)].¹¹⁶ In this study, the electrical control follows the same principle as in the two previous works: the patterned metallic structures on top of VO₂ serve as a means to distribute the flowing current throughout the device and trigger the VO₂ IMT via Joule heating. Using this reflectarray architecture, the authors report a phase modulation as high as 250°, accompanied by spectral tuning and intensity modulation. This is the first demonstration of electrically tunable continuous optical phase modulation using VO₂.

This “trilogy” of reports therefore nicely demonstrates that VO₂-based devices enable the active modulation of the intensity, spectrum, and phase of reflected fields via electrical means. These works therefore pave the way for future efficient electrical control of different aspects of light fields for free-space optics using metasurface-inspired configurations. Future works in that field may focus on improving metrics such as the optical efficiency and the switching time that are so far relatively modest. Designing devices that can separately modulate the amplitude and phase of the reflected/transmitted optical fields via electrical control appear to be very challenging but if successful would unlock such technologies for multifunctional integrated tunable devices.

The non-negligible optical absorption in both states of VO₂ could be a major drawback for most applications as a tunable optical medium. We review here two main strategies to mitigate this issue.

One straightforward solution is to design devices in spectral regions where the absorption is the lowest. As shown in Fig. 1(g), the extinction coefficient of VO₂ in its insulating state slowly decreases at longer wavelengths and reaches a minimum down to k ∼0.07 in the range 2 μm–10 μm. When excited to the metallic state, the extinction coefficient dramatically increases, up to values of k ∼ 5–10. In that range of the mid-IR, there is therefore room to design devices with low losses in one state and large absorption in the other state. This combination of features would be particularly useful for compact on–off modulators with large extinction ratios, both for guided and free-space optics. Looking back at the previously described results, we can indeed observe that the devices showing the highest modulation amplitudes are those designed for this wavelength range [see, e.g., Figs. 4(a), 4(b), and 7(e)] in which reflection modulation from ∼0% to ∼80% were demonstrated at wavelengths of 11 μm and 3 μm, respectively. This wavelength region appears also very promising for integrated guided wave optics, as many recent works demonstrated low-loss mid-IR waveguides on various platforms such as silicon, germanium, or SiGe.¹¹⁷ We foresee that the use of VO₂ for mid-IR guided optics should lead to very interesting devices and large-scale tunable systems in the near future.

Another solution to avoid the large losses in VO₂ is to appropriately engineer the hybrid photonic structure of VO₂. One way of doing so is to use ultra-thin films of VO₂. Indeed, the complex refractive index modulation of VO₂ is so large that a few nanometers of material suffices for active tuning in many devices. As previously mentioned, VO₂ is a complex material whose growing conditions are far from being straightforward. Fabricating ultra-thin films of VO₂ is therefore challenging in itself. Quackenbush et al. reported the successful growth of ultra-thin films (7.5 nm–1 nm) of VO₂ by molecular beam epitaxy.^118,119 Interestingly, they demonstrated that the IMT of VO₂ remains unchanged even for 1-nm-thick layers (i.e., about two unit cells). This important result holds promise for the future use of ultra-thin functional layers of VO₂. Recently, Guo et al. reported the use of atomic layer deposition (ALD) to grow ultrathin layers of VO₂.¹²⁰ With this method, they demonstrated the conformal coating of VO₂ layers on ITO nanorod arrays [see Figs. 8(a) and 8(b)]. Such a growing technique appears promising to seamlessly implement ultrathin layers of VO₂ on nanostructures with an overall low optical absorption.

It was recently shown that VO₂ nanocrystals (VO₂-NCs) implanted in SiO₂ as spherical nanoinclusions [see Fig. 8(c)] provide an alternative path toward low-loss tunable media.¹²¹ By exploiting the VO₂ IMT, it is possible to tune and switch multipolar modes supported by VO₂-NCs in the visible and to gradually produce a plasmonic mode in the NIR whose intensity is directly controlled by the VO₂ state. It was shown that the complex refractive index of such an effective medium (a slab of VO₂ nanospheres embedded in SiO₂) presents distinct optical tunability compared to unpatterned VO₂. By adjusting the VO₂-NC size, the effective medium can be designed to have a large refractive-index tunability without inducing modulation of the extinction coefficient at specific wavelengths. This zero-induced-extinction refractive index tuning opens up new regimes of record large figure of merit (Δn/Δk) and designer optical tunability, unattainable with conventional unpatterned PCM layers [see Fig. 8(c)], and suggests a new practical direction to produce low-loss tunable optical metamaterials.

As mentioned in Sec. III, VO₂ is known to present a broad hysteresis in its IMT cycle. This hysteresis can be, to some extent, engineered via strain or doping.^14,15,35,122 However, there is a surprisingly low number of papers that actually use this functionality in devices. Such an effect can be used to demonstrate memory effects. Lei et al. exploited this hysteresis to demonstrate an all-optical memory effect in hybrid plasmonic nanostructures.¹⁰⁶ More recently, Fan et al. reported an optoelectronic memory device with electrical writing and optical reading using epitaxial VO₂ thin films grown on GaN.¹²³ The memory effect has also been demonstrated in hybrid VO₂/Si waveguides.⁶³ Such memory devices are still volatile in nature though, as their “memory” state only lasts several microseconds at best. However, we describe in the following three recent works that elegantly exploited the hysteretic behavior of VO₂ to demonstrate all-optical non-volatile nanophotonic devices.

The VO₂-NCs described in Sec. VI B typically present an ultra-broad hysteresis behavior, with an IMT occurring at ∼80 °C upon heating and a relaxation to the insulating state at ∼25 °C upon cooling [see Fig. 9(a)]. Hence, by keeping the sample at temperatures as low as ∼30 °C, it is possible to maintain a persistent metallic state in VO₂-NCs. Jostmeier et al. exploited that functionality to optically imprint photonic elements onto an unpatterned VO₂-NC platform.¹²⁴ Using a visible laser scanned at specific points of the sample, they locally switch VO₂-NCs and define patterns such as gratings and zone plates, as shown in Fig. 9(a). This original technique appears as a very flexible way to fabricate reconfigurable photonic devices as they can later be erased by simply cooling down the sample to room temperature.

Two recent works used the same principle to optically imprint arbitrary reconfigurable patterns via locally switching the state of VO₂ thin films. The first one, reported in 2018 by Dong et al.,¹²⁵ demonstrated the dynamic writing and erasing of arbitrary patterns and reconfigurable photonic devices such as beam-steerers, linear polarizers, and concentric-ring gratings at a wavelength of 10.6 μm [see Fig. 9(b)]. Each of the patterns, fabricated using a 532 nm laser, can be “stored” in the VO₂ film by maintaining the sample at temperatures around the IMT, i.e., ∼60 °C. They coined this reconfigurable VO₂-based platform a “programmable metacanvas.” A following recent work used the same method to demonstrate a spatially resolved control of thermal emission in a large wavelength range of the mid-IR (8 μm–14 μm).¹²⁶ As shown in Fig. 9(c), they exploited both the hysteresis and the intermediate phases of VO₂ to write non-volatile multilevel states in the thin layer, each producing a different level of thermal emission.

These three works demonstrate very promising ways to fully exploit VO₂ for its hysteretic behavior and large optical modulation in the mid-IR. One may find the need for temperature control to be inconvenient, but this could alternatively be seen as a very practical method to easily erase and reconfigure devices. These demonstrations may pave the way for future real-time adaptive optical systems.

Finally, a very important open question remains on whether ultra-fast switching times, below the picosecond range, are actually feasible in practical hybrid integrated devices. All-optical switching seems to be the most promising approach based on previous experiments on non-integrated devices (see, for instance, Ref. 39). However, ultra-fast all-optical switching on hybrid integrated devices has only been very recently demonstrated.⁸¹ The key point is to control and minimize the thermal component associated with the VO₂ phase transition without penalizing the optical switching performance.

All in all, progress toward a better use and integration of VO₂ in nanophotonic devices will go in parallel with progress in better understanding and control of the complex physics of VO₂.

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

S.C. acknowledges funding from the French National Research Agency (ANR) under the project SNAPSHOT (Grant No. ANR-16-CE24-0004). Z.Z. and S.R. acknowledge AFOSR (Grant No. FA9550-18-1-0250) for support. P.S. acknowledges Generalitat Valenciana (No. PROMETEO/2019/123), Ministerio de Economía y Competitividad (MINECO/FEDER, UE) (No. TEC2016-76849) and Ministerio de Ciencia e Innovacion (MINECO/FEDER, UE) (No. PID2019-111460GB-I00). J.P. acknowledges the Ministerio de Ciencia, Innovación y Universidades for his grant (Grant No. FPU17/04224). We thank M. Kats for fruitful discussions and feedback on the manuscript.