1. INTRODUCTION

Moore’s law has governed the advancement of semiconductor devices since the 1950s, leading to a reduction in size, cost, and power consumption of integrated circuits. While this “law” has served the semiconductor industry very well since its inception, it has become clear that it cannot continue indefinitely with current technologies. In fact, such scaling is not unique to CMOS-integrated circuits and can be extended to previous computational technologies such as vacuum tubes (see Fig. 1) [1]. Here, the integrated circuit is seen as the fifth-generation technology to continue the trend in computational improvement. Such a depiction of technological scaling begs the question, “If the integrated circuit is near the end of its life, what is the next technology that will continue the trend?” This consideration does not include short-term improvement such as a movement to GaAs-based MOS, but is based on an entirely new technological architecture.

 

Fig. 1. Graph illustrating the scaling of computational technologies over time. Integrated circuits represent the fifth generation of scaling. To continue the scaling into the future a new technology platform should be investigated. (Adapted from Moore’s Law: The Fifth Paradigm [1]).

Download Full Size | PDF

While there are many potential technologies vying to be the sixth-generation such as quantum [2] and 2D materials [3,4], one of the most promising is photonics. Optics has been used for large-scale and high-speed network connectivity since the 1970s and has become the backbone for high-definition television and broadband Internet. Currently, 93% of American households (291.9 million people) are equipped to use high-speed cable Internet and television enabled by fiber optic networks, and 6% of Americans (18.8 million people) receive their services over an all-fiber connection [5]. As a result, peak Internet speeds have seen an exponential increase from 16 Mbps in 2007 to greater than 500 Mbps in 2013 with no signs of slowing.

The success of the fiber-optics industry has led many institutions to investigate the potential of integrated optical devices. Such photonic integrated circuits (PIC) not only offer the potential for a dramatic increase in the operational bandwidth, but also reduced cross talk between devices, immunity to electrical noise, and lower power dissipation [6]. In fact, several commercial integrated optical devices are available today such as Intel’s Thunderbolt and IBM’s signal routing chip [7,8]. These electro-optic devices are, and will continue to be, an extremely important area of integrated optics, since they represent a stepping stone between technological hierarchies. These unique devices enabled many of the benefits of photonics to be utilized today, without the need for a complete redesign of an all-optical system. However, photonics is not without its own problems.

In the past, the switch to a new technological base has historically resulted in a temporary decrease in the overall density of the resulting device compared with its predecessor; photonics is slightly different than the previous cases, however. The minimal size of devices, which can efficiently guide and modulate light is limited by the diffraction limit of light. Even under the best-case scenario, where a high index material is used, the size is limited to roughly 200 nm for the telecommunication wavelengths [6]. More compact designs have been demonstrated, though many of these designs do not utilize realistic or CMOS-compatible fabrication processes, limiting their applicability to the laboratory. While the lack of density may not be of great concern for near-term devices, this fundamental upper limit to the integration of photonic devices is a significant drawback for a long-term “sixth-generation” technology. These issues are one of the main reasons scientists began exploring how to break the diffraction limit of light, where yet another optical technology is waiting to be adopted into the realm of integrated devices: metal nanophotonics.

More recently, the subdiscipline of nanophotonics (or plasmonics) has emerged. Metallic nanophotonics presents the ability to confine and guide light by coupling with the momentum of free electrons at a metal–dielectric interface, called surface plasmon polaritons (SPPs) [9 –18]. The introduction of the metallic layer allows for the confinement of light, even down to the nanoscale. Of course, this gain in integration comes at a price. For metal nanophotonics, the advantage of enhanced light–matter interaction leads to unavoidable losses that vary with the modal confinement of the structure [19]. Consequently, while metal nanophotonics is capable of confining light down to 10’s of nanometers, the resulting losses make such structures extremely difficult to work with in current devices, although there are researchers studying the potential of compensating these losses with gain [20,21].

However, unlike photonics, metal nanophotonics does not have a fundamental limit to the size of confinement and, consequently, the integration density of devices. In theory, light can be confined entirely to the surface, reaching a nearly infinite wave vector. In practice, the confinement or maximum wave vector is limited by the losses in the structure. However, unlike the diffraction limit for photonic devices, this limitation is a matter of optimization since the losses in metals can be engineered. While losses cannot be removed from metals entirely while maintaining Kramers–Kronig relations, it is within the realm of science to engineer the losses in metals to be drastically reduced or eliminated over special frequency windows [22,23]. In this situation, the primary drawback of metal nanophotonics would be eliminated, and it could be a clear frontrunner for on-chip applications as well as in other fields.

While such a spectrally lossless material is theoretically achievable [22], it has yet to be demonstrated. However, there is truly only one component where losses cannot be tolerated (the waveguide), where energy must travel over a distance. In the comparison of conventional photonics and metal nanophotonics as a potential “sixth-generation” technology, one should consider that the field of photonics is significantly more mature than the field of nanophotonics. In fact, many photonic waveguide structures were not much better than today’s low-loss metal nanophotonic waveguides. Figure 2 shows a comparison of waveguide performance through time of the respective fields, plotting a figure of merit (FoM) defined as the inverse of the mode size times the attenuation ($\mathrm{FoM}=1/\delta \alpha$) [24].

 

Fig. 2. Performance figure of merit for integrated photonic and nanophotonic technologies in time. It can be seen that nanophotonics have reached a higher level of performance much quicker than photonics and are becoming competitive. Continual improvement in materials and geometries is expected to further increase the performance of both technologies, making them even more attractive for high-performance applications. For this comparison, photonics is represented by dielectric waveguides (using silicon or silicon dioxide for example), and nanophotonics is represented by insulator-metal-insulator strip plasmonic waveguides in a dielectric (using gold or silver, for example). References are listed forward in time by increasing performance.

Photonics: [27,29,29,28,31,32,33,34,67,68, and 69]

Nanophotonics: [70,71,72,72,73,74,39, and 39].

Download Full Size | PDF

Early studies of integrated photonic devices focused on the well-developed Corning optical glass, simply using silicon as a substrate [25,26]. Such structures were able to achieve very high performance with attenuation less than $0.1\mathrm{dB}/\mathrm{cm}$. Later, the development of thermal evaporation techniques for oxidizing silicon wafers enabled the direct formation of ${\mathrm{SiO}}_2$ waveguides on silicon [27,28]. However, these waveguides suffered from leakage into the higher index silicon substrate and had to be quite large to avoid this problem. It was not until the development of the separation by implantation of oxygen, now more commonly referred to as silicon-on-insulator (SOI) that light guiding in silicon was feasible. Since this time, many developments have been made to push guiding in silicon rib or ridge waveguides to the sub $1\mathrm{dB}/\mathrm{cm}$ mark, while achieving sizes of $\sim 1{\mathrm{\mu m}}^2$ [29 –34]. For more information on current and early-integrated photonic devices, please see the following reviews [26,35 –37]. Similarly, the field of low-loss metal nanophotonic structures started with humble beginnings, using gold and silver as the metallic core (note, only insulator-metal-insulator (IMI) strip waveguides are shown in Fig. 2) [38]. Through continued fabrication improvements and optimization, metal nanophotonics has also reached the level of sub dB/cm propagation [39]. However, because the size of the mode is larger than corresponding photonic structures, $\sim 10{\mathrm{\mu m}}^2$, the FoM is reduced for these structures. Varying geometries have recently been utilized in attempts to reduce the mode size while maintaining a low propagation loss (not shown in graph) [40 –42]. Consequently, hybrid and dielectric loaded waveguide structures have pushed the FoM to values greater than 1000 for $\sim 1{\mathrm{\mu m}}^2$ mode sizes [40,41]. As was seen with photonic waveguides, continued advancements in materials, fabrication, and structural design are likely to continue the improvement of metal nanophotonic waveguides into the future.

Unlike waveguides, nanophotonic dynamic devices are less affected by the loss in metals due to their ability to achieve ultracompact sizes such that the propagation loss is negligible. As a result, nanophotonic dynamic devices have become the leaders in compact size and performance [43 –46]. Thus, one would expect a realistic solution to be to use low-loss, compact silicon photonic waveguides to carry signals and ultracompact plasmonic active devices to switch and modulate the signals. Indeed, this seems a logical approach and one that many researchers are investigating [47 –51]. However, the inherent discrepancy in the essence of the energy profile (i.e., a photonic mode is Gaussian and a plasmonic mode is exponential [11]) and modal size is problematic and limits the coupling efficiency between the structures. While current coupling designs can achieve efficiencies $\sim 70\%$ per facet for ultracompact devices [52], such high-efficiency designs have largely been relegated to theoretical studies due to the intensive fabrication complexity and precision required. Coupling designs using nanoantennas to funnel light also have been reported and can achieve relatively high efficiencies under experimental conditions [53 –56]. However, these designs are best suited to couple light into the ultracompact waveguides from out of plane, a situation not likely to be experienced for integrated devices. The exception to this rule is low-loss strip plasmonic waveguides, which can achieve very high coupling efficiency with traditional photonic modes due to their dispersion being nearly on the light-line [10,57,58]. As a result, utilizing low-loss metal nanophotonic waveguides or traditional photonic waveguides to interface with ultracompact active devices is an important problem for integration, although metal nanophotonic waveguides may offer intrinsic benefits.

In addition, metal nanophotonic interconnects provide an inherent ability to efficiently couple with ultracompact active devices [44,59 –62], an extreme sensitivity to the metallic–dielectric interface [16,63,64], support for both electrical and optical signals [65,66], perfect polarization purity (SPPs only exist for TM polarization), and increased nonlinearities due to the field enhancement near the metallic–dielectric surface [75]. Because of these collective advantages for metal nanophotonic waveguides and the high performance of active devices, metal nanophotonics deserve a spot at the integrated optics table [76]. However, both photonic and electronic devices also offer significant advantages for on-chip situations that will outweigh nanophotonics in many applications. As such, both technologies will continue to play an important role in future integrated computational systems.

The field of integrated computation devices is at a crossroads, and the task before researchers in the foreseeable future is to effectively utilize the potential of the three available technologies (electronics, photonics, and plasmonics) so that they may compensate the weaknesses of each other (i.e., the losses in plasmonics with photonics, the size of photonics with plasmonics, and the interface to the outside world with electronics). In the remainder of this review, we focus on advances related to nanophotonic technologies, beginning with a review of the recent movement in nanophotonics to incorporate new materials. Following, we summarize integrated devices using traditional metallic components, as well as new materials, and provide several graphs illustrating the normalized performance of referenced works. Finally, we will conclude with a brief outlook and summary of the problems in the field of integrated nanophotonics.

2. ALTERNATIVE AND CMOS COMPATIBLE MATERIALS

The field of integrated nanophotonics contains many different devices, which all have separate requirements that should be met for optimal performance and integration. Initial studies of plasmonic components almost exclusively used noble metals due to their low ohmic losses and wide availability. However, in recent years it has become abundantly clear that the noble metals cannot address all of the application-specific requirements for the wide range of integrated nanophotonic devices [19]. For instance, the ability to alter the optical properties of gold and silver is limited, greatly restricting their potential applications for dynamic or active devices [77]. Also, due to their high surface energy, they have well-documented problems forming ultrathin uniform layers (ideal for many nanophotonic applications), tending to form islands below thicknesses of $\sim 10–15\mathrm{nm}$. While techniques such as wetting have been demonstrated, and there’s even been a report of epitaxial silver enabling ultrathin uniform films, such fabrication is still a significant challenge for the community at large [78 –80]. Finally, these materials are not compatible with standard semiconductor processing techniques, forming a proverbial wall for noble metal based structures in on-chip nanophotonic devices.

As a means to address the wide range of requirements for nanophotonic devices, the exploration of new materials has received significant attention in recent years [81 –84]. In this time, requirements such as low loss, high-quality thin films, adjustable optical properties, dynamic tuning of properties, chemical stability, and CMOS compatibility have become driving factors in the search for alternatives to silver and gold. Figure 3 illustrates the explosion of interest in new plasmonic materials, where more than half of the period table has been utilized. A similar trend has been noted in the semiconductor industry, as well as in the early days of integrated photonics devices [36], as new metallic components and high-k dielectrics have been introduced to improve the stability and performance of highly integrated devices [85 –88]. Silver and gold have also benefited from research into alternative materials, with recent work alloying them to improve various parameters of the materials [89 –91]. Clearly, by limiting ourselves only to the traditional metallic components, there is a significant range of new applications and performance that would have gone missing. Consequently, new materials such as yttrium hydride (light blue in Fig. 3) in the near-infrared [92], silicides (green in Fig. 3) in the near and mid-infrared [93 –95], the germanides (light red in Fig. 3) in the mid- and far-infrared [96 –99], the III–V semiconductors in the near- and mid-infrared (red in Fig. 3) [100,101], and graphene in the mid- and far-infrared [102,103] have enabled new and exciting research into the field of nanophotonics and beyond. As uses for additional materials are discovered and current materials are further optimized, the impact of new materials will only continue to grow. For more detailed information on these and other materials, see the recent review by Naik et al. [83] and references therein, as well as the recent review of metallic nanoparticles by Guler et al., which discusses similar trends for localized surface plasmon applications [104].

 

Fig. 3. Graphical illustration of the explosion of new materials research in the area of nanophotonics. In the past, the traditional metals were the only materials used, but now nearly half of the periodic table has been utilized for nanophotonic applications in various spectral ranges.

Download Full Size | PDF

There are two main approaches for obtaining new plasmonic materials in the visible and near-infrared spectral ranges: to dilute metals and to heavily dope semiconductors. By adding another material to a metal, i.e., nitrogen to titanium, the carrier concentration can be effectively controlled [82,105,106]. Thus, to reduce the losses due to scattering at longer wavelengths, the carrier concentration can be reduced to meet a desired specification. Alternatively, semiconductors can be doped to high levels ($\sim {10}^{20}{\mathrm{cm}}^{-3}$), providing them with plasmonic properties in the visible and near-infrared [101,107,108]. While many new plasmonic materials have been suggested using these principles, they are not all particularly well suited for applications in the field of nanophotonics. For instance, materials such as the alkali metals, copper, and aluminum possess high carrier concentrations and a strong plasmonic response, but suffer from chemical instability [106,109,110]. Additionally, the optical properties of copper and aluminum still cannot be adjusted, and they require a thin diffusion barrier layer [111]. Even still, they are popular materials to use in fully CMOS-compatible designs [40,112,113].

However, two classes of materials are worth mentioning individually. The first is the transition metal nitrides (TMN), depicted in orange in Fig. 3. The transition metal nitride family is a group of ceramic materials with a wide range of properties, but they are all able to withstand high temperatures $>2000°\mathrm{C}$ (refractory), are chemically stable, and can be grown epitaxially on materials such as magnesium oxide (MgO), c-sapphire, aluminum scandium nitride, and silicon [105,114]. Additionally, they are nonstoichiometric, which allows their optical properties to be tuned by the control of deposition parameters such as temperature, pressure, and gas ratios [105]. These properties make this class of materials attractive for many applications. Niobium nitride (NbN) has received attention for its superconducting properties [115 –117]. Titanium nitride (TiN), zirconium nitride (ZrN), hafnium nitride (HfN), and tantalum nitride (TaN) all exhibit plasmonic properties in the visible and near-infrared spectral ranges that mimic those of gold [118]. Additionally, TiN and TaN are currently used in CMOS processes [86,119,120]. Scandium nitride is a highly degenerate semiconductor that can be alloyed with aluminum to form a dielectric in the visible and NIR range, which can match the lattice constant and crystal structure of other TMNs, allowing for single crystal growth of metals and dielectrics [121]. In combination with the metallic nitrides, ScN and AlScN enable a full suite of tailorable materials, all within the nitride family. The TMNs thereby enable a good portion of our material wish list: low-loss, high-quality thin films, adjustable optical properties, and CMOS compatibility.

The second is a family of II-VI semiconductors within the group of oxides, depicted in blue in Fig. 3, more commonly known as transparent conducting oxides (TCOs). These materials are all able to sustain high doping densities ($>{10}^{19}{\mathrm{cm}}^{-3}$), but tin oxide (SnO), indium-doped tin oxide (ITO), zinc oxide (ZnO), gallium-doped zinc oxide (GZO), and aluminum-doped zinc oxide (AZO) have received most of the attention due to their ability to support extremely high doping densities ($\sim {10}^{21}{\mathrm{cm}}^{-3}$), enabling metallic properties in the near infrared [108,118,122,123]. Additionally, because these materials achieve their metallic properties from doping, their properties can be tuned by varying the doping density during deposition or through oxygen annealing [124 –126]. These properties have enabled many outstanding applications in tactile displays, integrated electronics, and photovoltaics, and the interest has led to the development of high-quality thin films for various fabrication processes such as sputtering and pulsed laser deposition [127 –133]. Additionally, the TCOs can be electrically or optically tuned, making them suitable for dynamic materials in modulators, metamaterials, and other nanophotonic applications [45,59,60,61,134]. In particular, the metal–dielectric transition wavelength for TCOs is in the range of 1200–1600 nm, enabling the potential for a dynamic dielectric-metal phase change at telecommunication wavelengths. Also, due to their NIR crossover wavelength and large bandgap energy, ITO, GZO, and AZO have very little loss in at the telecommunication wavelengths ($\mathrm{Im}\{\epsilon \}\sim 0.1$), making them suitable for high-efficiency nanophotonic applications [108,118,122,125]. Outside of the TCOs, vanadium dioxide (${\mathrm{VO}}_2$) has seen applications toward nanophotonic devices [135,136]. ${\mathrm{VO}}_2$ is a dielectric in the visible and NIR regions, but, when heated, it undergoes an abrupt phase change that causes it to become a strongly absorbing dielectric. As a result, ${\mathrm{VO}}_2$ has been used in integrated modulators and memory devices [136]. The TCO family of materials also meets many requirements of our alternative material wish list: low loss, high-quality thin films, adjustable optical properties, dynamic tuning, and CMOS compatibility.

3. METAL PLASMONICS

Before we begin the review of various plasmonic waveguides and modulator structures, it is good to briefly review the physics of plasmonics and their basic structures. For a more thorough analysis of plasmons, we direct the reader to the following books [9 –11,137].

Surface plasmons (SPs) are the coupling of electromagnetic energy to the motion of free electrons at the interface of a dielectric and a metal. These oscillations can exist in a closed environment, where they do not propagation, called SPs, or can propagation along the interface, called SPPs. While both situations are important in their own right, we will focus on propagating waves for means of integrated devices. As a starting point, we can investigate the properties of SPPs by considering a lower half-space of metal ($\mathrm{Re}\{{\epsilon }_m\}<0$) and an upper half-space of dielectric ($\mathrm{Re}\{{\epsilon }_d\}>0)$. By solving the wave equation for this system and applying the appropriate boundary conditions, we find the wave vector of the SPP is given by the formula

Multilayer structures of IMI or metal-insulator-metal (MIM) correct the problem of the SI structure, allowing one to tailor the properties of the SPP by altering the geometrical parameters (see Fig. 4 [10,11,139 –141]).

 

Fig. 4. Depiction of the three basic plasmonic structures. Left: single interface with a half-space of metal and a half-space of dielectric. Middle: insulator-metal-insulator where the metal is confined to a thin strip embedded in a uniform dielectric medium. Right: metal-insulator-metal where a thin strip of dielectric is imbedded in a uniform metallic medium.

Download Full Size | PDF

For the IMI geometry, the formulas for the wave vector are modified to

The IMI and MIM geometries enhance the two characteristics of plasmonic waveguides important for integrated devices: propagation length and modal confinement, respectively. The IMI geometry achieves long-range propagation at the expense of a smaller $k_{\mathrm{SPP}}$ and reduced confinement. Conversely, the MIM geometry $k_{\mathrm{SPP}}$ can be many times larger than for SI SPPs, enabling deeply subwavelength confinement at the expense of increased propagation loss.

4. NANOPHOTONIC WAVEGUIDES

Waveguides are the central component of a nanophotonic system, carrying information and providing a backbone for additional passive and active devices. As such, they have been widely studied, with a broad range of geometries and performance having been demonstrated. Traditionally, three categories exist into which such devices can be placed: SI, IMI, and MIM. While there is still research being conducted in the realm of SI plasmons, traditional SI waveguides and other 1D structures (as depicted in Fig. 4) have largely been set aside for on-chip applications due to the lack of tunable properties, horizontal confinement, and large propagation losses [see Eq. (1)] [58]. However, the early theoretical and experimental studies of 1D structures were instrumental in providing the groundwork for future designs. In addition, the 1D layout is still of fundamental importance for the preliminary analysis of the optical properties of novel plasmonic materials. For these reasons, we will briefly review them.

The work to improve the performance of plasmonic waveguides has led to a myriad of 2D structures, which introduce various alterations in the geometry and surroundings to provide additional design parameters with which the loss-confinement trade-off can be further optimized. In this review, we consider IMI waveguides consisting of 1D symmetric thin films [141] and 2D strip geometries [38], MIM waveguides consisting of the 1D nanowire [140] and 2D wedge [142], trench [143], gap [144], and channel geometries [145], dielectric loaded waveguides (DLW) [146], and hybrid plasmonic photonic waveguides (HPPW) [147]. Examples of considered waveguides geometries are shown in Fig. 5, although the referenced structures may vary. In the following sections, we will review the history and recent advances of metal nanophotonic waveguides, first with designs using noble metals and finally for designs using alternative and CMOS-compatible materials. In addition, we present a performance estimate for several new plasmonic materials that have yet to be studied experimentally. We conclude with a graph of numerous experimental and theoretical results, so that the many works can be readily compared and the state of the field can be assessed.

 

Fig. 5. Graphical depictions of the most widely used plasmonic waveguide geometries.

Download Full Size | PDF

A. Noble Metal Waveguides

Through many of the initial studies of SPP waveguides, gold and silver were the primary materials of choice for plasmonic devices. In this section, we will review the previous advances in waveguides which use noble metals as their base, including IMI, MIM, DLW, and HPPW geometries. While other devices have been suggested for waveguiding such as chains of nanoparticles or plasmonic crystals, due to their extremely high propagation losses or fabrication complexity, they are generally no longer considered the leaders for on-chip nanophotonic applications [148 –150].

1. Insulator-Metal-Insulator

Thin film IMI structures [Fig. 4(b)] that support SPPs were first studied with prism coupling [139]. While Kretschmann, Otto, and others demonstrated theoretical studies and experimental demonstrations of coupling into plasmonic modes, Kovacs was instrumental in uncovering many of the important features of the technologically important LR-SPP mode [151 –155]. One of the first experimental observations of the LR-SPP with measured attenuation was performed by Kuwamura et al. in 1983, where they achieved a propagation length of $\sim 100\mathrm{\mu m}$ for a 14.6 nm thick silver film [156]. This result was $10\times$ longer than the SI SPP propagation length. While these 1D studies were largely not useful for waveguiding applications due to the large propagation losses and lack of horizontal confinement, they illustrated the unique and important properties of the LR-SPP that would propel future studies of 2D structures.

Soon after, the 2D metal strip [Fig. 5(a)] became a popular structure of investigation for its even lower attenuation, ability for end-fire coupling, and control of lateral confinement [57,58]. Berini was instrumental in laying the theoretical framework of the 2D symmetric metal strip waveguide studying mode propagation, the loss-confinement trade-off, defining figures of merit, and the feasibility of end-fire coupling and data transmission [24,157]. Charbonneau et al. were the first to demonstrate propagation of the LR-SPP on 20 nm thick 8 μm wide gold strips excited by end-fire coupling from standard single mode fiber. They achieved a propagation length of 4.8 mm ($\alpha =0.9\mathrm{dB}/\mathrm{mm}$) at $\lambda =1.55\mathrm{\mu m}$ [38]. Over the next several years, 2D IMI metal strip structures improved upon the attenuation, demonstrating propagation lengths of 7 mm ($\alpha =0.62\mathrm{dB}/\mathrm{mm}$) for 10 nm thick 8 μm wide gold strips at $\lambda =1.55\mathrm{\mu m}$ to more than 100 mm ($\alpha =0.03\mathrm{dB}/\mathrm{mm}$) for 10 nm thick silver 2 μm wide silver strips at $\lambda =1.31\mathrm{\mu m}$ [39,71].

With low-loss propagation achieved and favorable fabrication requirements, 2D strip waveguides began to receive attention for their application toward integrated optical devices as well as in other fields such as sensors and quantum optics [63,158 –160]. Consequently, the demonstration of passive plasmonic devices as well as standard telecommunication studies received significant attention. The work by Ju et al. demonstrated transmission speeds for LR-SPP waveguides in excess of 10 Gbps for gold waveguides at $\lambda =1.55\mathrm{\mu m}$ with a propagation length of 30 mm ($\alpha =0.15\mathrm{dB}/\mathrm{mm}$) [161]. However, as initially discussed by Berini, these structures suffer from low modal confinement on the order of (or greater than) the wavelength, which presents a sizeable drawback for integrated devices (see [58] and references therein for more information).

2. Metal–Insulator–Metal

MIM structures primarily focused on achieving deeply subwavelength light confinement, a feature required for high-density optical interconnects. Initial works theoretically studied metallic nanowires [Fig. 5(b)], which demonstrated the ability to confine light to the scale of the nanowire radius as well as planar MIM [Fig. 4(c)] [140]. However, coupling light into these high-confinement structures is no trivial task due to the large mode mismatch, relegating studies primarily to the theoretical regime. However, several experimental works were successful in exciting to SPPs on metal nanowires using light scattered from the end of the nanowire, coupled from a nearby nanoparticle, or from an integrated photonic waveguide, although these studies were 10 years after the Takahara paper [162 –164]. The work by Sanders et al. coupled light through edge scattering on a silver nanowire, attaining a propagation length of 3 μm ($\alpha =1.4\mathrm{dB}/\mathrm{\mu m}$) at $\lambda =830\mathrm{nm}$ on silver nanowires with a mode size on the order of 70 nm (the nanowire diameter) [162]. Shortly following, the work by Pyayt et al. demonstrated coupling through a photonic waveguide with a propagation length of 9.1 μm ($\alpha =0.5\mathrm{dB}/\mathrm{\mu m}$) at $\lambda =628\mathrm{nm}$, enabling an effective on-chip option for coupling [164].

Later, several 2D variations of the MIM geometry arose such as the wedge, channel, slot, and gap geometries [Figs. 5(c)–5(f), respectively]. As with their 1D cousins, these structures are able to achieve subwavelength confinement. Theoretical studies of the properties and dispersion of wedge plasmonic modes were first investigated by Dobrzynski and Maradudin in 1972 for the electrostatic case and expanded to include time variation by Boardman et al. in 1981 [142,165]. Both the wedge and channel geometries were theoretically investigated by Moreno et al. for gold structures and shown to achieve a 40 μm propagation length ($\alpha =0.1\mathrm{dB}/\mathrm{\mu m}$) with a mode size of 0.37 μm and a propagation length of 120 μm ($\alpha =0.1\mathrm{dB}/\mathrm{\mu m}$) with a mode size of 0.6 μm at $\lambda =1.55\mathrm{\mu m}$, respectively [166]. These geometries became important for their ability to achieve high confinement while maintaining a modest propagation length. In 2005, Pile et al. demonstrated SPP propagation on silver wedge waveguides with a propagation length of 1 μm ($\alpha =3.4\mathrm{dB}/\mathrm{\mu m}$) and a mode size of 0.3 μm at $\lambda =633\mathrm{nm}$ [167]. Shortly thereafter, Boltasseva et al. experimentally demonstrated gold plasmonic wedge waveguides at $\lambda =1.55\mathrm{\mu m}$, achieving a propagation length of 120 μm ($\alpha =0.04\mathrm{dB}/\mathrm{\mu m}$) with a mode size of 1.3 μm [168]. The seminal work by Novikov and Maradudin in 2002 laid the theoretical framework for channel plasmon structures, investigating the dispersion relations for bound propagating modes [145]. Later, Bozhevolnyi demonstrated straight waveguides, along several passive components in gold, achieving a propagation length of 100 μm ($\alpha =0.4\mathrm{dB}/\mathrm{\mu m}$) for a mode size of 1.1 μm at $\lambda =1.55\mathrm{\mu m}$ [169]. Shortly thereafter, Volkov et al. demonstrated wavelength selective devices in gold channel waveguides [170]. It also is important to mention that an additional geometry called trench waveguides also has been proposed. These waveguides are essentially channel waveguides with a groove angle of 90°, and their performance is largely similar to the V-groove channel waveguides as mentioned above [143].

The gap and slot plasmonic structures also have received significant attention in the research community. Tanaka et al. theoretically investigated gap plasmonic waveguides and passive circuits discussing modal properties, propagation losses, and passive devices in silver films [144,171]. Bozhevolnyi and Jung simulated gap plasmonic structures in gold, which were found to achieve propagation lengths of 30 μm ($\alpha =0.14\mathrm{dB}/\mathrm{\mu m}$) for a gap width of 200 nm at $\lambda =1.55\mathrm{\mu m}$ [143]. In 2005, Pile et al. were the first to experimentally demonstrate gap plasmonic waveguides in silver and theoretically predicted a propagation length of 1.2 μm ($\alpha =3.62\mathrm{dB}/\mathrm{\mu m}$) for a 25 nm square gap at $\lambda =632.8\mathrm{nm}$ [172]. However, the slot plasmonic waveguide is typically easier to fabricate from a thin metal film and is more suitable for integration. Therefore, it has seen the bulk of the research associated with integrated MIM plasmonic devices. The theory for modal propagation and dispersion in the slot waveguide reaches back to radio-frequency slot waveguides, and, in 2005, Veronis and Fan investigated bound modes and their dispersion specifically related to plasmonics [173]. Unlike other MIM structures such as the wedge, channel, and gap plasmonic waveguides, which typically only support a tightly confined mode only near the plasmon frequency, the slot waveguide has two conductors, thereby enabling the structure to support TEM and quasi-TEM modes over a broad range of frequencies [173]. These quasi-TEM modes can have a high group velocity, appreciable propagation length, and deep subwavelength size. Since this time, many groups have demonstrated waveguides based on the slot geometry. In 2006, Dionne et al. demonstrated slot waveguides in silver, achieving propagation lengths of 6 μm ($\alpha =0.72\mathrm{dB}/\mathrm{\mu m}$) for a 100 nm wide slot at $\lambda =840\mathrm{nm}$ [174]. Recently, hyperbolic metamaterials have been suggested as replacements for solid metal claddings, numerically achieving a propagation length of 110 μm for a mode size of 1.2 μm at $\lambda =1.55\mathrm{\mu m}$ [175]. The performance was shown to be greater than MIM and IMI waveguides at the telecommunication wavelength.

All of these structures based on the MIM configuration are able to provide modal confinement in the subwavelength or deep subwavelength range, enabling ultracompact photonic circuitry with very sharp bends. Unfortunately, none of them were able to achieve propagation lengths larger than $\sim 100\mathrm{\mu m}$, which has limited their application for waveguiding. However, the tight modal confinement has proven useful for developing ultracompact modulators with high extinction ratios (ERs) and will be discussed in more detail to come (see [10,17] and references therein for more information).

3. Dielectric Loaded

Borrowing some ideas from standard photonics, DLW introduced a dielectric ridge placed directly on top of a thick or thin metal layer as shown in Fig. 5(g). The addition of this 2D ridge on a uniform thin metal film resulted in the localization of the SPP within the ridge and provided additional design parameters to modify the dispersion and affect confinement. DLWs were first proposed by Hohenau et al. where they theoretically investigated the change in dispersion for an unbalanced IMI configuration (i.e., superstrate and substrate permittivities are not equal) and experimentally investigated SPP propagation on a 50 nm gold layer on glass with a 150 nm thick silicon dioxide ridge using the Kretschmann geometry [146]. They demonstrated energy propagation localized within the ${\mathrm{SiO}}_2$ ridge, even though the size of the ridge was below the cut-off thickness. Shortly thereafter, Steinberger et al. demonstrated DLWs and passive components on 50 nm thick gold films with ${\mathrm{SiO}}_2$ ridges excited by the Kretschmann geometry at $\lambda =800\mathrm{nm}$, and suggested the effective refractive index method as a means for modeling the structures [176]. The structure was found to have a propagation length of 8 μm ($\alpha =0.54\mathrm{dB}/\mathrm{\mu m}$) for a ridge height of 60 nm and width of 300 nm, a value competitive with the current MIM structures. Using a finite metal strip with a dielectric ridge on top, Grandidier et al. demonstrated a propagation length of 25 μm ($\alpha =0.17\mathrm{dB}/\mathrm{\mu m}$) on a 1.75 μm wide, 50 nm thick gold film with a 600 nm square PMMA ridge at $\lambda =1.55\mathrm{\mu m}$ [177]. Following, an additional dielectric layer between the substrate and metal strip was proposed by Holmgaard et al. enabling the effective index above and below the waveguide to be balanced [178]. This balance mimics the case of the LR-SPP, but resulted in a hybrid plasmonic mode with tighter confinement than the standard LR-SPP while still maintaining low propagation loss. The resulting long-range DLW (LR-DLW) was found to have a propagation length of 3.1 mm ($\alpha =1.4\mathrm{dB}/\mathrm{mm}$) for a mode size of 1.8 μm. Subsequently, the LR-DLW was experimentally measured by Volkov et al. and was found to have a propagation length of 520 μm ($\alpha =8.3\mathrm{dB}/\mathrm{mm}$) for a mode size of 0.9 μm [179].

As a result of their adequate confinement, millimeter scale propagation, favorable fabrication, and relatively easy integration with photonic waveguides, LR-DLWs are one of the best potential candidates for on-chip plasmonic interconnects. Structures based on the traditional DLW geometry have shown the ability to carry 480 Gbps through wavelength division multiplexing, and similar experiments are expected for the LR-DLW structures in the near future (see [180] and references therein for more information).

4. Hybrid Plasmonic Photonic

Similarly to the DLW, HPPW geometries attempt to improve confinement and propagation length by introducing a photonic waveguide to the structure [see Fig. 5(h)]. HPPWs consist of a general group of geometries that may differ drastically in their structure and materials, but are commonly united in that their mode shape is a mixture of both Gaussian and exponential dependencies. HPPWs were first investigated theoretically by Oulton et al. in 2008, where they used a high index nanowire spaced above a thick silver film, achieving propagation lengths of 80 μm ($\alpha =54.3\mathrm{dB}/\mathrm{mm}$) for a mode size of 250 μm at $\lambda =1.55\mathrm{\mu m}$ [147]. Shortly thereafter, Salvador et al. theoretically studied a similar design for a hybrid slot waveguide providing analytical expressions for the field and comparing with numerical analysis of passive structures [181]. Following these initial designs, many other geometries of HPPWs began to arise. In 2010, Wu et al. demonstrated a metal cap waveguide that consisted of a silicon photonic waveguide with a 50 nm gold cap separated from the silicon by a ${\mathrm{SiO}}_2$ spacer layer. At $\lambda =1.55\mathrm{\mu m}$, they demonstrated a propagation length of 40 μm ($\alpha =108.5\mathrm{dB}/\mathrm{mm}$) for a mode size of 30 nm, essentially localized within the silicon dioxide spacer layer [42]. Shortly thereafter, Goykhman et al. demonstrated a similar structure using local oxidation of silicon to form the waveguide and demonstrated a propagation length of 400 μm ($\alpha =10.5\mathrm{dB}/\mathrm{mm}$) for a mode size of 400 nm at $\lambda =1.55\mathrm{\mu m}$ [41]. The metal cap structure was then slightly altered by Dai and He by depositing a thin oxide layer and a thick metal cladding over the silicon ridge [182]. The resulting structure, now called the metal-insulator-semiconductor-insulator-metal (MISIM), localized modes on either side of the waveguide where the oxide was thinnest. The authors experimentally achieved a propagation length of 50 μm ($\alpha =86.8\mathrm{dB}/\mathrm{mm}$) for a mode size of 50 nm [182]. LR-HPPWs, similar to the LR-DLW, also have been proposed, which can guide light for distances similar to conventional strip LR-SPP waveguides but with a smaller mode size. Chen et al. numerically investigated an LR-HPPW on a silicon platform with silver, achieving a propagation length of 1 mm ($\alpha =4.34\mathrm{dB}/\mathrm{mm}$) for a mode size of $\sim 350\mathrm{nm}$ [183]. Because of the outstanding ability of HPPWs to tune the loss-confinement trade-off, they are considered as one of the most promising alternatives for on-chip nanophotonics (see [184] and references therein for more information).

B. Alternative and CMOS Material Waveguides

Nanophotonic waveguides based on gold and silver have laid the framework for a myriad of geometries covering a broad spectrum of performance. However, for the performance, capabilities, and practicality of plasmonic interconnects to continue to grow, alternative materials should be considered. While metallic properties have been confirmed for a wide range of materials, only a small subset of these materials have actually been used to guide SPPs, and even fewer have been used in a 2D waveguide geometry. TCOs such as ITO, AZO, and GZO have been studied using prism or grating coupling to observe a dip in the reflectance due to SPP propagation [118,185,186]. A similar situation was noted for early tests on TiN films [187,188]. Also, more exotic methods such as topological insulators have been suggested as alternative materials for plasmonic waveguiding applications [189]. While these studies stand as important characterizations of new materials, as with other 1D studies on noble metals, we do not consider them further in this review, since they are ill-suited for integration.

As early as 2008, Soref began to investigate the CMOS-compatible options of highly doped silicon, other silicides, highly doped germanium, and other germanides for their applications toward plasmonic waveguiding [93 –95,99]. However, due to increased interband losses in the near-infrared and visible wavelengths, these materials are best suited for mid- and far-infrared nanophotonic applications. Palladium silicide (${\mathrm{Pd}}_2\mathrm{Si}$) and doped silicon were studied for varying doping concentrations in an unbalanced IMI waveguide configuration. For ${\mathrm{Pd}}_2\mathrm{Si}$, a propagation length of 10 μm ($\alpha =0.434\mathrm{dB}/\mathrm{\mu m}$) for a mode size of 2 μm was determined for $\lambda =1.55\mathrm{\mu m}$ and a propagation length of 1 mm ($\alpha =4.34\mathrm{dB}/\mathrm{mm}$) with a mode size of 8 μm for the same structure at $\lambda =4.1\mathrm{\mu m}$. Also, highly doped silicon was shown to achieve a propagation length of 3 mm ($\alpha =1.45\mathrm{dB}/\mathrm{mm}$) with a mode size of 50 μm at a wavelength of $\lambda =12.4\mathrm{\mu m}$. In addition to studying the silicides, Soref also suggested that germanides represent another potential CMOS-compatible alternative for integrated nanophotonic applications. In this work, various compositions and doping concentrations of n-Ge and $n\text{-}{\mathrm{Ge}}_x{\mathrm{Sn}}_y$ were studied in an unbalanced IMI waveguide structure. The performance of the materials presented were largely similar for a given carrier concentration, achieving a propagation length of 2 μm ($\alpha =2\mathrm{dB}/\mathrm{\mu m}$) for a mode size of 300 nm at $\lambda =3\mathrm{\mu m}$ to a propagation length of 40 μm ($\alpha =0.1\mathrm{dB}/\mathrm{\mu m}$) for a mode size of 60 μm at $\lambda =12\mathrm{\mu m}$ [99]. From these results, it is clear that at the technologically important range of 8–12 μm with thermal radiation and biological applications, highly doped silicon, germanium, the silicides, and germanides represent promising alternative material candidates for CMOS-compatible nanophotonics [190].

The group of III–V semiconductors also has received attention for integrated nanophotonics applications, mainly due to their ability to support high carrier concentrations through quantum well heterostructures. Additionally, these materials can be grown epitaxially one upon the other, enabling extremely sharp interfaces and low scattering losses. Li and Ning suggested an entirely plasmonic active system in 2011 using gallium antimonide (GaSb)/indium arsenide (InAs) quantum wells and aluminum antimonide (AlSb) as outer layers [100]. These quantum well waveguides were shown to achieve a FOM more than one order of magnitude larger than similar structures with gold or silver cores. The GaSb/InAs/GaSb quantum well waveguide structure was shown to achieve mode sizes down to 73 nm for a carrier concentration of $8\times {10}^{19}{\mathrm{cm}}^{-3}$. In addition, an entirely integrated plasmonic system was suggested, which borrowed from the very well established fabrication procedures and a wide range of devices using this materials system. The device included an electrically injected SPP source, in-line amplifiers, and an integrated detector, where one GaSb/InAs/GaSb quantum well served as the SPP waveguide and another AlSb/InAs/AlSb layer served to inject carriers and to provide gain, or in reverse bias as a detector. However, similar to the silicides and germanides, the III–V semiconductor structure has interband absorption starting in the near-infrared region. Thus, such a structure, while able to achieve carrier concentrations high enough to be metallic in the near-infrared, cannot currently be used at the telecommunication wavelengths. Regardless, this device represents a substantial leap forward in the work of integrated nanophotonic devices, especially with the growing interest within the semiconductor industry to include III–Vs into the CMOS process.

One of the most promising materials for nanophotonic devices is TiN due to its gold-like optical properties, CMOS compatibility, and adjustable optical properties. Earlier this year, Kinsey et al. demonstrated epitaxial quality IMI strip waveguides on c-sapphire that support the LR-SPP mode using TiN [191]. By means of index-matching oil, they demonstrated a propagation length of 5.5 mm ($\alpha =0.79\mathrm{dB}/\mathrm{mm}$) for a mode size of 9.8 μm at the telecommunication wavelength of $\lambda =1.55\mathrm{\mu m}$. This waveguide structure was able to outperform similar waveguides based on gold, despite having less ideal material properties, due to the roughness of gold in thin film form. This illustrates the importance of high-quality thin films for nanophotonic applications. Additionally, they proposed a solid-state alternative to the design using a stoichiometric ${\mathrm{Si}}_3{\mathrm{N}}_4$ upper cladding, forming a LR-HPPW, which was numerically shown to achieve a propagation length of 13 mm ($\alpha =0.34\mathrm{dB}/\mathrm{mm}$) for a mode size of 8.7 μm with an ultrathin, 6 nm thick TiN layer at $\lambda =1.55\mathrm{\mu m}$. While the structures proposed are not fully CMOS compatible, they represent the lowest loss alternative material nanophotonic waveguides to date, and demonstrate the potential performance of future fully CMOS-compatible designs.

In an effort to enhance the study of waveguides using alternative materials, numerical simulations in COMSOL multiphysics of 2D structures using published permittivity values at $\lambda =1.55\mathrm{\mu m}$ is presented for the strip and gap geometries. For the strip geometry, the dimensions of the metal waveguide were 1 μm by 10 nm, and for the gap geometry, 1 μm by 200 nm. Both designs were cladded in a uniform dielectric layer of silicon ($n=3.48$) to ensure CMOS compatibility of the design. The results of these tests are summarized in Table 1.

Table 1. Summary of Numerical Results for Strip and Gap Plasmonic Waveguides Using Alternative Materials^a

View Table

Nanophotonic waveguides that use the traditional CMOS-compatible metals copper and aluminum have seen more attention from the community. One of the first examples in 2010 was proposed by Krasavin and Zayats in the form of a silicon ridge DLW over a thick aluminum layer [192]. Here, they numerically demonstrate a propagation length of 5 μm ($\alpha =0.87\mathrm{dB}/\mathrm{\mu m}$) for a mode size of 141 μm at $\lambda =1.55\mathrm{\mu m}$. Shortly thereafter, Flammer et al. investigated an HPPW geometry with an $\sim 80\mathrm{nm}$ thick aluminum strip spaced from a SOI substrate by a thin layer of ${\mathrm{SiO}}_2$. This hybrid mode structure was experimentally found to achieve a propagation length of 0.82 mm ($\alpha =5.29\mathrm{dB}/\mathrm{mm}$) for a mode size of $\sim 5\mathrm{\mu m}$ at $\lambda =1.55\mathrm{\mu m}$ [193]. This structure was beneficial because the underlying SOI substrate was not required to be patterned to achieve guiding. By coupling to a relatively thin strip of aluminum on the surface, the mode effectively could be guided in the silicon device layer, removing the losses due to sidewall roughness that troubled standard ridge waveguides. In 2011 and 2012, Zhu et al., Kim et al., and Kwon et al. began to numerically and experimentally investigate geometries based on the MISIM geometry [112,194 –201]. This geometry was highly beneficial, as it mimics the structure of a metal-oxide-semiconductor (MOS) capacitor in traditional electronic circuits, increasing the potential for seamless integration into future devices. Straight waveguides by Zhu based on aluminum MISIM designs were able to achieve propagation lengths of 39 μm ($\alpha =0.11\mathrm{dB}/\mathrm{\mu m}$) for a mode size $\sim 180\mathrm{nm}$ at $\lambda =1.55\mathrm{\mu m}$ [194]. In 2013, CMOS-compatible designs were brought to LR-DLW designs by Shi et al. [202]. Using a ${\mathrm{Si}}_3{\mathrm{N}}_4$ ridge over an SOI substrate, designs based on copper were numerically shown to achieve a propagation length of 1 mm ($\alpha =4.34\mathrm{dB}/\mathrm{mm}$) for a mode size of 1.18 μm and as high as 4.4 mm ($\alpha =0.97\mathrm{dB}/\mathrm{mm}$) for a mode size of 2.32 μm, both at $\lambda =1.55\mathrm{\mu m}$. This structure, much like its noble metal equivalent, achieves a long propagation length for a relatively small mode size, making it highly desirable for on-chip applications. In 2014, a similar design was experimentally realized by Zektzer et al. in collaboration with Bozhevolnyi [40]. Using an SOI wafer with a ${\mathrm{Si}}_3{\mathrm{N}}_4$ ridge, the structure achieved a propagation length of 0.7 mm ($\alpha =6.2\mathrm{dB}/\mathrm{mm}$) with a mode size of roughly 1 μm at $\lambda =1.55\mathrm{\mu m}$. However, this device also is interesting because it can support two modes under TM excitation, enabling the potential for modal multiplexing for enhanced operational bandwidth. They also investigated bends, demonstrating a $6\mathrm{dB}/\mathrm{mm}$ loss, which could be improved with more optimized fabrication methods.

C. Performance Comparison

To finalize our discussion on waveguide geometries, it would be desirable to visually/numerically compare the performance of the structures. However, creating a full-featured FOM is quite difficult due to the wide range of important parameters for many applications. Here, we have taken two of the most important parameters for integrated interconnects, the modal size and the propagation length, and normalized them to the wavelength of operation to provide a qualitative evaluation of the performance for many works in the field of noble metal and alternative material waveguides. These two parameters also illustrate the intrinsic trade-off between loss and confinement whose optimization is one of the primary goals of these devices. In Fig. 6, a summary of performance of the various noble metal waveguide structures is illustrated, while in Fig. 7, a summary of waveguide structures using alternative and CMOS compatible designs is presented. In these graphs, circles represent experimentally measured performance, while triangles represent numerically predicted performance, and colors delineate the various waveguide geometries discussed in the previous sections. In addition, a solid line is presented, which represents a performance of $L_{\text{prop}}/D_z=1000$, such that the propagation length is 1000 times the size of the modal confinement.

 

Fig. 6. General overview of the performances of nanophotonic waveguides based on traditional noble metals. The graph represents the device’s normalized propagation length (PL) in $\lambda$’s versus normalized mode size (MS) in $\lambda$’s. The chart encompasses insulator-metal-insulator (blue icons), metal-insulator-metal (purple icons), dielectric loaded (green icons), and hybrid plasmonic photonic (red icons) waveguides. Dots and triangles represent experimental and numerical evaluations, respectively. The line represents a figure of merit (PL/MS) value of 1000. References for the graph are given in order or decreasing propagation length for each geometry group.

IMI: [39,39,161,74,161,71,203,72,203]

MIM: [166,168,204,174,54,166,205,206,207,167]

DLW: [178,208,208,179,209,210,211,209,176]

HPPW: [183,212,41,213,147,198,42,214,215,216]

Download Full Size | PDF

 

Fig. 7. General overview of the performances of nanophotonic waveguides based on alternative and CMOS-compatible materials. The graph represents the devices normalized propagation length (PL) in $\lambda$’s versus normalized mode size (MS) in $\lambda$’s. The chart encompasses insulator-metal-insulator (blue icons), metal-insulator-metal (purple icons), dielectric loaded (green icons), and hybrid plasmonic photonic (red icons) waveguides. Dots and triangles represent experimental and numerical evaluations, respectively. The line represents a figure of merit (PL/MS) value of 1000. References for the graph are given in order or decreasing propagation length for each geometry group.

IMI: [191], TiN (Table 1) [93,93], ZrN (Table 1), HfN (Table 1) [93,93,99,99,99]

MIM: TiN (Table 1), ZrN (Table 1), HfN (Table 1)

DLW: [202,202,192]

HPPW: [191,193,194,199,200,201,196,196,195,113,195]

Download Full Size | PDF

Considering the noble metal graph, several items are readily apparent. First, we can see that, at this time, structures with the largest FoM are the low-loss IMI waveguide structures. Despite their low modal confinement, they more than make up for this by the drastically reduced losses, with several experimental works above the 1000 FoM line. In addition, with the exception of the experimental work by Goykhman et al., no other geometry is capable of exceeding the 1000 FoM line in experiments [41]. With the addition of their simple fabrication, these structures currently may be the leaders for integrated interconnects. Second, the DLW and HPPW geometries provide the best compromise between the low loss of the IMI structures and the high confinement of the MIM designs. This is expected, since this is one of the main driving points of their development. Their modal confinement spans roughly $\lambda$ to $\lambda /100$ with many designs demonstrating performance very close to the 1000 FoM line. With continued development into these designs, their better modal confinement and relatively high propagation length may prove more suitable for integrated devices.

Next, if we consider the alternative materials graph, we can see that many of the structures shown here have demonstrated performance equivalent to other noble-metal-based designs, while providing additional technological benefits such as adjustable optical properties and CMOS compatibility. Two geometries even exceed the 1000 FoM line, the DLW design by Shi et al. and the HPPW design by Kinsey et al. [191,202], which represent two of the most recent advances in this field. This, therefore, demonstrates the potential of the field with continual development to meet or exceed the performance of noble metal devices while maintaining technological benefits. While there is currently a lack of experimental works demonstrating the performance of realistic devices, the movement to optimize waveguides using these materials has largely just begun and is expected to have a profound impact on future practical integrated nanophotonic devices in the near future.

5. NANOPHOTONIC MODULATORS

Modulators represent another key component for integrated devices, enabling data to be imparted onto a carrier wave. Drawing from similar components in radio-frequency electronics and photonics, plasmonic modulators can either adjust the amplitude of the light or the phase of light passing through the structure. The method these structures use to achieve modulation varies, although induced absorption, induced index change, and nonlinear interactions are among the most popular solutions today. While not desired for every application, some of these techniques are nonvolatile, which enables applications in optical memory devices and programmable nanophotonic circuits.

Recently, there has been great progress in the area of integrated photonic modulators with several groups reporting speeds greater than 10 GHz with low power dissipation [217 –220]. While these devices represent significant advances in their respective fields, and maintain compatibility with the CMOS process, there is still a need for devices that are significantly smaller to achieve high-density circuits. Modulators based on plasmonic designs are attempting to answer this call. In the following paragraphs, we will refer to a selected list of SPP modulators, which mark the most recent advances in nanophotonic communication systems. In our discussion, we identify three main groups of modulators: (1) electroabsorption modulators; (2) Mach–Zehnder modulators; and (3) other alternative designs.

The strategy based on the electroabsorption effects relies on the variation of the complex refractive index of a core material, which is triggered by an (optically or electrically) induced variation of the carrier concentration. This approach might exploit either a change in the real or the imaginary part of the dielectric permittivity. The former strategy achieves efficient modulation through a fast change in the allowed propagating modes, while the latter makes use of a considerable alteration of the propagation losses.

As opposed to the electroabsorption modulation, the Mach–Zehnder design takes advantage of the phase change of the propagating wave rather than its amplitude. On the one hand, because the phase change is normally a cumulative effect, which requires a remarkable propagation length, Mach–Zehnder modulators suffer from scalability problems that can only partially be solved by using the stronger light–matter interaction provided by SPPs. On the other hand, since the modulation is achieved by interference, this class of device normally shows very good modulation depth.

Even though electroabsorption and Mach–Zehnder modulators currently play the main role in the panorama of all possible options for achieving efficient SPP modulation, in recent years, other physical phenomena such as material phase transitions, nonlinear phase changes, induced formation of conductive bridges, resonant shifts, thermo-optic effects, and many others have been explored. Typically, these alternative strategies are successful in maximizing only one of the fundamental parameters for an ideal modulation process such as the modulation depth, temporal response, propagation losses, etc. However, they normally fail to deliver overall high-performance when all of the previously listed parameters have to be taken into account in one single FoM. Nevertheless, this class of devices can be proficiently used for specific applications where the full FoM does not need to be satisfied. In the following sections, we will review the advances in integrated plasmonic modulators that use noble metals, as well as those designs that use alternative and CMOS-compatible materials, and conclude with a graphical comparison and discussion of numerous experimental and theoretical results.

A. Noble Metal Nanophotonic Modulators

As with waveguides, noble metals have played a pivotal role in the development of nanophotonic modulators due to their desirable metallic properties and widespread availability in research centers. Here, we review nanophotonic modulator structures with noble metals, which utilize various techniques to achieve dynamic performance. Although some of these geometries also use alternative materials, from an integration standpoint, if the device uses noble metals, it is not friendly for current processing techniques. Consequently, devices that use noble metals within any part of the structure (with the exception of the electrical contacts that could easily be replaced with copper/aluminum without performance degradation) were considered noble metal designs and are reviewed in this section.

1. Electroabsorption Modulators

One of the most remarkable works of nanophotonic modulators is that of Dionne et al. based on the metal-oxide-Si planar structure, called the PlasMoStor (2009). The device has the capability of tightly localizing the propagating mode within a 10 nm thick oxide layer and achieves input/output coupling by means of subwavelength slits vertically etched into the top and bottom metallic layers, respectively [44]. The PlasMoStor was experimentally tested at telecom wavelengths and produced a modulation depth approaching 10 dB in few μm of propagation length with propagation losses in the off-state of the order of $1\mathrm{dB}/\mathrm{\mu m}$. Even though the intrinsic modulation speed of the device is theoretically limited to 15 GHz by the formation of the accumulation layer, this work for its completeness and originality is one of the cornerstones in nanophotonics.

Another interesting design is the waveguide-coupled MIM modulator by Cai et al., where a plasmonic cavity is formed inside an 80 nm thick dielectric layer by creating two nanogaps on one of the two metallic layers [221]. Efficient modulation is then achieved by electro-optical tuning of the cavity-mode resonance. This device is numerically proven to be capable of achieving 3 dB modulation in 1 μm with propagation losses of 3% per transit length of the cavity. The main limitations for this device are mainly related to the operative wavelength, which was set at 850 nm, and the lack of analysis with regards to the mode coupling in and out of the device (a critical parameter for MIM-based devices). However, the proposed layout allows for fast (hundreds of GHz) and deeply subwavelength modulation.

Another interesting SPP absorption modulator is the design proposed by Melikyan et al. in 2011, which is based on a metal/oxide/semiconductor/metal configuration [45]. The materials employed for the device fabrication were Ag, ITO, ${\mathrm{Si}}_3{\mathrm{N}}_4$, and Si, achieving a 1 dB modulation in 2 μm at $\lambda =1550\mathrm{nm}$ with a voltage-induced carrier density change in the ITO layer. Despite the consistent propagation losses (approaching $10\mathrm{dB}/\mathrm{\mu m}$) and the limited modulation depth, this work has the merit of analyzing the fundamental problem of coupling from a photonic waveguide and inspiring other modulators using different materials with a similar configuration.

An example of this inspiration is the ultracompact and broadband nanophotonic modulator by Sorger et al., which was published in 2012 [60]. In this case, an $\mathrm{Au}/{\mathrm{SiO}}_2/\mathrm{ITO}/\mathrm{Si}$ modulator is proposed and experimentally tested. The module achieves 1 dB modulation in 5 μm at $\lambda =1300\mathrm{nm}$ and realizes optimal coupling and propagation losses as low as 0.14 dB and $1\mathrm{dB}/\mathrm{\mu m}$, respectively. In addition, the operational bandwidth was estimated to span the remarkable wavelength range from 1.2 to 2.2 μm.

Another modulator layout following the same general structure as that of Sorger was proposed by Chen et al. in 2013 [134]. This numerical analysis refines the previous work by Sorger et al., also adding a general overview about photonic modulator performances.

Although the present review is centered on practical in-plane modulators for telecom applications, it is also important to mention the experimental demonstration of an Si-based SPP modulator using a gold/silicon grating coupler. The device was tested in a pulsed regime in the 1300–1700 nm wavelength range. The modulation is obtained by optically varying the carrier concentration of the upper silicon layer. Even though the overall time response of the device is limited by the relatively slow ($>100\mathrm{ps}$) carrier recombination in Si and represents more of a proof of concept than a finished device, it is one of the first experimental demonstrations of its kind [222].

2. Mach–Zehnder Modulators

Generally speaking, the Mach–Zehnder configuration is the most used solution for electro-optic modulation systems. This is mainly due to ease of fabrication, reliability, large engineering tolerances, good ER, and low production cost. However, as previously mentioned, this kind of device suffers from an intrinsic limit for the scalability of the device. Inspired by the possibility of exploiting the strong light–matter interaction provided by nanophotonic systems in order to drastically reduce the device footprint, numerous groups have used similar interferometric approaches for SPP modulation.

Among these works, we would like to mention the pioneering works of Prof. Bozhevolnyi’s group using the thermo-optic effect to modulate the SPP signal at the telecom wavelengths [66,223]. Even if these devices have switching times of the order of milliseconds, they have proven capable of achieving 3 dB modulation in short distances of $\sim 100\mathrm{\mu m}$ with low driving power, which is a drastic improvement when compared with equivalent thermo-optic photonic systems.

However, in order to fully overcome scalability problems, MIM and MISIM structures have been proposed and numerically analyzed. Among these, we recall the study proposed in 2010 by Pu et al. based on a nonlinear Mach–Zehnder modulator designed between two infinite silver slabs [224]. In this case, the active arm of the interferometer is modelled by using a highly nonlinear polymer (MEH-PPV) and theoretically produced a modulation depth of 16.8 dB in only 5 μm at $\lambda =1064\mathrm{nm}$. The device is also predicted to have limited coupling losses (on the order of 3 dB) and acceptable propagation losses in the off-state ($<2\mathrm{dB}/\mathrm{\mu m})$.

In the same year, another interesting layout was proposed by Zhu et al. [225]. Here, the SPP at telecom wavelengths propagates on a 3 μm long Ag–SiO₂–Si–Ag phase shifter, and subsequently through a 0.35 μm long combiner in order to produce 7.3 dB modulation with only 3 dB of insertion loss. The device is numerically proven to work at 500 GHz. The main drawbacks of this layout are the large switching voltage required (5.6 V) and the vertical geometry, which, in turn, produces numerous extra fabrication challenges.

3. Additional Modulators

The following section on additional modulator designs has the main purpose of underlining the exceptional versatility of nanophotonic structures in comparison with purely photonic elements. Most of the modulator layouts that are reported in this paragraph can be realized only in the plasmonic domain.

First, we would like to discuss those modulators that undergo a structural, physical, and/or morphological transformation in order to achieve signal modulation. A prime example is the Au/Ga waveguide-based modulator proposed by Krasavin et al. in 2004 [226]. In this device, the signal propagates in a metal-on-dielectric waveguide containing a gallium section a few micrometers long. Modulation is achieved by switching the structural phase of gallium, numerically shown to operate at $\lambda =1.3\mathrm{\mu m}$, and is expected to have a modulation depth approaching 7 dB in less than 6 μm.

Another important milestone is the $\text{metal-insulator}\text{-}{\mathrm{VO}}_2\text{-}\text{metal}$ modulator by Sweatlock et al., which exploits the large absorption of the metallic phase of vanadium dioxide to implement broadband modulation at telecom wavelengths [136]. The device is predicted to produce up to 20 dB of attenuation between the on- and off-states.

Recently in 2013, Emboras et al. proposed, and experimentally demonstrated, an induced conductive bridge modulator called a “memristor” [227]. The device’s switching capability is based on the formation/annihilation of nanoscale Ag filaments under the metal film layer inside the semiconductor core where most of the SPP propagates. The creation/removal of the filaments is electrically driven. The device has limited modulation capability of about 1% and an overall length of 3.3 μm. In general, the main limitations associated with the phase/structural change class of modulators are the slow time response (up to milliseconds) and the remarkable propagation losses. For these reasons, they are more suitable for data storage rather than data-processing applications.

Another alternative modulation approach is using nonlinear index modulation. In this direction, the work by Berini et al. who proposed a device consisting of a thin Au strip buried in z-cut lithium niobate (${\mathrm{LiNbO}}_3$) cladding designed to work at 1550 nm in 2007 [228]. Despite the simplicity of its design, this modulator is quite versatile. In fact, by engineering the crystallographic orientation of ${\mathrm{LiNbO}}_3$ with respect to the external applied field, it is possible to achieve modulation either in amplitude or in phase. The authors chose to modulate the amplitude of the propagating signal (z-cut crystal) in order to take full advantage of the highest electro-optic coefficient $r_{33}$. A 2 mm long device operating at $\lambda =1550\mathrm{nm}$ was experimentally demonstrated to have a 7 dB modulation with propagation losses of the order of $1\mathrm{dB}/\mathrm{mm}$.

In 2011, Sun et al. proposed a metal-polymer-Si nonlinear modulator [229]. In their numerical analysis, the plasmonic field enhancement in the low-index nonlinear polymer layer provides a subwavelength optical confinement and a fast optical modulation speed at telecom wavelengths. This design promises a $\pi$ phase modulation in 13 μm with good coupling losses and relatively low propagation losses ($<1\mathrm{dB}/\mathrm{\mu m})$.

Another nanophotonic modulator, which exploits the Pockels effect of nonlinear polymers, is the design realized by Melikyan et al. in 2014 [230]. This device, operating at $\lambda =1550\mathrm{nm}$, has a straightforward planar structure and considers the optical coupling to and from a Si nanowire. The device was shown to achieve a modulation depth approaching 10 dB in 30 μm with total I/O propagation losses of 12 dB.

The number of papers reporting nonconventional strategies for signal modulation at the nanoscale is simply enormous, and, for the sake of brevity, we conclude this paragraph by mentioning that elasto-optic effects [231], resonance shifts [232,233], and suppression/promotion of signal amplification in active media [234,235] have also demonstrated efficient modulation.

B. Alternative and CMOS-Compatible Nanophotonic Modulators

So far, we have reported only on noble-metal based modulators, which are easily attainable in a scientific lab, enabling a vast array of proof-of-concept devices. However, one key parameter for allowing low-cost mass production of nanophotonic devices is the compatibility with standard CMOS technology. In addition, it is in the realm of material science where we find solutions to enable full control over the optical properties of the device’s constituent materials. It is for these reasons that the research on alternative materials for nanophotonic modulators has been so active, if not frenetic, during the last decade. In the next section, we will briefly report on some of the most interesting SPP modulators that make use of alternative and CMOS-compatible materials.

One of the cornerstone works in CMOS-compatible nanophotonic modulators is the work by MacDonald et al. in 2009 [236]. This work is one of the very first experimental demonstrations of ultrafast modulation of SPP-based signals using CMOS planar technology. The proposed device is based on a SI SPP waveguide having aluminum and fused silica as constituent materials. Even though this modulator suffers from fundamental limitations such as a nontelecom operational wavelength ($\lambda =780\mathrm{nm}$), low propagation length, and limited ER (7.5% in 5 μm), it is a pioneering study that clearly pointed out the need for CMOS compatibility, in addition to underlining the great potentials of this technology for ultrafast data manipulation.

In addition to traditional CMOS-compatible materials, the work by Liu et al. investigated the potential for emerging alternative platforms [237]. In their study, using a $\text{graphene}\text{-}{\mathrm{Al}}_2{\mathrm{O}}_3\text{-}\mathrm{Si}\text{-}{\mathrm{SiO}}_2$ planar structure, a 4 dB modulation depth is achieved by tuning the Fermi level of the monolayer graphene sheet. The device operates at telecom wavelengths, and it is 40 μm long with propagation losses as low as $0.25\mathrm{dB}/\mathrm{\mu m}$.

Another important experimental demonstration of CMOS-compatible electroabsorption modulators is that of Zhu et al. [238]. In this structure, the device core is based on a horizontal MISIM nanoplasmonic slot waveguide, which achieves 3 dB of modulation in only 3 μm with $1\mathrm{dB}/\mathrm{\mu m}$ propagation losses. The main limitations related to this configuration are the fabrication complexity, the high driving voltage ($\sim 6.5\mathrm{V}$), and the parasitic capacitances limiting the maximum achievable speed.

Inside the realm of numerical studies, it is important to remember the ultracompact modulator by Babicheva et al. published in 2013 [59]. Among the many configurations reported in this work, the planar stack of $\mathrm{Si}\text{-}\mathrm{GZO}\text{-}{\mathrm{Si}}_3{\mathrm{N}}_4\text{-}\mathrm{TiN}\text{-}\mathrm{Si}$ stands out with a remarkable $46\mathrm{dB}/\mathrm{\mu m}$ modulation depth and propagation losses as low as $0.29\mathrm{dB}/\mathrm{\mu m}$. This study, despite being only numerical, clearly highlights the potential of TiN and gallium-doped zinc oxide for nanophotonic applications. In addition, the authors pay specific attention to the problem of the I/O coupling with LR-SPP waveguides, achieving less than 1 dB of coupling loss per facet in the off-state.

A similar technique was used by Lu et al. in 2012 [61]. Here, the authors analyze a $\mathrm{Si}\text{-}\mathrm{AZO}\text{-}{\mathrm{SiO}}_2\text{-}\mathrm{Si}$ multilayer device, where, in order to maximize the tunability of the optical properties, they vary the carrier concentration of the TCO across the range in which the material undergoes an optical transition at its epsilon near zero point. By doing so, a $17.4\mathrm{dB}/\mathrm{\mu m}$ modulation can be achieved at $\lambda =1310\mathrm{nm}$ with $1\mathrm{dB}/\mathrm{\mu m}$ of propagation losses in the low-loss state.

In addition to the electroabsorption strategy, several SPP modulators employing alternative/CMOS materials and a Mach–Zehnder configuration have been proposed. Specifically, the work by Zhu et al. published in 2013 investigated a horizontal $\mathrm{Cu}\text{-}{\mathrm{SiO}}_2\text{-}\mathrm{Si}\text{-}{\mathrm{SiO}}_2\text{-}\mathrm{Cu}$ waveguide as the guiding core of a Mach–Zehnder interferometer [239]. The device was shown to achieve a 9 dB modulation depth in just 1 μm with relatively low propagation losses ($0.58\mathrm{dB}/\mathrm{\mu m}$) and 1 dB of coupling loss per facet. Unfortunately, the device needs to be driven by a quite a high voltage (7 V) and sees its modulation capability drastically reduced when tested at relatively slow speed of 10 MHz, restricting its use for high-speed devices.

For numerical analyses, the study by Thomas et al. published in 2012 is important to mention [240]. In this case, a horizontal $\mathrm{Al}\text{-}{\mathrm{HfO}}_2\text{-}\mathrm{Si}\text{-}{\mathrm{HfO}}_2\text{-}\mathrm{Al}$ waveguide represents the core of the modulator, which is predicted to achieve an ER of 9.8 dB with propagation losses in the off-state of $1.7\mathrm{dB}/\mathrm{\mu m}$. The length of the active area where the phase modulation takes place is only 3 μm long.

Several other techniques also have been demonstrated for the efficient modulation of the SPP signals and are employed without using gold or silver, such as the work Pala et al. published in 2008 [241]. Their modulator consists of an aluminum film coated with a 65 nm thick layer of photochromic (PC) molecules. Grating couplers are used to couple the SPP signal in and out of the waveguide. Modulation is obtained by optically pumping the PC molecules, which switch between a transparent and absorbing state. The experimental examination shows 6 dB of modulation in 8 μm of propagation length with propagation losses approaching $2.5\mathrm{dB}/\mathrm{\mu m}$. The main drawback of the device is that it does not operate at telecom wavelengths.

In 2005, Krasavin et al. presented a device that exploited light-induced nanoscale structural transformations in a Ga-dielectric SPP waveguide in order to accomplish its switching capabilities [242]. The experimental results showed a modulation depth of 7 dB in only 2.5 μm of propagation length. The modulator was also characterized by propagation losses as low as $0.32\mathrm{dB}/\mathrm{\mu m}$ and was operated at $\lambda =1310\mathrm{nm}$. However, as in the previous case, the modulator was also speed (up to hundreds of MHz) was the main limitation.

In 2012, Ooi et al. proposed and numerically investigated a vanadium dioxide dual-mode waveguide modulator [135]. The modulation occurs inside a $\mathrm{Cu}\text{-}{\mathrm{SiO}}_2\text{-}{\mathrm{VO}}_2\text{-}{\mathrm{SiO}}_2\text{-}\mathrm{Cu}$ slot waveguide by controlling the phase transition of the ${\mathrm{VO}}_2$ core. A modulation of 9 dB and propagation losses of $5\mathrm{dB}/\mathrm{\mu m}$ are predicted for a device, which is only 200 nm long and operates at $\lambda =1550\mathrm{nm}$.

Finally, the modulator based on a $\mathrm{Si}\text{-}{\mathrm{Si}}_3{\mathrm{N}}_4\text{-}\mathrm{Si}$ slot waveguide with a buried graphene monolayer as the inner core was proposed in 2013 by Lu et al. [243]. The very strong mode confinement provided by the slot waveguide allows for a strong interaction between the propagating radiation at telecom wavelength and the graphene layer. For this layout, a $4.42\mathrm{dB}/\mathrm{\mu m}$ modulation is predicted with propagation losses as low $<0.2\mathrm{dB}/\mathrm{\mu m}$. Together, these devices utilizing alternative materials represent promising steps forward on the road toward fully integrated devices.

C. Performance Comparison

Making a fair comparison among the performances of electroabsorption, Mach–Zehnder, and alternative modulators is quite difficult. In this regard, the perfect approach would consist of three fundamental steps: (1) identifying all the fundamental parameters of the ideal modulator; (2) defining an overall FoM that can properly describe the intrinsic trade-off among these parameters; (3) applying this FoM to all the previously listed classes of modulators. However, this strategy, which would work for standard electronic devices, cannot be strictly applied to the nanophotonic modulators available in literature. This is mainly due to the incompleteness of the reported studies, which focus only on few important parameters but rarely tells the whole story. In fact, even if we exclude from our analysis the most “down-to-earth” parameters such as production cost, deterioration, reliability, etc. (which would be of paramount importance for industrial production), we still must consider other fundamental parameters such as operational speed, power consumption per bit, device size, modulation depth, propagation losses (low-loss state), and I/O coupling losses. Even just considering this second class of parameters, the available studies fail to be complete and exhaustive. In order to perform a relatively extensive analysis of the available SPP modulation technologies, our considerations are restricted to the modulation depth and the propagation losses, which describe the most intrinsic trade-off linked with nanophotonic devices, and are among the most widely reported values.

Figure 8 reports the normalized ER versus normalized propagation losses ($\alpha$) in the off-state (low-loss state) for the different noble-metal-based modulator configurations taken into account in the present review. Electroabsorption, Mach–Zehnder, and novel alternative scheme modulators are represented by blue, green, and red icons, respectively. Another distinction also is made between experimental (dots) and numerical (triangles) evaluations. A few important considerations are immediately evident from this plot. First, the fact that, in principle, we can fit all the reported data by a straight line means that there is no solution available for circumventing the scalability-loss trade-off. In this regard, it becomes difficult if not impossible to find the absolute perfect layout, and the optimal configuration should be chosen on a case-by-case basis with respect to the specific application such as switches, data storage, sensors, etc. For example, Fig. 8 illustrates that while the alternative designs perform quite well in most of the cases, electroabsorption modulators may be the best choice if high scalability must be met at any cost.

 

Fig. 8. General overview of noble-metal-based modulator performance. The graph represents the most fundamental trade-off for nanophotonic modulators, i.e., the device’s normalized extinction ratio ($\mathrm{dB}/\lambda$) versus the normalized propagation loss ($\mathrm{dB}/\lambda$) in the off-state or low-loss state. The chart reports electroabsorption (blue icons), Mach–Zehnder (green icons), and novel alternative scheme (red icons) modulators. Dots and triangles represent experimental and numerical evaluations, respectively. The line represents a figure of merit (ER/PL) value of 10. References for the graph are given in order or decreasing extinction ratio for each geometry group.

Electroabsorption: [44,244,44,45,221,60] and [245]

MZI: [224,225,223] and [66]

Others: [232,233,231,247,226,248,235,230,249,233,227,228] and [250]

Download Full Size | PDF

Other important considerations can be made taking into consideration Fig. 9 in which a plot similar to the one reported in Fig. 8 is made by considering only modulators based on CMOS compatible and/or novel alternative materials. What is immediately evident is that, despite the limited number of works where Au and/or Ag are not among the constituent materials, these novel designs are already competitive with or outperform most traditional noble-metal-based modulators.

 

Fig. 9. General overview of alternative and CMOS-compatible material modulator performance. The graph represents the most fundamental trade-off for nanophotonic modulators, i.e., the device’s normalized extinction ratio ($\mathrm{dB}/\lambda$) versus the normalized propagation loss ($\mathrm{dB}/\lambda$) in the off-state or low-loss state. The chart reports electroabsorption (blue icons), Mach–Zehnder (green icons), and novel alternative scheme (red icons) modulators. Dots and triangles represent experimental and numerical evaluations, respectively. The line represents a figure of merit (ER/PL) value of 10. References for the graph are given in order or decreasing extinction ratio for each geometry group.

Electroabsorption: [59,251,46,61,238,252] and [237]

MZI: [239,240] and [253]

Others: [135,243,242] and [241]

Download Full Size | PDF

Finally, we wish to conclude this paragraph with strong encouragement to the nanophotonic community. As mentioned, many of the current nanophotonic modulator studies do not perform a complete evaluation incorporating even the most fundamental parameters such as modulation depth, speed, energy dissipation, and coupling loss. This is especially critical for those studies focused on key devices that will inevitably be at the center of the next technological revolution. Completing these fundamental evaluations is the only way to optimize the effort for triggering such a revolution as well as to avoid dangerous and misleading evaluations, which have, unfortunately, already happened in other critical domains [254].

6. CONCLUSION AND OUTLOOK

While the large hype of subdiffraction nanophotonic circuits is still a looming goal for the field of nanophotonics, the research in this direction has produced many outstanding devices which can have an impact upon future technologies. As a result, the current task for the field of nanophotonics remains to take current designs and to realize them with materials and processes that are ready for industrial application. Specifically, the main challenge for waveguides is to continue developing new materials and to realize large-scale integration of components using completely CMOS-compatible materials and processes. The performance of plasmonic waveguides, through the foreseeable future, will continually be limited by the loss-versus-confinement trade-off such that their performance is not likely to see a drastic improvement. However, for many applications, the current performance of these waveguides is acceptable. As a result, the task turns to continuing to investigate new materials and to realize current designs in a highly integrated fashion. Likewise, for modulators, the main challenge is to continue investigating promising geometries and to complete full-characterizations, including the extinction ratio, propagation and coupling loss, speed, and energy consumption. Currently, many works only focus on a subset of these characteristics, but for realistic devices, all must be well understood. With a more in-depth evaluation, the true advantages and disadvantages of each design will be apparent, helping the community to better focus on the most promising materials and geometries.

Our current technological scaling methods are rapidly approaching their limits, requiring new technologies to assist electronics and continue the trend of exponential performance improvement. At the present time, photonics technologies are strong candidates for enabling the next-generation high-performance devices. Traditional photonics and more recent advances in nanophotonics provide attractive advantages for integrated devices. As a result, we have even begun to see industry recognize the potential of electrophotonic hybrid devices, with several companies having released commercial products for consumers and telecommunication companies. Future devices are likely to include photonics technologies at an even more intimate level with the potential for plasmonic devices to begin appearing on-chip as well. It is an exciting time to work in the area of integrated nanophotonics: at the eve of the transition between the lab and the fab.