Suspended LRSPP for the development of highly integrated active plasmonic devices

We present a novel long-range surface plasmon polariton (LRSPP) device consisting of a suspended dielectric matrix in which an electrically active, millimeter-long metallic waveguide is embedded. We show that, by opening an air gap under the lower cladding, the influence of the substrate is suppressed and the symmetry of the thermo-optical distribution around the LRSPP waveguide is preserved over extended ranges of applied electrical current with minimal optical losses. Experimental results show that, compared to a standard nonsuspended structure, our device allows either the induction of a phase change that is three times larger, for a fixed electrical power, or, equivalently, a scaling down of the device to one-tenth of its original length, for a fixed phase change. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
The design of novel integrated photonics devices can involve many different approaches and configurations depending upon the requirements of specific applications. Nevertheless, in general, one typically targets maximizing the device's response, simplifying its structure, and minimizing its size. For these purposes, the use of surface plasmon polaritons (SPPs) can be a suitable option due to the inherently large sensitivity to external perturbations of light propagating along metal-dielectric interfaces. This property triggered the use of SPP, for instance, in sensing applications, with the first papers being reported shortly after the first demonstrations in 1983 [1] and reaching almost 400 papers per year just a few years later in 1997 [2]. SPPs are transverse magnetic (TM) polarized optical surface waves that propagate along a metal-dielectric interface. SPPs were described at the beginning of the past century [3] and experimentally reported in the late sixties [4,5]. Since then, they have been extensively studied and most of their properties have been assessed [6]. One of the most significant characteristics of SPPs is their associated short propagation distance which can reach about 19 µm and 3 µm in an air-silver and air-gold interface, respectively, at visible and near-infrared wavelengths [6]. This high attenuation is a consequence of the propagation along the single metal-dielectric interface which confines a fraction of the optical field into the metal, and thus this high attenuation is caused primarily by free-electron scattering in the metal. Nevertheless, the propagation distance can be increased by more than two orders of magnitude by reducing the thickness of the metal layer and embedding the metal stripe in dielectric matrix, i.e., by sandwiching the metal stripe between claddings with similar optical properties [7]. The symmetric plasmonic structure reduces the confinement of the propagating mode, which in turns reduces the overlap with the metal stripe and the attenuation is significantly reduced. This kind of SPP is known as long-range SPP (LRSPP) since its characteristic propagation length is significantly longer as compared to SPP. Although reduced confinement can be an issue in some applications, the improved propagation length opens the door for a mirage of application for LRSPP.
To date, the vast majority of applications based on LRSPP deal with devices containing passive structures [8][9][10][11], such as straight waveguides, S-bends, Y-junction splitters, and Mach-Zehnder interferometers (MZI), among others [8,11,12]. However, there are specific applications where having the ability to change the amplitude and/or phase of the propagating optical signal through devices like interferometers, switches, or modulators, is critical. In the case of LRSPP, active devices can be easily achieved by adding contacts to the metal stripe and then applying an electrical current to the contacts [13]. For the case of a metal stripe embedded into a polymer matrix, this would allow the entire metal stripe to be heated up and, due to the large thermo-optic effects involved, appreciable effective refractive index (RI) changes can be induced. Using similar principles, different configurations for active LRSPP devices have been proposed and demonstrated for MZIs [14][15][16], directional coupler switches [14], variable attenuators [17][18][19][20] and tunable bandpass filters [21,22].
A critical limitation in the above-mentioned scenarios is the existence of a small range for the applied electrical current, i.e., in the case of active devices requiring pure phase changes, there exists an electrical current threshold after which the insertion losses of the LRSPP begin to increase and eventually become very large. This is due to the fact that the heat dissipation is not uniform between the bottom dielectric/substrate interface and the top dielectric/air interface. This anisotropic heat dissipation effectively results in an asymmetric temperature distribution in the bottom and top dielectric layers which, in turn, induces a difference in the effective RI of the claddings due to the thermo-optic effect of the dielectric matrix. Such RI difference will induce several effects such as mode distortion and radiation losses, as well as a conversion from a long-range regime to short-range, which will gradually attenuate the transmitted signal as the applied electrical current is increased [23,24]. In order to achieve appreciable phase changes while maintaining a long-range operation, one could attempt to use longer active regions over which the cumulative effects would be similar. Unfortunately, this would require active devices with significantly larger dimensions, which is not desirable not just because an important objective of photonic integration is the optimization of the use of wafer real estate, but also because longer devices will be more susceptible to adverse effects caused by the presence of surface defects.
In this work, we present a new device structure configuration for LRSPP which consists of a suspended active region achieved by removing material from the interface between the bottom cladding and the substrate. This approach allows having the top and bottom claddings with similar heat dissipation properties and, effectively therefore, with more symmetric temperature distributions. This, in turn, helps to minimize the RI difference between the two claddings, when an electrical current is applied to the metallic waveguide and the temperature is increased. This permits LRSPP operation over an extended range of values of electrical current and at the same time maintaining shorter device length requirements. Our approach is rather general and can be extended to sensing platforms in which the waveguide itself is the sensing element [25] or it can be incorporated into more complex active architectures, e.g., arms of interferometers [26,27]. In this particular work, we first show, both numerically and experimentally, that the suspended structure in the integrated device works as symmetric LRSPP device over an extended range of electric currents, as compared to a traditional (nonsuspended) structure with the same dimensions. Next, we experimentally demonstrate the use of this concept by embedding the device in one arm of a free space MZI, to show that significantly large phase changes can be induced while maintaining negligible attenuation, as opposed to the non-suspended structure. In section 2, we discuss the principles of operation for the proposed structure, and the hypothesis is well supported by our numerical results. In section 3, we explain, in detail, the microfabrication of the samples. In section 4, we describe the experime Finally, in sec numerically simulated to assess the influence of the substrate on the symmetry of the temperature distribution. Recall that an asymmetric temperature distribution will, in turn, produce a difference in the effective RI of the claddings that can take the plasmonic waveguide to a short-range regime. The simulation of the temperature distribution was performed by introducing the presence of an electrical current flowing along the metal stripe and in the same direction of the optical propagation. In this way, heat flows from the metal stripe outwards and the asymmetry of the temperature distribution in close proximity to the metal stripe can be evaluated. When the electrical current is applied to the metal stripe, the Joule effect is present and the heat from the waveguide is dissipated into the SU-8 matrix.

As
Simulations of Joule heating and thermal transfer were performed using the Heat Transfer module from COMSOL Multiphysics. The change of the RI by the SU-8 thermo-optic effect is given by n(ΔΤ) = n 0 + αΔΤ, where n 0 is the RI at room temperature, α is the thermo-optic coefficient (TOC), and ΔΤ is the change of temperature caused by the thermal transfer. As a boundary condition, natural convection is established by the convective heat flux physics. We consider a 5µm-wide gold film, the thermal conductivity and the heat capacity of the SU-8 polymer are 0.3 W/(m·K) [36] and 1200 J/(Kg·K) [37], respectively. The whole structure sits on top of a 500µm-thick silicon wafer with a top layer of 5 µm thick silicon dioxide, as shown in Fig. 2(a). In the simulation, we also consider that the electrical current is flowing into the page. Figure 2(c) shows the cross-sectional temperature distribution in the nonsuspended LRSPP (NS-LRSPP) structure, for four different values of applied electrical current (4, 5, 6 and 7 mA), as indicated. From these color maps, the asymmetry of the temperature distribution due to the presence of the substrate is evident. Another perspective is shown in Fig. 2(d), in which vertical cuts of the color maps in Fig. 2(c) (at the center of the metal stripe) are plotted. It can be seen that both the temperature in the metal layer and its temperature difference with respect to the edges of the SU-8 layer (at −10 μm and + 10 μm), increase with increasing electrical current. This difference of temperature between the boundaries of the SU-8 matrix could reach 3.0 +/− 0.5 °C and consequently the corresponding RI difference in the two extremes of the SU-8 layer oscillate around 5.61 × 10 −4 (SU-8 TOC = -1.87 × 10 −4 °C −1 ) [38]. Therefore, we can notice that even for very low electrical currents the RI difference can easily reach values of 5 × 10 −4 , which is a typical value of the threshold tolerance for defining the long-range operation, i.e., if the index difference is >5 × 10 −4 the operation is considered short-range [23]. This means that active NS-LRSPP devices require longer LRSPP waveguides to apply very low electrical currents.
In order to compare the effect of our proposed structure, we perform the same numerical analysis for the suspended LRSPP structure (S-LRSPP). The suspended structure is fabricated by opening an air gap underneath the embedded waveguide such that the lower cladding resembles the thermal properties of the top cladding (air), as shown in Fig. 2(b). In this way, the effective heat dissipation properties of the bottom cladding will be similar to those from the top cladding. We emphasize that the approach followed here allows for a simple fabrication process without the need of additional steps that may include, for instance, the deposition of more layers i.e., polymer matrix sandwiched between sets of layers that allow thermal isolation from the substrate and have, effectively, similar heat dissipation properties.  The next step is the fabrication of the gold stripes. Again, to improve the surface adhesion, a 1 min oxygen plasma etching was performed. We then applied NR9-1000 PY negative photoresist from Futurrex on the SU-8 bottom cladding, by spin-coating at 4000 rpm for 40 s to obtain a final thickness of 1 µm. The sample was soft baked at 150 °C for 1 min and then exposed to UV light for 18 s at an incident power density of 12 mW/cm 2 using the proper mask to delineate the metal stripe waveguides. After the UV exposure, the sample was baked at 100 °C for 1 min and the channels were revealed by immersing the sample into RD-6 resist developer for 11 s. Once the channels for the waveguides were open, Cr (1 nm)/Au (15 nm)/Cr (1 nm) were sequentially deposited by thermal evaporation (Edwards Auto 306 with quartz crystal thickness monitor). The bottom chromium layer is needed to improve adhesion of the gold layer and the top chromium layer is deposited to maintain the symmetry of the metal layers. Both the width of the resist channels and the thickness of the metal films were carefully monitored to avoid introducing additional losses in the device. After the metal evaporation, a standard lift-off process was used to remove the remaining metal on top of the negative photoresist while leaving only the metallic waveguides.
In similar fashion to the bottom cladding, the upper cladding was fabricated by spincoating undiluted SU-8 3010 at 3500 rpm for 40 s to obtain a thickness of 10 µm. This layer was also baked at 95 °C for 7 min and then exposed to UV light for 12 s at an incident power of 12 mW/cm 2 . The same mask containing the square windows was used when exposing the upper cladding to UV light for opening the access holes through the entire thickness of the device (20 µm-thick polymer matrix). After UV exposure, the sample was baked at 95 °C for 3 min and the uncured polymer was revealed using standard SU-8 developer (1.5 min immersion). Similarly to the bottom cladding, the SU-8 upper cladding was fully cured by hard baking the sample in a step-like fashion from 100 °C to 200 °C in increments of 25 °C every minute. Recalling how sensitive the plasmonic waveguides are to asymmetries in the RI distribution, it is worth mentioning that this last hard baking on the top cladding is important to balance the properties (both mechanical and optical) of the bottom and top cladding.
Finally, the LRSPP was isolated from the substrate by removing the portion of the sacrificial layer of SiO2 that is underneath the waveguide via wet chemical etching by immersing the sample in a 6:1 Buffered Oxide Etch (BOE) solution for a duration of 45 min. to form the suspended waveguide structure. The schematic cross-section of the final suspended SU-8/Au/SU-8 structure can be seen in the last panel of Fig. 3. We should also note that for the case of the NS-LRSPP structure the square holes are not defined besides the metal stripes and thus the BOE etching does not affect these structures at all. It is also important to mention that the schematic diagrams shown in Fig. 3 (specially the last panel) are not drawn to scale and that the holes opened to access the SiO 2 layer are significantly far away from the lateral end (edge) of the metal stripe. In other words, it means that there is sufficient polymer besides the metal stripe such that the medium can be assumed to be infinite in the lateral direction.

Experime
In order to te interferometri experiment A waveguide [s through the m produce an ou RI change via heated, it will of asymmetry suspended. A asymmetric a increased.  Fig. 4(a). A ntal line) and S order to guara ctures. In Fig. to the gold co ore clearly the S nected to gold in Fig. 4(c), w m both transmis in Fig. 5   The fact t applied electr LRSPP propa while still ach structures wa shown schem used to couple used at the in waveguide in fire coupling interference p processing an Figure 6(b cases, a linear is evident tha applied power power is incre which limits experiment w in the prior ex change induc W. This corre ell as a trans Nevertheless, in appear for elec . We can also n we believe are its surroundin hmic losses. U also noted that guide (from −8 SPP (from −8 gh to extend th es. 6 [39]. More interestingly, the experiments show that the linear responses of the NS-LRSPP and S-LRSPP structures have different slopes. This reveals a second aspect of the asymmetric structure that can be understood in terms of an effective TOC exhibited by the overall structure combined with a large thermal conductivity from the Si substrate of 148 W/m·K. In the case of the S-LRSPP waveguide, the TOC of the SU-8 (-1.87 × 10 −4 °C −1 [38]) mostly determines the temperature response of the device as the polymer matrix is considered to be somewhat isolated from the substrate. In addition, since the top and bottom claddings are surrounded by air, the net heat dissipation from the polymer is reduced due to the small thermal conductivity from the air of 0.026 W/m·K and heat dissipation via convection. However, in the case of the NS-LRSPP structure the presence of a solid conductive substrate consisting of materials with positive TOC, 1 × 10 −5 °C −1 and 1.8 × 10 −4 °C −1 for SiO 2 and Si, respectively [18], can decrease the thermally induced phase shift. This is an interesting feature that allows accessing the underlying interplay between the geometry of the structure and the properties of the materials involved. On the one hand the temperature distribution is changed to a more symmetric configuration by isolating the LRSPP from the substrate at the expense of having a stronger temperature dependence; on the other hand, the temperature dependence can be significantly reduced with a substrate with large thermal conductivity combined with a proper waveguide design to allow enough overlap of the mode with substrate having a TOC of opposite sign to that of the matrix in which the waveguide is embedded at the expense of sacrificing the symmetry of the RI distribution. This can be used to optimize the performance of the LRSPP for other applications by finding the proper balance between the symmetry and the side effects from both the architecture and the material's properties standpoint.
We should also highlight that the response time of the S-LRSPP is expected to be slower than the NS-LRSPP owing to the lack of thermal conductivity in the suspended structure. Nevertheless, it may be feasible to enhance the response time by modifying the S-LRSPP while maintaining the operational RI symmetry.
As a final remark, if we set a maximum 'acceptable' attenuation to be −11 dB, as indicated in Fig. 6(a), we can observe that this threshold value is reached for an electrical power of about 15 mW for the NS-LRSPP structure and about 55 mW for the S-LRSPP. Since the slopes of the induced phase change are linear, as shown in Fig. 6(b), we can consider a figure of merit of 0.06875 π/mW·mm and 0.191 π/mW·mm, for the NS-LRSPP and S-LRSPP, respectively, for an active LRSPP waveguide of 4 mm in length. Using the electric powers of 15 mW and 55 mW, respectively, for the slopes of the phase changes, we find that in 1 mm of propagation, it is possible to have a phase change of about π for the NS-LRSPP structure while, for the same length, the phase change in the S-LRSPP is about 10π. Equivalently, this means that we can obtain a phase change of about π in 1 mm of propagation for the traditional structure or in 0.1 mm of propagation in the suspended device. This represents a scale-down factor of one order of magnitude in the required length of the LRSPP waveguide. Based on this result, one can anticipate a two-fold advantage of the proposed suspended structure: on one hand, for active LRSPP-based devices, higher integrability can be possible with waveguides that are ten times shorter; on the other hand, for LRSPP-based sensors, one order of magnitude enhancement in the sensor's sensitivity can be achieved.

Conclusions
In conclusion, we have numerically and experimentally demonstrated that the inherent losses in active LRSPP plasmonic waveguides, which are mainly due to the asymmetric distribution of temperature in the top and bottom claddings imposed by the strong influence of the semiconductor substrate, can be significantly reduced by suspending the active region. We have demonstrated such a suspended polymer-embedded plasmonic waveguide by etching a sacrificial layer (SiO 2 ) from the interface between the bottom cladding and the substrate. This approach effectively removes the strong dependence on the thermal properties of the substrate and permits having similar heat dissipation properties above and below the waveguide, which results in a more symmetric temperature distribution around the waveguide over an extended range of applied electrical power. The result is that we can use higher values of electrical power in the S-LRSPP structure while minimizing the insertion losses. This allows us to use shorter lengths of S-LRSPP structure and still obtain large enough phase changes. In the experiments, we have compared side by side our proposed suspended structure with the conventional non-suspended one. Experimental results demonstrate that the suspended structure can operate at electrical powers which are 3 times larger than the maximum that can be held by the conventional structure. Additionally, the phase change rate for the suspended structure was measured to be 2.8 times larger than that obtained for the conventional structures. In other words, we have demonstrated that by using suspended structures, the fabrication of devices operating based on thermally induced phase changes can be scaled down almost 3 times when we use the same value of electric power. However, the most important result of this work is that since the suspended structure allows us to use more electric power, it means that we can obtain the same phase change with a waveguide length that is just 10% of the length of a non-suspended structure.