1. Introduction

The field of plasmonics offers unique capabilities for manipulating and controlling the propagation properties of electromagnetic radiation [1]. When metal films are appropriately structured, surface plasmon-polaritons (SPPs) can be excited from free-space radiation and propagate along metal-dielectric interfaces. A unique property of these excitations is that they exhibit dispersion properties that differ dramatically from that of free-space radiation, thereby enabling a broad range of new capabilities including subwavelength concentration of radiation [2,3], guided-wave propagation [4], enhanced extraction of light from photovoltaic devices [5] and optical filtering [6,7]. At present, much of this work has focused on passive demonstrations. However, there is great need for the development of active devices that can fulfill required needs across the electromagnetic spectrum. In the THz spectral range, such needs are particularly acute, since workable device technologies are largely absent.

In recent years, there have been a number of studies describing active control of the propagation properties of SPPs using a variety of different techniques. The most common approach involves coupling SPPs into optically active materials, such as photochromic molecules [8], semiconductor device structures [9] and liquid crystals [10]. However, other approaches have been developed that involve the use of transient optical nonlinearities [11,12], application of an external magnetic field [13], and pre-defining the phase properties of the incident optical radiation [14]. Recently, we showed that by injecting eutectic gallium indium (EGaIn), a liquid metal at room temperature, into a polydimethylsiloxane (PDMS) mold, we could create a periodic array of subwavelength apertures that was mechanically flexible and reversibly deformable [15]. Thus, the resonant frequencies could be tuned by mechanically stretching the device. In these demonstrations, the amplitude or frequency properties are typically tuned away from an equilibrium point via an external stimulus. However, it is not at all clear how the device geometries in any of these implementations could be made reconfigurable.

In this submission, we demonstrate a technique that allows for reconfigurable changes in the geometry of a plasmonic device, corresponding to large-scale changes in the optical response. We use conventional microfluidic technology to create channels in an elastomeric mold, where EGaIn can be injected into or withdrawn from. As a specific example, we have fabricated a bullseye structure [16, 17] in PDMS that is adhered to a gold (Au) coated metal foil that has a single subwavelength aperture. We measure the THz transmission properties as a function of the bullseye geometry, based upon which channels are filled with EGaIn. It should be noted that the present demonstration differs significantly from our earlier work with bullseye structures that were fabricated in free-standing stainless steel foils [18, 19]. In those cases, the geometries were fixed. Based on the present measurements, we develop a simple model that accounts for the temporal properties of the observed waveforms. This model differs from what we have previously observed with bullseye devices fabricated in metal foils [18, 19]. All of the measured data was performed using a single device in which different channels were filled with or emptied of EGaIn to obtain different geometries, demonstrating true reconfigurability. Results from different devices exhibited nearly identical results.

2. Experimental detail

The basic component of a reconfigurable bullseye structure is a replica fabricated in PDMS. In order to create this replica, we first fabricated a bullseye structure in a 150 µm thick free-standing stainless steel foil using a simple chemical etching process. The structure consisted of a series of concentric rectangular cross-section annular grooves that were 600 µm wide and 100 µm deep with a periodic spacing of 1 mm. We then prepared a PDMS pre-polymer that was mixed with a curing agent using a weight ratio of 11:1. This pre-polymer was degassed, poured onto the stainless steel bullseye structure, and cured for 2 hours at 60 °C yielding a 500 µm thick film. Since PDMS does not adhere well to untreated metal surfaces, which results from the low surface energy of the elastomer, the replica could be peeled off the metal master easily. Cured PDMS films can be adhered to select materials after being treated with a high voltage corona. Unfortunately, this process also does not work with most metals. Since the goal here is to use a metal substrate, an alternate procedure had to be used. In order to prepare the planar metal substrate, we initially used a free-standing planar 75 µm thick stainless steel foil with a single 490 µm diameter circular hole milled using a tripled Nd:YAG laser. This foil was then coated with a 10 nm thick Ti film, as an adhesion layer, and a 200 nm thick Au film. We then treated the Au layer with a monolayer of 3-mercaptopropyl trimethoxysilane (MPT). MPT creates a Au-Si bond on the metal surface that allows for preferential binding with PDMS. However, this approach does not still not work well for cured PDMS surfaces. Therefore, we applied a thin layer of uncured PDMS (<10 µm) to the bottom layer (not within the channels) of the cured, corona treated PDMS replica. After curing the entire device for two hours at 60 °C, good adhesion between the elastomeric bullseye structure and the Au coated metal foil was obtained.

As noted above, PDMS has a relatively low surface free energy (19.8 mN/m [20]). Therefore, when EGaIn is injected into a channel, the liquid metal does not adhere well to the PDMS sidewalls. However, it does adhere to the silanized Au surface. In order to minimize any liquid metal/Au or liquid metal/PDMS adhesion issues, which is necessary to reproducibly inject and withdraw EGaIn from a channel, we injected a fluorosilane solution (PFC504A-FS, Cytonix) into all of the channels and baked the device at 60 °C for an additional 2 hours. This reduced the surface free energy to ~6 mN/m on all four channel walls. In Fig. 1(a) , we show a schematic diagram of a final structure, where one channel is filled with EGaIn. The accompanying photograph in Fig. 1(b) shows an actual device in which EGaIn has been injected into the third annular channel. A syringe with a 33 gauge needle (200 µm outer diameter, 89 µm inner diameter) was used to manually inject and withdraw the liquid metal.

 

Fig. 1 (a) Schematic cross-section of the final bullseye device. The parameters for the device are d₁ = 400 µm, d₂ = 1 mm and d₃ = 490 µm. h₁ = 500 µm, h₂ = 250 nm, and h₃ = 75 µm, respectively. MPT is a silane that increases the adhesion between PDMS and the metal substrate. (b) Photograph of a bullseye pattern, in which EGaIn was injected into third annular channel.

Download Full Size | PDF

We used conventional time-domain THz spectroscopy to characterize the bullseye structures and bare apertures. In this approach, the time-domain properties of a single cycle electromagnetic transient transmitted through a structure can be measured using coherent detection with subpicosecond temporal resolution. Photoconductive devices were used for both emission and coherent detection. An off-axis paraboloidal mirror was used to collect and collimate the THz radiation from the emitter to the device. The THz beam, with a 1/e beam diameter of ~15 mm, was normally incident on the corrugated surface of the bullseye structure. It is important to note that the frequency content of the THz beam varies spatially, thus the temporal properties of the incident THz pulses are also spatially dependent.

3. Experimental results and discussion

In order to properly characterize the reconfigurable bullseye structures, we first measured the temporal properties of several different device structures to clarify the role of each component within the bullseye structure: (1) a single aperture in the metal foil without an attached PDMS mold, (2) a single aperture with an attached PDMS mold but without EGaIn, and (3) a single aperture with an attached PDMS mold and EGaIn injected into the second annular channel. The three corresponding waveforms are shown in Fig. 2 . The transmitted waveform for the single aperture without the PDMS replica is characterized by a single bipolar THz waveform. This waveform is similar, though slightly narrower, than the incident THz waveform (not shown). The slight temporal narrowing arises from the fact that low frequency components within the incident THz beam that lie below the cutoff frequency of the circular aperture experience greater loss than higher frequencies. For a 490 µm diameter circular aperture, the corresponding cutoff frequency occurs at ν = 0.36 THz (λ = 833 µm). Since the metal foil is only 75 µm thick, the electric field is only slightly attenuated at frequencies below cutoff. This is consistent with earlier observations with single subwavelength apertures [18, 21]. It should be noted that in the absence of the aperture, there was no transmitted THz radiation.

 

Fig. 2 Measured time-domain waveforms for the single aperture without PDMS (black), single aperture with only the PDMS bullseye replica (red) and single aperture with EGaIn injected into the second annular channel of the PDMS bullseye structure (blue). The waveforms are vertically offset for clarity.

Download Full Size | PDF

For the single aperture with an attached PDMS mold and no EGaIn in any channel, there are two notable observations. First, we do not observe any apparent oscillations after the initial bipolar waveform. This demonstrates that the structured PDMS mold does not appreciably couple freely propagating THz pulses to SPPs by itself. Second, the observed bipolar waveform exhibits a smaller peak-to-peak amplitude and is broadened slightly in time relative to the waveform associated with the single aperture. This can be explained by considering the absorption properties of PDMS in the THz spectral range. We define a frequency dependent absorption coefficient, α(ν) = 2πνn_i/c for the THz electric field, where the electric field decay is given by exp[-α(ν)d]). Here, n_i is the imaginary component of the refractive index, c is the speed of light in vacuum and ν is the THz frequency. We have previously found that that the complex refractive index of PDMS in this spectral range is given approximately by n = 1.57 + 0.04i and is relatively constant over the frequency range spanning 0.1 - 0.5 THz [15]. Therefore, in contrast to the aperture, which preferentially suppresses low frequencies, the PDMS layer preferentially suppresses higher frequencies. Both the amplitude reduction and pulse broadening can be properly accounted for, if we consider absorption within the PDMS layer. In the discussion that follows, we will refer to this waveform as the ‘reference.’

For the bullseye pattern with one annular channel filled with EGaIn, we observe the initial bipolar waveform followed by a single time-delayed oscillation. This waveform is similar in nature to that observed for a subwavelength aperture surrounded by a single concentric groove [18]. This is reasonable given theoretical analyses showing that metallic protrusions on top of the surface of a metal film scatter (couple) SPPs with efficiencies similar to that of grooves [22]. In contrast to earlier time-domain measurements on bullseye structures [18, 19], there are important differences dictating the temporal properties of the time-delayed oscillations here.

In order to elucidate these properties, we successively inject EGaIn into only one channel at a time, ensuring that all of the EGaIn has been withdrawn from the other channels. We then subtract the time-domain waveform associated with each resulting structure from that obtained with the reference aperture. In Fig. 3(a) , we show the temporal waveform associated with the reference aperture, along with each of the subtracted waveforms. It is clear that as the distance between the filled annular channel and the central aperture increases linearly, the time delay between the initial bipolar waveform and the time-delayed oscillation also increases linearly. This is consistent with our earlier finding that there are two independent, yet phase-coherent, transmission processes that contribute to the transmitted time-domain waveform [18]: one component related to transmission directly through the subwavelength aperture and a time-delayed component associated with coupling of free-space THz radiation to SPPs by the EGaIn annular ring. These coupled surface waves propagate towards and are sampled by the aperture. The extent of the sampling, corresponding in part to the magnitude of the observed time-delayed oscillation, is determined by the overlap between the spatial extent of the SPP wave and the subwavelength aperture.

 

Fig. 3 (a) Measured time-domain waveforms for the reference aperture and the contribution from each individual EGaIn filled channel (Ring 2 through Ring 5). The time-delayed oscillations were obtained by subtracting the time-domain waveform of the reference aperture from the waveforms associated with structures in which successive single annular channels were filled. The waveforms are vertically offset for clarity. (b) The corresponding normalized amplitude spectra of the reference aperture (red) and the time-delayed oscillation from Ring 3 (black).

Download Full Size | PDF

The measured time interval between successive oscillations (between Ring i and Ring i + 1, for i = 2-4 in Fig. 3(a)) is ~4.5 ps. Analytically, we find that this time delay, Δt, is given by

As with grooves fabricated into a metal foil, the EGaIn protrusions appear to also couple most of the incident THz to SPPs. To demonstrate this, in Fig. 3(b) we show the amplitude spectra for the initial bipolar pulse (reference aperture) along with the oscillation associated with Ring 3. While the spectra differ slightly, it is important to note that the frequency content of the THz beam varies spatially; therefore, the temporal properties of the incident THz pulse are also spatially dependent. In general, the THz beam has a frequency dependent beam diameter, with higher frequencies more closely concentrated near the beam axis.

Based on this insight, we are now in a position to fully exploit this capability to create reconfigurable bullseye devices of greater complexity by selectively injecting and withdrawing EGaIn into different channels to obtain the desired temporal response. In Fig. 4 , we show two sets of time-domain waveforms for three different bullseye geometries: (i) Rings 3 and 5 filled (ii) Rings 3-5 filled and (iii) Rings 2-5 filled. In each case, we show the experimentally measured waveform (blue) and a waveform (red) that has been reconstructed from the data in Fig. 3(a). The waveform reconstruction was performed by taking a superposition of the time-domain waveform associated with the reference waveform and that of the relevant filled rings. However, in contrast to what we have observed with stainless steel bullseye devices [18], the oscillations associated with a specific ring clearly change in timing and, to a lesser extent, amplitude depending upon which other rings are filled.

 

Fig. 4 Measured and reconstructed time-domain waveforms for bullseye structures with multiple filled annular channels. The blue waveforms corresponds to experimentally measured time-domain waveforms, while the red waveforms are reconstructed by taking a superposition of the relevant components in Fig. 3(a), with minor modifications, as described in the text. (top) Bullseye with Rings 3 and 5 filled with EGaIn (middle) Bullseye with Rings 3, 4 and 5 filled with EGaIn (bottom) Bullseye with Rings 2, 3, 4, and 5 filled with EGaIn. The waveforms are vertically offset for clarity.

Download Full Size | PDF

In order to explain why this happens, consider a bullseye structure in which only Ring 5 is filled with EGaIn. The resulting time-domain waveform can be reconstructed from the data in Fig. 3(a) using the reference aperture and Ring 5 waveforms. Now, suppose that we also fill Ring 3 with EGaIn. In this case, the resulting time-domain waveform cannot simply be reconstructed from the waveforms corresponding to the reference aperture, Ring 3 and Ring 5. By virtue of filling Ring 3, the temporal properties of the oscillation corresponding to Ring 5 will change. The reason for this timing change is because instead of having 4 sections of length d₁ with a refractive index of n_SPP1≈n_air = 1 and 4 sections of length (d₂-d₁) with a refractive index of n_SPP2≈n_PDMS = 1.57, by filling Ring 3, the oscillation arising from Ring 5 will now see 3 sections of length d₁ with a refractive index of n_SPP1≈n_air = 1, 1 section of length d₁ with a refractive index of n_SPP2≈n_PDMS = 1.57 and 4 sections of length (d₂-d₁) with a refractive index of n_SPP2≈n_PDMS = 1.57. Thus, the inclusion of a single filled annular channel closer to the aperture will push the temporal contribution from an outer filled channel to a larger time delay, Δτ = d₁ (n_SPP2 - n_SPP1) / c = 0.76 ps. As additional inner rings are filled with EGaIn, oscillations associated with outer filled rings will be pushed to slightly larger time delay values. These expected differences in timing match the experimental data well.

In addition to shifts in timing caused by filling empty inner channels, it is reasonable to assume that there would also be changes in the spectral content of the individual oscillations associated with the outer rings. Such changes would arise because of differences in the loss properties between SPP propagation along a metal-air interface and a metal-PDMS interface. However, changes in the temporal properties of outer ring generated oscillations are smaller than expected from the additional d₁ traversals through metal-PDMS sections as inner channels are filled.

4. Conclusion

In conclusion, we have demonstrated a means for creating reconfigurable plasmonic devices using liquid metals. As a specific example of this capability, we fabricated an elastomeric bullseye pattern using soft lithography techniques and bonded it to a metal substrate after appropriate surface treatment of the metal. By injecting and withdrawing a liquid metal, eutectic gallium indium, into or from individual channels, we are able to dramatically alter the device geometry. We note that EGaIn can only be withdrawn from a channel when all four interior channels walls are coated with a fluorosilane, which dramatically reduces the free surface energy and inhibits EGaIn from sticking to the PDMS and metal surfaces. We developed a simple description that accounts for the timing of all of the relevant time-domain oscillations. While the present device requires manual operation to inject and withdraw the liquid metal, well-developed micro-electromechanical systems (MEMS) technology can be used to fabricated micro-actuators and micro-pumps directly on the same substrate to enable significantly higher speed reconfigurability [24]. As an example, a silicon-based magnetohydrodynamic pump has been shown to effectively actuate liquid metals [25]. Plasmonic devices, such as the one described here, could be fabricated on the same substrate, allowing for reconfigurability using appropriate electrical signals to drive the micro-pumps. Thus, this approach offers significant promise for developing active devices. Finally, we note that that this general idea may be extended to develop active device that can operate in other regions of the electromagnetic spectrum. However, new challenges may arise. For example, as one moves to much shorter wavelengths, which requires smaller channel dimensions, use of syringes or micro-pump technology may not be viable to inject liquid metals, because significantly higher pressures may be needed to inject and withdraw liquid metals [26]. Thus, alternate approaches may need to be explored.