Information Transfer by Near-Infrared Surface-Plasmon-Polariton Waves on Silver/Silicon Interfaces

Electronic interconnections restrict the operating speed of microelectronic chips as semiconductor devices shrink. As surface-plasmon-polariton (SPP) waves are localized, signal delay and crosstalk may be reduced by the use of optical interconnections based on SPP waves. With this motivation, time-domain Maxwell equations were numerically solved to investigate the transport of information by an amplitude-modulated carrier SPP wave guided by a planar silicon/silver interface in the near-infrared spectral regime. The critical-point model was used for the permittivity of silicon and the Drude model for that of silver. The signal can travel long distances without significant loss of fidelity, as quantified by the Pearson and concordance correlation coefficients. The signal is partially reflected and partially transmitted without significant loss of fidelity, when silicon is terminated by air; however, no transmission occurs when silicon is terminated by silver. The fidelity of the transmitted signal in the forward direction rises when both silicon and silver are terminated by air. Thus, signals can possibly be transferred by SPP waves over several tens of micrometers in microelectronic chips.

The prolonged rapid growth of the capabilities of microelectronic circuitry, forecasted about 55 years ago 1 , is beginning to slow down as the silicon-based complementary metal-oxide-semiconductor (CMOS) fabrication technology has matured. The demand for smaller and faster electronic devices appears insatiable, leading researchers to other avenues to enhance the performance of silicon microelectronics.
Devices in a circuit must be interconnected, so faster interconnections will definitely speed up microelectronic chips. Thermal radiation associated with electronic interconnections inhibits much higher data-transfer rates than presently achieved or achievable. Furthermore, electronic interconnections are limited by signal delay and low interconnection density. Optical interconnections inside microelectronic chips can have higher data-transfer rates but the interconnection density is lower than of electronic interconnections because the size of the optical interconnection must exceed the diffraction limit 2,3 .
Optical interconnections exploiting surface-plasmon-polariton (SPP) waves may help greatly enhance data transfer rates in highly integrated optoelectronic circuits on the silicon platform. The strong localization feature of SPP waves can bridge the size mismatch between transistors and optical interconnections [4][5][6] . Progress on silicon-photonics devices has been impressive for about two decades 7,8 , and the realization of silicon-based electrical sources of SPP waves using CMOS technology 9 is a significant step towards integrated optoelectronic circuits.
The propagation of an SPP wave is guided by the planar interface of a metal and a dielectric material [10][11][12] . The plasmonic component of the SPP wave results from the collective oscillation of free electrons (plasma oscillation) on the surface of the metal illuminated by an electromagnetic wave. The polaritonic component of the SPP wave results from the collective oscillation of atomic and molecular dipole moments in the similarly illuminated dielectric material. When the dielectric partnering material is homogeneous, the amplitude of the electric field of the SPP wave is maximum at the interface and decreases with the distance on the both sides of the interface. Therefore, the electromagnetic fields of the SPP waves are highly localized to the interface. Not only is this localization a desirable quality for optical sensing 13 , but it is also attractive for thin optical interconnections in microelectronic chips 5,6 . www.nature.com/scientificreports www.nature.com/scientificreports/ The excitation and propagation of SPP waves are typically analyzed after assuming that the electromagnetic fields vary harmonically in time with frequency f [10][11][12] . While this analysis is suitable for optical-sensing applications, it is unwieldy for signals that must be transported by SPP waves in optical interconnections. This is because a signal exists for a finite duration so that frequency-domain analysis of the Maxwell equations must be undertaken for a wide frequency range 14,15 . Direct time-domain analysis 16,17 using the finite-difference time-domain (FDTD) method 18,19 is straightforward for theoretical investigation of SPP-wave-based optical communication.
With this motivation, we undertook a foundational investigation and solved the Maxwell equations in the time domain to investigate the transport of information by a carrier SPP wave guided by a planar silicon/silver interface. The critical-point model was used for the frequency-domain relative permittivity of silicon 20,21 and the Drude model for that of silver 22,23 , and the frequency-domain data was converted to the time domain using the inverse Fourier transform 24 . The carrier wavelength 1200 nm c λ = was chosen to lie in the near-infrared spectral regime. In the FDTD simulation, a carrier SPP wave with amplitude modulated by a signal pulse was launched at a specific location on the silicon/silver interface and allowed to propagate a certain distance before it encountered an upright wall between silicon and another material, which could be either air or silver, as shown in Fig. 1(a,b). The transmission of the signal beyond this wall was simulated in order to determine the fidelity of information transfer beyond the wall, as assessed using the Pearson 25 and the concordance 26 correlation coefficients.
The remainder of this paper is organized as follows. First, we briefly present the geometry of the problem and the time-domain constitutive relations of the various materials involved, and then we present numerical results to determine the fidelity of information transfer.

problem Geometry and constitutive Relations
The geometry of the chosen problem is shown in Fig. 1. Relevant to the FDTD simulation, the spatial domain is The subdomain A  is occupied by a homogeneous dielectric material (silicon) and the subdomains  B and  D by a homogeneous metal (bulk silver). Information transfer by the carrier SPP wave was determined for three different materials occupying the subdomain C  : (i) air (Fig. 1a), (ii) silver (Fig. 1b), and (iii) silicon (Fig. 1c).
In order to analyze the fidelity of information transfer, we identified a point labeled R x d z

as the point of transmission and a point labeled S
in C  as the point of reception, as shown in Fig. 1.
The frequency-domain relative permittivity of silicon is described by the critical-point model as 20,21 i i Figure 1. Schematic of the computational domain of the initial-boundary-value problem for information transfer by an amplitude-modulated SPP wave guided by a silver/silicon interface across a wall between silicon and another material. The signal is launched on the plane x a = − at time t 0 = and the wall between silver and either (a) air or (b) silver is identified as www.nature.com/scientificreports www.nature.com/scientificreports/ where δ(t) is the Dirac delta, t ( )  is the unit step function, and α = + Ω − Γ ∈ . The frequency-domain relative permittivity of bulk silver is described by the Drude model as 22,23 ε ω ω In our FDTD simulations, y / 0 ∂ ∂ ≡ is set in the time-domain Maxwell curl equations. Furthermore, the amplitude of the electric field of the carrier SPP wave on the plane x a = − is modulated by the pulse function is the angular frequency of the carrier SPP wave and = × − c 3 10 m s 0 8 1 is the speed of light in free space. Thus, the electric field where u x and û z are Cartesian unit vectors along the x and z axes, respectively. The free-space wavenumber of the carrier SPP wave in free space is the complex wavenumber describing the propagation and attenuation of the carrier SPP wave along the silicon/ silver interface [10][11][12] ; and the complex wavenumbers apply. Corresponding expressions for the magnetic field x z t H( , , ) on the plane = − x a for all ∈ t T are available elsewhere 16 .
The dimension a of the computational domain and the carrier wavelength λ c have to be chosen keeping the carrier SPP wave's propagation distance Δ = q 1/Im( ) prop c along the x axis in mind. The variation of Δ prop with λ c is shown in Fig. 2. Whereas Δ prop is very small in the visible spectral regime, it is considerably higher in the near-infrared spectral regime. Moreover, λ c must be sufficiently removed from the telecommunication regime [1260,1625] nm to prevent interference, if the microchip is to be used in a telecommunication network. Therefore we chose λ = 1200 nm c for all numerical results presented here, although our qualitative conclusions apply for higher values of λ c as well. Since Δ prop = 13.21 μm for λ = 1200 nm c , we fixed = a 5000 nm. Furthermore, the dimensions b and c must be much larger than the depth of penetration of the carrier SPP wave in silver and silicon, respectively 16 , so that reflections from the planes = − z b and = z c into  are minuscule when the FDTD method is implemented. We set = b 203 nm and c 949 nm = after examining the values of α c Ag and c Si α for λ = 1200 nm c . In order to implement the FDTD method,  is discretized into Δx × Δz rectangular cells,  is discretized into linear cells of duration Δt, and derivatives are approximated using the central difference formula 18 . The physical domain  is encapsulated by a perfectly match layer to the right of the plane = x a in order to prevent reflection into  16 . Details of the perfectly matched layer and the FDTD updating equations are available elsewhere 16  so as to satisfy the Courant-Friedrichs-Lewy criterion 19 .

numerical Results and Discussion
The amplitude-modulated carrier SPP wave is launched in the computational domain × R T from the plane = − x a at = t 0, as depicted in Fig. 1. In order to determine the transmission of the information, we evaluated the temporal variation of the instantaneous Poynting vector at the transmission point R and the reception point S. Figure 3 presents the temporal profile of the axial component of the Poynting vector at the transmission point for three different values of d R when the subdomain C  is occupied by air. All plots of the components of the instantaneous Poynting vector in this paper are normalized with respect to magnitude 6.8 × 10 −6 W m −2 . The top row of the Fig. 3 shows P x z t ( , , ) R c , respectively (see Fig. 3, top row). If we quantify the signal duration as the time interval for which P x z t ( , , ) x R R exceeds 50% of its peak value, the signal duration is 12.26 fs for λ = . ) is negligbly small. Figure 3 (bottom row) shows that the reflected signal begins to appear at = t 109 fs, t 103 fs = , and t 95 35 fs = . for R c , respectively. Most of the transmitted signal is reflected by the plane x 0 = because the subdomain  C is occupied by air. Furthermore, the peak intensity of the reflected signal at R increases, if that point is chosen closer to the plane x 0 = . The dependence of the reflected signal on the material occupying the subdomain  C can be gleaned from the plots of P x z t ( , , ) Fig. 4. Each plot contains the tail of the transmitted signal followed by the reflected signal at point R. When the materials in the subdomains  B and C  are different (silicon and air, or silicon and silver), their impedance mismatch is responsible for most of the transmitted signal being reflected by the discontinuity at x z { 0, 0} = > . Figure 4 (left) and (middle) show reflection at point R when  C is occupied by air and silver, respectively. Since the metal does not allow propagation of an electromagnetic wave inside it beyond a short distance, the peak intensity of the reflected signal is found higher when  C is occupied by silver than when it is occupied by air. Furthermore, the intensity of the reflected signal in Fig. 4 (right) is zero, since there is no discontinuity at the plane x 0 = when  C is occupied by silicon. The signal received at the reception point S also depends on the material in the subdomain  C . Figure 5 provides temporal profiles of P x z t ( , , ) x S S for λ ∈ . . . d / {1 0, 2 0, 3 0} S c and 1200 nm c λ = , when C  is occupied by air, silver, and silicon. A comparison of the left panel of Fig. 4 with the top row of Fig. 5 shows that the energy of the received signal is weaker than that of the reflected signal, when  C is occupied by air. Thus, reflection by the silicon/air interface = > x z { 0, 0} is highly significant. Furthermore, since the skin depth 28 of silver is minuscule, www.nature.com/scientificreports www.nature.com/scientificreports/ the middle row of Fig. 5 shows the intensity of the received signal is infinitesimal when C  is occupied by silver. Finally, the bottom row of Fig. 5 presents the received signal at S when C  is occupied by silicon. In this case, the intensity of the received signal is stronger than when C  is occupied by air. This is because of zero reflection of the transmitted signal by the plane x 0 = , as shown in right panel of Fig. 4. The signal received at S is a distorted version of the transmitted signal at R. The top and bottom rows of Fig. 5 show that the shape of the received signal is approximately the same as that of transmitted signal but the duration is not the same. The duration of the received signal is invariant with respect to d S when C  is occupied by air, which is explained by the fact that air is a non-dispersive material (in the present context). However, the duration of the received signal increases with d S when  C is occupied by the silicon, which is consistent with the signal broadening with increasing d R observed in the top row of Fig. 3.
The signal received at point S in Fig. 5 (top row) is definitely dependent on d S , because silicon is replaced by air across the plane x 0 = . The signal received at point S in Fig. 5 (middle row) is definitely independent d S , because the chosen values of d S are significantly larger than the skin depth in silver. The signal received at point S in Fig. 5 (bottom row) is definitely dependent on d S , because this signal is simply the transmitted signal in the absence of a material discontinuity across the plane x 0 = . As the transmitted signal propagates farther, it gets distorted more because different spectral components decay at different rates due to both silver and silicon being dispersive.
The first three rows of data in Table 1 provide a comparison of the energy x E ( ) f of the forward signal and the energy x E ( ) b of the backward (i.e., reflected) signal passing through different planes = .
x const when C  is occupied by air, silver, and silicon and D  by silver, as depicted in Fig. 1a-c. The net signal energy exceeds the net signal energy at the plane x 2 5 c λ = − . , regardless of the material occupying  C . This is because E b also diminishes with the propagation distance (in the backward direction).  reaches the plane λ = x 3 c , i.e., a distance of 9 μm or about 650 transistors of linear size 14 nm laid end to end. That number of transistors will increase in a few years. Even when C  is occupied by air, about 30% of the forward signal energy at λ = − . x 3 5 c reaches the plane λ = x 3 c . Of course, reflection is maximum when C  is occupied by silver.
Suppose in Fig. 1, if silver is abruptly terminated at x 0 = and both subdomains  C and D  are occupied by the same material. When this material is silver (Fig. 1b), practically no energy passes forward through the plane x 0 = + , as is clear from the middle row of Fig. 5 and the second row of data in Table 1. When the material occupying C D ∪   is air, the fourth row of data in Table 1 indicates that energy does pass forward through the plane x 0 = + and decays very slowly as x increases. When the material occupying C D ∪   is silicon, the fifth row of data in Table 1 indicates that even more energy passes forward through the plane = + x 0 but decays more rapidly as x increases.
The plane = + x 0 can be considered to be the virtual source of energy that passes into ∪ C D   . This source is of limited extent along the z axis because of the localization feature of SPP waves. As such, the Huygens' principle indicates that the energy would not pass only in the forward direction but also in other directions, as we   www.nature.com/scientificreports www.nature.com/scientificreports/ verified from our simulations. Therefore, it will decay in all directions. When C D ∪   is occupied by silicon rather than air, the decay rate is higher because silicon does have some dissipation because Im( ) 0 Si ε >  while air is assumed to be non-dissipative.
Reverting to the three situations depicted in Fig. 1a-c, we determined the Pearson 25 ( P RS ρ ) and the concordance 26 ( C RS ρ ) correlation coefficients, between the transmitted and received signals in order to quantify the similarity of the signal received at point S to the signal transmitted at point R.
The Pearson correlation coefficient [ 1,1] P RS ρ ∈ − compares the shapes and durations of two signals. This coefficient is defined as indicates stronger correlation or anticorrelation, as indicated by the sign of P RS ρ . If the received signal is independent of the transmitted signal, then 0 P RS ρ = . The concordance correlation coefficient compares the shapes of two signals and is defined as 26 www.nature.com/scientificreports www.nature.com/scientificreports/ duration of the received signal does not equal that of the transmitted signal, as indicated by the top and bottom rows of Fig. 5, we get the moderate value of ρ ≈ . 0 5 P RS . When both C  and D  are occupied by silver, the signal does not travel for an appreciable distance beyond the plane x 0 = + , as is clear from the middle row of Fig. 5, and both correlation coefficients become meaningless.
Both the shape and the duration of the received signal at point S are changed, if silver is abruptly terminated at = x 0 and both C  and D  are occupied by either air or silicon. Table 2 shows ρ P RS and C RS ρ are ~0.62 for all chosen values of d S when ∪   C D is occupied by air. The values of ρ P RS and ρ C RS indicate that the received signal's shape and duration are close to those of the transmitted signal, which is not surprising because air is nondispersive. The values of ρ P RS and C RS ρ are lower (~0.57) when   ∪ C D is occupied by silicon, because silicon is slightly dispersive in the near-infrared regime.

concluding Remarks
Motivated by the need for optical interconnections in silicon microelectronics, we numerically investigated the transfer of information by the amplitude modulation of a surface-plasmon-polariton wave guided by a silicon/silver interface, the carrier frequency lying in the near-infrared regime with dissipation of electromagnetic energy in silicon. As the signal pulse propagates guided by the silicon/silver interface, its temporal profile broadens and its amplitude reduces. The broadening is consistent with different spectral components of the signal having different phase speeds, and amplitude reduction occurs due to dissipation of the electromagnetic energy in both silver and silicon. The loss of fidelity, as quantified by the Pearson and concordance correlation coefficients, is not excessive.
The signal is partially reflected and partially transmitted without significant loss of fidelity, when silicon is terminated by air; however, no transmission occurs when silicon is terminated by silver. The fidelity of the transmitted signal in the forward direction rises when both silicon and silver are terminated by air.
Thus, our foundational investigation shows that the information can be transferred over distances on the order of a few tens of micrometers in microelectronic chips by SPP waves. Even if the metal is abruptly terminated, information continues to propagate in the forward direction. Effective strategies to reduce reflections do need to be devised, but even so our results are promising for SPP-wave-based optical interconnections because planar structures for propagating SPP waves for even longer distances have been devised and fabricated 29 .