Our device is composed of Si stripe and Au waveguide, as the schematic diagram shows in Fig. 1a. Although there are inherent losses in the top-down preparation of plasmonic waveguide, it has great potential in the field of optoelectronic integrated chips and plays an important role in the process of small footprint and high speed of optical communication device due to its simple and controllable fabrication. Surface plasmon polaritons (SPPs) are excited by focusing the incident laser on the grating region, then propagate to the terminal of Au waveguide. The near-field coupling between propagating SPPs and the waveguide modes in silicon is efficient when the gap between them is small and optimized. With external electric field, the electrical detection is achieved by the photocurrent response with photon induced conductance increasing in Si stripes. Thus from the change of photocurrent in silicon stripe, the propagating surface plasmons could be detected under external bias, which process is described as photoconductive effect. Figure 1b shows the SEM image of our prepared sample. The silicon stripe (220 nm thickness, \(\text{2.9 }\text{μm}\) width) is placed on the \(\text{Si}{\text{O}}_{\text{2}}\) substrate, and in close proximity to the Au waveguide (90 nm thickness, \(\text{6.5 }\text{μm}\) length). We first measured the on-chip electrical detection of SPPs in gold stripes and then moved to the subwavelength gold nanowire. This is because the propagation length of SPPs in the gold stripe is much longer, which reduces the difficulty of the experiment.
Coupling structure optimization.
The efficient near-field coupling between propagating SPPs and the waveguide modes is achieved by optimizing the gaps. We focus a 710 nm laser on the grating region to excite propagating surface plasmons on the surface of Au waveguide. In Fig. 1c, scattering photons are observed to come out in the other side of Si stripe rather than at the junction between stripe and waveguide. This interesting phenomenon indicates that the emission photons induced by plasmon radiative decay are coupling into the Si stripe through the near field. The fine gap size control is achieved by controlling the distance between electron beam lithography (EBL) exposure patterns to compensate the lateral errors due to development and RIE process. The corresponding simulation of the far-field image is shown in Fig. 1d, which confirms the coupling in silicon. In the image, some photons are scattered to the far-field at the junction due to the simulation model, but it is still to see that the photons coupled into silicon are dominant. As the wavelength changes, the scattered photons in the terminal of silicon also change accordingly, and we also observed in the experiment. In contrast, a serious leaking of light from the gap was observed for large gap (> 100 nm) or overlapping as shown in Fig. 2c. The detailed simulation of gap dependent absorption in silicon is shown in Fig. 2a, the absorption in Si (indicated by the color bar) is maximum with the gap about 10 nm, then decreases with larger gap size. When the gap size is smaller than 3 nm, light from 740 nm to 780 nm cannot be coupled. A gap of about 10 nm is also advantageous for experiment, because the process precision is limited, attempts to prepare a 0 nm gap will cause the Au waveguide to overlap on the Si, thereby affecting the coupling.
Another important factor for optimizing coupling is the relative height control between the gold SPP waveguides and the Si photonic waveguides. When the height of Au waveguide and Si stripe is close, almost all the photons are scattered at the junction, which suggests that no optical field could be coupled into the silicon stripe. In optoelectronic characterization of these samples, no electrical signal could be detected. When the height of Si waveguides is 220 nm, twice as the Au waveguide, we see the optimized coupling. The optimized geometry parameters for best coupling are verified by the FDTD simulation. Figure 2b shows the results of height control simulation with a fixed 90 nm Au waveguide, a threshold is found in the silicon thickness (90 nm), below which coupling efficiency drop sharply. From the results, the absorption in silicon stripe increases when it becomes thicker, so in our study, 220 nm thickness of silicon stripe ensures the best coupling efficiency. The electric field distribution dependence with Si thickness is shown in Fig. 2d. In 220 nm Si, a considerable field distribution is observed in Si with zigzag patterns. When the silicon stripe thickness is 110 nm, a large amount of scattered photons leak at the junction in the right upper and lower right directions, little energy is coupled into the stripe.
Absorption analysis in silicon strips.
In order to study the influence of the parameters of the detector model on the efficiency, we calculated the efficiency of the absorption of plasmon energy by the silicon strip under different conditions using the COMSOL simulation method. We change the thickness and width of the silicon strip, the length of the gold waveguide, and the size of the gap between the gold waveguide and the silicon strip. Here, we only consider the upper surface plasmon modes of the gold waveguide and the wavelength is 700 nm, in fact, we have similar conclusions for the modes on the lower surface. The absorption efficiency of silicon is represented by \({\text{η}}_{\text{Si}}\text{ }\text{/}\text{ }\text{(1-}{\text{η}}_{\text{Au}}\text{)}\), where \({\text{η}}_{\text{Si}}\) represents the ratio of the absorption power of silicon to the power of the excitation light, and \({\text{η}}_{\text{Au}}\) represents the absorption ratio of the gold waveguide. This absorption efficiency thus represents the energy conversion efficiency from the gold waveguide into the silicon strip. The strip has a higher absorption efficiency for the thicker silicon strip, as show in Fig. 3a, this is because the effective mode index of the silicon strip changes as the thickness increases. Although we use SOI substrates with 220 nm top layer silicon in our experiments, it can be seen from the figure that when the thickness of the silicon strip is greater than 200 nm, the corresponding absorption efficiency is above 52%. In fact, for sufficiently thick (above 600 nm) silicon strips, the absorption efficiencies close to 72%. The gap width between the silicon strip and the gold waveguide also matters to the coupling efficiency. Figure 3b shows that the smaller gap always lead to higher absorption efficiency, because the plasmon energy is more easily coupled into the silicon strip. As the gap decreases, the energy scattered into the space decreases, the coupling efficiency increases.
When the thickness of the silicon strip is fixed at 220 nm, the absorption efficiency increases with the width, as shown in Fig. 3c, this is because the wider the silicon strip is, the less energy scattered from the other end, which means the wider silicon strip is better. However, it can be seen from the trend in the figure that when the width of the silicon strip is above \(\text{3 }\text{μm}\), the increase rate decreases, and the strip has a absorption effect greater than 55%. So we used \(\text{2.9 }\text{μm}\) wide silicon strips in the first experiment. We also studied the effect of different length of the gold waveguide on the absorption efficiency (Fig. 3d). In fact, the change of the waveguide length has little effect on the absorption efficiency, so our detector is independent of the size of the device, which makes our detectors can be used in plasmonic components of different structures for optoelectronic integration. These simulations results for the absorption efficiency of the silicon strips explain our previous experimental results well, confirming our optimization of the structural parameters of the detector model. The higher absorption efficiency is important for photoelectric conversion efficiency, which is reflected in the larger photocurrent measured in the experiment.
The proofs of the SPPs detection.
The plasmon detection is verified by the polarization and wavelength sensitivity of the photocurrent. The polarization dependent photocurrents are measured (Fig. 4a). In the dark situation, the current driven by external bias 20 mV voltage is 27 pA, while the photocurrent varies as the cosine function of twice the polarization angle of incident laser. The contrast of polarization dependent photocurrent \({\rho }=7.6\) (\(\text{ρ=}\frac{{\text{I}}_{\text{max}}-{\text{I}}_{\text{dark}}}{{\text{I}}_{\text{min}}-{\text{I}}_{\text{dark}}}\)), representing a strong modulation by controlling the plasmon excitation. This is due to the efficient SPPs coupling when the excitation polarization is p-polarization (electric field vertical to the grating slits), and almost no coupling when that is in s-polarization (electric field parallel to the grating lines). Another proof of the SPPs detection is the wavelength dependence of the photocurrent shown in Fig. 4b. The incident laser wavelength is continuously tuning by the supercontinuum laser and AOTF. The photocurrent is normalized by photocurrent response with laser directly illuminating on Si to eliminate the effect of changes in laser intensity. The signature Fabry-Perot oscillation is found in the photocurrent–wavelength relation, which existing in the SPPs waveguide. This Fabry-Perot behavior is due to the interference of forward and backward propagation SPPs. The beating wavelength interval is sensitive to the wavelength and the length of the stripes, and the propagating length of the SPPs affects its appearance. In our measurement, the beating phenomenon is clear. The beating interval 23 nm is fitted well with the FDTD simulation (Fig. 4c). The corresponding electrical field in peak and valley of Fabry-Perot oscillation are calculated by FDTD solutions. From Fig. 4d, we find that standing wave forms in the electric field distribution on Au waveguide, and there is a zigzag pattern inside Si stripe. The optical energy inside Si with 714 nm laser is obviously stronger than 766 nm laser, which confirmed the photocurrent profile. The polarization dependence and the Fabry-Perot behavior in photocurrent spectra proves the on-chip SPPs detection in our devices.
Detection of SPPs in subwavelength plasmonic waveguide.
Then we moved to the subwavelength propagating surface plasmon on-chip electrical detection. The first issue we solved is the improvement of the propagation length of the SPPs in the top-down fabricated gold nanowires. One reason is that the surface roughness increases the loss of plasmon propagation. Another reason is that size restrains the possible SPP modes existing in nanowire waveguide, and the substrate leaky modes. We solved this by thermal annealing and inserting a high refractive index optimized Si layer between gold nanowires and the SiO2 substrate. The size of Au nanowire (280 nm wide and 130 nm thick) is comparable with the mainstream use in plasmonic circuits45. Next we integrated the subwavelength Au/Si nanowires with the on-chip silicon detectors. The SEM image of the device is shown in Fig. 5a, the device is composed of a 600 nm wide Si nanostripe and the Au/Si nanowire SPP waveguides. From the dashed orange rectangle inset, the SEM image confirms the good alignment structure with nanoscale gap between Au nanowire and Si stripe.
With incident 685 nm laser whose polarization parallel to the Au nanowire, we observe the good coupling with a signature O-shaped ellipse scattering pattern by far-field CCD imaging (Fig. 5b). In contrast, the bad coupling will cause a bright spot in the gap region in far-field CCD imaging. The height of Au nanowire (180 nm, including the inserted Si layer) is optimized and lower than Si stripe (220 nm). The polarization and wavelength dependence of photocurrent are measured. The photocurrent is maximum when the polarization of incident laser is parallel to Au nanowire, with the symmetric \(\text{m=0}\) plasmon mode is excited. While the polarization of incident laser is perpendicular with Au nanowire, we can also detect a net current which suggesting the excitation of \(\text{m=1}\) plasmon mode. The polarization dependence may be used in detect the plasmon router or logic circuit17, where electrical detection of different plasmon modes is needed. The relation of normalized photocurrent with laser wavelength is studied in Fig. 5e, several characteristic peaks are observed. Compared with the photocurrent dispersion spectra of the previous device, the period of the photocurrent oscillation is not very regular reflecting a low quality factor Q of the plasmon cavity. Due to the increased plasmonic loss of the nanowire and photon leakage in the junction of the coupling structure, less than 10% of the plasmon energy is reflected in the terminal of Au nanowire, leading to the experimental results.
Efficiency evaluation.
The device is characterized by optoelectronic platform, the intrinsic dark current response under 20 mV external bias is 21 pA, while the photocurrent under 20 mV bias with 685 nm laser excitation is 74 pA (Fig. 5c), and the switch ratio of photocurrent is 3.5. The noise equivalent power (NEP) is the incident laser power when the net photocurrent is equal to the dark current. The NEP of our device is \(\text{0.14 }\text{μW}\), in this power the CCD camera (\(\text{33 }\text{ms}\) integration time) cannot recognize the scattering photons in the terminal of the nanowire which demonstrate the near-field electrical detection is more efficient than the far field photodetector as SPP detector. The polarization dependence of net photocurrent is measured (Fig. 5d), obeying \({\text{I}}_{\text{net}}\left(\text{pA}\right)\text{=37.7+17.1}\text{cos}\text{2θ}\) where \(\text{θ}\) is the angle between Au nanowire and the polarization of incident laser. Figure 5f shows a linear relationship between photocurrent and incident laser power, which suggests a linear photon absorption efficiency and a intrinsic gain induced by external bias. We divide the whole detection process into two steps. In the first step, the incident laser excites propagating surface plasmon on the Au nanowire and decay as electromagnetic field distribution near the Si nanowire. The second step is photodetection, the near field photons are absorbed in Si then driven by external bias forming a photocurrent. For the carrier multiplication is excluded, the internal quantum efficiency and the external quantum efficiency could be estimated, where the latter considers the energy loss in plasmon excitation, SPP propagation and scattering efficiency. The net photocurrent between the electrodes is
Where \(\text{η}\) is the internal quantum efficiency, \(\text{α}\) is efficiency calculated from incident laser to the near-field surrounding Si nanowire. \(\text{τ}\) denotes the carrier lifetime, estimated as \(\text{10 μs}\)46, \(\text{ε}\) is the electric field in the Si nanowire, and \(\text{L}\) is the length between electrodes. The external quantum efficiency \(\text{α∙η}\) is calculated as \(\text{7.12×}{\text{10}}^{\text{-7}}\), considering a plasmon excitation efficiency as \(\text{~0.01}\)27, \(\text{~0.1}\) of the plasmon energy could propagate to the terminal of Au nanowire, \(\text{~0.1}\) of the plasmons are reflected by terminal. The simulation suggests \(\text{~0.01}\) of the near field photons is absorbed by Si stripe. The internal quantum efficiency of the photodetector \(\text{η}\) is calculated as \(\text{~7.9\%}\), the corresponding responsivity of photocurrent is \(\text{0.14}\text{ }\text{mA/W}\).