1. Introduction

Nanometallic structures are key to the conversion of light to surface plasmon-polaritons (SPPs) localized to small volumes. They are promising for several applications including telecommunications, solar cells and biosensing as they can provide highly enhanced fields, strong confinement (to sub-wavelength scales) and high bulk and surface sensitivities [1, 2].

A Schottky barrier photodetector formed at the interface between a metal and a lightly doped semiconductor can be used to detect radiation below the bandgap energy of the semiconductor via internal photoemission (IPE) [3–5]. Several Schottky photodetectors involving SPP excitations on the metal contact have been reported [6–8], including grating-coupled detectors [9, 10], waveguides [11, 12], and nanoparticles and nanoantennas [13–17], building on early work involving prism-coupling to SPPs on a suitable detection structure [18]. A grating structure can be used as an alternative to a prism to increase the in-plane momentum of the incident light in order to match that of the SPP, providing compactness and manufacturing advantages, and, potentially, high sensitivity. Metal-semiconductor Schottky grating structures have been investigated as SPP photodetectors [19, 20] and optimized for slit pitch and film thickness [10, 21].

Although studies have been carried out on various aspects of gratings and their detection applications, achieving high infrared sub-bandgap responsivities over a broad spectral range on low-cost Si remains a difficult and worthwhile challenge. In this paper we propose a novel plasmonic Schottky photodetector consisting of a Au grating on a thin metal patch on p-type Si (p-Si), providing detection via IPE of hot holes generated in the patch directly along the Schottky contact. We report infrared responsivities among the largest achieved to date. Given the small area of each detector, thousands can be densely integrated into arrays over the surface of the substrate. Packaging is simplified because butt-coupling and propagation through end-facets are not required, and simpler packages suitable for surface devices can be used [22]. Potential applications include infrared imaging, biosensors and optical receivers for interconnects [23, 24] and short-reach communications [25].

2. Proposed Structure

Figure 1(a) shows a schematic of the proposed structure. It includes a metal patch (thickness t) for guiding SPPs on a lightly-doped p-Si epitaxial layer on a heavily doped p-Si substrate. A metal grating on the metal patch is designed to couple normally-incident p-polarised infrared (λ₀~1550 nm, sub-bandgap) light to SPPs propagating along the bottom surface of the patch such that the SPP fields are coincident with the Schottky contact. As SPPs propagate they are absorbed in the metal, generating hot carriers therein primarily along the Schottky contact. Detection then occurs through the collection of hot holes in p-Si via IPE. The substrate is placed on a backside Al Ohmic ground contact.

 

Fig. 1 (a) Cross-sectional sketch of the proposed Au/Si Schottky surface plasmon photodetector. The structure comprises a metal patch of thickness t on p-Si, with a metal grating consisting of rectangular ridges of width W and thickness H arranged periodically in pitch Λ. The materials considered are Au for the metal, p-Si for the epitaxial layer and p⁺-Si for the substrate. The device is illuminated from the top with polarised infrared light perpendicular to the grating. (b) Energy band diagram of a metal contact to a p-type semiconductor and the 3-steps of the internal photoemission process: p: photoexcitation, t: transport, e: emission. E_C and E_V are the conduction and valence band edges, respectively, E_F is the Fermi level, E_g is the energy bandgap of Si, and Φ_B is the Schottky barrier height. (c) Hot hole photoemission across the Schottky junction between a metal and a p-type semiconductor.

Download Full Size | PDF

The IPE mechanism is sketched in Figs. 1(b) and 1(c). IPE is a 3-step process consisting of the photoexcitation of hot (energetic) carriers in the metal, transport with scattering of hot carriers towards the metal-Si interface, and the emission of hot carriers over the Schottky barrier into the semiconductor where they are collected as the photocurrent. IPE requires that the energy of the incident photons, hν (h is Plank’s constant and υ is the optical frequency), be greater than the Schottky barrier energy, Φ_B. This mechanism is useful for detection at energies below the semiconductor bandgap, E_g.

3. Theoretical

The grating coupler is highly suitable because arbitrary wavenumbers for diffracted light can be generated by simply changing the grating pitch. The grating coupling condition for SPPs on a metal surface is given by:

The substrate on which the devices are fabricated is a double-side polished wafer, consisting of a substrate (p⁺-Si) l₂ = 250 ± 25 μm thick, of carrier density N_A = 5 × 10¹⁸ to 3.5 × 10¹⁸ cm⁻³, bearing an epitaxial layer (p-Si) l₁ = 3 ± 0.5 μm thick, of carrier density N_A = 1.7 × 10¹⁵ to 1.0 × 10¹⁵ cm⁻³. We used this doping information to determine the refractive indices of these layers at λ₀ = 1550 nm as 3.477 and 3.4735 for the p-Si and p⁺-Si layers, respectively [26]. Au is selected as the metal for the patch and the grating, and its refractive index was taken as 0.55 + i11.5 [27].

The incident wave was set to a Gaussian beam at λ₀ = 1550 nm and its polarization was transverse magnetic (TM) (i.e., electric field parallel to the x-direction). The finite difference time domain (FDTD) method [28], with a discretisation mesh size of 1 nm × 0.5 nm × 0.0001 nm (x-y-z) around the grating, was used to model the optical performance of a cross-section (invariant along y) of several designs to determine good candidates for the device. Perfectly matched layers (PML) backed by perfect electric conductor (PEC) boundary conditions were used to terminate the computational domain in the z-direction along the top and bottom. The reflectance (R₀), transmittance (T₀), absorptance (A₀) and coupling coefficient (C₀) to SPPs are computed. The absorptance is calculated using:

Figures 2(a)-2(c) show the computed R₀, T₀, A₀ and C₀ for photodetectors having Λ = 390, 400 and Λ = 410 nm respectively. The subscript “d” (dashed curves) represents the initial design whereas the subscript “f” (solid curves) identifies the computed response of fabricated devices using their real dimensions. In general, the absorptance and coupling coefficient are largest at the wavelengths where Eq. (1) is satisfied, and the peak value red-shifts with Λ as expected. The absorptance of the photodetector is mostly due to coupling of light to SPPs which are then absorbed as they propagate: in the case of, e.g., the design having Λ = 400 nm and t = 20 nm (Fig. 2(b)), on resonance (λ₀ = 1560 nm) the peak value of A_0,d is 0.42 and the peak value of C_0,d is 0.38, the difference being due to direct optical absorption of the incident Gaussian beam in the metal grating. It is worth pointing out that the metal loss is usually considered harmful in most optoelectronic devices but here this parameter plays an important role because its consequence is the generation of useful hot carriers in the metal.

 

Fig. 2 Power ratios (absorptance A₀; transmittance T₀; reflectance R₀; and coupling coefficient C₀) of grating photodetectors operating near 1550 nm, with a duty cycle of 60% and t = 20 nm (solid lines), and a duty cycle of 62% and t = 28.9 nm (dashed lines) for grating designs having W = 200 nm, H = 80 nm and (a) Λ = 390 nm, (b) Λ = 400 nm, and (c) Λ = 410 nm. The resonance red-shifts with increasing pitch. Electric field distribution of the gratings corresponding to Λ = 400 nm, t = 20 nm, computed on resonance (λ₀ = 1560 nm); (d) Re{E_x}, (e) Re{E_z} and (f) |E| (also see Visualization 1 and Visualization 2). The incident electric field strength was 1 V/m.

Download Full Size | PDF

Figures 2(d)-2(f) show the distribution of electric fields over the cross-section of the device for the case Λ = 400 nm, t = 20 nm, computed on resonance (λ₀ ~1560 nm). Figure 2(d) shows Re{E_x}, associated primarily with the incident Gaussian beam. Portions of the incident beam are absorbed, coupled to SPPs, reflected and transmitted through the grating. Figure 2(e) shows Re{E_z} associated with SPPs propagating laterally (in the ± x directions) along the Au/p-Si interface, consistent with our sketch of Fig. 1(a). Figure 2(f) shows |E|, summarising the contribution of all field components. A movie of |E|, showing light propagation and scattering in the device is given in Visualization 1 and Visualization 2. It is observed from these field distributions that strong localised fields along the Au/p-Si interface are produced, which in turn will produce a considerable density of hot carriers in the metal along the Schottky contact as the SPPs are absorbed, ultimately enhancing IPE.

The detection wavelength can be tuned by varying the grating pitch Λ (the resonance red-shifts with increasing pitch). The photodetector can be optimized to operate at any infrared wavelength of sufficient photon energy to overcome Φ_B. (i.e.; from ~1100 to 4000 nm in this case). The full-width-at-half-maximum (FWHM) is another important parameter in the design of a resonant photodetector which is defined as the difference between the two wavelengths corresponding to the half-values of the peak of the absorptance curve, expressed in the frequency domain as:

The proposed photodetector is then analyzed by investigating its detection performance over a range of wavelengths. The photocurrent generation mechanism is IPE [29], wherein conduction carriers in the metal absorb energy quanta hν from the propagating SPPs. Gaining enough energy, hot carriers can cross over the Schottky barrier into the Silicon where they are collected as photocurrent under reverse bias as long as hν> Φ_B. The internal quantum efficiency for a thick-film Schottky barrier, η_i, which is dependent on the emission probability of hot carriers at the gold-silicon interface, is given by [30]:

4. Fabrication and experimental results

Au patches of diameter in the range of 4 to 30 μm were fabricated on p-Si on p⁺-Si wafers with the doping levels given above. The patches are circular to avoid field discontinuities and premature breakdown at corners, and each includes a contact arm with a probing pad. The patches were defined on the wafer’s topside (i.e.; on the p-Si epitaxial layer) using a bi-layer liftoff lithography technique with a UV contact mask aligner to expose the resist. Immediately prior to metal patch deposition, the wafers were placed in a buffered hydrofluoric (HF) acid solution to remove any native SiO₂ and allow intimate contact between the Au and the semiconductor. The grating duty cycles considered were 20% and 60% along with grating periods near 400 nm targeting telecom wavelengths (~1550 nm). The layout of the Au gratings was drawn using e-LiNE software [32] and e-beam lithography was used to define the grating ridges in a bi-layer PMMA lift-off process. Additional details on the fabrication of the patches and gratings can be found in [33]. Sintered Al was used for the Ohmic contact on the backside. No cover layer was used to allow for direct electrical probing and photocurrent measurements.

Figures 3(a) and 3(b) show a scanning electron microscope (SEM) image and an atomic force microscope (AFM) image of a fabricated grating photodetector. The metal thickness and roughness were deduced from AFM scans of the metal patch and grating ridges (e.g., Fig. 3(b)), revealing a thickness of t = 28.9 nm with a root mean square (RMS) roughness of 3 nm for the patch and H = 79 nm with an RMS roughness of 1 nm for the ridges. The width and spacing of the ridges are W = 252 nm and 154 nm, yielding a pitch of Λ = 406 nm (400 nm designed) and a duty cycle of 62% (60% designed). All dimensions are well in keeping with the target ones, except for the thickness of the patch (28.9 nm vs. the target of 20 nm). The low surface roughness on all metal surfaces (ridges and patch), attest to the good quality of the thermally evaporated Au.

 

Fig. 3 (a), (b) SEM and AFM images, respectively, of an uncoated Au/p-Si grating photodetector. The pitch of the device is measured as 406 nm (400 nm designed) and the thickness of the Au patch is measured as 28.9 nm (20 nm designed). The duty cycle of this device is 60%. (c) Optical microscope image of photodetectors of varying grating period, e-beam energy dose factors and duty cycles. Each color spot corresponds to one detector. The area shown is about 8 mm² and contains ~170 detectors.

Download Full Size | PDF

Figure 3(c) shows a microscope image of ~170 detectors distributed over a ~8 mm² area. The image was taken at a shallow angle in order to produce visible higher-order diffraction from each device. Each color spot corresponds to one detector, and the diffracted color changes with the period of the design. In each row, each yellow or orange spot corresponds to a detector with Λ = 390 nm but of e-beam energy dose factors of 1, 1.1, 1.2 and 1.3 and different duty cycles of 20% and 60%. Likewise, light green and dark green spots correspond to devices with Λ = 400 nm or Λ = 410 nm, respectively.

A photodetector under test is biased by probing its probe pad and grounding its Al contact (Fig. 1(c)). A cleaved polarisation-maintaining single-mode fibre (PM-SMF) is used to excite the photodetector under test from the top. The fibre is manipulated using a six-axis micro-positioner. A Photonetics TUNIC-Plus tunable external cavity laser is used to generate the incident light at λ₀ ~1550 nm. The laser is thermally stabilized and linearly polarized, with the polarisation aligned along the slow axis of the PM-SMF. The setup is placed on a vibration isolation optical table and is shielded from background light. Instrument control and data acquisition are performed via computer using LabVIEW (SP1 2013) [34]. The optical power incident on a photodetector is determined by removing the losses of all the elements in the input path of the setup.

The experiments are conducted by placing a die under test under microscope and using tungsten probes attached to micropositioners to make electrical contact to a device via its probe pad. A reverse-bias voltage applied between the probe pad and the Al Ohmic contact on bottom surface (ground). Another micropositioner is then used to align the PM-SMF (polarisation-aligned to launch TM light) to the photodetector. The cleaved fibre output field behaves as a Gaussian beam, with the beam waist coincident with the fibre output facet. The fibre is aligned to the photodetector by placing the device under reverse-bias and maximising the measured photocurrent. When the fibre is rotated to produce TE-polarised incidence, almost no photocurrent is generated. Once the best possible alignment is achieved, a computer automated routine controlling ancillary instrumentation is run to rapidly execute test routines and gather measurement data. Ancillary instruments consisted of a laser tunable over ~100 nm and a sourcemeter (Keithley 2400) used to apply a bias voltage and read the total current (dark + photo) in another channel simultaneously. All measurements were obtained at room temperature.

The red curve in Fig. 4(a) shows an average of five measured I–V curves for a Schottky photodetector without illumination, comprising a Au patch of diameter of 28.9 μm bearing a grating of pitch Λ = 400 nm designed to work at ~1550 nm. It shows the rectifying property of the Schottky structure. The reverse bias characteristics were rather soft for all devices, due to the low Schottky barrier and possibly to deep traps formed by the diffusion of Au into p-Si during fabrication (Fig. 4(c)). The dark current density, although high in low-barrier structures in general, produced a manageable dark current given the small area of the contacts. The contact area of our test device (including the pad) is A_c = 767 μm². Isolation of the pad from the Si can be achieved by adding an intervening dielectric, which would result in a reduction of the dark current and an increase in the photodetector’s signal to noise ratio.

 

Fig. 4 (a) The red curve shows an average of five dark I-V characteristics for a grating photodetector of diameter 28 μm and Λ = 400 nm. The purple curve is obtained after removing the effects of the shunt and series resistors shown in the non-ideal model in inset. (b) Measured photocurrent response of three grating photodetectors; V_B = −100 mV, Λ = 390, 400 and 410 nm, duty cycle of 62% and patch diameter of 25 μm. A 4-period moving average is plotted on each response as the bold curve. The rapid wavelength variations correspond to Fabry-Perot resonances - the upper right inset is an enlarged response showing such resonances and the lower right inset shows a response calculated using the TMM method, (c) Deep traps formed by the diffusion of Au into p-Si .

Download Full Size | PDF

The dark current of a Schottky diode is given by:

Figure 4(b) plots the p-polarized photocurrent response measured at normal incidence with P_inc = 1.4 mW and V_B = −100 mV for three grating detectors of different period. Two wavelength dependencies are observed on each response. The first one is a rapid variation which comes from Fabry-Perot resonances, as light transmitted through the grating and metal patch propagates through the semi-transparent substrate (p⁺-Si) and is reflected from the backside Ohmic contact. The right inset shows two plots, the upper is an enlarged plot of the measured photocurrent response near 1550 nm and the lower is the computed response of a three-layer structure comprising a 30 nm thick Au film, the p-Si epi layer, and the p⁺-Si substrate bounded by air on top and Al on the bottom, obtained using the transfer matrix method (TMM) [38]. The measured rapid variations and the computed response are in good agreement, as the free spectral range (FSR) of the Fabry-Perot responses (given by the difference between the frequencies of two adjacent peaks) is in good agreement; FSR ~1.87 THz. The Fabry-Perot resonances would disappear if a more heavily doped Si substrate was used. The second wavelength dependency observed in the main panel of Fig. 4(b) is the slow one highlighted as the moving average plotted in bold on each grating response, and is due to the incident light coupling into SPPs. We measured the photoresponse of each device three times; they are all repeatable with a normalized mean square error (NMSE) on the photocurrent relative to the averaged photoresponse of ~0.02 - 0.025 μA. These bold responses follow the absorptance responses labeled A_0,f in Figs. 2(a)-2(c), computed using the actual dimensions (t = 28.9 nm, duty cycle of 62%). We calculated the photocurrent enhancement factor of the photodetector, relative to one without the grating but a 100 nm thick Au film, as 12.86 μA / 2.48 μA = 5.18.

The peak of the bold responses clearly red-shifts as the grating pitch increases from 390 to 410 nm, following Eq. (1). Photocurrents were then measured as a function of incident optical power (emerging from the PM-SMF) at the resonance peak of each response, λ₀₁ = 1537, λ₀₂ = 1548 and λ₀₃ = 1570 nm with V_B = −100 mV applied. It should be noted that only the incident power overlapping with the device area couples to SPPs and contributes to the photocurrent and responsivity. As shown in the inset of Fig. 5(a), a portion of the incident power does not overlap with the device. There are two ways to refine the measurements to take this into account. One is to use a tapered fiber to focus the incident beam onto the photodetector area. The other is to correct the incident power assuming that a Gaussian beam emerges from the fibre, such that the beam diameter at the surface of the device can be estimated, and the fraction of overlapping incident power determined:

 

Fig. 5 (a) Photocurrent generated by grating detectors vs. incident power P_inc,d measured for three different pitches: Λ = 390 nm (λ₀₁ = 1537 nm), Λ = 400 nm (λ₀₂ = 1548 nm) and Λ = 410 nm (λ₀₃ = 1570 nm). A linear fit is applied to the data and the slope corresponds to the responsivity of the device. (b) Responsivity vs. reverse bias measured for the photodetector with Λ = 400 nm at λ₀ = 1548 nm; the dashed curve shows a trend curve for the responsivity obtained by taking into account the Schottky barrier lowering effect.

Download Full Size | PDF

Figure 5(b) plots the responsivity as a function of reverse bias, extracted from the measurements of Fig. 5(a) and from additional measurements at other bias points, revealing that the responsivity increases significantly with bias. Applying a reverse bias to a Schottky photodetector reduces Ф_B via the Schottky effect, so the responsivity is expected to increase with reverse bias [40]. A reverse bias reduces the Schottky barrier Φ_B by an amount ΔΦ_B via the Schottky effect [27]:

A trend curve for the responsivity was computed using Eqs. (6), (7) and (11) and plotted in Fig. 5 (b) (we added an offset to the trend curve such that it would coincide with the measurement at low-bias in order to illustrate the effect of ΔФ_B on the responsivity). For the computations, the Schottky barrier height and the built-in voltage were taken as Φ_B = 0.32 eV and V_bi = 0.26 V, as extracted from the I–V curve in Fig. 4(a), and the Fowler characteristics measured on a similar detector [11]. The change in Schottky barrier height over the range of reverse bias voltage considered does not explain the change in the measured responsivity. The large increase observed in Fig. 5(b) is more likely caused by deep traps in the Si bandgap, formed by Au migration into the Si during fabrication, which are uncovered with increasing reverse bias.

In perfect semiconductors, there exists a bandgap between the valence and conduction bands of energy E_g = E_C − E_V (Fig. 1(b)). During fabrication Au ions can migrate into the Si (driven by process temperatures) where they introduce energy states somewhere in the band gap, usually close to mid-gap (~E_g/2), and thus are termed deep traps (Fig. 4(c)). They generally act as recombination-generation centres and are not classified as donor or acceptor levels. The reason is that the generation-recombination centres have to capture/emit both electrons and holes. The capture/emission cross-sections of a deep trap will depend upon several factors such as whether the process is capture or emission, the ionisation energy of the trap, whether the trap is neutral, or positively or negatively charged, and whether the trap is in the bulk or at the interface. A deep trap (e.g. Au in Si) is most effective as a recombination centre if the hole capture time (from the valence band) is nearly the same as the electron capture time (from the conduction band). The trap levels for Au in Si are 0.54 eV for acceptors and 0.29 eV for donors [27], both of which are below the photon energy of the incident light, so traps can assist in producing photocurrent by acting as a mid-gap “bridge” for EHPs between the valence and conduction bands [41]. In Fig. 5(b), we suspect that such traps become increasingly uncovered as the reverse bias increases, thus enabling this photodetection channel resulting in increased responsivity.

5. Conclusion

In summary, we have presented a broadband silicon plasmonic Schottky photodetector with a high responsivity operating in the sub-bandgap regime exploiting internal photoemission. A metal grating is used to couple perpendicularly-incident light to SPPs propagating along a thin metal patch forming a Schottky contact to p-Si. The responsivity is enhanced by the absorption of SPPs directly along the Schottky contact leading to the creation of hot holes near the contact. These holes cross over the Schottky barrier (~0.32 eV) and are collected as a photocurrent. Our measurements reveal a high responsivity of ~13 mA/W at telecom wavelengths (~1550 nm) and under low reverse bias. The responsivity increases with reverse bias. The optical bandwidth (FWHM) is ~80 nm. The dark current performance of the photodetector could be significantly improved by using a smaller area and fabricating the contact arm and probing pad on a higher level (e.g., above a dielectric). These detectors have many important advantages such as speed, simplicity, compatibility with silicon, and low-cost fabrication.