n-type Ge/Si antennas for THz sensing

: Ge-on-Si plasmonics holds the promise for compact and low-cost solutions in the manipulation of THz radiation. We discuss here the plasmonic properties of doped Ge bow-tie antennas made with a low-point cost CMOS mainstream technology. These antennas display resonances between 500 and 700 GHz, probed by THz time domain spectroscopy. We show surface functionalization of the antennas with a thin layer of α -lipoic acid that red-shifts the antenna resonances by about 20 GHz. Moreover, we show that antennas protected with a silicon nitride cap layer exhibit a comparable red-shift when covered with the biolayer. This suggests that the electromagnetic fields at the hotspot extend well beyond the cap


Introduction
Plasmonics, with its ability to manipulate light into sub-wavelength volumes, could help boosting performances and add functionalities to photonic devices operating in the THz range of the electromagnetic spectrum [1][2][3].Indeed, one can exploit plasmonics in manipulating THz radiation, focusing it on subwavelength volumes and/or waveguiding [4].Plasmon-polaritons, i.e. quasiparticles arising from light-matter interactions that yield strong field confinement and enhancement, can be sustained either by a metallic-dielectric interface (surface plasmon polaritons -SPPs) or localized by subwavelength metallic structures (localized surface plasmons -LSPs).In particular, LSP resonances can be engineered by structuring metals on a subwavelength scale and plasmonic meta-surfaces can be specifically designed to tailor the dispersion and the confinement of waves traveling along their surface.Alternatively, one can obtain strongly localized fields in the infrared and in the THz spectral range relying on doped semiconductors and conducting polymers [5,6].Particularly, compared to metals, semiconductors have the advantage of a tunable plasma frequency by controlling their carrier concentration (N).In semiconductor-based plasmonic devices, negative permittivity is achieved over a wide frequency range below their screened plasma frequency ω * p , in the mid-and far-IR spectral range [7][8][9][10][11][12][13]. If, on one hand, III-V compound semiconductors (InP, GaAs) feature inherently higher carrier mobilities and thus can lead to lower plasmonic losses, on the other hand the use of elemental semiconductors opens up to the realization of mass-produced THz plasmonic devices.By leveraging on the low-point cost of Si-based CMOS mainstream technology, it is possible to enable the widespread use of plasmonics in consumer products for different application realms.A further advantage of CMOS technology, specifically SiGe-based technologies, is the capability to monolithically realize a single platform comprising mixed-signal electronics, biosensors and microfluidics.These advantages are highly relevant for the implementation of Lab-on-a-chip devices which can profoundly impact the public health sector.Indeed, several applications in the medical field can be envisioned, such as point of care testing, owing to the non-ionizing nature of THz radiation [14].Furthermore, the THz spectral range comprises the characteristic energies of molecular vibration, rotation, and libration, thus enabling e.g.environmental monitoring of volatile organic compounds [14,15].Several Si-based or Si-compatible approaches have been reported in the literature for devices operating in the THz region, such as plasmonic antennas [16][17][18][19], cavities [20], particles [21], and gratings [22].Aside from Si, in the past few years, Ge has entered routinely the CMOS foundry process flow, given the possibility to obtain high quality heteroepitaxial growth of Ge/Si structures that could be used for plasmonic applications.In fact, Ge has the advantage over Si of inherently lower losses [23], a lower electron effective mass [24], and can be doped with donor densities up to 10 20 cm −3 [25].Moreover, the use of doped Ge/Si heterostructures for the realization of resonators on Si substrates allows for a selective functionalization of Ge surfaces, in view of their application towards molecule sensing [26].
In this work, we demonstrate that THz bow-tie antennas fabricated using phosphorous-doped, n-type Ge/Si (001) can detect a deeply subwavelength layer of α-(Alpha-) Lipoic Acid (ALA) selectively absorbed on their surface.In order to assess the relevance of material quality on the plasmonic properties, we have investigated antennas made out of epitaxial doped Ge layers realized with different processes and featuring different electron mobility.By exploiting a CMOS-compatible manufactory pilot-line, we have realized several antenna array arrangements, in terms of geometry and density, either Si 3 N 4 encapsulated or left bare.For the different antenna arrangements we observed well-defined transmittance resonance in the 0.3-0.9THz spectral region, in good agreement with finite element Maxwell simulations.We also show that a deposited ALA layer causes a reproducible resonance red shift of ∼20 GHz on antennas, demonstrating the sensing capability of our devices.

Growth, fabrication and characterization
We first discuss the realization of the n-type Ge/Si heterostructures and their material properties.The deposition of n-doped Ge layers on slightly p-doped Si substrates Si(001) wafers (200 mm) has been carried out using a ASM Epsilon 2000 lamp-heated reduced-pressure chemical vapor deposition (RPCVD) single wafer system.Si(001) wafers were cleaned by means of a wet-chemical process (Radio Corporation of America, RCA, solution with a subsequent HF step).After RCA cleaning, the hydrogen terminated Si(001) wafers were transferred into the RPCVD system and prebaked at 1000 • C in H 2 for 45 s to remove remaining native oxide and moisture.After the cleaning, an n-type 100-nm buffer layer was deposited.A first set of samples (G1) was realized by depositing a 1.5 µm thick n-doped Ge layer by RPCVD on top of the Si(001) wafers at a growth temperature of 350 • C and 80 Torr chamber pressure, using N 2 as carrier gas for the reactant GeH 4 and the doping PH 3 gases.As will be later discussed, we have selected 1.5 µm as our target thickness as a trade-off between a strong plasmonic response and a technologically-relevant material process.The Ge layer is plastically relaxed, has a surface roughness of approximately 1-2 nm, and a threading dislocation density in the 10 7 cm −2 range [27,28].
Since a constant doping profile as well as a high donor activation ratio can be disrupted during growth by the out-diffusion of P, dopant dimer formation, or clustering of dopants [23], we have characterized the P doping profile using a Time-of-flight Secondary Ion Mass Spectroscopy (ToF-SIMS) in a ToF-SIMS V from IONTOF.The depth profiles were acquired by sputtering with a Cs + ion beam with an energy of 500 eV over a 300 × 300 µm 2 area and analyzing with a Bi 1 + beam with an energy of 25 keV over a 100 × 100 µm 2 area.The results of ToF-SIMS (Fig. 1(a)) show a constant doping profile of ∼1.5 × 10 19 cm −3 along the entire thickness of the Ge layer.The measured donor concentration is in agreement with the nominal one, which we have selected as a trade-off between the highest possible density of activated donors (and, consequently, larger ω * p ) and the quality of the material [27,28].We remark here that we found an excellent thickness and doping homogeneity over different areas of the 200-mm wafer (Fig. 1(a), inset), as only a negligible decrease in dopants concentration is found at the edge of the wafer.A second set of samples (G2) was grown by adding a 30 nm-thick silicon nitride (Si 3 N 4 ) encapsulation layer on n-Ge epilayers deposited using the same process of G1.The capping layer was deposited by a low-pressure CVD (LPCVD) batch reactor system using Bis(t-butylamino)silane (BTBAS) and was followed by annealing at 700 • C for 10 s in a rapid thermal annealing furnace system.Electrical transport characterization was carried out on representative sections of the n-Ge wafers for both G1 and G2.A shadow mask was used to deposit an Al layer by sputtering, which is known to provide reliable electrical contacts to n-Ge [29].Excellent agreement was found between four-point probe, van der Pauw and Hall bar contact scheme configurations, showing a low contact resistance.Additionally, temperature-dependent DC Hall measurements were performed.The contribution of the Si substrate carriers at low temperatures was removed thanks to the buried n-Si/p-Si isolating junction existing between the buffer layer and the substrate (see Supplement 1, Fig. S1(a) and S1(b)).In Fig. 1(b) the carrier mobility from Hall measurements (µ H ) at different temperatures is shown.At room temperature the Hall mobility was µ H = 308 cm 2 /Vs for G1 and µ H = 370 cm 2 /Vs for G2 sample, respectively.The mobility is in agreement with model [30] and experimental [31] values in literature at room temperatures for bulk n-type Ge featuring similar concentrations.This points to the high quality of the material realized, in which the electron scattering process is dominated by that on phosphorous ion.
Fourier Transform Infrared (FT-IR) spectroscopy was performed in order to estimate the doping level and losses, and compare it to Hall data.Near-normal incidence reflectivity (R) measurements were performed at ambient conditions in a Bruker FTIR Vertex 80v with an A510/Q-T accessory using a gold mirror as reference.It is possible to identify clearly in the spectra the drop in reflectivity, corresponding to the screened plasma edge as well as oscillations arising from Fabry-Perot fringes within the Ge layer above the plasma edge.The experimental data was fitted by the transfer-matrix method (TMM) using the Drude model (Fig. 1(c)).Even though a generalized Drude approach would better describe the data [23], a simple Drude model, with a single scattering rate coefficient (γ) can reasonably describe room temperature data, yielding a doping concentration of N IR = 1.4×10 19 cm −3 and N IR = 1.2 ×10 19 cm −3 and a mobility of µ IR = 310 cm 2 /Vs and µ IR = 400 cm 2 /Vs for G1 and G2, respectively, in reasonable agreement with Hall measurements.
We have then used these material data to feed the modeling of single antenna structures carried out using Finite Element Method (FEM) on an unstructured mesh [32,33] to obtain 3D vectorial solutions of the Maxwell equations.Starting from fixed values of doping N and mobility µ (N = 1.5×10 19 cm −3 and µ = 310 cm 2 /Vs) two different geometries, slab (dipole) vs bow-tie, were investigated, finding that for a target resonance frequency of 0.5 THz the extinction coefficient of the bow-tie geometry has a superior performance (>20%) against a simple slab geometry (see Supplement 1, Fig. S2).The gap width of 1 µm has been chosen for maximum field enhancement at the one hand, and reasonable size for sample loading and future microfluidic integration on the other hand (see [32]).Moreover, we have evaluated the behavior of the penetration depth δ(f ) = √︁ 1/πf µσ(f ), where σ(f ) is the frequency-dependent conductivity.For the above-mentioned values of doping and mobility, the penetration depth was calculated to be δ (0.5 THz) ≃ 3.4 µm.As it will be later shown, we can nevertheless obtain a reasonably strong and narrow resonance in the plasmonic antennas made of both G1 and G2 materials.
Antennas were fabricated by employing 210 nm low-deposition rate silane-based silicon oxide (SiO 2 ) as a hard mask.After lithographic patterning (i-liner process), reactive ion etching was used to obtain the final antenna shape (see Fig. 1(d)).The bow-tie resonator has 74 µm as the length of each (trapezoid) arm, 30 µm (10 µm) for the base (apex) of the trapezoid and 1 µm as the antenna gap.In order to better tailor the antenna response, we realized antenna arrays with different primitive cells, namely square (Sq), hexagonal (Hx) and a random (Rd) array (see Supplement 1, Fig. S3).For each array, samples have been fabricated with various antenna distances, as summarized in Table 1.For the fabrication of antennas on the G2 material an additional process step after wet chemistry was used.This consisted of 30 nm Si 3 N 4 additionally deposited by LPCVD batch reactor system using BTBAS, in order to cover the sidewalls and top of the created resonator structures for a final thickness of 30 nm and 60 nm Si 3 N 4 , respectively.

Antennas and sensing
Antennas extinction have been obtained by means of Terahertz Time Domain Spectroscopy (THz-TDS), with a spectral bandwidth of 5 THz, 90 dB dynamic range and a frequency resolution of less than 1.2 GHz (Terasmart from Menlo Systems, Germany and TERA K15 all-fiber coupled THz spectrometer from Menlo Systems, Germany [34]).In the TERA K15 systems the THz emission were obtained by photoconductive antennas excited by a femtosecond fiber-coupled laser (Menlo Systems T-Light) with center wavelength of 1560 nm, 100 MHz repetition frequency, and a pulse duration of nearly 90 fs [35].The THz radiation emitted by the photoconductive antenna was collected and collimated by means of a TPX (polymethylpentene) lenses.The samples were placed at the THz focus where the illuminated area was approximately 3 mm in diameter, so that the THz pulse could probe a large area, thus minimizing heterogeneity effects.Fabry-Perot interference effects were removed from spectral data by a numerical procedure [34].Measurements were performed in a transmission setup with a linearly polarized source.Spectra were collected by aligning the long axis of the antenna arms parallel to the electric field polarization (I ∥ ), thus exciting the antenna resonance, and, as reference, with the long antenna axis perpendicular to the electric field polarization (I ⊥ ).Measurements were performed in air.In order to eliminate the contribution of water absorption lines due to the environmental humidity, measurements of I ∥ and I ⊥ were carried out quickly one after the other (both acquisitions in less than 5 minutes).This ensured for a small variation of the humidity and a very good compensation of water absorption lines in the dichroic ratio obtained as DR=-ln(I ∥ /I ⊥ ).The dichroic ratio obtained is plotted in Fig. 2, showing a clear antenna resonance.Moreover, we also have measured a bare Si substrate as reference, in order to calculate the transmittance for the two polarizations of the electric field (Supplement 1, Fig. S4).In Fig. 2(a) we compare the experimental data with the theoretical calculations obtained for an increasing density of antennas (A, B, C), as well as for different antenna arrangements (Hx, Sq).The resonant frequency of the sample with the highest antenna density (C) displays a blue shift, when compared to the designed single antenna center frequency, as a result of antenna-to-antenna cross-talk (see Fig. 2(b), where the dichroic ratio has been normalized to the number of antennas in the THz focal spot).Indeed, the antenna resonance measured for larger antenna spacing (A and B) is comparable to that found when the antennas are placed on a random pattern.Mutual coupling between individual neighboring antennas and their near-fields thus results in shifted spectra [36,37].The theoretical simulations were adjusted to account for inter-antenna interactions within the array by treating each antenna as a scatterer in a total field scattered field (TFSF) approach.This resulted in a good agreement with the experimental data, demonstrating the capacity of the model to accurately predict the THz behavior.
We have, moreover, collected the experimental data from 10 replicas of G1 (gray circles) and G2 (red circles) sample from different parts of the 200 mm wafers showing high reproducibility throughout the wafer (Fig. 2(c)).A small difference was found, by averaging all data collected on each wafer among G1 and G2 antennas.To understand whether a small effect of mobility on the resonance can be seen we compared qualitatively with simulations (inset in Fig. 2(c)).Indeed simulations, where only the mobility difference among G1 and G2 was taken into account, show a comparable effect on the resonance, while the presence of the Si 3 N 4 capping layer was disregarded.
To demonstrate the sensing capability of our devices, we then explored selective functionalization of the Ge layers with α-lipoic acid (ALA).The intramolecular disulfide group of ALA mediates selective covalent bonding to the Ge surface while the remaining carboxyl moiety is suitable for immobilization of biomolecules.We have previously shown self-assembly of ALA monolayers on freshly cleaned Ge surfaces, which are stabilized by germanium-sulfur bonds [26].In a process previously reported [26], the native oxides and/or residual contaminants are removed from the Ge surface by a HF (12%)/H 2 O 2 (7%) cyclic cleaning procedure.To avoid possible etching effects and surface roughness, citric acid was used instead as an alternative for the removal of the native oxide [38].The H-passivated surfaces are then incubated in an ethanol/DI water/ALA solution (5 mM, pH 6) overnight at 4 • C.Under these conditions ALA forms a multilayer with a well-defined thickness of at least 20 nm [26].
The effect of ALA adlayer formation on the antennas is detected as a red-shift of the antennas resonances (Fig. 3(a) and 3(b)), which is comparable for both G1 and G2.Data were fitted with a Gaussian line shape to better estimate the central frequency.The resonances for both G1 and G2 are centered at about 0.7 THz before the incubation of ALA (in red).After the incubation period (in black), both show a red shift of about 20 GHz and a reduction of the resonance intensity of about 10%.In Fig. 3(c), we show simulations of the resonance behavior to the presence of a material, with varying refractive index (n), completely filling the antenna-gap and a 100 nm thick layer (lower value allowed by simulations) on its surface and rear-end of the antennas.Interestingly, the red shift of the resonance due to the ALA layer, if compared to the identical clean sample as reference, confirms the sensitivity of the plasmonic resonance to changes in their dielectric environment.The observed experimental red-shift qualitatively agrees with the simulations of a 100 nm layer (n > 1) covering the antennas and filling the 1 µm antenna-gap, suggesting a value of n ∼1.5 for the ALA layer at this frequency.These results are in complete agreement with the expected sensing behavior of the resonators.Within the antenna gap, the fields are concentrated and enhanced by orders of magnitude, making this volume the most responsive part of the sensor to the material therein.The identical center frequency between G1 and G2 antennas, i.e. one with the Si 3 N 4 encapsulation, may stem from a combination of factors.First of all, despite a full coverage, the Si 3 N 4 is thinner on the sidewalls than on the antenna surface.The enhanced fields arising due to the LSP resonance are mainly located along the antenna feed gap, and as the gap is filled by the Si 3 N 4 only by a small percentage of the entire volume it can result in the aforementioned behavior.Interestingly, the red-shift due to the presence of ALA is comparable in the two samples, despite the presence of Si 3 N 4 .Again, the behavior can be understood as the main field interaction takes place within the gap volume from which the thin encapsulation layer has a negligible effect on the antenna response.While we expect the field enhancement at the hot spot to exceed the Si 3 N 4 thickness, ALA binding on the surface of the encapsulation layer was not given for granted.
To thoroughly confirm the presence of the biolayer and its distribution on the antennas, a set of measurements consisting of nano-Auger Electron Spectroscopy (AES), ToF-SIMS and X-ray Photoelectron Spectroscopy (XPS) were performed in incubated and reference samples.AES mapping were performed using a Physical Electronics PHI 670 type Scanning Auger Microprobe with an electron energy of 10 keV by acquiring the Ge, Si and S signals with 256 pixel x 256 pixel resolution.AES maps of the characteristic signals of Ge, Si and S for an incubated sample are shown in Fig. 4(a), 4(b) and 4(c), respectively.The presence of sulfur confirms the binding of ALA on the antenna's surface.Indeed, thanks to the high resolution and surface sensitivity of AES measurements it is possible to show the presence of an S-containing layer in the incubated samples (Fig. 4(c)).The observed behavior can be explained by the anticipated ALA selectivity to Ge against SiO 2 [26].Even though the Si substrate was not deliberately oxidized, a native oxide can be expected in the surface of Si (see the O signal in Supplement 1, Fig. S5).By comparing the maps obtained for Ge and Si, clearly showing the gap between antennas, with the map obtained for S distribution we can infer an accumulation of S in the gap, which rules out any wettability problem of the hotspot region.In order to visualize the 3D distribution of ALA, depth profiles were measured by ToF-SIMS using S − ions as the marker for lipoic acid.The S accumulation, and thus the ALA binding to the antenna, is more clear in the 3D reconstruction of ToF-SIMS measurement (Fig. 4(d) and 4(e)) where it is noticeable that sulfur (seen as black shade) tends to accumulate at the edges (walls) and gap of the antenna.It can be seen that the presence of sulfur is distributed over the antenna and substrate surface at the beginning of the Tof-SIMS experiments (Fig. 4(e), 0 µm).As the sputtering continues (Fig. 4(e), 0.6 and 1.1 µm) the S signal is mostly coming from the gap and walls.This suggest that surface tension between the ALA solution and the antenna structure, as the liquid dries out after incubation, allows for an effective coverage of the antenna walls, also in the feed gap.The apparent higher signal of S on one antenna side of the ToF-SIMS results shown in Fig. 4(d) can be explained as an artifact produced by the 45°inclination of the sputtering gun with respect to the detector, giving the impression of a preferential side.Furthermore, a low ion energy (250 eV) was used to slowly sputter the sample's surface and follow the component concentration along the depth profile.ToF-SIMS highest mass resolution settings were used to confirm and avoid crosstalk between the S and (2x) oxygen signal.
The XPS analysis of reference and ALA-treated samples for both generations were acquired in a Ulvac-PHI VersaProbe II instrument equipped with a monochromatic Al K-Alpha X-ray source with a photoelectron takeoff angle of 45°, a hemispherical analyzer and Argon ion gun.A dual-beam charge neutralizer (8 eV Ar , and 10 eV-electron beam) was used to compensate the charge-up effect.The measurements were performed using a pass energy of 29 eV in order to ensure a good energy resolution.As expected, the XPS spectra in Fig. 5(c) shows the intrinsic differences of (bare Ge) G1 and (Si 3 N 4 encapsulated) G2 surfaces in their respective Ge and N signals.The complete absence of the Ge 2p 3/2 and Ge 3d signals on G2 samples demonstrated the effective encapsulation by the Si 3 N 4 .Similarly, the presence of the S 2p peak in ALA-treated samples (red lines) for both generations and its absence in the reference sample (blue lines), corroborate the existence of a shallow sulfur-containing layer on their surfaces.While the complete coverage of Ge atoms would hinder the binding of ALA, it can be expected that during the pretreatment and/or incubation procedure, the Si 3 N 4 layer can at least partially oxidize in its surface [39].These would produce Si/SiO x sites for the binding with ALA, which albeit being less effective than bare Ge, is nevertheless present (Fig. 4(c)).The measurements here presented endorse the non-selective binding of the biolayer to the encapsulated Ge antennas.The encapsulation layer is shown to have a minimum effect in its sensing capabilities which not only allows to improve the plasmonic material properties but also increases its robustness against aqueous solutions in microfluidic applications.In SEM images with lateral view of G1 and G2 incubated samples, a ∼20 nm-thick layer can be distinguished for both generations (Fig. 5(a) and Fig. 5(b), respectively), in agreement with the results of the above-mentioned measurements and those of Kazmierck et al. [26].The visible streaking of the outmost layer in Fig. 5(b) can be seen as an artifact influenced by the non-conductive Si 3 N 4 layer.Taking in consideration the vacuum conditions required for the characterization techniques mentioned above, a range, as an upper limit, for the thickness of the incubated ALA layer can be set between ∼100 nm and ∼1 µm given that the devices were kept in atmospheric conditions during the THz-TDS experiments.Indeed an extended exposure to atmospheric and/or vacuum conditions could influence the stability of the ALA layer.Such thickness range of the biolayer would in any case correspond to a non-specific subwavelength detection by the Ge resonant antennas between two and three orders of magnitude.

Conclusions
CMOS-compatible THz plasmonic antennas were fabricated in a pilot fabrication line.Up to 1.5 µm of highly n-doped Ge layers with a high carrier activation and mobility of up to 370 cm 2 /Vs were grown epitaxially on Si (001) substrates.Ge antennas show a strong, polarization dependent resonance at 0.5-0.7 THz that were probed for several geometric shapes and periodic arrangements.We have shown that it is possible to bind a layer of α-Lipoic Acid selectively on Ge surfaces and sense it via a resonance shift.This is as well observed for antennas protected with a Si 3 N 4 cap layer.These findings pave the way towards the possibility of achieving selective molecule bindings for microfluidic applications on a fully CMOS compatible sensing platform.

Fig. 1 .
Fig. 1.(a) ToF-SIMS doping profile of a ∼1.6 µm thick n-Ge layer.Inset: Analysis positions over the wafer.(b) Temperature sweep of Hall carrier mobility for G1 and G2 and comparison with mobility values extracted from Drude fitting of the FT-IR data.(c) Reflectivity measured between 500 and 3500 cm −1 for G1 and G2 compared to fitting based on Drude model.(d) SEM image of fabricated bow-tie n-Ge antennas in square arrangement.

Fig. 2 .
Fig. 2. (a) Experimental (solid lines) and TFSF simulations (dashed lines) of G1 THz resonance for increasing antenna density (A, B, C) in a Hx and Sq arrangements.(b) Dichroic ratio for increasing antenna density in Sq arrangement compared to random antenna arrangement normalized by the number of antennas.(c) Experimental and (inset) FEM simulation comparison of G1 and G2 THz resonances.

Fig. 3 .
Fig. 3. Experimental results (unfilled circles) and Gaussian fit (solid lines) of (a) G1 and (b) G2 before (red) and after (black) ALA incubation.(c) Simulation results of the effect on the resonance caused by a film covering the whole gap and 100 nm surface of the antennas at different n values.

Fig. 4 .
Fig. 4. (a) Germanium, (b) silicon and (c) sulfur AES maps of G1 ALA-treated samples.(d) 3D ToF-SIMS reconstruction and (e) sliced view at different heights showing the S concentration around one Ge antenna and especially in the gap.