Zero-Bias Long-Wave Infrared Nanoantenna-Mediated Graphene Photodetector for Polarimetric and Spectroscopic Sensing

Graphene has attracted great interest for integrated photonic platforms in the long‐wave infrared (LWIR) for spectroscopic and polarimetric sensing due to the capability of on‐chip integration, fast response, and broadband operation. However, graphene suffers from low photoresponsivity and thus poor sensing performance due to weak absorption. Polarization detection using graphene is hindered by its small in‐plane anisotropy. Here, nanoantenna‐mediated graphene photodetectors (NMGPDs) are proposed to enhance responsivity by tailoring the nanoantenna near‐field distribution to generate a strong photoresponse. The devices demonstrate a high responsivity of 6.3 V W−1 under zero bias at room temperature with low noise‐equivalent power of 1.6 nW Hz−1/2. Furthermore, polarization detection is enabled by artificial near‐field anisotropy enabled by double L‐shaped nanoantennas. The proposed LWIR NMGPDs achieve subtle polarization‐angle detection down to 0.05° thanks to the unusual negative polarization ratio of −1. To demonstrate the advances of LWIR NMGPDs for molecule detection, acetone is chosen as an analyte for spectroscopic sensing. The devices show a low limit of detection of 115 ppm, and fast dynamic gas sensing response of 6 s for real‐time monitoring. These results reveal the potential of the device as a multi‐functional on‐chip miniaturized optoelectronic platform for polarimetric and spectroscopic sensing toward real‐time environmental monitoring and biomedical screening.


Introduction
[32] Meanwhile, optical gas/ liquid sensing is a critical application for LWIR detection.[35][36][37][38][39][40] However, these optical sensors need bulky spectrometers to collect spectral information, which hinders them from on-chip integration.[43] However, they still need another commercial photodetector to detect the transmitted light to collect the sensing information.Therefore, integrating graphene photodetectors and optical sensors on a single platform can achieve high efficiency, low cost, and high-level miniaturization. [44]Particularly, polarization, another dimension of light in addition to the wavelength, phase and intensity, can provide supplementary but indispensable information for many analytes and enhance sensing performance. [45,46]Currently, polarization imaging and sensing techniques of biostructures have already been applied in various biomedical applications, including early disease diagnosis and enhanced Graphene has attracted great interest for integrated photonic platforms in the long-wave infrared (LWIR) for spectroscopic and polarimetric sensing due to the capability of on-chip integration, fast response, and broadband operation.However, graphene suffers from low photoresponsivity and thus poor sensing performance due to weak absorption.Polarization detection using graphene is hindered by its small in-plane anisotropy.Here, nanoantenna-mediated graphene photodetectors (NMGPDs) are proposed to enhance responsivity by tailoring the nanoantenna near-field distribution to generate a strong photoresponse.The devices demonstrate a high responsivity of 6.3 V W −1 under zero bias at room temperature with low noise-equivalent power of 1.6 nW Hz −1/2 .Furthermore, polarization detection is enabled by artificial near-field anisotropy enabled by double L-shaped nanoantennas.The proposed LWIR NMGPDs achieve subtle polarization-angle detection down to 0.05° thanks to the unusual negative polarization ratio of −1.To demonstrate the advances of LWIR NMGPDs for molecule detection, acetone is chosen as an analyte for spectroscopic sensing.The devices show a low limit of detection of 115 ppm, and fast dynamic gas sensing response of 6 s for real-time monitoring.These results reveal the potential of the device as a multi-functional on-chip miniaturized optoelectronic platform for polarimetric and spectroscopic sensing toward real-time environmental monitoring and biomedical screening.[49][50] Therefore, an integrated platform to realize multi-functional detection, including spectroscopic and polarimetric sensing is highly desirable.However, unlike black phosphorus and many IV-V compounds, graphene does not show in-plane anisotropy, which hinders graphene from polarization-dependent photodetectors. [51]Moreover, graphene also suffers from low absorption in IR of ≈2.3%, resulting a low responsivity and poor sensing performance. [52][62][63][64][65][66][67] Furthermore, the nanoantennas can serve as small metal electrodes to collect the carriers and thus give rise to the photocurrent. [68,69][72][73][74] Furthermore, strong artificial anisotropy in the metal-graphene surfaces can be achieved by fine designing the nanoantenna structures and hence realizing filterless polarization-sensitive detection. [75][77] It is desired to extend the operation wavelengths to the LWIR range and realize multi-function detection including polarization and molecule sensing.
Here, we propose chip-level LWIR NMGPDs for polarimetric and spectroscopic sensing (Figure 1a).The polarizationdependent photoresponse is realized by artificial anisotropy formed by designed double L-shaped nanoantennas (DLNAs).The polarization ratio (PR), an important figure of merit in polarization detection, is usually defined as the ratio of maximum and minimum polarization-dependent photoresponse.A negative PR is achieved in our experiments, which is unusual in the 2D materials polarization detectors based on intrinsic anisotropy (Table S1 and Note S8, Supporting Information), and enables our devices to be self-contained balanced detectors.Thanks to this, our NMGPDs demonstrate a sensitive noiseequivalent polarization-angle perturbation down to 0.05°, which is comparable to the reported detectors in the visible range, and still tricky in the LWIR range. [76,78]Meanwhile, the metal plasmons not only strongly enhance the responsivity, but also increase the light-matter interactions for sensing.Our devices work under zero bias (V DS = V g = 0) and show a high responsivity at room temperature with low noise-equivalent power (NEP) of 1.6 nW Hz −1/2 .To further demonstrate the advances of LWIR NMGPDs for molecule detection, we choose acetone as an analyte for spectroscopic sensing.Our NMGPDs show a low limit of detection (LoD) of 115 ppm, and a fast gas sensing response and recovery time of 6 s and 11 s, respectively.Our devices show promising potential in realizing chip-scale multifunctional integrated optoelectronic integrated platforms for simultaneously spectroscopic and polarimetric sensing toward real-time environmental monitoring and biomedical screening.

Nanoantenna-Mediated Graphene Photodetectors Design
A schematic perspective view of the designed NMGPDs is sketched in Figure 1b.The graphene flakes are mechanically exfoliated from natural graphite crystals onto the top of a silicon wafer with a 285-nm-thick thermal SiO 2 .An array of Pd/ Au (4/60 nm) DLNAs and source/drain electrodes are deposited on top of graphene layers.The measured atomic force microscopy (AFM) images of graphene flakes (Note S2, Supporting Information) demonstrate that the thickness of the multi-layer graphene flake is about 1.8 nm, indicating it is around 5-6 layers.A net photocurrent is generated under uniform illumination of linear polarization light due to artificial anisotropy.Our experiments are carried out under zero bias unless otherwise specified.Figure 1c shows the false-colored scanning electron microscopy (SEM) image of the fabricated device, and the right zoom-in view shows the geometric parameters.The nanoantennas are designed to have strong resonance at 6.5 µm (L x = L y = 1.2 µm) with a gap of 300 nm to avoid mutual near-field coupling (see Note S3, Supporting Information).The fabricated device (5.7 × 13.6 µm 2 ) is much smaller than the beam size (Gaussian beam, Note S5, Supporting Information), so the illumination on the device can be regarded as uniform.The photoresponse is due to the bulk photovoltaic effect (BPVE), which is usually observed in ferroelectrics, and only few research realized the artificial BPVE. [75]Here, we would like to briefly introduce the photoresponse generation mechanism.Due to the metal plasmons, the light is strongly localized and the near-field intensity is extremely enhanced to generate heat to form hot spots and hot carriers. [79,80]In this work, we consider that the photocurrent originates from the local heating hot spots of light and subsequent heat flow due to inhomogeneities of the Seebeck coefficient S. When the graphene is deposited with metal nanoantennas, a gradient of Seebeck coefficient, ∇S, is formed at the metal-graphene interfaces.And thus, the photocurrent I ph can be written as ∇ , where E is the amplitude of the electric field.Under illumination, the light is localized to the edges of the nanoantennas due to the metal plasma localization, and the electric field is enhanced by several orders.Therefore, the photocurrent can be written as: where E en is the enhanced electric field and n is the normal vector of ∇S.The direction of n is then along the normal of the metal-graphene interfaces, which points from graphene to metal or the opposite, depending on the relative Seebeck coefficients of graphene underneath and out of the metal.Intuitively, the photocurrent in the centrosymmetric systems is zero because photocurrents from opposite directions cancel each other out.Here, we give a quantitative mathematical analysis.
In a 2D plane, the current density , where r is the space coordinate vector, r = xe x + ye y , and the general expression of spontaneous of J is [76] : Adv. Optical Mater.2023, 11, 2202867 where A is the photoresponse tensor, E is the complex amplitude of the electric field, and E * is the complex conjugate of E. a, b, and c are the coordinate indices (i.e., x or y).We can write it in a matrix form: In a centrosymmetric system, it should have J(r) = −J(−r), E (r) = −E(−r), while the photoresponse tensor matrix A should remain the same because A is related to the structural symmetry.It is only possible when A = 0.And hence the photocurrent density is zero in centrosymmetric systems.To obtain a non-zero photocurrent, we introduce the non-centrosymmetry by using DLNAs.Under uniform illumination of linear polarization light, the DLNAs serve as non-centrosymmetric metaatoms, and the photocarriers of each meta-atom shift and form the photocurrents.The direction and magnitude of generated photocurrent are controlled by the polarization angle (θ) of the incident light Figure 1d,e shows the simulated near-field intensity, plasmonic modes, and the respectively predicted vector photocurrents in the unit cell at different polarization angles.The light is localized at the edges of the metal antenna, where the local electric field is enhanced by ≈3 orders.At 0°or 90°, the asymmetric mode or symmetric mode is excited respectively, and both left and right antennas are excited.Recent work shows that the photocarriers gain their momentum through two possible ways. [61]One is the hot carriers are driven through the gradient of the Seebeck coefficient.The other one is via the guidance of high conductance of metal nanoantennas.Therefore, there exists horizontal photocurrent from the left (I xl Figure 1f indicates the measured polarization-dependent photovoltages at 6.5 µm wavelength.The photovoltages are well-fitting with the function sin(2θ).More importantly, the sign-flipped property allows our devices to achieve a negative PR.Meanwhile, the PR of our device is −1, which can be leveraged for polarization-sensitive imaging.Because the photocurrent generated by un-polarized light will be zero in our device, polarized light overwhelmed in the background of unpolarized light will be detected with high contrast.Our devices can be used as an integrated photonic platform to demonstrate surface-enhanced spectroscopic sensing.[83] The red line indicates a typical acetone gas absorption spectrum measured by a Fourier transform infrared (FTIR) spectrometer, which shows that acetone has an absorption peak of around 7.3 µm.Hence, we use another device with a resonance wavelength of around 7.3 µm (L x = 1 µm and L y = 2 µm).The black dots indicate the reduction ratio of the photovoltage after the entry of acetone gas, where V 0 is the photovoltage when no analyte enters.

Device Simulation and Characterization
Figure 2a,b demonstrates the finite-difference time-domain (FDTD) simulated absorption and reflection of the DLNAs with L x = L y = 1.2 µm at different polarization angles, respectively.The resonant wavelengths of our device almost remain unchanged at different polarization angles.Figure 2c shows the respectively measured reflection using an FTIR spectrometer, which validates the simulation results.Tuning resonant wavelengths can be achieved by fine controlling the antenna lengths.Figure 2d,e shows the FDTD simulated absorption and reflection of different antenna lengths (L x and L y , we keep L x = L y in simulation and ranging from 0.9-1.5 µm) at 45°.The resonant wavelengths are redshift when increasing the antenna lengths.Therefore, we can fabricate an array of devices with strong resonance covering a wide LWIR range.

Figure 2f demonstrates the measured FTIR reflection of different antenna lengths
We fabricated an NMGPD with L x = L y = 1.2 µm.Figure 3a shows the measurement setup.A quantum cascade laser (QCL) was used as a light source of linear polarization at 6.5 µm wavelength, and a half-wave plate (HWP) to control the polarization.Figure 3b shows the I d -V DS characteristic curves at V g = 0 V, under dark conditions and illumination conditions of 45° and 135°, respectively.The resistance of our device is about 2.6 kΩ, which reveals the Ohmic contact between the graphene sheet and metal.The I d -V DS curves shift from the origin upward and downward at 45° and 135°, respectively, illustrating the opposite directions of generated photocurrents.Figure 3c indicates the measured gate-voltage-dependent current under dark conditions at V DS = 3 mV.Black and red lines indicate the forward (increasing V g ) and back sweeping (decreasing V g ), respectively.The graphene is p-type doped at V g = 0 V and the neutral critical point is about 40 V. Figure 3d demonstrates the dark noise.The noise is mainly from 1/f noise and Johnson noise, while the shot noise induced by dark current does not exist because our device operates under zero bias.The 1/f noise is significant at frequencies below ≈100 Hz.At higher frequencies, the Johnson noise becomes dominant.Since our device can operate at speeds well beyond 100 Hz, it is not significantly affected by the 1/f noise that is dominant at low frequencies.The dark noise is about 10 nV Hz −1/2 at high frequencies.The NEP, which is about 1.6 nW Hz −1/2 can be obtained by dividing the noise by the responsivity.Figure 3e shows the temporal response of our device at incident light of 45° polarization angle and modulated with an optical chopper of 1 kHz.Figure 3f demonstrates the measured and predicted frequency response of the graphene photodetector.We measured almost constant photovoltage up to 4 kHz (which is limited by our chopper speed) in our experiments.The rise time (t r ) of our device is estimated to be about 1 µs, [76] and t r is proportional to the network time constant τ = RC, which can be obtained by deriving from V t V t r ( ) (1 e ) i = − − , where V (t) is the voltage at time t, V i is the amplitude of the input step signal.At times t 1 and t 2 , the voltages are 0.1V i and 0.9V i , respec- where f 3 dB is the frequency when voltage is down to half (3 dB).Hence the f 3 dB can be estimated by using the rising time t r , and f t dB ln 9 2 3 r π = = 350 kHz. Figure 3g indicates the measured power-dependent photovoltage (polarization angle = 45°), which shows that the photovoltage has a strong linear dependence on incident power and the extracted responsivity is 6.3 V W −1 .As shown in Figure 3h, the responsivity of our device exhibits a wavelength dependence in the LWIR range.The peak around 6.5 µm is the resonance wavelength of our designed nanoantennas.It is worth noting that NEP is related to many factors, such as device temperature, bias, operating wavelength, and modulation frequency of incident light.Responsivity is also related to temperature, bias, and operating wavelength.Besides NEP and responsivity, there are many other important figures of merits.The NEP and responsivity of our devices are achieved at room temperature, in the LWIR range, and under zero bias.Furthermore, our devices can realize this low NEP with incident light modulated of only 100 Hz.We list recent MWIR and LWIR graphene photodetectors for benchmarking NEP and responsivity performance metrics (Table S2, Supporting Information).

LWIR Polarimetry and Small Polarization-Angle Perturbation
To validate the artificial anisotropy and polarization dependence of our devices, we fabricate three devices with different antennas on top of graphene: Device 1 (no antennas), Device 2 (DLNAs), and Device 3 (double-straight antennas).on intrinsic anisotropy.Figure 4c demonstrates the measured polarization-dependent photovoltage of Device 2 under the illumination of different incident powers.Device 2 keeps the polarization dependence and sign-flipped property even though the power decreases to half.The largest polarization-angle sensitivity is achieved at the point θ = 0°, where the photoresponse vanishes.This background-free operation allows us to eliminate the laser intensity noise that is usually dominant in practical applications.Figure 4d shows that Device 2 demonstrates high polarization-angle sensitivity of 5.9 µV per degree near 0°, and the inset indicates the measured photovoltage fluctuation δ is about 0.3 µV, leading to a highly sensitive noise-equivalent polarization-angle perturbation of about 0.05°.

LWIR Spectroscopic Sensing
To evaluate the feasibility of utilizing our NMGPDs for an onchip integrated platform for LWIR polarimetric and spectroscopic sensing, we select acetone as the analyte to examine the spectroscopic sensing performance.We fabricate another device with a resonant wavelength of around 7.3 µm with L x = 1 µm and L y = 2 µm (Note S7, Supporting Information).The sensing setup is depicted in Figure 5a, which can be divided into the optical module and the gas regulation module (see details in Experimental Section).The optical module is the same setup described in Figure 3a.For the gas regulation module, nitrogen (N 2 ) is selected as the buffered gas, which is divided into two flows (one flow is pure N 2 and the other flow is pumped through 99.5% acetone solutions to generate N 2 -acetone mixed gas).The acetone concentration in the dilution can be fine-tuned by controlling the flow rate and can be accurately calibrated by a commercial sensor.Figure 5b demonstrates the normalized measured photoresponse when pure N 2 and 750 ppm acetone-N 2 dilution are alternately injected into the gas chamber.A clear and repeatable drop of the photovoltage is observed under the injection of acetone.Figure 5c,d shows the gas sensing response and recovery stages extracted from Figure 5b, respectively, which indicate a gas sensing response time of 6 s and a gas sensing recovery time of 11 s, which are fast enough for real-time on-chip monitoring applications.Figure 5e indicates the normalized noise level of our sensing systems with pure N 2 (no analytes).The normalized standard error σ of the noise is about 0.23% and the corresponding 1−σ LoD is about 115 ppm.As Figure 5f shows, we measured the relative output drop (ΔV/V 0 ) with a smaller wavelength change (Δλ), and the output drop matches well with the typical absorption spectrum of acetone.Therefore, we can conclude that the output drop of our devices is due to the absorption of acetone. [25]urthermore, we also investigate the sensing performance of ethanol, which has an absorption peak near 7.2 µm.As shown in Note S7, Supporting Information, our devices can operate from 7.0-7.5 µm.Therefore, we use the same device as acetone sensing for ethanol detection under the illumination of 7.2 µm.
As demonstrated in Figure 5g, a clear and recoverable drop of the photovoltage is observed under the injection and ejection of 3250 ppm ethanol.Hence our integrated platform is ready to demonstrate simultaneously spectroscopic and polarimetric detection for potential biomedical applications.

Conclusion
In conclusion, we propose chip-level LWIR NMGPDs as an integrated platform for polarimetric and spectroscopic sensing.The polarization-dependent photoresponse is realized by the artificial anisotropy introduced by DLNAs.Furthermore, our devices demonstrate a sensitive noise-equivalent polarization-angle perturbation down to 0.05° in the LWIR range.Our devices work under zero bias and achieve high room temperature responsivity as well as a low NEP.To further investigate the feasibility of our devices as a monolithic integrated platform including polarimetric and spectroscopic sensing for potential biomedical applications, we demonstrate acetone spectroscopy with high performance.Our devices show a low LoD and fast dynamic gas sensing response.Our NMGPD paves the way to the realization of next-generation on-chip integrated optoelectronics for simultaneous spectroscopic and polarimetric sensing.In addition, by introducing a nanoantenna array of wavelength-scale devices, it is possible to develop high-resolution polarization imaging and a multifunctional monolithic integrated platform for real-time health monitoring and medical applications. [84]

Experimental Section
Fabrication: In the first step, the graphene flakes used in the experiments were mechanically exfoliated from graphite crystals onto the Si/SiO 2 substrate.The alignment marks were patterned using electron-beam lithography (EBL, Jeol) and deposited Ti and Au using E-beam evaporator (AJA) followed by the standard lift-off process.The device areas of graphene flakes were then patterned using EBL and the residue of graphene was removed by oxygen plasma.After that, the metal nanoantennas and electrodes consisting of Pd and Au, were deposited on top of graphene layers using EBL and E-beam evaporator followed by the lift-off process.
Simulation: The numerical simulations in this work were carried out using the finite-difference time-domainFDTD) method (Lumerical Inc.).The size parameters and period of nanoantennas were the same as experimental ones.Reflection, transmission, and electric field profiles were extracted by using three frequency-domain field and power monitors.
Characterization: A quantum cascade laser (QCL, Daylight Solutions, MIRCat-2300) with a wavelength of 6.2-8.6 µm was used as the linearly polarized light source.The polarization angle was controlled by a zeroorder HWP (Altechna).Light of different frequencies was modulated using an optical chopper (Stanford Research Systems, SR540).I d -V DS and I d -V g curves were measured using a semiconductor characterization system (Keithley 4200-SCS).In the characterization experiments, the polarization angle was always set to 45° unless otherwise stated.A lock-in amplifier (Stanford Research Systems, SR830) was used to measure the frequency response and dark noise of the device.An oscilloscope and preamplifier (Stanford Research Systems, SR560) were used to measure the time response.The power of the incident light was varied by using a neutral density filter (Thorlabs) and varying the laser current.The total laser power was measured by a power meter (843-R, Newport).
Polarization Dependence: The optical path and measurement setup were depicted in Section 2.3.In the measurement of small angle perturbation, the small changes in polarization angles were achieved by fine controlling the HWP.The photovoltage fluctuation was measured at a small angle close to 0°.
Spectroscopic Sensing: The spectroscopic sensing setup of these devices could be divided into the optical characterization module and the gas regulation module.For the optical module, described in Section 2.3, the polarization angle of the incident light is fixed at 45°.These units were integrated with a homemade gas chamber with a CaF 2 window to allow the passage of the LWIR laser.In the gas regulating module, nitrogen (N 2 ) was selected as the buffer gas and divided into two flows.One was pure N 2 and the other was pumped into an acetone solution to produce an acetone-N 2 mixture.Then, the two air flows were remixed, entering the air chamber from the air inlet, and finally exiting the air outlet.The concentration of acetone in the dilution could be precisely and dynamically controlled by adjusting the valves in both flows and calibrated with commercial sensors for volatile organic compounds.

Figure 1 .
Figure 1.Design concept and main results of NMGPDs.a) Illustration on the integration of conventional polarization selective detection and infrared spectroscopic sensors into a monolithic detector.b) Illustration of the designed NMGPDs, which consists of non-centrosymmetric sub-wavelength DLNAs as meta-atoms on top of graphene layers.Under uniform illumination of polarized light, directional photocurrents are generated from each meta-atom at zero bias (V DS = V g = 0 V).c) False-colored SEM image of the DLNAs and the geometric parameters.d) The simulated near-field intensity of different polarization angles.e) Simulated plasmonic modes and respective predicted vectorial photocurrent (blue) in a unit cell at different polarization angles (black) of incident light.The middle inset shows the experimentally measured photocurrent I ph (green), which is the scalar projection of I ph (blue) on the orientation of source-drain electrodes.f) Measured polarization-dependent photoresponse, the "+" and "−" represents the direction of photocurrent (along and opposite the source-drain direction), and the sign-flipped property of the photocurrent indicates a bipolar polarization detection and a negative PR. g) Spectroscopic sensing of acetone.The red line indicates a typical acetone gas absorption spectrum measured by FTIR.The black dots indicate the reduction ratio of the photovoltage output after the entry of acetone gas, where V 0 is the photovoltage output when no analyte enters.
) and right nanoantnena (I xr ) with opposite directions and two vertical photocurrents (I v ) with the same direction from both the left and right nanoantenna.The photocurrent generated in the x direction (I x ) is predicted to be zero because the photocurrent from the left (I xl ) and right side (I xr ) cancel each other out due to the mirror symmetry ( I x = I xl − I xr = 0), and only the vertical photocurrents (I v ) are left.While in experiments, we can only measure the scalar projection of photocurrent I ph in the x direction.At other polarization angles, both asymmetric and symmetric modes are excited but with a phase delay between them.Especially at 45° (135°), the phase delay is 0 (π) due to the symmetry, leading to constructive (destructive) interference on the left antenna and destructive (constructive) on the right.From 0° to 45° (90° to 135°), partially constructive interference in the left (right) antenna, while a destructive interference in the right (left) antenna, and thus I xl (I xr ) increases with I xr (I xl ) decreases and hence the absolute value |I x | = |I xl − I xr | increases.From 45° to 90° (135° to 180°), without surprise, the absolute value of I x decreases.The photocurrent at any other polarization angle (θ) is the difference between the contributions from the two eigenmodes, I xl and I xr .We can take the 45° and 135° as two orthogonal eigenmodes because pure I xl or I xr is achieved respectively.And thus, I (θ) = I 45 × cos 2 (θ − 45°) − I 135 × sin 2 (θ − 45°), where I 45 and I 135 are the absolute values of photocurrent when polarization angles are 45° and 135°, respectively.It should have I 45 = I 135 due to the mirror symmetry of the DLNAs.Hence, we can rewrite the photocurrent as:

Figure 2 .
Figure 2. Simulation and optical measurement.a,b) Simulated absorption and reflection of different polarization angles using FDTD.The bottom insets show the geometric parameters of the DLNAs and the definition of polarization angles.c) Measured reflection of different polarization angles using FTIR.d,e) Simulated absorption and reflection of different antenna lengths (L x and L y ) using FDTD.We keep L x = L y in simulation and experiments.f) Measured reflection of different antenna lengths using FTIR.
Figure 4a illustrates the SEM image of three fabricated devices.The graphene layers of these three devices are divided from the same graphene flake.Device 1 and Device 2 have the same width between the two electrodes.Device 2 and Device 3 have the same period of antennas in the x-direction, and Device 2 also has a resonant wavelength of around 6.5 µm (L x = 0, L y = 2.3 µm, Note S6, Supporting Information).As shown in Figure 4b, Device 2 demonstrates large and polarization-dependent photo voltage, while Device 1 and Device 3 have almost zero photovoltage.Comparing Device 1 and Device 2, we can conclude that the anisotropy is from the artificial nanoantenna structure.Comparing Device 2 and Device 3, we validate the analysis that non-zero photoresponse comes from non-centrosymmetric systems.It is worth noting that the photovoltage of Device 2 shows a flipped-sign property, achieving a negative PR, which is unusual in the 2D materials polarization detectors based

Figure 3 .
Figure 3. Device characterization.a) Measurement setup.b) Measured I d -V DS lines in dark conditions and illuminated conditions with polarized light of 45° and 135° at 6.5 µm wavelength.The used device was DLNAs with L x = L y = 1.2 µm at zero gate voltage (V g = 0).c) Measured I d -V g hysteresis curves at dark conditions.The applied V DS is 3 mV.Black and red lines indicate the forward and back sweeping respectively (forward sweep: increasing V g .back sweep: decreasing V g ).Graphene is p-type doped at V g = 0 V. d) Measured dark noise.The bottom dash line indicates the Johnson noise limit.e) Measured time response of our device with the illumination of polarization angle of 45° (Pol = 45°) and signal chopped at 1 kHz.f) Measured and predicted frequency response.g) Measured power-dependent photovoltage at zero bias and extracted responsivity of 6.3 V W −1 .h) Measured wavelength-dependent responsivity in the LWIR range.The peak around 6.5 µm is the resonance wavelength of our designed nanoantennas.

Figure 4 .
Figure 4. LWIR polarimetry and small polarization-angle perturbation.a) SEM images of three different-type devices: no antennas (Device 1), DLNAs (Device 2), and double-straight antennas (Device 3).b) Polarization dependence of Device 1-3.c) Polarization dependence of Device 2 under the illumination of different incident powers at 6.5 µm wavelength.d) Fine measurement of the polarization dependence, showing a polarization-angle sensitivity of 5.9 µV per degree at polarization angle θ = 0°.The inset shows the measured voltage fluctuation, δ = 0.3 µV.Therefore, the LoD of polarization-angle perturbation of our device is 0.05°.

Figure 5 .
Figure 5. Demonstration of spectroscopic sensing using acetone as analyte.a) Spectroscopic sensing measurement set up. b) Gas sensing response and recovery characteristic cycle curve of our device with 750 ppm acetone.c) Gas sensing response time.d) Gas sensing recovery time.e) Measured ΔV/V 0 of our device without acetone, which indicates the setup noise.The standard error σ is around 0.23% and the corresponding 1−σ LoD is about 115 ppm.f) The typical absorption spectrum of acetone gas and measured relative output drop (ΔV/V 0 ) of our devices.g) Gas sensing response and recovery characteristic cycle curve of our device with 3250 ppm ethanol.