Macroscopic Assembled Graphene for Silicon Mid-Infrared Photodetectors


 Graphene with linear energy dispersion and weak electron-phonon interaction is highly anticipated to harvest hot-electrons in a broad wavelength range from ultraviolet to terahertz. However, the limited absorption (~2.3%) and serious backscattering of hot-electrons associated with single-layer graphene result in inadequate quantum yields, impeding their practically broadband photodetection, especially in the mid-infrared range. Here, we report a macroscopic assembled graphene (MAG)/silicon heterojunction for ultrafast mid-infrared photodetection. The highly crystalline 2-inch scale MAG with tunable thickness from 10 to 60 nm is produced by scalable wet-assembly of commercial graphene oxide followed by thermal annealing. The MAG/Si Schottky diode exhibits broadband photodetection capability in 1-10 μm at room temperature with fast response (120-130 ns, 4 mm2 window) and high detectivity (1011 to 106 Jones), outperforming single-layer graphene/Si photodetectors by 2 to 8 orders in transient photocurrent. This optoelectronic performance is attributed to the superior advantages of MAG (~40% of light absorption, ~23 ps of carrier relaxation time, and high quasi-equilibrated hot-carrier-multiplication gain), atomic-scale contact interface of MAG and silicon, and impact-ionization avalanche gain (~100 times) from silicon. The MAG provides a long-range platform to understand the hot-carrier dynamics in stacked 2D materials, leading to next-generation broadband silicon-based image sensors.

number fluctuation from the surface effect (traps and disorders) 17 . Experimental results verify a 2 to 4 orders lower NSD of MAG/Si than that of SLG/Si at bias voltage -1 V to -70 V (Fig. 2f), respectively, which allows a larger signal-to-noise ratio (SNR) for MAG/Si, especially during the avalanche.
Based on the absorption and NSD properties, the SLG/Si device only shows detectable photocurrents at wavelength ≤ 1.5 μm ( Supplementary Fig. S10c, d). In contrast, the MAG/Si device exhibits distinguishable photoresponse in a wide wavelength range (1-10 μm, Fig. 2g). The enhanced absorption and suppressed noise give rise to 2 to 8 orders increasing of SNR of the MAG/Si compared with that of SLG/Si photodetector ( Supplementary Fig. S10c, d).
As a broadband photodetector, the MAG/Si diode exhibits both fast operation speed and high sensitivity. To explore the ultrafast optoelectronic dynamics in MAG/Si, we used the femtosecond (pulse width, Δt = 200fs) photoexcitations in MIR 18 . A typical external circuit limited response time τr~120-130 ns (4 mm 2 window, Fig. 3a) was obtained at wavelengths of 2 and 4 μm. By reducing the device parasitic capacitance (Supplementary Fig. S13) and resistance in external circuits, a higher speed can be achieved. While at the wavelength of 7 μm, the τr~300 ns indicates a different operation mechanism, which will be discussed in the following section. In terms of photoresponse, the MAG/Si exhibits a 10 11 to 10 6 Jones specific detectivity (D*) and a 10 -12 to 10 -6 W Hz -0.5 noise-equivalent-power (NEP) in the 1-10 μm at room temperature (Fig. 3b), far beyond the detection capabilities of previously reported silicon-based high-speed infrared photodetectors (Extended Data Fig. 1).
The broad photoresponse beyond the intrinsic absorption of silicon in MAG/Si features three distinct working mechanisms (Fig. 3c). The MAG/Si SBH, extracted from the temperature-dependent I-V characteristics ( Supplementary Fig. S14d-e), is~0.3 eV, corresponding to a cutoff wavelength of 2.1 μm (see Supplementary Note 5). For photon energies of 0.5hν > SBH (wavelength < 2.1 μm), photo-excited electrons directly transport over the barrier and contribute to the photocurrent through the internal photoemission (regime I in Fig. 3c) 19 , where the D* and NEP remain almost constant. When increasing the incident wavelength to 4.6 μm, the energy of direct photo-excited electrons is lower than SBH, so they cannot cross the SBH immediately but thermalize into a Fermi-Dirac distribution in MAG. The thermalized hot-electrons with energy higher than SBH can emit into silicon afterward ( photo-thermionic (PTI) emission, regime II) 4 . In this regime, the portion of hot-electrons with the energy higher than SBH in the Fermi-Dirac distribution drops as the incident photon energy decreases, lowering D* and increasing NEP greatly (Fig. 3c). In the region of 4.6-10 μm (regime III), negative photocurrents are collected, which possibly results from the screening, photo-gating, and photo-acoustic effects of the massive low energy hot-electrons at the MAG/Si interface 20-22 . Most of the hot-electrons generated in this regime cannot jump over the SBH, resulting in a slight change of photocurrent and thus D* and NEP as the wavelength increases. The overall three different trends of responsivity (as indicated with circles in Fig. 3d) as a function of incident laser irradiance at different wavelengths also confirm these three dominant electron-transfer mechanisms (Fig. 3d). The cutoff wavelengths of these three regimes can be tuned and the D* and NEP can be further optimized by tuning the doping state of the silicon (as shown in Supplementary Fig. S10e) or replacing silicon with a semiconductor with higher electron affinity (χ).
According to Fig. 3c, d, we find that the significantly enhanced hot-carrier scattering effect of MAG determines the 2-4 µm MIR detection. The reasons are two-fold. Firstly, MAG has an enhanced light absorption, generating more photo-excited hot-electrons than that of SLG.
Secondly, the 45-nm thick cross-section of MAG provides a larger space for hot-electrons to scatter and multiply. At the moderate electric field (~10 3 V cm -1 ), a large number of hot-electrons in MAG transport towards the MAG/Si interface facilitated by the AB-stacking dominant structure: (1) the prolonged hot-carrier lifetime and thus the long out-of-plane hot-carrier diffusion length 23 , (2) the relaxed conservation of in-plane momentum from carrier scattering 24 , and (3) the finite out-of-plane velocity (vz) from the momentum uncertainly due to the presence of a few misorientations in the stacking order (Fig. 1f). The synergistic combination of the effects (2-3) enhances the wave-like over particle nature of electrons until they completely transfer into silicon with a definite momentum, increasing the hot-electron density overcoming the MAG/Si barrier.
Hence, the charge transfer efficiency mainly depends on the out-of-plane hot-electron diffusion length (out-of-plane velocity multiplies relaxation time). Moreover, in SLG/Si case, based on the simplified Vickers-Mooney model, only a portion of hot-electrons with energy and momentum falling within the escape cone is transferred into silicon, and the rest scatter back to SLG 25 . In comparison, in the MAG/Si case, some of the backscattered hot-electrons are expected to be redirected into the escape cone through phonon and interface wall scattering due to the large available cross-section of MAG (Fig. 3c, regime II). Therefore, the trade-off between out-of-plane charge transfer and the recombination in MAG achieves the maximum quantum efficiency (QE) at 45 nm in our case (Fig. 3e).
To further investigate the hot-carrier dynamics in MAG/Si, we conducted a pump-probe transient absorption measurement (Supplementary Fig. S16) on our detectors. The pump laser wavelength is 3.5 μm, with the probe from 1.2-1.6 μm, which is just above the SBH. When pump laser fluences is > 1 mW mm -2 , the scattered electron occupation close to the charge neutrality  (Fig. 4a). The AR process dominates carrier kinetics in MAG until it reaches the optical phonon bottleneck (~200 meV) at~1 ps, and then the relaxation occurs via an interaction with acoustic phonons within~23 ps as displayed in Fig. 4b. Due to the thick crystalline MAG and non-polar property of silicon, the cooling from disorder and substrate-based surface plasmon polariton is weak in our devices 27 . The dominant carrier relaxation mechanism gradually shifts from AR to acoustic phonon emission at longer wavelengths, which is further confirmed from the elimination of the fast component (within 1.5 ps) in the bi-exponential fitting results (Fig. 4b). The out-of-plane hot-electron diffusion coefficient and cooling time are about 1 cm 2 s -1 and 23 ps, respectively. From the saturation absorption and power dependent analysis, the two-photon or multi-photon absorption from either MAG or silicon is ruled out in our case ( Supplementary Fig.   S17). By extracting the number of hot-electrons n(E) from transient absorption ΔA and the current collected from silicon, the hot-electron transfer efficiency ηe (3.5 μm) between MAG and silicon is obtained on the order of 10 -3 -10 -5 under the irradiance from 10 to 100 mW mm -2 . We also quantified the hot-electron scattering effect by calculating the hot-carrier-multiplication. The density of hot-electrons with higher energy than chemical potential exceeds that in the conduction band immediately after photo-excitation with a peak multiplication value~20 (1 mW mm -2 ) at 8 At the high electric field regime (~10 5 V cm -1 ), a~5 μm depleted region in silicon is induced, generating additional electron-hole pairs through impact ionization. The current amplification gain due to the avalanche in silicon reaches 10 2 as the bias Vb increases up to -70 V for all laser irradiance (Fig. 4d) with an excess noise factor of 10 (Supplementary Note 4). The increased NSD (Fig. 2f) with the bias also indicates the avalanche in silicon 28 . The drastic increase at -30 V indicates the beginning of the avalanche (Supplementary Fig. S19). Combining the carrier-carrier scattering in MAG and carrier multiplication through the avalanche in silicon, the MAG/Si shows an external QE of 0.1-10 % at 3.5 μm under the irradiance from 10 to 100 mW mm -2 (Fig. 4e).
MAG is a promising candidate for CMOS technology due to its robust mechanical property, uniform structure, and low cost. As shown in Fig. 5a, MAG was etched into a neat array composed of 50 μm-scale pixels by standard lithography and oxygen plasma 29,30 . Combined with CMOS-compatible process flow, MAG/Si was packed into a 9 × 9 pixel array image sensor (Fig.   5b, c). The D* variation among each pixel was tested less than one order (Fig. 3b) because of the uniformity of MAG in thickness and stacking order. Fig. 5d presents the room-temperature images captured by the MAG/Si with a mask of the Chinese character " 杭 (Hang)" at the wavelengths 1.55 μm, 3 μm, and 10 μm. The array-level CMOS compatibility of wafer-scale MAG offers opportunities to develop room-temperature silicon image sensors at broad infrared frequencies that conventional Schottky diodes are not readily applicable.
This work demonstrates a high-performance MAG/Si Schottky diode with fast response time, low NSD, and record response wavelength, far beyond state-of-the-art silicon-based high-speed infrared photodetectors. Such outstanding detection capability is achieved by introducing high crystalline MAG that compatible well with CMOS technology as an absorption layer, which is integrated with silicon with a clean and atomic-scale contact interface. This work opens a new avenue from macroscopic defective GO to high-performance optoelectronic devices, provides a strategy to explore hot-carrier dynamics in highly stacked two-dimensional systems, and demonstrates a feasible way to develop low-cost and large-scale broadband graphene-based photodetectors for CMOS image sensors at room temperature. Raman spectra of MAG on the silicon substrate. The 2D peak is fitted with 3 Lorentzian peaks: 2D1 (~2680 cm -1 ) and 2D2 (~2720 cm -1 ) for AB stacked graphite and 2DT (~2700 cm -1 ) for turbostratic graphite (see Supplementary Fig. S4d). wavelength under a fixed irradiance of 10 mW mm -2 (see Supplementary Fig. S14a-c). c, the corresponding energy-band diagrams for operation mechanism at the three wavelengths above. d, The responsivity of MAG/Si as a function of laser irradiance at different wavelengths. e, The internal-QE and hot-carrier life-time of MAG/Si as a function of MAG thickness at 4 μm laser wavelength with a fixed irradiance of 20 mW mm -2 (see Supplementary Fig. S15). were connected with an oscilloscope (Keysight DSO 9404A, 4 GHz bandwidth) to measure the photo-voltages. Periodic pulse lasers (Light Conversion, OPA-Series, 1 μm to 10 μm, 200 fs pulse width, and 100 kHz repetition rate) were used as the light source. The noise spectra were recorded by a noise measurement system (PDA NC300L, 100 kHz bandwidth). Infrared bandpass filters (FB2000-500 to FB4000-500, Thorlabs) were used to filter out the stray light from the MIR laser.
A calcium fluoride aspherical lens system was used to focus the infrared light on the device. The power was measured by a thermopile detector (Newport 1918-R). For the photograph, we built a scanning system with a focused laser and programmable X-Y stage. A data acquisition (DAQ) card was used to obtain the photocurrent data.
Ultrafast transient absorption spectroscopy measurement. The femtosecond transient absorption setup used for this study was based on a PHAROS laser system (Light Conversion, 1030 nm, < 190 fs, 200 uJ per pulse, and 100 kHz repetition rate), nonlinear frequency mixing techniques, and the Femto-TA100 spectrometer (Time-Tech Spectra). Briefly, the 1030 nm output pulse from the regenerative amplifier was split into two parts with an 80 % beam splitter. The reflected part was used to pump an ORPHEUS Optical Parametric Amplifier (OPA) which generates a wavelength-tunable laser pulse from 300 nm to 15 m. The transmitted 1030 nm beam was split again into two parts. One part with less than 50 % was attenuated with a neutral density filter and focused into a YAG window to generate a white light continuum (WLC) from 500 nm to 1600 nm used for probe beam. The probe beam was focused with an Ag parabolic reflector onto the sample. After the sample, the probe beam was collimated and then focused on a fiber-coupled spectrometer with CMOS sensors and detected at a frequency of 10 kHz. The intensity of the pump pulse used in the experiment was controlled by a variable neutral-density filter wheel. The delay between the pump and probe pulses was controlled by a motorized delay stage. The pump pulses were chopped by a synchronized chopper at 5 kHz and the absorbance change was calculated with two adjacent probe pulses (pump-blocked and pump-unblocked). All experiments were performed at room temperature.