Charge-coupled devices (CCD) and complementary metaloxide–semiconductor (CMOS) technology are widely used in high-end imaging applications. Both have their own merits and limitations. The simple metal–oxide–semiconductor photogate pixel and charge integration function in CCDs results in a high fill factor (FF), high sensitivity and low noise, but the serial charge transfer between wells requires complex multiphase biasing/clocking, limiting the readout speed. In CMOS sensors, the independent pixel structure allows random access, simple clocking, high-speed parallel readout, natural anti-blooming and monolithic integration with processing circuitry. However, CMOS usually has a lower FF and higher noise than CCD due to the relatively complex active pixel sensor structure and subsequent circuitry. Both silicon-based CCD and CMOS imagers are also generally not applicable beyond the visible range due to the intrinsic bandgap absorption limit of silicon. This is usually overcome through the heterogeneous integration of metal silicide Schottky barrier detectors; however, these have a complicated structure, low FF and high cost, and they require low-temperature operation. The integration of narrow-gap III–V or II–VI semiconductors can extend the spectral response regime, but this is technically challenging, costly and typically limited to the near-infrared (NIR) regime. Further expansion into the mid-infrared (MIR) regime has been a persistent challenge due to the intrinsic thermal noise in narrow-gap semiconductors. Graphene is of potential use in imaging devices due to its monolithic integration with CMOS, strong field effect for electrostatic photogating and broadband absorption spectrum. Incorporating graphene into silicon-based image sensors could thus be used to improve the responsivity and spectral performance.

In the traditional CCD structure, multiphase clock pulses are applied to the pixels to serially transfer photocharges to the readout circuit. Such a multiphase serial charge transfer is notably not necessary in our GCI as direct real-time charge readout can be realized from each potential well via a highly charge-sensitive photogating effect of the SLG. Specifically, the photo- and thermally generated holes are integrated in the deep-depletion well of silicon, whereas the same number of electrons are imaged in the SLG through capacitive coupling, resulting in a real-time change in drain current Id of the SLG. This real-time monitoring of charge in the GCI has several advantages over traditional CCD image sensors. First, the charges in the potential well are directly read out, avoiding serial charge transfer between the adjacent wells required in traditional CCDs. Second, the charge readout and charge integration process are independent, that is, the readout operation does not interrupt the charge integration process (the readout is non-destructive). Third, the photogating effect from the SLG, where the output signal of the GCI is amplified by the field-effect modulation of the graphene sensor with a high transconductance gain, contributes to high sensitivity and fast response time. The use of the photogating effect to improve both sensitivity and speed has been reported previously, which demonstrated the combination of photosensing and charge storage techniques. Here we advance this strategy by integrating a monolayer readout transistor with a deep-depleted potential well, which substantially improves the sensitivity via the photocharge integration effect in a controllable manner.

In this Article, we report graphene charge-injection (GCI) photodetectors that combine the charge integration feature of CCD and the independent pixel structure of CMOS. The photodetector consists of a deep-depletion SiO₂/Si well sandwiched between single-layer graphene (SLG) on top of the structure (for direct field-effect readout) and multilayer graphene (MLG) at the bottom (for broadband charge injection).  Combined with tunable responsivity, which can be used as the weight, the GCI can be potentially used to build neuromorphic networks. The combination of charge injection, integration and direct independent readout in each pixel could be used to make new imaging devices with applications in various fields.

In our device, the PTI emission allows a high Schottky barrier between the MLG and silicon that suppresses the thermal noise—the main limiting factor for room-temperature operation in traditional IR photodetectors. Although the high barrier results in a lower quantum efficiency, the synergetic combination of PTI, charge integration and transconductance amplification provides a large responsivity that is initially compromised for lower noise in our device structure. The superlinear relationship of ISch,pc ≈ Pγ and tunable EQE by VSch are two direct evidences that hot-carrier PTI emission from the MLG to silicon dominates the IR response. In addition, we exclude the contribution of heat from various factors. First, heating in SLG and MLG can lead to the bolometric effect. In SLG, an increase in resistance under IR laser illumination results in a negative photocurrent, a typical feature of the bolometric effect. However, the responsivity from the bolometric effect is much lower (approximately five orders of magnitude) than the response observed from the photocharge-injection effect (2.0 μA W⁻¹ of the bolometric effect of SLG  and 0.6 A W−1 of photocharge-injection effect. Besides, other thermal effects that could lead to the change in mobility of MLG cannot translate into a response in the SLG channel and thus can be ruled out. The heat transfer from the MLG to silicon and the consequent thermal carrier generation and integration in the deep-depletion well of silicon can also be excluded. Such a heat transfer effect is not controlled by the external voltage, contrary to that observed in Fig. 3d. We also considered the heat effects in the silicon substrate. In principle, thermally generated carriers in silicon integrate in the deep-depletion well that reflects in the channel current Id of graphene. However, the negligible IR response in the GOS structure without MLG on the bottom silicon surface excludes the heat effect from silicon. Since the hot electron generation and transport in MLG play key roles in IR detection via the PTI effect in our device, the photothermoelectric (PTE) effect can also be expected in MLG. The PTE effect in MLG emerges from the electron temperature gradient that transport hot electrons towards the heat sink or the junction. Considering the thickness of MLG, the PTE effect is primarily influenced by the interlayer hot-electron transfer rather than the Seebeck coefficient in MLG. As a result, hot electrons distribute into a wider area of MLG and thus helps to enhance the PTI effect at the MLG/Si junction, contributing to the charge-injection process. Besides, MLG could produce a small resistive change under light illumination and change the back-gate voltage accordingly, contributing to the photoresponse. However, this resistive change is too small to produce a detectable photoresponse even under large transconductance gain as a very little Vg drops on the MLG; thus, it cannot dominate the photoresponse in our devices. The photoresponse from surface-state absorption can also be eliminated in the MIR region as our control reference devices show a weak response.

https://www.nature.com/articles/s41928-022-00755-5

Nature Electronics volume 5, pages 281–288 (2022)