The pursuit of highly efficient photodetectors has garnered substantial
interest recently, driven by their wide-ranging applications spanning environmental
monitoring, communication systems, and imaging technologies. This chapter examines
how combining few-layer graphene (FLG), zinc oxide (ZnO), and silicon (Si) in a dual
heterojunction design can create a photodetector that captures light across a wide range,
from ultraviolet (UV) to near-infrared wavelengths. Utilizing the exceptional properties
of graphene, the wide-bandgap nature of ZnO, and the scalability of Si as a substrate,
the dual-heterojunction is optimized for UV-to-NIR light absorption and efficient
carrier transport. A comprehensive simulation and analysis of the photodetector's
characteristic parameters, including biased and unbiased operation, is conducted using
the Silvaco Atlas TCAD software. The results show enhanced performance, with a
least dark current density (Jdark) of 2.7×10–15 A/cm2, superior photocurrent density
(Jlight) of 0.26 µA/cm2, along with a remarkable Jlight/Jdark ratio of 9.77×107, a 3-dB
cut-off frequency of 7.36 THz, and a rapid rise (fall) time of 0.47 (0.88) ns at –1.0 V.
Under illumination conditions, the photodetector exhibits a peak external quantum
efficiency of 69.3%, photocurrent responsivity of 0.26 A/W, detectivity of 8.12×1015
cmHz1/2/W, and noise equivalent power of 7.79×10–20 W at –1.0 V bias. This research advances photodetection capabilities by demonstrating that combining p+
-
FLG, n-ZnO, and n-Si in a dual-heterojunction design enables effective light detection
from UV-to-NIR wavelengths. The findings suggest this integrated approach could lead
to improved sensors and imaging devices. Moreover, a comprehensive comparative
analysis is performed, utilizing the power of different machine learning (ML)
regression models to evaluate their predictive performance in modeling the
photodetector's behavior. The main goal is to determine the effectiveness of each ML
regression model in accurately forecasting the photodetector's characteristics. This will
be achieved by evaluating model performance across varying test set sizes using a
range of metrics such as mean squared errors, root mean squared errors, mean absolute
errors, and R2
scores. The findings highlight the utility of machine learning algorithms
in modeling and optimizing optoelectronic devices, offering valuable insights into their
potential applications in advancing photodetector research and development endeavors.
Keywords: Broadband, Dark current density, Dual-heterojunction, Few-layer graphene, Machine learning, Near-infrared, Noise current, Photocurrent density, Quantum efficiency, Ultraviolet, ZnO.
