Improving the near infrared region sensitivity of CMOS image sensors

Silicon-based CMOS image sensors inherently exhibit low near-infrared (NIR) absorption, which limits quantum efficiency and restricts their use in applications requiring extended spectral sensitivity such as biomedical imaging, LiDAR, surveillance, autonomous navigation, and machine vision. This thesis addresses this limitation by proposing four CMOS-compatible pixel architectures designed to improve NIR performance through sub-wavelength light trapping, optical confinement, photon recycling, and hybrid material integration. The structures are analyzed using three-dimensional Finite-Difference Time-Domain (FDTD) simulations. A 2 × 2 µm Bayer-patterned unit cell with a 3 µm silicon photodiode is used consistently, and optical efficiency (OE) is evaluated over the 400–1100 nm range. Absorption and confinement behaviors are examined through power and field-intensity monitors, while a conventional backside-illuminated (BSI) CMOS pixel is used as the reference. The four proposed architectures employ complementary enhancement mechanisms, including hemispherical photon trapping, Si–GaAs hybrid photodiodes, inverted-pyramid texturing with metallic DTI, and SiO2-based photonic structures integrated with reflective substrates. The results demonstrate that all designs significantly increase NIR sensitivity compared to the reference structure. The best-performing pixel achieves an OE of 0.824 at 1100 nm, corresponding to an approximate 285% improvement while maintaining visible-band performance. These findings confirm that geometrically engineered and hybrid-material-based strategies can effectively mitigate silicon's intrinsic absorption limitations and provide a strong foundation for developing next-generation broadband CMOS image sensors with enhanced NIR sensitivity.