Figure 1: Spectral engineering in CMOS image sensors.
1. Pre-reading
Complementary metal–oxide–semiconductor (CMOS) image sensors play a pivotal role in a wide range of applications that rely on digital imaging systems. Typically, a standard CMOS image sensor (CIS) comprises a configuration of microlenses, a color filter array, and photodetectors, as shown in Fig. 1(a). However, with the increasing demand for higher resolution, there has been a significant reduction in pixel size, leading to a notable decrease in optical efficiency. The ideal solution for overcoming these losses is to design a device capable of directing incident light to the appropriate photodetector, depending on the wavelength regions. These color-routing techniques have demonstrated efficient color separation using various approaches. However, studies on adjoint optimization-based color routers, which have demonstrated superior performance in color routing, have shown that they unintentionally direct a significant amount of light into neighboring pixels, i.e., “interpixel crosstalk”, as shown in Fig. 1(b).
The recent study conducted by Dr. Haejun Chung’s group at Hanyang University addresses the issue of interpixel crosstalk in color routing-based image sensors. They propose a series of novel structural designs and employ custom incident waves to mitigate this effect significantly. Their approach includes the insertion of a physical interlayer between the repeated color router structures and utilizing a customized incident source, which collectively demonstrates a remarkable reduction in interpixel crosstalk as shown in Fig. 1(c).
The color router developed in this study is expected to pave the way for commercializing color-routing-based CMOS image sensors.
2. Background
Color routing-based image sensors have been proposed to mitigate the optical losses associated with the reduction in pixel size of conventional image sensors. The color routers are designed to direct incident light to the appropriate photodetector depending on the wavelength regions and to serve as direct substitutes for the combination of microlenses and color filters. The color-routing techniques have demonstrated efficient color separation using various approaches, including single-layer metasurface-based strategies, laterally or vertically stacked metasurfaces, and inverse design methods.
Adjoint optimization, a representative method in inverse design, facilitates large-scale optimization of photonic structures. This algorithm calculates derivatives of the chosen figure of merit (FoM) resulting from permittivity changes using only a pair of simulation runs: a direct simulation and an adjoint simulation. Specifically, a slight adjustment in permittivity within the designable region can be approximated as a superposition of a simulation without permittivity changes and a simulation with dipole excitation at the permittivity change location. The Lorentz reciprocity principle then allows for replacing numerous simulations of dipoles at various positions within the designable region with a single dipole simulation at the detector location. This approach simplifies the calculation of FoM derivatives concerning permittivity changes, enabling the resolution of complex photonic problems with numerous design parameters within a feasible computational cost.
The previous studies on adjoint optimization-based color routers have demonstrated its ultimate performance in color sorting. However, the color routers have suffered from interpixel crosstalk, which refers to crosstalk between neighboring pixels, as shown in Fig. 2.
Figure 2: Intensity profiles of the five periods of optimized color router at three representative wavelengths (450, 550, and 650nm).
3. Innovative research
An innovative approach to mitigating the interpixel crosstalk issue of adjoint optimization-based color routers involves inserting a physical interlayer between the repeated color router structures. The materials employed for the interlayer include tungsten (Fig. 3) or air (Fig. 4), which have been utilized in CMOS image sensor processes for color filter isolation. This method substantially reduces interpixel crosstalk and enhances imaging resolution, albeit at the expense of elevated manufacturing costs due to the interlayer incorporation.
Figure 3: Interpixel crosstalk calculation of the optimized structure with tungsten isolation walls.
Figure 4: Interpixel crosstalk calculation of the optimized structure with air gaps.
Previous studies have utilized plane waves as incident waves within a Bloch boundary condition framework, where the incident waves positively contribute to optical efficiency even when absorbed by the photodetectors of neighboring pixels. In this research, the authors propose placing Perfectly Matched Layers (PMLs) on both sides of a structure comprising a repeated design area and employing an unfocused Gaussian beam to simulate a plane wave incident on a centrally positioned unit cell as shown in Fig. 5. This approach effectively suppressed interpixel crosstalk without incurring additional production costs.
Figure 5: Interpixel calculation of the optimized structure with unfocused Gaussian beam.
4. Applications and perspectives
Incorporating a physical interlayer and utilizing a customized incident wave have effectively mitigated the interpixel crosstalk observed in previously studied color routers. Notably, the 3D color routing approach, which combines interlayer and customized source methods, has demonstrated superior optical efficiency. The color router proposed in this study is expected to facilitate the commercialization of color routing-based CMOS image sensors. However, the proposed color routers may present challenges related to fabrication costs and the resolution limits of current lithography techniques. Therefore, advancements in lithography technology will be essential for successfully implementing these devices.
These research results are published online with the title “Inverse design of color routers in CMOS image sensors: toward minimizing interpixel crosstalk” on Nanophotonics.
The authors of this article are Sangbin Lee, Jaehyun Hong, Joonho Kang, Junjeong Park, Jaesung Lim, Taeho Lee, Minseok Jang, and Haejun Chung. Haejun Chung is the corresponding author of this work. Dr. Haejun Chung’s research group is affiliated with the Institute for Information and Communications Technology Planning and Evaluation (IITP), National Research Foundation of Korea (NRF), Ministry of Trade, Industry & Energy (MOTIE), Korea Semiconductor Research Consortium (KSRC) and the Department of Electronic Engineering, Hanyang University, Seoul, 04763, South Korea.