Near-infrared (NIR) CMOS image sensors (CIS) are increasingly used in various applications such as automotives, security systems, biological inspection, and mobile devices [1–4]. NIR sensitivity is an important optical parameter for improving the image quality in NIR CISs. However, silicon-based CISs exhibit poor NIR sensitivity at NIR wavelengths compared to visible wavelengths due to low absorption efficiency and limitations in silicon thickness, which results in more than a 50% reduction in efficiency for NIR wavelengths at the thickness used for visible pixels [5–7]. Therefore, numerous studies are being conducted not only on silicon-based CIS but also on alternative approaches to enhance the optical performance of NIR sensors [6–14].

Recent advancements in nanotechnology have led to the development of various nanostructured materials specifically designed to enhance NIR sensitivity [8–10]. By integrating Si nanowire (NW) arrays into CIS architecture, researchers have achieved significant improvements in light absorption across the NIR spectrum [8]. These Si NWs are designed with optimized geometry and doping profiles, resulting in substantial gains in quantum efficiency (QE). Although Si NWs show potential, challenges related to scalability and integration with existing CMOS processes still need to be addressed before they can be widely implemented. On the other hand, metasurface-enhanced photodetectors represent another effective approach [9]. These devices use engineered nanostructures on the sensor surface to manipulate light at sub-wavelength scales, thereby increasing the interaction time between light and the sensor material. This technique has shown potential in enhancing NIR sensitivity and is compatible with current CMOS technologies. However, manufacturing scalability and long-term operational stability have yet to be achieved. Additionally, quantum dot integration has been explored as a method of extending the absorption spectrum of CIS [10]. By incorporating quantum dots into the sensor design, researchers have been able to enhance NIR sensitivity significantly, though issues related to quantum dot uniformity and integration complexity persist.

Numerous research efforts in the field of CISs are also actively focused on developing high-performance silicon-based NIR sensors. These include increasing the thickness of the silicon layer, using plasmonic structures, implementing an inverted pyramid array (IPA) structure, and employing a backside scattering technique (BST) for light scattering. One of the primary methods involves increasing the thickness of the silicon photodiode. This approach has produced significant improvements in QE by allowing more NIR photons to be absorbed [6]. For instance, increasing the thickness of the silicon layer from 3 μm to 6 μm has been shown to improve sensitivity by up to 60% at 940 nm [7]. However, this method has drawbacks, including increased optical crosstalk and reduced spatial resolution, which are particularly problematic due to aspect ratio issues. Additionally, silicon-based NIR sensors are being explored through pixel design in addition to the development of plasmonic structures. Using the resonant condition of the plasmonic sensor, the optical path extends because of the diffraction of incident light [11]. Plasmon resonance converts much of the incident light on the surface into surface plasmon waves, reducing sensor surface reflectance and enhancing light efficiency. However, the integration of plasmonic structures, such as silver gratings, can also cause a dark current due to the additional electronic noise generated by near-surface plasmon resonance conditions. Using light diffraction, an IPA silicon surface has been developed to enhance both the optical path and the effectiveness of silicon thickness [12, 13]. To improve light trapping, an IPA structure with high-refractive-index c-Si and low-refractive-index materials such as SiO₂ is inserted on the Si surface to effectively confine light within the pixel, and deep-trench isolation (DTI) is employed to block crosstalk. Therefore, the use of a 400 nm pitch IPA surface combined with DTI showed an 80% improvement in sensitivity. However, issues such as the degradation of spatial resolution, increased color crosstalk, and manufacturing complexity are yet to be addressed. BST also improves NIR QE by scattering light within the pixel, increasing the optical path and enhancing light absorption in silicon [14, 15]. This technique effectively increases NIR sensitivity by using optimized scattering patterns, leading to up to a 150% improvement in QE at 940 nm. When combined with DTI and anti-reflection layers, BST further enhances performance while reducing crosstalk and optical efficiency. Furthermore, similarly to BST, there is the in-pixel DTI structure where DTI is inserted within the pixel to address sensitivity imbalance [16]. BST involves incorporating a scattering structure within the pixel, while in-pixel DTI was proposed as a solution to the sensitivity imbalance between pixels caused by incident light in the pixel structure [14–16]. In addition, in-pixel DTI is used to confine light within individual pixels, minimizing optical crosstalk and improving sensitivity and resolution in high-density CISs. However, the enhancement in NIR sensitivity remains poor when compared to visible light.

In this paper, improvement in the sensitivity of NIR image sensors based on various patterns of an in-pixel scattering trench (IST) is studied. The IST is a method used to extend the optical path through scattering structures. The design of the IST is similar to BST and in-pixel DTI. We analyze the differences between the IST patterns and shapes and optimize the IST’s dimensions (widths, lengths, and depths), along with the silicon-fill factor (Si-FF), to be consistent with the optimal pattern. Our goal is to provide the best design guide for the IST by analyzing sensitivity trends for various patterns and conducting optimization studies. Section 2 describes the concept behind the IST, and Section 3 presents the simulation results and discusses the results. Conclusions are presented in Section 4.

The most important factor for improving NIR sensitivity is achieving the optical path required for the absorption of NIR wavelengths. According to [17], to effectively absorb incident light in the NIR wavelength range of 0.8 μm to 1.0 μm, the optical path as the absorption length must be at least 10 μm. Increasing the thickness of silicon extends the optical path, resulting in the sensitivity of 6.0-μm-thick silicon increasing by 42% and 60% at NIR wavelengths of 850 nm and 940 nm, respectively, compared to 3.0-μm-thick silicon [7]. However, by increasing the thickness, the aspect ratio gets worse. Therefore, the optical crosstalk increases, and the spatial resolution becomes lower. For this reason, a scattering structure needs to be used to extend the optical path. Among the various scattering methods, BST and in-pixel DTI are the most convenient to fabricate. In this paper, we will analyze and optimize the IST structures to make them similar to BST and in-pixel DTI. It was decided to name it “in-pixel” DTI because it is a scattering structure within a pixel. This is also to eliminate confusion between BST and in-pixel DTI.

Next, we will explain the concept of the IST by describing the typical pixel and IST structures. Figure 1(a) shows a typical 3D back-side-illuminated (BSI) pixel structure. The Bayer color filter array is used by red, green, and blue CF. This silicon-based pixel structure is designed for absorption in the visible wavelength range. When NIR wavelengths are incident, longer wavelengths can pass beyond the photodiode, leading to optical loss, as shown in Fig. 1(b). However, in Fig. 1(c), the structure with the integrated IST demonstrates how the incident light is scattered, resulting in an increased optical path length. The scattering-induced extended optical path effectively enhances the absorption of NIR, specifically improving sensitivity to longer wavelengths. To help understand this visually, the beam profiles of pixels with and without the IST pattern are shown in Figs. 1(d) and 1(e). The beam profile represents the power flux density of the incident light and is used to analyze the intensity and direction of the electromagnetic wave. As shown in Fig. 1(d), the incident light passes through the photodiode, resulting in loss, as explained in Fig. 1(b). On the other hand, as shown in Fig. 1(e), the incident light is scattered by the IST, resulting in an extended optical path. As explained in Fig. 1(c), the extended light is absorbed at the bottom of the photodiode. As a result, it can be confirmed that the optical path is extended by the IST. Therefore, an analysis of various patterns in the IST structure is necessary, and with this analysis, a design with higher sensitivity will be achieved.

Figure 1. Concept drawings for 3D and 2D cross sections of the BSI pixel structure with or without IST. (a) Overview of 3D pixel structure. The 2D cross section of the pixel structures (b) without IST, (c) with IST. Beam profiles of the pixel structure (d) without IST, and (e) with IST. BSI, back-side-illuminated; IST, in-pixel scattering trench.

Next, the 12 proposed IST patterns used in the analysis of this paper will be explained. Figure 2 presents the 2D cross sections of the conventional structure and the 12 proposed IST patterns. Figure 2(a) shows a typical structure, composed only of DTI without the IST. Figures 2(b)–2(d) have a cross shape, Figs. 2(e)–2(g) show the merged shapes, 2(h) and 2(i) are square shapes, 2(j) and 2(k) represent the merged square shapes, and finally, 2(l) and 2(m) are net shapes. The merged shapes are based on both crosses and squares and are combined to create various forms. In this study, many patterns were simulated, and we selected and analyzed the 12 patterns that showed the highest levels of improvement in NIR sensitivity.

Figure 2. In-pixel scattering trench (IST) patterns: (a) No IST, (b) pattern A, (c) pattern B, (d) pattern C, (e) pattern D, (f) pattern E, (g) pattern F, (h) pattern G, (i) pattern H, (j) pattern I, (k) pattern J, (l) pattern K, and (m) pattern L.

Next, we will look at the analysis process, which begins by using the sensitivity of the typical structure and the 12-IST pattern-inserted structures to evaluate the degree of improvement. Secondly, the most and least effective beam profiles are analyzed to help us understand how light scattering varies according to the pattern and how it affects sensitivity. Finally, the Si-FF of the incident light z-axis is examined to determine the optimal Si-FF of the area for light incidence based on the pattern.

We investigated the optical properties of the proposed IST structure using a 3D optical simulator based on the finite difference time domain (FDTD) method [18]. The FDTD method is commonly employed for the numerical analysis of CMOS image sensors [19]. Figure 3(a) illustrates the simulated 3D pixel structure, which depicts a typical BSI pixel [20]. Figure 3(b) shows the 2D cross-sectional view of the 3D pixel. The structure used in the simulation has a pixel size of 2.0 μm and a depth of 3.0 μm. The height and radius of curvature of the micro-lens are optimized for NIR wavelengths, measured at 1.2 μm and 1.4 μm, respectively, while the DTI is optimized for the 2.0 μm pixel size and measured at 100 nm. Figures 3(c) and 3(d) show the y-axis and z-axis of the inserted IST pixel structure. The IST was optimized according to the cross, square, net, and merged shapes described in Fig. 2. The dimensions of each of the 12 structures were optimized, as shown in Table 1. The width varied from 50 to 500 nm for the cross and net shapes, and from 50 to 1,000 nm for the square shapes. For the square shapes, both the width and length were varied, while for the cross and net shapes, the length varied from 100 nm to 1,800 nm (the maximum length excluding DTI). The depth varied from 500 to 2,000 nm by 500 nm. Unlike other parameters such as the width and length of the IST, the sensitivity does not show a big difference with the interval. Next, sensitivity according to the IST will be analyzed.

Figure 3. Simulated BSI structures with or without in-pixel scattering trench (IST). (a) 3D pixel structure. 2D cross-section of the pixel structures (b) without IST, (c) with IST, and (d) with IST pixel z-axis cut view.

TABLE 1. Optimized simulation parameters of IST.

Pattern	Ty	A	B	C	D	E	F	G	H	I	J	K	L
Width (nm)	…	385	385	350	245	280	300	850	420	500	360	275	200
Length (nm)		1,400	1,130	1,600	1,600	1,800	1,800	850	1270	500	360	1,800	1,800
Depth (nm)		500	2,000	2,000	2,000	500	500	2,000	1,000	1,500	500	500	2,000

In this study, the absorbed photon density was used as an indicator of sensitivity to evaluate the optical characteristics of the pixel. The absorbed photon density was specified as the absorbed power density divided by photon energy. Figure 4(a) shows the simulated sensitivity of pixels with no IST and the 12 IST patterns. The absorption of NIR wavelengths is improved in all structures compared to the typical structure. Notably, the simulation results confirm that the best structure, pattern C, achieved an enhancement rate of 386%, while the least effective structure, pattern G, showed an enhancement rate of 267% when compared to the typical structure. The cross-shaped patterns A, B, and C exhibited different enhancement rates. Among the cross-shaped patterns, higher sensitivity was observed particularly in the +orientation. Additionally, the square-shaped patterns G and H exhibited lower sensitivity improvement rates. The square-shaped patterns exhibited relatively less scattering, resulting in a lower sensitivity improvement rate. This analysis will be explained in more detail later in the beam profile analysis. When comparing the merged and non-merged patterns, the improvement in sensitivity was much higher in the merged pattern than in the non-merged pattern due to the merging effect. Finally, even though the net patterns had the same shape, there were significant differences in sensitivity because the patterns are more complex on the silicon surface. Next, the analysis moves on to the Si-FF of the maximum sensitivity improvement for each of the 12 patterns.

Figure 4. Simulation results of near-infrared (NIR) wavelength sensitivity in 2.0 μm pixel size, with improvements observed in the 12 different IST patterns. (a) Simulation results of the normalized sensitivity with NIR wavelength. (b) Simulation results of the Si-FF with NIR wavelength.

Figure 4(b) illustrates the Si-FF for each structure. The Si-FF is an important parameter in the analysis since it reflects the shape of the pattern and is determined by the optimal width and length. It also represents the area of the silicon surface where light is incident. For the typical structure, excluding DTI, the silicon cross-sectional area represents the maximum Si-FF, calculated as 100%. As shown in Fig. 4(b), for pattern C, which demonstrated the best sensitivity enhancement, the optimal Si-FF was 69%. In contrast, for pattern G, which exhibited the lowest sensitivity enhancement, the optimal Si-FF was 78%. Excluding the worst pattern G, the Si-FF at which maximum sensitivity enhancement occurs for each pattern is observed between approximately 50% and 70%, and this trend can be analyzed.

For a visual understanding, Fig. 5 presents the beam profile of the power flux density for an incident NIR wavelength at an angle of 0 degrees. The beam profiles of the patterns with the best and worst sensitivity at a 940-nm NIR wavelength with 0 degrees compared with the typical structure, are provided to compare both cases in order to observe how light scattering and absorption vary depending on the shape of the IST. The power flux density and the region in the beam profile indicate where the most transmitted light could be absorbed by the photodiode. Figure 5(d) shows the scale bar of power flux density, where higher sensitivity corresponds to higher density, appearing in red. The structure with the integrated IST scatters the incident light, extending the optical path length and consequently improving NIR wavelength absorption. Figure 5(a) shows the typical structure. When light is incident, it can be observed that the incident light passes through the photodiode. However, as shown in Fig. 5(b), for the pattern with the best sensitivity, C, the incident light is scattered by the IST, extending the optical path, with strong absorption observed at the bottom of the photodiode.

Figure 5. Beam profiles of the simulated power flux density comparing the typical, best, and worst sensitivity structures at a 940-nm NIR wavelength with 0°. The power flux density and the region in the beam profile indicate where the most transmitted light could be absorbed by the photodiode. (a) Typical structure without a pattern, (b) best sensitivity pattern C, (c) and worst sensitivity pattern G. (d) The scale bar of power flux density.

In contrast, as shown in Fig. 5(c), for pattern G, the pattern with the worst sensitivity, the light is scattered by the IST, extending the optical path, but absorption appears weak at the bottom of the photodiode. Comparison of the best and worst sensitivity patterns shows the differences in the optical path extension length and absorption at the bottom of the photodiode.

Figure 6 presents the simulation results of the Si-FF to observe the trend analysis for each of the 12 patterns. The width of each of the 12 patterns varied from a minimum value of 50 nm to a maximum of 1,000 nm, while the length varied from a minimum of 100 nm to a maximum of 1,800 nm. The best Si-FF in the figure represents the Si-FF when the width and length are optimized. For the analysis, the patterns were categorized into cross-shaped forms A to C, merged forms D to F, square-shaped forms G and H, square merged forms I and J, and net-shaped forms K and L. The cross-shaped patterns A, B, and C show a trend with two peak points. However, while the best Si-FF differs considerably between the structures, Table 1 shows that the widths range from 350 to 385 nm, indicating little variation. Also, the square-shaped patterns G and H exhibit a single peak point, but their best Si-FF values differ. In contrast, the sensitivity and the best Si-FF of the merged square patterns I and J show similar trends and values. The net-shaped patterns K and L exhibit a single peak point and show similar best Si-FF values. Lastly, in the complex merged patterns D, E, and F, the best Si-FF values are similar, and it can be observed that the widths and lengths in Table 1 also show similar values, ranging from 245 to 300 nm and 1,600 to 1,800 nm, respectively. In summary, patterns within the same category present the same sensitivity trend line, but differences appear in their best Si-FFs. Additionally, the sensitivity enhancement rates show a variation of at least 10% to more than 20% even among patterns within the same categories, affected by the shape of the pattern and further affected by the optimization of width and length. Finally, when considering the best Si-FF for each of the 12 patterns, the values range between approximately 50% and 70%. This will be analyzed in more detail in Fig. 7.

Figure 6. Simulation results of Si-FF for NIR wavelength sensitivity in 2.0 μm pixel size, with improvements observed in the 12 different IST patterns. Each pattern is varied by width and length, and the best Si-FF in the figure corresponds to the optimal width and length. (a) pattern A with IST, (b) pattern B, (c) pattern C, (d) pattern D, (e) pattern E, (f) pattern F, (g) pattern G, (h) pattern H, (i) pattern I, (j) pattern J, (k) pattern K, and (l) pattern L. Si-FF, silicon full factor; NIR, Near-infrared; IST, in-pixel scattering trench.

Figure 7. Simulation results of Si-FF for NIR wavelength sensitivity at 2.0-μm pixel size, with improvements observed for all IST patterns, and Gaussian-fitted curve generated using a Gaussian filter. Si-FF, silicon full factor; NIR, near-infrared; IST, in-pixel scattering trench.

Figure 7 is a graph that plots all of the Si-FF variations for each of the 12 patterns as dot points. All the points represent sensitivity as a function of Si-FF for all simulated structures. To visualize the variations in the simulation results for Si-FF, a Gaussian-fitted curve was added, and an explanation of this curve is provided below. A Gaussian-fitted curve was created using a Gaussian filter, which smooths the data by applying a weighted average according to a Gaussian distribution. The weighting is based on the average sensitivity according to the Si-FF. Due to the reduction of the maximum sensitivity from 100% to approximately 80% based on the weighted average sensitivity, the peak value of the Gaussian-fitted curve is shown as 80%. An analysis of the best Si-FF shows that while the optimal Si-FF differs for each structure, the majority are within the range of 50% to 70%. The highest sensitivity was observed at a Si-FF of 60%, which is considered a significant optical parameter, and the point where the maximum sensitivity reaches 80% occurs at a Si-FF of 60%. Clearly, each pattern has its own specific best Si-FF. However, when proposing and simulating other shapes beyond the patterns used in this paper, it is reasonable to use Si-FF values between approximately 50% and 70% as the initial reference for varying width and length, as shown in Fig. 7. This is because it is the optimal silicon area that minimizes loss when NIR light is incident and limited by the IST. Therefore, the optimal IST width is 200 to 400 nm, resulting in the peak of the Gaussian-fitted curve for the best Si-FF appearing around 60%. This analysis can assist in optimizing the design, width, and length of the scattering structure.

In this study, we analyzed a high-sensitivity NIR pixel design incorporating IST. Many patterns were simulated, and we selected and analyzed the 12 patterns that showed the highest improvement in NIR sensitivity. The simulation results show that the best sensitivity, pattern C, achieved a 386% enhancement rate, while the worst, pattern G, had a 267% enhancement rate, both compared to the typical structure. The cross-shaped patterns A, B, and C showed varying enhancement rates, with higher sensitivity observed in the + orientation. In contrast, the square-shaped patterns G and H had lower levels of improvement in sensitivity due to less scattering. When comparing merged patterns, the improvement in sensitivity was greater. Finally, the net patterns had the same shape, but there were significant differences in their sensitivity because the patterns are more complex on the silicon surface. Using beam profile analysis, the scattering and absorption of incident light were examined based on the pattern shape. In particular, for the best sensitivity pattern C, the optical path was longer compared to the worst sensitivity pattern G, with power concentrated at the bottom of the photodiode, resulting in higher sensitivity. The analysis of Si-FF results shows that patterns within the same category present similar trend lines, but differences appear in their best Si-FF. However, the optimized widths and lengths appear to be similar. Additionally, the sensitivity improvement rates show a variation of at least 10% to more than 20% even among similar patterns, affected by the shape of the pattern and further affected by the optimization of width and length. Furthermore, the analysis revealed that the optimal Si-FF values for most patterns were within the 50% to 70% range, because it is the optimal silicon area that minimizes loss when NIR light is incident and limited by the IST. Therefore, the optimal IST width is 200 to 400 nm, resulting in the peak of the Gaussian-fitted curve for the best Si-FF appearing around 60%. As a result, the simulations demonstrate the effectiveness of the proposed design and provide valuable insights for optimizing NIR pixel structures.

NIR applications require highly sensitive pixel structures. To improve NIR sensitivity, the optical path for effective absorption length must be at least 10 µm. Extending the thickness of the silicon to increase the optical path can lead to negative effects such as optical crosstalk and reduced spatial resolution. Therefore, a scattering structure is needed. This study focuses on analyzing and optimizing the IST, similar to BST and in-pixel DTI, to enhance NIR sensitivity. Accordingly, we propose a high-sensitivity NIR pixel by analyzing important factors such as the various shapes of ISTs, optimization, and Si-FF. Many studies have been performed to improve NIR sensitivity by extending the optical path using scattering structures. This paper provides a reference for the various shapes and optimization factors that should come under consideration when designing scattering structures.

The EDA tool was supported by the IC Design Education Center (IDEC), Korea.

National Research Foundation of Korea (NRF) grant funded by the Korean government (MIST) (Grant No. NRF-2022R1A6A3A13070373).

The authors declare no conflicts of interest.

The data that support the findings of this study are available from the corresponding author upon reasonable request.