Spectral imaging system and method for manufacturing spectral imaging system

Abstract

The purpose of the present invention is to suppress variation in wavelength dependency of transmittance caused by differences in etching rates for each portion of an optical filter. A spectral imaging system according to the present invention is provided with a plurality of optical filters having mutually different wavelength dependencies of light transmittance, wherein: the optical filter has through-holes smaller than the pixel size of an image sensor; the optical filter is supported by a transparent material that transmits incident light; and the arithmetic average roughness on the transparent material-side of the interface between the optical filter and the transparent material is 10 nm or less (see fig. 8).

Description

Spectroscopic imaging system, method for manufacturing a spectroscopic imaging system

[0001] This invention relates to a spectral imaging system that detects incident light by spectrally analyzing it.

[0002] When observing substances or objects with cameras or microscopes, reflected or transmitted light from the substance or object is detected by the camera or microscope's detector (such as an image sensor) and converted into an image. Detectors capture all wavelengths of light for which they are sensitive; for example, a Si CMOS image sensor is sensitive to light with wavelengths of approximately 0.4 to 1 μm, and detects light within that wavelength by integration. In contrast, spectral imaging obtains images at specific wavelengths by spectrally analyzing reflected or transmitted light and detecting light of specific wavelengths with a detector. When the number of wavelengths (bands) is several to a dozen or so, i.e., when there are several to a dozen or so corresponding images, it is sometimes called multispectral imaging, and when there are more, it is called hyperspectral imaging. Generally, different substances or objects absorb different wavelengths, so by using images acquired through spectral imaging, it is possible to classify the type of substance or object, or to determine the amount of water it contains.

[0003] Multispectral imaging can be achieved by placing multiple types of optical filters on an image sensor that allow only specific wavelengths of light to pass through. For example, a color image sensor has color filters that allow red, green, and blue light to pass through, so it can be considered a type of multispectral imaging. On the other hand, when a similar method is used in hyperspectral imaging, the spatial resolution decreases due to the large number of optical filters, so a method is sometimes used in which light dispersed by a spectroscopic element such as a diffraction grating is irradiated onto the image sensor. However, since the light is dispersed in the x or y direction and irradiated onto an image sensor composed of a two-dimensional array, the remaining number of dimensions becomes 1, i.e., it becomes a line sensor, and it is not possible to acquire the entire image at once.

[0004] On the other hand, there is a known compressive sensing technique that significantly reduces the number of filters used in an optical filter-based approach compared to the number of bands for which the number of filters is desired. This technique uses multiple types of optical filters that transmit light of various wavelengths but have different transmittances depending on the wavelength, rather than an optical filter that transmits only light of a specific wavelength. The intensity of the light transmitted through each filter is detected by integrating it in the wavelength direction using an image sensor, and the spectrum of the incident light is reconstructed from the intensity information. To increase the accuracy of the reconstructed spectrum, it is preferable that the wavelength dependencies of the transmittances among different types of filters are as different as possible. This technique also uses multiple types of filters, and although the spatial resolution is inferior to that of an image without using filters, it can significantly improve the spatial resolution compared to the case of providing the number of filters corresponding to the desired number of bands.

[0005] The following Non-Patent Document 1 describes a technique for acquiring a spectroscopic image using compressive sensing.

[0006] Single-shot on-chip spectral sensors based on photonic crystal slabs, Zhu Wang et al., NATURE COMMUNICATIONS (2019) 10:1020

[0007] As a technique for forming multiple types of optical filters with different wavelength dependencies of transmittances on the same surface, there is a known technique of forming a structure (referred to as a sub-wavelength structure including a structure of about the wavelength) having a dimension of about the wavelength to be measured or shorter in an array. Since the wavelength dependence of the transmittance can be modulated according to the shape, dimension, and period on a two-dimensional plane such as a circle or a square, by changing these parameters for each region, that is, by forming multiple types of patterns, multiple regions with different wavelength dependencies of transmittances can be formed on the same surface. These optical filters are hereinafter referred to as photonic crystal filters.

[0008] While conventional semiconductor manufacturing processes are used to form subwavelength structures, the aspect ratio cannot be made very high, making single-film formation difficult; therefore, they are formed on a support substrate. Since the formed nanostructures are stacked on an image sensor, a material transparent in the wavelength range to be measured is used as the substrate. When forming subwavelength structures by etching a film deposited on the substrate, if various dimensions and periods are mixed in the plane, the etching rate will differ for each structure due to the loading effect. This can result in some patterns being etched only partway through the film, while others are etched all the way to the substrate. Even in these states, it is possible to modulate the wavelength dependence of the transmittance, but the wavelength dependence of the transmittance will differ compared to a structure etched to the depth of the film thickness. Furthermore, if in-plane distributions generated by semiconductor manufacturing equipment and variations during repeated processing are superimposed, the wavelength dependence of the transmittance will deviate from the value assumed in the design, even if the same pattern is intended to be formed.

[0009] When reconstructing the spectrum of incident light, information on the wavelength dependence of the photonic crystal filter's transmittance is needed, either directly or indirectly. In principle, it is possible to measure the wavelength dependence of transmittance at each location and use that information to correct the reconstruction model. However, this presents challenges such as the time required to retrain the reconstruction model, the enormous amount of information needed for correction straining memory, and the time required for correction. Furthermore, if the wavelength dependence of transmittance at a specific location is used as a representative value and applied to the entire region to reduce the amount of information, there is a problem in that the reconstructed spectrum does not faithfully reproduce the spectrum of the incident light.

[0010] This invention has been made in view of the above-mentioned problems, and aims to suppress variations in the wavelength dependence of transmittance caused by differences in etching rates for each part of the optical filter.

[0011] The spectroscopic imaging system according to the present invention comprises a plurality of optical filters whose light transmittance has different wavelength dependencies, the optical filters are composed of through-holes smaller than the pixel size of the image sensor, the optical filters are supported by a transparent material that transmits incident light, and the arithmetic mean roughness on the transparent material side of the interface between the optical filters and the transparent material is 10 nm or less.

[0012] The spectroscopic imaging system according to the present invention can suppress variations in the wavelength dependence of transmittance caused by differences in etching rates at different parts of the optical filter. Other problems, configurations, and advantages of the present invention will become clear from the following description of embodiments.

[0013] This is an overall configuration diagram of the spectroscopic imaging system according to Embodiment 1. This is a top view showing an enlarged portion of the photonic crystal filter 11. This shows an example of transmittance when light passes through the photonic crystal filter 11 and information on the wavelength dependence of the transmittance extracted therefrom. This is a schematic diagram showing the procedure by which the computing device 24 reconstructs hyperspectral image data. This is a side cross-sectional view of the photonic crystal filter 11. This shows an example of the configuration of data describing the wavelength dependence of transmittance. This is a side cross-sectional view illustrating a conventional process for manufacturing the photonic crystal filter 11. This is a side cross-sectional view illustrating the process for manufacturing the photonic crystal filter 11 according to Embodiment 1. This is a side cross-sectional view illustrating the process for manufacturing the photonic crystal filter 11 according to Embodiment 2. This shows an example in which the SiN film is formed in a tapered shape during the process of manufacturing the photonic crystal filter 11. This is a side cross-sectional view illustrating another problem when the SiN film is tapered. This is a side cross-sectional view illustrating the process for manufacturing the photonic crystal filter 11 according to Embodiment 3.

[0014] <Embodiment 1> Figure 1 is an overall configuration diagram of a spectral imaging system according to Embodiment 1 of the present invention. The spectral imaging system is a system for acquiring spectral images of an object to be measured 3, and comprises a filter 21, a lens 22, a camera 1, a cable 23, a computing device 24, and a light source 25. Reflected light from the object to be measured 3, illuminated by the light source 25, is focused on the camera 1 through the filter 21 and lens 22, and the camera 1 captures a spectral image.

[0015] The filter 21 consists of either a short-wavelength cut filter or a long-wavelength cut filter, or both, to limit the wavelength range to be measured. If it is not necessary to limit the wavelength range, the filter 21 may be omitted. The filter 21 is usually made of a multilayer film. The lens 22 is selected according to the spatial region to be measured, and the distance to the object to be measured 3 is set so that it is in focus.

[0016] Camera 1 comprises a photonic crystal filter 11 (optical filter), an image sensor 12, a readout circuit 13, and a casing 14. An example configuration of the photonic crystal filter 11 will be described later. Light transmitted through the photonic crystal filter 11 is detected by the image sensor 12, and the detection signal is then acquired by the readout circuit 13. The detection signal acquired by the readout circuit 13 is processed by the arithmetic unit 24 and recorded as a spectral image. A microlens array may be arranged on the image sensor 12, but for the sake of explanation, it is omitted in Figure 1.

[0017] Figure 2 is a top view showing an enlarged portion of the photonic crystal filter 11. The photonic crystal filter 11 is an optical filter in which subwavelength structures, each having dimensions similar to or shorter than the wavelength range to be measured, are formed in an array. Although only some of the holes constituting the subwavelength structures are shown in Figure 2, in reality, numerous holes are formed. The shape of the holes is not limited to circles or squares.

[0018] The upper layer of the photonic crystal filter 11 includes a substrate for supporting the photonic crystal filter 11, and an anti-reflective coating may be formed on the upper layer thereof. Since the photonic crystal filter 11 needs to transmit some light, it is formed thinly, with a film thickness of, for example, 100 nm to 500 nm. The photonic crystal filter 11 is manufactured separately from the image sensor 12, and the two are integrated by bonding them together.

[0019] In Figure 2, nine different hole shapes are arranged in a repeating pattern. Each pattern may correspond to one pixel of the image sensor 12 directly below it, or one pattern may be formed over multiple pixels. The dimensions of each pattern are formed to be an integer multiple of the pixel size of the image sensor 12. In Figure 2, a filter is formed by nine different hole shapes, but there may be any number of patterns.

[0020] Figure 3 shows an example of transmittance when light passes through the photonic crystal filter 11, and information on the wavelength dependence of the transmittance extracted therefrom. The pattern and material of the photonic crystal filter 11 are designed so that the light transmittance of each pattern is as different as possible from one another in the target wavelength range to be measured. The wavelength dependence of the transmittance is continuous, but the values ​​are extracted discretely according to the wavelength resolution to be measured, such as every 1 nm or every 10 nm. In the case of nine types of patterns repeated as shown in Figure 2, there are ideally nine types of transmittance, so nine types of transmittance values ​​are extracted for each discretely set wavelength value.

[0021] For each wavelength of light that passes through a given pattern, the transmittance value when light of that wavelength passes through the pattern can be obtained discretely. For example, the transmittance of the first wavelength among the wavelengths of light that pass through pattern 1 can be expressed as T11, for instance.

[0022] In addition, the effective transmittance may be used as the transmittance, which is a superposition of values ​​related to the characteristics of objects present in the path of light, such as the quantum efficiency of the image sensor 12 and the transmittance of the lens 22.

[0023] Figure 4 is a schematic diagram showing the procedure by which the computing unit 24 reconstructs hyperspectral image data. The computing unit 24 creates an encoded image as an image captured by the image sensor 12 through the photonic crystal filter 11. The computing unit 24 further stores the wavelength dependence of the transmittance of each pattern in the photonic crystal filter 11 in a storage device beforehand and reads it out. The computing unit 24 creates a hyperspectral image by inputting these into the reconstruction model. Note that by training the reconstruction model in advance using the wavelength dependence of the transmittance of each pattern in the photonic crystal filter 11, it is also possible to configure the system to use only encoded images as input data.

[0024] When a 3x3 pattern is considered as one pattern set, if the pattern set is manufactured uniformly regardless of its planar position on the photonic crystal filter 11, the transmittance characteristics will be common to all pattern sets (they will all be the same). Therefore, data describing the wavelength dependence (e.g., the matrix shown in the lower part of Figure 4) can be composed of nine pieces of information for each wavelength.

[0025] Figure 5 is a side cross-sectional view of the photonic crystal filter 11. To form the photonic crystal filter 11, holes of various sizes and densities must be formed. Generally, when forming holes, a film such as SiN is formed on a transparent substrate, a resist pattern is formed on top of it, and then holes are formed by etching. At this time, the etching rate differs depending on the size and density of the holes, so as shown in Figure 5, there are regions where the rate of hole formation is relatively fast and regions where it is slow. As a result, the depth of the holes differs from hole to hole. For example, as shown in Figure 5, there will be a mixture of regions where the holes do not penetrate the film on the transparent substrate and regions where the holes penetrate to the transparent substrate.

[0026] Figure 6 shows an example of data structure describing the wavelength dependence of transmittance. As shown in Figure 5, if the hole depth differs for each planar position on the photonic crystal filter 11, the transmittance will differ for each pattern set, as shown in Figure 6. Furthermore, if the in-plane distribution of film thickness of the deposition apparatus before hole formation and the in-plane etching distribution of the hole formation apparatus are superimposed, the transmittance will exhibit location dependence even if the same pattern is intended to be formed on the photonic crystal filter 11. This also contributes to the difference in transmittance for each pattern set.

[0027] Even in such cases, it is possible to make the transmittance different for each pattern. However, if a hyperspectral image is created using information from only a single pattern set as the wavelength dependence of transmittance, as shown in Figure 4, the hyperspectral image will not faithfully reproduce the actual spectrum. Therefore, in order to faithfully reproduce the actual spectrum of the object being measured 3, it is necessary to input the wavelength dependence of transmittance for each location into the reconstruction model. This leads to problems such as the reconstruction model taking a long time to retrain, or the amount of information required for correction becoming enormous and putting a strain on memory, or the correction taking a long time.

[0028] In view of the above-mentioned problems, this embodiment proposes a method in which the hole formation is allowed to continue without stopping midway, thereby ensuring that the hole depth equals the film thickness and improving the uniformity of the depth.

[0029] Figure 7 is a side cross-sectional view illustrating a conventional process for manufacturing a photonic crystal filter 11. A film for forming a hole pattern (in this case, a SiN film) is deposited on a transparent substrate for transmitting light (1). A resist pattern is formed on the SiN film (2), and the SiN film is patterned by etching (3). Through these steps, a photonic crystal filter 11 with a hole pattern is completed. However, according to this process, as explained in Figure 5, the hole formation rate differs depending on the position, so even when the same pattern set is formed, the wavelength dependence of the transmittance will differ for each pattern set.

[0030] Figure 8 is a side cross-sectional view illustrating the process for manufacturing a photonic crystal filter 11 according to Embodiment 1. A film for forming a hole pattern (in this case, a SiN film) is deposited on a substrate such as a Si substrate (1). A resist pattern is formed on the SiN film (2), and the SiN film is patterned by etching (3). A transparent substrate is attached to the side of the SiN film opposite to the substrate (the top surface in Figure 8) by, for example, surface plasma treatment (4). The substrate is ground from the back side (the side opposite to the transparent substrate) to the extent that it does not reach the bottom surface of the hole pattern (5). Finally, the substrate is removed from the back side by etching (e.g., wet etching) (6).

[0031] In the above process, since there are no holes created by the etching process at the interface between the transparent substrate and the SiN film (the upper surface of the SiN film in Figure 8), the transmittance does not differ from place to place due to differences in hole formation speed, as shown in Figure 7. Furthermore, the interface between the transparent substrate and the SiN film is a surface formed by the process of depositing the SiN film, and since deposition generally provides better in-plane uniformity than etching, this surface has a smaller arithmetic mean roughness than the surface formed by etching. Therefore, although the in-plane thickness distribution of the SiN film still remains, the in-plane uniformity of the hole pattern depth can be improved compared to the conventional technique. As a result, the in-plane uniformity of the wavelength dependence of transmittance can also be improved compared to the conventional technique.

[0032] <Embodiment 1: Summary> In the spectroscopic imaging system according to Embodiment 1, the photonic crystal filter 11 is equipped with a subwavelength structure consisting of a large number of holes, and the dimensions and periods of the hole patterns of the subwavelength structure are different from each other. The arithmetic mean roughness of the interface between the SiN film constituting the hole pattern of the photonic crystal filter 11 and the transparent substrate supporting the SiN film is smaller than when the hole pattern is formed by etching. As a result, the transmittance when light passes through the photonic crystal filter 11 is approximately uniform across each planar position of the photonic crystal filter 11. Therefore, the processing load for correction when reconstructing hyperspectral image data and the capacity required to store data describing wavelength dependence can be suppressed. Furthermore, it is possible to further improve the reconstruction accuracy of hyperspectral image data by correcting the in-plane distribution of transmittance that remains slightly due to the in-plane thickness distribution of the SiN film, but since the in-plane distribution of transmittance is small, it is sufficient to simply fine-tune the existing model, and this can be done with low computational cost.

[0033] <Embodiment 2> Figure 9 is a side cross-sectional view illustrating the process for manufacturing a photonic crystal filter 11 according to Embodiment 2 of the present invention. Compared to the process described in Embodiment 1, steps (4) and (5) are different, while the other steps are the same. In step (4), a transparent substrate is attached to the SiN film with an adhesive such as an organic material. This fills the SiN film with the organic material in the same layer. In step (5), the substrate is removed by grinding from the back side. By doing so, a photonic crystal filter 11 can be manufactured, and similar to Embodiment 1, the depth of the pore pattern is generally uniform across each planar position of the photonic crystal filter 11.

[0034] <Embodiment 3> Figure 10 shows an example in which a SiN film is formed in a tapered shape during the process of manufacturing the photonic crystal filter 11. When etching the SiN film, it is desirable to etch it as vertically as possible, but this does not prevent it from becoming tapered. For example, in step (3) described in Figure 8, the SiN film may be etched into a tapered shape.

[0035] However, as shown in Figure 10, if the pore pattern is etched so that it tapers from the Si substrate towards the opening, the SiN film pattern is prone to collapse. If areas where the pattern has collapsed and areas where it has not coexist within the plane of the photonic crystal filter 11, the wavelength dependence of the transmittance will differ between areas with a lot of collapse and areas with little collapse, which is undesirable.

[0036] Figure 11 is a side cross-sectional view illustrating another problem when the SiN film has a tapered shape. In this example, the photonic crystal filter 11 and the image sensor 12 are bonded together such that the transparent substrate faces the image sensor 12, and the pore pattern in the SiN film is configured to taper from the image sensor 12 towards the opening. In this case, because the transparent substrate is relatively thick, the incident light diffracted by the photonic crystal filter 11 may be incident on adjacent pixels instead of the intended pixels. This can result in an unclear image.

[0037] Figure 12 is a side cross-sectional view illustrating the process for manufacturing a photonic crystal filter 11 according to Embodiment 3 of the present invention. In step (3), when etching the SiN film, the etching is performed so that the hole pattern tapers from the opening toward the Si substrate. The rest is the same as in Embodiment 1. The image sensor 12 is attached to the completed photonic crystal filter 11 so that it is in contact with the SiN film (7). By following the above steps, the problems described in Figures 10 to 11 can be suppressed.

[0038] <Regarding Variations of the Invention> The present invention is not limited to the embodiments described above, and various variations are included. For example, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment, without departing from the spirit of the invention. Furthermore, it is possible to add, delete, or replace parts of the configuration of each embodiment with other configurations.

[0039] The positions, sizes, and ranges of the components shown in the drawings and other illustrations of this specification may not represent their actual positions, sizes, and ranges for the sake of ease of understanding. Therefore, the present invention is not necessarily limited to the positions, sizes, and ranges disclosed in the drawings and other illustrations.

[0040] In the embodiments described above, it was explained that the interface between the SiN film (optical filter) of the photonic crystal filter 11 and the transparent substrate is flatter than when formed by an etching process. Specifically, it is desirable that the arithmetic mean roughness on the transparent substrate side of the interface be about 10 nm or less, and more preferably about 1 nm or less.

[0041] In the embodiments described above, the spectroscopic imaging system needs to be able to detect the wavelength range of the reflected light from the object to be measured 3. Therefore, the image sensor 12 needs to have detection sensitivity to incident light in that wavelength range. Furthermore, the photonic crystal filter 11 needs to be able to modulate its transmittance by its hole pattern when incident light in that wavelength range is incident on it. For example, it can be configured as follows: (a) When the image sensor 12 has detection sensitivity to incident light with a wavelength of 500 nm, the refractive index of the material constituting the photonic crystal filter 11 is 1.5 or more and 2.8 or less at room temperature (e.g., 25°C, the same applies below) and a wavelength of 500 nm; (b) When the image sensor 12 has detection sensitivity to incident light with a wavelength of 1500 nm, the refractive index of the material constituting the photonic crystal filter 11 is 1.9 or more and 3.5 or less at room temperature (e.g., 25°C, the same applies below) and a wavelength of 1500 nm; (c) When the image sensor 12 has detection sensitivity to incident light with a wavelength of 2500 nm, the refractive index of the material constituting the photonic crystal filter 11 is 2.5 or more and 4.5 or less at room temperature (e.g., 25°C, the same applies below) and a wavelength of 2500 nm.

[0042] In the above embodiments, the material constituting the photonic crystal filter 11 can include, for example, any one or more of silicon nitride, titanium oxide, silicon carbide, silicon, germanium, and gallium nitride. Further, the material of the transparent substrate can include, for example, at least one of glass or quartz.

[0043] In Embodiment 2, the organic material constituting the adhesive needs to transmit incident light. For example, when transmitting incident light with a wavelength of 500 nm, the refractive index of the organic material is 1.4 or more and 1.7 or less at 25 °C and a wavelength of 500 nm. Even when the wavelength is longer than this, the same organic material can be used.

[0044] In the above embodiments, it is assumed that, for example, the nine types of hole patterns shown in FIG. 2 all have the same hole depth. However, if each hole pattern has different transmittance characteristics with respect to incident light, any one of the hole depths of each hole pattern may be different from the hole depths of other hole patterns. However, even in this case, it is still desirable that the transmittance characteristics be the same regardless of the planar position between the pattern sets.

[0045] The present invention relates to a spectroscopic imaging system, and by using these, it can be applied when classifying the types of objects and substances or specifying the amount of contained moisture in a factory production line, a farm, a forest, etc.

[0046] 1: Camera 11: Photonic crystal filter 12: Image sensor 13: Readout circuit 14: Casing 21: Filter 22: Lens 23: Cable 24: Arithmetic unit 25: Light source 3: Measurement object

Claims

1. A spectral imaging system for detecting incident light by spectrally analyzing it, comprising: a plurality of optical filters having wavelength-dependent light transmittances that differ from each other; an image sensor for detecting the intensity of incident light transmitted through the optical filters; and a computing device for calculating the spectrum of the incident light based on the intensity, wherein the optical filters are composed of through-holes smaller than the pixel size of the image sensor; the optical filters are supported by a transparent material that transmits the incident light; and the arithmetic mean roughness on the transparent material side of the interface between the optical filters and the transparent material is 10 nm or less.

2. The spectroscopic imaging system according to claim 1, characterized in that the arithmetic mean roughness of the interface on the transparent material side is 1 nm or less.

3. The spectral imaging system according to claim 1, characterized in that the refractive index of the material constituting the optical filter is 1.5 or more and 2.8 or less at 25°C and a wavelength of 500 nm.

4. The spectral imaging system according to claim 3, characterized in that the image sensor is sensitive to the incident light with a wavelength of 500 nm.

5. The spectral imaging system according to claim 1, characterized in that the refractive index of the material constituting the optical filter is 1.9 or more and 3.5 or less at 25°C and a wavelength of 1500 nm.

6. The spectral imaging system according to claim 5, characterized in that the image sensor is sensitive to the incident light with a wavelength of 1500 nm.

7. The spectral imaging system according to claim 1, characterized in that the refractive index of the material constituting the optical filter is 2.5 or more and 4.5 or less at 25°C and a wavelength of 2500 nm.

8. The spectral imaging system according to claim 7, characterized in that the image sensor is sensitive to incident light with a wavelength of 2500 nm.

9. The spectroscopic imaging system according to claim 1, characterized in that the material constituting the optical filter includes one or more of the following: silicon nitride, titanium oxide, silicon carbide, silicon, germanium, and gallium nitride.

10. The spectroscopic imaging system according to claim 1, characterized in that the transparent material includes at least one of glass or quartz.

11. The spectroscopic imaging system according to claim 1, characterized in that the transparent material is an inorganic material that transmits the incident light, the optical filter and the inorganic material are joined to each other by an organic material that transmits the incident light, and the refractive index of the organic material is 1.4 or more and 1.7 or less at 25°C and a wavelength of 500 nm.

12. The spectral imaging system according to claim 1, characterized in that the aperture size of the through hole on the side farther from the image sensor is greater than or equal to the aperture size on the side closer to the image sensor.

13. A method for manufacturing a spectral imaging system that spectrally analyzes and detects incident light, comprising the steps of: forming a plurality of optical filters having wavelength dependences of light transmittance that differ from each other; bonding the optical filters to an image sensor that detects the intensity of incident light transmitted through the optical filters, wherein in the step of forming the optical filters, the optical filters are formed by forming through holes smaller than the pixel size of the image sensor; in the step of forming the optical filters, the through holes are formed by dry etching after the material of the optical filters is formed on a temporary substrate; in the step of forming the optical filters, the optical filters are bonded to a transparent material, and then the temporary substrate is removed, and the arithmetic mean roughness on the transparent material side of the interface between the optical filters and the transparent material is 10 nm or less.

Images