PIXEL SEMICONDUCTOR COMBINING DIFFRACTION GRID AND INTERFEROMETRIC FILTER, IMAGE SENSOR

Abstract

An image sensor consists of backlit pixels (110) formed in and on a semiconductor substrate (101). Near-infrared radiation reaches the backlit photodetector (112) of the pixels via an upper illuminated surface (102) of the photodetector. A hybrid structure combining a diffraction grating (114) at this upper surface and an interferometric filter with non-regular layers (115) deposited on the diffraction grating traps the photons in the silicon of the photodetector. The radiation is diffracted in the photodetector at a sufficient angle so that, after reflection from trenches (113) isolating adjacent photodetectors and from the metallic layer (111) deposited on the opposite surface of the backlit photodetector, it is reflected by the interferometric filter. The photons are trapped in the substrate; their absorption, and therefore the quantum efficiency of the backlit pixel, is improved.Figure for the abbreviation: Figure 2.

Description

Title of the invention: PIXEL SEMICONDUCTOR COMBINING DIFFRACTION NETWORK AND FILTER INTERFEROMETRIC IMAGE SENSOR technical field

[0001] Embodiments and implementations relate to the field of image sensors, typically of cameras operating in the near-infrared.

[0002] TECHNICAL CONTEXT

[0003] A photographic or image sensor, of the CMOS type (for "complementary metal-oxide-semiconductor"), classically comprises a plurality of pixels, for example arranged in a matrix according to rows and columns, integrated in and on a semiconductor substrate.

[0004] Each pixel conventionally comprises a photosensitive area or "photosite"—for example, a photodiode—formed in the semiconductor substrate and acting as a photodetector, that is, absorbing incident electromagnetic radiation and converting it into an electrical signal. The electrical signals from the pixels are processed and assembled to form an image.

[0005] The pixel photodetectors are separated laterally from each other, electrically and / or optically, by trenches or isolation walls made in the substrate. An electrode (or more generally metallic layers) is usually located on one face of the photodetector to connect the photodetector to the decoding circuits.

[0006] It is known that the silicon composing the photodetector exhibits excellent absorption of radiation in the visible range (380 to 780 nanometers - nm). However, this absorption weakens considerably in the near-infrared (or NIR, for "near-infrared", around 850-1000 nm) range, and more specifically around 940 nm, where many applications are found, notably distance detectors such as FacelD, LIDAR, etc.

[0007] To avoid detecting unwanted wavelengths (e.g., visible light), current technologies use NIR filters in the module, generally deposited on the module's lenses, which focus the incident electromagnetic radiation onto the pixels. These NIR filter deposition operations on the module's lenses are complex and represent significant design costs.

[0008] Furthermore, the low quantum efficiency of silicon in the NIR range impairs the quality of the images acquired by the sensor.

[0009] There is a need to improve this situation.

[0010] SUMMARY

[0011] To this end, a semiconductor pixel is proposed comprising: a substrate, a backlit photodetector in the substrate between a first illuminated surface of the substrate and a second surface of the substrate in contact with a metallic layer and opposite the first illuminated surface, a diffraction structure on the back-illuminated photodetector, formed at the first illuminated surface of the substrate, and an interferometric filter deposited on the diffraction structure.

[0012] A back-side illuminated (BSI) sensor is a sensor in which light rays from the scene to be imaged reach the substrate from the side opposite an interconnect stack comprising interconnecting elements (such as electrodes and metallic layers) of the sensor pixels. The "illuminated" surface of a pixel therefore refers to the side through which the light rays from the scene to be imaged arrive.

[0013] An "interferometric filter" is defined as a stack of alternating dielectric (and / or metallic) thin layers with different refractive indices. Such a filter generally allows for the filtering of a desired band of wavelengths. Furthermore, the interferometric filter is extremely sensitive to the angle of incidence of the light radiation, as the filter's transmission coefficient decreases rapidly with the angle and thus shifts towards shorter wavelengths.

[0014] Also, by providing such an interferometric filter directly on the diffraction structure of the backlit photodetector, the filtered incident radiation is deflected by the diffraction structure into the photodetector substrate. Despite reflections within the substrate (notably by the metallic layer on the second surface) and diffraction on the metallic layer(s) opposite the illuminated face, the radiation, or a significant portion thereof (depending on the design of the diffraction structure), retains an angle of incidence when it returns to the interferometric filter, preventing back transmission. The photons are thus trapped within the substrate of the backlit photodetector. Their absorption, and therefore the quantum efficiency QE of the backlit pixel, is substantially improved.

[0015] Furthermore, by directly combining the interferometric filter and the diffraction structure, a compact hybrid photon trapping structure is obtained.

[0016] According to another aspect, a method for manufacturing a semiconductor pixel is proposed, comprising the following steps: to form a backlit photodetector in a substrate, between a first illuminated surface of the substrate and a second surface of the substrate in contact with a metallic layer and opposite the first illuminated surface, to form a diffraction structure on the backlit photodetector, at the level of the first illuminated surface of the substrate, and forming, by successive deposition of thin layers, an interferometric filter on the diffraction structure.

[0017] An image sensor comprising a plurality of semiconductor pixels as defined in this disclosure is also proposed.

[0018] Optional embodiment characteristics are defined below with reference to the device, while they can be transposed into process characteristics.

[0019] In one embodiment, the interferometric filter is formed by a stack of irregular layers. This irregularity in the dielectric layers (particularly irregularity in thickness) allows the angular transmission selectivity of the filter to be adjusted within a desired wavelength band. This arrangement makes it possible, in particular, to adapt the interferometric filter to the angular aperture of the sensor's main lens to illuminate all the pixels.

[0020] In one embodiment, the substrate is made of silicon. Silicon advantageously has a high refractive index at 940 nm (3.6 versus 1 for air), ensuring total reflection of photons beyond a certain relatively small limiting angle. Therefore, photon trapping is improved.

[0021] In one embodiment, the diffraction structure is designed to diffract incident radiation of a given wavelength at a diffracted angle greater than a minimum target reflection angle to the normal of the interferometric filter for the given wavelength at the interface with the diffraction structure. This minimum angle can typically be the limiting angle defined at the interface between the substrate and the first layer of the interferometric filter. Such a configuration further increases the trapping of photons in the substrate. The "target reflection" can be aimed at a low or even near-zero threshold rate (such as 1% or even a few percent) to achieve near-total reflection beyond said minimum angle.

[0022] In one embodiment, the diffraction structure comprises a diffraction grating fabricated on the first illuminated surface of the back-illuminated photodetector. For example, the diffraction grating may have a pitch (or period) between 400 and 500 nm, so as to diffract NIR radiation at an angle of approximately 30°, suitable for strong (or even total) reflection by the interferometric filter. By way of example, with such a pitch in a silicon substrate, a single order is diffracted at 940 nm, between 30° and 45°.

[0023] In one embodiment, the substrate comprises one (or more) isolation trenches around the backlit photodetector, for example made of SiO2, to optically isolate it from one or more backlit photodetectors of adjacent pixels. The presence of one or more isolation trenches at the edge of the backlit photodetector limits crosstalk between neighboring pixels, and thus enhances photon trapping in each pixel. The quantum efficiency (QE) of the pixel is thereby improved. BRIEF DESCRIPTION OF THE FIGURES

[0024] Other advantages and features of the invention will become apparent upon examination of the detailed description of embodiment and implementation, which is by no means limiting, and the accompanying drawings in which:

[0025] [Fig.l] ;

[0026] [Fig.2] ;

[0027] [Fig.3] ;

[0028] [Fig.4A] ;

[0029] [Fig.4B];

[0030] [Fig.5A] ;

[0031] [Fig.5B] ;

[0032] [Fig.6A] ;

[0033] [Fig.6B] ; and

[0034] [Fig.7] schematically illustrate methods of implementation and realization of the invention. DETAILED DESCRIPTION

[0035] The same elements are designated by the same reference numerals in the different figures. In particular, structural and / or functional elements common to the different embodiments may have the same reference numerals and may have identical structural, dimensional and material properties.

[0036] For the sake of clarity, only the steps and elements necessary for understanding the described embodiments are shown and detailed. In particular, the electronic circuits for decoding and controlling the described image sensors are not detailed, as the described embodiments are compatible with common implementations of these components. Furthermore, the module comprising primary lenses focusing the radiation onto subgroups of pixels is not necessarily shown, but may conform to common implementations, including, for example, lenses with a numerical aperture of F#2. In addition, the applications of the described image sensors are not detailed.

[0037] In the following description, when reference is made to absolute position qualifiers, such as the terms "front", "back", "top", "bottom", "left", "right", etc., or relative position qualifiers, such as the terms "above", "below", "superior", "inferior", etc., or to orientation qualifiers, such as the terms "horizontal", "vertical", etc., reference is made, unless otherwise specified, to the orientation of the figures.

[0038] Unless otherwise specified, the expressions "approximately", "roughly", and "in the order of" mean within 10%, preferably within 5%.

[0039] According to one aspect of the described embodiments, an image sensor consisting of backlit pixels formed in and on a semiconductor substrate is provided. Near-infrared radiation reaches the backlit photodetector of the pixels via an illuminated upper surface of the photodetector. A hybrid structure combining a diffraction grating at this upper surface and an interferometric filter with non-regular layers deposited on the diffraction grating traps the photons in the silicon of the photodetector. The radiation is diffracted in the photodetector at a sufficient angle so that, after reflection from trenches isolating adjacent photodetectors and from the metallic layer deposited on the opposite surface of the backlit photodetector, it is reflected by the interferometric filter. The photons are trapped in the substrate; their absorption, and therefore the quantum efficiency of the backlit pixel, is improved.

[0040] Figures 1 and 2 illustrate, in perspective and cross-section, an image sensor 100 formed in and on a semiconductor substrate 101, for example, silicon, for example, monocrystalline silicon. However, the embodiments described are not limited to this particular example. As an alternative, the sensor 100 can be made from a substrate of a type III-V semiconductor material, a quantum film, or any known photosensitive material, organic or inorganic.

[0041] The sensor comprises a plurality of pixels (or sensor pixels) 110, for example arranged in a matrix according to rows and columns. The pixels of the sensor are, for example, all identical, within manufacturing variations, or similar.

[0042] The sensor 100 is a back-illuminated or rear-side illumination (BSI) sensor: the light radiation from the scene to be imaged reaches the substrate via its rear side, that is, its side opposite an interconnect stack comprising metallic layers interconnecting the sensor pixels to pixel decoders. Back-illuminated sensors are contrasted with front-side illumination (FSI) sensors.

[0043] Each backlit pixel 110 comprises a backlit photodetector 112 formed in the substrate 101 between a first surface 102 of the substrate, called the "illuminated" surface, and a second surface 103 of the substrate opposite the first surface, called "Interconnection" surface insofar as it is in contact with the interconnection stack and therefore a metallic layer.

[0044] Figure 2 illustrates backlit pixels in different manufacturing states. for illustrative purposes. In practice, several pixels are manufactured in parallel on the same substrate 101, so that all pixels are in the same manufacturing state at any given time. The manufacturing steps of the sensor 100 are described below with reference to [Fig. 7].

[0045] To simplify the illustrations, the entire area 112 of the substrate of a backlit pixel 110 is represented as corresponding to the backlit photodetector. However, in practice, the active area of ​​the backlit photodetector 112 may represent only a portion of the substrate area of ​​the backlit pixel.

[0046] The back-illuminated photodetector 112 is typically a photodiode comprising, in the semiconductor substrate 101 (of thickness 6 pm for example), a P-doped semiconductor region at least partially surrounding an N-doped semiconductor region, thus forming a PN junction. The N-doped region extends from the illuminated surface, here upper, 102 of the substrate 110 to a depth (for example 2.5 pm) in the P-doped region.

[0047] As illustrated, the backlit photodetector 112 is substantially in the shape of a block.

[0048] Each pixel comprises, on the interconnection surface 103, one or more electrodes 111 connected to control and decoding circuits 120 for acquiring a signal from the pixel. Generally, a metallic layer is deposited on the interconnection surface 103 of the substrate.

[0049] In the example shown, isolation trenches or walls 113 (or CDTI for "Capacitor Deep Trench Isolation"), extending vertically into the substrate 101, laterally separate the back-illuminated photodetectors 112 from each other, electrically and / or optically, to limit crosstalk between adjacent pixels. Typically, deep isolation trenches can be formed in a substrate by etching the substrate, followed by the deposition of one or more dielectric materials (such as silicon dioxide SiO2). Excess dielectric material can be removed using appropriate techniques, such as chemical-mechanical planarization.

[0050] In this example, a backlit pixel 110 includes a diffraction structure 114, having in particular a periodic or regular structuring pattern - such as a diffraction grating, on the backlit photodetector, formed at the illuminated surface 102. The diffraction structure 114 can consist of bumps formed by dry etching of the substrate 101 at the first surface 102. Alternatively, pyramids can be formed using wet etching.

[0051] A person skilled in the art is aware of simple grating diffraction formulas enabling them to determine the dimensions of this grating (in particular the width and height of the dots as well as their pitch or period) as a function of a given wavelength (or band) and a target diffraction angle.

[0052] By way of illustration, for an application in the near-infrared (NIR) range, a single-order light at 940 nm is diffracted, at a diffraction angle between 30° and 45°, by a diffraction grating fabricated in a silicon substrate with a pitch of 400 to 500 nm. The diffraction grating may have pads approximately 200 nm wide and between 200 and 400 nm high. These values ​​are, of course, only examples. Furthermore, the wavelength of 940 nm is given as a preference. However, other wavelengths in the NIR, typically between 850 nm and 1000 nm, can be considered.

[0053] In the illustrated example, the backlit pixel 110 finally includes an interferometric filter 115 deposited on the diffraction structure 114. The filter is intended to filter a narrow band of NIR, for example around 940 nm. The interferometric filter 115 can form a structure with a thickness between 1 and 10 pm, although smaller or larger thicknesses are also possible.

[0054] The interferometric filter 115 consists of a stack of alternating thin films with different refractive indices. Typically, two different dielectric materials are alternated. For example, the number of thin films can be between 2 and 100, for example between 5 and 50.

[0055] By way of illustration, the following pairs of materials may be used: SiO2 and TA2O5 SiO2 and a-SI (amorphous silicon), SiO2 / Nb2O5> SiO2 / SiN.

[0056] Preferably, the interferometric filter is formed from a stack of irregular layers in that the thicknesses from one layer to the next are irregular and do not follow a regular pattern. The irregularity of the layers makes it possible to adjust the range of incidence angles effectively transmitted by the filter, in particular to meet varying angular aperture requirements for illuminating all the pixels.

[0057] The design of the interferometric filter 115 can be computer-aided (typically using thin-film software such as Optilayer or OTF Studio, trade names) taking into account the wavelength used (e.g., 940 nm) and a target bandwidth at half maximum of the transmission diagram (e.g., 23 nm). The design seeks, in particular, to maximize transmission over a band of wavelengths around the wavelength The desired range of incidence angles, for example 0° to 14° or 0° to 30°, is specified. Indeed, the interferometric filter is sensitive to the angle of incidence of the light radiation; the filter's transmission coefficient decreases rapidly with the angle and shifts towards shorter wavelengths. Therefore, the range of transmitted incidence angles and the target full width at half maximum (FWHM) are closely related: the higher the FWHM, the greater the range of incidence angles (α) that guarantee transmission through the filter. A minimum incidence angle for (near-)total reflection can thus be defined, beyond which the transmission of radiation at the target wavelength is considered too low (compared to a threshold, for example).The minimum angle of incidence of (quasi-)total reflection on the filter-air interface (therefore on the surface 115A for the outward path corresponding to light radiation from the scene to be imaged) is noted a0. .

[0058] Symmetrically, we can denote ai the minimum angle of incidence of (quasi-)total reflection on the substrate-filter interface (therefore on the surface 115B for the return path corresponding to the light radiation reflected by the substrate) since the interferometric filter 115 is deposited on the substrate 101.

[0059] The design of the interferometric filter 115 can take into account the angles a0 and ai in addition to the full width at half maximum.

[0060] Fig. 3 illustrates, for example, an image sensor arrangement 100 in which the radiation is focused on all or part of the pixel matrix 110 using a main lens 30, e.g., of numerical aperture F#2. This lens illuminates the center pixels with a maximum angle of incidence of 14° (a=14° with respect to the normal A), and the outermost pixels with a maximum angle of incidence between 25 and 30°, for example 27° (a=27°).

[0061] Figures 4A and 4B show transmission diagrams of an example of an interferometric filter deposited on a silicon substrate. The interferometric filter consists of non-regular SiO2 and TA2O5 layers, designed by computer for a wavelength band around 940 nm, an illumination cone of F#2, and an angle a0 of 14°. Figure 4A shows the transmission with an incidence cone of F#2, thus corresponding approximately to the forward direction (light rays from the scene), and Figure 4B shows the transmission without an incidence cone, thus corresponding approximately to the reverse direction (rays from the substrate 101).

[0062] The interferometric filter 115 in this example is formed of 35 alternating non-regular thin layers of SiO2 and Ta2O5, for a total thickness of approximately 6 pm. The diagrams are obtained by simulating NIR radiation transmission (collimated light) in this filter as a function of different angles of incidence a, using the software Lumerical (trade name).

[0063] Figure 4A shows that the interferometric filter is a bandpass filter with a full bandwidth at half maximum (FWHM) of 23 nm. It also exhibits good transmission tolerance (90% or more) at least up to an angle of 15°, and approximately up to 17-18°. Therefore, such a filter appears suitable for an image sensor configuration where the maximum angle of incidence of the radiation α (e.g., from the primary lens F#2) is limited to 17-18°. In this case, the minimum angle of incidence for (quasi-)total reflection for the forward path is α0 = 17°.

[0064] Figure 4B shows that up to an angle of incidence of 16°, the radiation is partially transmitted by the interferometric filter. Beyond 16°, transmission is almost zero. In this case, the minimum angle of incidence for (quasi-)total reflection on the return path is ai = 16°. The (quasi-)total reflection in the return direction is due in particular to the properties of silicon which, with a high refractive index, ensures total reflection beyond a certain angle: photons that strike the lower surface 115B of the interferometric filter 115 at an angle greater than ai are all reflected and remain in the photodetector 112.

[0065] This is the phenomenon illustrated by the arrows in pixel 11Od of [Fig. 2]. When the incident NIR radiation is transmitted through the interferometric filter 115 to the photodetector 112, it is deflected by an angle, e.g. 30°, greater than ai by the diffraction structure 114. Given the geometry of the photodetector (typically a cuboid), the lateral isolation trenches 113 and the metallic layer 111 on the lower surface of pixel 110 ensure reflection of the photons in the photodetector 112, which return to strike the lower surface 115B of the filter 115 at the same large angle (approximately 30° in the example) and are therefore totally reflected back to the silicon of the photodetector 112.

[0066] In other words, the hybrid structure combining the diffraction grating 114 and the interferometric filter 115 forms an asymmetric structure that prevents the return (from the substrate 101) of transmitted NIR radiation, and thus traps photons in the silicon of the photodetector 112, resulting in improved absorption. The quantum efficiency QE of the pixel is thereby improved.

[0067] Figure 5A illustrates the quantum efficiency QE of a backlit pixel (with a pitch of 3.2 pm) equipped with the interferometric filter 115 of Figures 4A and 4B and a diffraction grating 114 with a pitch of 500 nm. The curves were obtained using the aforementioned Lumerical simulation software, with the backlit pixels illuminated at 940 nm by an illumination cone F#2. Curve A represents the QE of pixel 110 with the interferometric filter 115, and curve B represents the QE of the pixel without the interferometric filter. 115, and curve C the QE of the pixel without the interferometric filter 115 but with a conventional microlens positioned opposite the pixel (according to classical techniques). Indeed, the pixel 110 used in the NIR can be without an opposite microlens, contrary to the techniques usually accepted in the visible range.

[0068] A gain of 10 points in QE is observed at 940 nm, allowing a substantial improvement in the quality of the acquired signal, and therefore of the image acquired by the image sensor 100.

[0069] This 10-point gain corresponds to a halved reflection at 940 nm, as shown in [Fig. 5B], which illustrates the reflection of the backlit pixel on radiation from the scene—that is, by surface 115A of the interferometric filter 115. Again, curve A represents the reflection with the interferometric filter 115 present, curve B the reflection without the interferometric filter 115, and curve C the reflection without the interferometric filter 115 but with a microlens positioned opposite the pixel. The curve in [Fig. 5B] is essentially complementary to the curve in [Fig. 5A].

[0070] The interferometric filter 115 used for the simulations in Figures 4A, 4B, 5A and 5B offers good transmission at 940 nm for angles of incidence up to a0=17-18°. This may prove insufficient for image sensor arrangements implementing higher angles of incidence.

[0071] Other interferometric filters can be designed using, for example, Optilayer software to obtain a larger minimum incidence angle of (quasi-)total reflection a0.

[0072] Figures [Fig.6A] and [Fig.6B] represent the quantum efficiency QE of a backlit pixel (pitch 3.2 pm) equipped with another example of an interferometric filter illuminated by an F#2 cone.

[0073] This other interferometric filter was designed under 940 nm conditions, using SiO2 and TA2O5 materials, with a bandwidth greater than that of the previous figures to achieve transmission at 940 nm even at 30° of incidence. The resulting interferometric filter consists of 28 alternating non-regular thin layers of SiO2 and TA2O5, for a total thickness of approximately 4.3 pm.

[0074] Fig. 6A illustrates the quantum efficiency QE of pixel 110 in the presence of the interferometric filter 115 (curve A) as opposed to a pixel without the interferometric filter 115 (curve B) and a pixel without the interferometric filter 115 but with a microlens positioned opposite the pixel (curve C).

[0075] A widening of the half-width at half-height is observed at 60 nm, while maintaining a 7-point gain in QE at 940 nm, still allowing for improvement substantially the quality of the acquired signal, and therefore of the image acquired by the image sensor 100.

[0076] Fig. 6B illustrates the quantum efficiency QE of pixel 110 as a function of three angles of incidence of radiation from the scene on the interferometric filter 115, namely 20°, 24° and 28°.

[0077] It is observed that the quantum efficiency curve QE shifts towards lower wavelengths as the angle of incidence increases. It is therefore possible to widen the half-height bandwidth to maintain a high QE over a wider range of angles of incidence.

[0078] In the example given, the QE efficiency remains greater than 0.6 up to an angle of 28°. This interferometric filter 115 is suitable for a back-illuminated image sensor arrangement where the chief ray angle (CRA) is less than 28°.

[0079] By widening the half-width, it is thus possible to obtain an interferometric filter capable of transmitting incident radiation from the scene regardless of the angular opening of the illumination cone of the pixels, while continuing to trap photons in the pixels.

[0080] Fig. 7 illustrates, using a flowchart, manufacturing steps of an image sensor with backlit semiconductor pixels.

[0081] In step 700, the hybrid structure {diffraction grating 114, asymmetric interferometric filter 115} is designed for the desired wavelength, typically 940 nm, and desired angles of incidence. This design can be computer-aided.

[0082] Typically, thin-film software is used to design an interferometric filter 115 satisfying constraints such as a band of wavelengths to be filtered, the angles of incidence transmitted by the filter (e.g., a transmission at 940 nm greater than a threshold of up to 30°) and / or a full-width filter.

[0083] A simulation of the return transmission makes it possible to obtain the minimum angle of incidence of (quasi-)total reflection ah typically the limiting angle of the substrate-filter interface.

[0084] The diffraction grating 114 (in particular its pitch) is then determined so that it diffracts an incident light with an angle greater than ai. In other words, the diffraction grating 114 is designed to diffract an incident radiation (at 940 nm for example) with a diffracted angle which is greater than the minimum (quasi-)total reflection angle (to the normal) of the interferometric filter at the interface with the diffraction grating (or substrate).

[0085] In step 702, a P-doped semiconductor substrate 101 is obtained on which photodiodes 110a forming pixels are fabricated. In a known manner, trenches Insulating electrodes 113 separate the photodiodes from each other, typically in a row-column arrangement. The substrate, at each pixel, is locally N-doped in the already P-doped region to form a PN junction. An electrode 111 is positioned at the rear of the pixel to form a backlit pixel 110a.

[0086] In step 704, the diffraction grating 114 is formed on the upper - therefore illuminated - surface 102 of the photodiode 100b, according to the parameters determined in step 700. Etching (by dry or wet method) is typically used.

[0087] In step 706, the interferometric filter 115 is formed above the diffraction grating 114 by the successive deposition of non-regular and alternating thin films with different refractive indices. Preferably, each thin film is deposited on all the photodiodes 100c formed on the substrate 101, in order to create the photodiodes in parallel.

[0088] In step 708, the photodiode array is cut and integrated with an optical module and connected to control and decoding circuits, to form the backlit image sensor 100.

[0089] Of course, this disclosure is not limited to the embodiments described above by way of example; it extends to other variations. Other embodiments are possible.

Claims

Demands

1. Semiconductor pixel (110) comprising: a substrate (101), a backlit photodetector (112) in the substrate between a first illuminated surface (102) of the substrate and a second surface (103) of the substrate in contact with a metallic layer and opposite to the first illuminated surface, a diffraction structure (114) on the backlit photodetector, formed at the level of the first illuminated surface of the substrate, and an interferometric filter (115) deposited on the diffraction structure.

2. Semiconductor pixel (110) according to claim 1, wherein the interferometric filter (115) is formed of a stack of non-regular layers.

3. Semiconductor pixel (110) according to claim 1 or 2, wherein the substrate is formed of silicon.

4. Semiconductor pixel (110) according to any one of claims 1 to 3, wherein the diffraction structure (114) is designed to diffract incident radiation of a given wavelength with a diffracted angle that is greater than a minimum target reflection angle (aj, to the normal) of the interferometric filter (115) for the given wavelength at the interface with the diffraction structure.

5. Semiconductor pixel (110) according to any one of claims 1 to 4, wherein the diffraction structure (114) comprises a diffraction grating made on the first illuminated surface (102) of the backlit photodetector (112).

6. Semiconductor pixel (110) according to claim 5, wherein the diffraction grating (114) has a pitch between 400 and 500

7. 11111. Semiconductor pixel (110) according to any one of claims 1 to 6, wherein the substrate (101) comprises an isolation trench (113) around the backlit photodetector (112) to optically isolate it from one or more backlit photodetectors of adjacent pixels.

8. Image sensor (100) comprising a plurality of semiconductor pixels (110) according to any one of the preceding claims.

9. Method of manufacturing a semiconductor pixel (110), comprising the following steps: forming (702) a backlit photodetector (112) in a substrate (101), between a first illuminated surface (102) of the substrate and a second surface (103) of the substrate in contact with a metallic layer and opposite to the first illuminated surface, forming (704) a diffraction structure (114) on the backlit photodetector at the level of the first illuminated surface of the substrate, and forming (706), by successive depositions of thin films, an interferometric filter (115) on the diffraction structure.