High dynamic range photodetector and photoelectric image sensor
The photodetector element with a control device and computing unit corrects transition inconsistencies between linear and logarithmic modes, addressing noise and sensitivity issues to achieve a high dynamic range and low noise level for reliable image capture in neuro-mimetic recognition systems.
Patent Information
- Authority / Receiving Office
- FR · FR
- Patent Type
- Patents
- Current Assignee / Owner
- GPIXEL CHANGCHUN MICROELECTRONICS INC
- Filing Date
- 2023-10-27
- Publication Date
- 2026-06-19
Smart Images
Figure 00000022_0000 
Figure 00000022_0001 
Figure 00000022_0002
Abstract
Description
Title of the invention: High dynamic range photodetector and photoelectric image sensor. Field of the invention
[0001] The present invention relates in general to photoelectric image sensors with pixel matrix.
[0002] Background of the invention
[0003] Image sensors, historically developed for shooting (photography, cinematography, etc.), are becoming essential components in intelligent systems such as autonomous cars, mobile robots, etc.
[0004] In such applications, the image captured by an image sensor is analyzed by software often based on neural networks. The reliability of these analyses depends largely on the quality and properties of the image provided by the sensor. In this case, the desired image quality and properties are very different from those desired for photography or cinematography, where interpretation is performed by the human vision system.
[0005] Fundamentally, for neuro-mimetic recognition systems, small variations in the quality and characteristics of the image capture are essential. Indeed, this recognition process is essentially a comparison between the input image and the images "seen" by the same system previously.
[0006] It is well known that an image sensor with a logarithmic response can contribute to such small variations, as evidenced by all vision systems of living beings. However, realizing a logarithmic response image sensor capable of operating with performance similar to that of vision systems of living beings is difficult.
[0007] The realization of a logarithmic sensor is necessarily based on a physical phenomenon which is in itself exponential, because a simple mathematical logarithmic conversion of an image taken conventionally with a linear response does not allow to cover the desired operating dynamics for a logarithmic sensor.
[0008] Fig. 1 illustrates a logarithmic response photodetector based on a non-linear conversion element.
[0009] The simplest implementation of such a photodetector is based on a non-linear resistor that transforms a photoelectric current in a photodiode into a voltage with a logarithmic response. This non-linear resistor can be either a A PN diode, or a transistor connected as a diode. Despite its simplicity, this photodetector structure has many critical flaws, in particular: - a large disparity in response from one photodetector to another, generating very strong, fixed spatial noise, - a loss of sensitivity in low light due to leakage current in the non-linear resistance, - a strong variability in response with temperature variations, and - a strong persistence at low light levels.
[0010] That is why such a basic approach has not had industrial success.
[0011] A second approach is based on the use of a photodiode in mode In a solar cell, the open-circuit voltage across a photodiode is used as the image signal. This solution significantly reduces fixed spatial noise. However, an anti-blooming structure is necessary to maintain image resolution. Under these conditions, photocharge collection is greatly reduced. A reset operation is required to compensate for pixel response dispersion and eliminate afterglow. The sensitivity of this type of photodetector is very low compared to that of conventional photodetectors. Patent EP1354360B1 describes such a solution.
[0012] A third approach consists of creating a photoreceptor that initially operates in charge accumulation mode (linear mode) and then has its response evolve towards a logarithmic mode. Operating the photodetector initially in accumulation mode allows the photodiode to be read in charge mode. A readout in charge mode enables a double correlated readout, thus eliminating switching noise during the readout. All modern conventional photodetectors use this type of operation. In a charge-transfer photodetector, switching noise can be almost completely eliminated, the only residual noise being the basic noise in the readout transistor.For example, the read noise in a modern charge-transfer photodetector is typically a few electrons (the most efficient can reach % electrons), while the read noise in a logarithmic response photodetector is a few tens or even hundreds of electrons. Documents US9413991B2, US9843750B2, and US20150281607A1 provide examples of photodetectors with mixed linear / logarithmic response.
[0013] This mixed linear / logarithmic solution is very interesting in that it allows a considerable improvement in sensitivity at low light levels and a conservation of logarithmic operation in strong light.
[0014] But a major drawback is that the transition point between the linear mode and the logarithmic mode is very sensitive to certain physical parameters of components; in particular, it changes significantly depending on the temperature.
[0015] Numerous attempts have been made to correct this non-uniformity, but none have been satisfactory. Summary of the invention
[0016] The present invention aims to provide a solution that makes a LIN-LOG response photodetector free of fixed spatial noise while retaining the advantages in low light level and in high light level, thus obtaining a photodetector, and more generally a pixel matrix sensor, which exhibits a high dynamic range and a low noise level, this type of sensor being particularly expected nowadays especially for intelligent systems based on artificial vision.
[0017] For this purpose, according to a first aspect, a photodetector element is proposed, intended in particular to form a matrix image sensor, comprising: - a buried photodiode exposed to a radiation source, - a diffusion node linked to the photodiode and capable of transforming a photoelectric charge variation in the photodiode into a voltage, - a controlled reset switch capable of applying a voltage to the broadcast node, - a charge transfer switch capable of transferring the charge accumulated in the photodiode to the linked diffusion node, and - a photodiode exposure control device capable of controlling the reset switch and the charge transfer switch to generate a response signal from the photodetector element, this response signal intrinsically having an initially linear phase followed by a logarithmic phase, with a potentially indeterminate transition between the two modes, an element characterized in that the control device is capable of causing two successive exposures of the photodiode with two different operating parameters, and in that it further comprises a computing device receiving as input the two response signals corresponding to the two exposures and capable of generating a unified logarithmic response signal over the entire range of the response from a combination of the two response signals, taking into account on the one hand the determination of a slope (K1; K2) of the logarithmic response of the photodetector and an offset (OS1;OS2) between the two signals. ;
[0018] According to a second aspect, a photodetector element is proposed, intended in particular to form a matrix image sensor, characterized in that it comprises in combination: - a first buried photodiode exposed to a radiation source, - a second buried photodiode exposed to the same radiation source, - at least one diffusion node linked to each photodiode and capable of transforming a photoelectric charge variation in the photodiode into a voltage, - at least one controlled recovery switch capable of applying a voltage to the diffusion node, - two charge transfer switches capable of transferring the charge accumulated in the two photodiodes to the diffusion node, and - a photodiode exposure control device capable of controlling the recovery switch and the charge transfer switch with different operating parameters, to generate two different response signals from the two photodiodes, these response signals intrinsically having an initially linear phase followed by a logarithmic phase, with a potentially indeterminate transition between the two modes, the exposures of the two photodiodes exhibiting at least partial temporal overlap, the photodetector further comprising a computing device receiving as input the two response signals obtained with the two photodiodes respectively and capable of generating a unified logarithmic response signal over the entire response range from a combination of the two response signals, taking into account on the one hand the determination of a slope (K1; K2) of the logarithmic response of the photodetector and an offset (OS1;OS2) between the two signals. ;
[0019] Optional aspects of the two photodetector elements described above include the following additional features, considered individually or in any combinations that a person skilled in the art will understand to be technically compatible with each other:
[0020] * The different operating parameters include exposure times different.
[0021] * the exposure periods defining the two exposure durations end at roughly the same time for each capture.
[0022] * the calculation device is capable of:
[0023] - determine a ratio between the values of the two signals, and
[0024] - depending on the value of this ratio:
[0025] - form an output signal by converting the signal value obtained with the duration of shortest exposure in its logarithm, by applying a pre-established slope coefficient (Kl), or
[0026] - form an output signal by applying a predetermined offset to the signal value obtained with the longest exposure time.
[0027] * The different operating parameters include capacity values of different accumulations at the level of the photodiode(s).
[0028] * the different storage capacity values are fixed and obtained by construction.
[0029] * different storage capacity values are obtained by polarization variable at the level of the load transfer switch(s).
[0030] * the calculation device is capable of:
[0031] - determine a difference between the values of the two signals, and
[0032] - depending on the value of this difference:
[0033] - form an output signal by converting the signal value obtained with capacitance of greatest accumulation in its logarithm, by applying a pre-established slope coefficient (K2), or
[0034] - form an output signal by applying a predetermined offset to the signal value obtained with the smallest accumulation capacity.
[0035] According to another aspect, a pixel matrix photoelectric sensor is proposed, characterized in that each pixel comprises a photodetector element as defined above, and in that it comprises a common digital processing device forming the computing devices respectively associated with the different pixels.
[0036] According to yet another aspect, a photodetector element is proposed, intended in particular to form a matrix image sensor, characterized in that it comprises in combination:
[0037] - a first buried photodiode exposed to a radiation source,
[0038] - a second buried photodiode exposed to the same radiation source,
[0039] - at least one diffusion node linked to each photodiode and capable of transforming a variation of photoelectric charge in the photodiode under voltage,
[0040] - at least one controlled reset switch capable of imposing a voltage the diffusion node,
[0041] - two charge transfer switches capable of transferring the accumulated charge in both photodiodes towards the bound scattering node, respectively, and
[0042] - a photodiode exposure control device capable of controlling the a recovery switch and a charge transfer switch with different operating parameters, to generate two different response signals from the two photodiodes, these response signals intrinsically exhibiting an initially linear phase followed by a logarithmic phase, with a potentially indeterminate transition between the two modes, the exposures of the two photodiodes exhibiting at least partial temporal overlap, such so that the response signals are delivered essentially simultaneously to obtain by processing a fully logarithmic response.
[0043] Finally, a calculation device is proposed that is configured to be associated with a buried photodiode(s) photodetector element capable of generating two different responses in linear / logarithmic mode for the same capture, the calculation device being capable of generating a unified logarithmic response signal over the entire range of the response from a combination of the two response signals, taking into account on the one hand the determination of a slope of the logarithmic response of the photodetector and an offset between the two signals. Brief description of the drawings
[0044] Other aspects, objects, and advantages of the present invention will become more apparent upon reading the following description of preferred embodiments thereof, given by way of non-limiting example and with reference to the accompanying drawings. In the drawings:
[0045] [Fig. 1] illustrates three versions of a photodetector with a logarithmic response of a known type,
[0046] [Fig.2] illustrates in its left part a photodetector with a linear / logarithmic mixed response and in its right part a timing diagram of the signals involved in such a photodetector,
[0047] [Fig.3A] illustrates in its upper part the physical structure of a conventional charge-transfer photodetector and in its lower part the structure of the energy bands in such a photodetector,
[0048] [Fig.3B] is a timing diagram of the signals involved in the photodetector of [Fig.3A],
[0049] [Fig.4A] illustrates the energy bands in two operating modes of such a photodetector,
[0050] [Fig.4B] is a timing diagram of the signals involved in the photodetector of [Fig.4A],
[0051] [Fig.4C] shows the different responses of the photodetector in [Fig.4A] as a function of its operating mode,
[0052] [Fig.5] shows the variation in response of this same photodetector by varying the exposure time and the height of an energy barrier under the grid,
[0053] [Fig.6] shows in its left part the response of this same photodetector with two different exposure times and shows in its right part the evolution of the ratio between the signals respectively obtained,
[0054] [Fig.7] shows in its left part the response of this same photodetector with two levels of maximum potential wells and shows in its right part the evolution of the difference between the signals respectively obtained,
[0055] [Fig.8] shows in its left part the responses obtained with two different exposure durations and in its right part the shape of the corrected response obtained by processing the output signals,
[0056] [Fig.9] shows in its left part the responses obtained with two different maximum potential well levels and in its right part the shape of the corrected response obtained by processing the output signals, and
[0057] [Fig. 10] schematically illustrates two ways of obtaining the two different signals in the photodetector.
[0058] Detailed description of preferred embodiments
[0059] We will first recall the structure and behavior of a linear / logarithmic or LIN / LOG photodetector.
[0060] As mentioned above, the logarithmic response in the photodetector is created by a controlled leakage of the photoelectric charge. For example, in a photodetector with a conversion diode, the charge leakage is achieved by the conduction of the PN junction of the conversion diode. If, in this case, it is possible to pre-charge the signal node voltage to a voltage that initially blocks the conduction of the PN junction, such that there will be no charge leakage at the beginning of the exposure. The photonic charge accumulates and the signal node voltage decreases linearly. The evolution of the signal node voltage SIG is related to the light intensity by a linear law, before the conversion diode unblocks and the evolution of the SIG voltage gradually becomes logarithmic with respect to the light intensity. [Fig.[2] illustrates such a LIN / LOG photodetector where the cathode of a PD photodiode is connected on one side to a reference voltage VREF via a zeroing transistor MRST and on the other side to a voltage VDD via an NMOS LOG conversion transistor.
[0061] It is understood that if the initial voltage of the output signal SIG is taken just after the switching off of the reset transistor MRST, and it is subtracted from the voltage taken at the end of the exposure, the switching noise introduced by the MRST transistor can be eliminated.
[0062] In this basic version of a LIN-LOG photodetector, it can be seen that: a) at low light levels, the photodetector operates in the linear regime and the switching noise of the MRST transistor is compensated by determining the difference between two readings taken at the beginning and end of the exposure, as just indicated; very good sensitivity is obtained here; b) the transition between the LIN-LOG regimes depends not only on the intensity of light but also on the duration of exposure or exposure time: the longer the exposure time, the lower the intensity of light at the transition point; c) The transition point between the two regimes is strongly dependent on both the type of conversion element (a PN diode, a MOSFET junction, etc.) and the perturbation that the MRST transistor introduces at the end of the reset.
[0063] In such a photodetector, good image quality is obtained in the linear regime thanks to double reading and differentiation. However, in the logarithmic regime, image quality is rather poor due to dispersion and instability at the transition point between the two regimes. Furthermore, while sensitivity is improved, the suppression of zeroing noise requires a reading at the beginning of the exposure and another at the end. This necessitates not only image memory, but also the relatively long time between the two readings, which leads to signal degradation due to low-frequency noise in the transistors. Finally, the dark current is very high due to the necessary contact on the photodiode. These considerations indicate that it would be preferable to use a charge-transfer photodetector.
[0064] Figure 3A illustrates the microelectronic structure of a conventional charge-transfer photodetector and, in the lower part, the representation of the energy levels. The photoelectric charge accumulates in a buried photodiode PPD, the surface of whose N-doped region, denoted NPD, is protected by a P+ doped region connected to the substrate SUB. The transfer transistor TX and the reset transistor RST are shown schematically. The PW regions denote P-doped wells.
[0065] This configuration ensures a very low dark current and thus creates the possibility of complete desertion of the N zone. A transfer transistor TX connects this buried photodiode PPD to a diffusion node FD. A reset transistor RST is mounted between a voltage source VDD and the node FD. To read the photoelectric charge of the buried photodiode PPD, the node FD is first reset via the transistor RST. Then the gate of the transistor TX is biased to a high voltage, and a conduction channel is thus formed under this transistor TX to transfer the charge of the photodiode PPD to the node FD.
[0066] The physical parameters of the PPD photodiode and the TX transfer transistor are adjusted so that all of the photon-origin charge is transferred to the FD node at the end of the TX transistor's operation. It is understood that this read operation also discharges the PPD photodiode, which is then ready for another acquisition cycle.
[0067] With particular reference to the timing diagram in [Fig. 3B], two voltage signals are taken, one after the action of transistor RST and the other after the action of transistor TX. The difference between these two voltages gives the output signal. This This operation is commonly called "Correlated Double Sampling (CDS)" in English. Thanks to this differentiation and the very short time interval between the two readings, reset noise can be almost completely eliminated.
[0068] If a charge leakage path is now introduced between the PPD photodiode and a charge drain during exposure, a hybrid LIN-LOG response can be obtained for this charge-transfer photodetector. It is known that charge can naturally overflow from the photodiode during exposure in a charge-transfer photodetector. Particular attention must therefore be paid to the presence of a charge drain to prevent excess charge from propagating to neighboring photodetectors in the case of a matrix sensor; otherwise, image quality will be greatly degraded. For this reason, a charge-transfer photodetector with a LIN-LOG response has a special arrangement for dissipating excess charge in the PPD diode.
[0069] A simple implementation for draining excess photon-induced charge is to lower the energy level barrier under the gate of the TX transistor so that it is lower than that of the substrate. In this case, the excess charge takes this path to the FD node instead of flowing into the substrate and contaminating neighboring pixels. The RST transistor must then remain conducting.
[0070] As illustrated in [Fig.4A], when the accumulated charge level reaches the energy barrier under the TX gate, the charge spills over onto the FD node and is efficiently drained to a voltage source via the RST transistor.
[0071] The operating timing diagram, shown in [Fig. 4B], is very similar to that of a conventional charge-transfer photodetector. The essential difference is that the reset signal RST must be maintained for the duration of the photodetector's exposure time, as illustrated in the timing diagram.
[0072] It is possible to physically demonstrate that the amount of residual charge in the buried PPD photodiode during this overflow is proportional to the logarithm of the light intensity reaching it. The article "QLog Solar-Cell Mode Photodiode Logarithmic CMOS Pixel Using Charge Compression and Readout," Yang Ni, Sensors magazine 2018, 18, 584, provides a detailed explanation of this operation.
[0073] Fig. 4C shows the shape of the LIN-LOG response of the photodetector of Fig. 4A as a function of the photon intensity.
[0074] The LIN-LOG transition point can be adjusted either by the voltage applied to the TX transfer transistor during exposure or by the exposure time. Figure 5 shows, on the left, the evolution of the LIN-LOG response as a function of barrier height and exposure time.
[0075] It is well known that the threshold voltage of a MOS transistor depends strongly on the interface state between the gate and the channel in the substrate. This interface state is unstable and exhibits very high dispersion. Consequently, the transition point of a charge-transfer LIN-LOG response photodetector is highly dispersed from one photodetector to another. An image sensor composed of this type of photodetector for each pixel therefore suffers from very high fixed spatial noise. It should be noted that considerable effort has been devoted to correcting this fixed spatial noise, without, however, obtaining a simple and stable solution.
[0076] If we analyze the response of a photodetector with LIN-LOG response, whatever its concrete realization, this response is composed of 3 zones, as illustrated on [Fig.4C] where the abscissa scale is a logarithmic scale: a first linear zone (LIN), an intermediate zone (X) then a logarithmic zone (LOG).
[0077] As explained, the initial LIN part of the response is very accurate because the voltage conversion is based on counting the number of photon-derived electrons accumulated in the photodiode and is performed on the intrinsic capacitance of the FD node. This process is very stable and relatively accurate. Variations in the capacitance of the floating scattering node FD are easily corrected by a gain, and the dark thermal load can be corrected by a simple offset voltage. In general, a LIN-LOG response photodetector made using CMOS technology requires virtually no correction.
[0078] In the following discussion, it is assumed that the linear response of the LIN-LOG photodetector is accurate, either intrinsically or after correction.
[0079] It can be observed that the LOG portion of the response does indeed follow a logarithmic law, as this process is governed by thermodynamic laws that affect electrons. The dispersion lies in the shift value and the value of the logarithmic slope. The slope of a LOG response is directly related to the absolute temperature and can also be modulated by the readout gain in a particular design. It can be measured and its correction is straightforward, for example, by measuring the response at two different light levels to determine this slope.
[0080] In the following discussion, we will assume that the slope of the LOG response of a LIN-LOG response photodetector is accurate either intrinsically or after a correction.
[0081] The problem lies in the intermediate X region between the LIN and LOG regions, as this transition region is difficult to model and varies with the physical parameters of the components and the operating temperature. The signal from a LIN-LOG photodetector can be located in one of these regions with a completely different fixed spatial noise: very low fixed spatial noise in the LIN region, and very high in the X or LOG region. Therefore, this region cannot be identified by a simple response from this photodetector.
[0082] We will now describe a solution according to the present invention for: - determine the intermediate zone X of a LIN-LOG response photodetector, - correct the offset of the LOG response and obtain a unified logarithmic response over the entire operating range of a LIN-LOG response photodetector.
[0083] It should be noted that these operations are carried out by a digital processing circuit associated with the photodetector and receiving its output signals, or in the case of an implementation in the form of a pixel matrix image sensor, associated with the set of pixels.
[0084] We will first describe the process of determining the intermediate zone X.
[0085] It is known that the transition point between the LIN and LOG responses can be influenced by various physical or functional parameters. For example, a long exposure time advances the transition point, while a short exposure time moves it back. A greater accumulation capacity moves the transition point back, and a smaller accumulation capacity moves it forward.
[0086] It is recalled here that the LIN response is precise and stable in absolute value, that the LOG response is precise and stable in slope, and that between the LIN response and the LOG response, there is an unstable and uneven intermediate response, which is difficult to calibrate.
[0087] In order to correct the dispersion of the LIN-LOG response, it is necessary to know the exact position of the LIN-LOG transition zone, however, if we only look at the value of the signal from a photodetector with a LIN-LOG response, we cannot know this position.
[0088] A first approach to determining the transition zone is to operate the same LIN-LOG response photodetector with two different exposure times T1 and T2. Let R be the ratio between the two exposure times: R = T1 / T2 and T1 and T2 are chosen so that R> 1
[0089] SIG1 and SIG2 are defined as the signals obtained with exposure times T1 and T2, respectively.
[0090] As illustrated in [Fig. 6]: - when the report: RS = SIG1 / SIG2 is equal to R, this means that SIG1 and SIG2 are both in the LIN zone, - when this same ratio is equal to 1, this means that SIG1 and SIG2 are both in the LOG zone. - when it is determined that RS is between 1 and R, this means that SIG1 or SIG2 is in the intermediate zone X.
[0091] A second approach to determining the transition zone consists of modulating the maximum accumulation capacity FW (for "Full Well") of the LIN-LOG response photodetector. A first accumulation capacity value FW1 is used to obtain a first response signal SIG1 and a second accumulation capacity value FW2 to obtain a second response signal SIG2, and it is assumed that the light intensity is stationary during both exposures.
[0092] When the SIG1 and SiG2 signals are in the LIN region, the response is not impacted by the variation in storage capacity, since the voltage is proportional to the number of photons arriving at the photodiode.
[0093] Conversely, if the light intensity is such that the signals SIG1 and SIG2 are in the LOG region, a relatively constant difference D is observed between SIG1 and SIG2. Analyzing the difference: DS = SIG1 - SIG2 It is possible to determine the operating area of SIG1 and SIG2 as illustrated in [Fig.7].
[0094] The first approach based on exposure times and the ratio between signals is used first to determine the operating area of SIG1 and SIG2.
[0095] Thus, when the RS ratio is between R-TH1 and 1+TH2 (TH1 and TH2 being tolerance coefficients determined experimentally), this means that SIG1 (which corresponds to the longest exposure time) has entered the LOG regime while SIG2 is still in the LIN regime. Recall that SIG2 then exhibits a precise and stable response, whereas SIG1, located in the LOG zone, suffers from a highly dispersed shift. Typically, the coefficients TH1 and TH2 are approximately equal to Rx0.2 and Rx0.8, respectively.
[0096] Now with reference to [Fig.8], according to the invention, the signal SIG2 is transformed into its logarithmic value LSIG2 weighted with the slope of the logarithmic response of SIG1, i.e.: LSIG2 = Kl*log(SIG2)
[0097] When the RS ratio is between R-TH1 and 1+TH2, the value of SIG1 must be equal to LSIG2. In this case, the photodetector has two equal measurements, SIG1 and LSIG2, for the same light intensity (which is assumed here to be constant between the two exposures).
[0098] Since LSIG2 is a precise and stable response, because it is obtained when the photodetector is still operating in the LIN zone, it is therefore possible to obtain a correction offset value OS1 that allows SIG1 to be aligned with LSIG2: OS1 = LSIG2-SIG1
[0099] This OS1 parameter can be calculated either during a calibration phase or each time that, for a given exposure, the value of RS falls within the interval between R-TH1 and 1+TH2.
[0100] It is thus possible to generate a unified logarithmic response FS of the photodetector which is determined as follows:
[0101] FS = LSIG2 if RS > 1+TH3 FS = SIG1 + OS1 if RS <= 1+TH3
[0102] The TH3 threshold is preferably approximately equal to R / 2. Figure 8 illustrates this. graphically this correction and obtaining the unified output signal FS.
[0103] Now with reference to [Fig.9], when using the second approach based on modulation of the accumulation well depth FW, two exposures of the photodetector are made with two different energy levels of the barrier TX.
[0104] The output processing circuit then calculates the difference DS between the two response signals obtained SIG1 and SIG2.
[0105] If the DS value is zero or less than a tolerance value TH4 (chosen to account for potential measurement noise), this means that the values of signals SIG1 and SIG2 are both in the LIN range. When this difference DS is greater than TH4, then the values of signals SIG1 and SIG2 are in the LOG range.
[0106] It is possible to set two tolerance thresholds, TH5 and TH6, and when it is determined that DS lies within this interval, this means that the value of the signal SIG1 is in the LIN zone while the value of the signal SIG2 is in the LOG zone. The values of TH5 and TH6 are chosen to allow for the certain determination that SIG1 is indeed in the LIN zone and that SIG2 is indeed in the LOG zone. These values can be determined by simulations and / or by actual measurements. Typically, TH5 can be set to Dx0.2, and TH6 to Dx0.8.
[0107] The digital processing device then mathematically transforms SIG1 into its logarithm weighted by the logarithmic slope of SIG2 (K2):
[0108] LSIG1 = K2*LOG(SIG1)
[0109] Recall that the LSIG1 value is a precise and stable response, as it is obtained when the photodetector is still operating in the LIN regime. It constitutes a reference value. Recall also that the value of SIG2 must be corrected to coincide with LSIG1 when DS is in the interval [TH5, TH6], the offset value being: OS2 = LSIG1 - SIG2
[0110] It therefore becomes possible to generate a unified logarithmic response of the FS pixel, which is calculated as follows:
[0111] FS = LSIG1 if DS < TH7
[0112] FS = SIG2 + OS2 if DS >= TH7
[0113] A value approximately equal to D / 2 is preferably chosen for the TH7 threshold. Figure 9 graphically illustrates this correction and the obtaining of the unified signal FS.
[0114] We will now describe a possible approach for the dynamic updating of correction parameters.
[0115] As seen above, the main correction parameters are the logarithmic slope Kl or K2 and the offset value OS1 or OS2. The parameter Kl or K2, essentially related to the temperature and the gain of the reading chain, can either be measured during a calibration phase and extracted from a reference circuit integrated into the sensor.
[0116] The OS1 or OS2 offset value is determined respectively from the responses with two different exposure times or from the responses with two different integration capacitance values. The initial offset value can be obtained in a calibration procedure with a stable light intensity. Because its value changes in a rather complex way with temperature and operating conditions, it is advantageous to update this value frequently during the use of the LIN-LOG response photodetector.
[0117] It can be anticipated that the processing circuit calculates the value of RS (respectively DS) during operation, and that each time the calculated value falls within a given range, the value of OS1 (respectively OS2) is recalculated. This continuous update makes it possible to keep pace with changes in the operating conditions of the photodetector and to maintain the effectiveness of the correction.
[0118] It should be noted that in real-world use, light intensity varies, and this variation can introduce errors between two shots. Continuous and adaptive correction can then be adopted to keep pace with changing conditions.
[0119] It can also be expected that the digital processing circuit of the photodetector applies a smarter update procedure to: - apply a smoothing function on the updates, - discard abnormal correction values, which can occur when the light varies greatly between the two exposures.
[0120] An example of an architecture for implementing the invention in the context of a pixel array image sensor comprises a substrate on which the photoelectric conversion elements and the control transistors are implemented. A specialized companion processor is attached to this substrate, programmed to apply the corrections described above to the output signals of each pixel and to generate a unified logarithmic response (FS) image. This image can be directly used by a conventional downstream processing unit.
[0121] We will now describe an approach to optimizing the signal-to-noise ratio.
[0122] It is worth recalling here that when using the exposure time modulation approach, the unified low-light response relies on the SIG2 response obtained with the shortest exposure time. However, in low-light conditions, this solution is not optimal. In low light, both the SIG1 and SIG2 signals are in LIN mode, but the SIG2 signal has a worse signal-to-noise ratio than the SIG1 signal, which is generated with a longer exposure time. To optimize performance, the unified FS response can be generated using the response given by the SIG1 signal (divided by the value of R) instead of SIG2. This results in better image quality in low light. The decision threshold between generating the FS response as described above and as described in this paragraph can be determined experimentally on a real sensor.
[0123] When the modulation approach to the integration capacity is implemented, under low-light conditions, the SIG1 and SIG2 signals can be averaged to obtain a better signal-to-noise ratio. Indeed, under low light conditions, the SIG1 and SIG2 signals theoretically have the same value, and calculating an average of the two values reduces random noise. As in the previous case, the decision threshold can be determined experimentally.
[0124] We will now describe an implementation of the invention allowing for a superposition in time of the two exposure phases leading to the generation of the SIG1, SIG2 signals.
[0125] First, it is recalled that the approaches to developing a unified logarithmic response as described above are based on two sequential readings of the same photodetector with the different operating parameters, namely: - either the variation of the exposure time, - or the variation of the accumulation capacity.
[0126] It should be noted here that it would also be possible to combine these two approaches.
[0127] This implies a lengthening of the overall duration of an acquisition, which can sometimes pose problems for matrix sensor applications equipped with such photodetectors and which have to process rapid movements, for example in automotive vision.
[0128] In this case, with reference to [Fig. 10], a dual photodetector is implemented according to the invention, that is, with two photodiodes and two associated control circuits whose image capture parameters are different, either by programming or by construction. Each dual photodetector simultaneously or almost simultaneously provides the two signals SIG1 and SIG2 from which the same correction than that described previously to obtain a unified logarithmic response.
[0129] Thus, on the left side of [Fig. 10] we have illustrated the two sequential approaches based on a single photodetector, one with modulation of the exposure time (designated by T1 / T2), and the other with modulation of the accumulation capacity (designated by FW1 / FW2).
[0130] The right part of the figure shows the same approaches with two photodetectors for the same pixel, which has the advantage of ensuring at least partial simultaneity of captures and thus shortening the overall acquisition and processing time for an image.
[0131] For example, if the approach based on two different exposure times is used, the control signals are programmed so that the two exposure phases end at the same time or at approximately the same time. In this case, the SIG1 and SIG2 signals are delivered at the same instant, and it is not necessary to provide a buffer to store the signal that is delivered first, as in the single-photodetector embodiment.
[0132] If we use the approach based on the modulation of the accumulation capacity, this modulation can be implemented in a material way, by providing different physical characteristics for the two photodiodes, for example by using different doping levels.
[0133] Of course, the present invention is in no way limited to the embodiments described and represented in the drawings, but a person skilled in the art will be able to make many variations or modifications to it.
Claims
Demands
1. A photodetector element intended in particular for forming a matrix image sensor, comprising: - a buried photodiode (PPD) exposed to a radiation source and having an accumulation capacity, - a scattering node (FD) linked to the photodiode and capable of transforming a change in charge of photoelectric origin in the photodiode into a voltage, - a controlled recovery switch (RST) capable of imposing a voltage on the scattering node, - a charge transfer switch (TX) capable of transferring the charge accumulated in the photodiode to the linked scattering node, and - a photodiode exposure control device capable of controlling the recovery switch and the charge transfer switch to generate a response signal from the photodetector element, this response signal intrinsically having an initially linear phase (LIN) followed by a logarithmic phase (LOG),with a potentially indeterminate transition (X) between the two modes, an element characterized in that the control device is capable of causing two successive exposures of the photodiode with two different exposure times, and in that it further comprises a computing device receiving as input the two response signals corresponding to the two exposures and capable of generating a unified logarithmic response signal over the entire range of the response from a combination of the two response signals: - by determining a ratio (R) between the values of the two signals, and, as a function of the value of this ratio: - by forming an output signal (FS) by converting the signal value obtained with the shortest exposure time into its logarithm, by applying a pre-established slope coefficient (Kl),or - by forming an output signal (FS) by applying a predetermined offset (OS1) to the signal value obtained with the longest exposure time.
2. A photodetector element intended in particular to form a matrix image sensor, characterized in that it comprises:
3. - a first buried photodiode (PPD) exposed to a radiation source and having an accumulation capacity, - a second buried photodiode (PPD) exposed to the same radiation source and having an accumulation capacity, - at least one diffusion node (FD) linked to each photodiode and capable of transforming a photoelectric charge variation in the photodiode into a voltage, - at least one controlled recovery switch (RST) capable of applying a voltage to the broadcast node, - two charge transfer switches (TX) capable of transferring the charge accumulated in the two photodiodes to the diffusion node, and - a photodiode exposure control device capable of controlling the reset switch and the charge transfer switch with different exposure times, to generate two different response signals from the two photodiodes, these response signals intrinsically having an initially linear phase (LIN) followed by a logarithmic phase (LOG), with a potentially indeterminate transition (X) between the two modes, the exposures of the two photodiodes having at least partial temporal overlap, the photodetector further comprising a computing device receiving as input the two response signals obtained with the two photodiodes respectively and capable of generating a unified logarithmic response signal over the entire response range from a combination of the two response signals: - by determining a ratio (R) between the values of the two signals, and, depending on the value of this ratio: - by forming an output signal (FS) by converting the signal value obtained with the shortest exposure time into its logarithm, by applying a pre-established slope coefficient (Kl), or - by forming an output signal (FS) by applying a pre-established offset (OS1) on the signal value obtained with the longest exposure time. Element according to claim 2, characterized in that the exposure periods defining the two exposure durations end substantially at the same instant for each capture.
4. A photodetector element intended in particular to form a matrix image sensor, comprising: - a buried photodiode (PPD) exposed to a radiation source and having an accumulation capacity, - a scattering node (FD) linked to the photodiode and capable of transforming a photoelectric charge variation in the photodiode into a voltage, - a controlled recovery switch (RST) capable of imposing a voltage on the scattering node, - a charge transfer switch (TX) capable of transferring the charge accumulated in the photodiode to the linked scattering node, and - a photodiode exposure control device capable of controlling the recovery switch and the charge transfer switch to generate a response signal from the photodetector element, this response signal intrinsically having an initially linear phase (LIN) followed by a logarithmic phase (LOG),with a potentially indeterminate transition (X) between the two modes, an element characterized in that the control device is capable of causing two successive exposures of the photodiode with two different accumulation capacitance values, and in that it further comprises a computing device receiving as input the two response signals corresponding to the two exposures and capable of generating a unified logarithmic response signal over the entire range of the response from a combination of the two response signals by determining a difference (DS) between the values of the two signals, and, depending on the value of this difference: - by forming an output signal (FS) by converting the signal value obtained with the largest accumulation capacitance into its logarithm, by applying a pre-established slope coefficient (K2),or - by forming an output signal by applying a predetermined offset (OS2) to the signal value obtained with the smallest accumulation capacitance.
5. A photodetector element intended in particular to form a matrix image sensor, characterized in that it comprises: - a first buried photodiode (PPD) exposed to a radiation source and having a first accumulation capacitance value, - a second buried photodiode (PPD) exposed to the same radiation source and having a second accumulation capacitance value different from the first, - at least one diffusion node (FD) linked to each photodiode and capable of transforming a photoelectric charge variation in the photodiode into a voltage, - at least one controlled recovery switch (RST) capable of applying a voltage to the broadcast node, - two charge transfer switches (TX) capable of transferring the charge accumulated in the two photodiodes to the diffusion node, and - a photodiode exposure control device capable of controlling the reset switches and the charge transfer switches to generate two different response signals from the two photodiodes, these response signals intrinsically exhibiting an initially linear phase (LIN) followed by a logarithmic phase (LOG), with a potentially indeterminate transition (X) between the two modes, the exposures of the two photodiodes exhibiting at least partial temporal overlap, the photodetector further comprising a computing device receiving as input the two response signals obtained with the two photodiodes respectively and capable of generating a unified logarithmic response signal over the entire response range from a combination of the two response signals by determining a difference (DS) between the values of the two signals, and, depending on the value of this difference: - by forming an output signal (FS) by converting the signal value obtained with the largest accumulation capacity into its logarithm, by applying a pre-established slope coefficient (K2), or - by forming an output signal by applying a pre-established offset (OS2) on the signal value obtained with the smallest accumulation capacity.
6. Element according to claim 5, characterized in that the different accumulation capacity values are fixed and obtained by construction.
7. Element according to claim 4 or 5, characterized in that different accumulation capacity values are obtained by variable biasing at the charge transfer switch(s) (TX).
8. Pixel matrix photoelectric sensor, characterized in that each pixel comprises a photodetector element according to any one of claims 1 to 7, and in that it comprises a common digital processing device forming the computing devices respectively associated with the different pixels.