WO2009141590A2

WO2009141590A2 - Image sensor

Info

Publication number: WO2009141590A2
Application number: PCT/GB2009/001235
Authority: WO
Inventors: Steve Collins; Hsiu-Yu Cheng
Original assignee: Isis Innovation Limited
Priority date: 2008-05-22
Filing date: 2009-05-19
Publication date: 2009-11-26
Also published as: GB0809337D0; GB2460260A; WO2009141590A3

Abstract

An image sensor for an electronic imaging device includes an array of pixels. Each pixel (2) includes a switch device (8) having a first switch input, a second switch input and a switch output that is switchably connected to the first switch input. The switch device (8) is constructed and arranged to disconnect the switch output from the first switch input at a capture moment determined by comparing a first input signal V_In1 and a second input signal V_In2. A photosensor device (4) has a photosensor output that is connected to the first switch input. A first signal -generator (18) generates alternately a reference signal V_ref and a reference calibration signal V_refcal, and is connected to the second switch input. A second signal generator (30) generates a photosensor reference signal V_pcal, and is switchably connected to the first switch input. A readout circuit (10) is arranged to capture the switch output signal V_s at the capture moment and provide an output signal that is related to the switch output signal.

Description

IMAGE SENSOR

The present invention relates to an image sensor for an electronic imaging device, and to a method of sensing. The invention also relates to an imaging device having such an image sensor.

An image sensor for an electronic imaging device such as a camera generally consists of an array of photosensitive picture element detectors ("pixels"). Light falling on the image sensor is detected by the pixels, which generate output signals corresponding to the amount of light falling on the pixels. The output signals of the pixels are digitised and stored in an electronic file that contains the image information. The image sensor may be incorporated into either a still camera for taking single images, or a video camera, or any other electronic imaging device. There is a huge market for electronic imaging sensors, which may be incorporated in a wide-range of systems including digital cameras, mobile telephones, personal computers, surveillance systems and various automobile applications, including driving aids such as night vision cameras and reversing/manoeuvring aids. These cost-sensitive markets have spurred the development of single chip CMOS (complementary metal oxide semiconductor) cameras. Like the CCD (charge coupled device) sensors that they have largely superseded the vast majority of CMOS cameras create an image by integrating the photocurrent within each pixel for a predetermined period. Integrating the photocurrent within each pixel creates a voltage that is proportional to the photon flux at the corresponding pixel. This voltage is then converted to an equivalent binary number. Usually this binary number contains 10 or fewer bits and the dynamic range of each pixel is therefore 6OdB or less.

Integrating the photocurrent is a simple imaging strategy that achieves a large enough dynamic range for scenes with relatively uniform levels of illumination. However, natural scenes can have a dynamic range of 6 decades and as a result some naturally illuminated scenes will be either underexposed in the darker areas, overexposed in the brighter areas or both, leading to a loss of detail in the affected areas. While this is inconvenient for consumers, it is a real problem for surveillance systems and may be actually dangerous for „^■ _} some automotive applications. Several techniques have therefore been developed to increase the dynamic range of imaging sensors.

5 For example, a method of extending the dynamic range of linear pixel detectors has been devised by Stoppa et al (David Stoppa, Andrea Simoni, Lorenzo Gonzo, Massimo Gottardi and Gian-Franco Dalla Betta: 'Novel CMOS Image Sensor with a 132-dB Dynamic Range' IEEE JSSC 37(12) 1846-1852 (2002)). To achieve an increase in dynamic range, a comparator is integrated into each pixel. The comparator compares the voltage within the

10 pixel with a threshold voltage. If the pixel voltage reaches the threshold value, the comparator disconnects two capacitors in the pixel from two analogue input voltages that together represent the time at which this event occurs. At the end of the integration process, the pixel voltage and the two time voltages are sampled from each pixel. These three analogue voltages are then digitised to 8-bits each, creating a 24-bit signal that

15 encodes the photocurrent within the pixel. The large number of bits per pixel produced by the above system is typical of the results of trying to represent a high dynamic range (HDR) signal in a linear format.

However, for many surveillance and automotive applications the captured image must be displayed to the user and acquiring a high dynamic range image is only the first step in any

20 useful system. An equally important aspect of the system is the ability to display the image once it has been acquired. In fact, the dynamic range of display devices is at most 8-bits per pixel, which is smaller than the dynamic range of a conventional digital camera. Displaying high dynamic range images so that they are perceptually acceptable is therefore another challenge. A complete imaging system must both acquire high dynamic range

25 images and process those images so that they can be displayed, whilst retaining as much detail as possible in all parts of the displayed image.

The problems associated with displaying high dynamic range images on conventional displays has been an active area of research for several years. One class of techniques that have been developed to solve this problem uses a global tone-mapping algorithm in which 30 a common function is employed to transform each pixel value in the captured high dynamic range image to a displayed value that falls within the low dynamic range of the display device. This allows detail to be seen in both the brightest and darkest regions of the image. However, applying a tone mapping algorithm to a captured image is a relatively slow process, which requires additional image processing. This precludes its use in certain time- or cost-sensitive applications. One method of capturing a HDR image and recording it with a comparatively small number of bits per pixel is to use an image sensor that has a logarithmic output. A high dynamic range pixel having a logarithmic output is described in the inventor's international patent application WO2007/051964A. This pixel resolves some of the problems associated with earlier pixels. The logarithmic output allows the image sensor to image a scene with a high dynamic range and represent that scene on a display with a much lower dynamic range without losing detail in either the highlight or dark regions of the image. Also, as the conversion to a logarithmic value takes place within the pixel, it is very quick and does not significantly increase the cost of the pixel.

However, using a pixel with a logarithmic output does not resolve all problems associated with imaging HDR scenes. One problem with logarithmic compression is that if the input range of the pixel array does not match the range of photocurrents in the scene the image can be clipped, resulting in a loss of detail at one or other end of the dynamic range. The pixel is also limited to a single tone mapping algorithm (a logarithmic algorithm). Although many tone-mapping algorithms have been proposed, there is little agreement about the relative merits of those algorithms and the best algorithm in any particular case may depend on the application and/or the specific scene. There would therefore be considerable merit in providing a pixel that has the ability to support different tone- mapping algorithms, and that allows the user to select the algorithm that is to be implemented. Another problem that affects all CMOS sensors is fixed pattern noise (FPN), which is caused by variations between the nominally identical electronic components of different pixels. This degrades the quality of the resulting images. Minimisation and correction of fixed pattern noise is a key aspect of the design of any imaging sensor.

The FPN in pixels is characterised by variations in an additive (offset) component of the output and variations between the responses of different pixels. Experiments have shown that the dominant form of FPN in conventional integrating pixels is generally the additive component, which is caused mainly by threshold voltage variations in the readout circuit. To reduce FPN, double sampling schemes are used, in which an image is formed from the difference between the output voltage when the pixel is reset and the output voltage after the pixel has integrated the pixel current. However, such an approach is not effective in a high dynamic range pixel of the type described in the inventor's international patent application WO2007/051964A. This pixel suffers from all the sources of fixed pattern noise that are known from conventional pixel designs, plus an additional source of noise caused by threshold voltage variations in a disconnect transistor that links the photosensor to a readout circuit. This additional source of fixed pattern noise means that the conventional methods of correcting fixed pattern noise are not as effective as in previous designs. A new fixed pattern noise correction procedure is therefore required in pixels of this type.

Another important aspect of the design of CMOS image sensors is the size of the pixels, which affects both the light sensitivity of the sensor and also its maximum resolution. Reducing the size of the pixel increases the maximum resolution of the sensor by allowing it to carry more pixels for a fixed price. However, reducing the size of the photosensor reduces its sensitivity to light. It is therefore desirable to reduce the number of transistors per pixel, so that the pixel size can be reduced without reducing the size and sensitivity of the photosensor. WO 02/054759 A3 describes a CMOS pixel cell that includes a photodiode and a pMOS voltage comparator working as a one bit analogue to digital converter. The circuit relies on the fact that the effective threshold voltage of a transistor depends upon its substrate voltage, which is less than unity. As a result, the method is les sensitive than using the gate voltage directly. The circuit only provides a digital output. US 6069377 describes an image sensor that uses a timer to measure the integration period and increase dynamic range. The pixel has two outputs that have to be merged to create the final image. Although the need to reduce the number of transistors is acknowledged, this aim is only partially achieved.

It is an object of the present invention to provide an image sensor that mitigates at least some of the aforesaid problems. According to one aspect of the invention there is provided an image sensor for an electronic imaging device, the image sensor including an array of pixels, each pixel including a switch device having a first switch input for a first input signal Vi_nI, a second switch input for a second input signal Vi₁^, and a switch output for a switch output signal V₅ that is switchably connected to the first switch input, the switch device being constructed and arranged to disconnect the switch output from the first switch input at a capture moment determined by comparing the first input signal Vi_n] and the second input signal Vi₁₁₂, a photosensor device for detecting incident light, said photosensor having a photosensor output for a photosensor signal V_p that represents a time integral of the detected light intensity, said photosensor output being connected to said first switch input, a first signal -generator that is constructed and arranged to generate alternately a reference signal V_ref and a reference calibration signal V_refcai, said first signal generator being connected to said second switch input, a second signal generator constructed and arranged to generate a photosensor reference signal V_pcai, said second signal generator being switchably connected to said first switch input, and a readout circuit that is arranged to, capture the switch output signal V_s at the capture moment and provide an output signal that is related to the switch signal V_s.

This arrangement allows fixed pattern noise (FPN) to be reduced by eliminating the additive component of FPN. Since the additive component is the major source of FPN in CMOS sensors, the reduction in FPN can be substantial.

Advantageously, the first signal generator is constructed and arranged to generate alternately a reference signal V_ref and a reference calibration signal V_refcai.

Advantageously, the pixel is constructed and arranged such that, in use, it carries out separate integration and calibration steps, wherein during the integration step, the first switch input receives the photosensor signal V_p, the second switch input receives the reference signal V_ref, and the readout circuit captures a first switch output signal V_sl, and during the calibration step, the first switch input receives the photosensor calibration signal V_pcai, the second switch input receives the reference calibration signal V_refcai, and the readout circuit captures a second switch output signal V_s2, the amount of light falling on the photosensor being determined from the difference between the first switch output signal V_si and the second switch output signal V_s2. Optionally, the photosensor calibration signal V_pcai comprises a voltage that changes at a uniform rate. Preferably, this ramp voltage represents a photocurrent approximately in the middle of the dynamic range. Alternatively, a constant voltage equivalent to an average or typical output voltage may be used to minimise responsivity variations. Preferably, the reference signal V_ref and the reference calibration signal V_refcai are similar in form. This is convenient, as it means both signals can be generated in the same way. However, they may vary at different rates. Preferably, the reference calibration signal V_refcai changes at a faster rate than the reference signal V_ref, so that the duration of the calibration step is shorted than that of the integration step. Alternatively, the forms of the reference signal V_ref and the reference calibration signal V_refcaimay be different.

The switch device may be constructed and arranged to disconnect the switch output from the first switch input when the difference between first input signal Vi_n i and the second input signal V^ equals a switch threshold voltage V_th.

According to another aspect of the invention there is provided an image sensor for an electronic imaging device, the image sensor including an array of pixels, each pixel including a switch device having a first switch input for a first input signal Vi_nJ, a second switch input for a second input signal V_h12, and a switch output for a switch output signal V₅ that is switchably connected to the first switch input, the switch device being constructed and arranged to disconnect the switch output from the first switch input at a capture moment determined by comparing the first input signal Vi_nI and the second input signal V_h2, a photosensor device for detecting incident light, said photosensor having a photosensor output for a photosensor signal V_p that represents a time integral of the detected light intensity, said photosensor output being connected to said first switch input, a signal generator that is constructed and arranged to generate a reference signal V_ref, said signal generator being connected to said second switch input, and a readout circuit that is arranged to capture the switch output signal V_s at the capture moment and provide an output signal that is related to the switch signal V_s; characterised in that said pixels are arranged in a plurality of groups where each group includes a plurality of pixels, and wherein at least two pixels within said group share a common readout circuit. Arranging the pixels in groups that include one or more shared components reduces the number of transistors per pixel and so helps to reduce the size of the pixel without affecting the size of the photosensor. The maximum resolution of the sensor can thus be increased without reducing the sensitivity of the pixels.

Advantageously, said common readout circuit is also shared with one or more pixels in another group. Advantageously, each group includes a plurality of read switches, for ^"5 selecting which pixel is connected to the readout circuit. Advantageously, each group includes a reset switch and at least two pixels within said group share said reset switch.

According to another aspect of the invention there is provided an image sensor for an electronic imaging device, the image sensor including an array of pixels, each pixel including: a switch device having a first switch input for a first input signal Vi_nJ, a second0 switch input for a second input signal Vi_n2, and a switch output for a switch output signal V_s that is switchably connected to the first switch input, the switch device being constructed and arranged to disconnect the switch output from the first switch input at a capture moment determined by comparing the first input signal Vi_nI and the second input signal Vuα, a photosensor device for detecting incident light, said photosensor having a5 photosensor output for a photosensor signal V_p that represents a time integral of the detected light intensity, said photosensor output being connected to said first switch input, a signal generator that is constructed and arranged to generate a reference signal V_ref, said signal generator being connected to said second switch input, and a readout circuit that is arranged to capture the switch output signal V₈ at the capture moment and provide an0 output signal that is related to the switch signal V_s; characterised in that said reference signal V_ref varies according to a predetermined function V_ref (t) that may be selected from a plurality of different functions.

This arrangement allows the response of the sensor to be adjusted according to the contrast range of the scene to be imaged or the tone mapping algorithm that is required. It is5 therefore possible to customise the image sensor according to the user's requirements.

Advantageously, the image sensor includes selector means for selecting a function from said plurality of different functions.

Advantageously, the signal generator is programmable with different predetermined functions V_ref (t). Advantageously, at least one of said different functions is such that the output signal is related to the intensity of light incident on the photosensor device by a function that is logarithmic over one portion of the pixel's dynamic range and non-logarithmic over another portion of the pixel's dynamic range. Advantageously, said predetermined function V_ref(t) is defined to create a desired response R by creating a pixel with an effective integration time tφ where C is the capacitance of the pixel and _ CAV_p

Preferably, where I_add is an additional current discharging the pixel, the effective integration time t_eff is

Advantageously, the image sensor includes a reset device for applying a reset signal V_K55 to the photosensor device.

Advantageously, the photosensor includes a capacitance and a photosensor element that is arranged to conduct a current depending on the detected light intensity, and said photosensor signal represents a voltage across the capacitance as it is charged or discharged by the photosensor current.

Advantageously, the image sensor includes a reset device that applies an initial voltage to said capacitance, wherein said capacitance is subsequently discharged by said photosensor current.

Advantageously, the reference signal varies according to a non-linear function. Preferably, the rate of change of the reference signal decreases with time.

Advantageously, the switch device includes a transistor switch with a gate connected to receive the reference signal, a source connected to receive the photosensor signal and a drain connected to the readout circuit. Preferably, the transistor switch is a p-channel MOSFET.

The transistor switch may be constructed and arranged to disconnect the photosensor signal from the readout circuit when the difference between the reference signal and the photosensor signal is less than a threshold value. According to another aspect of the invention there is provided a method of sensing images using an electronic imaging device having an image sensor that includes an array of pixels, the method comprising performing an integration step by detecting incident light with a photosensor and providing a photosensor signal V_p, generating a reference signal V_ref, comparing the photosensor signal and the reference signal, determining a capture moment from the comparison and capturing a first output signal V_sl, performing a calibration step by generating a photosensor calibration signal V_pcai and a reference calibration signal V_refcai, comparing the photosensor calibration signal with the reference calibration signal, determining a capture moment from the comparison and capturing a second output signal V_S2, and deriving a pixel output signal from the difference between the first output signal V_si and the second output signal V_s2.

Advantageously, the photosensor calibration signal Vp_cai comprises a voltage that changes at a uniform rate. Preferably, the reference signal V_ref and the reference calibration signal V_refcai are similar in form. According to another aspect of the invention there is provided a method of sensing images, using an electronic imaging device having an image sensor that includes an array of pixels, the method comprising: detecting incident light with a photosensor and providing a photosensor signal V_p, generating a reference signal V_ref, comparing the photosensor signal and the reference signal, determining a capture moment from the comparison and capturing an output signal; characterised in that said pixels are arranged in a plurality of groups where each group includes a plurality of pixels, and wherein the output signals of at least two pixels within said group are captured by a common readout circuit.

Advantageously, said common readout circuit also captures the output signals of one or more pixels in another group. Preferably, said pixels are connected sequentially to said common readout circuit.

Advantageously, at least two pixels within said group are reset by a common reset switch.

According to another aspect of the invention there is provided a method of sensing images using an electronic imaging device having an image sensor that includes an array of pixels, the method comprising detecting incident light with a photosensor and providing a photosensor signal V_p, generating a reference signal V_ref, comparing the photosensor signal and the reference signal, determining a capture moment from the comparison and capturing an output signal; characterised in that said reference signal V_ref varies according to a predetermined function V_ref (t) that may be selected from a plurality of different functions.

Advantageously, at least one of said different functions is such that the output signal is related to the intensity of light incident on the photosensor device by a function that is logarithmic over one portion of the pixel's dynamic range and non-logarithmic over another portion of the pixel's dynamic range.

Advantageously, said predetermined function V_ref(t) is defined to create a desired response

R by creating a pixel with an effective integration time tφ, where C is the capacitance of the pixel and CAV

Advantageously, said predetermined function V_ref(t) is defined to create a desired response R by creating a pixel with an effective integration time tφ where C is the capacitance of the pixel, I_add is an additional current discharging the pixel and

Advantageously, the capture m '*o is determined by sensing when the difference between the photosensor signal and the reference signal reaches a predetermined value.

Advantageously, the photosensor conducts a current that depends on the detected light intensity, and said photosensor signal represents a voltage across a capacitance that is charged or discharged by said photosensor current.

Advantageously, the method includes applying an initial voltage to said capacitance and subsequently discharging said capacitance by said photosensor current.

Advantageously, the reference signal varies according to a non-linear function. Advantageously, the rate of change of said reference signal decreases. Advantageously, the method includes capturing the photosensor signal at the capture moment, said output signal being related to the captured photosensor signal. Advantageously, the method includes disconnecting said readout circuit from said photosensor signal at the capture moment. Certain embodiments of the invention will now be described, by way of example, with reference to the accompanying drawings, in which:

Figure 1 is a schematic circuit diagram of a pixel that forms part of an electronic image sensor according to a first embodiment of the invention; Figure 2 is a graph showing the relationship between the pixel output voltage and the illumination intensity for a pixel of the type shown in Fig. 1 ;

Figure 3 is a schematic circuit diagram of a pixel that forms part of an electronic image sensor according to a second embodiment of the invention;

Figure 4 is a graph showing the variation with time of a reset voltage, a reference voltage and a readout voltage in a pixel of the type shown in Fig. 3; and

Figure 5 is a schematic circuit diagram showing an arrangement of pixels within an electronic image sensor according to a third embodiment of the invention.

A CMOS image sensor for a camera conventionally includes an array of pixels. Figure 1 is a schematic circuit diagram of a single pixel 2 that forms part of such an array. The pixel 2 includes a photosensor device 4, a transistor disconnect switch 8 (M2), a readout circuit 10 and a transistor reset switch 12 (Ml).

The photosensor device 4 comprises a photodetector, for example a photodiode or phototransistor, having a small capacitance C. The required capacitance can be provided either by the intrinsic capacitance of the photodetector and other circuit components, or by a separate capacitor (not shown) connected in parallel with the photodetector. The photosensor device 4 provides an output signal at photosensor connection 14, which is represented by voltage V_p. This output signal is connected to one terminal of the transistor disconnect switch 8.

The other terminal of the disconnect switch 8 provides a readout voltage V_s at readout connection 16, which is connected to the readout circuit 10. The gate terminal of the disconnect switch 8 is connected to the output of a signal generator 18. In this embodiment, the disconnect switch 8 is a p-channel MOSFET (metal oxide semiconductor field effect transistor), which only conducts when the gate voltage V_gate is less than the source voltage V_source minus the transistor threshold voltage V_th- When the gate voltage is equal to or greater than the source voltage minus the threshold voltage, the disconnect switch 8 ceases to conduct.

The reset switch 12 comprises another MOSFET transistor. The source terminal of the reset switch 12 is connected to a fixed power supply voltage V_dd, the gate terminal is connected to receive a reset signal and the drain terminal is connected to the photosensor device 4 through photosensor connection 14.

The signal generator 18 generates a time-dependent control signal represented by the reference voltage V_re/, which is connected to the gate of the disconnect switch 8 to control operation of the switch. When the reference voltage V_ref applied to the gate is less than the photosensor voltage V_p minus the threshold voltage V_th the switch conducts, so applying the photosensor voltage V_p to the readout circuit 10 via the readout connection 16. When the reference voltage V_ref is greater than the photosensor voltage V_p minus the threshold voltage V_th, the disconnect switch 8 ceases to conduct, thus isolating the readout circuit 10 from the photosensor voltage V_p. The readout circuit 10 comprises a conventional selectable source follower read out circuit, which includes a source follower transistor 20 (M3), a select transistor 22 (M4), and an output node 24 for an output signal V_ouh which is connected to the source connection of the select transistor 22. The readout circuit 10 is designed to sense the voltage V₅ at the readout connection 16 of the disconnect switch 8, without drawing any current from the output. The gate 26 of the select transistor 22 is connected to receive a select voltage V_seι and in use selects which pixel in a row or column of pixels is connected to the output node 24. The source terminal of the select transistor 22 is also connected to a biasing transistor 27, which receives a biasing voltage V _as at the gate terminal 28.

In use, the image collection process is initiated by applying a low reset voltage to the gate of the transistor reset switch 12, causing the switch to conduct. This applies a high voltage

VD_D to the photosensor 4, which stores a charge owing to the capacitance C of the circuit components. The reset voltage then goes high and the reset switch 12 stops conducting, allowing the stored charge to discharge through the photodetector 4. The rate at which the charge discharges depends on the current I_p through the photodetector 4, which in turn depends on the intensity of light incident on the photodetector. The voltage V_p at photosensor connection 14 thus decreases at a rate that depends on the intensity of light falling on the photosensor 4.

At a time t after the reset voltage has gone high, the photosensor voltage V_p will be given by the equation: V_p = V_dd - I_pΛ/C

The photosensor 4 thus provides an output signal V_p at photosensor connection 14 that is proportional to a time integral of the detected light intensity (the constant of proportionality being negative). The output signal V_p of the photosensor 14 is applied to the input of the readout circuit 10 via disconnect switch 8 and the readout connection 16. The signal generator 18 is constructed and arranged to generate a time dependent reference signal F_re/that varies according to a predetermined function V_ref(t). Various functions can be applied but generally the reference signal V_ref will increase with time t, and the rate of change of the reference signal will decrease with time.

The transistor disconnect switch 8 receives the photosensor signal V_p at its source terminal and the reference signal F_re/at its gate terminal. The drain terminal of the transistor switch

8 is connected to the readout. circuit 10. When the reference voltage F_re/ is less than the photosensor voltage V_p minus the threshold voltage Vu, (i.e. when V_re/ < V_p - V,/,) the disconnect switch 8 will conduct, applying the photosensor signal V_p to the readout circuit

10. However, when the reference voltage F_re/is equal to or greater than the photosensor voltage V_p minus the threshold voltage Vu, (i.e. when V_re/ > V_P - V₁/,), the disconnect switch

8 will stop conducting, thus disconnecting the photosensor signal V_p from the readout circuit 10. hi other words, the readout circuit 10 is isolated from the photosensor at a capture moment, when the difference between the photosensor voltage V_p and the reference voltage V_ref is less than the threshold voltage Vu, (i.e. when V_p - V_ref < Vu₁). It should be noted that with p-channel MOSFETs, the threshold voltage is sometimes given as a negative number. The readout circuit 10 thus samples the photosensor signal V_p at the capture moment and provides a read out signal at output node 24.

The output signals from all of the pixels in the image sensor array are fed to a conventional electronic processor (not shown), which digitises the signals and combines them to form an image file. This file may be viewed and/or stored for later viewing. The capture process can if required be repeated at suitable intervals to provide a video signal.

The function that defines the form of the reference signal V_ref(t) is designed such that the captured output signal V_p is related by a predetermined tone mapping algorithm to the integrated detected light intensity. Different relationships between the photocurrent and the pixel output voltage can be obtained using different functions for the reference signal V_ref(t). For example, it is possible to create a pixel in which the voltage sampled onto the readout circuit is proportional to the logarithm of the photocurrent. Different tone mapping algorithms can be generated by varying the form of the reference signal V_re/t). Alternatively, if a linear response is required, this can be achieved by changing the reference signal V_ref (t) to a constant low voltage value. It is therefore possible to switch easily between different kinds of response.

The reference signal needed to provide a required response can be generated using a similar approach to that used in arbitrary waveform generators. For example, a read-only memory (ROM) and a digital analogue converter (DAC) can be used to create the reference voltage. Different sets of data can be stored within the signal generator allowing a required response to be selected, or it can be designed to be reprogrammed as required.

One way to create a pixel with a user defined response is to adjust the "capture moment" when the readout circuit is disconnected from the photosensor signal so that the effective integration time ^depends on the photocurrent. In an integrating pixel the photocurrent is used to discharge a capacitance. The resulting change in voltage within the pixel at a time t after integration begins is then

p C

The effective integration time must be controlled so that this change in voltage also represents the desired response R

ΔV_p = R(I_p)

It is the change in voltage rather than the photocurrent itself that is detectable. The first step in obtaining an expression for the effective integration time needed to achieve this desired response is to invert this response to obtain an expression for the photocurrent as a function of the change in pixel voltage

This can then be combined with the first equation to show that

One phenomenon that has not be taken into account so far is that the pixel capacitance will be discharged by both the photocurrent arising directly from the imaged scene, I _p and an additional current I₀₁₁₁₁ , so that:

Even with this additional contribution to the current discharging the pixel the output voltage must still only be a function of the photocurrent. To achieve this the effective integration time must be:

A simple tone-mapping function that has been proposed is a logarithmic compression. To achieve this, where S is the gain and /_re/is a reference current, we require

Λ(/_p) = SLn(I_p II_ref) This means that

I_p = I_ref exp(AV_p /S)

and hence to obtain this response we require an effective integration time

A logarithmic response can compress a high dynamic range scene and retain the contrast information that is important for the human visual system. A possible problem with the logarithmic compression is that if the input range of the pixel array does not match the range of photocurrents in the scene, the image can be clipped. This can be avoided using a response that is logarithmic over a central part of the pixel's dynamic range but gradually saturates at one or both ends of the range. One response of this type is

R(I _p) = AV_max /2 x (t_arύi(S(Ln(I_p /I_ref))) + 1)

The inverse of this is

and hence the effective integration time is:

As will be clear, an equivalent procedure can be used to obtain an expression for the effective integration time required to obtain any response. Therefore, by altering the reference signal V_re/t) it is possible to programme the pixel to provide any desired response or tone mapping algorithm. COLOUR RENDITION

In some applications, such as those in which accurate colour rendition is important, it will be critical to know the ratio of photocurrents in different pixels. This means that the rate of change of the sensed pixel output voltage when the photocurrent is changed should be well controlled. In this situation another effect that should be taken into account is that the voltage within the pixel is sensed using a readout circuit. This is conventionally a source- follower circuit, in which the change in the output voltage of the source follower is only a fraction of the change in the voltage within the pixel. To ensure that the source-follower output voltage changes at a particular rate, the rate of change of the pixel voltage must be increased to account for this attenuation. The pixel circuit shown in figure 1 is similar to a conventional integrating pixel, the only additional device being the disconnect switch 8_. This transistor is connected so that it can be used to isolate the readout circuit 10 from the photodiode 4 at a time that depends upon the photocurrent I_p and the time dependent reference voltage V_re/t). This device disconnects these two parts of the pixel when

V_ga,_e(t)-V_source(t) = - V_th

In the absence of the additive current a logarithmic response with a gain of S will be obtained if the time dependent reference voltage V_ref(t) at a time t after integration has started is given by <KX(V_dd-V_lrf(t)-V_th)exp[(V_iel(t)-V_dd+V_tll)/S]yi_ref

This equation can be solved numerically to generate V_ref(t) and when this voltage is supplied to a test pixel manufactured on an 0.25 micron process the result, as shown in figure 2, is a logarithmic response with a gain that depends upon the user defined reference voltage. To calculate the reference voltage we have assumed that the photodiode and the readout circuit are disconnected at a time τ when

However this fails to take into account the threshold voltage of the disconnect switch 8. This transistor is a pMOS device and therefore it will conduct when the gate voltage is more than the threshold voltage below the voltage of the source connection. With V_p decreasing and V_ref increasing the condition when this transistor will stop conducting is

V_ref = V_p - V_th = V_dd -I_pτlC - V_th

This condition is sufficient to obtain a logarithmic response. However to process colour images it is necessary to know the response of the pixel (typically quoted in mV/decade) so that the ratio of the responses of pixels with different spectral sensitivities, and hence the colour of this area of the image, can be determined.

The first effect that must be taken into account so that the reference voltage can be specified to achieve a known pixel response is the back-gate effect in the disconnect switch 8. This effect means that the effective threshold voltage of this device depends upon the reference voltage. Measurements show that for a pixel manufactured on an 0.25 micron process there is a linear relationship between the effective threshold voltage and the reference voltage: V_th = a V_ref - V_m

where α is a constant and V_tllo is the threshold voltage at time = 0. This means that the equation for the reference voltage becomes:

K_ef^ + oc)= V_dd -I_pr/C-V_lh0

Finally, the readout circuit 10 will reduce the response of the pixel detected at the output node 24, as compared the response that occurs on the readout connection 16. This is a linear attenuation and its effects can be negated by aiming for a larger response within the pixel than the one that is required at the output node 24. With these two corrections the reference voltage can be designed to give a well controlled response from a typical pixel. FIXED PATTERN NOISE

All CMOS image sensors suffer from fixed pattern noise (FPN), which is caused by variations in the performance of the nominally identical electronic components making up the different pixels. This degrades the quality of the resulting images. Minimisation and correction of any residual fixed pattern noise is a key aspect of the design of any imaging sensor.

The FPN in CMOS image sensors typically comprises both an additive (offset) component that affects the output of the pixels incrementally, and variations in the sensitivity or response of the pixels. The dominant form of FPN is the additive component. Sources of fixed pattern noise include threshold voltage variations in the readout circuit 10 and in the disconnect switch 8. A scheme for correcting fixed pattern noise will now be described with reference to figures 3 and 4.

The circuit shown in figure 3 is similar to that shown in figure 1, except that the reset transistor switch 12 is connected to a reset signal generator 30 that can generate a variable reset voltage V_res, instead of being connected to the constant supply voltage VM- Operation of the circuit is split into two phases: an integration phase depicted in the left- hand part of figure 4, and a calibration phase shown in the right-hand part of the figure. Typical variations in various voltage signals during each of these phases are illustrated in Fig. 4, the reset voltage signal V_res being represented by a broken line, the reference signal V_ref by a solid line and the readout signal V₅ by a dotted line.

During the integration phase shown in the left half of Fig. 4, the reset voltage V_res is held high. Initially, the reset switch 12 is closed, thus resetting the photosensor voltage V_p to the reset voltage V_res. The disconnect switch 8 is also closed and the readout voltage V_s is therefore equal to V_p. At the commencement of integration, the reset switch 12 opens, disconnecting the photosensor 4 from the reset voltage V_res. The charge stored in the circuit then starts to discharge through the photosensor 4, the photosensor current I_p depending on the intensity of light falling on the photosensor. This causes the photosensor voltage V_p and the readout signal V_s to fall at a steady rate. At the same time, the reference signal V_ref applied to the gate terminal of the disconnect switch 8 rises at a rate determined by the chosen predefined function V_ref(t). The disconnect switch 8 only conducts for as long as the source voltage V_p exceeds the reference voltage V_ref by at least the threshold voltage V,_h of the switch. Therefore, as soon as the difference between the source voltage V_p and the reference voltage V_ref equals the threshold voltage V,_/,, the disconnect switch 8 disconnects the read- out circuit 10 from the photosensor 4 and the read-out circuit captures the readout signal V_s at that moment. This signal is stored as a first output signal Vi.

During the calibration phase, the reset transistor switch 12 closes to reconnect the photosensor 4 to the reset voltage V_res. Respective calibration signals are then generated by both the reset signal generator 30 and the reference signal generator 18. As illustrated in the right-hand part of Fig. 4, the reset signal generator 30 generates a photosensor calibration signal V_pcaι that decreases at a steady, known rate, thus emulating steady discharging of the photosensor. This photosensor calibration signal V_pcaι is applied to the source terminal of the disconnect switch 8. At the same time, the reference signal generator 18 generates a reference calibration signal V_ref_caι that resembles the normal reference signal V_ref but increases at a faster rate. This signal is applied to the gate terminal of the disconnect switch 8. The disconnect switch 8 operates as before and only conducts for as long as the photosensor calibration signal V_pcaι exceeds the reference calibration voltage V_ref_cai by at least the threshold voltage V_th. As soon as the difference between the photosensor calibration signal V_pcaι and the reference calibration voltage V_ref_caι equals the threshold voltage V_th, the disconnect switch 8 disconnects the read-out circuit 10 from the reset signal generator 30 and the read-out circuit captures the readout signal V_s at that moment. This signal is stored as a second output signal F_?.

The output of each pixel is then acquired by subtracting the first output signal V_/ from the second output signal V₂. As the first and second output signals both include the same amount of fixed pattern noise, subtracting one from the other effectively removes all additive fixed pattern noise from the output of the pixel.

In summary, during normal integrating operation the reset voltage V_res is held high and the pixel output is sampled after a conventional integration period to measure the response of the pixel to the photocurrent. The other sampling operation for calibration follows a short period during which the reference voltage increases quickly. During this time the reset voltage is reduced quickly whilst the reset transistor is conducting to discharge the pixel capacitance. The decreasing voltage V_pcaι emulates the pixel response to a photocurrent. Since each pixel is stimulated by the same reset voltage V_pcaι the variations in the response of different pixels to this signal will represent the fixed pattern noise, including the contribution due to variations in the threshold voltage of the disconnect switch 8. An image formed from the difference between the two outputs Vj, F_? from each pixel will therefore be corrected for the dominant form of fixed pattern noise in these pixels.

The effect of this procedure on the pixel voltage is shown in Figure 4. The broken line denotes the reset voltage V_res, the dotted line denotes the readout voltage V_s at the readout circuit 10_, and the solid curve represents the reference voltage V_ref. As illustrated in the figure, the readout voltage V_s stops decreasing whenever it reaches a threshold voltage V_th higher than the V_ref. This shows that the output voltages sampled after integrating the photocurrent (to obtain Vj) and after decreasing the reset voltage {to obtain Vi) both depend upon the threshold voltage of the disconnect switch 8. Creating an image from the difference between these two outputs removes the additive part of fixed pattern noise, including variations in the threshold voltage V_th of the disconnect switch 8. SHARED TRANSISTORS

Another important aspect of the design of pixels is the reduction of their area whilst maintaining sufficient light sensitivity. This can be achieved by sharing the readout transistors between different pixels. A possible scheme for sharing transistors between four transistors is shown in figure 5.

In the arrangement shown in figure 5, the photosensor devices 4 making up the image sensor are arranged in a number of groups 40, each group containing four photosensors 4 (PD-I, PD-2, PD-3, PD-4). One such group 40 is marked by a broken line in Fig. 5. Each photosensor 4 is associated with a PMOS disconnect switch 8 (Ml, M4, M5, M8), the gate terminal of which is connected to a reference signal line 42, and a read transistor 44 (M2, M3, M6, M7), the gate terminal of which is connected to either a first read line 46a or a second read line 46b. The first two read transistors 44 (M2, M3) are connected to a first float diffusion connection 48 (FD-I), and the second two read transistors 44 (M6, M7) are connected to a second float diffusion connection 48' (FD-2). Each group 40 also includes a reset transistor 12 (M9), the gate terminal of which is connected to a reset line 50,50', a source follower transistor 20 (MlO), the gate terminal of which is connected to a respective float diffusion connection 48,48', and a select transistor 22 (Ml 1), the drain terminal of which is connected to an output line 54a,54b. These three transistors are shared between two of the photosensors (PD-I, PD-2) within the group 40 and two photosensors (not shown) in an adjacent group.

In the first step of pixel operation, to reset the photodiode 4 and convert the signal lines 54a,54b (Outl and Out2) to a reference level, short pulses are supplied via the read lines 46a,46b to the read transistors 44 (M2 and M6), via the first select line 52 to the select transistor 22 (Ml 1), and via the reset line 50 to the reset transistor 12 (M9). In the second step, the light signal from the photosensor 4 is integrated with the reference signal Vref applied from the reference line 42 to the PMOS disconnect switch 8. In the third step, the readl and select 1 lines 46a, 52 are again pulsed to read out the output levels of the first pair of photosensors 4 (PD-I and PD-3). From the fourth step to the sixth step, the voltages in the second pair of photosensors 4 (PD-2 and PD-4) are readout by the same principle of operation. Thus, within the pixel unit consisting of the group of four photodiodes, the signal read-out operation is made by transferring photo-charges from a pair of the photodiodes in the odd columns, followed by transferring photo-charges from another pair of the photodiodes in the even columns. It is noted that, for reading the charges of the third and fourth photosensors 4 (PD-3 and PD-4), the floating diffusion connection 48' (FD-2) located in the neighbouring group is used. In addition to increasing the dynamic range, decreasing the pixel size is an important requirement of current CMOS image sensor design for high resolution applications, hi the conventional 4T configuration, each pixel contains a photodiode, a reset, transistor, a transfer transistor, a source follower transistor and a select transistor. An intuitive way to reduce the pixel size is to use a more advanced manufacturing process with smaller transistor sizes. However, the demand for high aperture ratio makes it impractical to go beyond 0.18μm process unless some process modifications are adopted, such as reducing the metal layers and dielectric thickness. In order to minimize the pixel area occupied by transistors while keeping the photo-sensing area without changing the process, the shared- transistor concept is adopted. The approach is to share the peripheral transistors, reset transistor and readout circuits among a group of pixels (normally 4). Different ways of sharing the transistors are possible and minimum pixel size of 2*2μm² can be achieved with 1.5 transistors per pixel.

In summary, the PMOS disconnect switch 8 is not shared by different pixels. A transfer transistor 44 is adopted next to the disconnect switch 8 in order to isolate the output signal from the float diffusion connection 48. The reset transistor 12, source follower transistor 20 and select transistor 22 are shared by four pixels. It is noted that the first floating diffusion connection 48 (FD-I) is shared by two photosensors 4 (PD-I, PD-2) from the first group and by two photodiodes located in the pixel group one column above. Thus, with eleven transistors per four pixels, 2.75 transistors per pixel are used in this new configuration. As a result, a reduced pixel size of 2^χ2μm² can be realized.

The image sensor is suitable for use in various applications, including in particular cameras (for still and video images) designed for use in uncontrolled lighting conditions, or conditions with a very wide dynamic range of luminance. These may include for example traffic monitoring cameras, security cameras, number plate identification cameras and night-vision cameras for use in cars. The image sensor may also be useful for various robotic applications, for example for machine vision.

Claims

1. An image sensor for an electronic imaging device, the image sensor including an array of pixels, each pixel including: a switch device having a first switch input for a first input signal Vi_nI, a second switch input for a second input signal V_h2, and a switch output for a switch output signal V_s that is switchably connected to the first switch input, the switch device being constructed and arranged to disconnect the switch output from the first switch input at a capture moment determined by comparing the first input signal Vi_n i and the second input signal V₁₁₁₂, a photosensor device for detecting incident light, said photosensor having a photosensor output for a photosensor signal V_p that represents a time integral of the detected light intensity, said photosensor output being connected to said first switch input, a first signal generator that is constructed and arranged to generate alternately a reference signal V_ref and a reference calibration signal V_refcai, said first signal generator being connected to said second switch input, a second signal generator constructed and arranged to generate a photosensor reference signal V_pcai, said second signal generator being switchably connected to said first switch input, and a readout circuit that is arranged to capture the switch output signal V_s at the capture moment and provide an output signal that is related to the switch signal V_s.

2. An image sensor according to claim 1, wherein the first signal generator is constructed and arranged to generate alternately a reference signal V_ref and a reference calibration signal V_refcai.

3. An image sensor according to claim 2, the pixel being constructed and arranged such that, in use, it carries out separate integration and calibration steps, wherein during the integration step, the first switch input receives the photosensor signal V_p, the second switch input receives the reference signal V_ref, and the readout circuit captures a first switch output signal V_si, and during the calibration step, the first switch input receives the photosensor calibration signal V_pcai, the second switch input receives the reference calibration signal V_refcai, and the readout circuit captures a second switch output signal V_s2, the amount of light falling on the photosensor being determined from the difference between the first switch output signal V₅₁ and the second switch output signal V_s2.

4. An image sensor according to claim 3, wherein the photosensor calibration signal V_pcai comprises a voltage that changes at a uniform rate.

5. An image sensor according to claim 3 or claim 4, wherein the reference signal V_ref and the reference calibration signal V_refCai are similar in form.

6. An image sensor according to claim 3, wherein the photosensor calibration signal V_pcai comprises a constant voltage.

7. An image sensor according to any one of the preceding claims, wherein the switch device is constructed and arranged to disconnect the switch output from the first switch input when the difference between first input signal Vi_nI and the second input signal V^₂ equals a switch threshold voltage V_tn.

8. An image sensor for an electronic imaging device, the image sensor including an array of pixels, each pixel including: a switch device having a first switch input for a first input signal Vi_n i, a second switch input for a second input signal Vi_n2, and a switch output for a switch output signal V_s that is switchably connected to the first switch input, the switch device being constructed and arranged to disconnect the switch output from the first switch input at a capture moment determined by comparing the first input signal Vi_nI and the second input signal V_L12, a photosensor device for detecting incident light, said photosensor having a photosensor output for a photosensor signal V_p that represents a time integral of the detected light intensity, said photosensor output being connected to said first switch input, a signal generator that is constructed and arranged to generate a reference signal V_ref, said signal generator being connected to said second switch input, and a readout circuit that is arranged to capture the switch output signal V_s at the capture moment and provide an output signal that is related to the switch signal V_s; characterised in that said pixels are arranged in a plurality of groups where each group includes a plurality of pixels, and wherein at least two pixels within said -group share a common readout circuit.

9. An image sensor according to claim 8, wherein said common readout circuit is also shared with one or more pixels in another group.

10. An image sensor according to claim 8 or claim 9, wherein each group includes a plurality of read switches, for selecting which pixel is connected to the readout circuit.

11. An image sensor according to any one of claims 8 to 10, wherein each group includes a reset switch, and wherein at least two pixels within said group share said reset switch.

12. An image sensor for an electronic imaging device, the image sensor including an array of pixels, each pixel including: a switch device having a first switch input for a first input signal V_h11, a second switch input for a second input signal V^, and a switch output for a switch output signal V₅ that is switchably connected to the first switch input, the switch device being constructed and arranged to disconnect the switch output from the first switch input at a capture moment determined by comparing the first input signal Vj_nI and the second input signal V^, a photosensor device for detecting incident light, said photosensor having a photosensor output for a photosensor signal V_p that represents a time integral of the detected light intensity, said photosensor output being connected to said first switch input, a signal generator that is constructed and arranged to generate a reference signal V_ref, said signal generator being connected to said second switch input, and a readout circuit that is arranged to capture the switch output signal V_s at the capture moment and provide an output signal that is related to the switch signal V₅; characterised in that said reference signal V_ref varies according to a predetermined function V_ref (t) that may be selected from a plurality of different functions.

13. An image sensor according to claim 12, including selecting means for selecting a functions from said plurality of different functions.

14. An image sensor according to claim 12 or claim 13, wherein the signal generator is programmable with different predetermined functions V_ref (t).

15. An image sensor according to any one of claims 12 to 14, wherein at least one of said different functions is such that the output signal is related to the intensity of light incident on the photosensor device by a function that is logarithmic over one portion of the pixel's dynamic range and non-logarithmic over another portion of the pixel's dynamic range.

16. An image sensor according to any one of claims 12 to 15, wherein said predetermined function V_ref(t) is defined to create a desired response R by creating a pixel with an effective integration time t_eff, where C is the capacitance of the pixel and

17. An image sensor according to any one of claims 12 to 15, wherein said predetermined function V_ref(t) is defined to create a desired response R by creating a pixel with an effective integration time tφ where C is the capacitance of the pixel, I_add is an additional current discharging the pixel and

18. A method of sensing images using an electronic imaging device having an image sensor that includes an array of pixels, the method comprising: performing an integration step by detecting incident light with a photosensor and providing a photosensor signal V_p, generating a reference signal V_ref, comparing the photosensor signal and the reference signal, determining a capture moment from the comparison and capturing a first output signal V_si, performing a calibration step by generating a photosensor calibration signal V_pcai and a reference calibration signal V_refcai, comparing the photosensor calibration signal with the reference calibration signal, determining a capture moment from the comparison and capturing a second output signal V_s2, and deriving a pixel output signal from the difference between the first output signal V_sl and the second output signal V_s2.

19. A method according to claim 18, in which the photosensor calibration signal Vp_03I comprises a voltage that changes at a uniform rate.

20. A method according to claim 18 or claim 19, wherein the reference signal V_ref and the reference calibration signal V_refc_ai are similar in form.

21. A method according to claim 18, in which the photosensor calibration signal V_pcai comprises a constant voltage.

22. A method of sensing images using an electronic imaging device having an image sensor that includes an array of pixels, the method comprising: detecting incident light with a photosensor and providing a photosensor signal V_p, generating a reference signal V_ref, comparing the photosensor signal and the reference signal, determining a capture moment from the comparison and capturing an output signal; characterised in that said pixels are arranged in a plurality of groups where each group includes a plurality of pixels, and wherein the output signals of at least two pixels within said group are captured by a common readout circuit.

23. A method according to claim 22, wherein said common readout circuit also captures the output signals of one or more pixels in another group.

24. A method according to claim 22 or claim 23, wherein said pixels are connected sequentially to said common readout circuit.

-25. A method according to any one of claims 22 to 24, wherein at least two pixels within said group are reset by a common reset switch.

26. A method of sensing images using an electronic imaging device having an image sensor that includes an array of pixels, the method comprising: detecting incident light with a photosensor and providing a photosensor signal V_p, generating a reference signal V₁^_f, comparing the photosensor signal and the reference signal, determining a capture moment from the comparison and capturing an output signal; characterised in that said reference signal V_ref varies according to a predetermined function V_ref (t) that may be selected from a plurality of different functions.

27. A method according to claim 26, wherein at least one of said different functions is such that the output signal is related to the intensity of light incident on the photosensor device by a function that is logarithmic over one portion of the pixel's dynamic range and non-logarithmic over another portion of the pixel's dynamic range.

28. A method according to claim 26 or 27, wherein said predetermined function V_rei(t) is defined to create a desired response R by creating a pixel with an effective integration time t_eff, where C is the capacitance of the pixel and

29. A method according to claim 26 or 27, wherein said predetermined function V_ref(t) is defined to create a desired response R by creating a pixel with an effective integration time t_eff, where C is the capacitance of the pixel, I_add is an additional current discharging the pixel and