A Wide Dynamic Range CMOS Image Sensor with an Adjustable by fdjerue7eeu


More Info
									 A Wide Dynamic Range CMOS Image Sensor with an Adjustable
                   Logarithmic Response
                      Bhaskar Choubey, Hsiu-Yu Cheng and Steve Collins
                   Department of Engineering Science, University of Oxford,
                             Parks Road Oxford, UK, OX1 3PJ

A wide dynamic image CMOS image sensor with a user adjustable logarithmic photo-response is presented.
A pMOS switch and a time-dependent reference voltage are integrated into a three-transistor (3T) pixel
structure to implement a logarithmic response. Several pixels have been manufactured using a 0.25µm
standard CMOS technology. Compared to the conventional logarithmic response pixel based on a
diode-connected transistor, the proposed pixel combines a wide dynamic range of 120dB with much higher
responsivity (250mV/decade) and better dark response.

Keywords : CMOS image sensor, wide dynamic range, logarithmic response, pMOS switch

                                         1. INTRODUCTION

A wide dynamic range CMOS image sensor that can capture a scene containing both bright and dark areas
is highly desirable for applications including automobile driver aids, security cameras and consumer
products. Numerous approaches have been proposed to expand the dynamic range of CMOS image sensors.
Most of these can be divided into one of three principal groups. The first group convert photocurrents into a
time-to-saturation signal by integrating a comparator in each pixel [1-2]. However, this approach increases
the pixel area with the result that these pixels are at a disadvantage when costs must be controlled or
reduced whilst increasing pixel count. The second more evolutionary group samples the photocurrent
several times within one or more integration periods and then synthesizes the wide dynamic range image
[3-6]. The main disadvantage of these systems is the cost of the processing needed to synthesize the final
image. The last group realizes a logarithmic compression of the input photocurrent to the output voltage
using the current-voltage characteristics of MOSFETs working in weak inversion [7-10]. The small
maximum output swing (typically 0.3V) and responsivity (50mV/decade) of these pixels make them
vulnerable to both fixed pattern and temporal noise. Furthermore, although the continuous output available
from these pixels can be an advantage in some applications, it means that these pixels are slow to respond
to a sudden decrease in photocurrent.

The ideal pixel should combine the speed of response of an integrating pixel with the dynamic range
compression of logarithmic pixels. This can be achieved using a comparator within each pixel to vary the
effective integration time of the pixels so that the output voltage is proportional to the logarithm of the
photocurrent [11]. However, the large pixel size and low fill factor resulting from the use of an in-pixel
comparator makes it impractical for most applications. To overcome these problems a novel wide dynamic
range CMOS pixel has been developed that is described for the first time in this paper.

The rest of this paper is organized as follows. In section 2, the pixel and its operation are described. The
design and characterization results of the prototype pixel are presented in section 3. In section 4, the
measurement results of the low dark signal pixel is presented.

                                             2. THE PIXEL

A schematic circuit diagram of the proposed wide dynamic range pixel is shown in Fig. 1. As in the
conventional 3T active pixel, the process of forming an image starts when the voltage in the pixel, Vdiode, is
reset. To obtain an increased voltage swing a pMOS device, Mp1, is preferred as the reset transistor. This
allows the voltage to be reset to the maximum allowed voltage, which for this process is 2.5V. The wide
dynamic range operation is achieved by integrating a second pMOS device, Mp2, into the standard 3T active
pixel. This device is connected between the photodiode and the input to the source follower output circuit
with the gate of Mp2 connected to an externally generated reference voltage, Vref(t). Whilst the pixel voltage
is being reset this reference voltage is held low so that it conducts. This means that the input node to the
source-follower circuit, the storage capacitor Cs, is also reset.

When the reset voltage, rst, suddenly increases to the maximum allowed voltage, Mp1 stops conducting and
the photocurrent (Iph) starts to discharge the pixel capacitance, including the diode capacitance. Once the
reset voltage has been increased so that the photocurrent has started to discharge the pixel capacitance the
reference voltage is also increased. As the voltage Vdiode decreases, this voltage becomes the source voltage
for Mp2. This device therefore has a decreasing source voltage and an increasing gate voltage. Eventually,
these changing voltages cause the channel resistance of Mp2 to increase suddenly and Mp2 stops conducting.
This isolates the gate voltage of the source follower transistor Mn1 which decides the output voltage from
the pixel. The effective integration time for the pixel output is therefore determined by the time at which
Mp2 stops conducting.

                                          Vdd                                      Vdd

                            rst             Mp1                                      Mn1
                                                Vdiode      Mp2           Vs

                                                Iph      CP         Cs       sel     Mn2

Figure 1. A schematic circuit diagram of the proposed pixel including the PMOS reset and switch transistors (Mp1 and
                               Mp2) and a soure follower readout circuit (Mn1 and Mn2).

To understand the effects of the operation of Mp2 consider Fig. 2. This shows a typical time dependant
reference voltage needed to obtain a logarithmic response, the voltage across the photodiode and at the
input to the source-follower for two different photocurrents. Initially transistor Mp2 is conducting in both
cases and the photodiode is connected to the source-follower. The only difference between the responses of
the pixel to the two different photocurrents is that the larger of the two photocurrents is discharging the
pixel more quickly. This means that with this photocurrent the source voltage of Mp2 reaches a threshold
voltage of Mp2 above the gate voltage of this device more quickly. Since this is a pMOS device the further
reduction in the voltage across the photodiode combines with the increase in the reference voltage to drive
Mp2 into weak inversion. Ideally this device will stop conducting when

                                                Vgate(t)-Vsource(t) = – Vth,Mp2

After this condition has occurred the input to the source-follower circuit, which is the pixel output,
corresponds to the voltage in the integrating pixel when the source-follower is isolated from the photodiode.
Fig. 2 makes it clear that this occurs more quickly for a large photocurrent than for a small photocurrent.
The effective integration time is therefore dependant upon both the photocurrent and the user controlled
reference voltage. The relationship between the photocurrent and the pixel output voltage can be varied by
using different reference voltages.



                             rst                     Integration Period

                      Figure 2. Vdiode and Vstop of the proposed pixel when a Vref(t) is applied.

To create a wide dynamic range pixel with a logarithmic response, the aim is to turn off the switch at a time
ts so that the change in output voltage is proportional to the logarithm of photocurrent (Iph). The photodiode
voltage at a time t seconds after integration has started from an initial value of Vdd is

                                                Vdiode = Vdd – Iph· t / Cp                     (1)

where Cp is the pixel capacitance when the photodiode and source-follower are connected. The voltage
isolated at the input to the source-follower will depend on the value of Vdiode when Vdiode = Vref(t) + Vth,Mp2.
To obtain a logarithmic response the change in voltage that has occurred before this condition is reached
should be proportional to the logarithm of the photocurrent in the pixel. This change occurs because the
photocurrent has discharged the pixel capacitance. To obtain a logarithmic output with a responsivity S, the
two parts of the circuit should be isolated at a time ts such that the change is voltage

                                           Iph· ts / Cp = S·Ln ( Iph / Iref )                  (2)

where Iref is a reference current. The source-follower circuit is isolated from the photodiode when

                                       Vref(t) = Vdd – Vth,Mp2 – Iph· t / Cp                   (3)

For this to occur when the change of voltage represents the logarithm of the photocurrent

                                   Vref(t) = Vdd – Vth,Mp2 – S·Ln ( Iph / Iref )               (4)

This can be rearranged to give

                                   Iph = Iref ·exp – [ Vref(t)–Vdd+Vth,Mp2 ]/S                 (5)

Finally, this expression for the photocurrent can be substituted into (3). Ideally the result should be an
expression for the reference voltage as a function of time, however, the equations led to an expression for
time as a function of the reference voltage
                                        Cp{(Vdd–Vref(t)–Vth,Mp2) ·exp[(Vref(t)–Vdd+Vth,Mp2)/S]}
This expression can be evaluated to determine the time at which a particular reference voltage will occur.
Fig.3 shows the reference voltages for two sets of parameters of the logarithmic function calculated using


                    Voltage (V)


                                  0.5                       Vref1(t) for S=0.25, Iref=10

                                                            Vref2(t) for S=0.35, Iref=10

                                        0.000      0.005      0.010        0.015           0.020
                                                            Time (s)
                     Figure 3. The Vref(t)s needed to realize a logarithmic response over a wide
                                        dynamic range with different slopes.

                                            3. EXPERIMENTAL RESULTS


            Figure 4. Layout of the proposed pixel with pixel size of 7.5μm×7.5μm and 45% fill factor.

The proposed pixel has been fabricated using the UMC 0.25μm, 1P4M, 2.5V CMOS process. The resulting
layout of the pixel with pixel size 7.5μm×7.5μm and 45% fill factor is shown in Fig. 4. The pixel size can
be further reduced for higher resolution demands with the trade-off in fill factor. The reference voltages
used were calculated in MATLAB and then supplied to the pixel using the Agilent Intuilink Waveform
Editor and 33250A arbitrary waveform generator. For all the experiments, the time between starting the
integration of the photocurrent and sampling the output voltage was set to 20ms and the pixel output
voltage was monitored by HP 4155B.
                                   1.8                                                         74 pA
                                                                                              530 pA


                        Vout (V)



                                                0.000          0.005          0.010      0.015          0.020
                                                                             Time (s)

              Figure 5. The measured output voltage of two different Iph in the integration time of 20ms

                                    1.5                 Vref(t)=0V
                       Vout (V)




                                          -15            -13           -11       -9      -7        -5            -3
                                     10                 10           10        10       10       10             10
                                                                      Photocurrent (A)
      Figure 6. The measured change in pixel output voltage as a function of the ‘photocurrent’ in the test pixel.
In order to demonstrate the functionality of the proposed design, particularly its very wide dynamic range, a
test pixel was manufactured in which the photodiode was replaced by a MOSFET acting as a voltage
controlled current sink. Each gate voltage needed to sink particular current through the MOSFET was
estimated from Cadence Spectre simulation results. Fig. 5 illustrates the measured output signal from the
pixel under two different photocurrents, which demonstrates that VS can be isolated from the Vdiode at
certain level when Mp2 is switched off. Results in Fig. 6 show that the pixel has a logarithmic response over
a wide dynamic range operation. In particular these results show that by using two different reference
voltages, Vref1(t) and Vref2(t), it is possible to vary the dynamic range of the same pixel, in this case from 96
dB to 137dB. Compared to the response of a conventional logarithmic pixel with a load transistor
operating in weak inversion, the responsivity of the pixel is also significantly enlarged to 250mV/decade
and 350mV/decade. This will make the pixel less vulnerable to noise. Finally, the results in Fig 6 show that
if the reference voltage is held low enough, the pixel acts as a conventional integrating pixel with a linear

The optical response of the pixel with a photodiode was then measured using the OL Series 455
high-intensity integrating sphere with a digital controller. A precision silicon detector-filter combination
with an accurate photonic response is mounted in the sphere wall and monitors the sphere luminance. The
luminance can be varied over six decades without changing the color temperature. The results in Fig. 7
show logarithmic responses with different slopes obtained using two different shaped reference voltages.
The change in output voltage per decade change in illumination intensity was found to be 249mV/decade
and 298mV/decade, respectively. As expected these results are almost identical to the results shown in Fig.
6 that were obtained with the same reference voltages.


                            Voltage (V)



                                          0.0    -1    1           3         5          7
                                                10    10         10         10        10
                                                       Illumination (Lux)

          Figure 7.   The optical measurement result of the pixel with photodiode using Vref1(t) and Vref2(t).

                              4. IMPROVED LOW LIGHT SENSITIVITY

The dark current that flows in the pixel in the absence of light determine the low light sensitivity of the
pixel. This current can be measured using the integrating mode of operation of the pixel when the reference
voltage is held low. In the experiments to measure the dark current the pixel was covered and allowed to
integrate the dark current for 1 seconds. The resulting rate of change of the output voltage was found to be
267mV per second at room temperature. To convert this to an equivalent illumination level the
experiment was repeated with the pixel illuminated with 0.73 lux, as measured using a Sekonic L-508 light
meter. Under these conditions the rate of change of the pixel voltage was 541mV per second. This means
that the dark current for this pixel layout corresponds to 0.36 lux.

                                                                                 Photon Sensing Area

                Figure 8. Layout of the proposed pixel with poly-silicon ring around the photodiode

To improve the low light sensitivity of a pixel, a dark current reduction technique can be adopted [12]. As
illustrated in Fig. 8, a poly-silicon ring is added around the photo-sensing area to keep the photodiode away
from the surface defects results from shallow trench isolation (STI). As expected, the fill factor is slightly
decreased by using this approach (45% to 38%). Fig. 9 shows the measured dark signal of the proposed
pixel with and without the poly-silicon ring. These results show that the rate of change of pixel voltage is
reduced from 267mV/s to 113mV/s by using the poly-silicon ring. This means that the corresponding dark
current has been reduced from 19.2fA to 8.1fA. The measured optical response of the pixel using the
time-variant Vref(t) is illustrated in Fig. 10. The dynamic range of the pixel with the poly-silicon ring is
found to be identical to that of the normal pixel with the improved minimum detectable illumination of 0.15


                      Voltage (V)

                                                     STI isolated
                                    1.5              Normal Pixel

                                          0.0          0.4            0.8        1.2        1.6
                                                                 time (s)
           Figure 9. The measured dark signal of the two proposed pixel structures perated in linear mode

                      Voltage (V)


                                                 Normal Pixel
                                                 STI isolated

                                            -1               1               3          5
                                          10             10                 10         10
                                                  Illumination Intensity (Lux)
                   Figure 10. The measured optical response of the two proposed pixel structures

                                                   5. CONCLUSION

Several methods have been proposed that increase the dynamic range of CMOS pixels. Each of these
methods has different advantages and disadvantages, but they suggest that the best high dynamic range
pixels will match the compactness and speed of response of integrating pixels with the dynamic range
compression of logarithmic pixels. An integrating pixel has been described which can achieve a logarithmic
response by making the effective integration time of each pixel dependant upon its photocurrent. The result
is a pixel that can have a dynamic range of up to 137dB. Furthermore, the response of the pixel is
controlled by a user generated reference voltage. This means that the user can easily change this voltage to
vary the response of the pixel to match a particular application and/or scene. The result is a pixel that seems
particularly well suited to applications, such as driver aids and security cameras, which demand a high
dynamic range.

[1]    Xin Qi, Xiaochuan Guo and Harris, J. G., “A time-to-first spike CMOS imager,” in Proc. of 2004
       IEEE International Symposium on Circuits and Systems, vol. 4, pp. 840-843 May 2004.
[2]    D. Stoppa, Andrea Simoni, and Andrea Baschirotto, “A 120-dB Dynamic Range CMOS Image Sensor
       with Programmable Power Responsivity,” in Proc. of 2006 IEEE European Solid State Circuits, vol.1,
       pp.420-423, Sep. 2006.
[3]    O. Yadid-Pecht and E. Fossum, “Image Sensor with Ultra High Linear Dynamic Range Utilizing Dual
       Output CMOS Active Pixel Sensors,” in IEEE Trans. on Electron Devices, vol. 44, pp.1721-1724,
[4]    O. Schrey, J. Huppertz, G. Filimonovic, A. Bubmann, W. Brockherde and B. J. Hosticka, “A 1K×1K
       High Dynamic Range CMOS Image Sensor With On-Chip Programmable Region-of-Interest
       Readout,” in IEEE Journal of Solid-State Circuits, vol.37, no.7, pp. 911-915, July 2002.
[5]    P. M. Acosta-Serafini, I. Masaki, C. G. Sodini, “A 1/3” VGA Linear Wide Dynamic Range CMOS
       Image Sensor Implementing a Predictive Multiple Sampling Algorithm With Overlapping Integration
       Intervals,” in IEEE Journal of Solid-State Circuits, vol.39, no.9, pp. 1487-1496, Sep. 2004.
[6]    Masaaki Sasaki, Mitsuhito Mase, Shoji Kawahito, and Yoshiaki Tadokoro, “A Wide-Dynamic-Range
       CMOS Image Sensor Based on Multiple Short Exposure-Time Readout With Multiple-Resolution
       Column-Parallel ADC,” in IEEE Sensors Journal, vol.7, no.1, pp. 151-158, Jan. 2007.
[7]    S. Kavadias, B. Dierickx, D. Scheffer, A. Alaerts, D. Uwaerts, and J. Bogarets, “A Logarithmic
       Response CMOS Image Sensor with On-chip Calibration,” in IEEE Journal of Solid-State Circuits,
       vol.35, no.8, pp. 1146-1152, Aug. 2000.
[8]    M. Loose, K. Meier, and J. Shemmel, “A Self Calibrating Single Chip CMOS Camera with
       Logarithmic Response,” in IEEE Journal of Solid-State Circuits, vo36, no.4, pp. 586-596, Apr. 2001.
[9]    L. W. Lai, C. H. Lai and Y. C. King, “A Novel Logarithmic Response CMOS Image Sensor with High
       Output Voltage Swing and In-Pixel Fixed Pattern Noise Reduction,” in IEEE Sensors Journal, vol. 4,
       pp. 122-126, Feb. 2004.
[10]   S. O. Otim, B. Choubey, D. Joseph and S. Collins, “Simplified Fixed Pattern Noise Correction for
       Logarithmic Sensors,” in Proc. of 2005 IEEE Workshop on Charge-Coupled Devices and Advanced
       Image Sensors, Kruizawa, Japan, June 2005.
[11]   H. Y. Cheng, B. Choubey, and S. Collins, "A High-Dynamic-Range Integrating Pixel With an
       Adaptive Logarithmic Response" in IEEE Photonics Technology Letters, Vol. 19, No. 15, Aug.1 2007.
[12]   B. Choubey and S. Collins, "Wide dynamic range CMOS pixels with reduced dark current", in Analog
       Integrated Circuits and Signal Processing, Online First, 2007.

To top