BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to electronic imaging devices and, in particular, to CMOS imagers having a minimum of analog components in each pixel.

2. Description of Related Art There presently exists many alternatives to CCD sensors for generating video or still images. The various schemes can be grouped into two basic classes, depending upon whether signal amplification is performed at each pixel site or in support circuits outside the pixel array. Passive-pixel sensors perform amplification outside the array. Passive pixel sensors exhibit pixel simplicity and maximized optical fill factor.

Active-pixel sensors include an amplifier at each pixel site. Active pixel sensors optimize signal transfer and sensitivity.

The simplest passive pixel comprises a photodiode and an access transistor.

The photo-generated charge is passively transferred from each pixel to downstream circuits.

The integrated charge must, however, be efficiently transferred with low noise and non- uniformity. Since each column of pixels often share a common row or column bus for reading the signal, noise and non-uniformity suppression are typically facilitated in the "column"buffer servicing each bus. One example of a passive pixel implementation is shown in Figure 1. It uses a buffer consisting of a transimpedance amplifier with capacitive

feedback to yield reasonable sensitivity considering the large bus capacitance. Such charge- amplification was not generally practical for on-chip implementation in early MOS imaging sensors. Accordingly, alternative schemes compatible with NMOS technology were used.

The basic scheme shown in Figure 2 was mass-produced by Hitachi for camcorders. The key refinements with respect to the Figure 1 scheme include anti-blooming control and circuitry for reducing fixed pattern noise. Though these imagers were inferior to the emerging charge coupled device (CCD) imagers available at the time, similar MOS imagers are still being offered commercially today.

Subsequent efforts at improving passive-pixel imager performance have also focused on column buffer enhancements. The column buffer was improved by using an enhancement/depletion inverter amplifier to provide reasonably large amplification in a small amount of real estate. Its 40 lux sensitivity was still nearly an order of magnitude below that of competing CCD-based sensors. Others worked to enhance sensitivity and facilitate automatic gain control via charge amplification in the column buffer. More recently, the capacitive-feedback transimpedance amplifier (CTIA) concept of Figure 1 has served as a basis for further development, as exemplified by U. S. Patent Nos. 5,043,820 and 5,345,266. The CTIA is nearly ideal for passive-pixel readout if the problems with temporal noise pickup and fixed-pattern noise are adequately addressed.

Though much progress has been made in developing passive-pixel imagers, their temporal S/N performance is still fundamentally inferior to competing CCD imagers.

Their bus capacitance translates to read noise of ~100 e-. CCDs, on the other hand, typically have read noise of 20 to 40 e-at video frame rates. The allure of producing imagers with conventional MOS fabrication technologies rather than esoteric CCD processes (which usually require many implantation steps and complex interface circuitry) encouraged the development of active-pixel sensors. In order to mitigate the noise associated with the bus capacitance, amplification was added to the pixel via the phototransistor. One such approach called a Base-Stored Image Sensor (BASIS) used a bipolar transistor in emitter follower configuration with a downstream correlated double sample to suppress random and temporal noise. By storing the photogenerated-signal on the

phototransistor's base to provide charge amplification, the minimum scene illumination was reduced to 10-3 lux in a linear sensor array. However, the minimum scene illumination was higher (10-2 lux) in a two-dimensional BASIS imager having 310,000 pixels because the photoresponse non-uniformity was relatively high (52%). These MOS imagers had adequate sensitivity, but their pixel pitch was too large at about 13 llm. This left the problem of shrinking the pixel pitch while also reducing photoresponse non-uniformity.

Since the incorporation of bipolar phototransistors is not strictly compatible with mainstream CMOS processes, some approaches have segregated photodetection and signal amplification. United States Patent Nos. 5,296,696 and 5,083,016, for example, describe active-pixel sensors essentially comprising a three- transistor pixel with photodiode. These implementations still exhibit inadequate performance. The'696 patent, for example, augments the basic source-follower configuration of the'016 patent with a column buffer that cancels fixed pattern noise, but adds a fourth transistor that creates a floating node vulnerable to generation of random offsets for charge-pumping and concomitant charge redistribution. The'016 patent offers a method for reducing offset errors, but not with adequate accuracy and resolution to be useful for competing with CCDs. Furthermore, these and other similar approaches require 3-4 transistors in the pixel (at least one of which is relatively large to minimize 1/f noise) in addition to the photodiode. These implementations also require off-chip signal processing for best S/N performance because none addresses the dominant source of temporal noise. In order to eliminate or greatly suppress the reset noise (kTC) generated by resetting the detector capacitance, a dedicated memory element is usually needed, either on-chip or off-chip, to store the reset voltage, to apply correlated double sampling and coherently subtract the correlated reset noise while the photo-generated voltage is being read.

This basic deficiency was addressed in U. S. Patent No. 5,471,515 by developing an active pixel sensor (APS) that uses intra-pixel charge transfer to store the reset charge at each pixel at the start of each imaging frame. The floating gate APS facilitates correlated double sampling with high efficiency by adding several transistors

and relying on a photogate for signal detection. The concomitant drawbacks, however, are intractable because they increase imager cost. The former adds several transistors to each pixel and several million transistors to each imager thereby reducing production yield. The latter is not compatible with standard CMOS gate fabrication so a non- standard process must be developed. These deficiencies were tackled in U. S. Patent Nos.

5,576,763 and 5,541,402 issued to Ackland et al. and U. S. Patent Nos. 5,587,596 and 5,608,243 issued to Chi et al. Ackland addressed the image lag issues associated with the intra-pixel charge transfer means. But his approach still requires a non-standard CMOS process. Chi reduced pixel complexity by using the simplest possible active pixel comprising only a phototransistor and reset MOSFET. Chi's implementation still suffers from reset noise and compromises spectral response at longer wavelengths because the photodiode is in an n-well.

OBJECTS AND SUMMARY OF THE INVENTION It is an object of the present invention to provide an active-pixel low-noise imaging system for implementation in CMOS or in other semiconductor fabrication technologies.

It is another object of the present invention to provide a low-noise amplifier for an imaging system that efficiently suppresses reset noise.

It is yet another object of this invention to provide an improved electronic imaging system having an integral low-noise amplifier that reduces cost and power consumption while providing low temporal noise and low fixed pattern noise.

These objects and the advantages of the present invention are accomplished by circuitry at each pixel consisting of a photodetector and four transistors. The first transistor serves as the driver of a source follower during signal read and as the driver of a transimpedance amplifier during signal reset to suppress reset noise without having to implement correlated double sampling using either on-chip or off-chip memory. The second transistor is an access MOSFET used to read the signal from each pixel and multiplex the signal outputs from an array of pixels. The third transistor is a MOSFET that

resets the detector after the integrated signal has been read and the detector sense node has effectively been"pinned"by the transimpedance amplifier. The fourth transistor is a MOSFET that acts successively as a switch during signal readout and as a current source during reset. In a typical two-dimensional display, multiplexing can be performed by horizontal and vertical shift registers.

BRIEF DESCRIPTION OF THE DRAWINGS Other objects and the many attendant advantages of this invention will be readily apparent upon reference to the following detailed descriptions when considered in conjunction with the accompanying drawings in which like reference numerals designate like parts throughout the figures, and wherein: Figure 1 is a schematic circuit diagram illustrating the amplifier system for a passive-pixel MOS photodiode array of the prior art; Figure 2 is a schematic circuit diagram illustrating the amplifier system for a passive-pixel imaging sensor of the prior art; Figure 3 is a schematic circuit diagram illustrating the amplifier system for an active-pixel imaging sensor of the prior art; Figure 4 is a schematic circuit diagram illustrating a preferred embodiment of the low-noise active-pixel of the present invention; Figure 5 is a schematic circuit diagram illustrating the operation of the low- noise amplifier system of the present invention during signal reset; Figure 6 is a schematic circuit diagram illustrating the operation of the present invention during readout; Figure 7 is a schematic circuit diagram illustrating a preferred embodiment of the column-based source supply circuit; Figure 8 is a schematic circuit diagram illustrating the small-signal equivalent circuit for the active-pixel sensor of the present invention; Figure 9 is a signal diagram showing representative clocking for the tapered- reset waveform.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Visible imaging systems implemented in CMOS have the potential for significant reductions in cost and power requirements in components such as image sensors, drive electronics, and output signal conditioning electronics. A video camera, for example, can be configured as a single CMOS integrated circuit supported by only an oscillator and a battery. Such a CMOS imaging system requires lower voltages and dissipates less power than a CCD-based system. These improvements translate into smaller camera size, longer battery life, and applicability to many new products.

Because of the advantages offered by CMOS visible imagers, there has been considerable effort to develop active-pixel sensor (APS) devices. Active-pixel sensors can provide low read noise comparable or superior to scientific grade CCD systems. The active circuit in each pixel of an APS device, however, utilizes cell"real estate"that could otherwise be used to enable imagers having optical format compatible with standard lenses and/or to maximize the sensor optical fill factor for high sensitivity. Active-pixel circuits also may increase power dissipation relative to passive-pixel alternatives, increase fixed pattern noise (possibly requiring additional circuitry to suppress the noise), and limit scalability.

The low noise amplifier system of the present invention is formed by the aggregate circuitry in each pixel and the waveform generation circuits servicing the columns and rows of pixels. The signals from the active pixels are read out by a low-noise signal amplification system consisting of the active-pixel, the waveform generators, a standard column buffer and an output buffer. In addition to means for suppressing the detector's reset noise, the column buffer in the downstream electronics often performs correlated double sampling, sample-and-hold, optional video pipelining, and column amplifier offset cancellation functions to suppress the temporal and spatial noise that would otherwise be generated by the column buffer.

The low-noise system of the present invention provides the following key functions: (1) suppression of reset noise without having to provide means for analog memory and signal subtraction for each pixel; (2) global correlated reset to minimize image distribution due to object motion; (3) high sensitivity signal readout via small sensor mode capacitance; (4) adequate amplifier bandwidth to avoid generation of fixed pattern noise due to variations in amplifier time constant and stray capacitance; (5) adequate power supply rejection to enable development of cameras-on-a-chip that do not require elaborate support electronics; and (6) compatibility with application to imaging arrays having pixel pitch as small as 5 microns when using 0.25 llm lithography.

The invention has the advantages of full process compatibility with standard silicided submicron CMOS. This helps maximize yield and minimize die cost because the circuit complexity is distributed amongst the active-pixels and peripheral circuits, and exploits signal processing capability inherent to CMOS. The invention's spectral response is broad from the near-ultraviolet (400 nm) to the near-IR (>800 nm).

Because the low-noise system of the present invention has only four MOSFETs in each pixel, the invention offers as-drawn optical fill factor > 15% at 7 u. m pixel pitch using 0.5 pm design rules in CMOS. The actual optical fill factor is somewhat larger due to lateral collection and the large diffusion length of commercial CMOS processes. A final advantage is the flexibility to collocate digital logic and signal-processing circuits due to its high immunity to electromagnetic interference.

When fully implemented in a desired camera-on-a-chip architecture, the low- noise active pixel sensor (APS) can provide temporal read noise as low as 15 e- (at data rates compatible with either video imaging or still photography via electronic means), fixed pattern noise significantly below 0.1% of the maximum signal (on a par with competing CCD imagers), <0.5% non-linearity, >1 V signal swing for 3.3 V power supply, large charge-handling capacity, and variable sensitivity using simple serial interface updated on a frame-by-frame basis via digital interface to a host microprocessor.

A prototype embodiment of the low-noise APS invention formed a visible imager comprising an array of 1032 (columns) by 776 (rows) of visible light detectors

(photodetectors). The rows and columns of pixels were spaced 7 microns center-to-center using standard 0.5 u. m design rules to provide 15% as-drawn optical fill factor. Subsequent layouts using 0.25 u. m rules show that the invention can also provide similar fill factor at 5 u. m pitch. Several columns and rows of detectors at the perimeter of the light-sensitive region were covered with metal and used to establish the dark level for off-chip signal processing. In addition, the detectors in each row were covered with color filters to produce color imagers. For example, the odd rows may begin at the left with red, green, then blue filters, and the even rows may begin with blue, red, then green filters, with these patterns repeating to fill the respective rows.

A low-noise active-pixel sensor 10 according to the present invention is illustrated in Figure 4. Each pixel 10 in a sensor array (not shown) comprises a photodetector 12, such as a photodiode, for example, connected to the gate of a dual-driver MOSFET 14, and one leg of a reset MOSFET 16. The other leg of MOSFET 16 is connected to a leg of MOSFET 14 and a leg of MOSFET 20. MOSFET 20 acts as a current source during global reset and as a switch during pixel readout. A row select MOSFET 18 has one leg connected to MOSFET 14 and the other leg connected to columns bus 24.

Column bus 24 connects all the pixels in a column of the photodetector array by way of the row select MOSFET 18 to a source supply 30. Row bus 22 connects all the pixel resets in a row to an access supply Vdd. Tapered reset supply 50 supplies an optimized active-pixel reset waveform (Figure 9) to the gate of MOSFET 16. Photodiode 12 may be a substrate diode, for example, with the silicide cleared. In this embodiment, it is necessary to clear the silicide because it is opaque to visible light. Pixel 10 is designed as simply as possible to obtain the largest available light detecting area while providing broad spectral response, control of blooming and signal integration time, and compatibility with CMOS production processes.

For maximum compatibility with standard submicron CMOS processes, photodiode 12 may be formed at the same time as the lightly doped drain (LDD) implant of n-type MOSFETs for the chosen process; this creates an n-on-p photodiode junction in the p-type substrate. Since no additional ion implantation is necessary, the process and wafer

cost for active-pixel circuit 10 are the same as those of standard, high volume digital electronic products.

Figure 5 illustrates operation of the circuit of Figure 4 during pixel reset.

Figure 6 illustrates the operation during readout. In the preferred embodiment, all the photodetectors 12 in the array are reset globally to synchronize the start of image capture.

Reset is initiated by fully enabling the row select MOSFETs 18 of the pixels in the selected row, thereby connecting a low-impedance voltage source (located in source supply 30) to one leg of MOSFET 14 for all the pixels in the row. Dual purpose MOSFET 20 is biased as a current source by waveform Vbias on gate 26 so that all the pixel amplifiers in the imager are configured as transimpedance amplifiers with capacitive feedback provided by MOSFET'S 14 Miller capacitance. MOSFET 14 thus acts as a transconductance, and reset MOSFET 16 acts as a resistance controlled by the tapered reset supply 50. The series resistance of MOSFET 16 is gradually increased by applying a decreasing ramp waveform (Figure 9) to the gate of MOSFET 16 to give the feedback transconductance of MOSFET 14 the opportunity to null the reset noise (kTC) via feedback.

This active-pixel implementation resets within an aperture of tens of microseconds using standard 0.5 micron CMOS technology.

Figure 6 shows the same pixel circuit configured in readout mode. The signals from photodetectors 12 are subsequently read out after the prescribed integration time, one row at a time, from bottom to the top of the array. Within each row, photodetectors 12 are read out from left to right. Readout is initiated by turning on the access MOSFETs 18 of all the photodetectors 12 in a selected row and fully turning on MOSFET 20 so that the upper leg of MOSFET 14 is now connected via row bus 22 to low- impedance voltage source Vdd. The lower leg of MOSFET 14 is connected to a current source at the periphery via column bus 24. MOSFET 14 is now a source follower driver so that the amplified signal from each row-selected photodiode 12 is efficiently transferred to column bus 24.

Figure 7 shows a preferred embodiment for source supply 30. Supply <BR> voltage VSRC is buffered by a unity gain amplifier 44. To facilitate configuring the active-

pixels as transimpedance amplifiers for low-noise reset, the gate 42 of transistor 46 is pulsed by the supply voltage to fully turn on transistor 46 and connect VSRC to the output column bus 24. To facilitate active-pixel readout, the gate 42 of transistor 46 is pulsed low to open- circuit transistor 46. Current source 48 then supplies the active-pixels with supply current I$RC which is established by a constant voltage Vnbias.

The application of the tapered reset waveform (Figure 9) to the transimpedance amplifier enables the reset noise (kTC noise) envelope to decay before the reset switch 16 (Figure 4) is completely opened. The invention also reduces the fixed- pattern offsets from MOSFET 14 in each pixel because the photodiode node charges to a voltage that cancels MOSFET 14 variations from pixel-to-pixel. By using a tapered reset, a row is resettable to within several tens of microseconds for full noise suppression, or a shorter time for moderate noise reduction.

Figure 8 shows the generalized small-signal equivalent circuit model for the pixel 10 during reset. This circuit allows calculation of the steady-state noise envelope at the reset node depending on reset switch resistance, Rw. If the reset voltage is ramped down too slowly, too much time is needed to reset each row and operation at video frame rates becomes problematic. Application to digital still cameras, however, is still quite feasible. If the tapered-reset waveform is ramped down too quickly, then the kTC noise envelope will not decay sufficiently to suppress reset noise before the switch is completely opened.

Figure 8 shows the photodiode 12 node as having a voltage Vl and capacitance Cl to ground. The amplifier 14 output node has voltage V2, output capacitance Co and output conductance Go to ground. The capacitance Co is associated with the entire reset access bus, most of which comes from the MOSFET 14-MOSFET 16 junction of all the rows. The transconductance of MOSFET 14, possibly degenerated by MOSFET 18, is shown as a controlled current source gmVt. The feedback capacitance, Cfb, is the parasitic Miller capacitance of MOSFET 14. Noise from MOSFET 14 is represented by current source in. Noise from MOSFET 16 (which is operated in the ohmic region) is represented by voltage source Vn. Not included in this simplified model is the noise from capacitive feedthrough of the tapered-reset waveform.

Using the small-signal equivalent circuit, a simplified noise formula can be derivedsince: n-4kT g. ; vn2=4kTRsw Assuming that the amplifier's dc gain, Adc, is much greater than 1, then the rms reset noise is: RswG0C1 RswgmCfb<BR> =andk2=wherek1 <BR> <BR> C0 C0+C1C1 The tapered-clock waveform's time constant is thus appropriately selected so that the dimensionless quantity (kl + k2) is significantly >1. The reset noise is hence reduced to the much smaller quantity stemming from the transconductance amplifier's feedback capacitance. In the present invention, this feedback capacitance is the parasitic Miller capacitance of MOSFET 14.

A preferred embodiment of the present invention has the approximate design values: 1000x700 format, 7 µm x 7 ure pixel, gm=20 umho; Go=0.08 umho, Adc=300; Cl=15 fF; Co=3.0 pF and Cfb=0. 3fF. The desired tapered-clock frequency of 25 kHz that is fully compatible with video rate operation hence requires Rsw=50 G# and an optimum tapered-clock time constant of 25 µs. This yields kl+k2=58 for the preferred embodiment, and an equivalent noise capacitance of 1.8 fF. Since the nominal detector capacitance is 15 fF and kTC noise is proportional to the square root of the relevant capacitance, the reset noise is suppressed from about 55 e-to only 14 e-.

Rsw must be tailored to support any changes in line rate. Increasing the line rate hence requires lower switch resistance. The table numerically illustrates the impact on reset noise as the tapered clock time constant is appropriately shortened. At a time constant of 2.7 gsec, the reset noise degrades to 55 e-.

Impact on Reset noise for Preferred Embodiment Rsw(GQ) 5020105210. 50.1 kl+k2 58 23. 2 11. 6 5. 8 2. 32 1. 16 0. 58 0. 12 Reset Noise (e-) 14 17 21 26 35 41 47 55 T (sec) 25 25 24 22 18 14 9. 5 2.7

The column bus 24 is preferably monitored by a standard column buffer to read the video signal when it is available. The key requirements on the column buffer are similar to conventional designs having to handle voltage-mode signals and are well known in the art.

The reset clock signal (Figure 9), for reset circuit 10, and the clocking of source supply 30 (*Figure 7) which facilitate active-pixel reset and readout, is generated on- chip using standard CMOS digital logic. This digital logic scheme thus enables "windowing,"wherein a user can read out the imager in various formats simply by enabling the appropriate support logic to clock the appropriate subformat. With windowing, the 1032 x 776 format of the prototype embodiment can be read out as one or more arbitrarily sized and positioned M x N arrays without having to read out the entire array. For example, a user might desire to change a computer-compatible"VGA"format (i. e., approximately 640 x 480) to either Common Interface Format (CIF; nominally 352 x 240) or Quarter Common Interface Format (QCIF; nominally 176 x 120) without having to read out all the pixels in the entire array. This feature simplifies support electronics to reduce cost and match the needs of the particular communication medium. As an example, a personal teleconference link to a remote user having only QCIF capability could be optimized to provide QCIF resolution and thus reduce bandwidth requirements throughout the teleconference link. As a further example, an imager configured on Common Interface Format (CIF) could provide full-CIF images while supplying windowed information for the portions of the image having the highest interest for signal processing and data compression.

During teleconferencing the window around a person's mouth (for example) could be

supplied more frequently than the entire CIF image. This scheme would reduce bandwidth requirements throughout the conference link.

Although the present invention has been described with respect to specific embodiments thereof, various changes and modifications can be carried out by those skilled in the art without departing from the scope of the invention. Therefore, it is intended that the present invention encompass such changes and modifications as fall within the scope of the appended claims.