Focal Plane Readout Unit Cell Background Suppression Circuit and Method

1. Field of the Invention

This invention relates to focal plane arrays, and particularly to focal plane readout unit cells.

2. Description of the Related Art

A focal plane array is made from an array of photodetectors that are coupled to respective "readout unit cells". The unit cells integrate the photocurrent produced by their respective photodetectors over a specific integration period, and the integrated currents are multiplexed and amplified to produce a single video output.

The unit cells are typically arranged into an array and integrated together on a common substrate. A capacitor within each unit cell performs the photocurrent integration, which is necessary to reduce the effect of photocurrent shot noise and improve signal-to-noise (S/N) ratio. Integrated capacitors, however, tend to have small capacitance values due to the small surface area typically allotted them. The small integrated capacitors have a very limited charge capacity, and thus may reach maximum charge before the integration period is over. To prevent the loss of signal information, the integration period must be kept short, which increases the effect of photocurrent shot noise and lowers S/N ratio.

The photocurrent output of a photodetector includes an inherent "background" component, with the signal information of interest riding on top of the background component. As such, some of the integrating capacitor's limited charge capacity is consumed by the unwanted background component. This background component is particularly large in the photocurrent produced by infrared-sensitive photodiodes, and the limited capacity of integrated unit cell capacitors is particularly troublesome for infrared focal plane arrays.

Several approaches to circumventing the limited charge capacity problem involve "skimming" some of the charge from the integrating capacitor. That is, some of the charge accumulating on the capacitor during an integration period is removed so that the capacitor can accommodate a higher maximum photocurrent, with the skimmed charge ideally comprising just the unwanted background component. One such approach involves connecting a MOSFET current source to the integrating capacitor. The MOSFET is operated continuously throughout each integration period to bleed off charge from the capacitor. There are several problems with this approach, however: with charge being withdrawn over the full integration period, typically about 8 ms, the MOSFET only need carry a very small current. To prevent the quantity of charge withdrawn from exceeding that due to the background component, the MOSFET must be operated in its weak inversion mode, where the drain current is exponentially related to the effective gate drive--i.e., the gate voltage minus the MOSFET's threshold voltage (V.sub.gs -V.sub.th). However, due to processing variations, there is some variability in the threshold voltages of the focal plane array's charge skimming MOSFETs. Because of the exponential relationship, the threshold variability can cause the amount of charge skimmed to vary widely from cell to cell, rendering the technique virtually unusable. Furthermore, the low MOSFET transconductance and the long charge skimming period serve to increase the amount of noise contributed by each MOSFET.

Another approach is described in U.S. Pat. 5,128,534 to Wyles, et al. Conventionally, one terminal of the integrating capacitor is connected to the photocurrent and the other terminal connected to ground. The Wyles circuit takes the terminal which is normally grounded and connects a ramping voltage to it instead. This voltage is ramped up for the duration of the integration period, reducing the voltage across the capacitor and allowing a greater amount of photocurrent to be integrated.

The Wyles approach suffers from several drawbacks. The integrating capacitors must be able to operate with both plates electrically accessible, so that the voltage ramp can be applied. This effectively excludes the use of r a MOSFET configured as an inversion capacitor, since the ramp voltage can bias the MOSFET below its threshold voltage. A capacitor made from two polysilicon layers above the substrate can be employed, but the thick oxide dielectric layer required by this type of capacitor lowers its capacitance per unit area. The preferred capacitor implementation for Wyles' circuit uses a gate electrode and a substrate diffusion as plates, with gate oxide as a dielectric, but fabricating such a capacitor requires special processing and significantly reduces yield.

A background suppression circuit and method are presented, suitable for use with a focal plane readout unit cell, which solve many of the problems noted above. A charge skimming operation having a duration much shorter than an integration period reduces the quantity of charge stored on an integrating capacitor by a highly predictable amount. No special fabrication processes are required, and very little noise is introduced.

A transistor, preferably a MOSFET, is connected to an integration capacitor from which charge is to be removed. During each integration period, the MOSFET's gate is driven with at least one "skimming pulse", which briefly turns on the transistor. While on, the MOSFET provides a conductive path between the capacitor and a current sink through which stored charge is removed from the capacitor. The skimming pulse has two unique characteristics: 1) it is of short duration--significantly shorter than an integration period--which reduces the amount of noise contributed by the MOSFET's noise mechanisms; and 2) it has an amplitude great enough to drive the MOSFET into its strong inversion mode, which makes the quantity of removed charge relatively insensitive to variations in threshold voltage.

The background suppression circuit and method described herein do not require access to both capacitor terminals, and thus allow the use of MOSFETs as inversion capacitors. The charge skimming MOSFETs can be used to serve multiple purposes: with appropriate driving circuitry, each MOSFET can sequentially reset its respective integrating capacitor to an initial known state, skim a fixed amount of charge from the capacitor during each integration period, and read out the charge on the capacitor at the end of the integration period.

The charge skimming pulse is preferably timed to be applied just as the capacitor nears its maximum charge capacity, and is arranged to reduce the charge on the capacitor almost, but not quite, to zero--so that the entire integration period remains utilized.

Variations described herein include applying more than one skimming pulse per integration period to enable additional charge can be skimmed, and skimming while briefly interrupting integration to reduce noise that might otherwise be coupled into the signal path.

Further features and advantages of the invention will be apparent to those skilled in the art from the following detailed description, taken together with accompanying drawings.

FIG. 1 is a schematic diagram of a background suppression circuit per the present invention.

FIG. 2a is a graph plotting charge storage versus time when charge skimming is performed per the present invention.

FIG. 2b is a graph plotting charge storage versus time when two charge skimming events are performed per integration period.

FIG. 3 is a schematic diagram illustrating the use of a charge-skimming transistor per the present invention to perform additional functions.

FIG. 4 is a block diagram of an N.times.N array of unit cells which each include a background suppression circuit per the present invention.

A background suppression circuit suitable for use with a focal plane readout unit cell is shown in FIG. 1. A photodetector 10, typically a photodiode, has its cathode connected to a DC bias voltage V.sub.bias. When light to which it is sensitive impinges on the photodetector, it produces a DC photocurrent I.sub.ph at its anode. The photocurrent I.sub.ph is preferably buffered with a detector interface circuit 12, which preferably presents a low impedance to the photocurrent at its input 14, and produces a signal I.sub.ph1 representing photocurrent I.sub.ph at a high impedance output 16.

A capacitor C.sub.int receives current I.sub.ph, at a terminal 18, with the capacitor's other terminal tied to a fixed voltage, typically ground. Also connected to terminal 18 is a reset switch 20, which is closed to set the voltage on C.sub.int to an initial value V.sub.reset. For a positive photocurrent, V.sub.reset is preferably set to the most negative power supply voltage available. Switch 20 is typically controlled with a logic signal .phi..sub.reset. C.sub.int and detector interface circuit 12 make up part of a "unit cell", a number of which are arranged in an array to form an imaging device.

In practice, the capacitor C.sub.int is first set to V.sub.reset. Switch 20 is then opened to begin an "integration period", during which incoming current I.sub.ph1 causes a charge q.sub.cap to build up on C.sub.int ; i.e., C.sub.int performs an integration on current I.sub.ph1 by converting it to charge q.sub.cap. An integration period ends when the flow of photocurrent to C.sub.int stopped. At the end of an integration period, the charges stored on C.sub.int and on the other capacitors in the array are read out and multiplexed together to create a video frame.

The photocurrent is integrated over time to reduce the effect of photocurrent shot noise, with longer integration periods providing more noise reduction than shorter ones. A problem arises, however, due to the physical constraints placed on the size of the capacitor C.sub.int. Focal plane array unit cells are preferably integrated, and the amount of wafer surface area allotted to capacitors is typically limited. The resulting small capacitors have a correspondingly small maximum charge capacity q.sub.limit. If q.sub.cap reaches q.sub.limit before the end of an integration period, any current I.sub.ph coming to C.sub.int after q.sub.limit is exceeded is not integrated, and its signal information is lost. Thus, unless otherwise countered, the constraints on C.sub.int generally force integration periods to be shorter than would be preferred for purposes of shot noise reduction.

The graph in FIG. 2a plots the integrated photocurrent .intg.I.sub.ph .multidot.dt (charge) stored on an integrating capacitor with respect to time. The photocurrent I.sub.ph generated by photodetector 10 includes a significant "background" component, labeled i.sub.bkg, on top of which rides the signal of interest. The current labeled i.sub.max includes both background and signal components, and represents the maximum current that can be generated. The background component is integrated along with the signal of interest and consumes some of the integrating capacitor's charge capacity. The graph covers one integration period, labeled .tau..sub.int. If the integrating capacitor had sufficient capacity, integrating i.sub.max and i.sub.bkg over one entire integration period would result in a stored charge of between q.sub.max and q.sub.min.

An integrating capacitor's charge capacity is limited, however. As seen in FIG. 2a, the maximum charge capacity q.sub.limit of the capacitor can potentially be reached at a time t.sub.sk, prior to the end of integration period .tau..sub.int. When this occurs, photocurrent received at the capacitor between t.sub.sk and .tau..sub.int is not integrated, and the signal information contained therein is lost. The invention solves this problem by "skimming" some of the charge from the capacitor during the integration period, i.e., reducing the quantity of charge stored on the capacitor, to prevent q.sub.limit from being reached prior to the end of the integration period and thus allowing more photocurrent to be integrated than could otherwise be accommodated. Ideally, a precisely known amount of charge is removed each time a skim is performed, without introducing any noise into the signal path. In FIG. 2a, charge is skimmed at time t.sub.sk, with a quantity of charge q.sub.sk being removed. Charge quantity q.sub.sk is preferably equal to that amount of charge reduction that will reduce i.sub.bkg almost, but not quite, to zero. Current i.sub.bkg is preferably not reduced to zero to insure that the integrated background noise is retained; if reduced to zero, the effective integration time for i.sub.bkg is reduced to .tau..sub.int --t.sub.sk, which would result in a higher shot noise level. Once charge q.sub.sk has been removed, integration of the photocurrent can continue over the entire integration period without the capacitor's charge capacity q.sub.limit being reached, with the amount of charge stored on Cift at time .tau..sub.int being between q.sub.sigmin and q.sub.sigmax.

Implementing the invention as described herein nearly achieves the ideal charge skimming performance noted above. Referring back to FIG. 1, a transistor Q1, preferably a MOSFET, is connected as a controlled current source between integrating capacitor terminal 18 and a current sink terminal at a voltage V-, which can include ground. To perform a charge skim operation, a "charge skimming pulse" V.sub.sk is applied to Q1's control input. The pulse briefly turns Q1 on, providing a conductive path between C.sub.int and V- through which a controlled amount of charge stored on C.sub.int is removed and dumped to the current sink.

To fully realize the invention's advantages, charge-skimming pulse V.sub.sk is required to have particular characteristics. Due to manufacturing processing variations, some variation will be found between the respective threshold voltages of an array of integrated MOSFETs such as those used to perform the charge-skimming. These V.sub.th variations affect the amount of charge removed from an integrating capacitor for a given skimming pulse. To minimize the impact of V.sub.th variations, MOSFET Q1 is made to operate in its strong inversion mode. In strong inversion, Q1's drain current is proportional to (V.sub.gs -V.sub.th).sup.2 rather than the exponential relationship present when operated in weak inversion, making the quantity of removed charge relatively insensitive to variations in threshold voltage. This is accomplished by supplying a skimming pulse V.sub.sk with an amplitude sufficient to drive Q1 into strong inversion.

With Q1 in strong inversion, all of the charge on C.sub.int can be quickly removed. But as noted above, it is preferable that the charge removed (q.sub.sk) not reduce the charge on C.sub.int to zero. This is prevented by making the skimming pulse V.sub.sk substantially shorter than an integration period. A short skimming pulse limits the amount of charge removed, and also reduces the amount of noise contributed by Q1's noise mechanisms. Thus, a skimming pulse having a duration substantially shorter than an integration period and an amplitude sufficient to drive a MOSFET into strong inversion is applied to Q1, and a highly predictable amount of charge q.sub.sk is thereby removed from C.sub.int.

Because the charge removed q.sub.sk by Q1 is ideally just that developed by the background component of photocurrent i.sub.ph, Q1 as driven by the skimming pulse V.sub.sk described above is referred to as a "background suppression circuit". Though useful with all focal plane readout unit cells which employ integrating capacitors, this circuit is particularly beneficial when the photodetector 10 driving the unit cell is an infrared-sensitive photodiode, which tend to generate photocurrent with a high background component.

A detector interface circuit such as interface circuit 12 preferably serves as a buffer between the output of a photodetector such as photodetector 10 and an integrating capacitor such as C.sub.int. The interface circuit 12 buffers the photodetector 10 from the large voltage swings which occur on the integrating capacitor during integration by preferably presenting a low impedance to the incoming photocurrent and having a high impedance output. Detector interface circuit 12 is typically either a direct injection circuit or a feedback-enhanced direct injection circuit (FEDI). An example of a direct injection circuit is shown in FIG. 1, in which a common-gate MOSFET Q2 is used as interface circuit 12. A voltage V.sub.gate applied to the gate of Q2 biases the photodetector 10 during an integration period, then is made more positive to turn Q2 off and stop the flow of photocurrent at the end of the integration period.

To remove a predictable amount of charge, the amplitude and duration of V.sub.sk are critical. The proper values for these skimming pulse characteristics will be particular to a specific system design, and are dependent on several factors. For example, the characteristics of the scene being detected and the photodetector's leakage characteristics should be considered. The characteristics of the charge skimming transistors must also be known, because the amount of charge skimmed is largely determined by the effective gate drive (V.sub.gs -V.sub.th) applied to them. The he expected variability in V.sub.th for the skimming transistors in an array should be determined or estimated for the fabrication process being used. The amount of variability in the charge skimming process which can be tolerated should also be ascertained. This leads to a minimum value of effective gate drive and to suitable dimensions for the charge-skimming transistors that, when followed, reduce the variability in q.sub.sk to an acceptable value. A charge-skimming MOSFET that is long and narrow is preferred: because a higher voltage is required to drive it into strong inversion, such a device is less sensitive to V.sub.th variability.

Skimming pulse V.sub.sk is produced by a pulse generator circuit 22, arranged to generate pulses having the amplitude and duration characteristics required by a particular imaging system, as discussed above. The pulses are preferably derived from a master clock signal which is typically generated on the same chip that contains the unit cells, using conventional means that are well-understood in the art.

The invention contemplates a least one skimming pulse per integration period. For a single skimming event per integration period, the skimming pulse V.sub.sk preferably occurs when the stored charge is about at the capacitor's charge capacity limit q.sub.limit, which generally occurs at about the middle of the integration period .tau..sub.int. The background suppression circuit can be arranged to remove even more charge per period by performing more than one skim per period, though additional charge-skimming events are likely to couple more noise into the unit cell's signal path than would a single event. When employing two skimming events per period, the pulses preferably occur at approximately 1/3 and 2/3 of .tau..sub.int. A graph depicting the use of two skimming events, occurring at times 1/3.multidot..tau..sub.int and 2/3.multidot..tau..sub.int, respectively, each of which removes a quantity of charge q.sub.sk, is shown in FIG. 2b. For a system with a variable integration time, the choice of when and how many times to do the charge skimming is part of the overall system design. The invention also contemplates the possibility of briefly interrupting an integration to perform a charge skimming operation, which could reduce the possibility of introducing noise via the coupling of the skimming pulse into the signal path.

Integrating capacitor C.sub.int is typically connected between terminal 18 and ground, so that only one of its terminals need be accessible. This allows C.sub.int to be implemented with a MOSFET used as an inversion capacitor. Alternatively, a conventional capacitor such as a MOS capacitor can be used.

A unit cell per the present invention can be configured such that its charge-skimming MOSFET performs capacitor reset and readout functions in addition to its charge-skimming duties. The schematic diagram of FIG. 3 shows one such unit cell 23 from a row and column array of unit cells. The drain of a FET Q3 is connected to the integrating capacitor C.sub.int, and its source is connected to a column output bus 24 which it shares with all other unit cells 25 in the same column. Q3's gate is connected to a row select bus 26, with other rows of unit cells connected to respective row select buses 27. A switch S1 is connected between the column output bus and voltage V.sub.reset.

Q3 is used to reset the voltage on C.sub.int as follows: switch S1 is closed and row select bus 26 is pulsed with a signal .phi..sub.reset. This turns Q3 on and connects C.sub.int to V.sub.reset. The amplitude and duration of .phi..sub.reset are not critical as long as Q3 turns on fully and all the integrating capacitors in the array are preset to V.sub.reset.

Q3 is used to perform a skimming operation by closing S1 during the integration period and pulsing row select bus 26 with a signal .phi..sub.skim, which is analogous to the V.sub.sk skimming pulse discussed above. As stated above, the timing, amplitude and duration of .phi..sub.skim are critical to the skimming operation.

Q3 is used to readout the charge on C.sub.int by pulsing row select bus 26 with a signal .phi..sub.rowselect with switch S1 open. This transfers the charge on C.sub.int to the column output bus 24, and eventually to a column amplifier (not shown). The amplitude of .phi..sub.rowselect is not critical as long as Q3 turns on fully. The duration of .phi..sub.rowselect is also not critical as long as it is sufficient to allow the column amplifier to settle adequately.

Clock generation circuitry 28 generates signals .phi..sub.reset .phi..sub.skim, and .phi..sub.rowselect, which are preferably derived from a master clock signal by means well-understood in the art.

A number of unit cells per the present invention can be assembled into an N.times.N array to form an imaging chip; the typical architecture of such a chip is shown in FIG. 4. Each unit cell 30 in the array is as described above; i.e., with a charge-skimming FET driven by a charge-skimming pulse having a period substantially shorter than an integration period and an amplitude sufficient to operate the FET in its strong inversion mode. The configuration shown forms a "staring" array in which image data is read out one row at a time, with the outputs of the N unit cells in a row being multiplexed out to a linear array of N column amplifiers and sample and hold circuits.

In operation, three sequential timing periods are required for each video frame: the unit cell reset time, the photocurrent integration time, and the data readout time. The unit cell reset time occurs at the beginning of the video frame, during which the integrating capacitors in all N.times.N unit cells are set to the V.sub.reset. The photocurrent integration period is then started, typically by either opening the switch that connects the integrating capacitor to V.sub.reset, or by setting the gate voltage V.sub.gate on detector interface transistor Q2 to permit photocurrent to pass through to the capacitor. Manipulation of V.sub.gate is also commonly used to terminate the integration period.

Data readout is performed immediately following the integration period. Clock generation circuitry 32 inserts a logic "one" into the first stage 34 of a row shift register, and another "one" is inserted into the first stage 36 of a column shift register. This "one" causes the signals stored on the integrating capacitors of the first row of unit cells to be transferred simultaneously to an array of N column amplifiers/sample and hold circuits/multiplexers. The clock generation circuitry 32 then shifts the "one" down the N stages of the column shift register, causing the signal voltages on the sample and hold circuits to be sequentially read into an output driver 38.

After data in the first row of unit cells has been read out, the clock generation circuitry 32 shifts the "one" in the row shift register to the second stage 40, and the row data readout sequence is repeated for the second row. This sequence of events continues until the Nth row is read out.

Only MOSFETs have been discussed herein to perform the invention's charge-skimming operation, but the invention is not limited to this type of transistor. Other types of transistors, including JFETs, MESFETs, and bipolar transistors could also be employed--it is only necessary that the device be able to turn on and provide a well-controlled conductive path when driven with a given skimming pulse. However, focal plane arrays are typically operated at very low temperatures at which bipolar transistors perform poorly.

While particular embodiments of the invention have been shown and described, numerous variations and alternate embodiments will occur to those skilled in the art. Accordingly, it is intended that the invention be limited only in terms of the appended claims.