|Publication number||US20020179820 A1|
|Application number||US 09/993,886|
|Publication date||Dec 5, 2002|
|Filing date||Nov 27, 2001|
|Priority date||Nov 27, 2000|
|Also published as||EP1353791A2, EP1353791A4, US6822213, US20030201379, WO2002043113A2, WO2002043113A3|
|Publication number||09993886, 993886, US 2002/0179820 A1, US 2002/179820 A1, US 20020179820 A1, US 20020179820A1, US 2002179820 A1, US 2002179820A1, US-A1-20020179820, US-A1-2002179820, US2002/0179820A1, US2002/179820A1, US20020179820 A1, US20020179820A1, US2002179820 A1, US2002179820A1|
|Original Assignee||Moshe Stark|
|Export Citation||BiBTeX, EndNote, RefMan|
|Referenced by (6), Classifications (19), Legal Events (1)|
|External Links: USPTO, USPTO Assignment, Espacenet|
 This application claims the benefit of U.S. Provisional Patent Application Ser. No. 60/252,915, filed Nov. 27, 2000, and entitled “Noise floor reduction in image sensors,” incorporated herein by reference in its entirety.
 The present invention relates to imaging electronics in general, and more particularly to noise floor reduction in CMOS process Active Pixel image sensor systems.
 CMOS process Active Pixel Sensor (APS) technology is foreseen as the next generation technology for image sensors, which will replace the currently dominating CCD technology. Among the advantages that APS technology has over CCD technology are the ability to integrate image sensor and camera electronics onto a single chip, low power dissipation due to the inherently lower CMOS process voltage as compared with CCD voltage, and significantly-lower manufacturing costs.
 Dramatic advance in the CMOS process technology are also expected to lead to the implementation of imagers with a 5 μm pixel pitch on a submicron CMOS process, which is approximately equal to the diffraction limit of the camera lens. This limit offsets one of the major advantages of CCD technology, namely the high fill factor afforded by a very simple pixel circuit.
 The ability to implement photographic-quality imagers using CCD technology is severely limited by the large array dimensions that would be required, having thousands of pixel columns and rows. It is difficult to implement such large arrays using CCD technology due to the CCD Charge Transfer Efficiency (CTE) factor which dictates that image quality severely deteriorates as the size of the image sensor array increases. It is not commercially feasible to produce 3,000×2,000 pixel CCD arrays as would be required for near photographic quality images due to the prohibitive manufacturing costs involved.
 Although the transition from CCD-based technology to APS-based technology for commercial image sensors appears inevitable, APS technology has several limitations that have yet to be overcome. The ability to implement large CMOS-based APS image sensor arrays is limited by readout bus capacitance that originates from multiplexing all pixels within each column into a single column line. The parasitic output capacitances of the multiplexing circuits and of the line interconnect, normally implemented with metal, are the major contributors to column capacitance. Thus, for a given CMOS process and pixel unit cell size, the column capacitance is proportional to the number of multiplexed rows.
 The column capacitance is the dominant contributor to the input-referred noise and it governed by the so-called “kTC” noise mechanism. One technique that may be used to reduce the kTC noise effect involves introducing an amplification stage in each pixel's unit cell by including an in-pixel Source-Follower circuit. The Source-Follower amplifier “de-couples” the in-pixel integration capacitor from the column capacitance, which results in a reduced input-referred readout noise. However, this technique leads to a reduction in gain due to the attenuation of the signal as a function of column bus capacitance. This can be costly in terms of signal-to-noise ratio (SNR) for large-format circuits with a high column capacitance and for applications where the charge that is involved is small. Thus, although implementing a Source-Follower circuit results in a reduced input-referred readout noise, its effect diminishes as the imager's size increases due to the increasing column capacitance.
 The present intention seeks to provide methods and apparatus for noise floor reduction in CMOS-based APS image sensor arrays that overcomes disadvantages of the prior art. The present invention substantially reduces the column capacitance in large image sensor arrays, resulting in a reduced noise floor and a better signal-to-noise ratio. A Direct Injection (DI) circuit approach is employed in place of the Source-Follower circuit per unit cell approach. A DI circuit is relatively simple to implement and deploys less transistors per unit cell, which results in a higher unit cell fill-factor, a smaller pixel, or both. Furthermore, the Fixed Pattern Noise (FPN) of a DI circuit is considerably lower than that of the Source-Follower-based unit cell. The DI circuit of the present invention directly injects the charge accumulated by the integration capacitor into the column. This results in a significant input-referred readout noise that is higher than that of the Source-Follower-based unit cell. By reducing column capacitance the present invention significantly reduces the image sensor's noise floor and improves its signal-to-noise ratio, particularly in large image sensor arrays.
 In one aspect of the present invention an image sensor array is provided including a first plurality of unit cells coupled to a first sense amplifier, and a second plurality of unit cells coupled to a second sense amplifier, where the first plurality and the second plurality are substantially electrically isolated from each other.
 In another aspect of the present invention each of the first and second pluralities of unit cells includes at least one column line.
 In another aspect of the present invention the unit cells are arranged in two or more clusters of two or more of the unit cells each, and the unit cells within each of the clusters are coupled to a cluster line which is coupled to the column line,
 In another aspect of the present invention only one of the clusters is actively connected to the column line at any given time.
 In another aspect of the present invention the unit cells are direct injection unit cells.
 In another aspect of the present invention the first plurality and the second plurality are substantially electrically isolated from each other by at le 10M Ohms.
 In another aspect of the present invention an image sensor array is provided including a plurality of columns, each column including a plurality of unit cells coupled to a column lines a first sense amplifier coupled to a first plurality of the unit cells in each of the columns, and a second sense amplifier coupled to a second plurality of the unit cells in each of the columns, where the first and second pluralities of the unit cells in each of the columns are substantially electrically isolated from each other.
 In another aspect of the present invention each of the columns includes a plurality of clusters, each cluster including two or more of the unit cells coupled to a cluster line which is coupled to the column line.
 In another aspect of the present invention only one of the clusters is actively connected to the column line at any given time.
 In another aspect of the present invention the unit cells are direct injection unit cells.
 In another aspect of the present invention the first plurality and the second plurality are substantially electrically isolated from each other by at least 10M Ohms.
 In another aspect of the present invention a method for reducing noise floor in an image sensor is provided, the method including sensing a first plurality of unit cells with a first sense amplifier, and sensing a second plurality of unit cells with a second sense amplifier.
 In another aspect of the present invention either of the sensing steps includes sensing different subsets of the unit cells at different tunes.
 In another aspect of the present invention either of the sensing steps includes sensing mutually exclusive subsets of the unit cells at different times.
 In another aspect of the present invention each of the sensing steps includes sensing its associated plurality of unit cells in substantial electrical isolation from the other the plurality of unit cells.
 In another aspect of the present invention each of the sensing steps are performed alternatingly.
 The disclosures of all patents, patent applications, and other publications mentioned in this specification and of the patents, patent applications, and other publications cited therein are hereby incorporated by reference in they entirety.
 The present invention will be understood and appreciated more fully from the following detailed description taken in conjunction with the appended drawings in which:
FIGS. 1A and 1B are schematic flow illustrations of a Direct Injection (DI) unit cell 100, useful in understanding the present invention;
FIG. 2 is a schematic illustration of an image sensor array segment, useful in understanding the present invention;
FIGS. 3A and 3B, taken together, are top-view and side-view illustrations of readout transistor T2 of FIGS. 1A, 1B, and 2, useful in understanding the present invention;
FIG. 4 is a schematic illustration of an image sensor array, constructed and operative in accordance with a preferred embodiment of the present invention; and
FIG. 5 is a schematic illustration of an alternative image sensor array column arrangement for use with the image sensor array of FIG. 4, constructed and operative in accordance with a preferred embodiment of the present invention.
 Reference is now made to FIGS. 1A and 1B, which are schematic illustrations of a Direct Injection (DI) unit cell 100, useful in understanding he present invention. DI 100 is shown as having a photodiode PD 102, an integration capacitor Cmt 104, a transistor T1 106, column line capacitance Ccol 108, a readout transistor T2 110, a column line 112, and transistor gates 114 and 116.
 When an image sensor's background photon noise is low, its noise floor is determined by the image sensor's electronics, particularly the readout electronics associated with photocurrent signals, as well as its input stages, which in turn determines the image sensor's sensitivity.
 The major noise components which determine the noise floor are Fixed Pattern Noise (FPN), 1/f noise, and white noise. A Direct Ejection stage typically features a very low FPN, and conventional techniques may be applied to remove 1/f noise. If generated at later stages, white noise may also be removed using conventional techniques
 Among the major factors which set the lower bound to noise originating from the pixel readout, the dominant noise component is the so-called kTC noise, which originates from transferring charges from their origin to a collecting capacitor C. The kTC noise originates from a resistor which charges a capacitor. The noise output on the capacitor, <vn>, which originates from a resistor, is expressed:
 where k is Boltzmann's constant of 1.38×10−23 Joul/° K., T is the resistor's/capacitor's temperature expressed in degrees Kelvin, C is the capacitor's capacitance expressed in Farads, and <vn> is the capacitor's PMS noise voltage expressed in Volts.
 The noise may be expressed in terms of “noise electrons,” that is the number of electrons that would cause the RMS noise on capacitor C. The RMS number of noise electrons, <Nn> may be derived from (EQ. 1) as:
 where e is the electron charge of 1.6×10−19 Coulomb.
 In FIG. 1A a photon-generated photocurrent Iph flows from photodiode PD 102 into integration capacitor Cint 104 through transistor T1 106. The kTC noise source is the transistor T1 channel resistance. The photocurrent integration stage is depicted in FIG. 1A, and thus the noise source may be defined as the integration noise. The integration noise may be expressed in terms of RMS voltage as:
 where <vn int> is the RMS integration noise and Cint is the charge integration capacitance. The RMS number of noise electrons <Nn int> may be expressed as:
FIG. 1B shows the path taken by the integrated charge readout from the Cint to the column line 112. The column line's capacitance is shown as Ccol 108. The charge transfer is embodied as a current flow through a readout transistor T2 110. This current flow generates a kTC noise on the column line, and the noise component is translated into an equivalent noise source on the integration capacitor Cint. This represents the noise on the integration capacitor that would result in the same noise on the column line. This noise is referred to as “input-referred noise,” and is expressed as <vn col> for the RMS noise voltage and <Nn col> for the RMS number of noise electrons. It may be shown that
 Typically, Ccol>>Cint. Thus, it may be seen that the dominant factor which contributes to noise floor is not the integration noise, but rather the noise that originates from the integrated charge readout to the image sensor's column.
 By way of example, given a 0.6 μm process, a 10 μm×10 μm pixel, and a 1,000-row image sensor, the column capacitance is approximately 4 pF, and the integration capacitance is approximately 0.1 pF. In this example, the column input-referred readout noise is approximately 6.5 times greater than the integration noise. The readout noise is approximately 40 μV RMS, while the input-referred readout noise is approximately 1.5 mV. The integration noise is approximately 6 μV rms.
 Thus, it may be seen that the column readout noise is the dominant factor and may be considered to be the noise floor. Significant reduction of the column readout capacitance would therefore result in a significant noise floor reduction, as the column readout noise is determined by the
 ratio and reduction of Ccol would result in noise floor reduction.
 Improvement in the signal-to-noise ratio may also be achieved as follows, Let vint represent the highest possible signal that may be collected on the integration capacitor Cint at reaching saturation. Given that column readout noise a dominant contributor to noise floor, the signal-to-noise ratio may be approximated as:
 where SNR is the signal-to-noise ratio at the column line, vint is the near-saturation voltage on the integration capacitance, and <vn col> is the input-referred column line noise RMS voltage. Thus, for a 5 volt process vint is approximately 1.5 Volts. Continuing with the previous example, given a 373° K. junction temperature, the input-referred column readout noise may be as much as ˜1.5 mVolts, resulting in a signal-to-noise ratio of approximately 1,000. Where column readout noise is negligible, the signal-to-noise ratio is limited mainly by the charge integration noise, being approximately 6.5 times better than the signal-to-noise ratio in this example.
 Reference is now made to FIG. 2, which is a schematic flow illustration of an image sensor array segment, useful in understanding the present invention. In FIG. 2 a single column 200 of an X by V-rows image sensor array is shown having multiple unit cells 202 connected to a column line 206, where each unit cell includes a Direct Injection (DI) circuit 204 as described hereinabove with reference to FIGS. 1A and 1B. In the configuration shown, when a row is read out, its readout transistors T2 conduct a charge, and the charge accumulated on the integration capacitors of the row is transferred to its respective column line. All the other readout transistors which reside on each column are in a cutoff state.
 The column capacitance Ccol in FIG. 2 may be approximated by:
C col ≅V·(C d +c M ·a) (EQ. 8)
 where V is the number of image sensor rows, Cd is the readout transistor drain capacitance when in cutoff and when the column is biased approximately to 0 Volts, cM is the column metal capacitance per unit length, and a is the pixel pitch for square pixels. Ccol≅Cd+cM·a is thus the column capacitance per pixel, as is shown at reference numeral 208.
 It may thus be seen that column capacitance, which determines the noise floor, is directly proportional to the number of rows in the image sensor, and, thus, the larger the image sensor array, the greater the noise floor.
 Reference is now made to FIGS. 3A and 3B, which are top-view and side-view illustrations of readout transistor T2 of FIGS. 1A, 1B, and 2, useful in understanding the present invention. In FIGS. 3A and 3B a transistor assembly 300 is shown including a transistor T2 element 302 including a gate 304, a drain 306, a column metal line 308, a field oxide element 310, a connection 312 of column metal line 308 to drain 306, all overlying a bulk 314.
 In FIGS. 3A and 3B, transistor T2 is shown with its contact and an adjacent section of column line, typically constructed from M1 metal. It may be seen that the width of transistor T2 transistor is not minimal due to the drain-to-column contact rules which require the width of transistor T2 to be more than double the minimal possible transistor channel width. The drain diffusion capacitance and the overlapping gate-drain capacitance determine the drain capacitance Cd as follows:
C d ≅C gd +C db (EQ. 9)
 where Cgd is the gate-to-drain overlapping capacitance, and Cdb is the drain bulk diode capacitance at zero volts. It may further be seen that:
C gd ≅W·L OV ·c g (EQ. 10)
 where W is the T2 transistor's width, LOV is the overlapping distance between the gate and the drain, which is usually derived in an empirical manner, and Cg is the gate-bulk capacitance per unit area determined by the gate oxide thickness. And finally:
C db ≅c jd 0 ·A d +c jdsw 0 ·P d (EQ. 11)
 where cjd 0 is the drain junction capacitance at zero voltage bias per area unit, Ad is the drain junction area, cjdsw 0 is the drain junction sidewall capacitance per unit length at zero voltage bias, and, Pd is the junction periphery length which includes all the junction sidewalls excluding the gate side.
 By way of example, for a typical 0.6 μm CMOS process, the T2 transistor has
 Ad≅2 μm2
 Pd≅4 μm
 cjd o≅0.4 fF/μm2
 cjdsw o≅0.45 fF/μm
 Lov≅0.1 μm
 Therefore, Cd may be expressed as:
Cd≅3fF (EQ. 12)
 The metal capacitance per unit length CM is given by:
c M =c M A ·W M+2·c M P (EQ. 13)
 where cM A is the metal line capacitance per area unit, and cM P is the metal capacitance, per line side, per unit length. Thus, for the example 0.6 μm CMOS process, the typical Metal 1 capacitances are
c M A≅0.04 fF/μm 2
c M P≅0.03 fF/μm
 and a metal width WM≅0.6 μm,
 giving a metal capacitance per unit length CM as
c M≅0.08 fF/μm (EQ. 14)
 Given EQ. 12 and 14, for a pixel pitch a=10 μm the total column capacitance per pixel ccol may be calculated as:
c col≅3.8 fF/pixel (EQ. 15)
 Thus if V=1,000, the column capacitance is approximately 3.8 pF.
 The integration capacitor's capacitance value may also be calculated. This value for a 0.6 μm CMOS process is:
Cint≅0.1 pF (EQ. 16)
 Reference is now made to FIG. 4, which is a schematic illustration of an image sensor array 400, constructed and operative in accordance with a preferred embodiment of the present invention. The sensor array 400 of FIG. 4 includes one or more columns 402, each having one or more Direct Injection (DI) unit cells 404 configured as described hereinabove with reference to FIGS. 1A, 1B, 2, 3A, and 3B. As in most any two-dimensional array, sensor array 400 may be alternatively referred to as having one or more rows of unit cells 404. Each column 402 of sensor array 400 is separated into two or more electrically isolated portions, such as into an upper half 406 and a lower half 408 as shown in FIG. 4, thus forming one or more upper half rows and one or more lower half rows. The separation of the portions of each column is such that there is little or no conductivity between the portions, such as a resistance of greater that 10M Ohm. Each electrically isolated portion is arranged to be read out through a separate sense amplifier, such as is shown in FIG. 4 where each upper half row is arranged to be read out through a top sense amplifier set 410, and each lower half row is arranged to be read out through a bottom sense amplifier set 412. Sensor array 400 is also preferably configured with a row decoder 414 and an output buffer 416.
 Since the upper and lower halves 406 and 408 of each column 402 are electrically isolated, the column capacitance which each sense amplifier set faces may be expressed as:
 The associated noise floor is of each sense amplifier set is thus,
 where typically Ccol>>Cint. The reduction in the noise level may then be calculated as:
 Thus, by splitting each column in the array into two halves, the noise floor is reduced to about 70% of what it would be were the columns not split.
 The signal-to-noise ratio of the output data improves by the same factor as the noise floor reduction as follows:
 where SNR1 is the signal-to-noise ratio of the split array, and SNR is the signal-to-noise ratio an undivided array as described hereinabove with reference to FIGS. 1A, 1B, 2, 3A, and 3B. In an array where the columns are split into N portions, the improvement of the SNR will be on the order of the square root of N.
 Reference is now made to FIG. 5, which is a schematic illustration of an alternative image sensor array column arrangement for use with the image sensor array of FIG. 4, constructed and operative in accordance with a preferred embodiment of the present invention. The sensor array column of FIG. 5, now referred to as column 500, includes one or more Direct Injection (DI) unit cells 502 configured as described hereinabove with reference to FIGS. 1A, 1B, 2, 3A, and 3B, and is separated into electrically isolated upper and lower halves as is described hereinabove with reference to FIG. 4. Each half of column 500 is segmented into k clusters 504 (shown as 504A, 504B, 504C, and 504D), typically numbering two or more. Each cluster 504 may be expressed as
 rows, where V is the number of unit cells 502 in cluster 504, and is interconnected to a column line 506 via a cluster select transistor 508 (shown as 508A, 508B, 508C, and 508D) that is controlled by a cluster selector 510 (shown as 510A, 510B, 510C, and 510D).
 Typically, during row readout of column 500, only one out of 2·k cluster select transistors 508 in column 500 is “ON”, and all other cluster select transistors 508 in column 500 are “OFF”. The row readout typically starts with the top cluster 504A, when cluster select transistors 508A transistor is “ON”. The first cluster rows are then selected and sequentially read, starting with row 0, and ending with row
 After all the rows in the top cluster are read out, cluster select transistor 508A of the top cluster 504A switches from “ON” to “OFF” (i.e., cluster selector 510A goes from “High” to “Low”), and cluster select transistor 508B of the next cluster 504B switches from “OFF” to “ON” ((i.e., cluster selector 510B goes from “Low” to “High”). The rows of cluster 504B are then read sequentially. This operation continues until all the rows of the all of the columns 500 of the image sensor array are read out. As is described hereinabove with reference to FIG. 4, the top half rows of each column 500 is read through a top sense amplifier set, while the bottom half rows of each column 500 is read through a bottom sense amplifier set.
 It may be seen that since only one cluster 504 is actively connected to column line 506 at a time while all the other clusters are not actively connected to column line 506, the total parasitic load of column line 506 is significantly reduced. By way of explanation, assume that cluster select transistors 508 are of the same dimensions as the readout transistors T2 (FIGS. 1A, 1B, 2, 3A, and 3B) in every unit cell. This assumption may be justified, since the transistor's width is much wider the minimum width, where the width is determined by metal-to-drain contact dimensions and overlap design rules. Thus, putting such two transistors in series does not significantly slow down the readout. Therefore, the dimensions of cluster select transistor 508 may be given the same dimensions as that of readout transistor T2. This being the case, the drain capacitance of cluster select transistor 508 is identical to that of readout transistor T2. Since k−1 cluster select transistors 508 are “OFF” at any given time, their associated clusters are not actively connected to column line 506 and, therefore, do not load the column line.
 The capacitance of column line 506 may be calculated as:
 is the capacitance associated with a single currently-read cluster,
 approximates the capacitance associated with the column and the cluster select transistors parasitic capacitance.
 Since it is highly desirable to minimize column capacitance, the optimal value for k may be found for EQ. 21 as:
 Once the optimal number of rows in a cluster is determined, the column capacitance may be also derived from EQ. 21.
 Continuing with the example presented hereinabove, for an image sensor with 1,000 rows, on a 0.6 μm CMOS process, kopt≅25. Thus, in this example, each half of the image sensor array should be divided into 25 clusters, with 20 rows in each, in order to achieve a minimum column capacitance of Ccol 2≅0.55 pF.
 Thus, through column segmentation and image sensor array halving, column capacitance may be reduced approximately by a factor of 7. This reduces the noise floor from about 1.5 mVolts to less t 0.7 mVolts in the present example, and improves the signal-to-noise ratio from approximately 1,000, to approximately 2300.
 It is appreciated that several options are available by which the necessary circuitry described hereinabove may be implemented in a minimum of space. For example, the cluster line may be implemented in M1, while the column line may be implemented over the cluster line in M2. Alternatively, the cluster line may be implemented in Poly, while the column line is implemented over the cluster line in M1. The cluster selector lines may also be implemented in Poly. This is feasible since the cluster selection is done infrequently, once every
 rows. Thus, the longer time it takes to precondition the cluster for readout is insignificant. It is also possible to alternate the readout between rows in the upper half of the image sensor array and rows in the bottom half. This would effectively double the readout time per row, and leave significant slack time for cluster switching. If the cluster select line is implemented in Poly, it could be placed over a ground line or a signal line which runs in metal, thus reducing space requirements even further.
 It is appreciated that one or more of the steps of any of the methods described herein may be omitted or carried out in a different order than that shown, without departing from the true spirit and scope of the invention.
 While the methods and apparatus disclosed herein may or may not have been described with reference to specific hardware or software, it is appreciated that the methods and apparatus described herein may be readily implemented in hardware or software using conventional techniques.
 While the present invention has been described with reference to one or more specific embodiments, the description is intended to be illustrative of the invention as a whole and is not to be construed as limiting the invention to the embodiments shown. It is appreciated that various modifications may occur to those skilled in the art that, while not specifically shown herein, are nevertheless within the true spirit and scope of the invention.
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7088394 *||Jul 9, 2001||Aug 8, 2006||Micron Technology, Inc.||Charge mode active pixel sensor read-out circuit|
|US7336309||Jul 3, 2001||Feb 26, 2008||Vision-Sciences Inc.||Dynamic range compression method|
|US7649558 *||Jun 20, 2006||Jan 19, 2010||Aptina Imaging Corporation||Charge mode active pixel sensor read-out circuit|
|US9036068 *||Oct 14, 2009||May 19, 2015||Sony Corporation||CMOS image sensor with fast read out|
|US20040169752 *||Jan 7, 2004||Sep 2, 2004||Moshe Stark||Multi-photodetector unit cell|
|US20100097510 *||Oct 14, 2009||Apr 22, 2010||Sony Corporation||Image pickup element and control method therefor, and camera|
|U.S. Classification||250/208.1, 348/E03.021, 257/E27.132|
|International Classification||H04N5/365, H04N5/341, H04N5/345, H01L27/146, H04N3/14, H01L27/00|
|Cooperative Classification||H04N5/3415, H01L27/14643, H04N5/3454, H01L27/14609, H04N5/3658|
|European Classification||H01L27/146F, H04N5/365A3, H04N5/345B, H04N5/341A, H01L27/146A4|
|Apr 22, 2002||AS||Assignment|
Owner name: VISION - SCIENCES INC., NEW YORK
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:STARK, MOSHE;REEL/FRAME:012823/0800
Effective date: 20020122