Low-voltage CMOS comparator
||Garavan, Patrick J.;
A MOS comparator which includes a capacitor connected in an electrical path between two amplification stages. The comparator also includes a voltage source, and a switch is provided between the voltage source and the input of the second stage. A variability of electrical parameter of the voltage source can be matched with a parameter of the amplification stage. The comparator can also include another switch between another voltage source and a third stage, with the two voltage sources providing different voltages. A comparator gain stage includes circuitry for deriving a differential current from the two voltages. Circuitry is also provided for loading the differential current to derive an amplified difference voltage. Further circuitry is provided for bypassing the loading circuity to reduce a quiescent voltage drop associated with the loading circuitry.
FIELD OF THE INVENTION
This invention relates to CMOS comparators and more particularly to CMOS comparators for use in low-voltage, low-power analog-to-digital conversion applications.
BACKGROUND OF THE INVENTION
CMOS semiconductor manufacturing technologies are relatively inexpensive, and permit the design of integrated circuits including both digital and analog functions. CMOS technology has been used to build multi-stage differential amplification circuits for analog-to-digital converters. Such circuits have employed offset memorization to compensate for input offset voltages in the amplification stages.
Offset memorization is generally performed by dividing circuit operation into two phases, an offset memorization phase and an amplification phase. During the offset memorization phase, a capacitor in series with the output of an amplification stage is grounded, allowing the stage's input offset voltage multiplied by the stage's gain to be stored in the capacitor. During the subsequent amplification phase, the memorized offset voltage cancels by subtraction the actual offset voltage of the amplification stage. Alternatively, offsets can be memorized by providing a feedback path around an amplification stage during the offset memorization phase.
Different CMOS differential amplification stages have been proposed for these analog-to-digital conversion applications. These stages have not been particularly well suited to low-voltage operation, however, because too many transistors are connected between the supply rails or because resistors are used to load the outputs of the gain stages.
SUMMARY OF THE INVENTION
In general, the invention features a CMOS comparator which includes a capacitor connected in an electrical path between two amplification stages. The comparator also includes a voltage source, which can provide a common-mode voltage for use in offset memorization, and a switch is provided between the voltage source and an input of the second stage. A variability of an electrical parameter of the voltage source can be matched with a variability of a parameter of the amplification stage. The comparator can also include another switch between another voltage source and an input of a third stage, with the two voltage sources providing different voltages.
In another general aspect, the invention features a comparator gain stage, which includes circuitry for deriving a differential current from two input voltages. Circuitry is also provided for loading the differential current to derive an amplified difference voltage. Further circuitry is provided for bypassing the loading circuity to reduce a quiescent voltage drop associated with the loading circuitry.
The gain stage of the invention is particularly well suited for low voltage operation because a minimum of number of devices are connected in any series path between the supply rails, and because the load devices are driven with only part of the current from the input differential pair. The gain stage structure of the invention is also advantageous in that its gain is dependent on a ratio defined by the transconductance of matched input and load transistors. Gain is therefore process-independent and insensitive to supply voltage and temperature changes. Clamping is also process-independent and insensitive to changes in supply voltage and temperature, because it is performed by devices which are matched to the devices that they clamp.
In another aspect, the invention is advantageous in that the voltage source devices used in connection with memorizing the offset are matched with the gain stage output devices. This improves responsiveness to drifts and variations and helps to prevent the inputs of gain stages from being driven outside the supply voltages. An offset memorization voltage level can also be different for the different gain stages of the device, which allows for stage-by-stage optimization of the comparator circuitry.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a partially-block partially-schematic circuit diagram of an example of a comparator according to the invention;
FIG. 2 is a simplified circuit diagram of the first gain stage of the comparator of FIG. 1;
FIG. 3 is a simplified circuit diagram of the latching stage of the comparator of FIG. 1;
FIG. 4 is a detailed schematic circuit diagram of the comparator of FIG. 1;
FIG. 5 is a detailed schematic circuit diagram of the gain stage of FIG. 2; and
FIG. 6 is a detailed schematic circuit diagram of the latching stage of FIG. 3.
Referring to FIG. 1, an example of a comparator 10 according to the invention includes a differential input gain stage 12, a first differential intermediate gain stage 14 responsive to the input stage, a second differential gain stage 16 responsive to the first intermediate stage, and a high-speed output latching stage 18 responsive to the second intermediate stage. A first offset memorization circuit 20 is provided between the input gain stage and the first intermediate stage, a second offset memorization circuit 22 is provided between the first and second intermediate stages, and a third offset memorization circuit 24 is provided between the second intermediate stage and the output latching stage. The stage voltage gains are set at 20, 20, and 15 for the input stage and the first and second intermediate stages, respectively. Although this embodiment includes four stages, the principles of the invention are, of course, applicable to circuits with different numbers of stages.
The first offset memorization circuit 20 includes first and second capacitors 26, 28 each located in one of the two signal paths between the input stage 12 and the first intermediate stage 14. A first NMOS reset switch 30 is provided between a first plate of the first capacitor and a first voltage source 34. A second NMOS reset switch 32 is provided between a second plate of the second capacitor and the voltage source 34. The voltage source 34 is referenced to a ground rail 35.
The second offset memorization circuit 22 includes third and fourth capacitors 36, 38 each located in one of the two signal paths between the first and second intermediate stages 14, 16. A third NMOS reset switch 40 is provided between the third capacitor and a second voltage source 44. A fourth NMOS reset switch 42 is provided between the fourth capacitor and the second voltage source. The second voltage source is referenced to the ground rail 35.
The third offset memorization circuit 24 includes fifth and sixth capacitors 46, 48 each located in one of the two signal paths between the first intermediate stage 16 and the output latching stage 18. A fifth NMOS reset switch 50 is provided between the fifth capacitor and a third voltage source 54. A sixth NMOS reset switch 52 is provided between the sixth capacitor and the third voltage source. The third voltage source is referenced to the ground rail 35.
In operation, offset errors due to device mismatches are removed by auto-zeroing each of the input stage 12 and the two intermediate stages 14, 16 in an offset memorization operating state prior to an amplification operating state. This is achieved by shorting inputs 15, 17 of the input stage 12 together, and closing the reset switches 30, 32, 40, 42, 50, 52 in the offset memorization circuits 20, 22, 24. This auto-zeroing also removes other errors introduced by, for example, low-frequency power supply variations, leakage currents, and offset drifts.
The auto-zeroing operation independently sets the common-mode input voltage of each stage 14, 16, and 18 to the value of the respective one of voltage sources 34, 44, and 54. This is an important attribute for a low-voltage system, because the common-mode input level of each stage can be set independently to optimize each stage by assigning different output voltages to the voltage sources 34, 44, and 54. For example, the common-mode voltage at the input to the latching stage might be required to be different (e.g., lower by about 200 mV) from the common-mode voltage at the inputs of the other stages, because its input devices differ from those of the other stages. The ability to set the common-mode voltage of each stage independently obviates the need for level-shifting circuitry.
Another feature of this system is that the NMOS reset switches 30, 32, 40, 42, 50, 52 have their sources biased at a positive voltage. The gates of these devices are driven by the output of a CMOS inverter or buffer (not shown); therefore, when the switches are turned off, the gate-source voltage of these switches is negative. This significantly reduces the sub-threshold leakage current that flows in these switches and, hence, the error such a current would contribute to the operation of the comparator. The use of NMOS reset switches also improves power supply rejection since these switches have their backgates tied to ground 35 and, when turned off, their gates are at ground potential.
In the present embodiment, the three voltage sources 34, 44, 54 are each set to approximately 1 V output. Further, the output voltage swing of each gain stage is restricted to less than 1 V by means of clamping devices (not shown), which will be discussed in more detail below in connection with FIG. 4. This choice of source and clamping voltages allows the comparator to be operated without the input nodes of each gain stage going more negative than ground potential.
Setting the common-mode input level to each stage at 1 V also allows each gain stage to use PMOS input devices for its differential pair, as discussed below. Since these devices are isolated in N-wells, they provide good power supply rejection. PMOS devices also have lower threshold voltage (Vt) hysteresis than do NMOS devices. This significantly reduces another potential source of error in the comparator.
Referring to FIG. 2, the input gain stage 12 includes a folded-cascode gain stage circuit with differential input and differential output. This circuit includes first and second PMOS input devices 62, 64, which form a differential pair. The gates of these devices receive the voltages at the stage's's first and second inputs 66, 68, respectively; and the sources of these devices receive current from a first current source 70. The drains of these first and second devices act as the first and second outputs 72, 74 of the differential pair, respectively.
Each gain stage further includes first and second PMOS load devices 76, 78. The sources of these first and second load devices receive the voltage at the positive supply rail 61. The gates and drains of each load device are connected together, and their connected terminals are respectively tied to the outputs of current sources 80, 82 at nodes 84, 86, which nodes also serve as the first and second outputs of the gain stage.
Between the first gain stage output node 84 and a second supply rail 35, which may be at ground potential, is a first NMOS cascode device 88 in series with a first NMOS current source device 90. Similarly, between the stage's second output 86 and ground is a second NMOS cascode device 92 in series with a second NMOS current source device 94. The gates of the cascode devices both receive a first bias voltage 96, and the gates of the current source devices both receive a second bias voltage 98. The first output 72 of the differential pair is provided to a node 73 between the stage's second cascode device 92 and the second current source device 94. The second output 74 of the differential pair is provided to a node 75 between the first cascode device 88 and the first current source device 90.
In operation of the input gain stage 12, the first and second PMOS input devices 62, 64 in the differential pair convert the differential input voltage across the first and second stage inputs 66, 68 to a pair of currents. This signal current pair is then converted back to a voltage by the first and second load devices 76, 78. The first and second bias voltages 96, 98, bias the first and second current source devices 90, 94, and the first and second cascode devices 88, 92 respectively.
The stage gain is set by the ratio of input device 62 or 64 transconductance (gm1) to load device (76 or 78) transconductance (gm2), i.e.: ##EQU1## where W1 and L1 are the respective widths and lengths of the input devices 62, 64; Ids1 is the drain current in the input devices 62, 64; and Ids2 is the drain current of the load devices 76, 78. Since both the input and load devices are of the same polarity type, i.e., PMOS, and the current ratio can be accurately defined, the gain is set only by the relative sizes of the input and output devices. This makes the gain insensitive (to a first order) to temperature, supply levels and variations, and magnitude of the bias currents. This is particularly important for a low-power design, because the bias currents can be scaled while retaining the same stage gain.
Another important characteristic of a comparator is its noise performance. For the single gain stage as shown above, the total rms thermal noise referred-to-the-input is given by: ##EQU2## where C.sub.1 is the total capacitance on each output node 84, 86;
A is the stage gain as given in Eq. 1;
gm3 is the transconductance of the output current source devices 90, 94; and
gm4 is the transconductance of the device forming the second and third current sources 80, 82.
As with the stage gain, the total noise is dependent on transconductance ratios and load capacitance. To a first order approximation, the noise is independent of the bias current level. This again allows the bias current levels to be scaled and still retain the same input noise level or precision.
The use of PMOS input devices 62, 64 also provides an isolated device, because the input devices are situated in an N-well. This results in improved power supply rejection. In addition, a PMOS input device gives low Vt hysteresis, which is another desirable property for a precision comparator.
Due to the differential nature of the gain stage, improved rejection of power supply variation and leakage current errors also results. These types of errors are now dependent on the matching of devices, i.e., the matching of the load PMOS devices 76, 78 with the input PMOS devices 62, 64, and the matching of the second and third current sources 80, 82 with the first and second current source devices 90, 94. In essence, electrical parameters of these devices vary in a matched way in response to variations end drifts. With careful layout techniques, this matching can be quite accurate.
The "Miller" gain, Am, is defined as follows: ##EQU3## where W5 and L5 are the width and length respectively of the cascode devices 88, 92; .mu..sub.n and .mu..sub.p are the mobilities of elections and holes, respectively; Ids1 is the drain current in the input devices 62, 64; Ids5 is the drain current of the output devices 76,78; go5 is the output conductance of the first cascode device 88; and gm5 is its transconductance. In an ac-coupled system, it is desirable to have this gain as low as possible, since the input capacitance of each stage is determined by:
Cin=Cgsm1+[1+Am]x Cgdm1 (Eq4)
where Cgsm1 is the gate-source capacitance of the input devices 66, 68 and Cgdm1 is the gate-drain capacitance of these devices. The input capacitances (Cin) of the stages are ideally to be reduced as much as possible, because they cause undesirable attenuation between stages. Since the input devices are PMOS and the cascode devices 88, 92 are NMOS, and the NMOS devices have a higher mobility (.mu.), the Miller gain is low.
Another important feature of each gain stage is its ability to operate from a low supply voltage. The minimum supply voltage Vsupply.sub.-- min is determined by the following:
Vsupply.sub.-- min=Vgsm9+Vdsatm5+Vdsatm7 (Eq. 5)
where Vgsm9 is the gate-source voltage of the load devices 76, 78, Vdsatm5 is the saturation voltage of the cascode devices 88, 92; and Vdsatm7 is the saturation voltage of the current source devices 90, 94.
Reducing the term "Vgsm9" reduces the supply voltage required. This is achieved by the addition of the second and third current sources 80, 82, which allows current to bypass the load devices 76, 78. These therefore carry a lower current than the input devices 62, 64, and as a result, their quiescent gate-source voltages (Vgs) are reduced.
There are some minor differences in implementation between the stages. In particular, the input gain stage 12 is generally built to be larger than the second two stages 14, 16 and is typically driven by larger currents. This is a matter of scaling, however, and the structure and operation of the first and second intermediate gain stages 14, 16 are in essence the same as for the input gain stage 12. These intermediate stages will therefore not be discussed further.
Referring to FIG. 3, the output latching stage 18 includes a first PMOS input device 102 with its gate connected to a first input 101 of the stage, and a second PMOS input device 104 with its gate connected to a second input 103 of the stage. The first and second input devices 102, 103 receive current from first and second current sources 106, 108, respectively. The drains of the first and second input devices are connected to the second supply rail 35, and the sources of the first and second input devices are connected respectively to the gates of first and second NMOS differential pair devices 110, 112. The sources of these differential pair devices are connected together and to a third current source 114.
The sources of first and second current source PMOS devices 116, 118 receive the supply voltage at the power supply rail 61. The gates of these devices are biased by a first bias voltage 120, and the drains of these devices are respectively connected to first and second PMOS cascode devices 122, 124, which are biased by a second bias voltage 126. The drain of the first differential pair device 110 is connected to a node 111 between the first current source device 116 and the first cascode device 122. The drain of the second differential pair device 112 is connected a node 113 between the second current source device 118 and the second cascode device 124.
An NMOS reset device 128 is provided between the drains of the first and second cascode devices 122, 124. Also provided between these drains are first and second cross-coupled NMOS latching devices 130, 132. These latter devices have their sources connected to the second supply rail 35, and their gates connected to each other's drains.
In operation, the latching stage converts a small signal input voltage at the inputs 101, 103 to a full-CMOS output signal at the outputs 134, 136. The first and second input devices 102, 104 together with the first and second current sources 106, 108 form source follower input stages to reduce kickback from the latching stage. These source followers also level-shift the input signal to drive the differential pair formed by the differential pair devices 110, 112. This input level shifting allows the latching stage to operate at low supply voltages. The NMOS differential pair converts the input signal voltage to a differential current and this differential current drives the cross-coupled latching devices 130, 132.
During the initial part of the comparison operating state, the reset device 128 is "on", which prevents the latching devices 130, 132 from regenerating. When the reset device's gate signal 129 goes low, the reset device 128 is turned off, the latch enters its regeneration mode, and the outputs 134, 136 go to within approximately 50 mV of the power rails 35, 61, respectively. The cross-coupled NMOS latching devices 130, 132 cause this regeneration to occur at very high speed.
Referring to FIGS. 1 and 4, the complete comparator will now be discussed further. The first voltage source 34 includes first and second PMOS current devices 31, 33 together with a first PMOS voltage source device 37. The second voltage source 44 includes third and fourth PMOS current source devices 41, 43, together with a second PMOS voltage source device 47. The third voltage source 54 includes fifth and sixth PMOS current source devices 51, 53, together with a third PMOS voltage source device 57.
Bias voltages and currents are generated from a separate circuit (not shown and within the level of skill in the art to implement) and supplied to the first and second amplification stages 12, 14 via a multiconductor bus 45. Further bias voltages are then derived by circuitry in these stages to bias the gates of the current source devices 31, 33, 41, 43, 51, and 53. First, second, and third reset lines 39, 49, 59 and the latching stage gate signal 129 are driven by a signal supplied via control bus 55, to which they are all connected.
Referring to FIGS. 2 and 5, the input gain stage circuit now will be discussed further. The first current source 70 is implemented with two serially-connected PMOS devices 140, 142, which are biased by reference to first and second separately generated bias currents I144 and I146, respectively, on leads 144, 146 from a bias bus 148.
The second current source 80 is implemented with two more serially-connected PMOS devices 150, 152, which are also biased by reference to the first and second separately generated bias currents I144, I146. The third current source 82 is implemented with two further serially-connected PMOS devices 154, 156, which are also biased with reference to the first and second separately generated bias currents I144, I146. An offset or hysteresis may be introduced to the stage output by making the second and third current sources provide unequal currents. A series of PMOS devices 180, 182, 184, 186, and 188 derive on lines 141, 143 the bias for these three current sources, based on the separately generated bias currents. The first and second bias voltages on lines 96, 98 are generated by four PMOS devices 158, 160, 162, 164 and six NMOS devices 166, 168, 170, 172, 174, 176.
The first and second PMOS input devices 62, 64 are actually formed physically of two devices each. In particular, the first device 62 is made up of first and second parallel-connected PMOS devices 190, 192. Similarly, the second input PMOS device is made up of two parallel-connected PMOS input devices 194, 196.
A pair of PMOS clamping devices 197, 198 is provided across the output of the gain stage circuit. The drain of the first output clamping device is connected to the stage's first output 84 and to the gate of the first output clamping device. The source of the first output clamping device is connected to the stage's's second output 86. Similarly, the drain of the second output clamping device 198 is connected to the stage's second output and to the gate of the second output clamping device. The source of the second output clamping device is connected to the stage's first output. These clamping devices are each PMOS devices, and are therefore matched with the PMOS load devices 76, 78 and the PMOS voltage source devices 37, 47, 57 (see FIGS. 1 and 3).
This matching is advantageous in low power operation, as it leads to cancellation of variations due to temperature changes, power supply changes, or processing variations. For example, an increase in the threshold voltage of the load transistors 76, 78 would lead to a larger output swing of the gain stage. This increased swing will be accompanied by an increase in the common mode voltage, because the threshold voltage of the voltage source devices 37, 47, 57 (see FIG. 4) will also be increased. The result is that the increased voltage swing will not cause the output of one stage to go more negative than the second supply line 35.
The clamping PMOS devices restrict the output swing of the stage to approximately .+-.0.7 V. The addition of these clamping devices reduces the response time of the comparator and allows the common-mode (CM) voltage at each stage input to be set at approximately 1 V. Another advantage of this architecture is that the load devices exhibit an inherent clamping action for negative output transitions because of their square-law behavior.
Referring to FIGS. 3 and 6, the latching stage will be further discussed. The first and second current sources 106, 108 are formed by two pairs of PMOS devices 204, 206 and 200, 202, respectively. These current sources also share a series of five more PMOS devices 216, 218, 220, 222, 224, which receive the first and second bias currents I144, I146 provided by way of the bias bus 148.
Two modified CMOS NOR gates 290, 292 are also shown in the detailed schematic. The first NOR gate is made up of a series of PMOS and NMOS devices 258, 260, 262, and is followed by an inverter 264, 266, a second NOR gate 268, 269, 270, and a second inverter 272, 274. Another pair of inverters 292 is provided for delay and is made up of four PMOS and NMOS devices 280, 282, 284, and 286.
The above-described comparator circuitry is particularly useful in connection with a monolithic successive approximation analog-to-digital converter. Further analog and digital circuitry in this analog-to-digital converter interacts with the above-described circuitry in performing analog-to-digital conversion operations. For example, referring to FIG. 4, this circuitry is responsible for generating control signals such as the reset signals 39, 49, 59 and the latching stage gate signal 129 on the control bus. It is also responsible for tying the inputs 15, 17 of the input stage 12 together during an offset memorization phase. An analog-to-digital converter which employs comparator circuitry according to the present invention is discussed in my co-pending application titled "Low-voltage" "CMOS Analog-to-Digital Converter", filed on the same day as the present application, and herein incorporated by reference. Also particularly useful in connection with the present application are the calibration methods and circuitry described in a co-pending application entitled "Charge Redistribution Analog-to-Digital Converter with System Calibration", filed on the same day as the present application, and herein incorporated by reference.
While there have been shown and described various embodiments of the present invention, such embodiments are provided by way of example only and it will be obvious to those skilled in the art that various changes and modifications may be made therein without departing from the scope of the invention as defined by the appended claims.