Method and system for reducing hazards in a flip-flop

[0002] This application claims priority from and incorporates by reference in its entirety U.S. Provisional Application Serial No. 60/323,198, titled “A Method And System For Reducing Hazards In A Flip-Flop” by Nikola Nedovic, et. al., filed Sep. 17, 2001.

[0003] The invention relates generally to the field of electronic circuit design, and in particular to techniques for reducing hazards in a digital logic circuit, for example, a digital logic flip-flop circuit.

[0004] The Semi-Dynamic Flip-Flop (SDFF) is one of the high performance flip-flops based on the hybrid concept. In part due to its size, low clock-to-output delay, negative set-up time, and simple topology, it is considered to be one of the fastest flip-flops today. However, the SDFF is susceptible to a hazard condition, when both the input and output are at a high logic value.

[0005] FIG. 1 shows a schematic circuit diagram of a typical prior art SDFF. The data input is D 312, the clock signal is CLK0314, and the outputs are Q 316 and Qbar 317. The two inverters inv5318 and inv6319 are a “keeper” circuit which maintains the value of output Qbar 317 and hence output Q 316. A transparency window for the SDFF is given by the propagation delay of the two inverters, inv1350 and inv2352, and the NAND gate 354. The internal node X 320 of the first stage 330 of the SDFF is set to a high logic level (H), when the clock CLK0314 is at a low logic level (L), i.e., the first stage 330 is pre-charged. When the input signal D 312 is H, node X 320 transitions from H to L in the transparency window where both CLK0314 and S 356 are H (transistors Mn1346, Mn2344, and Mn3342 are on). The second stage 332 captures the transition on node X 320 generated by the first stage 330 and produces output Q 316. In this case node X 320 sets output Q 316 to H via transistor Mp2374. If input D 312 is L, Mn2344 is off and node X 320 remains high during the transparency window. With node X 320 at H, output Q 316 is set at L during the transparency window (transistors Mn4370 and Mn5372 are on).

[0006] FIG. 2 is an example timing diagram for the SDFF schematic circuit diagram of FIG. 1 showing a glitch in the output. The timing diagram shows the clock signal CLK0410 representing the CLK0314 in FIG. 1. D414, X 416, and Q 418 show the signals for D 312, node X 320, and Q 316 in FIG. 1 respectively. From FIG. 2, after the rising edge 430 of CLK0410 and with D 414 set to L, X 416 remains at H and output Q 418, due to transistors Mn4370 and Mn5372, transitions from H to L 434. After another rising edge 440 of CLK0410 and with D 414 at H, X 416 transitions from H to L 442 due to transistors Mn1346, Mn2344, and Mn3342 turning on. Next output Q 418 transitions from L to H 444 due to transistor Mp2374. Thus the L to H transition of output Q, e.g., 444, is done using in effect an inverting intermediate node X 320, while the transition of output Q, e.g., 434, from H to L is done directly via nMOS transistors and avoids the slower pMOS transistors. The SDFF is used where the time critical output transitions are from L to H, e.g., 444, on output Q 316, and thus the node X transition, e.g., H to L 442, is important.

[0007] However, the asymmetrical transition times of the SDFF lead to a “static-one-hazard” at the output Q when both input D and output Q are H. In FIG. 2, before the rising edge 450 of the CLK0410, X 416 is set (or reset) to H by transistor Mp1340. Because the first stage 330 has a non-zero propagation delay from the time of the rising clock edge 450 to the time X 416 transitions from H to L 454, the second stage 332 uses the previous X (H). Hence during the time window between the rising edge 450 of the clock CLK0410 and the falling edge 454 of X 416, both Mn4370 and Mn5372 in FIG. 1 are on and the output Q 316 is pulled to low logic level (e.g., transition 452). After the propagation delay, i.e., the falling transition 454 of X 416, the transistor Mp2372 turns on (and Mn4370 turns off), and the output Q 316 is pulled to H (e.g., transition 456). Thus a glitch 462 is caused on the output Q 418 (and Qbar 420) and makes the use of the SDFF hazardous. In addition the glitch consumes power unnecessarily, as output Q 316 should not change, since input D 312 has not changed.

[0008] There is also a problem of power consumption in the unconditional keepers of the SDFF (back-to-back inverters, inv3360 and inv4362, and back-to-back inverters, inv5318 and inv6319, of FIG. 1). The keeper is used to hold the value of a dynamic node, e.g., node X 320 or output Q 316, that would otherwise be in high impedance and thus sensitive to leakage current effects and noise, especially in low-power applications where clock gating techniques are typically employed. The problem is that in order to change the value of the dynamic node, the keeper (two keepers, in the case of the SDFF) has to be overpowered, i.e., the output logic level of the keeper needs to be switched, which increases power consumption.

[0009] The power consumption and hazard problems associated with the SDFF, are demonstrative of the same or similar problems with hybrid-type flip-flops in general. Another example of a flip-flop having the same or similar type problems is the Hybrid Latch Flip-Flop (HLFF).

[0010] Therefore with the problems of hazard and power consumption with the conventional hybrid-type flip-flop, e.g., SDFF, there is a need for an improved flip-flop with fewer problems, such as fewer hazards or no hazards at all.

[0011] The present invention provides techniques, including a system and method, for reducing hazards in a conventional flip-flop, having a pre-charged stage coupled to an evaluation stage, by allowing a change in the pre-charged stage to settle before the evaluation stage processes the change. One embodiment has a substantially similar delay as the SDFF with fewer hazards and a significant reduction in power consumption.

[0012] Broadly, the present invention provides a method that reduces hazards in a flip-flop. In one embodiment this method includes a delayed reset of the output Q. The setting of the output Q, i.e., the second or evaluation stage, is disabled from being set to a low logic level by a delayed clock signal. This leaves time for the internal node X, i.e., output of the first or pre-charged stage, to transition from the high to the low logic level after the rising edge of the clock, without affecting the output Q. Consequently, the glitch that appears at Q in case of SDFF, when both the input D and output Q are at a high logic level, is prevented. In addition, a delayed pre-charge of the first stage is provided in order to prevent another hazard on the output Q. This hazard may occur because the second stage remains enabled until the delayed clock is pulled low. Hence, if the internal node X quickly charges before the second stage is disabled, a glitch on or false switching of output Q may occur. Thus, this embodiment provides a reduced-hazard flip-flop.

[0013] In another aspect of the present invention a method is provided for reducing hazards in a flip-flop, with the method including a pre-charged stage coupled to an evaluation stage by at least an internal node. First, the pre-charged stage sets the internal node based on a data input. The evaluation stage is prevented from evaluating the internal node for a predetermined time period. After the predetermined time period, the internal node is evaluated by the evaluation stage to determine an output of the flip-flop.

[0014] Yet another aspect of the present invention provides a system for reducing hazards in a hybrid flip-flop including: a pre-charge stage for determining a pre-charge stage output depending upon a data input during a transparency window; and an evaluation stage for evaluating the pre-charge stage output to produce a data output during the transparency window, where the evaluation stage is enabled to change the data output to a low logic level only after the pre-charge stage determines the pre-charge stage output. The transparency window includes a time period when the pre-charge stage output is logically equivalent to an inverted value of the data input. Optionally, after the transparency window, the pre-charge stage output is restored to a default logic level, when the evaluation stage is next disabled.

[0015] The present invention also includes a system for reducing hazards in a hybrid flip-flop. The system includes a pre-charge stage for determining a pre-charge stage output depending upon a data input during a transparency window; and an evaluation stage for evaluating the pre-charge stage output to produce a data output during the transparency window, where the evaluation stage is enabled to change the data output after the pre-charge stage determines the pre-charge stage output.

[0016] A further aspect of the present invention provides a method for reducing hazards in a flip-flop, including, a pre-charged stage coupled to an evaluation stage by at least an internal node. First the method determines a first logic level for the internal node based on a data input to the pre-charged stage within a first section of a transparency window. The evaluation stage is disabled from setting a data output to a second logic level during the first section of the transparency window. Next the data output is determined during a second section of the transparency window based on the first logic level, where the second section follows the first section.

[0017] Another embodiment of the present invention provides a method for reducing power consumption in a hybrid flip-flop, including a pre-charged stage coupled to an evaluation stage by at least an internal node. First, an input of the pre-charged stage is set to a high logic level. The output of the evaluation stage also has a high logic level. Responsive to the setting of the input, the internal node is set to a low logic level within a beginning part of a transparency window. The evaluation stage is disabled from setting the output to a low logic level, during the beginning part of the transparency window. And the evaluation stage maintains the high logic level on the output during a remainder of the transparency window, using the low logic level of the internal node.

[0018] A further embodiment of the present invention provides a system for reducing hazards in a hybrid flip-flop including: a pre-charge stage for determining a pre-charge stage output depending upon a data input during a first part of a transparency window. The pre-charge stage includes a first conditional keeper for keeping the pre-charge stage output. And an evaluation stage for evaluating the pre-charge stage output to produce a data output during a second part of the transparency window, where the evaluation stage includes a second conditional keeper for keeping the data output.

[0019] These and other embodiments, features, aspects and advantages of the invention will become better understood with regard to the following description, appended claims and accompanying drawings.

[0020] FIG. 1 shows a schematic circuit diagram of a typical prior art SDFF;

[0021] FIG. 2 is an example timing diagram for the SDFF schematic circuit diagram of FIG. 1 showing a glitch in the output;

[0022] FIG. 3 illustrates a simplified circuit diagram of a flip-flop of a first embodiment of the present invention;

[0023] FIG. 4 shows an example of a simplified timing diagram for the embodiment of FIG. 3;

[0024] FIG. 5 is a schematic circuit diagram of a second embodiment of the present invention;

[0025] FIG. 6 is a schematic circuit diagram of a third embodiment of the present invention with an unconditional (i.e., conventional) keeper at the output;

[0026] FIG. 7 is a schematic circuit diagram of a fourth embodiment of the present invention with an unconditional (i.e., conventional) keeper at both the internal node X and at the output Q;

[0027] FIG. 8 is a schematic circuit diagram of a fifth embodiment of the present invention with an unconditional (i.e., conventional) keeper at the output;

[0028] FIG. 9 is a schematic circuit diagram of a sixth embodiment of the present invention with an unconditional (i.e., conventional) keeper at both internal node X and at the output Q;

[0029] FIG. 10 is a schematic circuit diagram of another embodiment of the present invention with the keeping function of the second stage performed by a feedback NAND gate;

[0030] FIG. 11 is a schematic circuit diagram of yet another embodiment of the present invention with the keeping function of the second stage performed by a feedback NAND gate and an unconditional keeper in the first stage; and

[0031] FIGS. 12-1, 12-2, 12-3 and 12-4 show examples of the second stages of two flip-flops being combined.

[0032] In the following description, numerous specific details are set forth to provide a more thorough description of the specific embodiments of the invention. It is apparent, however, to one skilled in the art, that the invention may be practiced without all the specific details given below. In other instances, well known features have not been described in detail so as not to obscure the invention.

[0033] FIG. 3 illustrates a simplified circuit diagram of a flip-flop of one embodiment of the present invention. The D-type flip-flop 510 shown has input D 512, clock CLK0514, and output Q 516 (inverted output Qbar 518). Clock CLK1566 is the clock CLK0514 delayed by the inverter inv1520. Clock CLK2524 is the clock CLK0514 delayed by the inverters inv1520 and inv2522. Clock CLK3528 is the clock CLK0514 delayed by the three inverters inv1520, inv2522, and inv3526. There is an internal node X 530 which connects the first or a pre-charge stage 550 to the second or evaluation stage 560. In addition, the clock CLK2524 is also input into the second stage 560.

[0034] There are two conditional keepers that in general maintain the value of the outputs of the first and second stages mostly outside of the transparency window, where the transparency window includes the delay of the three inverters inv1520, inv2522, and inv3526. The first conditional keeper maintains output Q 516 (and Qbar 518) and includes inverter inv5562 connected to the inverted tri-state 564. The inverted tri-state 564 has outputs: high logic level (H), low logic level (L), and open circuit, i.e., high impedance (Z). The inverted tri-state 564 is conducting when both the clock CLK1566 is H and the clock CLK2524 is L. The inverted tri-state 564 is open or high impedance (Z), when either CLK1566 is L or CLK2524 is H. The second conditional keeper maintains node X 530 and includes inverter inv4532 connected to the inverted tri-state 534. The inverted tri-state 534 is conducting when both the clock CLK2524 is H and the clock CLK3528 is L. The inverted tri-state 534 is open or high impedance (Z), when either CLK2524 is L or CLK3528 is H.

[0035] The basic operation of FIG. 3 is explained as follows: when CLK0514 goes H, CLK2524 disables the ability to set the output Q 516 of the evaluation or second stage (via transistor Mn5570) from being set L for a predetermined time period, e.g., the delay of the two inverters inv1520 and inv2522, so that the output of the pre-charge or first stage 550, i.e., node X 530, can be set. Then after node X 530 is set, output Q 516 is set. This prevents the glitch 462 of the SDFF illustrated in FIG. 2 from occurring. In addition, CLK2524 via transistor Mp2546 delays the pre-charging of node X 530 to its default H value, until after the evaluation stage 560 is disabled from being set L. This prevents the fast charging of node X 530, when the CLK0514 goes L, from affecting the output Q 516.

[0036] FIG. 4 shows an example of a simplified timing diagram for the embodiment of FIG. 3. The signals shown in FIG. 4, i.e., clock CLK0612, clock CLK2616, clock CLK3618, input D 619, X 620, output Q 622, and output Qbar 624, correspond respectively to clock CLK0514, clock CLK2524, CLK3528, input D 512, node X 530, output Q 516, and output Qbar 518 in FIG. 3. For illustration purposes only, there are three clock CLK0612 cycles shown, cycle 630, cycle 632, and cycle 634. An example of a transparency window is the time interval from the rising edge 662 of clock CLK0612 to the falling edge 664 of clock CLK3618, i.e., both CLK0612 and CLK3618 are H. An example of a disable window 660 is shown for the time period from after the clock CLK0612 rising edge 662 to after the rising edge 666 of clock CLK2616. During this disable window the evaluation stage 560 is disabled from having its output set to L. X 620 needs to transition from its default (H) to L, e.g. transition 672, in this disable window. In other embodiments the disable window may be from before, during or after the rising edge 662 of clock CLK0612 to before, during or after the rising edge 666 of clock CLK2616. The main criteria for setting the disable window is that X 620 and CLK2616 are not both simultaneously at a high logic level (H) in this disable time period. Otherwise, Mn5570 and Mn6572 may both be on, pulling output Q 516 to L and potentially causing a glitch on the output Q 516 like the SDFF circuit.

[0037] In FIG. 4 cycles 632 and 634 show the operation of the D type flip-flop 510 in changing the output Q 622 from H to L 656 and from L to H 646. The rising edge 642 of cycle 634 of clock CLK0612, when D 619 is H 640, causes transistor Mn1540 to turn on and transition X 620 from H to L 644 via transistors Mn1540, Mn2542, and Mn3544. Next output Q 622 is pulled to H (transition 646) by transistor Mp3574. The rising edge 650 in cycle 632 of clock CLK2616, when D 619 is L 652, maintains H 654 on X 620, and causes transistor Mn5570 to pull output Q 622 to L (transition 656). Thus like the H to L transition 434 of output Q 418 (FIG. 2) of the SDFF, the H to L transition 656 of the output Q 622 of this embodiment is done directly, i.e., does not involve a transition of X 620.

[0038] For cycle 630 of clock CLK0612 the rising edge 662 causes transistor Mn1540 to turn on. When the input D 619 is H 670, node X 620 is pulled to L via transistors Mn1540, Mn2542, and Mn3544, e.g., transition 672. This H to L transition 672 is fast and is within the disable period 660. Next, the L to H transition 666 of clock CLK2616 turns Mn5570 on and since X 620 is L, Q 622 is set to H via transistor Mp3574. As Q 622 is already H 674, no change on Q occurs. Thus there is no glitch on output Q 622 (or Qbar 624) in the area 678.

[0039] Referring to FIG. 3 and FIG. 4, the node X conditional keeper, i.e., inverter inv4532 and inverted tri-state 534, maintains the node X 530 logic value (H or L). For example in FIG. 4, the falling edge 664 on CLK3618 turns the inverted tri-state 534 on (CLK2616 is H at falling edge 664), and hence the node X conditional keeper is turned on. Before the falling edge 664 the inverted tri-state 534 and thus the node X conditional keeper is an open circuit or off. Thus a change in X 620, for example, transition H to L 672, can occur within the transparency window without consuming the power necessary to change the state of a conventional keeper, i.e., fighting the keeper.

[0040] The output Q conditional keeper, i.e., inverter inv5562 and inverted tri-state 564, maintains the output Q 516 and Qbar 518. For example, in FIG. 4, the falling edge (not shown) on CLK1 is one inverter delay after the rising edge 662 on CLK0612. This falling edge turns the inverted tri-state 534 off, and hence the output Q conditional keeper is turned off. Thus the output Q conditional keeper is off during about the second half of the disable window 660 and the remainder of the transparency window. Therefore the output Q 516 may be set without consuming the power necessary to change the state of a conventional keeper.

[0041] Generally with some overlap in the transition window, the node X conditional keeper and the output Q conditional keeper alternate, with one being on while the other is off (i.e., open circuit). When the node X conditional keeper is on, CLK2616 is H, and Mn5570 is on. The transistors Mn6572 and Mp3574 then function as an inverter, passing an inverted node X to output Q 516. Hence the node X conditional keeper maintains the output Q 516. Thus one of the conditional keepers maintains the output Q outside of the transparency window.

[0042] Another feature of the embodiment of FIG. 3 is the delayed pre-charge of node X 530. This prevents another hazard on the output Q 516, when node X 530 transitions from L to H, i.e., node X 530 is pre-charged. For example, when CLK0612 transitions from H to L (transition 680), transistor Mp1548 is turned on. If Mp1548 was directly connected to node X 530 (no intervening transistor Mp2546), node X 530 could pre-charge to H before CLK2524 turns off transistor Mn5570. If output Q 516 is at H, then node X 530 via path Mn6572 and Mn5570, could cause a false switching or glitch (i.e., an unwanted voltage or current spike of short duration, usually unintentionally produced) on output Q 516. Transistor Mp2546 controlled by clock CLK2524 insures that transistor Mn5570 is turned off before node X 530 is pre-charged by transistor Mp1548.

[0043] FIG. 5 is a schematic circuit diagram of one embodiment of the present invention. The D type flip-flop 710 has two functional stages: the first or pre-charge stage 740 and the second or evaluation stage 742. The first stage 740 produces a signal at internal node X 734, and the second stage 742 produces the outputs Q 716 and Qbar 718. The first stage 740 is pre-charged (set to the default high logic level(H)) when the input clock CLK0714 is at the low logic level(L). The first stage 740 allows a closed circuit path (Mn3750, Mn2752 and Mn1754) from node X 734 to ground in the transparency window (i.e., when CLK0=CLK3=H), if the data input D 712 is H. This turns transistor Mp5760 on and sets the output Q 716 to H. Once the internal node X 734 is pulled to L, transistors Mn4766 and Mn5768 keep node X 734 at L for the duration of the high logic level(H) of the delayed clock CLK2730.

[0044] If the input D 712 is L in the transparency window, node X 734 remains at H. The H on node X 734 is kept (restored) after the transparency window, when CLK3730 goes to L. The H on node X 734, after the rising edge of the delayed clock CLK2730 causes the output Q 716 to be pulled to L (transistors Mn6764 and Mn7762 are on). During the time the input clock CLK0714 is at L, and when the forward path (transistors Mn6764, Mn7762, Mp5760) is off, the output Q 716 is kept at its logic level by the conditional keeper (transistors Mn8770, Mn9772, Mp7774, Mp6776 and inverter inv5778).

[0045] The SDFF glitch occurrence (462 in FIG. 2) is prevented by delaying the reset of the output Q 716, i.e., the output Q 716 is connected to ground when the delayed clock CLK2730 goes to H. This allows the node X 734 to evaluate to L, when input D 712 is H, before node X's transient high level is used to reset the output Q 716. However, this solution may create another potential hazard. When the clock CLK0714 transitions from H to L, a fast pre-charge of node X 734 (assuming transistor Mp2780 is absent and Mp1782 is connected directly to node X 734) and slow propagation of the low logic level of the clock through the inverters inv1720 and inv2722 would cause the overlap between the node X 734 at H and CLK2730 at H. As a result, the transistors Mn7762 and Mn6764 would turn on, and another glitch, or even a false switching of the node Q could potentially occur. This is avoided by the delayed pre-charge of the node X 734 via transistors Mp2780 and Mn4766. The transistor Mp2780 inhibits the pre-charge of the first stage 740 before the falling edge of the delayed clock CLK2730. The role of the transistor Mn4766 is to additionally ensure the low level of the node X 734 when CLK2730 is at H. Without the transistor Mn4766, there would be no path from the node X 734 to ground in the timing window between falling edges of input clock CLK0714 and delayed clock CLK2730. Therefore, the node X 734 would remain at L (preventing the false switching of the flip-flop), but it would be more sensitive to noise in the flip-flop environment. Since the size of the transistor Mn4766 is minimal, it adds additional robustness to the design with minimal performance penalty.

[0046] In sum, the embodiment of FIG. 5 shows both a delayed reset of the output Q and a delayed pre-charge of the first stage. The delayed reset is achieved by controlling the forward path from Q to the ground and by delaying the clock signal (CLK2). This leaves time for the internal node X to transition to the low level after the rising edge of the clock CLK0. The output Q is disabled until the correct value of node X is set. Consequently, the glitch that appears at Q in case of SDFF, when the H to L transition is controlled by input clock CLK0, is prevented.

[0047] This delayed reset causes the need for a delayed pre-charge of the first stage, since the second stage forward path to the ground remains open until the delayed clock (CLK2) is pulled to L. Therefore, the pre-charge of internal node X should not start before CLK2 goes to L. This is achieved by the pre-charge pMOS transistors Mp1 and Mp2 and the keeping nMOS transistor Mn5, that provides keeping node X at L after the falling edge of clock CLK0 and before the falling edge of delayed clock CLK2.

[0048] In FIG. 5 the conditional keeper of node X keeps or maintains the low level of node X, when CLK2 is H, and keeps the high level of node X, when CLK3 is L. The conditional keeper of output Q keeps the low level of output Q, when CLK1 is H, and keeps the high level of output Q, when CLK2 is L. Each conditional keeper is off during some or all of the transparency window in order to allow node X or output Q to change without fighting their respective keepers. In one embodiment, the conditional keeper of node X is off during a beginning portion of the transparency window and the conditional keeper of output Q is off during an ending portion of the transparency window, where the beginning and ending portions may overlap either partially or fully.

[0049] Table 1 below shows the SPICE simulation results of the embodiment of FIG. 5 with the conditional keepers and the removal of the glitch, as normalized to the conventional SDFF. The input data activity is 50%.

[0050] The EDP is the Energy Delay Product. The EDP is a figure of merit and in this case is the total power dissipated (P_diss) times the delay (t_d) divided by the clock frequency (f_CLK0), e.g., 500 MHz. The delay includes the flip-flop setup time+clock-to-output time. The un-normalized EDP formula is thus: $\mathrm{EDP}=\frac{t_d\times P_{\mathrm{diss}}}{f_{\mathrm{CLK0}}}$

[0051] As Table 1 illustrates the embodiment of FIG. 5 has substantially the same delay as the SDFF, but consumes significantly less power.

[0052] In another embodiment, the delayed reset/pre-charge technique described above may be used together with transistor sizing to equalize the high-to-low and/or low-to-high delay for the optimum set-up times.

[0053] FIG. 6 is a schematic circuit diagram of a second embodiment of the present invention with an unconditional (i.e., conventional) keeper at the output. The embodiment uses the delayed reset technique to avoid the glitch at the output Q. A partially conditional keeper in the first stage 860, that includes transistors Mp3820 and Mn5824 and inverter inv4822, is used. Node X 830 keeps a high logic level unconditionally, and keeps a low logic level only if the input clock CLK0832 is H, i.e., the slow pre-charge operation is performed without the contention with the keeper. The evaluation path of the node X 830 (transistors Mn1840, Mn2842, and Mn3844) has to overpower the high-level keeper (transistor Mp3820), i.e., the nMOS transistors need to pull enough current to switch node X 830 to L. While this may slow down the transition of node X and increase power consumption of the flip-flop, it may improve noise immunity of the flip-flop. The size of the transistor Mp3820 can be used to tradeoff the noise immunity for the performance of the flip-flop, when needed.

[0054] The second stage 870 of the flip-flop 810 is made fully static with an unconditional keeper (inv6850 and inv5852). The delayed pre-charge of node X 830 is accomplished via the transistors Mp2846 and Mn4848 similar to the embodiment of

[0055] FIG. 7 is a schematic circuit diagram of a third embodiment of the present invention with an unconditional (i.e., conventional) keeper at both the internal node X and at the output Q. FIG. 7 is similar to FIG. 6 where the second stage 970 is the same as or similar to the second stage 870 of FIG. 6. The first stage 960 has an unconditional (conventional) keeper (back-to-back inverters, inv5930 and inv4932) in place of the partially conditional keeper in the first stage 860 of FIG. 6. The operation of FIG. 7 is similar to that of FIG. 6.

[0056] FIG. 8 is a schematic circuit diagram of a fourth embodiment of the present invention with an unconditional (i.e., conventional) keeper at the output. FIG. 8 is similar to FIG. 6, except the transistors Mn21022 (for input D 1020) and Mn31032 (for clock CLK31030) have been switched (In FIG. 6 Mn2842 for CLK3843 is below Mn3844 for input D 834). The embodiment of FIG. 6 with the input D 834 at the top of the first stage 860 nMOS transistor stack minimizes the data-to-output time at the optimum set-up time. The flip-flops in FIGS. 6 and 8 may have negative optimum set-up times. Thus while FIG. 8 illustrates another embodiment of the present invention, the use of the embodiment of FIG. 6 is preferred.

[0057] FIG. 9 is a schematic circuit diagram of a fifth embodiment of the present invention with an unconditional (i.e., conventional) keeper at both internal node X and at the output Q. FIG. 9 is similar to FIG. 7, except the transistors Mn21122 (for input D 1120) and Mn31132 (for clock CLK31130) have been switched (In FIG. 7 Mn2942 is below Mn3940 in the input nMOS stack). The embodiment of FIG. 7 with the input D 944 at the top of the first stage 960 nMOS transistor stack minimizes the data-to-output time at the optimum set-up time. The flip-flops in FIGS. 7 and 9 may have negative optimum set-up times. Thus while FIG. 9 illustrates another embodiment of the present invention, the use of the embodiment of FIG. 7 is preferred.

[0058] FIG. 10 is a schematic circuit diagram of another embodiment of the present invention with the keeping function of the second stage performed by a feedback NAND gate. The first stage 1220 is similar to and has the function of first stage 860 in FIG. 6. Node X 1215 and clock CLK11240 (clock CLK01214 inverted and delayed through inverter inv11230) are outputted from the first stage 1220 and inputted into the second stage 1222. The second stage 1222 has the delayed reset function of CLK2 (the clock reference signal seen by the second stage is delayed through the inverter inv1 and the NAND gate NAND11242). When output Q 1216 is H, then NAND11242 functions as an inverter for CLK11240 and the NAND output 1243 is CLK2. When output Q 1216 is L, NAND output 1243 is H, transistor Mn71252 is on, and output Q 1216 gets the inverted value of node X 1215 via transistors Mn61250 and Mp41254.

[0059] The embodiment of FIG. 10 provides delay improvements. These delay improvements are from reducing the complexity of the layout of the critical path, so that the parasitic drain diffusions and contact capacitances in the critical path are minimized.

[0060] The sizing of the transistors Mn61250, Mn71252, Mp41254, and Mp51256 in FIG. 10 can be used to improve the driving capability of the output Q, compared to the conventional flip-flop. This sizing allows the flip-flop 1210 of FIG. 10 to drive large loads even when noise spikes occur, because the strong NAND gate NAND11242 is much less sensitive to noise compared to the weak keeper of the second stage of the conventional flip-flop, e.g., FIG. 1 inv5318 and inv6319. In a noisy environment, the typical component following the flip-flop that needs the output of the conventional flip-flop, e.g., FIG. 1, uses Qbar 317 and inverts it again, rather than using Q 316. Qbar 317 is used in order to prevent noise from being injected back into output Q 316, overpowering the weak keeper, and undesirably switching the output Q 316. However, this incurs an additional delay of two inverters, over using Q directly or only one inverter, if the consumer, is willing to use Qbar instead of Q. The use of the NAND gate NAND11242 in FIG. 10 allows for output Q 1216 to be used directly.

[0061] FIG. 11 is a schematic circuit diagram of yet another embodiment of the present invention with the keeping function of the second stage performed by a feedback NAND gate and an unconditional keeper in the first stage. FIG. 11 has the same second stage 1340 as the second stage 1222 of FIG. 10. The first stage 1320 is similar to the first stage 1220 of FIG. 10, except the conditional keeper in the first stage 1220 of FIG. 10 has been replaced by an unconditional keeper, i.e., back-to-back inverters inv41324 and inv51322.

[0062] From looking at the second stages of FIGS. 10 and 11 above, it can be seen that other digital logic may be embedded in the second stage of the flip-flop. For example, in the second stage 1222 of FIG. 10, the transistors Mn61250, Mn71252, Mp41254, and Mp51256 form a static CMOS NAND gate with two inputs node X 1215 and the output 1243 of NAND11242, and one output, output Q 1216. It is easier to embed an additional digital logic circuit directly into the second stage 1222 of FIG. 10 using its static CMOS NAND gate, rather than waiting for the output Q 1216 of the second stage 1222 to be evaluated first, and then using the output Q 1216 in the additional digital logic circuit.

[0063] FIGS. 12-1, 12-2, 12-3 and 12-4 show examples of the second stages of two flip-flops being combined. For example, the circuits in FIGS. 12-2 and FIGS. 12-4 show the equivalent of the circuits shown in FIGS. 12-1 and FIGS. 12-3, respectively.

[0064] FIGS. 12-1 has two D-type flip-flops. The first flip-flop has inputs D11410 and CLK01412 into the first stage 1414, an internal node X11416 that connects the first stage 1414 to the second stage 1420, and output Q11422. The second flip-flop has inputs D21430 and CLK01432 into the first stage 1434, an internal node X21436 that connects the first stage 1434 to the second stage 1440, and output Q21442. The outputs Q11422 and Q21442 are combined in an OR gate 1450 to give output 1452.

[0065] FIGS. 12-2 has the logical equivalent circuit of FIGS. 12-1 with the two second stages 1420 and 1440, and the OR gate 1450 combined. The outputs X11416 and X21436 of the two first stages (1414 and 1434) are input along with the output of a feedback NAND gate 1462 (similar to NAND11242 of FIG. 10) into a three input NAND gate 1460. The output of the NAND gate 1460 is the combined circuit output OUT 1452 and is inputted into feedback NAND gate 1462, which also receives CLK11465, i.e., CLK01432 through inverter 1464.

[0066] FIGS. 12-3 has two D type flip-flops. The first flip-flop has inputs D11510 and CLK01512 into the first stage 1514, an internal node X11516 that connects the first stage 1514 to the second stage 1520, and output Q11522. The second flip-flop has inputs D21530 and CLK01532 into the first stage 1534, an internal node X21536 that connects the first stage 1534 to the second stage 1540, and output Q21542. The outputs Q11522 and Q21542 are combined in an AND gate 1550 to give output 1552.

[0067] FIGS. 12-4 has the logical equivalent circuit of FIGS. 12-3 with the two second stages 1520 and 1540, and the AND gate 1550 combined. The outputs X11516 and X21536 of the two first stages (1514 and 1536) are input into OR gate 1560. The output of OR gate 1560 is input along with the output of a feedback NAND gate 1566 (similar to NAND11242 of FIG. 10) are inputted into a two input NAND gate 1562. The OR gate 1560 and NAND gate 1562 form an Or-And-Invert logic (OAI) single stage circuit 1570. The output of the OAI 1570 is the combined circuit output OUT 1552 and is inputted into feedback NAND gate 1566, which also receives CLK11565, i.e., CLK01532 through inverter 1564.

[0068] When the additional digital logic function needs to be implemented anyway, the effective delay of the flip-flop becomes the difference between the delay of the flip-flop with additional logic embedded and the delay of the additional digital logic circuit alone. Thus the circuitry is embedded with minimal, if any, delay penalty. This embedding when done together with logic embedding in the first stage of the flip-flop, allows the entire flip-flop delay overhead to be masked.

[0069] Although specific embodiments of the invention have been described, various modifications, alterations, alternative constructions, and equivalents are also encompassed within the scope of the invention. While a transparency window of three inverters is given in some embodiments in this disclosure, the transparency window is not so limited. In an alternative embodiment there may be one or more inverters. More specifically there may be an odd number of inverters. The transparency window in other embodiments of the present invention may include any time duration in which both the pre-charge or first stage and the evaluation or second stage are active. For example, the transparency window in FIG. 4 is when both CLK0612 and CLK3668 are H, but other activation schemes as determined by one of ordinary skill in the arts may be used and are within the scope of this invention. In addition while the disclosed embodiments show a D-type flip-flop, the scope of this invention includes other flip-flops, latches, and memory circuits in which there is a first or pre-charge stage and an evaluation or second stage coupled together by at least one internal node, where this internal node is charged or discharged.

[0070] The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. It will, however, be evident that additions, subtractions, deletions, and other modifications and changes may be made thereunto without departing from the broader spirit and scope of the invention as set forth in the claims.