Noise tolerant wide-fanin domino circuits

1. Technical Field of the Invention

The present invention relates to circuits and, more particularly, to domino circuits.

2. Background Art

Technology scaling combined with aggressive design practices have made high performance digital circuits more susceptible to deep submicron noise. Among various noise disturbances, one major concern is leakage current degradation, which can become substantial in sub- 1 V, 0.1 μm technologies. This is primarily because the reduced supply voltage and high speed requirements force designers to use low threshold voltage (Vt) transistors. Reducing threshold voltage, on the other hand, increases the leakage current exponentially. Large leakage current results in higher DC offsets at the inputs of wide domino circuits, degrading the noise tolerance. The term noise immunity refers to the degree to which a circuit is noise tolerant.

Wide-fanin gates are gates having numerous input ports. Wide-fanin gates are routinely employed on critical delay paths of high-performance datapaths, such as in a microprocessor, digital signal processor, or other semiconductor device. Dynamic/Domino logic techniques have been used to achieve substantially higher performance than are provided by static complementary metal oxide semiconductor (CMOS) technology for wide-fanin gates.

As an example, FIG. 1 illustrates a conventional prior art wide-fanin OR domino gate or circuit 10 having a domino stage 12 and an output stage 14 joined by a node Q which carries a domino stage output signal. Domino stage 12 includes an evaluate network 16 , precharge and keeper p-channel metal oxide semiconductor field effect transistors (PMOSFET) M 2 and M 3 and an inverter 18 . Output stage 14 is illustrated as an inverter, but may be a dual function generator or other output stage. Evaluate network 16 includes eight n-channel metal oxide semiconductor field effect transistors (NMOSFET) M 1 - 0 . . . M 1 - 7 , the gates of which receive corresponding input signals Vin 0 . . . Vin 7 . A wide-fanin gate may have a greater or lesser number of inputs.

During a precharge phase, input signals Vin 0 . . . Vin 7 and a clock signal (Clk) are low (Vss). When Clk goes low, pull-up PMOSFET M 2 is turned ON and node Q is pulled high to a power supply node Vcc (sometimes called Vdd). As signal Q goes high, an inverter 18 turns on PMOSFET M 3 which keeps signal Q high after Clk transitions high and PMOSFET M 2 is off. During an evaluate phase, if each input remains low, Q remains high and the output signal Out at the output conductor 22 of output stage 14 remains low. If one or more of input signals Vin 0 . . . Vin 7 goes high, the corresponding NMOSFET(s) M 1 - 0 . . . M 1 - 7 is turned ON pulling signal Q low. When signal Q goes low, Out on conductor 22 goes high.

To improve noise tolerance, NMOSFETs M 1 - 0 . . . M 1 - 7 have a high Vt so that a greater noise voltage is required to turn on the transistor. However, this sacrifices pull-down speed. To enable the use of low Vt transistors for high speed, certain noise tolerance mechanisms have been proposed.

One such noise tolerant technique is illustrated in FIG. 2 . Referring to FIG. 2, a domino circuit 30 includes a domino stage 32 and an output stage 14 . Evaluate network 34 includes NMOSFET M 1 - 0 . . . M 1 - 7 , the gates of which receive corresponding input signals Vin 0 . . . Vin 7 . The sources of M 1 - 0 . . . M 1 - 7 are at a node X. A pull-up PMOSFET diode M 4 increases the source potential (node X) of evaluate network 34 . An NMOSFET M 5 isolates node X from ground during the evaluate stage. The Vt of an NMOSFET increases when it is reverse body biased, which occurs when the source voltage is greater than the body voltage. NMOSFET M 1 - 0 . . . M 1 - 7 have a grounded body. A reverse body bias, and hence increase in Vt, is created in NMOSFETs M 1 - 0 . . . M 1 - 7 , when node X is greater than ground, thus reducing the domino precharge node dip. Leakage is reduced when two off NMOSFET transistors are stacked in series. During evaluate stage, leakage through evaluate network 34 is reduced because NMOSFET M 5 is stacked when M 1 - 0 . . . M 1 - 7 , respectively. While circuit 30 is easy to implement, it suffers from a large static power dissipation through M 4 and M 5 during the evaluate phase.

Accordingly, there is a need for a better noise tolerant wide-fanin domino circuit technique.

The invention involves a die having domino circuits. In some embodiments, at least some of the domino circuits include an output stage and a domino stage including a domino stage output node coupled to the output stage. The domino stage includes a wide-fanin evaluate network including the domino stage output node and at least one intermediate node. The domino stage has improved noise immunity and reduced leakage through reverse body biasing transistors in the evaluate network by raising voltage of the at least one intermediate node without static power consumption through the evaluate network.

In other embodiments, at least some of the domino circuits include an output stage and a domino stage including a domino stage output node coupled to the output stage. The domino stage includes a wide-fanin evaluate network including the domino stage output node and wherein the domino stage further includes a diode transistor having a gate and an additional terminal connected to the domino stage output node.

Still other embodiments are described and claimed.

The invention will be understood more fully from the detailed description given below and from the accompanying drawings of embodiments of the invention which, however, should not be taken to limit the invention to the specific embodiments described, but are for explanation and understanding only.

FIG. 1 is a schematic representation of a prior art domino circuit.

FIG. 2 is a schematic representation of another prior art domino circuit.

FIG. 3 is a schematic representation of a domino circuit with a boosted source and a dual function generator according to some embodiments of the invention.

FIG. 4 is a schematic representation of a domino circuit with a dual stack evaluate network according to some embodiments of the invention.

FIG. 5 is a schematic representation of a domino circuit with a diode pull-up at the output of the domino stage according to some embodiments of the invention.

FIG. 6 is a schematic representation of a domino circuit similar to that of FIG. 3 but with body bias voltage circuitry.

FIG. 7 is a schematic representation of a domino circuit similar to that of FIG. 4 but with body bias voltage circuitry.

FIG. 8 is a schematic representation of a domino circuit similar to that of FIG. 5 but with body bias voltage circuitry.

FIG. 9 is block diagram representation of a die in a system that includes domino circuits according to one or more embodiments of the invention.

The present invention involves noise tolerant domino circuits. Various embodiments of the invention fall into three categories ( 1 ) domino circuits with a boosted source and a dual function generator, ( 2 ) domino circuits with a dual stack evaluate network, and ( 3 ) domino circuits with a diode pull-up at the output of the domino stage. In some embodiments, there is no static power consumption. In other embodiments, there is no static power consumption during the precharge phase and a relatively small amount of static power consumption during evaluate. In some embodiments, leakage is reduced through stacked transistors.

A. Embodiments with boosted source and a dual function generator

In some embodiments, the invention includes a boosted source evaluate network in combination with a dual function generator (DFG), which provides true and complementary output signals Vout and Vout*. For example, referring to FIG. 3, a boosted source wide-fanin domino OR circuit 50 includes a domino stage 54 and a DFG output stage 62 . Domino stage 54 includes an evaluate network 58 having a node Q. Node Q is an output of domino stage 54 and a signal on node Q is a domino stage output signal.

Evaluate network 58 includes n-channel field effect transistors (NFETs) (e.g., NMOSFETs) M 11 - 0 . . . M 11 -X, the gates of which receive data input signals Vin 0 . . . VinX, respectfully. Node D is a source node because it is at the sources of transistors M 11 - 0 . . . M 11 -X. Node D may also be called an intermediate node because it is between transistors Ml 1 - 0 . . . M 11 -X and a boost NFET transistor M 12 . Transistor M 12 is called a boost transistor because it helps boost the voltage of node D. In the illustrated embodiment, the body (or bodies) of transistors M 11 - 0 . . . M 11 -X are at ground (Vss), although they could be at another voltage bias. Although there is no single definition of wide-fanin, as used herein wide-fanin means at least four inputs. For example, X may be 7, such that circuit 50 is an 8-wide OR gate, or X may be more or less than 7. A wider fanin tends to result in greater power consumption and noise sensitivity problems. Although the boosted source domino circuit is illustrated as an OR gate, those skilled in the art having the benefit of this disclosure could easily implement it in other logic forms such as AND, NOR, and NAND.

During a precharge phase, a clock signal (Clk) goes low turning on a p-channel field effect transistors (PFETs) (e.g., PMOSFETs) M 13 , M 14 and M 15 to precharge nodes A, B, and Q high with a supply voltage (e.g., Vcc which is sometimes called Vdd). Input signals Vin 0 . . . VinX are low during precharge so that, absent noise, transistors M 11 - 0 . . . M 11 -X are off. The Clk signal is inverted through an inverter I 1 to produce a slightly delayed inverted clock signal at node C. Node C is at the gate of NFET M 12 and at the input to an inverter 12 . During precharge, with M 12 on, node D is at ground (Vss) and transistors M 11 - 0 . . . M 11 -X are zero biased.

The output of inverter 12 is referred to as Clksa (or clock sense amplifier), although the particular name for the signal is not important. Clksa could be called a delayed clock signal. While Clksa is low, a block NFET M 20 is off. With nodes A and Q high, NFET devices M 18 and M 19 are on. However, until blocking transistor M 20 is on, neither node A or B can discharge. In the particular embodiment of FIG. 3, charge means to raise voltage and discharge means to lower voltage. In other embodiments, the opposite could be the case.

During the evaluate phase, Clk goes high and transistors M 12 and M 13 turn off. After a brief delay Clksa goes high so M 20 turns on. Assuming each of the inputs Vin 0 . . . VinX remain low, node Q remains high and M 19 remains on. Node A discharges to ground. As node A falls, M 18 begins to turn off and M 17 begins to turn on pulling node B high. Vout and Vout* are true and complementary output signals of output stage 62 and domino circuit 50 . Vout is low and Vout* is high, on conductors 66 and 68 respectively, indicating that the inputs are all low. Note that 13 and 14 are not necessary but can increase the output voltage drive and allow the transistors of DFG 62 to be smaller.

Now assuming one or more of inputs Vin 0 . . . VinX goes high, charge redistributes between nodes Q and D, lowering the voltage of node Q so that M 19 turns off. However, because M 12 is off, node Q does not discharge full rail to ground. This increases the noise immunity because a higher Vin at one or more of M 11 - 0 . . . M 11 -X is needed to allow enough charge move between nodes D and Q to turn off M 19 . If node D were at ground, a smaller Vin could cause M 19 to turn off. With M 19 off, node A cannot discharge, but node B falls to ground. As node B falls to ground, M 16 turns on pulling node A high. (Note that M 16 is not necessary, but keeps the input to inverter 14 high.) Vout is high and Vout* is low indicating at least one of the input signals Vin 0 . . . VinX is high. Accordingly, DFG 62 acts as a sense amplifier responding to (amplifying) small swing signals on the output of domino stage 54 (node Q), to create full rail (full swing) output signals Vout and Vout*. (Of course, the DFG could use less than full rail inverters to produce low voltage swing outputs if desired.)

A purpose of the delay in M 20 turning on after the beginning of the evaluate phase is to allow time for charge from node Q to redistribute to node D during evaluate phase. However, the delay is not necessary and M 20 could be tied directly to the Clk signal. The delay between Clk and Clksa is tunable (selected through design). Merely as an example, the delay between Clk and Clksa might be {fraction (1/10+L )}th of a cycle or some other value. The delay could be dynamically controlled.

The relative sizing of transistors can affect the noise tolerance. Transistors M 16 , M 17 , M 18 , and M 19 may be sized so that if node Q is high, the node A side of DFG 62 is stronger than the node B side. If node Q is low, the node B side of DFG 62 is stronger than the node A side. For example, M 19 may be sized larger than M 18 . However, such sizing may not be required.

There is no static power consumption through evaluate network 58 because during precharge M 11 - 0 . . . M 11 -X are off and during evaluate M 12 is off. In some embodiments, the timing of the turning off of M 12 is such that none of M 11 - 0 . . . M 11 -X can turn on until M 12 is off. In other embodiments, one or more of M 11 - 0 . . . M 11 -X could be momentarily on before M 12 is off.

Further, if there is any leakage through one or more of M 11 - 0 . . . M 11 -X, the voltage of node D will increase creating a reverse bias on M 11 - 0 . . . M 11 -X, thereby increasing their threshold voltages making it necessary for noise at the input gate to have a higher voltage to turn on one or more of M 11 - 0 . . . M 11 -X. Reverse biasing reduces leakage. This leakage reduction is referred to herein as active leakage reduction. The amount of leakage is also reduced because M 11 - 0 . . . M 11 -X are stacked with M 12 . (Note that it was known by others before the present invention that stacked transistors result in substantially reduced leakage as compared to a single transistor.)

In circuit 50 , because of the boosted source, transistors M 11 - 0 . . . M 11 -X may have lower Vts than would otherwise be acceptable. This offers better delay characteristics than a high Vt implementation. Moreover, the circuit speed is determined to a great extent by the DFG, which can be tuned independently without severely interfering with the domino device sizes. The boosted source design does not necessarily increase fanin capacitance, and hence does not impose any negative impact on the driving capability of the previous datapath stages (not shown). Further, because circuit 50 can produce true and complementary outputs from a single input, it prevents the performance bottlenecks and other disadvantages imposed by dual-rail circuit implementations of wide-fanin OR gates.

B. Domino circuits with a dual stack evaluate network

In some embodiments, the invention includes a domino circuit with a dual stack wide-fanin evaluate network. For example, FIG. 4 illustrates a dual stack wide-fanin OR circuit 100 having a domino stage 104 and an output stage 108 . Output stage 108 may be various circuits including an inverter as in FIG. 1 or a dual function generator similar to or somewhat different from that of FIG. 3 (in which case it would include Out and Out* signals). Domino stage 104 includes an evaluate network 112 , pull-up PFET M 23 , and a keeper PFET M 24 and associated inverter 118 .

Evaluate network 112 includes upper NFETs M 21 - 0 . . . M 21 -X and lower NFETs M 22 - 0 . . . M 22 -X separated by intermediate nodes A 0 . . . AX, respectively. Lower transistors M 22 - 0 . . . M 22 -X may be considered boost transistors because they help boost the voltages of the intermediate nodes. In some embodiments, transistors M 21 - 0 . . . M 21 -X and M 22 - 0 . . . M 22 -X may be what is considered low threshold voltage transistors at zero bias, although that is not required. Transistors M 21 - 0 and M 22 - 0 share a data input Vin 0 , transistors M 21 - 1 and M 22 - 1 (not shown) share a data input Vin 1 (not shown), and so forth such that transistors M 21 -X and M 22 -X share a data input signal VinX. Merely as an example, X may be 7 or a greater or lesser number. Although circuit 100 is illustrated as an OR gate, those skilled in the art having the benefit of this disclosure could easily implement it in other logic forms such as AND, NOR, and NAND.

During a precharge phase, Clk and data inputs Vin 0 . . . VinX are low so that transistors M 21 - 0 . . . M 21 -X and M 22 - 0 . . . M 22 -X are off and M 23 is on. Nodes A 0 . . . AX have inherent capacitance which is charged due to leakage from node Q. Accordingly, M 21 - 0 . . . M 21 -X are reverse biased because their sources have a voltage higher than their body (which in the illustrated embodiment is at ground, but could be at another voltage). Reverse biasing reduces leakage. This leakage reduction is referred to herein as active leakage reduction. The amount of leakage is also reduced because of the stacked transistors so that Q remains higher.

During the evaluate phase, if all the inputs Vin 0 . . . VinX remain low, intermediate nodes A 0 . . . AX remain charged to a voltage level slightly higher than ground. Thus, upper transistors M 21 - 0 . . . M 21 -X remain reverse biased. The threshold voltages of M 21 - 0 . . . M 21 -X are increased by the reverse bias so that the input voltage signals Vin 0 . . . VinX must be greater to turn on transistors M 1 - 0 . . . M 1 -X, respectively. On the other hand, if one or more of data input signals Vin . . . VinX goes high, the signal Q is discharged through the corresponding transistors in evaluate network 112 as in the prior art. Unlike the prior art circuit of FIG. 2, there is no static power consumption through evaluate network 112 .

In circuit 100 , because of the capacitance of nodes A 0 . . . AX, the transistors of evaluate network 112 may have lower Vts than would otherwise be acceptable. This offers better delay characteristics than a high Vt implementation. The pull-down penalty caused by the stacked transistors can be mitigated by tuning the sizes of the stacked transistors, thus providing an optimal trade-off with other design specifications. However, in some embodiments, the stack evaluate network may increase fanin capacitance, which may mitigate some of the performance improvement achieved by using all low Vt transistors. However, circuit 100 has good decoupling capability like the original domino implementation.

C. Domino circuits with a diode pull-up at the output of the domino stage

In some embodiments, the invention includes a domino circuit with diode pull-up transistor at the output of the domino stage. For example, FIG. 5 illustrates domino wide-fanin OR circuit 150 having a domino stage 154 and an output stage 162 . Output stage 162 may be various circuits including an inverter as in FIG. 1 or a dual function generator similar to or somewhat different from that of FIG. 3 . Domino stage 154 includes a wide fan-in evaluate network 158 , pull-up PFET M 32 , and a keeper PFET M 33 and associated inverter 118 . Domino stage 154 also includes an diode pull-up PFET transistor M 34 on a node Q, an output of the domino stage. Node Q carries a domino stage output signal. Evaluate network 158 includes NFET transistors M 31 - 0 . . . M 31 -X, the gates of which receive data input signals Vin 0 . . . VinX. Merely as an example, X may be 7 or a greater or lesser number. Although circuit 150 is illustrated as an OR gate, those skilled in the art having the benefit of this disclosure could easily implement it in other logic forms such as AND, NOR, and NAND.

During a precharge phase, Clk and Vin 0 . . . VinX are low so that transistors M 31 - 0 . . . M 31 -X are off and M 32 is on. Accordingly, the domino stage output signal at node Q is pulled high. Diode pull-up transistor M 34 is off because V GS of M 34 is zero.

During the evaluate phase, Clk goes high so that M 32 turns off. As long as node Q is high, keeper PFET M 33 is on pulling up node Q. If, one or more data inputs signals Vin 0 . . . VinX goes high, the corresponding one(s) of M 31 - 0 . . . M 31 -X turns on, pulling Q low which turns off M 33 turns off. As Q does low, diode transistor M 34 turns on. However, diode M 34 does not have enough strength to overcome the NFET(s) of evaluate network 158 .

If each of inputs Vin 0 . . . VinX remain low, node Q remains high and M 34 remains off.

However, if there is noise of one or more of inputs Vin 0 . . . VinX so that the corresponding one(s) of M 31 - 0 . . . M 31 -X barely turns on, node Q may start to go low and M 34 will start to turn on pulling up Q. Further, M 33 will also assist in pulling up Q.

If Vin 0 . . . VinX remain low, but there is leakage through one or more of M 31 - 0 . . . M 31 X, even though V GS of M 34 may be less than VT, transistor M 34 will be in a subthreshold region of operation that replenishes charge at node Q.

The Vt and size of M 34 can be selected to achieve desired noise immunity, speed, and power consumption tradeoffs.

D. Body Biasing

A FET transistor has a zero body bias when its body has the same voltage potential as its source (Vsource). The body voltage of a PFET is referred to as Vbbn, because it has an n-type body and the body voltage of an NFET is referred to as Vbbp, because it has a p-type body. An NFET is forward body biased when Vbbp>Vsource and reversed body biased when Vbbp<Vsource. A PFET is forward body biased when Vbbn<Vsource and reversed body biased when Vbbn>Vsource. When a transistor is forward body biased, its Vt decreases and leakage increases as compared to a zero body bias or reverse body bias. When a transistor is reverse body biased, its Vt increases and leakage decreases as compared to a zero body bias or forward body bias. For the reasons explained above, with increased leakage, circuits tend to have lower noise tolerance. They also tend to have higher performance with a lower Vt. In many cases, performance and noise immunity are traded off.

In some embodiments, the present invention may be used to give higher performance to the circuits described above through applying body biases other than Vss to NFETs and Vcc to PFETs. For example, referring to FIG. 6, a circuit 170 is similar to circuit 50 in FIG. 3, except that the body of the NFETs is tied to bias generation circuitry 172 and the body of the PFETs is tied to bias generation circuitry 174 . If Vbbp is greater than Vss, the NFETs would be forward body biased. The NFETs would have a lower Vt and greater leakage. This would tend to give greater performance, but possibly also lower noise tolerance and noise immunity. However, with a forward body biased condition, they may conduct better so that the voltage of intermediate node D could rise more easily and M 11 - 0 . . . M 11 -X could assume a reverse body biased condition. (Vbbp could be chosen to obtain a desired reverse bias.) With the same noise immunity as the prior art circuits of FIGS. 1 and 2, circuit 170 could have higher performance, or circuit 170 could have better noise immunity and better performance than the prior art circuits.

FIGS. 7 and 8 show circuits 190 and 200 , which are similar to circuits 100 and 150 of FIGS. 4 and 5, except that the body of the NFETs are not necessarily at Vss and the body of the PFETs are not necessarily at Vcc. Bias generation circuitry 172 and bias generation circuitry 174 may be used in connection with circuits 190 and 200 .

Bias generation circuitry 172 and bias generation circuitry 174 may provide voltages to reverse body bias some or all of the transistor of FIGS. 6, 7 , and 8 to increase Vt and decrease leakage and increase noise immunity. Circuitry 172 and 174 may provide constant or changing voltages depending on the mode of the circuits. In some circuits, different NFETs could get different body voltages and different PFETs could get different body voltages. For example, transistors M 11 - 0 . . . M 11 -X might get different body voltages than other NFETs. In some circuits, some NFETs may have their body tied to Vss, while other NFETs have their body at a non-Vss voltage; and some PFETs may have their body tied to Vcc, while other PFETs have their body at a non-Vcc voltage.

In some circuits, where Vcc is relatively low (e.g., 700 millivolts), the body of NFETs may be tied to Vcc, while the body of PFETs may be tied to Vss, to obtain a forward body bias.

It is noted that as described above in connection with circuits 50 and 100 in FIGS. 3 and 4, transistors M 11 - 0 . . . M 11 -X and M 21 - 0 . . . M 21 -X may have an intrinsic reverse body biased even though the body of these transistors is at ground.

(Note that it was known by others before the present invention that changing the body voltage with respect to the source voltage could change the bias of the transistor and change the Vt and leakage of the transistor.)

E. Other Information and Embodiments

Different features of the circuits of FIGS. 3-8 can be mixed and matched. For example, the dual stack evaluate network of circuit 100 could be incorporated into circuit 50 and 150 .

FIG. 9 illustrates a semiconductor die 250 , that includes many circuits, of which circuits 252 , 254 , and 256 are examples. Circuits 252 , 254 , and 256 may be any one of circuits of FIGS. 3-8 or other circuits. Die 250 could be for any of a various types of electrical devices including a microprocessor, DSP (digital signal processor), embedded controller, ASIC (application specific integrated circuit), in connection with register files, chipset, networking, communications including (RF) radio frequency, etc. Referring to FIG. 9, die 250 is part of a system 260 , which may be a personal computer, mainframe computer, computer with a DSP, microcomputer, hand held computer, communication system, or a variety of other systems.

As used herein, Vcc and Vss are examples of power supply and ground voltages. However, the power supply and ground voltages of circuits of FIGS. 3-8 may be, but do not have to be, the same power supply and ground voltages as are used by other circuits in die 250 . Also, as is well known, power supply and ground voltages are not necessarily constant, but rather have fluctuations because of noise or other reasons.

Although in the illustrated examples, node Q is high during precharge and data input signals Vin and the output stage output signal Vout are low during precharge, this is not required.

FETs other than MOSFETs could be used. Although the illustrated embodiments include enhancement mode transistors, depletion mode transistors could be used with modifications to the circuit which would be apparent to those skilled in the art having the benefit of this disclosure.

Circuits according to the present invention could include additional structure not illustrated in the present drawings.

If the specification states a component, feature, structure, or characteristic “may”, “might”, or “could” be included, that particular component, feature, structure, or characteristic is not required to be included. Reference in the specification to “an embodiment,” “one embodiment,” “some embodiments,” or “other embodiments” means that a particular feature, structure, or characteristic described in connection with the embodiments is included in at least some embodiments, but not necessarily all embodiments, of the invention. The various appearances “an embodiment,” “one embodiment,” or “some embodiments” are not necessarily all referring to the same embodiments.

Those skilled in the art having the benefit of this disclosure will appreciate that many other variations from the foregoing description and drawings may be made within the scope of the present invention. Accordingly, it is the following claims including any amendments thereto that define the scope of the invention.