Background of the Invention
1. Field of the Invention
The present invention pertains to low-power semiconductor circuitry. More particularly, the present invention pertains to complementary transistor gates for adiabatic circuits.
2. Discussion of Related Art
Circuits that operate with low power requirements are desirable for use in many electronic devices, particularly portable devices that are powered by batteries, solar cells, or other limited-capacity power supplies. One of the advantages of CMOS circuitry has been its low rate of power dissipation relative to emitter-coupled logic (ECL), bipolar, and other types of semiconductor circuit devices.
However, even as mask technology reduces the size of CMOS structral elements to the submicron range, these elements are increasingly called upon to perform their switching functions at the high end of the frequency spectrum. As a result, some recent CMOS circuit designs have wattages more appropriate for the common lightbulb than for submicron, heat-sensitive VLSI semiconductor packages.
Generally, the energy dissipated in switching can be reduced to provide adiabatic performance if logic switching is performed quasi-statically: the dissipation of energy through a dissipative medium is reduced if the transfer is made gradually enough. Despite their name, so-called "hot clock" circuits are unconventional circuit designs that apply this principle to provide adiabatic performance. These devices reduce the rate of change in the circuit, but "gradually" is a relative concept. Known "hot clock" circuits are capable of operating at frequencies as high as 20 MHz in 0.9 micron CMOS. Under these conditions a 5x to 10x power dissipation reduction can be achieved by hot-clock designs.
Hot-clock logic designs supply all power to the gates via their clock lines, as described by Seitz, et al. at the Chapel Hill VLSI Conference in 1985. Quasi-static logic operations have two cardinal rules, rules familiar to those in the DC to DC converter art: (1) Never enable a switch unless the potential across it is zero; (2) Never disable a switch if there is a current flowing through it.
Hot-clock circuit performance in CMOS has been improved by adding blocking diodes, as disclosed by Avery, et al., in Ser. No. 08/175,709 filed Dec. 30, 1993. However, the resulting circuit may require that clock levels be boosted asymmetrically, or that level shifters be provided between gates or low-threshhold diodes be used to overcome the resulting diminished signal voltage swings at the circuit's output.
Summary of the Invention
A CMOS gate in accordance with the present invention includes a gate transistor pair, an output node between the gate transistors in the pair, and a pair of gate clock signals. Each gate clock signal is connected to the source electrode of a respective CMOS gate transistor.
An ancillary transistor preferably has its channel adapted for connection to the drain of each gate transistor and the output node. A respective one of a pair of complementary ancillary clock signals is connected to the gate electrode, the "control element" of each respective ancillary transistor. Said ancillary clock signals are phase-shifted from the gate clock signals so that said four clock signals have four distinct phases, whereby the gate transistor pair is enabled to establish and maintain full-rail voltage values at the output node. Preferably, these four clock signals are phase-shifted a full 90 degrees from each other.
In particular, the ancillary transistors and offset clock signals of the present invention provide adiabatic performance while producing full-rail voltage swings at gate outputs. In a preferred embodiment this is achieved in homogeneous CMOS device packages. Homogenous packages are simpler to fabricate than the composite packages required for circuits that include functions such as output level shifting or an asymmetric boost on the clock leads. In particular, the Schottky diodes that are preferable for use as blocking diodes are non-MOS devices and their use can be avoided, for many applications, in accordance with the present invention.
In another embodiment the interaction of blocking diodes with offset clock signals in accordance with the present invention provides a further reduction of non-adiabatic device charging in a circuit, for specially demanding applications.
Brief Description of the Drawings
The features and advantages of the present invention will be better understood when the detailed description of a preferred embodiment given below is considered in conjunction with the drawings provided, wherein:
FIG. 1 is a diagram of a prior art logic circuit;
FIG. 2 is a timing diagram for the circuit of FIG. 1;
FIG. 3 is a diagram of a CMOS logic circuit in accordance with a first embodiment of the present invention;
FIG. 4 is a timing diagram for the circuit of FIG. 3;
FIGS. 5a through 5c are diagrams of other types of logic gates in accordance with the present invention;
FIG. 5d is a schematic diagram of a memory cell that incorporates ancillary transistors controlled by multiple clock signals of different phases, according to the principles of the present invention;
FIG. 5e is a graph illustrating the amplitude and relative phase of multiple clock signals, which are preferably applied to the memory cell circuit illustrated in FIG. 5d.
FIG. 6 is a timing diagram for the circuit of FIG. 3, showing a different input signal In; and
Like reference numerals indicate like structures.
Detailed Description of the Preferred Embodiments
FIG. 1 shows a string of CMOS inverters using two offset trapezoidal "hot clock" signals C.sub.1 and C.sub.2, and their complements, to obtain adiabatic operation. In FIG. 2, it can be seen that output waveforms A, B and D do not swing "full rail" through the full voltage range V.sub.SS -V.sub.DD in response to full-rail swings in the input signal In.
In the FIG. 1 circuit, the input signal that is supplied to the control element of each gate transistor changes during a SETUP PHASE (#1) when the clock signal input to that gate has been entirely disabled, as required for adiabatic operation, and the gate's output floats, as indicated by the dashed line. However, the gate's output is, by then, already floating in midrange rather than holding the full logic values V.sub.SS and V.sub.DD.
For example, in FIG. 2, when C.sub.1 goes high during the ENABLE PHASE (#2), MN1 becomes enabled and output A makes the transition to full V.sub.SS and holds that value. However, because the clock signal is trapezoidal, during the DISABLE PHASE (#4) gate transistors MP1 and MN1 being disabled, and the active gate transistor MN1 remains active long enough to permit a current to pass through MN1, causing the output A to follow the clock C.sub.1 to a mid-range voltage. Then, MN1 finally becomes disabled, and the output A floats at midrange during the rest of the DISABLE PHASE (#4) rather than hold its full-rail value of V.sub.SS. The output A continues to float during the subsequent SETUP PHASE (#1), when the gate transistors have completed their transitions and the path between C.sub.1 and its complement C.sub.1 is once again cut off.
The input signal In in this timing diagram changes value from V.sub.DD to V.sub.SS, or V.sub.SS to V.sub.DD during the SETUP PHASE (#1). In the ENABLE PHASE (#2), that full-rail change in the input signal In during the SETUP PHASE (#1) produces a belated, partial change in the output signal A from midrange to a full V.sub.DD or V.sub.SS value in the ENABLE PHASE (#2). This distorted signal appearing at the output of the gate A then propagates through the rest of the inverter gates B, D and Out, in this orderly but unsatisfactory fashion.
FIG. 3 shows another serial string of inverters that provides full-rail signal transitions and improved adiabatic performance in accordance with a first embodiment of the present invention. Again, each pair of gate clock signals, such as C.sub.1 and C.sub.1, is connected to the source electrodes of a respective pair of CMOS gate transistors MP1 and MN1, and the drain electrode of each gate transistor MP1 and MN1 is connected to the gate's output node through ancillary transistors MP2 and MN2.
However, in FIG. 3, an ancillary transistor MP2 or MN2 is connected between each respective gate clock and the output node. Specifically, the source electrode of each ancillary transistor, MP2 and MN2 is connected to the drain electrode of the respective gate transistor MP1 or MN1 and the drain electrode is connected to the output node A. Also, a respective one of a pair of complementary clock signals C.sub.2 and C.sub.2 that are offset from the gate clock signals of that gate, C.sub.1 and C.sub.1, is connected to the gate electrode, the "control element" of each respective ancillary transistor MP2 or MN2.
In FIG. 3 the gate clock pair in each stage, for example C.sub.1 or C.sub.2 and its complement, is also supplied to a previous stage, a stage that is driven by another, offset clock, C.sub.2 or C.sub.1 respectively. This assures that the ancillary transistors MP2 or MN2 will be cut off during the gate's DISABLE PHASE (#4), so that the path between the clock signals and the output A is disrupted. In contrast to FIG. 1, each ancillary transistor MP2, MN2 is fully cut off while the gate transistor MP1, MN1 connected in series with it is in transition in FIG. 4, and the output A remains at its full-rail value. This action of the gate's ancillary transistors produces a full-rail voltage swing at output A during the subsequent ENABLE PHASE (#2), as in FIG. 4.
FIG. 5a shows a CMOS NAND gate circuit in accordance with the present invention. FIG. 5b shows a CMOS NOR gate circuit in accordance with the present invention. In these logic gates the gate transistors are controlled by the input signals IN1 and IN2. Offset clock C.sub.2, and its complement, control the auxiliary transistors MN3 and MP3, respectively. These ancillary transistor pairs prevent the outputs A from following the trapezoidal clocks by disrupting the path to the gates' outputs during the transistions of clock C.sub.1 and its inverse, as in FIGS. 3 and 4.
Of course, inverters and other gates in accordance with the present invention can be connected in series as shown in FIG. 3, or in parallel as shown in FIG. 5c, or in any combination thereof. Also, rather than being a logic gate, the gate may be a memory cell, as shown in FIG. 5d. In each instance, the ancillary transistors between the respective gate clocks and the gate output prevent the trapezoidal clocks from wasting power and distorting the output signal, as explained above.
It will be appreciated by one skilled in the art that variations and modifications of the disclosed apparatus are possible within the spirit and scope of this invention. For example, the clocks C1, C2 need not be fully 90 degrees offset to enjoy some of the benefits of the present invention. The embodiments described are provided to illustrate presently preferred ways of making and using this invention. The invention is defined by the claims appended below.