5-Transistor Memory Cell Which Can Be Reliably Read and Written

This invention relates to a static memory cell and in particular to a five-transistor memory cell which can be reliably read and written.

FIG. 4 shows a prior art 6 transistor CMOS memory cell similar to the Intel 5101 cell. Transistors T'.sub.1, T'.sub.2, T'.sub.3 and T'.sub.4 constitute a cross-coupled latch that typically draws a steady state current of approximately one nanoampere. Transistors T'.sub.5 and T'.sub.6 are gating devices (pass transistors) that couple the bit lines (data lines) to the latch when the voltage on the row select line (address line) is high (5 volts). The output signal Q is a logical 1 when N channel enhancement mode transistor T'.sub.3 is off and P channel enhancement mode transistor T'.sub.4 is on, and it is a logical zero when these states are reversed. Reading and writing are accomplished through the left and right bit lines. For example to read the data out of the memory cell in FIG. 4, a high signal is applied to the row select, turning on transistors T'.sub.5 and T'.sub.6. If a logical 0 (0 volts) is on node A and a logical 1 (5 volts) is on node B, the left bit line is charged to a lower level than the right bit line. These two bit lines are typically connected to a differential amplifier (not shown) that amplifies the difference in voltage levels on the bit lines. The amplified difference is then interpreted as a logical 0 or a logical 1, according to some design convention.

To write a bit into the memory cell, the row select line is brought high (to 5 volts) and the left and right bit lines are charged to opposite states by the write driver (not shown in FIG. 4), which drives node A to the same logic level as the left bit line and node B to the same logic level as the right bit line.

The six transistor memory cell requires two gating devices (pass transistors) and two bit lines to be reliably read and written. Note that the six transistor memory cell can also be implemented in NMOS. See Holt, Electronic Circuits, John Wiley and Sons, Inc., pp. 293-294 (1978) which are incorporated herein by reference.

In contrast to the prior art, the present invention describes a five transistor memory cell which can be reliably read and written from a single data line. The memory cell includes a first and a second inverter with the output of the first inverter connected to the input of the second inverter and the output of the second inverter connected to the input of the first inverter and only a single gating (pass) transistor which is connected between the input lead of the first inverter and the single bit line.

In one embodiment of the invention the memory cell also includes a first and a second output node (lead) which continuously provides the state of the memory cell to circuitry external to the memory, for example to control the gates of external pass transistors or to provide an input signal to a logic gate.

Typically a plurality of five transistor memory cells are connected to the same data line. As one feature of the invention, means are provided for increasing the rise time on the gate of the pass transistor in order to reduce the possibility of disturbing the content of the memory cell during the read operation. As another feature of the invention, the trigger point of the first inverter is selected to be more than one threshold voltage with body effect below the voltage applied to the gate of the pass transistor and channel dimensions of the pass transistor relative to channel dimensions of the N channel and of the P channel transistors in the second inverter are selected to insure that the memory cell can be successfully written.

In another embodiment, circuitry is provided for precharging the data line to a first selected voltage level prior to reading a stored bit in order to reduce read disturbance.

In another embodiment circuitry is provided for charging the gate of the pass transistor to a first level during the read operation and to a second level during the write operation and for precharging the data line to a third selected voltage level prior to the read operation. The voltage levels are selected to minimize read disturbance.

The invention will be more readily understood by reference to the drawings and the detailed description.

FIG. 1 shows a five transistor memory cell according to the present invention.

FIG. 2 shows a circuit for precharging the data line for the memory cell shown in FIG. 1.

FIG. 3 shows an address supply voltage source and an address driver for supplying a first selected voltage level to the address line for the memory cell of FIG. 1 during the read operation and a second selected voltage level during the write operation; FIG. 3 also shows a precharge circuit for precharging the data line to a third selected voltage level prior to the read operation.

FIG. 4 shows a prior art six transistor memory cell.

FIG. 1 shows one embodiment of memory cell 10 of the present invention. Memory cell 10 includes N channel enhancement mode pass transistor N.sub.3 and inverters INV.sub.1 and INV.sub.2.

Inverter INV.sub.1 includes P channel enhancement mode transistor P.sub.1 and N channel enhancement mode transistor N.sub.1. Source 1 of transistor P.sub.1 is connected to the positive voltage supply having magnitude V.sub.CC, which is typically 5 volts plus or minus 10% although other voltages may be employed for V.sub.CC. Drain 2 of transistor P.sub.1 is connected to drain 4 of transistor N.sub.1 whose source is connected to ground. Gates 3 and 6 of transistors P.sub.1 and N.sub.1, respectively are connected to sense node A. The common drains 2 and 4 are connected to output node B.

Inverter INV.sub.2 includes P channel enhancement mode transistor P.sub.2 and N channel enhancement mode transistor N.sub.2. Source 7 of transistor P.sub.2 is connected to the positive voltage supply having magnitude V.sub.CC. Drain 8 of transistor P.sub.2 is connected to drain 10 of transistor N.sub.2 whose source is connected to ground. Gates 9 and 12 of transistors P.sub.2 and N.sub.2, respectively, are connected to output node B. Drains 8 and 10 are connected to sense node A. In operation the output signal on node B is continuously available to other circuits (not shown in FIG. 1), for example, as a gate control signal for other transistors.

Data line D.sub.M is connected to sense node A via pass transistor N.sub.3. Gate 15 of transistor N.sub.3 is controlled by the voltage signal on address line A.sub.N. Source/drain 13 of transistor N.sub.3 is connected to data line D.sub.M and source/drain 14 of transistor N.sub.3 is connected to sense node A.

One advantage of memory cell 10 is that while the signal on output node B is used, typically continuously, to control other circuits (not shown in FIG. 1), the content of memory cell 10 (i.e. the signal stored on node A) can be repeatedly checked by a read operation to verify the integrity of the memory cell's data content without degrading the output signal on node B. Furthermore, if desired, for example if the complement of the signal on node B is required to control other circuits, sense node A can also be used as an output node. This is indicated by the dotted arrow in FIG. 1. The voltage level at sense node A may be somewhat degraded during the read operation.

When sense node A stores a logical 0 and it is desired to write a logical 1 to cell 10, the signal provided at source/drain 14 of transistor N.sub.3 must be sufficient to increase the voltage on sense node A above the trigger point of inverter INV.sub.1 despite the pulldown effect of current flowing through transistor N.sub.2 of inverter INV.sub.2 (the trigger point of an inverter is the voltage at which the gate (input) voltage of the inverter equals the output voltage of the inverter). Conversely, when sense node A stores a logical 1 and it is desired to write a logical 0 to sense node A, the signal provided at source/drain 14 of transistor N.sub.3 must be sufficient to decrease the voltage on sense node A below the trigger point of inverter INV.sub.1 despite the pullup effect of transistor P.sub.2 of inverter INV.sub.2.

The operation of writing data into cell 10 and the selection of the parameters for transistors N.sub.1, N.sub.2, N.sub.3, P.sub.1, and P.sub.2 may be understood by considering the following examples.

Suppose that memory cell 10 stores a logical 0, i.e. the voltage on node A is 0 volts (logical 0) and that the output signal of inverter INV.sub.1 is V.sub.CC (logical 1). Suppose that a logical 1 having a voltage level of V.sub.CC on data line D.sub.M is to be stored on node A and that transistor N.sub.3 is turned on by applying the signal V.sub.CC to gate 15. A voltage level of V.sub.CC on source/drain 13 and a voltage level of V.sub.CC on gate 15 results in a voltage on source/drain 14 not higher than V.sub.CC -V.sub.TH (N.sub.3). V.sub.TH (N.sub.3) is the threshold voltage of transistor N.sub.3 with body effect. Thus the trigger point of inverter INV.sub.1, denoted by V.sub.TRIG (INV.sub.1), is selected to be less than V.sub.CC -V.sub.TH (N.sub.3). This is accomplished by selecting the ratio of the ratio of the channel width to channel length of pullup transistor P.sub.1 to the ratio of channel width to channel length of pulldown transistor N.sub.1 of inverter INV.sub.1 to be sufficiently small. For example, if V.sub.CC equals 5 volts and the channel width and channel length of transistors N.sub.1 and P.sub.1 are as given in the following table:

then the trigger point of inverter INV.sub.1 will be less than 2 volts. Having selected the channel width and channel length of transistors P.sub.1 and N.sub.1 so that the trigger point of inverter INV.sub.1 is less than V.sub.CC -V.sub.TH (N.sub.3) the channel dimensions of N.sub.2 relative to the channel dimensions of N.sub.3 are selected so that the voltage at node A rises above the trigger point TP of inverter INV.sub.1. When transistors N.sub.3 and N.sub.2 are both on, they act as a voltage divider, and the voltage at sense node A is given by V.sub.CC (R(N.sub.2)/(R(N.sub.2)+R(N.sub.3))) where R(N.sub.2) is the channel resistance provided by transistor N.sub.2 and R(N.sub.3) is the channel resistance provided by transistor N.sub.3. R(N.sub.2) is directly proportional to L(N.sub.2)/W(N.sub.2) and R(N.sub.3) is directly proportional to L(N.sub.3)/W(N.sub.3) where L(N.sub.2) is the channel length of transistor N.sub.2, W(N.sub.2) is the channel width of transistor N.sub.2, L(N.sub.3) is the channel length of transistor N.sub.3 and W(N.sub.3) is the chanel width of transistor N.sub.3. By appropriately choosing the channel length and channel widths, we may ensure that V.sub.CC (R(N.sub.2)/(R(N.sub.2)+R(N.sub.3))) is greater than the trigger point TP of inverter INV.sub.1. In one embodiment, the channel length of pass transistor N.sub.3 is 2.5 microns and the channel width is 7.5 microns. Transistor N.sub.2 has a channel length of 4 microns and a channel width of 4 microns. In this case R(N.sub.2)/(R(N.sub.2)+R(N.sub.3)) equals 0.6. Hence the voltage on node A will rise above the trigger point of inverter INV.sub.1. Once the voltage on sense node A rises above the trigger point, V.sub.TRIG(INV1), the output signal on node B goes low and the output signal of inverter INV.sub.2 goes high driving sense node A to the V.sub.CC level.

In writing a logical 0 to memory cell 10, assume the voltage on data line D.sub.M is 0 volts, address line A.sub.N is charged to V.sub.CC, and a voltage signal V.sub.CC (logical 1) is stored on node A. When both transistors P.sub.2 and N.sub.3 are on, pullup transistor P.sub.2 in inverter INV.sub.2 and transistor N.sub.3 act as a voltage divider and the voltage at sense node A is given by V.sub.CC (R(N.sub.3)/(R(N.sub.3)+R(P.sub.2))) where R(P.sub.2) is the channel resistance provided by transistor P.sub.2 and R(N.sub.3) is the channel resistance provided by transistor N.sub.3. The channel resistance of P.sub.2 is directly proportional to L(P.sub.2)/W(P.sub.2) where L(P.sub.2) is the channel length of transistor P.sub.2 and W(P.sub.2) is the channel width of transistor P.sub.2. The channel resistance of N channel transistor N.sub.3 is directly proportional to L(N.sub.3)/W(N.sub.3) where L(N.sub.3) is the channel length of transistor N.sub.3 and W(N.sub.3) is the channel width of transistor N.sub.3. These channel lengths and widths are chosen so that the voltage on sense node A falls below the trigger point of inverter INV.sub.1. In one embodiment, transistor P.sub.2 in inverter INV.sub.2 has a channel length of 4 microns and a channel width of 6 microns. The fraction R(N.sub.3)/(R(N.sub.3)+R(P.sub.2)) equals 0.1. In this case, the voltage on sense node a will fall below the trigger point of inverter INV.sub.1 where the channel dimensions of inverter INV.sub.1 are specified in the above table. Once the voltage on sense node A falls below the trigger point, the output signal on node B goes high and the output signal of inverter INV.sub.2 goes low driving sense node A to 0 volts. The above analysis assumes that the channel resistance of the pullup and pulldown transistors of the write driver (not shown) are significantly smaller (less than 10%) than the channel resistance of transistors P.sub.2, N.sub.2 and N.sub.3.

It is also desirable to be able to read the data signal stored on sense node A by transmitting this signal via pass transistor N.sub.3 to data line D.sub.M disturbing the content of the memory. The value read is the value that appears on source/drain 13 of transistor N.sub.3. Typically data line D.sub.M, which may be connected to many cells similar to cell 10 of FIG. 1, has a large capacitance compared to the capacitance of sense node A. When address line A.sub.N goes high to turn on pass transistor N.sub.3 in order to read the value stored on node A, the content of the memory (the value stored on node A) may be disturbed due to charge sharing. The following techniques can be employed to reduce the danger of disturbing the memory cell during the read operation. First, one may increase the rise time of the address line A.sub.N by slowing the rate of increase of the voltage of address line A.sub.N. Then transistor N.sub.3 turns on more slowly, allowing memory cell 10 to react to the disturbance caused by charge sharing without changing the content of the data stored on node A. For example, if V.sub.CC is stored on node A, the rise time must be sufficiently long that the voltage on node A does not fall to V.sub.TRIG (INV.sub.1) when transistor N.sub.3 turns on. If 0 volts is stored on node A, the rise time of the signal on address A.sub.N must be sufficiently long that the voltage on node A does not rise to V.sub.TRIG (INV.sub.1) when transistor N.sub.3 turns on. A typical address rise time should be 200 ns or more. The rise time of address line A.sub.N is increased by using a "weak" (small channel width to channel length ratio) pullup transistor (not shown) in the address driver.

A second technique for avoiding disturbing the content of cell 10 during the read operation is to precharge the data line D.sub.M to the value V.sub.TRIG (INV.sub.1).

Assume data line D.sub.M is precharged to the value V.sub.TRIG (INV.sub.1). Assume also that a read signal of magnitude V.sub.CC is applied to address line A.sub.N. If V.sub.CC (logical 1) is stored on sense node A, then pullup transistor P.sub.2 of inverter INV.sub.2 and pass transistor N.sub.3 form a voltage divider network and sense node A does not fall below V.sub.TRIG (INV.sub.1). Similarly, if 0 volts (logical 0) is stored on sense node A, then sense node A does not rise above V.sub.TRIG (INV.sub.1), since in this event transistors N.sub.2 and N.sub.3 form a resistor divider network and data line D.sub.M is precharged to V.sub.TRIG (INV.sub.1). In one embodiment, the circuit shown in FIG. 2 is used to precharge data line D.sub.M to V.sub.TRIG (INV.sub.1).

The V.sub.TRIG (INV.sub.1) precharge circuit shown in FIG. 2 includes P channel enhancement mode transistor T.sub.1, N channel enhancement mode transistor T.sub.2, and N channel enhancement mode pass transistor T.sub.3. As shown in FIG. 2, source 20 of transistor T.sub.1 is connected to the positive voltage supply V.sub.CC. Drain 21 of transistor T.sub.1 is connected to drain 23 of transistor T.sub.2 whose source 24 is connected to ground. Gates 22 and 25 of transistor T.sub.1 and T.sub.2, respectively, are connected to the common drain connection of transistors T.sub.1 and T.sub.2 which also connects to drain 26 of pass transistor T.sub.3. Source 27 of transistor T.sub.3 is connected to data line D.sub.M and gate 28 of transistor T.sub.3 is connected to precharge signal, .phi..sub.precharge. The "inverter" comprising transistors T.sub.1 and T.sub.2 is designed to have the same trigger point as inverter INV.sub.1, shown in FIG. 1. In the precharge cycle, the precharge signal, .phi..sub.precharge is set to V.sub.CC, which turns on N channel pass transistor T.sub.3, and data line D.sub.M is precharged to a voltage level of V.sub.TRIG (INV.sub.1) (assuming V.sub.TRIG (INV.sub.1) is lower than the voltage level of .phi..sub.precharge minus V.sub.TH,T3). The precharge signal .phi..sub.precharge is then brought low by control circuitry (not shown), turning off pass transistor T.sub.3 just before the address line A.sub.N connected to gate 15 of pass transistor N.sub.3 is brought high.

The third technique for avoiding disturbing the content of memory cell 10 during the read operation is to precharge data line D.sub.M to V.sub.CC and set the high level of address line A.sub.N to the value V.sub.TRIG (INV.sub.1). These conditions are implemented using the circuitry shown in FIG. 3. Under these conditions, when the value stored in cell 10 is V.sub.CC (logical 1), pass transistor N.sub.3 remains off and the value sensed at source/drain 13 is V.sub.CC (logical 1), and sense node A is undisturbed. On the other hand, when the value stored in cell 10 is 0 volts (logical 0), the highest voltage that sense node A can be charged to is V.sub.TRIG (INV.sub.1)-V.sub.TH (N.sub.3) since N.sub.3 is cut off when the voltage on source/drain 14 reaches this value. Hence the read "0" operation has a noise margin of V.sub.TH (N.sub.3). This is the preferred technique because the memory cell 10 is guaranteed not to be disturbed by the read operation regardless of the rise time of address line A.sub.N, the imbalance between the capacitance on the data line D.sub.M and the capacitance on the sense node A, or the ratio of channel resistance between transistor N.sub.3 and transistor P.sub.2 or N.sub.2. This third technique requires that the address line A.sub.N be charged to V.sub.CC for a write operation and to V.sub.TRIG (INV.sub.1) during a read operation. Symbolically, ##EQU1## The address supply voltage source can be implemented as shown in FIG. 3. Address supply circuit 90 shown in FIG. 3 includes P channel enhancement mode transistor TA.sub.1, N channel enhancement mode transistor TA.sub.2, N channel enhancement mode transistor TA.sub.4, and P channel enhancement mode transistor TA.sub.3. As shown in FIG. 3, source 30 of transistor TA.sub.1 is connected to the positive voltage supply V.sub.CC. Drain 31 of transistor TA.sub.1 is connected to drain 33 of transistor TA.sub.2 whose source 34 is connected to drain 36 of transistor TA.sub.4 whose source 37 is connected to ground. Source 39 of P channel transistor TA.sub.3 is connected to V.sub.CC and drain 40 of transistor TA.sub.3 is connected to gates 32 and 35 of transistors TA.sub.1 and TA.sub.2 and to the common drain connection of transistors TA.sub.1 and TA.sub.2. Gates 41 and 38 of transistors TA.sub.3 and TA.sub.4 are controlled by the signal on line R/W. In the write mode, a signal of 0 volts is applied to line R/W, which turns off N channel transistor TA.sub.4. P channel transistor TA.sub.3 then charges V.sub.ADDRESS SUPPLY to V.sub. CC. Note that transistor TA.sub.3 should be designed to be sufficiently large to provide the current to address driver 70 to charge up address line A.sub.N in the write mode. In the read mode, V.sub.CC (logical 1) is applied to line R/W. This turns P channel transistor TA.sub.3 off and turns on N channel transistor TA.sub.4. By appropriately choosing channel lengths and channel widths, the circuit comprising transistor TA.sub.1, transistor TA.sub.2 and transistor TA.sub.4 is designed such that the voltage at the output mode 45, V.sub.ADDRESS SUPPLY is the same as the trigger point of inverter INV.sub.1 shown in FIG. 1. Thus, V.sub.ADDRESS SUPPLY is equal to V.sub.TRIG (INV.sub.1). Note that transistor TA.sub.1 should be designed to be sufficiently large to provide the current to address driver 70 to charge up address line A.sub.N in the read mode. In one embodiment, transistors TA.sub.1 and TA.sub.3 have a channel length of 2.5 microns and a channel width of 30 microns and transistors TA.sub.2 and TA.sub.4 have a channel length of 2.5 microns and a channel width of 108 microns.

Address driver 70 is logically a NOR gate having input lead 54 for receiving the signal address clock and input lead 55 for receiving the signal address select.

Lead 54 provides the signal address clock to gate 58 of P channel enhancement mode transistor 52 and to inverter 56 whose output signal controls gate 61 of N channel enhancement mode transistor 50. Lead 54 is also connected to gate 65 of N channel enhancement mode transistor 66.

Lead 55 provides the address select signal to gate 59 of P channel enhancement mode transistor 53 and to inverter 57 whose output signal controls gate 62 of N channel enhancement mode transistor 51. Lead 55 is also connected to gate 63 of N channel enhancement mode transistor 64.

Transistors 50, 51, 52, and 53 comprise four transmission gates forming two parallel pairs of gates with the gates of each pair connected in series. Note that when signals address clock and address select are both low (0 volts) all four transistors 50, 51, 52 and 53 are on and N channel transistors 64 and 66 are off and thus the voltage V.sub.ADDRESS SUPPLY is transmitted to address line A.sub.N.

P channel enhancement mode transistor 80 is connected between the voltage supply V.sub.CC and data line D.sub.M. Data line D.sub.M is precharged to V.sub.CC by applying a low (0 volts) .phi..sub.precharge signal to gate 81 on lead 82.

The third technique described above in conjunction with FIG. 3 can be modified by replacing P channel transistor 80 by an N channel enhancement mode transistor (not shown) whose gate is controlled by the signal .phi..sub.precharge, the complement of .phi..sub.precharge. In this embodiment, the data line is precharged to V.sub.CC -V.sub.T where V.sub.T is the threshold voltage of the N channel transistor.

Typically a plurality of memory cells identical to cell 10 are connected to data line D.sub.M. FIG. 3 shows two such memory cells having address lines A.sub.N connected to address driver 70 and A.sub.N+1 which is connected to a corresponding address driver (not shown). In another embodiment (not shown), a rectangular memory array is formed which comprises a plurality of data lines, a plurality of address lines, and a plurality of memory cells, the memory cells attached to a given one of said data lines forming a column in the array and the memory cells attached to a given one of the address lines forming a row in the rectangular array.

The above embodiments are intended to be exemplary and not limiting. For example, while the circuits described above are implemented using CMOS technology, they may also be implemented using NMOS technology.

Many other modifications will be obvious to one of average skill in the art without departing from the scope of the invention.