Memory Core Organization

Technical Field

The invention relates to semiconductor memory arrangements and methods, and more particularly to semi¬ conductor sense amplifier circuit arrangements and methods.

Background Art

Sense amplifiers are circuits used in reading data from storage transistors, i.e. core memory, in semiconductor memory arrays. One example of a known sense amplifier arrangement is shown in U.S. Pat. No. 5,198,997, to H. Arakawa, entitled "Ultraviolet Erasable Nonvolatile Memory". The arrangement shown in this patent is of particular interest in that it is directed toward a sense amplifier arrangement which includes a reference cell which is connected to the core memory sensing circuitry in another column through a current mirror arrangement.

Fig. 3 shows another sense amplifier arrange- ment according to the prior art, which differs from Ara¬ kawa. In particular, Fig. 3 shows a sense amplifier cir¬ cuit 300 effective for detecting the conductive state of selected memory cells in the core memory of a selected semiconductor memory chip. The particular sense amplifi- er circuit 300 shown is a generally known approach ap¬ plicable in particular to the field of variable threshold memory devices. The field of such devices includes the subfields of EPROMs, i.e., electrically programmable and ultra-violet (UV) erasable read only memories and EE- PROMs, i.e., electrically programmable and electrically erasable read-only memories.

For purposes herein, and by convention, the conductive state of a memory cell containing a variable threshold device is defined as "low", if the cell is conductive or "on", and as "high", if the cell is non- conductive or "off".

The information regarding the conductive state of a particular core cell 310 in chip memory is transmit¬ ted by currents passing along a bit line 320 which is connected through several additional devices to a sense node 340 connected to the input side of one or more inverters 350, which amplify the voltages produced at sense node 340, in order to suitably drive an output device, not shown.

As is well known, core cell 310 may comprise one or more transistors; for example, core cell 310 may comprise a select transistor 311 connected at its drain to a bit line 320 and at its source to the drain of a variable threshold transistor 312 and driven at its gate by a word line 313, and a variable threshold transistor 312 driven at its gate by a sense line 314. A particular core cell is selected in read mode by applying appropri¬ ate bias voltages to word line 313 and sense line 314 and by turning on one or more core select transistors, such as core select transistor 321. Such a combination of bias conditions will effectively connect core cell 310 to a data bus 330, which acts as the core side input to sense amplifier 300. In addition, sense amplifier 300 has a reference side input line 430 which is additionally connected to a reference cell 410 similar to core cell 310. Reference cell 410 is driven at the gate of refer¬ ence cell select transistor 411 by a reference cell select line 413, and at the gate of a reference cell variable threshold transistor 412 by a reference sense line 414. The core side input of the sense amplifier, i.e., data bus 330, acts as inverting input to differen¬ tial amplifier 333 and is also connected to the source of first pass transistor 331. The drain of first pass

transistor 331 is connected to sense node 340, which in turn is connected to the source of second pass transistor 342, to the drain of current mirror transistor 341, to the input of amplifying inverter 350 and to the drain and gate of a clamping transistor 3^0.

The reference side input of the sense amplifi¬ er, i.e., reference data node 430, is connected to the source of third pass transistor 431. The drain of third pass transistor 431 is connected to reference node 440, which also connects to the gate and drain of a current mirror transistor 441, as well as to the gate of a cur¬ rent mirror transistor 341.

In addition, the non-inverting input of differ¬ ential amplifier 333 is connected to a bias voltage source VB at node 332, the gate of first pass transistor 331 is connected to the output 334 of differential ampli¬ fier 333 and the gate of second pass transistor 342 is connected to the positive side 335 of bias voltage source VD, which in turn is connected at its negative side to output 334 of differential amplifier 333. In addition, clamping transistor 360 is connected at its source 370 to clamping voltage source VC.

In a practical implementation of this sense amplifier arrangement, the differential amplifier 333 and the various voltage supplies, VB, VD, and VC, would actually be implemented with a combination of transis¬ tors; they are shown here as ideal components for the purpose of explaining the operation of the sense amplifi¬ er. Differential amplifier 333, in conjunction with bias voltage source VB, forces the voltage potential on data bus 330 to a controlled level effective for biasing the bit line connected terminal of the core cell. This bias approximately equals VB, with a typical value of ca. 2V. The function of the first and second pass transis¬ tors 331 and 342 is to provide a current path for main¬ taining a voltage potential on data bus 330 equal to VB. If data bus 330 were discharged in the process of inquir-

ing as to the status of a particular core cell, then the combination of first and second pass transistors 331 and 342 would provide a current path for restoring the volt¬ age potential on data bus 330 approximately to VB. Voltage source VD provides an adequate drain-to-gate bias to keep first pass transistor 331 saturated under steady state bias conditions, i.e., when data bus 330 is charged up to a level essentially equal to VB. Thus, first pass transistor 331 will be maintained in an operational region in which the current is quadratically dependent on the gate-to-source bias, a condition favorable for opera¬ tional faster response times by the circuit.

The function of third pass transistor 431 is to limit the voltage potential on reference data node 430 to a value essentially equal to the voltage potential on the data bus 330. This arrangement provides essentially similar drain bias conditions to both the selected core cell 310 as well as to reference cell 410. When applying appropriate bias voltages to reference cell select line 413 and to reference sense line 414, a current IR P flows through the reference cell. This reference current is mirrored via the two current mirror connected transistors 441 and 341 and is reflected at sense node 340 as the current I _SENSE . The various device sizes and bias levels are typically chosen in such a manner as to obtain a current I _SENSE essentially equal to about one half of I _Q ORE' where corresponds to the current of a "low" state core cell; thus, when reading the contents of a conduc- tive cell, the sum of currents at sense node 340 would be such as to keep first and second pass transistors 331 and 342 in a conductive state and the voltage potential relatively low, while when reading a nonconductive cell, the sum of currents at sense node 340 will be such as to charge up the sense node to a higher voltage potential than that corresponding to the previously described conductive core cell case. The voltage potential at sense node 340 is limited at the lower end to a value

close to VB+VD, as explained earlier, and the upper value is limited by the combination of clamping transistor 360 and clamping voltage source VC connected to the source of clamping transistor 360 at node 370. Reduced voltage swing at sense node 340 is desirable since it translates into reduced charge transfer and thus faster response time. Additionally, the trip point of inverter 350 should be above the lowest sense node potential, i.e., VB+VD, and below the highest sense node potential, as determined by clamping transistor 360 and clamping volt¬ age source VC.

The arrangement described above senses the conductive state of the core cells. However, its per¬ formance is limited by several factors. By employing an individual, typically UV erased, reference cell associat¬ ed with a sense amplifier in conjunction with a multitude of core cells programmed under varying voltage and tem¬ perature conditions, it is difficult to obtain optimum matching between the current of the reference cell, IR _E , and the current of the core cell, IcORE' ^{over a w} i^ ^e temperature and voltage range.

Further, when switching from reading a "high" state to reading a "low" state, data bus 330 starts out from being fully charged and has to be discharged just enough so that first pass transistor 331 will conduct a current somewhat larger than IgENSE' ^tnus starting a downward movement of sense node 340. The voltage change required to turn on first pass transistor 331 is brought about by the fact that the conductive core cell must discharge the initially fully charged data bus 330. This is typically the limiting factor in response time, since the relatively small cell current "fights" the relatively large capacitive load of the data bus.

Additionally, while the clamping transistor 360 is meant to limit the voltage swing at sense node 340, it also adds capacitive loading which counteracts the bene¬ fit of reduced voltage swing.

Further ore, since inverter 350 operates in a high gain region, it will present a relatively large capacitive load to sense node 340, due to inherent gate capacitance, as well as Miller reflected capacitance. An object of the invention herein is according¬ ly to increase the speed of reading core memory by im¬ proving on or eliminating the limiting factors indicated above.

Summary of the Invention

The above object has been achieved by replacing the individual reference cell with a "column" of refer¬ ence cells, which are programmed simultaneously with the core cells sharing the same "word line". In the sense amplifier, the first pass transistor separating the data bus from the sense node is maintained in a conductive state, irrespective of core cell current. The sense node clamping diode is eliminated and replaced with an "ac¬ tive" clamping arrangement, and the sense node is buff- ered from the first inverting stage through a low input capacitance and relatively high output drive buffer stage.

By employing a "column" of reference cells which share the same word lines and, as the case may be, sense lines with the core cells, one can achieve very good matching of reference cell current and core cell current, because the core and reference cell sharing the same word line will be programmed at the same time under similar temperature and voltage conditions. Furthermore, by physically placing the reference cells dispersed within the core, one achieves very good dimensional matching of reference to core cells. Thus, the current of a conductive memory cell will be essentially equal to the current of the reference cell. The pass transistor separating the sense node from the data bus is "lubricated" by maintaining it in a conductive state irrespective of the conductive state of the core cell being sensed. Thus, the current flowing

through the pass transistor will change from a finite value, I _L TJB, to a value equal to I _L UB ^{+ I} _CO RE' ^{instead of} changing from zero (non-conductive cell) to I-coRE ( ^con ~ ductive core cell) . The gate-to-source voltage change required to bring about this current change is smaller than in the case of the "non-lubricated" pass transistor due to the quadratic relationship between transistor current and gate to source voltage. Thus, the voltage swing required on the data bus to bring about a change from sensing a non-conductive cell to sensing a conduc¬ tive cell is reduced, and the response time is improved. In order to maintain the desired relationship between core cell current and reference cell current, the "lubri¬ cation" current is mirrored to cancel out at the sense node. Speed of the circuit operation is improved by reducing sense node capacitance through the buffer cir¬ cuitry employed.

Brief Description of the Drawings Fig. 1 is a memory plan view illustrating the improved core arrangement in accordance with the inven¬ tion herein, which includes interspersed reference col¬ umns distributed within the core;

Fig. 2 shows a circuit arrangement according to the invention which is effective for improving sense amplifier performance by "lubricating" the pass transis¬ tor connecting the data bus to the sense node and by buffering the sense node from the first inverter through the use of a low input capacitance/high output drive stage; and

Fig. 3 shows a sense amplifier arrangement according to the prior art.

Best Mode for Carrying Out the Invention Figure 1 shows a core arrangement according to the invention herein. In particular, there is shown memory core 10 including, according to one embodiment, 2048 active (and additionally 32 redundant) memory core

columns each associated with a predetermined number of core cells including a select transistor and a floating gate transistor, as noted above. The memory core 10 is organized according to the invention into a plurality of core memory column sets, 10a through lOn, and reference columns 15a through I5n. According to one version, each memory column set, 10a...lOn, includes 128 active (and additionally two redundant) memory core columns which are selectively connectable to one of data nodes 18 which in turn serve as an input point to one of sense amplifier circuits 24. In total, there are "n" data nodes, 18a...18n. Additionally, each of sense amplifier cir¬ cuits 24 has connected at its input an associated refer¬ ence column node 25a...25n connected to a respective one of reference columns 15a...15n. Accordingly, as to sense amplifier circuit 24a, there is provided a corresponding pair of input nodes, respectively 18a and 25a, which are the data node and the reference column node inputs in that order. It is accordingly apparent that memory core 10 includes a plurality of column sets 10a...lOn each associated with a corresponding reference column 15a...15n having a reference column node 25a...25n. The reference columns are interspersed between the column sets 10a...lOn and are distributed within memory core 10. A result of this interspersed distribution within memory core 10, is the establishment of automatic matching of core cell parameters and reference cell parameters over the entire effective voltage and temperature range. Each row of memory and reference core cells is programmed at a particular voltage and temperature. Accordingly, the reference cell of the applicable reference column is automatically programmed to match the conditions of the corresponding memory core cell at the particular row within memory core 10. Memory column sets 10a...lOn and reference columns 15a...15n are interconnected by word lines 19a...19n.

Fig. 2 shows a sense amplifier circuit 4 ac¬ cording to the invention which includes a "lubrication"

stage having transistors 501 and 502 and a low input capacitance buffer stage 600 including an arrangement of transistors 601, 602, 603 and 604, explained in greater detail below. Further shown in Fig. 2 are respective transistors 336, 337, 338 and 339 which form a network effective to maintain an essentially stable bias voltage at the inputs to the sense amplifier, i.e., data input node 330 and reference input node 430. The gates of transistors 336 and 337 are connected to the data input node 330, which in turn connects to the source of first pass transistor 331 and to the drain and source of "diode" connected "lubrication" transistor 501. The source of transistor 336 is connected to the ground terminal, and the source of transistor 337 is connected to the positive supply terminal. The drain of transistor 337 is connected to the second bias node 335, which is also connected to the gate of second pass transistor 342 and to the drain and gate of transistor 339. The source of transistor 339 is connected to first bias node 334, which also connects to the drain of transistor 336, the drain and gate of transistor 338, the gate of first pass transistor 331 and the gate of third pass transistor 431. The source of transistor 338 is connected to the ground terminal. Transistors 336, 337, 338 and 339 constitute a modified inverter which has an input at data input node 330 and two outputs at first bias node 334 and second bias node 335. If first pass transistor 331 is relative¬ ly large in relation to the currents flowing through it, and if it is a "0" threshold transistor then at a steady state, the voltage level of first bias node 334 will be just slightly higher than the voltage level on data input node 330. The voltage level at second bias node 335 will be still higher than the voltage level at first bias node 334 due to the voltage drop across transistor 339 as a result of the currents flowing through it. The voltage level on the drain of first pass transistor 331 will be close to the voltage level at second bias node 335, if second pass transistor 342 is built using a "0'! threshold

transistor and is large enough in relation to the cur¬ rents flowing through it. Therefore, the voltage on the drain of first pass transistor 331 is higher than the voltage on the gate of that transistor, and the transis- tor is kept saturated, a condition desirable for higher transconductance.

At steady state, both data input node 330 and reference input node 430 will be at stable levels corre¬ sponding to the sensing of either a conductive or non- conductive core cell. The variable threshold core cell, as well as variable threshold reference cell, are connected to the two sense amplifiers inputs 330 and 430, as shown in Fig. 1.

When sensing a "low" state, i.e., conductive cell, a current I oRE ^w ^ ¹¹ flow from the data input 330 towards the selected core cell, and a current I _REE will flow from reference node 430 to the selected reference cell. In the case of the core cell/reference cell ar¬ rangement shown in Fig. 1, where all the core cells within one memory output share a reference cell driven by the same word and/or sense line and also similar in size to the active core cell, I _f oRE ^w iH ke substantially equal to IR _EE -

The combination of devices formed by first pass transistor 331, second pass transistor 342, and modified inverter transistors 336, 337, 338 and 339 will force a voltage level onto data input node 330, which has a value close to the trip point of the modified inverter, where the trip point is defined as the voltage applied at the data input node 330 that results in an equal voltage level at the first bias node 334. The voltage "forced" onto the data input node will determine a current ILUBE to flow through "lubrication diode" connected transistor 501. A "mirror" image of current IL _T JBE ^w iH flow through transistor 502 which has the gate connected to data input node 330, the drain connected to the reference input node 430 and the source connected to the ground terminal. The magnitude of IpoR _C E' ^{the m} i ^rrored current flowing through

transistor 502, will depend on the ratio between the physical sizes of transistors 501 and 502. The size ratio between "lubrication" transistors 502 and 501 is preferably the same as that between the "top" current mirror transistors 441 and 341, so as to result in a can¬ cellation of "lubrication" current at sense node 340.

In a preferred embodiment of the sense amplifi¬ er, the size ratio between transistors 502 and 501 and 441 and 341 is approximately 2 to 1, i.e., at a similar voltage bias, the larger transistor will conduct twice the current of the smaller transistor. Thus, the current flowing through the reference branch, i.e., through top mirror transistor 441 and third pass transistor 431 is:

^■ • ^• RB ^{= I} FORCE ^{+ I} REF ^{= 2xI} LUBE ⁺ "'"CORE' and the current flowing through top mirror transistor 341 tends toward I _SENSE = IRB / 2 = I _LUBE + I _C0RE / 2.

Therefore, in the presence of a conductive memory cell, when I ORE ^S flowing ^ou of data input node

330 into the active core, the current flowing out of the source of first pass transistor 331 is:

^I TOT ^{= I} CORE ^{+ I} LUBE ^*

Since I _SENSE = ^I CORE/ ^{2 + I} LUBE' ^second P ^ass transistor 342 will be on and will deliver a current equal to ICO E/ ² ^ ^{n order} to maintain current balance at sense node 340. If second pass transistor 342 is on, then the voltage level at sense node 340 is relatively "low".

Conversely, if a non-conductive core cell is being read, and the current flowing from the data input node 330 towards the core is 0, then the current flowing out of the source of first pass transistor 331 is:

^I TOT ⁼ ° ^{+ I} LUBE ^{= I} LUBE ^*

Since tends toward being equal to

I _CO R _E /2 + lj_,UBE' ^{anc s} ^ ^{nce I} SENSE ^{can not} ^ ^e l ^ar 9 ^er than I _jj UBE' ^ follows that the voltage level on sense node 340 will rise until the drain of transistor 341 is high enough so that a current equal to I _L UBE ^{W;L11 f} l° ^w through transistor 341. This can only happen if transistor 341

enters into the linear region. As a result, sense node 340 is charged "up" to a relatively "high" level, but this level is below the positive supply voltage, i.e., current will still be flowing through transistor 341, thus actively clamping sense node 340 at a level below the positive voltage supply. This active clamping action limits the voltage swing at the sense node, thus improving speed, as compared to a "non-clamped" approach, without burdening the sense node with additional clamping devices as used in the prior art, seen in Fig. 3.

It follows that at steady state, the current flowing through first pass transistor 331 will vary from a low value equal to ILUB _E to a high value equal to I _jU BE + Ic _O RE* ^Tne voltage level on the data input node 330 will also vary accordingly. This voltage will be some¬ what lower when the current flowing through the first pass transistor 331 is higher. The voltage difference at data input node 330 between the "high" and respectively "low" state will be smaller in the presence of ILUBE' thus resulting in faster switching time compared to the case where Ij_, _U BE ^woulα - ^not ^e present. This is a result of the quadratic dependence of transistor current on gate to source voltage when the MOS transistor is saturated. "Lubrication diode" connected transistor 501 also plays an important role in the transient behavior of the sense amplifier. Thus, when the sense amplifier is charging up a previously discharged data input node 330, due to the finite phase lag between the first bias node 334 and the data input node 330, it is possible for first pass transistor 331 to "overcharge" the data input node

330 to the point where first pass transistor 331 is actu¬ ally shut off. If first pass transistor 331 shuts off during the data input node charge-up process and a con¬ ductive cell is being sensed, then the data input node 330 has to be discharged again until first pass transis¬ tor 331 turns on, thus permitting the data input node 330 to communicate with the sense node 340. In the presence of Ij _j UBE' ^an o ^vercnar 9 ^ed data input node 330 is

discharged faster, thus again contributing to faster response time. The overcharging scenario also applies in the case of a sudden drop in supply voltage level, when the voltage at the first bias node 334 could drop, thus shutting off first pass transistor 331. In such a case, the presence of IJ_,UBE ^s P ^{eeds U} P recovery substantially. Diode connected transistor 338 is used to control the gain of the modified biasing inverter and to improve its response time. In contrast to the prior art where the sense node 340 was connected directly to the input of inverter 350, in the present invention, the sense node is buffered from the first inverting stage. Specifically, sense node 340 is connected to the gate of "follower" connected transistor 601 and the gate of "0" threshold and "follow¬ er" connected transistor 602. The internal reference node 440 is connected to the gate of transistor 603, which acts as a current switch. The drain of transistor 603 is connected at node 650 to the input of inverter 350 and to the source of "follower" transistor 602. The source of transistor 603 is connected at "follower node" 640 to the source of "follower" transistor 601 and to the drain of "current source" connected transistor 604. The drains of transistors 601 and 602 are connected to the positive supply, and the source of transistor 604 is connected to the ground terminal. The gate of "current source" connected transistor 604 is connected to first bias node 334. Since the sense node 340 is driving "follower" type inputs, it follows that the "following" nodes 640 and 650 move "in phase" with the driving sense node, and therefore the capacitive loading presented to the sense node 340 is light compared to the case where the sense node directly drives an inverter input.

When a "low" state, i.e., conductive core cell, is detected, the voltage level at sense node 340 is below the voltage level at internal reference node 440. There¬ fore, the current flowing through "current source" con¬ nected transistor 604 is steered through transistor 603,

rather than through transistor 601, and "follower" tran¬ sistor 602 will be weak or "off". This allows inverter input node 650 to discharge to a low level which should be below the trip point of inverter 350. Conversely, when a "high" state is detected, then sense node 340 will be at a relatively high level, higher than the voltage level at the internal reference node 440. Consequently, the current flowing through "current source" connected transistor 604 will be steered through "follower" tran- sistor 601, rather than through transistor 603. Also, "follower" transistor 602 will pull inverter input node 650 high without significant "opposition" from "current switch" transistor 603. The input to inverter 350 is thus clamped "high" below the highest level of the sense node 340 and clamped "low" at a bias approximately one enhancement threshold below the relatively stable level of the internal reference node 440. Whereas the current moving the sense node 340, when it is close to the quies¬ cent state, is approximately Ϊ-CORE/ ² ' ^ ^{e cur} rents moving the input of inverter 350 can be much stronger, thus resulting in faster response time. Additional gain is provided by driving the current source transistor 604 from the bias node 334; thus, when detecting a "low" state, the bias node 334 will be relatively higher and a stronger current will be forced through transistor 604, thus discharging buffer output node 650 faster than a constant current would. Conversely, when detecting a "high" state, bias node 334 will be relatively lower and a weaker current will be forced through transistor 604, thus presenting less opposition to the upward movement of node 650.

In summary, according to the arrangement of the invention, sense time has been improved through the use of reference columns associated with groups of core col- umns, by maintaining relatively higher transconductance in the pass transistors, by actively clamping both the sense node as well as inverter input voltage swings and by reducing sense node capacitive loading through follow¬ er buffering circuitry.