Method and apparatus for identifying SRAM cells having weak pull-up PFETs

The present invention relates generally to integrated circuit memory devices and, more particularly, to identifying weak pull-up PFETs contained within a Static Random Access Memory (SRAM) cell.

A typical static random access memory (SRAM) cell includes an array of individual SRAM cells. Each SRAM cell is capable of storing a binary voltage value therein, which voltage value represents a logical data bit (e.g., “0” or “1”). One existing configuration for an SRAM cell includes a pair of cross-coupled devices such as inverters. With CMOS (complementary metal oxide semiconductor) technology, the inverters further include a pull-up PFET (p-channel) transistor connected to a complementary pull-down NFET (n-channel) transistor. The inverters, connected in a cross-coupled configuration, act as a latch which stores the data bit therein so long as power is supplied to the memory array. In a conventional six-transistor cell, a pair of access transistors or pass gates (when activated by a word line) selectively couple the inverters to a pair of complementary bit lines.

As memory devices continue to shrink in size over time, so do the individual cells and hence the individual devices within the cell. One problem associated with SRAM device miniaturization occurs as a result of an unstable threshold voltage (V T ) of the PFETs. In a typical six transistor SRAM cell layout, the PFETs are minimally sized in order to allow for maximum array density. A relatively small PFET, in turn, may exhibit an excessively high shift in V T after module stress. As the V T of the PFET approaches the supply voltage (V DD ), the six-transistor cell becomes a dynamic four-transistor cell. Moreover, in a cell with a weak pull-up PFET, a “high” cell node will eventually drift to ground over time, since the NFET to ground current (when the NFET is “off”) is the dominating leakage current within the cell. In such a case, a cell which has a “1” bit stored therein may, over time, flip to a “0” bit before a read operation is later performed. The reverse may also be true if both PFETs are weak or if the PFET associated with the complementary bit line is weak.

Unfortunately, during conventional SRAM testing, a cell with a weak PFET(s) may appear healthy. In addition, the SRAM bit line circuitry can accidentally refresh the cell nodes during a read operation, thereby helping to hide the problem. As a selected word line is activated, all cells connected thereto are exposed to the bit line bias, which was set to V DD during a restore operation. If both PFETs within the cell are equally weak, and if the high cell node had drifted down near ground, both cell nodes will be pulled up to about the same level through the passgates. Eventually, if the inactive standby time is long enough, both cell nodes will drift back down toward ground and at about the same voltage level. In such cases where the voltage difference between the left cell node and the right cell node is about zero, the cell is considered to have been disturbed to the “X” state.

During a subsequent read operation, an SRAM cell in the “X” state may or may not fail the test. A cell in the “X” state which passes the test is considered a problematic escapee. If only one of the PFETs is weak, and the cell is at the state where the weak PFET should be turned off, then the defect is virtually undetectable. Unlike a dynamic random access memory (DRAM) cell where the data loss is mainly from the “1” state to the “0” state, an SRAM cell that has lost data may in certain instances be read as a healthy cell. Present screening methods for identifying weak pull-up PFETs, therefore, are a “hit or miss” venture. In practice, some screening methods involve shortening the test cycle time and lowering the power supply V DD , thereby trying to make the weak cells fail the write operation. Other existing methods modify the write cycle to allow a “weak write” operation to flip the weak cells. However, these existing weak write operations are too strong, thereby allowing good cells to be screened out by mistake.

The above discussed and other drawbacks and deficiencies of the prior art are overcome or alleviated by a method for determining the memory cell stability of individual memory cells included within a memory array. In an exemplary embodiment, the method includes presetting each memory cell to a first logic state and then applying a gradually increasing, controlled leakage current to a node within each memory cell. The voltage of each of the nodes within each corresponding memory cell is then monitored. Then, for each memory cell within the memory array, the level of leakage current which causes the memory cell to be changed from the first logic state to a second logic state is determined. The level of leakage current which causes the memory cell to be changed from the first logic state to the second logic state corresponds to the threshold voltage of a pull-up PFET within the memory cell.

Referring to the exemplary drawings wherein like elements are numbered alike in the several Figures:

FIG. 1 is a schematic of an existing six-transistor SRAM cell structure;

FIG. 2 is a graph which illustrates the noise margin of a given SRAM cell as a function of the threshold voltage (V T ) of the PFETs therein;

FIG. 3 is another graph which illustrates the noise margin of a given SRAM cell as a function of the threshold voltage (V T ) of the PFETs therein;

FIG. 4 is a schematic of a screening circuit coupled to the SRAM cell of FIG. 1, in accordance with an embodiment of the invention;

FIG. 5 is an alternative embodiment of the screening circuit in FIG. 4; and

FIG. 6 is an alternative embodiment of the screening circuit in FIG. 5 .

Referring initially to FIG. 1, there is shown a conventional SRAM cell structure 100 which represents a single memory cell included within a memory array arranged in rows and columns. The SRAM cell structure 100 includes a six-transistor memory cell 102 which is capable of storing a binary bit of information. Specifically, the memory cell includes a pair of cross-coupled, complementary metal oxide semiconductor (CMOS) inverters. One inverter includes an NFET storage transistor N 1 and a PFET load transistor P 1 . Similarly, a second inverter includes an NFET storage transistor N 2 and a PFET load transistor P 2 . Transistors P 1 and P 2 are often referred to as pull-up transistors because of their coupling to the voltage source V DD . The memory cell 102 further contains NMOS access transistors (i.e., passgates) NL and NR serving as switches, each of which are coupled between the bistable circuit (P 1 , N 1 , P 2 and N 2 ) and a pair of complementary bit lines BL and BR, respectively. Passgates NL and NR are activated by an appropriate signal generated on a word line WL.

The junction node A (i.e., the drains) of transistors P 1 and N 1 , as well as the gates of transistors P 2 and N 2 , are coupled through passgate NL to bit line BL. The junction node B (i.e., the drains) of transistors P 2 and N 2 , as well as the gates of transistors P 1 and N 1 are coupled through access transistor NR to complementary bit line BR. In the above described SRAM cell structure 100 , data is stored as voltage levels within the two sides of the bistable circuit (P 1 , N 1 , P 2 and N 2 ) in opposite voltage configurations; that is, node A is high and node B is low in one state, and node A is low and the B is high in the other state, thereby resulting in two stable states. Node B is thus the logical complement of node A.

Again, as indicated earlier, in the event that one or both of the PFETs P 1 , P 2 are weak, such a condition may escape detection in conventional SRAM testing and thus effectively result in a “dynamic” SRAM which (unless periodically refreshed) may lose data over a sustained period of inactivity.

FIG. 2 is a graph which illustrates the noise margin (i.e., tolerable leakage current versus voltage) of a given SRAM cell as a function of the threshold voltage (V T or V TP ) of the PFETs therein. As the threshold voltage V T (defined as the difference between the supply voltage V DD and the gate to source voltage V GS which causes the PFET to conduct) increases, the cell noise margin degrades. In other words, as V T increases, the cell node is less tolerant to outward leakage current (and inward leakage current in the case of two bad PFETS).

The “N-curves” shown in FIG. 2 are determined by isolating the cell from the bit lines and then forcing the cell node voltage from ground to V DD (approximately 1.2 volts) with an external voltage source. As the cell node voltage is switched from ground to V DD , the current from and to the voltage source is plotted. In each plot, the current from the voltage source is positive, while the current to the voltage source is negative. The middle zero-current intercept represents the metastable state of the cell, which drifts to the “0” state to the left or the “1” state to the right if disturbed. The positive portion of each curve represents the cell tolerance of leakage current into the node when the node is in the low or “0” state. If an SRAM cell contains two bad cells therein, the inward leakage current tolerance may be as low as 24 μA for a cell to flip from “0” to “1”.

Similarly, when the data stored in the cell node is high or in the “1” state, the negative current values (on the right portion of each curve) represent the current leakage that the cell can tolerate without losing the data. In the case of a healthy cell having PFETs with a V T of about 0.4 volts, for example, it can be seen that the cell will tolerate up to about 25 μA of NFET leakage current. However, as V T starts to increase, the leakage current tolerance (and thus cell stability) decreases. For example, if V T of one or both of the PFETs is about 0.8 volts, it may still be within an acceptable range, but it can be seen that the leakage current tolerance is reduced to about 6 μA. Further, if the PFET threshold voltage V T is high enough such that it equals the supply voltage V DD , there is zero leakage tolerance and any NFET leakage current will eventually pull a high node down to ground.

FIG. 3 is another graph illustrating the noise margin characteristics of an SRAM cell where only one of the two PFETs are weak. In this case, the logic state where the weak PFET should be “off” is as stable as that in a normal good cell. Thus, the left side of the curves shown in FIG. 3 (representing the leakage tolerance to inward leakage current) is substantially the same regardless of the threshold voltage of the weak PFET. On the other hand, a high threshold voltage in just one of the two cell PFETs will still result in a decreased leakage current tolerance for a cell node at the high logic value.

Therefore, in accordance with an embodiment of the invention, a test circuit is provided which not only provides a method for identifying “good” and “bad” pull-up PFETs, but which also provides a method for identifying the specific distribution of threshold voltages for all PFETs within a memory array such as an SRAM.

Referring now to FIG. 4, there is shown a test circuit 200 which is included within each bit column in the memory array. Test circuit 200 may be characterized as a current source which provides a leakage current path for the internal cell junction nodes A and B. In this manner, the precise threshold voltage for both PFETs within each memory cell may be determined. Test circuit 200 includes a first pull-down leakage transistor T 4 , and a second pull-down leakage transistor T 5 . Transistors T 2 and T 4 provide pull-down leakage for junction node A through input terminal LEAK 0 , while T 3 and T 5 provide pull-down leakage for junction node B through input terminal LEAK 1 . Effectively, T 2 is a pass gate which selectively couples pull-down transistor T 4 to BL, and T 3 is a pass gate which selectively couples pull-down transistor T 5 to BR. In a preferred embodiment, transistors T 2 , T 3 , T 4 and T 5 comprise PFET transistors to allow for a smaller layout; however, it will also be appreciated that NFETs may also be used to provide a more elaborate leakage current source with precise control.

In order to determine the threshold voltage V T of P 1 in each cell within a given column, each cell within the column is first initialized to the “1” state. That is, junction node A has an initial voltage of V DD (e.g., 1.2 volts). A control signal applied to LEAK 0 will then implement a soft write “0” function by causing T 2 to conduct, thereby generating a pull-down current through T 4 to ground. Thus, if the magnitude of the applied leakage current exceeds the leakage tolerance of P 1 (depending upon its threshold voltage), the voltage value at junction node A will be flipped to the “0” state.

Similarly, the threshold voltage of P 2 may be determined by initializing junction node B to the “1” state and applying a control signal to input terminal LEAK 1 . This time, a leakage current is drawn from junction node B, through complementary bit line BR and transistors T 3 and T 5 , to ground. It will be noted that in applying the leakage current to each cell junction node A and B, the word line WL is activated so that the junction nodes are coupled to the bit line pair.

In one aspect, test circuit 200 may be used to screen out PFETs having a V T above a certain predefined value by applying a particular voltage value to LEAK 0 and LEAK 1 . For example, if the input voltage at LEAK 0 were selected such that T 2 were to draw about 10 μA of controlled leakage current from junction node A, this soft write function would screen out those PFETs P 1 having a V T of greater than 0.6 volts. This will be understood by recalling from FIG. 2 that a cell PFET with a V T of about 0.61 volts has a maximum leakage current tolerance of about 14 μA, whereas a cell PFET with a V T of about 0.81 volts has a maximum leakage current tolerance of about 6 μA. Because a cell PFET with a V T of about 0.61 volts will tolerate a leakage current of 10 μA without allowing a “1” bit stored in the junction node to be flipped or lost, those PFETs having threshold voltages which are unacceptably higher will be screened out. The cell voltage at junction node A is monitored during the application of controlled leakage current thereto. If the data is lost as a result of the application of a leakage current, then the PFET P 1 has a threshold voltage greater than 0.61 volts.

In another aspect, test circuit 200 may also be configured for (in addition to establishing a pass/fail status for a PFET) determining the threshold voltage of each PFET to determine a distribution of V T values throughout the entire SRAM array. Rather than applying a preset input voltage value to LEAK 0 and LEAK 1 (and thus drawing a fixed leakage current), a range of input voltages may be applied thereto. By starting at a very small input voltage and drawing a very small leakage current at the outset (e.g., on the order of nanoamperes), those PFETs with the highest threshold voltages will first be determined and measured. Then, the threshold current may be gradually increased until each cell node voltage begins to flip from the “1” state to the “0” state. The value of the threshold voltage for a given PFET will be determined by the amount of leakage current which caused the cell to flip from high to low.

FIG. 5 illustrates an alternative embodiment of the test circuit 200 shown in FIG. 4 . In order to prevent any latching circuits connected to the bit columns from masking off the leakage current control, pull-up PFETs T 0 and T 1 are used to selectively couple BL and BR, respectively, to the supply voltage V DD . Passgate T 2 (in addition to selectively coupling T 4 to BL) selectively couples pull-up transistor T 0 to BL, while passgate T 3 (in addition to selectively coupling T 5 to BR) selectively couples pull-up transistor T 1 to BR. During a soft write “0” test operation while current from BL is being pulled down, BR is thus held at V DD , since the signal applied at LEAK 0 is also connected to the gate of T 1 , thereby coupling BR to V DD . This is so, despite the pull-down effect of N 2 , since T 1 is much larger than N 2 . Conversely, during a pull-down test on junction node B, bit line BL is held to V DD through transistor T 0 .

FIG. 6 illustrates still an alternative embodiment for the test circuit 200 in FIG. 5 . In this embodiment, an additional pair of input terminals LEAK 0 P and LEAK 1 P are used in order to separate a pull-down signal from the complementary pull-up signal, and can originate from a separate chip pad. Thus configured, test circuit 200 provides a sufficient pull-up to counteract any NFET pull-down effects which might otherwise result in the complementary side from being pulled toward ground and confuse the screening process.

Through the above described embodiments, the test circuit 200 may be used to qualify individual pull-up PFETs within a memory cell as “good” or “bad”, as well as to determine the distribution of threshold voltages thereof for each PFET in the array. In the latter case, such information may prove to be quite useful in analyzing the overall technology process control. It will also be appreciated that the disclosed method and apparatus embodiments may also be applied in principle to an instance where it is desired to screen for possible weak pull-down NFETs in SRAM cells (e.g., N 1 and N 2 ). Although as a practical matter, it is the PFETs that are susceptible to weakness during the manufacturing process thereof, the testing process may nonetheless apply to the NFETs.

While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for elements thereof without departing from the scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from the essential scope thereof. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed as the best mode contemplated for carrying out this invention, but that the invention will include all embodiments falling within the scope of the appended claims.