Flash memory cell & array with improved pre-program and erase characteristics

[0002] The present invention relates to flash memory cell structures and methods that facilitate more efficient pre-programming and erasing of cells in flash memory arrays. The disclosed embodiments improve reliability and durability of such cells and reduce time and power requirements for pre-programming and erase operations in flash memory cell arrays.

[0003] Flash EPROM is differentiated from EPROM by the fact that Flash devices can be programmed and erased up to a guaranteed multiple number of times, whereas EPROM can only be programmed one time. In typical operation of Flash EPROM, programming, or writing, of the memory array is done on a byte-by-byte (or bit-by-bit) basis; in contrast, erasing (or resetting) of the memory array is done on a global (whole block) basis. Various sophisticated in-circuit algorithms are typically employed by Flash manufacturers to ensure that the Flash memory array can be reliably programmed and erased for a high number of erase cycles (typically>100,000 cycles). It has been found that the number of achievable program/erase cycles is directly correlated to the threshold voltage distribution of the erased bits. In other words, the greater the disparity in threshold voltages between cells, the fewer the number of available program/erase cycles. This is due to the fact that over-erased bits can cause difficulty in subsequent re-programming of the memory array, and this results in failure of the entire device. In addition, over-erased bits can potentially have high enough leakage current to cause a false reading when sensing of data is done on neighboring bits sharing the same bit line, even if these neighboring bits are at high V_t. Therefore, efforts in the field of Flash programming research have focused on tightening the V_tdistribution of the memory array after erase operation.

[0004] In most of the prior art, to facilitate a tight erase V_tdistribution, all the memory cells are first programmed before commencing an erase operation. This programming operation, commonly known as “pre-programming”, is performed to ensure that all the cells are at a uniformly high V_tbefore they are then globally erased. This pre-program operation can be thought of as a pre-conditioning of the array prior to erase operation. The theory underlying this approach is based on an assumption that if pre-programming is not done before an erase operation, then cells with stored electrons or data will be at a high V_t, while cells without stored electrons or data will be at a low V_tat the beginning of the erase operation. This relatively wide V_tdistribution at the beginning of an erase operation will translate into a wide erased V_tdistribution at the end of the same operation. If instead all the cells are placed at a uniformly high V_tprior to an erase operation, the occurrence of some of the bits having over-erasure problems is minimized. Thus the benefits of pre-programming are well-understood in the art, including the fact that it results in enhancement of the overall reliability of a Flash memory array.

[0005] To effectuate a pre-programming (conditioning) operation, the prior art typically uses the same mechanism as used in a normal array programming operation. Currently, the most preferred method for programming, and thus for pre-programming, is the use of Channel Hot Electron Injection, or CHEI. This method typically takes about 3-10 μs per bit, with each bit consuming about a peak current of 300 μA. The operation typically is done one byte, or 8 bits, at a time, until a whole sector is pre-programmed.

[0006] Unfortunately, the use of CHEI for pre-programming can exacerbate another degradation mechanism of flash cells. In particular, CHEI involves the use of high energy (hot) electrons. These hot electrons, along with the hot holes they generate in the channel of the device, can severely degrade the trans-conductance (known as Gm), and thus limit the current handling capability of the Flash memory cell. As is to be expected, this effect is more pronounced as the channel length of the Flash cell is progressively scaled downward. As currently understood, the Gm degradation mechanism is irreversible. A Flash memory cell with a severely degraded Gm will yield significantly reduced current during reading, and this results in yet another failure mechanism for the memory array.

[0007] In addition, the CHEI mechanism tends to saturate at a high threshold voltage, on the order of the control gate voltage (V_pp) of the cell. Thus, even if a pre-programming operation is performed to ostensibly set threshold voltages at a value of V_tp, the threshold voltage distribution at end of such pre-programming can nevertheless have a range between V_tpand V_pp. In some common applications, this distribution can still be as wide as 3 volts, which is excessive and can lead to further mis-operation of the device.

[0008] Yet another drawback of using CHEI for pre-programming is that as the supply voltage for the technology is scaled down below 3V, it takes progressively longer to perform the pre-program operation (and thus an erase operation as well). This is because the CHEI mechanism is not easily scalable with supply voltage. During a CHEI pre-programming operation, the drain of the cell being pre-programmed must be charge pumped to a higher voltage than the source voltage; typically this difference is in excess of 4.5 volts. Consequently, instead of being capable of performing pre-program of 8 bits simultaneously, the pre-program operation might need to be done with less than 4 bits at a time. This in effect doubles the time required for pre-programming. Furthermore as cell sizes are scaled down, a large amount of current is consumed in supply the charge pump, making this prior art method even less efficient. To compensate against this effect, the charge pump must be made larger in order to supply the same amount of current for a reduced voltage. This trend would only get worse as the supply voltage is further scaled to 1.8V and below, and the die size increases.

[0009] It can be seen therefore that the goal of decreasing the charge pump size (to scale it proportionately in accordance with the memory cell array) is not realized easily in practice because to do so defeats the goal of keeping (or certainly improving) the pre-program (erase) times. Thus, to maintain the same erase speeds, a proportionately larger (relative to the cell size) charge pump is required, and this results in overall reduced device integration density, higher fabrication costs, etc.

[0010] Accordingly, there is a substantial need for a pre-programming mechanism that does not significantly degrade device performance, and that provides a tighter V_tdistribution after a pre-programming operation. In addition, it is extremely desirable for such pre-programming method to be easily scalable with the supply voltage of flash devices.

[0011] An object of the present invention, therefore, is to provide a method of pre-programming that eliminates the need for deleterious CHEI operations in flash memory arrays.

[0012] A further object of the present is to improve the reliabily and durability of cells in a flash array so that device failures are reduced.

[0013] Yet another object of the present invention is to provide a flash memory array and pre-programming method that require less time and power to perform a pre-programming operation.

[0014] Another object of the present invention, is to provide a method of pre-programming that yields a safer, faster and more reliable erase operation for flash memory array.

[0015] Another purpose of the present invention is to provide a flash memory array structure and pre-programming method that is easily scalable with device size and voltage supply.

[0016] An additional objective of the present invention is to provide a flash memory array structure and pre-programming method that effectuates a pre-programming operation on a global (sector) basis so that the time for pre-programming of such array is invariant by sector size.

[0017] These and other objectives are accomplished by locating a memory cell in a first well of a first conductivity type, and then situating the first well within a second well of a second conductivity type. Through suitable biasing of the cell, and first and second wells, a bipolar pre-programming current can be injected from the second well into a floating gate of the cell. Since cells within any logical sector all share the same geometric well, pre-programming can be accomplished essentially simultaneously for any other memory cells within the same sector as said cell. This fact results in remarkably faster pre-programming times (on the order of 100 ms per sector, which for a typical device corresponds to less than 2 microseconds per cell in such sector). Moreover, since the inventive method relies on an injected current from the substrate, hot electrons are not generated in the cell channel, and this improves reliability and durability of a device utilizing such cell structures. Furthermore, because the present invention does not use a charge pump to generate the pre-programmed injected current, this further improves device integration density and power considerations.

[0018] The inventive process can be embedded in conventional fashion in a control circuit typically employed in flash devices for programming cells by adjusting the threshold voltages of the cells.

[0019] FIG. 1 depicts a conventional Flash memory cell as it is biased during a prior art pre-programming operation;

[0020] FIG. 2 illustrates flash memory array structure and general operational characteristics of a programming method employed in the present invention;

[0021] FIG. 3 is a graphical chart showing how threshold voltages of cells that vary significantly prior to pre-programming in a flash memory array with a well bias of 0.7 nevertheless converge rapidly in time after the pre-programming method of the present invention is employed;

[0022] FIG. 4 is similar to FIG. 3, except that a well bias of 0.9 volts is utilized for the graphical comparison.

[0023] FIG. 1 illustrates how a Flash memory cell is biased during a typical prior art pre-programming operation. This bias condition is the same as during a programming operation, and, as explained above, relies on a charge pump to maintain a drain-source differential of approximately 4.5 volts or greater to induce CHEI. An individual cell draws about 350-500 μA of peak current during pre-programming for a duration of 3 μs to 10 μs. The total programming time for a whole sector of 64 Kybte, using optimistic values, would be approximately given by the formula:

[0024] The total current supplied to 8 bits lines therefore is around 3-4 mA. This does not include the current consumed by the charge pump to supply the drain current to the cell. It can be seen immediately that this method is inefficient for a number of reasons, including the fact that it is performed in a serial manner, and therefore the time and power required increases directly as a function of device size. Furthermore, as the voltage of the array is reduced from generation to generation, the size of the charge pump must increase proportionately in order to supply the same amount of programming current. The only way to counteract against this effect is to program fewer cells at a time, and thus device pre-programming, erase operations take significantly longer. This latter effect, of course, is equally undesirable and becomes exacerbated with increasing array sizes.

[0025] FIG. 2 illustrates a preferred embodiment of a Flash memory cell array structure that is most suitable for a new method of pre-programming that eliminates many of the adverse effects associated with CHEI. In this embodiment, a Flash memory cell array is situated inside a P-Well 210 in turn located within a deep N-Well 220. This diagram figure also illustrates how the flash array structure should be biased during a pre-programming operation employed in the preferred method of the present invention. As can be seen from the figure, P-Well 210 is forward biased with respect to deep N-Well 220, with drain 260 and/or source 270 of the Flash cell biased at V_cc. In essence, it can be seen that this arrangement effectuates a bipolar transistor configuration with P-Well 210, deep N-Well 220, and drain 260 and/or source 270 of the Flash memory cell being the base, emitter, and collector respectively. In one sense, since the floating gate is also positively biased with a voltage V_pp(typically between 8.5 to 9 volts), it can be considered a collector of the bipolar transistor as well. Under these bias conditions an electron current flows from N-well 220, passing through P-well 210 and is then injected into floating gate 255, thus placing such cell in a pre-programmed state. The structure of FIG. 2 can be manufactured using a number of well-known conventional semiconductor manufacturing techniques. While FIG. 2 depicts a preferred embodiment of the present invention, it will be apparent to those skilled in the art that other suitable array geometries and structures, bias conditions and signal timing variations can be utilized for achieving the electron injection that pre-programs the flash cells. For example, another well geometry could be used, or alternatively, in lieu of a bipolar configuration, a single carrier embodiment could be implemented with a parasitic MOSFET structure configured in any one of many known methods. The key aspect of the present invention lies in the fact that it relies on a pre-programming mechanism that is substantially less harsh than conventional CHEI techniques so that device reliability is enhanced. Moreover, the present invention also does not rely on Fowler-Nordheim tunneling, another commonly used technique for programming and erasing memory cells which unfortunately has a number of significant process and structure requirements that make it difficult to scale as device size and channel lengths decrease. These include the fact, of course, that FN tunneling requires an extremely large field in the oxide layer—typically greater than 10 MV/cm to be effective—and this results in abbreviated device lifespan as well from oxide breakdown effects.

[0026] FIG. 3 illustrates the improved pre-programming performance of the present electron injection method. In this graph, different initial threshold voltages are used to represent a wide initial V_tdistribution for the cells in an array prior to pre-programming. In other words, the different symbols (squares, circles, etc.) represent cells having different initial V_tlevels; for example, the curve made of interconnected diamond shapes illustrates the behavior of cells having an initial threshold voltage of 3 volts. From the figure, it is seen that the V_tdistribution decreases from about 7V in variation initially (i.e., from −1 Volt to +6 volts) to less than 2V (i.e., from about +4 to +6 volts) after about 100 ms of pre-programming, and a Vt distribution of around 1V can be obtained after about 200 ms of pre-programming. This configuration assumes a P Well 210 bias of 0.7V (V_be). At a P Well 210 bias of 0.9V, it is seen from FIG. 4 that the V_tdistribution decreases from 7V initially to less than 2V after about 10 ms of pre-programming. After 100 ms of pre-programming, the V_tdistribution decreases further to less than 1 volt.

[0027] After this first portion of an erase operation is completed (pre-programming is only typically used as part of an erase procedure) a subsequent threshold reducing (erasing) signal can be applied to the cells in a second operation to place them into a fully erased step. This can be done in any number of well-known ways, and it can be seen that the overall erase operation thus takes place in a faster and more reliable fashion than that of prior art methods. Subsequent to this operation, the device is ready for programming with user data again using any conventional techniques, including FN, CHEI, etc.

[0028] To those skilled in the art it will be clear that this mechanism of pre-programming (erasing) does not rely on channel hot electrons that are produced by impact ionization in the channel region. Thus the electrons that are injected into the floating gate with this mechanism will not introduce significant device trans-conductance degradation. Experimental measurements performed on an embodiment of the type shown in FIG. 2 suggest that this bipolar electron injection mechanism consumes about 1 mA to 10 mA for pre-programming a whole sector of the Flash memory array. Compared to the prior art, it can be seen that the time for programming a cell in the present invention is essentially invariant, even with increasing array sizes. For a 64 Kbyte device, the present invention yields a figure of merit of less than approximately 1 μsec of pre-programming time per cell, which is substantially less than that of the prior art. While not confirmed at this time, the amount of current required for the present invention pre-programming operation is expected to increase slightly with larger sector sizes (because of a larger area to be mass-programmed). Looking at yet another performance benchmark of Time*Current, for a given sector size (i.e., an array of cells) the prior art has a figure of merit of 3 μs/cell*350 μA/cell (using the generous values above) which is 3 to 5 times worse than the present invention which yields 1 μsec/cell* 200 μA/cell. Moreover, since the present invention performs a pre-program operation on an entire sector, it is easily scaled to smaller supply voltages and denser sectors.

[0029] Furthermore, in contrast to the prior art, the present invention achieves greater device integration density and flexibility because it does not require a charge pump to generate the pre-programming voltage. This is due to the fact that the small bias voltage V_beon the P Well 210 is sufficient to supply the requisite bipolar current. This fact has been confirmed at least preliminarily in experiments conducted by the applicants in which it was discovered that varying the voltage at the source drain terminals did not change the amount of injected current into the floating gate. Thus, while the details of the physical mechanism are not entirely known to applicants, it appears that the charge pump terminals merely provide extra energy to the electrons of the injected current, and do not provide an additional source of current itself.

[0030] It is evident that a flash memory device integrated circuit can be manufactured using conventional processing means to include a control circuit for effectuating the above bias signals in the manner described above. Such article of manufacture could include a control circuit for generating the appropriate timed bias signals for the cells and wells, taken in combination with a typical flash memory cell array and conventional supporting peripheral circuitry (power supplies, address decoders, I/O data buffers, sense amplifiers, reference arrays, counters, timers, etc.). Such control circuit, processing means and peripheral circuitry can be implemented using any of a number of structures and methods well-known in the art, and are therefore not described here in substantial detail. In any event, finished integrated circuit articles embodying the present invention will exhibit superior performance since better, more uniform voltage threshold populations will be effectuated during operation of the device in the field.

[0031] The above routines for implementing the inventive processes are provided merely by way of example, and are not intended to be limiting of the present invention in any respect. Other variations of the routines will become evident to those skilled in the art based on the teachings herein. Accordingly, it is intended that the all such alterations and modifications be included within the scope and spirit of the invention as defined by the following claims.