Adaptive reference cells for a memory device

[0002] Electronic devices generally include at least one processor and a memory device. Typically, the memory device is used by the electronic device to store data and instructions for the processor. A variety of memory devices are available in the art, including such common examples as Read Only Memory (ROM), Read Access Memory (RAM) and flash memory. Additionally, each of these examples is available in several different versions to satisfy differing design considerations that may be important in particular applications.

[0003] Recently, flash memory devices have found growing commercial success in the electronic device market. This commercial success is due in part to the ability of flash memory devices to store electronic data over long periods of time without an electric power supply. In addition, flash memory devices can be erased and programmed by the end user after they are installed in an electronic device. This combined functionality is especially useful in electronic device applications, such as cellular telephones, personal digital assistants, and computer BIOS storage, and other applications where power supply is intermittent and programmability is desired.

[0004] Like other types of memory devices, flash memory devices typically include an array of individual memory transistors that are oriented in rows and columns. This array is sometimes referred to as the core, and the memory transistors are often referred to as cells, or core cells. As is common practice in the memory device art, the control gates of the memory cells in each row of the core are usually connected to a series of word lines, thus forming individual rows of cells that can be accessed by selecting the corresponding word line. Similarly, the drain regions of the cells in each column of the core are connected to a series of bit lines, thus forming individual columns of cells that can be accessed by selecting the corresponding bit lines. Finally, the source regions of all of the cells in the array are connected to a common source line. In some flash memory devices the array of transistors is further subdivided into sectors of separate transistor arrays to provide added flexibility for the programming and erasing operations.

[0005] The data stored in each memory cell represents a binary 1 or 0, as is well known in the art. To perform a program, read or erase operation on a particular cell in the array, various predetermined voltages are applied to the control gate, drain region and source region of the memory cell. Thus, by applying these predetermined voltages to a particular bit line column, a particular word line row and the common source line, an individual cell at the intersection of the bit line and word line can be selected for reading or programming.

[0006] In one common type of flash memory device, non-volatility of the memory cells is achieved by adding a floating gate between the control gate and the substrate region of the transistors. Typically, the cells of the flash memory device are programmed by applying a predetermined raised voltage to the control gate and the drain region of the cell and grounding the source region. As a result, the voltages on the control gate and the drain region cause the generation of hot electrons that are injected onto the floating gate, where they become trapped. This electron transfer mechanism is often referred to as Channel Hot Electron (CHE) injection. When the programming voltages are removed, the negative charge on the floating gate remains, thereby raising the threshold voltage of the cell. The threshold voltage is then used during reading operations to determine if the cell is in a charged state, i.e., programmed (0), or whether the cell is in an uncharged state, i.e., erased (1).

[0007] Cells are read by applying a lower predetermined voltage to the control gate and the drain region and grounding the source of the cell. The current in the bit line is then sensed with a sense amplifier. If the cell is programmed, the threshold voltage will be relatively high and the bitline current will be zero, or at least relatively low, thus registering a binary 0. On the other hand if the cell is erased, the threshold voltage will be relatively low and the bit line current will be relatively high, thus registering a binary 1.

[0008] In contrast to the programming procedure, flash memory devices are usually bulk-erased by simultaneously erasing all the cells in a memory sector. One procedure for erasing an entire memory sector involves applying predetermined voltages to the common source line and all the word lines of the sector while the drain regions of the cells are left to float. This causes electron tunneling from the floating gate to the source region through Fowler-Nordheim (F-N) tunneling, thereby removing the negative charge from the floating gate of each of the cells in the memory sector.

[0009] Typically, the memory device is provided with a number of address pins that allow the user to specify individual groups of memory cells for various operations. As is well-known in the art, the number of address pins usually provided for selecting the rows of cells is equal to log₂X, where X is the number of word lines in the memory device. Similarly, the number of address pins provided for selecting column groups of cells is equal to log₂Y, where Y is the number of bytes or words in each row of cells (a byte being eight cells and a word being sixteen cells). When the memory device is performing internal embedded functions, the address bits for the row and column bits will generally be generated by a state machine within the memory device instead of being provided by the user through the address pins. The memory device also provides a number of data pins for inputting and outputting the data stored in the core cells. Commonly, the number of data pins is equal to the number of cells in the column groups that are selected by the column address bits, e.g., eight or sixteen.

[0010] In order to translate the row and column address bits into the specific word lines and bit lines that must be selected for an operation, an X-decoder and a Y-decoder are usually provided in the memory device. As is well-known in the art, the X-decoder receives the row address bits and connects the selected word line that corresponds to the address signal to the appropriate circuits. For example, in the case of a reading operation, the X-decoder will connect the selected word line to a voltage boosting circuit.

[0011] Likewise, the Y-decoder receives the column address bits and connects the selected bit lines that correspond to the address signal to the appropriate circuits. Thus, in reading operations, the Y-decoder will connect each of the selected bit lines to a sense amplifier.

[0012] Typically, memory devices are also provided with a number of reference cells. Reference cells are well known in the art and are commonly used in a variety of comparing functions which compare the reference cell to the status of the core cells. One example of a reference cell is an erase verify reference (“ERV_rf”) cell. The ERV_rfcell is used during erasing operations to verify that each of the core cells have been successfully erased. Thus, the threshold voltage (“V_t”) of the ERV_rfcell is compared with each of the previously erased core cells to verify that the V_tof each core cell is less than the V_tof the ERV_rfcell.

[0013] Similarly, a program verify reference (“PGMV_rf”) cell is also commonly provided in the memory device. The PGMV_rfcell is used during programming operations to verify that each of the core cells have been successfully programmed. In contrast to the erase verify operation, the programming verify operation compares the V_tof the PGMV_rfcell with each of the previously programmed core cells to verify that the V_tof these cells is higher than that of the PGMV_rfcell. Generally, the V_tof the PGMV_rfcell is higher than the V_tof the ERV_rfcell, with the difference between the V_tof the PGMV_rfcell and the ERV_rfcell representing a minimum change in V_t(“ΔV_t”) between programmed and erased states. Typically, the ΔV_tmay be about 2 V.

[0014] Another reference cell that is normally provided is a read reference (“R_rf”) cell. The V_tof the R_rfcell is usually somewhere between the V_tof the PGMV_rfcell and the ERV_rfcell. In one example, the V_tof the R_rfcell is about 0.5 V higher than that of the ERV_rfcell. Accordingly, the R_rfcell is used in a comparing function with the core cells to determine whether each of the cells is in a programmed or erased state.

[0015] Typically, predetermined fixed values for the V_tof each of the reference cells is selected by the memory device designer prior to manufacturing the memory devices. In selecting these fixed V_tvalues, the memory device designer must take into account the amount of variation that occurs in the initial V_tof the core cells during manufacturing of the memory devices. The designer must also consider the amount of time that will be required to adjust the V_tof the core cells during set up of the memory device so that all the cells have a V_tbelow the predetermined fixed value of V_tfor the ERV_rfcell. In some memory devices, however, the time required to adjust the V_tof the core cells can be especially time consuming. Furthermore, some types of memory devices have a limited capacity to adjust the V_tof the core cells. In these cases the ΔV_tmay become unacceptably small and may affect the performance of the memory device after the expected manufacturing variations are factored in when predetermining the V_tvalue of the reference

[0016] A memory device setup procedure is provided for adapting reference cells to core cells. One advantage of this procedure is that the initial threshold voltages of the core cells do not need to be changed during setup of the memory device. In one stage of the method, an erase verify reference cell is adapted to the core cells until substantially all of the core cells have threshold voltages less than the threshold voltage of the erase verify reference cell. In another stage of the method, a program verify reference cell is setup by increasing the threshold voltage of the program verify reference cell by a desired change in voltage above the threshold voltage of the erase verify reference cell. A read reference cell is setup by changing the threshold voltage of the read reference cell so that it is intermediate between the erase verify reference cell and the program verify reference cell. Comparing circuits are also provided for changing the threshold voltages of the reference cells to the desired levels.

[0017] The invention, including its construction and method of operation, is illustrated more or less diagrammatically in the drawings, in which:

[0018] FIG. 1 is a chart of the prior art, showing the threshold voltages of core cells and reference cells from a memory device;

[0019] FIG. 2 is a chart, showing the threshold voltages of core cells and adaptive reference cells;

[0020] FIG. 3 is a flow chart, showing a method for adapting the reference cells to the core cells;

[0021] FIG. 4 is a comparing circuit for changing the threshold voltage of an erase verify reference cell;

[0022] FIG. 5 is a comparing circuit for changing the threshold voltage of a read reference cell;

[0023] FIG. 6 is a comparing circuit for changing the threshold voltage of a program verify reference cell; and

[0024] FIG. 7 is an alternative comparing circuit for changing the voltage of the program verify reference cell.

[0025] Referring now to the drawings, and particularly to FIG. 1, a chart is shown to demonstrate the relationship between the threshold voltage (“V_t”) of the core cells 10, 12, 14, 16 in a memory array and the reference cells 90, 92, 94 . FIG. 1 shows the prior art, in which the V_tof the first group of core cells 10, 12 from one memory device varies from a second group of core cells 14, 16 from another memory device. As is well known by those skilled in the art, manufacturing variations cause a variation 18 in the initial V_tof the core cells 10, 14 between different memory devices.

[0026] Typically, the core cells of the memory device are changed to a programmed state by applying various voltages to the gate, drain region and source region of each of the cells, thereby resulting in an increase in the V_tof the programmed cells 12, 16. Thus, in FIG. 1 the first group of core cells 10, 12 and the second group of core cells 14, 16 are also shown in programmed states 12, 16 with their V_tincreased. Since the increase in V_tbetween erased states 10, 14 and programmed states 12, 16 is generally constant for all the core cells, the initial variation 18 between the V_tof the first group 10 and the second group 14 of core cells remains after the cells are programmed 12, 16.

[0027] The use of reference cells is generally known by those in the memory device art. In the present embodiment, three separate reference cells 90, 92, 94 are provided. The first reference cell, erase verify reference (“ERV_rf”), 90 is used during erase operations in a limit test to verify that all of the erased core cells 10, 14 have been properly erased. Accordingly, during erase verification each erased core cell 10, 14 is compared to the ERV_rfcell 90. If an erased core cell 10, 14 passes less electrical current through the core cell 10, 14 than the ERV_rfcell 90, the core cell 10, 14 is determined to be in a more programmed state than the ERV_rfcell 90. Thus, the ERV_rfcell 90 is used as an upper limit in the verification test.

[0028] Likewise, the program verify reference (“PGM_rf”) cell 94, or second reference cell, is used as a lower limit to verify that the core cells 12, 16 are properly programmed. Thus, each programmed core cell 12, 16 is compared with the PGM_rfcell 94 during programming verification to ensure that the core cells 12, 16 are not in a more erased state than the PGMV_rfcell 94. The difference between erased states 10, 14 and programmed states 12, 16 of the core cells can be represented as a change in V_t(“ΔV_t”) 20. Therefore, the ΔV_t20 extends between the limits provided by the ERV_rfcell 90 and the PFMV_rfcell 94. Typically, a ΔV_t20 of about 2 V is preferred for flash memory devices.

[0029] The third reference cell, read reference (“R_rf”), 92 is used during reading operations to determine whether a selected core is in an erased state 10, 14 or a programmed state 12, 16. Accordingly, the R_rfcell 92 is compared with the selected core cell. If the core cell passes more electrical current through the core cell than is passed through the R_rfcell 92, the core cell is determined to be in an erased state 10, 14. On the other hand, when the opposite comparison occurs, the core cell is determined to be in a programmed state 12, 16. In other words, when the core cell is in an erased state 10, 14, the V_tof the core cell will be less than the V_tof the R_rfcell 92. Similarly, when the core cell is in a programmed state 12, 16, the V_tof the core cell will be higher than the V_tof the R_rfcell 92.

[0030] Typically, the value of V_tfor all three reference cells 90, 92, 94 is predetermined by the memory device designer. Therefore, the V_tvalues are usually fixed during the design of the memory device and are not changed for each memory device as they are manufactured. However, because the initial V_tof the core cells 10, 14 varies from memory device to memory device, this variation 18 must be accounted for to ensure that all the core cells have a V_tvalue below the V_tof the ERV_rfcell 90 when erased 10, 14 and a V_tabove the PGMV_rfcell 94 when programmed 12, 16. Several possible solutions are available to memory device designers to ensure proper operation of the reference cells 90, 92, 94. Usually, the initial V_tof the core cells 10, 14 can be changed by some degree during setup of the memory device. This usually involves applying erase pulses to the core cells 10, 14 to lower the V_tof the cells 10, 14. This solution can be demonstrated in FIG. 1 by visualizing the second group of erased core cells 14 being shifted to the left towards the first group of erased core cells 10 as the erase pulses are applied. One problem with this solution is the extra amount of time that is required to apply the necessary erase pulses to achieve the desired V_tshift 18 in the core cells. Moreover, in some memory devices, such as non-floating gate flash memory devices, the characteristics of the cells allows only a relatively small shift in the initial V_tof the core cells. Another solution to ensure proper values for V_tof the ERV_rfcell 90 and the PGMV_rfcell 94 is to narrow the ΔV_t20 between the two reference cells 90, 94. This, however, is an undesirable solution because the range of ΔV_t20 between the ERV_rfcell 90 and the PGMV_rfcell 94 can become so small after accounting for expected manufacturing variations in the initial V_tof the core cells 10, 14 that the operating performance of the memory device can be adversely affected.

[0031] Turning now to FIG. 2, a similar chart to that of FIG. 1 is shown to demonstrate an advantage of the invention. In the present invention the V_tof the reference cells 90, 92, 94 is adapted individually to each memory device during setup of the memory devices. Therefore, the V_tof the reference cells 90, 92, 94 will vary depending on the initial V_tof the core cells 22 of each memory device. Accordingly, as shown in FIG. 2, the initial V_tvalues of the core cells 22 in the memory device do not need to be changed from their initial values 22. Instead, the V_tof the ERV_rfcell 90, the R_rfcell 92 and the PGMV_rfcell 94 are changed to appropriate values for each individual memory device. Therefore, the V_tof the ERV_rfcell 90 is changed to position it slightly above the highest V_tof any of the initially erased core cells 22. Likewise, the V_tof the PGMV_rfcell 94 can be positioned so that it is slightly below the lowest V_tof any of the programmed core cells 24. As a result, the time required to set up the memory device is reduced because the initial V_tvalues of the core cells 22 do not need to be shifted by applying numerous erase pulses. A wider ΔV_t26 is also possible because manufacturing variations that may occur between different memory devices do not need to be accounted for. In addition, this method of adapting reference cells 90, 92, 94 to individual memory devices can be used on non-floating gate flash memories, where shifting the initial V_tof the core cells 22 is difficult.

[0032] Turning now to FIG. 3 in combination with FIGS. 4-7, a method to adapt the reference cells 90, 92, 94 to a memory device is shown. In the first stage 30-46, the V_tof the ERV_rfcell 90 is changed in response to the initial V_tvalues of the core cells 22. In the first step of the method the initial V_tof the ERV_rfcell 90 is determined. This is accomplished by applying a voltage (“V_g”) to the gate of the ERV_rfcell 90 and incrementing V_g30 until the ERV_rfcell 90 compares 32 with a master reference (“M_rf”) 98.

[0033] A comparing circuit 91, or setup circuit, for the ERV_rfcell 90 is shown in FIG. 4. Accordingly, a voltage source 96 is provided to apply V_g30 to the gate of the ERV_rfcell 90. Various voltage sources 96 are possible, but preferably a digital to analog converter with binary weighted resistors and a four or five bit counter is used. The voltage source 96 increments V_g30 in 0.1 V increments and can remember the voltage level by storing the count used in the counter. Resistors 74, 76 are also provided between separate V_ccnodes 70, 72 and intermediate nodes a and b 78, 80. Preferably, the resistors 74, 76 are cascode sense amplifiers with equal amounts of resistance. The ERV_rfcell 90 is positioned between node a 78 and a ground 86. The M_rf98 is positioned between node b 80 and a ground 88. Preferably, the M_rf98 is a resistor 98 that produces a small current at node b 80. Thus, in one example, the V_ccnodes 70, 72 have a voltage of about 3 V, and the M_rfresistor 98 has a resistance that results in an electrical current of 4.5 μA and a voltage of 1.0 V at node b 80. Node a 78 and b 80 are then connected to a comparator 82 that generates an output signal 84 when the electrical current at node a 78 and node b 80 compare 32.

[0034] The operation of the first stage 30-46 of the method is now apparent to those skilled in the art. Initially, the V_gapplied 30 to the gate of the ERV_rfcell 90 will be below the V_tof the ERV_rfcell 90. Therefore, no current flows through the ERV_rfcell 90, and the comparator 82 does not generate an output signal 84 since nodes a and b 78, 80 do not compare 30. However, when V_gis raised 30 slightly above the V_tof the ERV_rfcell 90, current begins to flow through the ERV_rfcell 90. The current at node a 78 and node b 80 now compare 32 and the comparator 82 generates the output signal 84.

[0035] Next, an erase verify operation 34 is performed on a sample of the core cells 22. The erase verify operation 34 could be performed on all the core cells 22, but a smaller sample is selected in order to minimize the time of the erase verify operation 34. Preferably, the sample includes core cells 22 located along imaginary diagonal lines extending between the corners of the memory array. Usually, the initial erase verify operation 34 will fail because some core cells will have V_tvalues that are higher than the ERV_rfcell 90.

[0036] When the erase verify operation 34 fails, the V_tof the ERV_rfcell 90 is changed 36 to adapt the ERV_rfcell to the V_tvalues of the core cell sample 22. Thus, in the described embodiment, the voltage source 96 increases the V_g36 of the ERV_rfcell by 0.1 V. Because V_gis now greater than the V_tof the ERV_rfcell 90, more current flows through the ERV_rfcell 90 and node a 78 than is flowing at node b 80. As a result, the comparator 82 does not generate the output signal 84; 38. The V_tof the ERV_rfcell 90 is then changed 40 until node a 78 and node b 80 compare 38. Preferably, the V_tof the ERV_rfcell 90 is changed by programming 40 the ERV_rfcell 90 at a rate of 5 mV/μs, thereby raising the V_tuntil the applied V_gprovides a current at node a 78 that is equal to the current at node b 80.

[0037] After the V_tof the ERV_rfcell 90 has been changed, the erase verify operation 34 is performed again. As before, if the erase verify operation 34 fails, the V_tof the ERV_rfcell 90 is changed again 36-40 and the erase verify operation 34 is retried. When the erase verify operation 34 passes, a V_tfor the ERV_rfcell 90 has been set that is slightly higher than the highest V_tfrom the core cell sample 22. The last V_gapplied 36 to the ERV_rfcell 90 is then increased by a margin amount 42. Similar steps to those previously explained are then repeated to change 46 the V_tof the ERV_rfcell 90 and compare 44 node a 78 and node b 80 until the V_tof the ERV_rfcell 90 has been increased by the margin amount. The V_tof the ERV_rfcell 90 has now been setup and is marginally higher than all of the core cells 22.

[0038] The next stage 48-52 of the method involves setting the V_tof the R_rfcell 92 to the appropriate value. In the preferred embodiment, the R_rfcell 92 has a V_tthat is 0.5 V higher than the V_tof the ERV_rf90 cell. Turning now to FIG. 5, an R_rfcomparing circuit 93 is shown for changing the V_tof the R_rfcell 92. FIG. 5 is similar to FIG. 4, with the R_rfcomparing circuit 93 including V_ccnodes 70, 72, nodes a and b 78, 80, a comparator 82, the ERV_rfcell 90 and a first voltage source 100 connected to the gate of the ERV_rfcell 90. Unlike the ERV_rfcomparing circuit 91, the R_rfcell 92 is connected to node b 80 and a ground 88. A second voltage source 102 is connected to the gate of the R_rfcell 92.

[0039] The V_tof the R_rfcell 92 is then set to the desired value by simultaneously applying a V_gto the gate of the R_rfcell 92 that is 0.5 V higher than a V_gapplied to the gate of the ERV_rfcell 90; 48. Preferably, the first voltage source 100 connected to the gate of the ERV_rfcell 90 is the same voltage source 96 that was used in FIG. 4. Furthermore, the V_gprovided by the first voltage source 100 is preferably the last voltage 42 that was used in setting up the ERV_rfcell 90. Because the initial V_tof the R_rfcell 92 is likely to be less than the V_gapplied to the gate of the R_rfcell 92 by the second voltage source 102 (i.e., V_gof first voltage source+0.5 V), more current will flow at node b 80 then at node a 78. Therefore, the comparator 82 will not generate the output signal 84; 50. The V_tof the R_rfcell 92 is then changed 52 by programming the R_rfcell 92 at a rate of 5 mV/μs. Once the V_tof the R_rfcell 82 has been increased so that it is only slightly less than the V_gapplied to the gate of the R_rfcell 92, the current flow at node a 28 and node b 80 will compare 50, and the comparator 82 will generate the output signal 84. Accordingly, the R_rfcell 92 is now setup and has a V_tthat is 0.5 V higher than the V_tof the ERV_rfcell 90.

[0040] Turning now to FIG. 6, a PGMV_rfcomparing circuit 95 is shown for performing the third stage 54-60 of the method. The PGMV_rfcomparing circuit 95 functions similarly to the R_rfcomparing circuit 93 shown in FIG. 5. Thus, the first voltage source 100 preferably applies the same V_gto the gate of the ERV_rfcell 90 that was applied in FIG. 5. The second voltage source 104 then applies a V_gto the gate of the PGMV_rfcell 94 that is 2.0 V higher than the V_gsupplied by the first voltage source 100; 56. Like the R_rfcomparing circuit 93, the current at node b 80 will initially be more than at node a 78, thus the comparator 82 will not generate the output signal 84; 58. Therefore, the V_tof the PGMV_rfcell 94 is increased 60 by programming the cell at a rate of 5 mV/μs until the current at node a 78 and node b 80 compares 58 and the comparator 82 generates the output signal 84. Accordingly, the PGMV_rfcell 94 is now setup and has a V_tthat is 2.0 V higher than the V_tof the ERV_rfcell 90. All three reference cells 90, 92, 94 have now been setup and are adapted to the V_tvalues of the core cells 22, 24; 62.

[0041] Turning now to FIG. 7, an alternative PGMV_rfcomparing circuit 97 is shown. This alternative circuit 97 is similar to the PGMV_rfcomparing circuit 95 shown in FIG. 6, but instead the M_rfresistor 98 is substituted in place of the ERV_rfcell 90 and the first voltage source 100. In this embodiment the V_gapplied to the gate of the PGMV_rfcell 94 by the voltage source 106 is 2.0 V higher than the last V_g42 that was applied to the gate of the ERV_rfcell 90 during setup of the ERV_rfcell 90; 54. Like the previous PGMV_rfcomparing circuit 95, the V_tof the PGMV_rfcell 94 is then changed 60 until the current at node a 78 and node b 80 compare 58. As is apparent to those skilled in the art, a similar alternative circuit could be provided to setup the R_rfcell 92.

[0042] While a preferred embodiment of the invention has been described, it should be understood that the invention is not so limited, and modifications may be made without departing from the invention. The scope of the invention is defined by the appended claims, and all devices that come within the meaning of the claims, either literally or by equivalence, are intended to be embraced therein.