Local reference voltage generator for non-volatile memories - Patents.com

The use of a local reference voltage generator in a block-divided NVM array stabilizes detection margins, addressing statistical and temperature-related issues in 1T1C F-RAM cells, improving reliability and extending the operating life of non-volatile memory devices.

JP7583156B2Active Publication Date: 2024-11-13INFINEON TECHNOLOGIES LLC
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
JP2023519852
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2020-12-15
Filing Date
2021-09-30
Publication Date
2024-11-13
Estimated Expiration
2041-09-30

AI Technical Summary

Technical Problem

Conventional non-volatile memory (NVM) devices face issues with statistical variations and temperature dependence in the P-term and U-term values, leading to reduced detection margins and premature end-of-life, particularly in large arrays using single-ended sensing architectures like 1T1C F-RAM cells.

Method used

The implementation of a memory device with an array of NVM cells divided into blocks, utilizing a local reference voltage generator that provides address-dependent reference voltages for each block, optimizing the detection circuit by determining and applying a local reference voltage based on the specific block being read, thereby stabilizing the margin between P-term and U-term.

Benefits of technology

This approach enhances the reliability, yield, and operating life of the memory device by maintaining consistent detection margins across varying temperatures and cell variations, allowing for larger array sizes and improved repair methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007583156000005
    Figure 0007583156000005
  • Figure 0007583156000006
    Figure 0007583156000006
  • Figure 0007583156000007
    Figure 0007583156000007
Patent Text Reader

Abstract

A memory device including a reference voltage (VREF) generator and a method of operation for improving the memory sensing margin of the device and extending the operating temperature range and lifetime of the device are disclosed. Generally, the device further includes an array of non-volatile memory cells divided into a plurality of blocks, and a sensing circuit coupled to the array for receiving a memory signal from the array and comparing it to VREF to read data from the cells. The local reference voltage generator is configured to provide one of multiple reference voltages to the sensing circuit based on which block is being read. The array can be divided based on row and column addresses of cells within the block. The cells include 1T1C ferroelectric random access memory (F-RAM) cells, and the reference voltage is selected based on the lowest P-term or highest U-term of the cells within the block being read.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] (CROSS REFERENCE TO RELATED APPLICATIONS) This application is an international application of U.S. Nonprovisional Application No. 17 / 122,284, filed December 15, 2020, which claims priority under 35 U.S.C. 119(e) to U.S. Provisional Application No. 63 / 085,823, filed September 30, 2020, all of which are incorporated by reference in their entireties.

[0002] The present disclosure relates generally to semiconductor memories, and more particularly to a memory device comprising an array of non-volatile memory (NVM) cells divided into segments or blocks and including a local reference voltage generator, and a method for operating the same. [Background technology]

[0003] Integrated memory devices using arrays of non-volatile memory (NVM) cells employing single-ended sensing, such as 1T1C (one transistor, one capacitor) architecture, provide high memory density. NVM memory cells store data as a charge, such as in silicon-oxide-nitride-oxide-silicon (SONOS) based charge trapping NVM cells, or as the polarization state of a ferroelectric capacitor in ferroelectric random access memory (F-RAM) cells. In single-ended sensing / reading, the data stored in the cell is read by comparing the voltage developed on the cell's bit line due to the cell's charge or polarization state to a reference voltage. The NVM cell is considered to be programmed or erased based on the result of the comparison of the bit line voltage or current to the reference. For example, a 1T1C F-RAM cell is shown in FIG. 1. The cell 100 includes a ferroelectric capacitor 102 having a first plate connected to a plate line (PL) and a second plate coupled to a bit line (BL) via a transistor 104 controlled by a word line (WL). Data is written to the cell by applying an electric field by charging both plates of the ferroelectric capacitor 102, forcing the atoms in the ferroelectric capacitor to point up or down (depending on the polarity of the charge) and storing a P-term or logic "1" or a U-term or logic "0". In other embodiments, the logic / binary states of the "P-term" and "U-term" can be reversed. In single-ended detection, the cell 100 is read by applying a read voltage between the plate line PL and the bit line BL, or pulsing the plate line, and activating the transistor 104. The voltage of the "P-term" or "U-term" is output via the bit line BL and compared to a reference voltage (VREF) using a sense amplifier 106 in the memory device.

[0004] A reference voltage (VREF) is generated by performing a margin sweep on an array of memory cells to define the expected bitline voltage from reading programmed (P-term) and erased (U-term) cells. According to one embodiment, VREF is selected to be between the highest U-term (U0) and lowest P-term (P0) voltages measured across the memory array, so that when memory cell 100 is read, a resulting bitline voltage above the reference is recognized as programmed and a voltage below the reference is recognized as erased. Figure 2 shows an example margin sweep of an F-RAM memory array, illustrating the statistical variation of bitline voltage (memory signal) from reading P-terms and U-terms. The difference between the lowest P-term (P0) and the highest U-term (U0) represents the overall detection margin (U-margin + P-margin).

[0005] FIG. 2 illustrates a margin sweep of memory signals from memory cells in an F-RAM array that shows statistical variations in P-term and U-term voltages (bit line voltages) by different cells in an array of cells in the NVM, and the impact on the overall sensing margin of the NVM. With reference to FIG. 2, it can be seen that one problem with conventional NVM devices employing single-ended sensing using a global VREF applied via a reference bit line (referred to as BL) to all or many cells in the device, such as an array of 1T1C F-RAM cells, is the statistical variation of the P-term and U-term values ​​of individual cells. In FIG. 2, each black dot between 0 and approximately 55 millivolts (mV) corresponds to the unprogrammed term or U-term of one cell. The U-margin 202 of the NVM device corresponds to the voltage difference between VREF and the best or worst case U-term (U0) of one or more cells in the NVM device. Similarly, each black dot between approximately 140 and 200 mV corresponds to the programmed term or P-term of one cell. The P-margin 204 of an NVM device corresponds to the voltage difference between VREF and the lowest or worst-case P-term (P0) of a cell in the NVM device. These statistical variations are particularly problematic for the P-margin because retention losses and other time-dependent factors have been observed to cause more degradation of the P-term (a downward movement of P0) than the U-term.

[0006] A further problem with memory devices using an array of 1T1C cells in an F-RAM array is due to the temperature dependence of the P-term and U-term. FIG. 3 is a graph showing the temperature dependence of the P-term and U-term of a memory cell when VREF is set 21 mV higher than the worst case U-term (U0), i.e., about 83 mV. Referring to FIG. 3, it can be seen that as the temperature of the NVM device increases, the P-margin (the difference between the worst case P-term (P0 302) and VREF) decreases. For example, at 0° C., the P-margin of the NVM device is about 57 mV, while at about 125° C., the P-margin drops to about 22 mV. Compounding this temperature dependence is the time-dependent degradation of the temperature-dependent P0 302 as it moves from an initial high voltage to a lower voltage over the life of the NVM device. This time-dependent degradation at a particular temperature is illustrated in FIG. 3 by the vertical line 302. For example, over the life of an NVM device, P0 302 at 0° C. may drop from a high of about 155 mV to about 128 mV, or from a P-margin of about 72 mV to about 45 mV. This time-dependent degradation combined with temperature dependency may induce the P-margin to drop to virtually zero at high operating temperatures, resulting in premature end-of-life when there is no longer enough P-margin between P0 302 and VREF at the highest anticipated operating temperature to reliably determine the state of the memory cell via a read operation.

[0007] Referring again to Figure 3, it is noted that the worst case or best U-term (U0 306) is also subject to temperature- and time-dependent degradation. However, it is observed that the degradation induced by both factors is less in the U-term (U0 306) than in the P-term (P0 302).

[0008] Finally, because the range of P-term and U-term values ​​for individual cells is the result of statistical variation, and it is assumed that the larger the array, the greater the observed variation, it should be understood that the size of the array is limited by reduced margins in worst-case memory cells when a single global VREF is employed across the entire array.

[0009] Therefore, what is needed is a memory device including an array of 1T1C NVM cells and a method of operating the same that optimizes the margin between VREF and the P and U terms to enable memory devices with larger arrays, improves reliability and yield, and extends the operational life of the device. Summary of the Invention

[0010] A memory device or system including an array of cells divided into segments or blocks, a reference voltage generator capable of providing local reference voltages for reading different blocks, and a method of operating the same are disclosed. In general, the device includes a reference voltage generator, an array of non-volatile memory (NVM) cells divided into a number of blocks, and a detection circuit coupled to the array for receiving a memory signal from the array to read data from the cells and comparing the memory signal to a local reference voltage. The reference voltage generator is configured to provide one of a number of reference voltages to the detection circuit based on which block is being read. The NVM cells are comprised of a number of rows, each of which shares a word line and a plate line, and a number of columns, each of which shares a bit line, and the array is logically divided into a number of blocks based on row and column addresses of the NVM cells in each block, and the reference voltage generator is configured to provide one of a number of reference voltages to the detection circuit based on row and column addresses of the NVM cells in the number of blocks.

[0011] A method of operating a memory device including an array of memory cells divided into a plurality of blocks and a reference voltage generator capable of providing an optimized local reference voltage for each block is also disclosed. In general, the method begins with logically dividing the array of memory cells in the memory device into a plurality of blocks based on row and column addresses of cells in each block. Next, multiple margin sweeps are performed for each block to determine an optimized local reference voltage for each block. From the local reference voltage for each block, a base reference voltage for the array and an adjusted reference voltage for each block are determined, the adjusted reference voltage corresponding to a voltage by which the base reference voltage is offset to provide the local reference voltage for each block. The adjusted reference voltage for each block is stored in a lookup table coupled to the memory device or the reference voltage generator. During a read operation of a memory cell in one of the plurality of blocks, a final reference voltage or a local reference voltage substantially equal to the optimized local reference voltage for the block is generated by referencing the adjusted reference voltage for the block from the lookup table and combining it with the base reference voltage. The local reference voltage is applied to an input on a reference bit line of a sense amplifier coupled to the bit line of the memory cell being read.

[0012] The reference voltage generators and methods of the present disclosure are particularly useful when an array includes one transistor-one capacitor (1T1C) ferroelectric random access memory (F-RAM) cells and the reference voltage used to read one of the blocks is selected based on the lowest P-term (P0) or highest U-term (U0) of the F-RAM cells in the block being read.

[0013] Embodiments of the present invention will now be described, by way of example only, with reference to the accompanying schematic drawings in which corresponding reference characters indicate corresponding parts, and in which: FIG. 1 illustrates an embodiment of the present invention; and FIG. 2 illustrates an embodiment of the present invention; and FIG. 3 illustrates an embodiment of the present invention; [Brief description of the drawings]

[0014] [Figure 1] Schematic diagram of a one-transistor-one-capacitor (1T1C) ferroelectric random access memory (F-RAM) cell. [Diagram 2] FIG. 1 illustrates a margin sweep of memory signals from memory cells in an F-RAM array, illustrating the statistical variation of P-terms and U-terms and their impact on the overall sensing margin. [Diagram 3] 1 is a graph showing temperature and time-dependent degradation of P-term and U-term of memory cells in an F-RAM array. [Figure 4] 1 is a block diagram of a system including a memory array and an address-dependent local reference voltage generator. [Diagram 5] Block diagram of a memory array divided into several blocks for which address-dependent reference voltages are generated. [Figure 6A] FIG. 2 is a schematic block diagram illustrating a reference voltage generator including a base VREF generation circuit and an on-the-fly adjusting VREF circuit. [Figure 6B] FIG. 2 is a schematic block diagram illustrating a reference voltage generator including a base VREF generation circuit and an on-the-fly adjusting VREF circuit. [Figure 6C] FIG. 2 is a schematic block diagram illustrating a reference voltage generator including a base VREF generation circuit and an on-the-fly adjusting VREF circuit. [Figure 7] 7 is a schematic diagram of a detection circuit and switches coupling the local reference voltage generator of FIG. 6 to a memory array in a system. [Figure 8] FIG. 6D is a timing diagram of signals to the local reference voltage generator of FIG. 6A through FIG. 6C. [Figure 9A] FIG. 1 is a schematic diagram of a local reference voltage selection circuit for obtaining a VREF adjustment value from a lookup table stored in a monitor memory based on a memory address. [Figure 9B] FIG. 1 is a schematic diagram of a local reference voltage selection circuit for obtaining a VREF adjustment value from a lookup table stored in a monitor memory based on a memory address. [Figure 10] FIG. 1 is a flow diagram of a method for determining base and trim voltages for a memory device or system including multiple blocks and using address-dependent local reference voltages and operating the memory device or system. [Figure 11A] 1 is a graph illustrating memory margin results for a memory system operating using a single global reference voltage. [Figure 11B] 1 is a graph illustrating memory margin results for a memory system operating using a single global reference voltage. [Figure 12A] 1 is a graph showing memory margin results for a memory system operating with an address-dependent local reference voltage. [Figure 12B] 1 is a graph showing memory margin results for a memory system operating with an address-dependent local reference voltage. [Figure 13] FIG. 1 is a flow diagram of a method for repairing a failed bit in a memory device using a local VREF segment (LVS) repair method. [Figure 14] 1 is a diagram of the P-margin of an unrepaired memory device, a memory device repaired using a global repair method, and a memory device repaired using an LVS repair method. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] A memory device or system including an array of cells divided into segments or blocks, a reference voltage generator capable of providing local reference voltages for reading different blocks, and methods for operating them are provided that improve yield and reliability, extend the operating life and temperature range of the device, and are particularly useful for use in or with non-volatile memories (NVM) employing single-ended sensing, such as 1T1C ferroelectric random access memories (F-RAMs).

[0016] In the following description, for purposes of explanation, numerous specific details are set forth in order to facilitate a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without these specific details. In other instances, well-known structures and techniques are not shown in detail or are shown in block diagram form in order to avoid unnecessarily obscuring the understanding of this specification.

[0017] Reference to an embodiment or embodiments herein means that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of the invention. The phrase "in one embodiment" in various places in this specification does not necessarily all refer to the same embodiment. The term "coupled" as used herein may include both a direct electrical connection between two or more components or elements and an indirect connection through one or more intervening components.

[0018] Briefly, the memory device includes an array of non-volatile memory (NVM) cells divided into a number of blocks, and a detection circuit coupled to the array for receiving signals from the array and comparing the memory signals to a reference voltage to read data from the NVM cells. The memory device further includes or is coupled to a reference voltage generator that provides a reference voltage to the detection circuit. The reference voltage generator is configured to provide one of a number of reference voltages to the detection circuit based on which block is being read. The block being read may be identified by an address or addresses of the NVM cell(s) being read.

[0019] FIG. 4 shows a block diagram of a system 400 including a memory device 402 having a memory array 404 divided into a number of segments or blocks 406, and an address-dependent local reference voltage (VREF) generator 408. According to FIG. 4, the memory device 402 further includes several detection circuits 410 coupled between the memory array 404 and the local reference voltage generator 408 in this embodiment. The detection circuits 410 receive address-dependent local reference voltages from the local reference voltage generator 408, determine the binary or logical state (i.e., programmed or erased state) of the memory cells in the block 406 being read, and output the read data via an interface or data path 412. In one embodiment, the memory array 404 can include 1T1C F-RAM cells whose P-terms or U-terms are compared to the local reference voltages provided by the local reference voltage generator 408 to determine their binary state (programmed with a logical 1 or not). In other embodiments, the memory device may include other types of NVM cells that employ single-ended sensing configurations, such as charge-trapping silicon-oxide-nitride-oxide-silicon (SONOS) based cells, floating gate cells, R-RAM cells, M-RAM cells, and the like.

[0020] The system 400 further includes a reference voltage selection circuit 414 for operating or controlling the local reference voltage generator 408, which provides an address-dependent reference voltage optimized for the block 406 being read. In this embodiment, these circuits include a configuration circuit 416 for providing the local reference voltage generator 408 with a base reference voltage (VREF) setting that is common or substantially similar across all blocks 406 in the memory array 404, and a digital control circuit 418 for providing an adjusted reference VREF to the local reference voltage generator. As will be described in more detail below, the adjusted VREF setting enables the local reference voltage generator 408 to generate an adjusted reference voltage that is combined with or used to offset the base reference voltage to generate an address-dependent reference voltage optimized for the particular block 406 being read.

[0021] Generally, as in this embodiment, system 400 also includes an address buffer 420 coupled to digital control circuitry 418 and a non-volatile monitor memory 422. Address buffer 420 receives the NVM cell address(es) being read and communicates with digital control circuitry 418, allowing the digital control circuitry to identify the block 406 being read (in which the requested NVM cell(s) are located). Monitor memory 422 includes one or more lookup tables that store information used by digital control circuitry 418 to generate the adjustment reference voltage settings.

[0022] Although the circuits and elements of system 400 of FIG. 4 described above are shown as separate blocks, it will be appreciated that each of these circuits and elements, including memory device 402, may be integrally formed on a single semiconductor substrate or chip or packaged in a single multi-chip package with the memory device or other elements of the system not shown, and that a system may refer to the same thing as a memory device.

[0023] The circuits and elements of the system 400 of FIG. 4, and their operation, will now be described in more detail with reference to FIGS.

[0024] 5 is a block diagram of a memory array 500 divided into multiple segments or blocks 502 for which address dependent reference voltages are provided. Each segment or block 502 may be addressed by a column and row address decoder in a memory device (not shown) and is identified or defined by a range of column and row addresses of memory cells in the memory array 500. For example, the memory array 500 may include a 4 Mb memory array divided into sixteen (16) 256 Kb subarrays 504, each subarray including one or more blocks 502. A margin sweep is performed for each block 502 and an optimized local reference voltage is determined for each block 502 individually. From these local reference voltages, a base reference voltage for the entire memory array 500 or all blocks 502 in the memory array 500 is determined along with the appropriate adjustment reference voltage for each block 502.

[0025] Additionally or alternatively, the memory array 500 may be physically (but not logically) divided into a number of physically separated blocks 502, with each individual block being provided with a local reference voltage from one of a number of local reference voltage generators coupled to one or a few individual smaller blocks 502.

[0026] An embodiment of a reference voltage generator will now be described in more detail with reference to Figures 6A-6C. Figure 6A is a detailed schematic diagram of a portion of an embodiment of a reference voltage generator including a base reference voltage (VREF) generator or generation circuit 602 and an on-the-fly adjusted VREF generation circuit 604. In general, the reference voltage generator includes an array of metal oxide silicon (MOS) capacitors arranged in binary groups, each of which is coupled to a positive voltage supply or voltage source (Vdd) and a negative supply voltage (Vss) or ground via a number of first MOS switches 606 to precharge the capacitor, and to a distribution bus 618 via a number of second MOS switches 608 to share the charge on the capacitors in the base VREF generation circuit 602 to generate the base reference voltage. The capacitors are split between the base VREF generation circuit 602 and the adjusted VREF generation circuit 604. The total number of capacitors depends on the values ​​of Vdd and Vss and the number of increment voltages between Vdd and Vss selected for the potential reference voltages. Similarly, the number of capacitors included in the base VREF generation circuit 602 and the adjusted VREF generation circuit 604 depends on the number of increments between Vdd and Vss selected for the adjusted reference voltage.

[0027] According to this embodiment, the base VREF generation circuit 602 includes a first array of 225 metal oxide silicon (MOS) capacitors 603 arranged in binary groups: a single capacitor (X1), a group of two capacitors (X2), a second group of four capacitors (X4), etc., with the number of capacitors doubling with each successive group. As will be explained in more detail below, the only exception to this configuration is that the largest set or group (X113) has 113 capacitors instead of 128, with the remaining capacitors included in a second array 605 of 15 capacitors in the adjusted VREF generation circuit 604 to facilitate address-dependent adjustment of the reference voltage.

[0028] The input signals pVdd[0:7] and pVss[0:7] precharge a set of capacitors to Vdd or Vss, respectively, by controlling the switches 606 in the base VREF generation circuit 602. The binary arrangement of MOS capacitors allows the capacitors to be incrementally controlled in discrete steps from 1 to 255 capacitors based on the state of the base VREF setting provided by the pVdd[0:7] and pVss[0:7] signals. The pVdd[0:7] and pVss[0:7] signals are generated by a first logic circuit 610, as shown in FIG. 6B and included in the reference voltage generator. According to FIG. 6B, a precharge signal (prch) is provided to the first inputs of the first and second AND gates 612, 614. The base VREF setting signal is then provided from the configuration circuit (configuration circuit 416 of FIG. 4) to a second input of a first AND 612 gate and through a first inverter 616 to a second input of a second AND 614 gate to generate the pVdd[0:7] and pVss[0:7] signals, respectively. Although only one first logic circuit 610 is shown in FIG. 6B, it should be understood that the reference voltage generator includes multiple logic circuits, one for each group of MOS capacitors X1 through X113.

[0029] One advantage of the reference voltage generator design is that it takes some time for the capacitors to be reliably precharged to Vdd or Vss and ready for the next memory cycle. Therefore, the pVdd[0:7] and pVss[0:7] signals cannot be set on the fly based on the incoming address. In the illustrated design, these input settings (pVdd[0:7] and pVss[0:7]) are set so that the base VREF generation circuit 602 can provide a common or base reference voltage to all blocks in the memory array. After the precharge step, all the capacitors in the base VREF generation circuit 602 are connected together and onto the distribution bus 618 coming out of the reference voltage generator (charge sharing) by closing the switch 608 controlled by the share signal.

[0030] The regulated VREF generation circuit 604 includes a second array or block of two sets of 15 capacitors, each of which is made up of binary groups X1, X2, X4, and X8. These sets or groups of capacitors are precharged by a switch 620 controlled by a precharge signal (prch). All of the 15 capacitors in the first set are precharged to Vdd, and all of the second set are precharged to Vss. After the precharge step, selected groups from these sets of 15 capacitors are shared on a distribution bus 618 by a switch 622 selectively controlled by sVdd[0:3] and sVss[0:3] input signals. The sVdd[0:3] and sVss[0:3] signals are generated by a second logic circuit 624 included in a reference voltage generator as shown in FIG. 6C. According to FIG. 6C, the same shared signal used to couple the capacitors in the base VREF generation circuit 602 to the distribution bus 618 is applied to a first input of a third AND gate 626. The adjusted VREF setting signal provided by the digital control circuit 418 is then applied to a second input of a third AND gate 626 to generate the sVdd[0:3] signal, and via a second inverter 628 to generate the sVss[0:3] signal. Again, although only one second logic circuit 624 is shown in FIG. 6C, it should be understood that the reference voltage generator includes multiple logic circuits for generating the signals sVdd[0] through sVdd[3] and the signals sVss[0] through sVss[3].

[0031] An important feature of the design is that the sVdd[0:3] and sVss[0:3] signals can be changed on the fly based on the incoming address, since there is a complete set of 15 capacitors precharged to Vdd and Vss ready to be selectively shared to the distribution bus 618 without delay.

[0032] FIG. 7 is a schematic diagram showing a circuit for coupling the local reference voltage generator of FIG. 6A to FIG. 6C to a detection circuit and through the detection circuit to a bit line (BL) in a memory array. According to FIG. 7, the circuit 700 includes a first MOS switch 702 for coupling a distribution bus 618 from the reference voltage generator to a bus capacitor 704, and a second MOS switch 706 for coupling the bus capacitor to a staging capacitor 708 in the detection circuit 710. While a share-to-bus signal is applied or asserted to the first and second MOS switches 702, 706, the charge generated on the capacitor array of the reference voltage generator is shared on the staging capacitor 708. The detection circuit 710 further includes a third switch coupled between the staging capacitor 708 and a comparator or sense amplifier (sense amplifier 714) in the detection circuit 710. While the shared-to-reference BL signal (share-to-ref-BL) is asserted, and typically after the shared-to-bus signal is deasserted, an address-dependent local reference voltage generated by the charge shared on the staging capacitor 708 is applied to the reference bitline (BL) input of the sense amplifier 714. It should be understood that the elements shown on the left side of FIG. 7, i.e., the first and second MOS switches 702, 706, and the bus capacitor 704, may be included inside the local reference voltage generator or detection circuit 710.

[0033] Although only one detection circuit 710 is shown in FIG. 7, it should be understood that a memory device or system (not shown) may typically include multiple detection circuits, from 1 to x, where x is equal to the number of bit lines (BLs) in each block of the memory array to be read.

[0034] The operation of an embodiment of a local reference voltage generator for generating an address dependent local reference voltage will now be described with reference to Figures 6A-6C and 7 and Figure 8 showing timing diagrams of the signals shown in Figures 6A-6C and 7. According to Figure 8, the operation begins with asserting a precharge signal (prch 802) between t0 and t1 to precharge the capacitors in the regulated VREF generation circuit 604. During this time, the base VREF setting or signal (base VREF[0:7] 804) is static and is coupled to a logic circuit 610 which, together with the precharge signal, generates signals pVdd[0:7] and pVss[0:7] to precharge the capacitors in the base VREF generation circuit 602. At time t1, the precharge signal is deasserted and between times t1 and t2, an adjusted VREF setting based on an address in the block being read (adj.VREF[0:3] 806) is coupled to the second logic circuit 624 in preparation for selectively sharing multiple capacitors in the adjusted VREF generation circuit 604 to the distribution bus 618. At time t2, the share signal 808 is asserted, closing the second switch 608 to share the charge on the capacitors in the base VREF generation circuit 602 to the distribution bus 618. At substantially the same time, a share signal applied to a third AND gate 626 in the second logic circuit 624 generates sVdd[0:3] and sVss[0:3] signals to close switch 622 and share the charge on selected capacitors in the adjusted VREF generation circuit 604 to the distribution bus 618. At time t3, the share signal is deasserted and switches 608 and 622 are opened. Between time t4 and time t5, a shared-to-bus signal 810 is asserted, closing the first switch 702 and the second switch 706 and allowing the charge generated on the reference voltage generator capacitor array to be shared on the staging capacitor 708. Between time t6 and time t7, a share-to-ref-BL signal 812 is asserted, allowing an address-dependent local reference voltage generated by the charge shared on the staging capacitor 708 to be applied to the reference BL input of the sense amplifier 714.

[0035] With reference to the schematic diagrams of FIG. 9A and FIG. 9B, a local reference voltage selection circuit used with or included in the reference voltage selection circuit 414 for obtaining a VREF adjustment value from a lookup table based on a memory address and adjusting or trimming the reference voltage on the fly is described below. According to FIG. 9A, the selection circuit 900 receives an adjustment signal 902 as a vector containing predetermined adjustment values ​​for each segment or block of the memory array. Each segment uses the same number of bits for adjustment. The vector signal is a concatenation of all adjustment values ​​in the order in which the addresses of the segments appear in the memory array. The adjustment signal or vector is coupled to an even adjustment value lookup table 904 and an odd adjustment value lookup table 906. The even adjustment value lookup table 904 stores a selected subset of its vectors for only segments with even-numbered segment addresses. The odd adjustment value lookup table 906 is a selected subset of its vectors for only segments with odd-numbered segment addresses. The live_segment_addr signal 908 is received as a vector containing addresses of segments that are being actively accessed. It is updated only when a transaction is initiated and when a burst transaction crosses an address boundary between segments. It remains static at the end of the transaction unless cleared by a reset. The buffer_segment_addr signal 910 is received as a vector signal containing addresses of segments that have not yet been initiated, and although it may not yet be a complete value in that the address is still being transmitted from the address source to the selection circuit 900, it always remains available and static once the entire value is received by the selection circuit. A first multiplexer 912 selects between live_segment_addr and buffer_segment_addr based on a live_select signal 914 received from a portion of the selection circuit shown in FIG. 9B, and outputs the selected address.A burst transaction naturally increments the address during a data phase such as burst progress. To facilitate burst transactions, the selection circuit 900 includes an increment / decrement block 916 for calculating a successor address of a currently selected address based on the state of a received cr_decrement signal 918. When asserted, the cr_decrement signal 918 indicates that the successor address should be a decrement operation from the current address in this burst transaction. The increment / decrement block 916 provides the successor address to a second multiplexer 920 coupled to the even adjust value lookup table 904 and a third multiplexer 922 coupled to the odd adjust value lookup table 906.

[0036] The least significant bit (LSb) of the currently selected address determines whether the segment is considered even or odd. The LSb is used as a selector for second and third multiplexers 920, 922. If the LSb is 0, the second multiplexer 920 coupled to the even adjust value lookup table 904 selects the current address and the selected address, with the LSb truncated, is provided as an index into the even adjust value lookup table 904. If the LSb is 1, the second multiplexer 920 coupled to the even adjust value lookup table 904 selects the subsequent address and the selected address, with the LSb truncated, is provided as an index into the even adjust value lookup table 904. The result of the lookup is provided on an even_Vref vector signal 924. If the LSb is 0, a third multiplexer 922 coupled to the odd adjust value lookup table 906 selects the subsequent address, and the selected address, with the LSb truncated, is provided as an index into the odd adjust value lookup table 906. If the LSb is 1, a third multiplexer 922 coupled to the odd adjust value lookup table 906 is selected, and the selected address, with the LSb truncated, is provided as an index into the odd adjust value lookup table 906. The result of the lookup is provided on the odd_Vref vector signal 926.

[0037] The state of the live_select signal 914 is determined by the selection circuit shown in FIG. 9B. According to FIG. 9B, the test_scan_mode signal 928 selects the operation mode of the selection circuit 900. When the test_scan_mode signal 928 is 1, the mode is test scan mode, where all flip-flops 930, 932 operate with a synchronous clock signal, and all other elements in the functional path are assumed not to operate. When the test_scan_mode signal 928 is 0, the selection circuit 900 is in operation mode, and the clock signal to the flip-flops 930, 932 may be asynchronous. The first multiplexer 934 selects the clock signal to be communicated to the flip-flops 930, 932 and other elements based on the test_scan_mode signal 928. When the test_scan_mode signal 928 is 1, the first multiplexer 934 communicates a synchronized clk_tc_ms_tile clock signal 936, where the active edge of the clock signal is on the falling edge. When the test_scan_mode signal 928 is 0, the signal passing through the first multiplexer 934 is the result of a logical NOR 938 of the active read strobe signals (tc_ms_rd_strb_b1 clock signal 940 and tc_ms_rd_strb_b0 clock signal 942) that indicate the start of a macro read operation. The output of the first multiplexer 934 is communicated through an inverter 944 to a clock gating cell (scan_gater 946) that is enabled by the test_scan_mode signal 928. The output from the scan_gater 946 is inverted by an inverter 948 and provided to the scan control / observation flip-flop 930, toggling the flip-flop only when the selection circuit 900 is operating in scan mode.

[0038] The local VREF lookup operation is enabled by the local_vref_en signal 950, which is a control signal provided to generate the live_select signal 914 by enabling a clock signal to the live_select flip-flop 932 via a clock gating cell (rd_gater 952). The rd_gater 952 receives the inverse of the clock signal that passes through the first multiplexer 934. The output of the rd_gater 952 is then inverted by an inverter 954 such that the phase of the output matches the phase of the clock that passed through the first multiplexer 934, and the clock to the live_select flip-flop 932 only toggles when the local_vref_en signal 950 is 1. In the event of a rising edge of the tc_ms_rd_strb_b1 signal 940 or a rising edge of the tc_ms_rd_strb_b0 signal 942 when the local_vref_enable signal 950 is 1, the Q output of flip-flop 930 operating through an inverter 955 causes the flip-flop 932 to take in data at its D input, thus indicating on the live_select signal 914 that the address to be used at that time is the live_segment_address signal 908.

[0039] When a new transaction is started, it is determined whether it is a read transaction or not, and if it is a read transaction, it is determined whether it includes an address phase or not. If the current transaction is a read transaction that does not include an address phase, the destination address is address zero, and a tc_ms_opc_rcont signal 956, which is a pulse one clock period in length, is received indicating that a starting address of zero is found on the buffer_segment_addr signal 910 shown in FIG. 9A. If the current transaction is a read transaction that includes an address phase, a tc_ms_addr_ld1 signal 958, which is a pulse one clock period in length, is received indicating that a starting address of the transaction is found on the buffer_segment_addr signal 910. If the new transaction is not a read transaction, neither the tc_ms_opc_rcont signal 956 nor the tc_ms_addr_ld1 signal 958 is asserted, and the starting address of the transaction is found on the live_segment_addr signal 908.

[0040] The tc_ms_opc_rcont signal 956 and the tc_ms_addr_ld1 signal 958 are combined in a logical OR operation by OR gate 960 to generate the return_to_buf_adr signal, which is inverted via inverter 962 and combined with an active low reset signal (rst_tc_ms_tile_n 964) in AND gate 966. The resulting logical AND is input to a second multiplexer 968 which, when selected, clears or resets the live_select flip-flop 932 if the tc_ms_opc_rcont signal 956 or the tc_ms_addr_ld1 signal 958 is received. The output from AND gate 966 is also combined with the output from NOR gate 938 in exclusive OR gate 970 to provide the data input to the scan control / observation flip-flop 930. The data input to scan control / observation flip-flop 930 can be logically expressed as: ((NOT(return_to_buf_adr) AND rst_tc_ms_tile) XOR (tc_ms_rd_strb_b0 NOR tc_ms_rd_strb_b1)). As a result, stuck-at faults can be observed on any signal used as an input to this circuit when 900 is operating in scan mode. The reset state of the output of scan control / observation flip-flop 930 is 0, and the default state of the D input to live_select flip-flop 932 (coupled from the Q output of flip-flop 930 operating through inverter 955) is 1 when test_scan_mode signal 928 is 0. The test_reset_control signal 972, when asserted, forces the subsequent active-low reset signal to an inactive high state. If test_scan_mode is 1, then the second multiplexer 968 selects the logical OR of test_reset_control and rst_tc_ms_tile_n 964. If test_scan_mode 928 is 0, then the second multiplexer 968 selects the logical AND of rst_tc_ms_tile_n 964 and (NOT(return_to_buf_adr)).The output of the second multiplexer 968 is called rst_local_async_n. It provides an active low reset signal to the live_select flip-flop 932.

[0041] With reference to the flow diagram of FIG. 10, a method for determining base and trim voltages and operating a memory device or system using address-dependent local reference voltages is described below. According to FIG. 10, the method generally begins with dividing a memory array in a memory device into multiple blocks based on row and column addresses of cells in each block (1002). As described above, this can include simply logically dividing the array, or physically dividing the array, whereby each individual block is provided with an optimized local reference voltage from one of many local reference voltage generators coupled to one or a small number of blocks in the larger memory array. The division can also include logically or physically dividing the larger memory array into multiple sub-arrays, each of which includes one or more blocks.

[0042] Next, multiple margin sweeps are performed for each block to determine an optimized local reference voltage for each block (1004). The local reference voltage is a voltage that provides a pre-determined or defined minimum margin (U-margin) between the local reference voltage and the highest erase term or U-term in the block. As mentioned above, the erase term or U-term is generally observed to be less subject to statistical fluctuations and more stable over time and temperature, so the pre-determined minimum margin is generally selected to be substantially smaller than the margin (P-margin) between the local reference voltage and the lowest program term or P-term in the block.

[0043] Margin sweep means sweeping or monotonically and gradually increasing or decreasing the reference voltage applied to the memory array or block and repeatedly reading the memory cells to determine the bit line signals or voltages from the read, program or P-term, and erase or U-term. As mentioned above, statistical variations in the bit line signals are expected from reading different memory cells due to process, voltage, and temperature. In particular, in F-RAM, statistical variations in the P-term and U-term are expected due to variations in the dimensions or dielectric constant of the ferroelectric layer of the ferroelectric capacitor.

[0044] To perform a margin sweep, the entire array or block is written to the P-term (internal 1) and the memory is read using a reference voltage set to a level below the minimum expected statistical variation in the P-term. At this reference voltage, 100% should pass, i.e., all cells should be correctly read as programmed or P-term. The reference voltage is then stepped up and the write and read are repeated. At some point, the reference voltage exceeds the lowest P-term in the block (P0) and the associated memory cell fails, i.e., it is erroneously read as erased or U-term. As the reference voltage is increased, more and more programmed cells are erroneously read as erased, i.e., read errors (expected due to statistical variation) occur. A graph can be made that shows the P-term bit fail count as a function of the swept reference voltage. After a predetermined number of bit fail counts or when all or substantially all of the P-terms have failed, the process is repeated by writing erased or U-terms to the entire array or block and sweeping the reference voltage from an initial high voltage above the maximum expected statistical variation in the U-terms to a lower voltage, producing a graph of U-term bit fail count as a function of reference voltage. These two graphs can be combined as shown in FIG. 2 above. The distance between the lowest P-term (P0) and a particular reference voltage indicates the P-margin for that reference voltage. The distance between the highest U-term (U0) and the reference voltage indicates the U-margin.

[0045] The base reference voltage for the entire array is then determined from the local reference voltage found for each block in the previous step, and an adjusted reference voltage for each block is determined (1006) for an offset from the base reference voltage. According to one embodiment, the base reference voltage is determined or selected by setting the base reference voltage to a voltage a pre-determined or defined amount away from the highest internal "0" or U-term in the entire array. The adjusted reference voltage can then be determined by calculating the voltage to which the base reference voltage must be adjusted for each block to achieve or obtain the local reference voltage found for each block. This results in a constant or nearly constant margin (U-margin) between the address-dependent final or local reference voltage and the highest internal "0" or U-term of each block across the array. Moreover, the margin is substantially lower than the margin (P-margin) between the local reference voltage and the lowest program term or P-term of each block, so that the P-margin is also maximized.

[0046] The trimmed reference voltages for each block are then stored in a look-up table in a monitor memory coupled to the reference voltage generator (1008). Next, during a read operation of a memory cell in one of the blocks, a final or local reference voltage is generated (1010) by generating a base reference voltage, looking up the trimmed reference voltage for the block being read, and generating and combining the trimmed reference voltage with the base reference voltage. Finally, the local reference voltage is coupled or applied (1012) to a reference bit line of a sense amplifier coupled to the bit line of the memory cell being read.

[0047] 11A and 11B and 12A and 12B, improvements in memory margins for a memory device or system that operates using address-dependent local reference voltages versus conventional memory systems that operate using a single global reference voltage are described.

[0048] 11A and 11B show memory merge results for a memory device that includes a 4Mb F-RAM array divided into sixteen 256Kb blocks and conventionally operates using a single global reference voltage. FIG. 11A shows a graph of the lowest P term (P0) 1102 and highest U term (U0) 1104 of each block in the array, and a single global reference voltage 1106 that is used to read from all blocks in the array. The global reference voltage is selected in a conventional manner by selecting it as a fixed offset from the highest U term (U0) or lowest P term (P0) in the array. According to this embodiment, the global reference voltage is selected to have a fixed offset of 19mV from the highest U term, shown here as being in block 10. FIG. 11B shows the resulting P margin and U margin across each block in a memory device that conventionally operates using a single global reference voltage 1106. In Figure 11B, line 1108 represents the P margin and line 1110 represents the U margin. The resulting minimum, maximum and average P and U margins are shown below in Table 1. Note that operating with a single global reference voltage results in a low P margin of 18mV, which is problematic as it may lead to bit or read failures at elevated temperatures and may reduce the operational life of the memory device. [Table 1]

[0049] 12A and 12B are diagrams illustrating memory margin results for the same memory device operated using address-dependent local reference voltages as described herein. FIG. 12A illustrates a diagram of the lowest P term (P0) 1202, highest U term (U0) 1204, and sixteen (16) address-dependent local reference voltages 1206 for each block in the array. As described above, the address-dependent local reference voltages 1206 are determined using the method described above. That is, the base reference voltage is determined by setting a predefined offset from the highest U term (U0) in the array, i.e., 19 mV higher than U0 for block 10, determining the trimmed reference voltages for each of the 16 segments, and generating the local reference voltages for each of the 16 segments or blocks. FIG. 12B illustrates the resulting P margin 1208 and U margin 1210 across each block in a memory device operated using address-dependent local reference voltages. The resulting minimum, maximum and average P and U margins are shown in Table 2 below. Note that the address-dependent local reference voltage results in a substantially constant U margin across all segments or blocks. Note further that operating with the address-dependent local reference voltage increases the minimum P margin by 63% and the average P margin by 33%. [Table 2]

[0050] In another embodiment, a local VREF segment (LVS) repair method is disclosed for repairing failed bits in a memory device having an array of NVM cells divided into multiple blocks and including a local reference voltage generator.

[0051] In a conventional global repair method, the bit or memory cell having the lowest program term or P-term of all memory cells in the array is replaced with a previously unused spare bit or memory cell in the array or fabricated on the die or chip with the array. Briefly, repair and replacement is typically accomplished by opening links isolating the memory to be replaced from the word lines, plate lines and bit lines in the array before packaging the die. A spare memory cell, typically formed coupled to one or more of the same word lines, plate lines and bit lines as the memory cell being replaced, is coupled to the array in place of the failed memory cell by updating row and column decoders in the memory device so that bits addressed to the failed memory cell are instead directed to the spare memory cell.

[0052] The global repair method is sufficient for memory devices with a single large array and / or using a single global reference. However, in memory devices that include an array of 1T1C cells divided into multiple blocks and use local reference voltage generators as described above, the margin between the lowest program term or P margin is determined by the difference between the lowest program term or P0 in a block or segment and the address-dependent local VREF. Thus, the lowest program term or P0 in the entire array does not mean that the associated cell or bit has the lowest P margin, since the local VREF is not the same in all local blocks or segments. Therefore, using the global repair method for a memory device with multiple blocks and using local reference voltage generators will result in over-repair in some segments with lower references and under-repair in some segments with higher references.

[0053] In contrast, the LVS repair method repairs the bits with the lowest program term or P term (P0) only in blocks or segments whose program term margin or P margin is less than a predefined value. Thus, by using the LVS repair method, over-repair of strong segments is avoided, under-repair of weak segments is avoided, and the program term margin or P margin of a die is maximized.

[0054] FIG. 13 is a flow diagram of an embodiment of an LVS repair method for repairing failed bits in a memory device including an array of 1T1C NVM cells divided into multiple blocks and further including the above-mentioned local reference voltage generator. According to FIG. 13, in general, the method begins by dividing an array of memory cells in the memory device into multiple blocks based on row and column addresses of cells in each block (1302). A number of margin sweeps are performed for each block to determine an optimized local reference voltage for each block (1304). Next, a block is identified in which the margin between the memory cell having the lowest program term or P0 in the block and the local reference voltage is below a predefined minimum value (1306). The block is then repaired by isolating the memory cell having the lowest program term or P0 in the identified block that has a margin below the predefined minimum value and combining a spare memory cell in the block instead (1308).

[0055] With reference to the graph shown in FIG. 14, the advantages and improvements of the LVS repair method over the conventional global repair method for repairing failed bits of a memory device having an array of 1T1C cells divided into multiple blocks will be described below. The graph in FIG. 14 shows the P margin of an unrepaired memory device, a memory device repaired using a 40-bit global repair method, and a memory device repaired using the LVS repair method. The data for the graph shown in FIG. 14 was obtained using a 4 Mb F-RAM array divided into sixteen 256 Kb segments or blocks operated using sixteen (16) address-dependent local reference voltages. Bits or memory cells with failed P-terms were first repaired using the global repair method and then repaired using the LVS repair method. Bits with failed U-terms in both cases were repaired using a three-bit global replacement. According to FIG. 14, line 1402 shows the P margin of each block or block of an unrepaired memory device. Line 1404 shows the P margin of each block or blocks of the same memory device repaired using the global repair method, and line 1406 shows the P margin of each block or blocks of the memory device repaired using the LVS repair method. The resulting average P margin, minimum P margin, maximum P margin, and margin of the memory device are shown in Table 3 below. Note that all values ​​in Table 3 are expressed in units of the local reference voltage VREF. It should be further noted that the margin of the memory device is substantially equal to the most restrictive P margin of the blocks in the memory device, i.e., the minimum P margin. With reference to FIG. 14, it can be seen that by applying the LVS repair method, under-repaired segments (indicated by P term 1408) are eliminated and over-repairs (indicated by P term 1410) are substantially minimized. With reference to Table 3, it can be seen that the P margin of the memory device repaired using the LVS repair method is improved by 11% (from 44 to 49 VREF units) compared to the global repair method. [Table 3]

[0056] In the following, we describe a method for performing LVS repair of 1T1C F-RAM by performing an independent n-step P-margin binary search on each block or segment, where each step of the n-step P-margin binary search is performed on every segment before proceeding to the next step. A final repair solution is obtained in the final step such that no segment is under-repaired and over-repair of segments is minimized.

[0057] Before performing an LVS repair, three values ​​must be determined: 1. The highest U-term (U0) of each segment. This value is used as the starting point for each LVS during repair and for calculating the P-margin. 2. Predefined minimum 1T1C margin limits (P0-U0), which are based on reliability and yield evaluation. If no repair solution is found for certain parts, these parts are rejected. The minimum 1T1C margin limits (P0-U0) ensure the reliability of the passed parts. 3. A predefined P-margin search range. This is based on the statistics of the device's P-margin distribution including 6 sigma of the distribution. The search range determines the number of search steps. For example, 2 5 i.e. 32 and 2 6 That is, to perform a binary search of a search range of 64 VREF units, 5 and 6 search steps are required, respectively.

[0058] In general, the repair search starts from the first interval of a binary search with VREF=U0+margin limit+N steps. The first interval corresponds to the midpoint of the binary search range. In each interval, repair is attempted for all segments. A particular interval is considered repairable only if all segments are repairable. Below, a five-step binary search is used for illustration purposes. In this example, the following values ​​are used for U0 of an 8-segment device: [Table 4]

[0059] The 1T1C margin limits (P0-U0) are predefined or set in 32 VREF units, and the binary search range is set in 32 VREF units. Therefore, five binary search steps are required in this example. In the first interval, an attempt is made to repair each segment at VREF=U0+32. If successful, the second interval is calculated by adding 8 to the first interval. If unsuccessful, the second interval is calculated by subtracting 8 from the first interval.

[0060] An attempt is made to repair each LVS with VREF=U0+32+2nd interval. If successful, the 3rd interval is calculated by adding 4 to the 2nd interval. If unsuccessful, the 3rd interval is calculated by subtracting 4 from the 2nd interval. This sequence continues for two more intervals: ±2 and ±1. The repair solution determined with the highest repairable VREF is stored in the device.

[0061] The embodiments of the present invention have been described above using functional block diagrams and schematic block diagrams showing implementation of specific functions and relationships related to the embodiments. The boundaries of these functional building blocks have been arbitrarily defined in this specification for the convenience of description. Other boundaries may be defined as long as the specific functions and relationships related to the embodiments are appropriately implemented.

[0062] The foregoing description of the specific embodiments fully reveals the general nature of the present invention, so that others may easily modify and / or adapt the specific embodiments for various applications by applying the knowledge to the extent possessed by those skilled in the art, without undue experimentation and without departing from the general concept of the present invention. Such adaptations and modifications are therefore intended to be within the meaning and range of equivalents of the disclosed embodiments, based on the teaching and guidance presented herein. It is to be understood that the terms or phrases in the present specification are for the purpose of explanation and not for the purpose of limitation, and that the terms or phrases in the present specification should be interpreted in light of the teaching and guidance by those skilled in the art.

[0063] It is understood that the Detailed Description section, and not the Summary and Abstract sections, are intended to be used to interpret the claims. The Summary and Abstract sections may describe one or more example embodiments of the invention, but not all, as contemplated by the inventor(s), and are therefore not intended as limiting the scope of the invention and the appended claims.

[0064] The breadth and scope of the present invention should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.

Claims

1. 1. A ferroelectric random access memory (F-RAM) device, comprising: an array of F-RAM cells arranged as a number of rows with each row sharing a word line and a plate line and a number of columns with each column sharing a bit line, the array being divided into a number of blocks based on row and column addresses of the F-RAM cells in each of the blocks; a detection circuit including at least one comparator having a first input coupled to a bit line of one of the blocks being read to receive a memory signal, and a charge sharing element coupled to a second input of the comparator through at least one switching circuit; It is equipped with the coupled charge-sharing element is further coupled to a local reference voltage generator for receiving a reference voltage based on an address for the one of the plurality of blocks being read; the F-RAM cells comprise one transistor, one capacitor (1T1C) F-RAM cells, and the reference voltage is selected based on a highest U-term (U0) of the F-RAM cells in the one of the blocks being read. F-RAM device.

2. the local reference voltage generator comprises: a base reference voltage generation circuit that provides a base voltage; and an adjustment reference voltage generation circuit that provides an adjustment voltage based on which of the plurality of blocks is being read to offset the base voltage to generate the reference voltage that is provided to the detection circuit.

2. The F-RAM device of claim 1.

3. The F-RAM device further comprises a control circuit coupled to the local reference voltage generator, the control circuit comprising: a configuration circuit for controlling the base reference voltage generating circuit; and a regulation control circuit for controlling the regulated reference voltage generating circuit.

3. The F-RAM device of claim 2.

4. the F-RAM device further comprising a monitor memory coupled to the regulated reference voltage generation circuit, the monitor memory including one or more look-up tables that store information enabling the regulated reference voltage generation circuit to operate to provide the regulated voltage based on which of the plurality of blocks is being read.

4. The F-RAM device of claim 3.

5. the base reference voltage generation circuit may comprise a first array of capacitors, one or more of which may be coupled to transfer charge to the charge-sharing element in the detection circuit and provide the base voltage to the charge-sharing element; the regulated reference voltage generation circuitry may comprise a second array of capacitors, one or more of which may be coupled to transfer charge to the charge sharing element to offset the base voltage and generate the reference voltage provided to the detection circuit; 5. The F-RAM device of claim 4.

6. 1. A system comprising: an array of non-volatile memory (NVM) cells divided into a plurality of blocks; a detection circuit coupled to the array for receiving a memory signal from the array and comparing the memory signal to a reference voltage to read data from the NVM cells; a local reference voltage generator coupled to the detection circuit and providing a reference voltage to the detection circuit; It is equipped with the local reference voltage generator is configured to provide one of a number of reference voltages to the detection circuit based on which of the plurality of blocks is being read; the NVM cells comprise one transistor, one capacitor (1T1C) memory cells, and the one of the multiple reference voltages is selected based on a highest erase term of the NVM cells in the one of the blocks being read. system.

7. a first reference voltage of multiple reference voltages provided to the detection circuit for reading one of the multiple blocks is different from a second reference voltage of multiple reference voltages provided for reading another one of the multiple blocks; The system of claim 6.

8. the NVM cells are arranged into a number of rows, each row sharing a word line, and a number of columns, each column sharing a bit line; the array being logically divided into a plurality of blocks based on row and column addresses of NVM cells in each of the plurality of blocks; the local reference voltage generator is configured to provide one of the multiple reference voltages to the detection circuit based on the row and column addresses of NVM cells in each of the plurality of blocks. The system of claim 6.

9. the local reference voltage generator comprises: a base reference voltage generation circuit that provides a base voltage; and an adjustment reference voltage generation circuit that provides an adjustment voltage based on which of the plurality of blocks is being read to offset the base voltage to generate the one of the multiple reference voltages that is provided to the detection circuit. The system of claim 6.

10. The system further comprises a control circuit coupled to the local reference voltage generator, the control circuit comprising: a configuration circuit for controlling the base reference voltage generating circuit; and a regulation control circuit for controlling the regulated reference voltage generating circuit. The system of claim 9.

11. The system further comprises a monitor memory coupled to the regulated reference voltage generation circuit, the monitor memory including one or more look-up tables that store information enabling the regulated reference voltage generation circuit to operate to provide the regulated voltage based on which of the plurality of blocks is being read. The system of claim 10.

12. the NVM cells are arranged into a number of rows, each row sharing a word line, and a number of columns, each column sharing a bit line; the array being logically divided into a plurality of blocks based on row and column addresses of NVM cells in each of the plurality of blocks; the system further comprising an address buffer coupled to the regulated reference voltage generating circuit to identify which of the plurality of blocks is being read. The system of claim 11.

13. the base reference voltage generation circuit may comprise a first array of capacitors, one or more of which may be coupled to transfer charge to a staging capacitor in the detection circuit and provide the base voltage to the staging capacitor; the regulated reference voltage generation circuit comprises a second array of capacitors, one or more of which may be coupled to transfer charge to the staging capacitor to offset the base voltage and generate the one of the multiple reference voltages provided to the detection circuit; The system of claim 9.

14. 1. A method of operating a memory device, the method comprising: logically dividing an array of memory cells in said memory device into a plurality of blocks based on row and column addresses of cells in each block; performing multiple margin sweeps for each block to determine a local reference voltage for each block; determining a base reference voltage for the array and an adjustment reference voltage for each block from the local reference voltage for each block; storing the adjusted reference voltage for each block in a look-up table; during a read operation of one of the plurality of blocks, looking up and generating the adjusted reference voltage for the block, and combining the adjusted reference voltage with the base reference voltage to generate the predetermined local reference voltage for the block being read; applying the local reference voltage to a reference bit line of a sense amplifier coupled to a bit line of the memory cell being read; Including, The step of performing multiple margin sweeps for each block comprises: determining the highest internal "0" for the block by writing a "0" to each memory cell in the block, starting with an initial sweep reference voltage, reading all memory cells in the block, decreasing the initial sweep reference voltage in steps, repeating the writing and reading until no memory cell can be read correctly, and recording the sweep reference voltage at which the memory cell fails as the highest internal "0"; method.

15. The step of performing multiple margin sweeps for each block comprises: determining a lowest program term for the block by writing a "1" to each memory cell in the block, starting with an initial sweep reference voltage, reading all memory cells in the block, incrementally increasing the initial sweep reference voltage, repeating the writing and reading until no memory cell can be properly read, and recording the sweep reference voltage at which the memory cell fails as the lowest program term; the local reference voltage is set to a predefined voltage higher than the sweep reference voltage having the highest internal "0" failure, and a first margin between the local reference voltage and the sweep reference voltage having the highest internal "0" failure is less than a second margin between the local reference voltage and the sweep reference voltage having the lowest program term failure; The method of claim 14.

16. The memory cells comprise ferroelectric random access memory (F-RAM) cells, the highest internal "0" being a highest U-term, and the local reference voltage being set to a predefined voltage higher than the sweep reference voltage having a fault in the highest U-term. The method of claim 14.

17. performing the multiple margin sweeps for each block includes identifying a block in which a margin between the memory cell having the lowest program term in the block and the local reference voltage is less than a predefined minimum value; The method further includes the step of isolating the memory cell having the lowest program term in the block and repairing the block by coupling in its place a spare memory cell in the block. The method of claim 14.

Citation Information

Patent Citations

  • Magneto-resistive device and magnetic thin-film memory

    JP2001243760A

  • Semiconductor storage device

    JP2010009687A

  • Reference voltage generation circuit and semiconductor memory device

    JP2010160851A

  • Setting reference voltage for data sensing in ferroelectric memory

    JP2018514891A

  • Hybrid reference generation for ferroelectric random access memory.

    JP2019505945A