JP2007250187A - Nonvolatile semiconductor memory and method of testing same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a nonvolatile semiconductor memory that makes it possible to reduce test time after manufacturing and reduces cost by using an inexpensive test system, and to provide a method of testing the same. <P>SOLUTION: The nonvolatile semiconductor memory includes: a memory cell array where nonvolatile memory cells are arranged in a matrix, and that can perform erasure in units of blocks formed with a plurality of pages; row decoders that are provided per block and that decode row address signals and select rows of the memory cells; nonvolatile storage sections that are provided in these row decoders, respectively; a section that stores, in part of the nonvolatile memory cell, information of a defective block address determined defective in a verifying and reading process in a test mode; a writing section that writes flag data in the nonvolatile storage section provided in the row decoder corresponding to the defective block address stored in the memory cell to make the defective block a nonselective state upon reading. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a nonvolatile semiconductor memory and a test method therefor, and more particularly to a technique for reducing test time and cost of a nonvolatile semiconductor memory.

  Conventionally, hard disks have been generally used as data storage means. However, with the recent increase in capacity of flash memory, nonvolatile semiconductor memories such as flash memory have come to be used as data storage means.

  As the flash memory, a NAND flash EEPROM (Electrically Erasable and Programmable Read Only Memory) and an AND flash EEPROM are generally known. When using flash memory as a data storage means, it is important how to reduce the bit cost and realize a large capacity memory, so even if there are less than the specified number of defective blocks such as hard disks, it will be shipped as a product The form is taken. For this reason, a technique for managing defective blocks is required on the host side using the memory. As one of the technologies for managing this bad block, some data is stored in the bad block at the time of shipment, and the host side using the memory first detects this data and creates a block management table for prohibiting the use of the bad block. The block management method used is widely used. Taking a NAND flash EEPROM as an example, since many of the failure modes are failures in which “1” data changes to “0”, “0” data is written to the defective block and all remaining good blocks are set to “1”. Data is shipped as a form.

  A test flow from the wafer sort process of the flash memory to shipping will be described with reference to FIG. As shown in the figure, after the pre-process for forming the semiconductor memory element on the wafer is completed, a wafer sort process is first performed to determine whether each chip is a good product or a defective product in the wafer state. In this wafer sort process, a DC item check (Step S30) and an operation check test (Function Check) for checking write / erase are performed (Step S32). Whether the check is good or bad in the DC item is determined (step S31), the DC defective chip determined to be defective is discarded (S32-1), and the chip determined to be non-defective is then subjected to an operation check test ( S32-2). The defective block in the chip determined by the good / bad determination (step S33) in the operation check test is replaced with the redundant block in the next R / D replacement step (step S34). After the R / D replacement, the operation confirmation test (step S35) is performed again, and the following chips having the prescribed number of defective blocks are regarded as non-defective products and assembled into packages in the next assembly process (step S36). Thereafter, a burn-in test for screening by accelerating the failure mode at high temperature and high voltage is performed (step S37). After the burn-in, the DC item check (step S38) and the operation confirmation test (step S39) are performed again using the memory tester, and the address information of the defective block is stored in the fail memory of the memory tester. Based on this information, “0” data is written in the defective block of the memory (step S40) and shipped.

  FIG. 13 is a test flow showing details of the operation check test after the burn-in. The content of the operation check test is to write several types of pattern data to all blocks in the memory, read the written data, and compare it with expected value data with a memory tester. As shown in the flow of FIG. 13, a pattern check sequence 1 for writing the first pattern data is first performed (step S50). First, the first pattern data is written in all the blocks of the memory (step S51), the written data is read, and it is checked whether or not the writing is performed accurately (step S52). Next, the block that has not been correctly written is recognized as a defective block, and the block address is stored in the fail memory of the memory tester (step S53). All blocks are erased (step S54).

  This pattern check sequence is performed N times, and if it is not the Nth time (step S55), the pattern check sequence in the next pattern (step S56) is performed.

  The above processing is also performed for the next pattern check sequence i (i is a natural number equal to or less than N). First, the i-th pattern data different from the first to (i-1) pattern data is written in all blocks, the write data is read and checked, and the defective block address is additionally stored in the fail memory of the memory tester. Erase.

  The memory tester stores bad block information on the fail memory each time a write / read check is performed on the N types of pattern data. The bad block information in each pattern data is added to the bad block information on the fail memory stored so far. In this way, the result of the accumulated defective block address of the defective block in the test using the N types of pattern data is stored in the fail memory after the completion of the write operation check with the N types of pattern data (i = N). Yes. After the write operation confirmation test for all the pattern data is completed, “0” data is finally written to the defective block based on the accumulated defective block information (step S57).

  Generally, the operation check test after the burn-in test shortens the test time by simultaneously measuring a plurality of chips.

  FIG. 14 shows a test system for simultaneously measuring 64 NAND type flash EEPROM chips. Each chip is selected by a chip select signal CE and an operation check test is performed. In the NAND flash EEPROM, command data and address data are commonly input to each device from the I / O bus, and a write / erase / read mode is set. At the time of writing, write data for one page (512 bytes) is input to the data latch via the I / O bus, and writing is performed to the memory cell of the page corresponding to the address held in the address register. A plurality of chips are simultaneously performed in the above writing operation. In the next read operation, the memory cell data of the page corresponding to the address input from the I / O bus to the address register is read to the data latch common to the S / A, and externally in synchronization with the read clock signal RE. Output serially. In this case, the read operation is performed for each chip.

  FIG. 15 is a time chart of chip select signals CE1 to CE64 supplied to the respective chips 1 to 64 when the 64 chips are measured simultaneously. When writing certain pattern data to all chips, first, the chip select signals CE1 to CE64 of all chips are simultaneously enabled. The same command and address data are input to all chips. Subsequently, pattern data for one page is input simultaneously for all 64 chips. Further, by inputting a write start command at the same time for all chips, the auto-write operation is executed simultaneously for all chips. The above writing operation takes 226 μsec per page, and in the case of a 128 Mbit NAND flash EEPROM, it takes 7.23 sec. After the auto write operation is completed, a read operation is performed in order to check whether writing has been performed correctly on each chip. This read operation is performed continuously from chip 1 to chip 64, but since the read operation needs to be performed for each chip, the read time per page is only 41.4 μsec. When 64 chips 1 to 64 are read, it takes 172.8 seconds. If a read data error occurs in a certain chip, the defective block address information is stored in the defective block storage area for each chip existing on the fail memory of the memory tester. Thereafter, in order to erase the written pattern, an erase command is input in common to the 64 chips 1 to 64, and all the chips 1 to 64 are simultaneously erased. Since the erase operation takes 1 msec per page, it takes 2 sec to erase all the chips.

  In this way, the auto function of the chip itself can be used for the write / erase operation, so multiple processes can be performed in parallel. However, when checking the write data, the result of the individual check for each chip (Pass / Fail result) Are stored in the bad block storage area of the memory tester, and a plurality of them cannot be measured simultaneously. For this reason, there is a problem that the test time becomes long. Also, since it is necessary to store the Pass / Fail result in the fail memory, it is necessary to use an expensive memory tester having a fail memory.

  As described above, in a conventional nonvolatile semiconductor memory, when an operation check test of a plurality of chips is performed after manufacturing, writing and erasing operations can be performed simultaneously for all chips, but a reading operation is performed for each chip individually. There was a need. Therefore, there is a problem that the test takes time.

  Further, in order to store the address of the defective block determined to be defective as a result of the operation check test, it is necessary to use a memory tester having a fail memory. However, the tester having a fail memory is expensive, and there is a problem that the cost of the operation confirmation test becomes high.

  The present invention has been made in view of the above circumstances, and its purpose is to shorten the test time after manufacture and to reduce the cost by using an inexpensive test system, and to provide a highly reliable nonvolatile semiconductor. To provide a memory and a test method thereof.

  A nonvolatile semiconductor memory according to one embodiment of the present invention includes a memory cell array in which nonvolatile memory cells are arranged in a matrix, read and write are performed in units of pages, and can be erased in units of blocks each including a plurality of pages. Provided for each block of the memory cell array, a row decoder for decoding a row address signal to select a row of memory cells, a non-volatile storage means provided in each of the row decoders, and verify reading in a test mode Means for storing defective block address information determined to be defective in the process in a part of the non-volatile memory cells arranged in the matrix, and in order to deselect the defective block at the time of reading, In the row decoder corresponding to the bad block address stored in part The vignetting nonvolatile storage means and a writing means for writing the flag data.

  In addition, a test method for a nonvolatile semiconductor memory according to an aspect of the present invention is a test method for a nonvolatile semiconductor memory in which writing and reading are performed in units of pages and erasable in units of blocks configured by a collection of a plurality of pages. A first step of inputting an arbitrary data pattern generated by a memory tester to the nonvolatile semiconductor memory as write data; and a step of writing the data pattern to the nonvolatile memory in units of pages. Step 3, the third step of performing a verify read if the data pattern has been normally written in the second step, and if a defect is found as a result of the verify read, the defect has been found Do not pass the address of the block containing the page through the memory tester. A fourth step of writing to any one of the blocks of the nonvolatile semiconductor memory, and a fifth step of performing a data erasing operation for a plurality of the blocks other than the block in which the address is written. It has.

  According to the present invention, a test time after manufacture can be shortened, and a cost can be reduced by using an inexpensive test system, and a highly reliable nonvolatile semiconductor memory and a test method thereof can be provided.

  Embodiments of the present invention will be described below with reference to the drawings. In the description, common parts are denoted by common reference symbols throughout the drawings.

  The nonvolatile semiconductor memory according to the first embodiment of the present invention will be described by taking a NAND flash EEPROM as an example.

  FIG. 1 is a block diagram showing a schematic configuration by extracting a main part of a 128 Mbit NAND flash EEPROM, FIG. 2 is an enlarged view around the memory cell array in FIG. 1, and FIG. 3 is each memory cell in FIG. It is a circuit diagram of a block.

  As shown in FIG. 1, the NAND flash EEPROM according to the present embodiment includes a memory cell array 10, an interface circuit (I / F circuit) 11, a data latch (S / A) 12, an address register 13, a command register 14, and a column. The decoder 15, the row decoder 16, the sequence control circuit 22, the high voltage generation circuit 23, the status register 24 and the like are configured.

  As shown in FIG. 2, the memory cell array 10 is divided into 2048 memory cell blocks BLK0 to BLK2047, and further, a ROM block ROMBLK corresponding to the storage capacity of one memory cell block is provided. In each of the memory cell blocks BLK0 to BLK2047 and the ROM block ROMBLK, NAND cells as shown in FIG. 3 are arranged in a matrix. Each NAND cell is formed by connecting a plurality (16 in this case) of memory cells MC, MC,... In series so that adjacent ones share a source and a drain. The drain on one end side of the NAND cell column is connected to the bit lines (data lines) BL0 to BL4095 via the selection transistors ST1, respectively. The source on the other end side of the NAND cell column is connected to the source line SL via the selection transistor ST2. Select gate lines SGD and SGS extending along the row direction of memory cell array 10 are connected to the gates of select transistors ST1 and ST2 in the same row, respectively. Similarly, word lines WL0 to WL15 extending along the row direction of memory cell array 10 are connected to control gates CG0 to CG15 of memory cells MC, MC,. In the case of a NAND flash EEPROM, one page is constituted by 512 bytes of memory cells MC, MC,... Connected to one word line WL, and the memory cell blocks BLK0 to BLK2047 and the ROM block ROMBLK are composed of 16 pages. One block is configured. That is, since one block is composed of 8 kbytes, one chip of the 128 Mbit NAND flash EEPROM is composed of 2048 blocks. Note that writing to and reading from the memory cell array 10 are performed in units of one page, and erasing is performed in units of blocks.

  Various commands, address signals, cell data to be written, and the like are input to the interface circuit 11, and data read from the memory cell array 10 and latched in the data latch (S / A) 12 is output. It has become so. The row address signal and the column address signal input to the interface circuit 11 are supplied to the address register 13 and latched, and the command is supplied to the command register 14 and latched.

  The column address signal latched in the address register 13 is supplied to the column decoder 15 and decoded. In the data latch (S / A) 12, cell data to be written input to the interface circuit 11 at the time of writing is latched, and from the selected memory cell blocks BLK 0 to BLK 2047 in the memory cell array 10 at the time of reading. The cell data read to each bit line is latched.

  The row address signal (block address signal, page address signal) latched in the address register 13 is supplied to the row decoder 16 and decoded. As shown in FIG. 2, the row decoder 16 has a row main decoder circuit 17 and a row sub-decoder circuit 18 corresponding to the memory cell blocks BLK0 to BLK2047, respectively. Signals from page 0 to page 15 are supplied to all the row sub-decoder circuits 18 through the level shifter circuit 19. The row sub-decoder circuit 18 is a circuit for supplying a predetermined voltage to the 16 word lines WL0 to WL15 in the selected block, and includes a plurality of transfer gate transistors and a booster circuit. The row main decoder circuit 17 receives the signal obtained by predecoding the block address signal by the predecoder circuit 20 and turns on the selection transistor of the selected block. The row sub decoder circuit of the ROM block ROMBLK has the same configuration as the row sub decoder circuit 18 of the memory cell blocks BLK0 to BLK2047, but the row main decoder circuit is replaced with a ROMBLK selector circuit 21. The ROMBLK selector circuit 21 receives a MODE signal for enabling the ROM block ROMBLK. The MODE signal is a signal that is enabled when testing the NAND flash EEPROM chip according to the present embodiment, and is a test signal that becomes “H” level when a predetermined test command is input. For this reason, the ROM block ROMBLK is not selected by an address selection method in normal writing / erasing, and can be accessed only after a predetermined test command is input. In addition, the normal flash memory has a test mode in which all blocks are selected and the write and erase operations are performed in order to shorten the test time, but this ROM block ROMBLK is not selected even when all blocks are selected. Designed to be in state.

  The command supplied to the command register 14 is decoded by a command decoder circuit (not shown) and supplied to the sequence control circuit 22. The sequence control circuit 22 includes a chip enable signal CE, a command latch enable signal CLE, an address latch enable signal ALE, a write enable signal WE, a read enable signal RE, and a write protect signal WP from the outside. Based on these signals, control is performed according to the read operation, write operation, erase operation, verify operation, and the like of each circuit in the memory. Further, based on the output of the sequence control circuit 22, the high voltage generation circuit 23 supplies a high voltage to the row decoder 16 and the memory cell array 10.

  The status register 24 stores information (Pass / Fail flag) indicating whether or not the operation has been normally completed after the auto write or auto erase operation. This information can be output to the outside by inputting predetermined command data. However, the status register 24 only stores the result of the write or erase operation performed immediately before, and this information is reset at the start of the next operation.

  A test flow for testing the NAND flash EEPROM having the above configuration will be described with reference to FIG.

  FIG. 4 shows the contents of the write check operation in one pattern among the multiple data pattern write check operations at the time of the test after burn-in, and the flow is described with particular attention to the write operation of the first page. First, a command code normally used by the user is input, and an address for writing and predetermined 512-byte write data are loaded into the chip (step S10). A checkerboard pattern is commonly used as write data, and such a regular pattern can be generated by a simple tester having an inexpensive pattern generator function. Next, by inputting a predetermined test command and inputting a write start command after data loading, an auto-write operation is executed for the page (step S11). Information indicating whether or not the auto-write operation has been normally performed is stored in the status register (step S12). If the writing is not completed normally and the Fail flag is stored in the status register, that is, if there is a bit failure or a word line voltage failure, the block address data is transferred to the data latch (step S13), and the ROM An operation of storing the defective block address in the block ROMBLK is performed (step S14). Then, automatic writing to the next page is started (step S15).

  FIG. 5 shows a block diagram of the memory space of the ROM block ROMBLK. As described above, the total number of blocks of the 128 Mbit memory chip is 2048 blocks. Therefore, the defective block address data of all the memory blocks BLK0 to BLK2047 are stored by configuring the defective block table using 4 pages out of 16 pages in the ROM block ROMBLK. One page of the ROM block ROMBLK is composed of 512 bytes like the first page of the memory cell blocks BLK0 to BLK2047, and each 1 byte from the first byte to the 512th byte of the first page is from the memory cell block BLK0. It is assigned to each block address of BLK511. That is, block addresses from address 0 to address 511 are assigned to the first group and stored in the first page of the ROM block. Further, addresses 512 to 1023 are assigned to the second group and stored in the second page of the ROM block ROMBLK. Similarly, the addresses from 1024 to 1535 are assigned to the group of address 3, and the addresses from 1536 to the last address of 2047 are assigned to the fourth group. The group at address 3 is stored on page 3 and the group at address 4 is stored on page 4. When this storage operation is completed, all blocks are simultaneously erased. After all the block data except the ROM block ROMBLK are erased, the next pattern data write pattern check is performed.

  A method of storing defective address information in the ROM block ROMBLK will be described in detail with reference to FIG. FIG. 6 is a circuit diagram of the address register and its periphery.

As shown in the figure, the three register circuits of the 9-bit column address register 30, the 4-bit page address register 31, and the 11-bit block address register 32 constituting the address register 13 operate as a binary counter. It is configured as follows. When operating as a binary counter, the signal CLK is normally supplied from the outside via the multiplexer 33 to the clock input of the column address register 30. This signal CLK is formed in synchronization with an externally input signal RE during serial reading, and in synchronization with an externally input signal WE during data loading for writing. In the test mode, the signal TCLK is supplied to the column address register 30 via the multiplexer 33. The TCLK signal is used when block address information is transferred to the data latch as will be described later, and 512 clock signals (not shown) are automatically generated in the test mode in the internal control circuit. The output signal of the column address end detection circuit 34 is supplied to the clock input of the page address register 31. This is because the NAND flash memory is designed so that the page address is automatically incremented after the serial read operation of one page and the random read of the next page is continuously executed. For the same reason, the output signal of the page address end detection circuit 35 is input to the clock input of the block address register 32. The binary counters of the column address register 30, the page address register 31, and the block address register 32 are supplied with a CLR signal from the internal control circuit as a counter reset signal, and each register is reset at the beginning of address input. Is done. In the address data input mode, the internal data of these three registers is initialized to data supplied from the outside of the chip via the interface circuit 11. Address data input from the outside is temporarily stored in the 8-bit input data latch circuit 37 via the input buffer 36, and then the data is supplied to the internal bus by the internal bus control circuit 38. Ordinary address data is input from the outside in several steps of 8 bits, and the total of 9 bits of 8-bit data of the first column address information and 1-bit command flag data indicating the upper / lower of 512 bytes.
The bits are sent to the 9-bit column address counter 30 by the internal bus control circuit 38 and stored as initial data. Of the 8 bits input for the second time, 4 bits are sent to the page address counter 31 and the remaining 4 bits are sent to the block address counter 32 and stored therein. All the 8-bit input addresses after the third time are sent to the block address counter 32 and stored as initial data. The output of the 11-bit block address counter 32 is divided into buses for the lower 9 bits from addresses 0 to 8 and the upper 2 bits from addresses 9 to 10 for output. Internal address data of a total of 11 bits, the lower 9 bits and the upper 2 bits, is supplied to the row decoder 16 via the predecoder circuit 20. The lower 9-bit data is output to the internal bus during the operation of storing the defective block address information in the ROM block ROMBLK, and is stored as initial data in the 9-bit counter 39 by the internal bus control circuit. However, the 9-bit output data of the block address counter circuit 32 output at this time is inverted and stored. The 9-bit data of the counter circuit 39 is input to the NOR logic circuit 40, and the output signal of the NOR logic circuit 40 is input to the AND logic circuit 41 with the MODE signal.

  In the test mode in which defective block address information is stored in the ROM block ROMBLK, TCLK is supplied from the internal control circuit and the column address is counted up from address 0, and at the same time, the 9-bit counter 39 is counted up. Since the initial value of the 9-bit counter 39 is a complementary value of the defective block address to be written to the ROM block ROMBLK, when the X address is defective (X + 1), all the block address signals are “ 0 ”. At this time, the output signal WDATASET of the AND logic circuit 41 becomes “1”, and data 0 to 7 for loading data into the data latch (S / A) are set to the “L” level (“0” write data). The Since all the 9-bit outputs become “0” only once in 512 clocks, the remaining 511 load data is equal to the data on the internal bus. In a data load operation during normal writing other than the test mode, a signal from the chip I / O bus is supplied to the data line via the input buffer 36 and the 8-bit input data latch circuit 37, and the data latch (S / A) is loaded with data. However, the internal bus is fixed at “H” level during the data load operation in the test mode. For this reason, the data latch (S / A) at the column address equal to the defective block address is loaded with 1 byte of “0” write data, and the data latches (S / A) at the other 511 column addresses are loaded. “1” write data is loaded.

  The above operation will be specifically described with reference to FIGS. As described above, there are memory cell blocks from 0 to 2047, but when 0 to 2047 are represented by binary numbers, the result is as shown in FIG. Looking at the upper 2 bits, 0 to 511 is “00”, 512 to 1023 is “01”, 1024 to 1535 is “10”, and 1536 to 2047 is “11”. That is, the page to be written to the ROM block ROMBLK can be specified with the upper 2 bits, and the column address can be specified with the lower 9 bits. For example, it is assumed that the address 2, that is, the memory cell block BLK2 is defective. At this time, the data of the block address counter 32 is “01000000000000”. Of these, the lower 9 bits are output to the 9-bit counter 39 with its data inverted. That is, as shown in FIG. 7B, the data of the 9-bit counter 39 is “101111111”. Therefore, by counting up 2 addresses + 1 times = 3 times, which is a defective block, all the data of the 9-bit counter become “0” and WDATASET = “1”. As a result, “0” data is written in the ROM block ROMBLK corresponding to the second address.

  A method for controlling which page of the four pages in the ROM block is to be written, after the block address is transferred to the data latch (S / A) by 512 clocks and the automatic write operation to the memory cell starts. Will be described. As described above, since the page address can be designated by the upper 2 bits of the 11-bit data of the block address counter 32, the upper 2 bits of the block address register 32 are input to the ROM block page decoder circuit 42. The ROM block page decoder circuit 42 operates as a decoder circuit for the upper block address signal in response to the MODE signal in the test mode, and sets all decoder output signals at other times to the “0” level. The ROM block page decoder circuit 42 forms pages 0 to 3 which are 4-bit page signals supplied to the row sub-decoder through the multiplexer circuit 43 in the test mode. Except in the test mode, a normal page decoder circuit 44 that decodes a page address forms a page signal for these four pages via a multiplexer circuit 45. Further, during the test, the signals of the upper 4 to 15 pages are set to the “0” level non-selected state by the multiplexer circuit 45. During normal operation, the page decoder circuit 44 forms these upper page signals via the multiplexer circuit 77.

  Each time such a defective page is found, this defective block address information is stored in the 8-bit memory cell of the corresponding column address in the corresponding page of the ROM block ROMBLK. In a normal NAND flash EEPROM, writing “0” data corresponds to changing the threshold voltage of the memory cell from negative to positive, and writing “1” data corresponds to not changing the negative threshold voltage of the memory cell as it is. For this reason, if 512 byte data is overwritten on the same page in the ROM block ROMBLK many times, "0" data is accumulated and stored on the "1" data, and all bad block information remains at the end. It will be. That is, OR data of 512 byte data corresponding to the number of times of writing is stored. The present invention takes advantage of this feature of flash EEPROM write operations. As shown in FIG. 4, when this write pattern check is completed for all pages of one chip, all blocks other than the ROM block are erased in response to an erase command input from the outside. Then, a write check is performed again on all pages using the next pattern, and the block address is stored in the ROM block each time a defective page is found. In addition, by performing the operation of erasing each block in the middle without using the all block erasing mode for all erasing operations, it is possible to store the address that causes the erasure failure in the ROM block ROMBLK based on the erasure Pass / Fail information. . When the check for all patterns is completed in this way, the ROM block ROMBLK stores all the write patterns and the accumulated defective block addresses in the block erase operation.

  Such defective block information is stored in the ROM block ROMBLK and shipped, and the controller can refer to this information to construct a block management table.

  As described above, according to the nonvolatile semiconductor memory of this embodiment, the ROM block which is a special redundant block that cannot be written or erased by normal address input is provided. The defective block address information is stored in the ROM block. This eliminates the need for a read operation for verifying the result of the write / erase operation when simultaneously testing a plurality of nonvolatile semiconductor memories. As a result, the test time can be shortened and the test can be performed with an inexpensive test system having no fail memory, so that the test cost of the nonvolatile semiconductor memory can be reduced.

  Next, a nonvolatile semiconductor memory according to a second embodiment of the present invention will be described by taking a NAND flash EEPROM as an example.

  The circuit configuration of the NAND flash EEPROM according to the present embodiment is the same as that shown in FIGS. 1 to 3 described in the first embodiment, and a description thereof will be omitted.

  FIG. 8 is a circuit diagram of the address counter of the NAND flash EEPROM according to the present embodiment and its periphery.

  In the present embodiment, unlike the first embodiment, in the test mode in which the defective block address information is stored in the ROM block ROMBLK, the lower 9 bits of the block address are not stored in the 9-bit test dedicated counter 39, The data is directly stored in the 9-bit column address counter 30. That is, the sequence control circuit 22 transfers the lower 9-bit data of the block address counter 32 to the column address register 30 using the internal bus. Then, 1-byte “0” data write is loaded only into the data latch (S / A) indicated by the column address. Prior to this loading operation, all the data latches (S / A) are simultaneously reset to “1” data, so that the page of the corresponding ROM block ROMBLK contains “0” data only for the column address corresponding to the defective block address. Is memorized. Normally, the NAND flash EEPROM sets all data latches to “1” data before the start of the write operation. Therefore, if this function is used, it is not necessary to load data to all addresses, and bad block address information is stored in the ROM block. Test mode time can be shortened.

  According to the above embodiment, the test time of the nonvolatile semiconductor memory can be further shortened as compared with the first embodiment.

  Next, a nonvolatile semiconductor memory according to a third embodiment of the present invention will be described by taking a NAND flash EEPROM as an example.

  Since the configuration of the NAND flash EEPROM according to the present embodiment is the same as that of the first embodiment, description thereof is omitted.

  In the first and second embodiments, the defective block address is stored in the ROM block ROMBLK, and the memory tester is used to actually write “0” data in the defective block at the end of the test process. . In the present embodiment, in the above circuit configuration, the sequence control circuit 22 performs the operation of the flowchart of FIG. 9 and has a function of automatically writing “0” data to the defective block without using a memory tester in the final test process. It is

  First, all memory cell blocks BLK0 to BLK2047 are checked, and defective block data is stored in the ROM block ROMBLK.

Then, as shown in FIG. 9, the data at the address of the first page of the ROM block ROMBLK is read (steps S20 and S21) and stored in the data latch (S / A). Next, the sequence control circuit 22 generates a clock for the column address counter 30 shown in FIG. 6, and sequentially increments the output address of the column address counter 30 from address 0 (step S22). Further, the determination circuit provided in the sequence control circuit 22 checks whether or not all the data stored in the data latch corresponding to each incrementing column address is “0” data (step S23). If the data latch data at a certain address is “0” data, the contents of the column address counter 30 at that time are transferred to the lower 9 bits of the block address 11-bit counter 32 via the internal bus. Further, the lower 2-bit data of the 4-bit page address counter 31 indicating the first page of the ROM block ROMBLK is transferred to the upper 2 bits of the block address 11-bit counter 30 via the internal bus (step S24). . When the address setting operation for such a defective block is completed, the defective block in the row decoder indicated by this address is selected. Thereafter, all the outputs of the page address counter 31 are set to the “1” level, and all the page addresses of the block are multi-selected (step S25). Next, the data of all the data latches are reset to “0” data (step S26), and the write operation of 1 msec longer than the normal write time of 20 μsec is performed, so that all the memory cells on the 16 pages of the defective block have “ 0 "data is written (step S27). The reason why the writing time is set longer than the normal writing time is that “0” data is correctly written even when the memory cell block is a defective block due to a defective mode in which the word line voltage drops. This is because it is necessary to consider. Instead of setting the time longer, a write voltage higher than the normal write voltage may be used. After the writing of the defective block is completed, the first page of the ROM block ROMBLK is read again to the data latch, and the column data scan is restarted from the column address next to the column address remaining in the column address counter 30. And another column address “
When the “0” data hits, a sequence of writing “0” data to all pages of the defective memory cell block by the same method is executed. In this way, the column data scanning operation is performed up to the final column address. (Step S28) Then, the current number of pages is determined (Step S29), and the process proceeds to the second page (Step S29 '), and then the second page in the ROM block is read and the same operation is repeated. When the fourth page in the ROM block ROMBLK is completed (step S29), the test sequence for automatically writing “0” data in this defective block is completed. "Data will be written. By using this test mode, the host side In addition to a system that accesses a memory block to form a block management table, the system also scans all memory cell blocks BLK0 to BLK2047 at the time of shipment and determines a block in which “0” data is detected as defective. Applicable.

  According to the nonvolatile semiconductor memory as described above, not only the defective block address data is stored in the ROM block, but also “0” data is automatically stored in the memory cells of all pages of the defective block without using the memory tester. Can write. Therefore, since the test of the nonvolatile semiconductor memory can be simplified, the cost of the nonvolatile semiconductor memory can be further reduced.

  Next, a nonvolatile semiconductor memory according to a fourth embodiment of the invention will be described by taking a NAND flash EEPROM as an example.

  As described in the first to third embodiments, “0” data is written in the memory cell MC in the defective block of the normal NAND flash EEPROM before shipment. Further, when the degree of destruction of the memory cell MC is severe and “0” data cannot be written, the memory cell MC is discarded. However, the NAND flash EEPROM has a characteristic that only “0” data can be read from the NAND cell by always turning off the selection transistor.

  In the present embodiment, in view of the above characteristics of the NAND flash EEPROM, the row decoder is set so that the defective block selection transistor is always turned off instead of writing “0” data in the memory cell of the defective block. Is.

  FIG. 10 is a circuit diagram showing a configuration of 2048 row decoders corresponding to the memory cell blocks BLK0 to BLK2047, respectively.

  As shown in the figure, the output of the decoding unit 50 to which the block address signal is input is input to the gate of the N-channel transistor 51. The source of the N-channel transistor 51 is input to the drain of an N-channel transistor 52 whose gate is supplied with a signal C, and the source of the N-channel transistor 52 is connected to the power supply Vss. The drain of the N-channel transistor 51 is connected to one end of an electrical fuse 53 formed of a narrow polysilicon filament. When a current exceeding a certain value flows through the electrical fuse 53, the electrical fuse 53 is melted and is not electrically connected. Further, the other end of the electrical fuse 53 is connected to the source of the latch circuit 54 and the N-channel transistor 55 whose gate is supplied with the signal B. The drain of the N-channel transistor 55 is connected to a terminal corresponding to each of the memory cell blocks BLK0 to BLK2047 and to the drain of the P-channel transistor 56 to which the signal A is supplied to the gate in common. The source of this P-channel transistor 56 is connected to the power supply VDD. The output of the latch circuit 54 is connected to the drain of an N-channel transistor 57 whose signal D is input to the gate, and the source of the N-channel transistor 57 is connected to the power supply Vss. The output of the latch circuit 54 is input to a level shifter 58 to which Vpgm is supplied as a power source. The output of the level shifter 58 is connected to the gates of all transfer gate transistors in the row sub-decoder circuit 59. When the transfer gate transistor in the selected memory cell block is turned on, the global select gate signals GSGD and GSGS and the page signals from page 0 to page 15 are selected in the select gate lines SGD and SGS of the memory cell block and the memory. It is supplied to the word lines WL0 to WL15 of the cell. An inverted signal of the output of the latch circuit 54 is input to the gate of the N channel transistor 61 in the row sub decoder 59 via the inverter 60. The drain of this N channel transistor 61 is connected to the select gate line, and its source is connected to the SE line in common for each block.

  Next, the operation of the row decoder configured as described above will be described. During normal reading, writing, and erasing operations, the P-channel transistor 56 and the N-channel transistor 55 are nonconductive. During normal read, write, and erase operations, first, the signal D becomes “1” level, and the data in the latch circuits 54 of all blocks is reset to “0”. Next, when the block address signal is determined, the signal C becomes “1” level and the N-channel transistor 52 becomes conductive. In addition, since the output node of the decoding unit 50 of the selected block becomes “1” level and the N-channel transistor 51 also becomes conductive, when the electrical fuse 53 is not blown, the latch circuit 54 has “1” level. Remembered. The output of the latch circuit 54 is supplied to the level shifter 58, and a voltage higher than the power supply VDD by a predetermined level is supplied to the transfer gate transistor at the time of reading. Further, the output signal of the inverter 60 becomes “0” level, and the N-channel transistor 61 becomes non-conductive. As a result, a predetermined read voltage supplied from GSGD, GSGS, and CG0-15 is supplied to the select gate line and the 16 word lines. In the non-selected block, the output of the latch circuit remains at “0” level, so that the transfer gate transistor becomes non-conductive and the N-channel transistor 61 becomes conductive. At the time of reading, since the SE line is at the power supply Vss level, the select gate line of the non-selected block is at the Vss level, and the select gate transistor of the non-selected block is turned off.

  At the time of writing, in the selected block, a voltage higher than the write voltage Vpgm by the threshold voltage of the N-channel transistor is supplied to the transfer gate transistor, and the select gate line and 16 word lines are supplied by GSGD, GSGS, and CG0-15. A predetermined write voltage is supplied. In the non-selected block, the N channel transistor 61 is in the conductive state as in the read operation, and the SE line is at the Vss level, so the select gate line is at the Vss level and the select gate transistor is in the nonconductive state.

  Further, at the time of erasing, the voltage of the power supply VDD is supplied to the transfer gate transistor, and the 16 word lines are at the power supply Vss level. At the time of erasing, the power supply VDD level is supplied to GSGD and GSGS. In the selected block, the gate of the N-channel transistor 61 is set to the “1” level, but the SE line is set to the VDD level during the erase operation, so that the N-channel transistor 61 is turned off. For this reason, the drain-side select gate line SGD and the source-side select gate line SGS are charged to a voltage lower than the VDD by the threshold voltage of the N-channel transistor and then enter a floating state. Thereafter, the substrate potential of the memory cell rises to the erase voltage, but at the same time, the select gate line SGD is also raised to almost the same potential by coupling, so that no electric field stress is applied to the oxide film of the select gate transistor 61. The control gates of the memory cells connected to the 16 word lines are at the Vss level, and all the memory cells in the block are erased. In the non-selected block, all the transfer gate transistors are turned off, and the N-channel transistor 61 is also turned off, so that the 16 word lines and the select gate line are in a floating state. As a result, the word lines are also coupled to the substrate. It is lifted by the ring and the memory cell is not erased.

  In order to blow the electrical fuse 53, the signal A is set to the “0” level, and the signals B and C are set to the “1” level. At this time, it is desirable to reduce the conduction resistance of the transistor 55 by supplying a boosted voltage higher than the power supply voltage to the signal B. By inputting the address signal of the block to be blown in this state to the row decoder, a predetermined current flows through the electrical fuse 53 of the selected block, so that the fuse can be blown.

  When the block in which the electrical fuse 53 is blown is selected, the same operation as that of the non-selected block is performed for the read operation, the write operation, and the erase operation. That is, even if the output of the decoding unit 50 becomes “1” level and the N-channel transistor 51 becomes conductive, the output of the latch circuit 54 is “non-selected” because the electrical fuse 53 is electrically nonconductive. 0 ″ remains stored, and the power supply Vss is supplied to the transfer gate. Further, the N-channel transistor 61 becomes conductive. Therefore, even if a block in which the electrical fuse 53 is cut at the time of reading is selected, the select gate line becomes the Vss level, and no current flows from the bit line through the memory cell. As a result, only “0” data is read from the defective block. When a block in which the electrical fuse is cut at the time of writing and erasing is selected, the electric fields for writing and erasing are not applied to the memory cell as in the non-selected block.

  As described above, in the nonvolatile semiconductor memory according to the present embodiment, the defective block information is stored in the electrical fuse 53, and the selection transistor is always in a non-conductive state no matter what defect exists in the memory cell. Can read only “0” data. As a result, the reliability of bad block information can be improved. Although the electrical fuse is used in this embodiment, the same effect can be obtained by arranging the flash memory cell in the row decoder and inserting the current path of the flash memory cell instead of the electrical fuse 53. . For example, normally, the flash memory cell is kept in the erased state where the threshold voltage is 0 V or less. In the test mode in which defective block information is stored in the row decoder, if the selected block is a defective block, the threshold voltage is set to 0 V or higher by supplying the Vpgm voltage to the gate of the flash memory cell in the row decoder. Change to writing state. By providing the Vss level to the gate of the flash memory cell at the time of accessing the row decoder in the read, write, and erase operations, the same effect as the above-described electrical fuse can be obtained.

  Next, a nonvolatile semiconductor memory according to a fifth embodiment of the invention will be described by taking a NAND flash EEPROM as an example.

  In this embodiment, instead of providing a nonvolatile memory element in the row decoder as described in the fourth embodiment, a volatile memory element is provided, and this volatile memory element is based on a power-on detection signal after power-on. Is stored with bad block information.

  FIG. 11 shows a row decoder circuit, which is different from the circuit of FIG. 10 described in the fourth embodiment in that the electrical fuse 53 and its fuse cutting transistors 56 and 55 are deleted, and the output of the decoding unit 50 is shown. The inverted signal of the output of the latch circuit 54 is input to the NAND circuit 62, and the inverted signal of the output of the NAND circuit 62 is input to the level shifter 58. A method for storing defective block information in the row decoder having the above-described configuration will be described.

  Based on the power-on detection signal after the power is turned on, an internal control circuit (not shown) reads the defective block information stored in the ROM block ROMBLK into the data latch. Thereafter, in order to store the defective block flag in the volatile memory element in the row decoder, the signal D is enabled and the output of the latch circuit 54 as the volatile memory element of all the row decoders is reset to “0” level. Disable D. In this state, a column data scan operation is performed based on the sequence shown in FIG. 9, and when defective block information is detected in the data latch 54, the data latch information is transferred to the block address register 32. Thereafter, the signal C is enabled to set the N-channel transistor 52 to the conductive state, and the output of the latch circuit 54 of the defective block indicated by the block address register is changed from “0” level to “1” level. The bad block flag information in the latch circuit 54 is stored until the power is turned off. When this operation ends, the column data scan operation is continued according to the sequence of FIG. When flag setting of all defective blocks is completed, the flash memory of the present invention can be accessed from the outside. Thus, by setting the output of the latch circuit 54 in the defective block to the “1” level, the selection transistor is always in the non-selected state.

  With the above configuration, the select gate line can be set to Vss at the time of reading in a defective block by using a power-on detection signal without using a nonvolatile memory element. Further, the memory cell is accessed after power-on, usually after 100 msec to 1 sec, and the above operation can be completed within this time.

  By using the embodiment of the present invention, a test process for storing defective block information at the time of shipment can be simplified, and an inexpensive test system can be used, so that an inexpensive flash memory can be realized.

  As in the first to fifth embodiments, a ROM block which is a special redundant block that cannot be written or erased by normal address input is provided, and defective block address information is stored in the ROM block. . Therefore, when testing a plurality of nonvolatile semiconductor memories at the same time, the read operation can be performed simultaneously on all the chips as well as the write / erase operation.

  Further, the test of the nonvolatile semiconductor memory can be simplified by providing the function of automatically writing “0” data to all pages in the defective block after checking the defective block.

  Further, the reliability of the defective block information can be further improved by setting the row decoder so that the selection transistor of the defective block is always turned off instead of writing “0” data to the memory cells in the defective block. I can do it.

  As a result, the test time can be shortened, and the test can be performed with an inexpensive test system having no fail memory. Therefore, the test cost can be reduced and a highly reliable nonvolatile semiconductor memory can be realized.

  The first to third embodiments have been described by taking the NAND flash EEPROM as an example. Needless to say, the present invention can also be applied to other nonvolatile semiconductor memories such as a NOR flash EEPROM. Changes can be made as appropriate without departing from the spirit of the invention.

1 is a schematic configuration diagram of a NAND flash EEPROM according to a first embodiment of the present invention. 1 is an enlarged view around a memory cell array of a NAND flash EEPROM according to a first embodiment of the present invention. 1 is a circuit diagram of each memory cell block of a NAND flash EEPROM according to a first embodiment of the present invention. 2 is a flowchart showing a part of a test flow of the NAND flash EEPROM according to the first embodiment of the present invention. 1 is a block diagram showing a memory space of a ROM block in a NAND flash EEPROM according to a first embodiment of the present invention. 1 is a circuit diagram of an address register and its periphery of a NAND flash EEPROM according to a first embodiment of the present invention. 2A and 2B are diagrams for explaining a write operation of the NAND flash EEPROM according to the first embodiment of the present invention, in which FIG. 1A is a binary display from 0 to 2047, and FIG. 2B is a change of bit data during writing. FIG. FIG. 6 is a circuit diagram of an address register and its periphery in a NAND flash EEPROM according to a second embodiment of the present invention. 9 is a flowchart showing a part of a test flow of a NAND flash EEPROM according to a third embodiment of the present invention. FIG. 10 is a circuit diagram of a row decoder in a NAND flash EEPROM according to a fourth embodiment of the present invention. FIG. 10 is a circuit diagram of a row decoder of a NAND flash EEPROM according to a fifth embodiment of the present invention. 10 is a flowchart showing a test flow of a conventional semiconductor memory. The flowchart of the operation confirmation test of the conventional semiconductor memory. The figure which shows the conventional test system of semiconductor memory. The time chart of the chip enable signal in the conventional semiconductor memory test system.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Memory cell array, 11 ... Interface circuit, 12 ... Data latch, 13 ... Address register, 14 ... Command register, 15 ... Column decoder, 16 ... Row decoder, 17 ... Row main decoder circuit, 18, 59 ... Row sub decoder circuit , 19 ... level shifter circuit, 20 ... predecoder circuit, 21 ... ROM block selector circuit, 22 ... sequence control circuit, 23 ... high voltage generation circuit, 24 ... status register, 30 ... column address register, 31 ... page address register 32 ... Register for block address, 33, 43, 45 ... Multiplexer, 34 ... Column address end detection circuit, 35 ... Page address end detection circuit, 36 ... Input buffer, 37 ... Input data latch, 38 ... Internal bus control circuit, 39 ... Bit counter, 40 ... NOR logic circuit, 41 ... AND logic circuit, 42 ... ROM block page decoder circuit, 44 ... page decoder circuit, 50 ... decoding unit, 51, 52, 55, 57, 61 ... N channel transistor, 53 ... Electrical fuse, 54 ... Latch circuit, 56 ... P channel transistor, 58 ... Level shifter, 60 ... Inverter, 62 ... NAND logic circuit

Claims (6)

A memory cell array in which nonvolatile memory cells are arranged in a matrix, read and write are performed in units of pages, and can be erased in units of blocks composed of a plurality of pages;
A row decoder provided for each block of the memory cell array, for selecting a row of memory cells by decoding a row address signal;
Non-volatile storage means provided in each of these row decoders;
Means for storing defective block address information determined to be defective in the verify read process in the test mode in a part of the nonvolatile memory cells arranged in the matrix;
A writing means for writing flag data into a non-volatile storage means provided in a row decoder corresponding to the defective block address stored in a part of the memory cell in order to make a defective block unselected at the time of reading; A non-volatile semiconductor memory comprising:   2. The nonvolatile semiconductor memory according to claim 1, wherein the nonvolatile storage means includes an electric fuse element.   2. The nonvolatile semiconductor memory according to claim 1, wherein the nonvolatile storage means includes a nonvolatile memory cell. A method for testing a nonvolatile semiconductor memory in which writing and reading are performed in units of pages and erasable in units of blocks configured by a collection of a plurality of pages,
A first step of inputting an arbitrary data pattern generated by a memory tester as write data to the nonvolatile semiconductor memory;
A second step of writing the data pattern to the nonvolatile memory in units of pages;
A third step of performing a verify read to determine whether the data pattern has been normally written in the second step;
If a defect is found as a result of the verify read, the address of the block including the page in which the defect is found is transferred to one of the blocks of the nonvolatile semiconductor memory without going through the memory tester. A fourth step of writing;
And a fifth step of performing a data erasing operation for a plurality of the blocks other than the block in which the address is written. A test method for a nonvolatile semiconductor memory, comprising: A test method for testing a non-volatile semiconductor memory in which writing and reading are performed in units of pages and erasable in units of blocks constituted by a collection of a plurality of pages,
A first step in which an arbitrary data pattern is generated by a tester, and the data pattern is input to the nonvolatile semiconductor memory;
A second step in which a command is generated by the tester to write the data pattern into the memory cell to the nonvolatile semiconductor memory;
A third step of determining whether the writing of the data pattern has succeeded or failed by reading the data pattern written in the memory cell by the tester and comparing it with an expected value; If it is determined in step 3 that the failure has occurred, the address of the block including the page that has failed to be written is not stored in the fail memory of the tester, but is stored in any of the blocks of the nonvolatile semiconductor memory. A test method for a non-volatile semiconductor memory, characterized by being written. The nonvolatile semiconductor according to claim 5, further comprising a fourth step of writing fixed data to the memory cell in the defective block based on the address written in any one of the blocks. How to test memory.

Images