The present application is related to U.S. Pat. No. 7,353,438 B2, entitled “Transparent Error Correcting Memory”; and U.S. Pat. No. 7,051,264 B2, entitled “Error Correcting Memory and Method of Operating Same”. This application is also related to U.S. patent application Ser. No. 12/378,248 entitled “Data Restoration Method for a Non-Volatile Memory” by Jeong Y. Choi. The disclosure of the aforementioned patent application and patents are incorporated by reference herein.
The present invention relates to non-volatile memories (NVM). More particularly, this invention relates to non-volatile memory cells fabricated using an application specific integrated circuit (ASIC) or conventional logic process. This invention further relates to a method of operating a non-volatile memory to ensure maximum data retention time. The invention also relates to a system-on-chip (SoC) architecture which includes both non-volatile memory and dynamic random access memory (DRAM) that operates like static random access memory (SRAM), wherein the SoC is fabricated using a conventional logic process.
A nonvolatile memory cell has two gates separated by an insulator layer. The floating gate electrode consists of polysilicon and is vertically separated from a control gate by the insulator layer. The control gate functions to provide external current control to the nonvolatile memory. Examples of nonvolatile memory as used herein include, but are not limited to, flash memory, EEPROM, CAM, and CD-ROM. Whenever the term flash memory is used herein, it is understood not to limit the invention to strictly flash memories, but to include all of the aforementioned nonvolatile memories.
A flash memory stores data in the form of charge on a floating gate that is insulated by two insulator layers. To store data in a floating gate, the memory cell is either programmed or erased. Conventionally, a logical “0” is stored by a programmed cell and a logical “1” is stored by an erased memory cell. Over time, the logical values in a memory array can be inadvertently changed due to charge leakage or disturbs resulting from accesses to adjacent rows.
Programming in a nonvolatile memory is performed by storing electrons on the floating gate to obtain a logical state of “0”. In a “0” state the bit draws current of approximately greater than 10 microAmps when the read voltage is applied to the memory cell. Erasing of data is performed by removing electrons from the floating gate to generate a logical state of “1”. In a “1” state, the bit draws a small current of less than 1 microAmp when the read voltage is applied to the memory cell.
In a flash memory data, a byte or word can be read in a mere 20 to 100 ns period. On the other hand, programming a byte or word requires a few microseconds, while erasing can consume up to several seconds of time.
For system-on-chip (SoC) applications, it is desirable to integrate many functional blocks into a single integrated circuit. The most commonly used blocks include a microprocessor or micro-controller, memory blocks, and various functional logic blocks, all of which are fabricated on the same chip. The memory blocks can include any combination of volatile random access memory (SRAM), non-volatile memory and/or register-based memory. The register-based memory is typically used where a small amount of high speed storage is required (e.g., for register files and/or small tables to be used by one or more functional logic blocks within the SoC).
If a large non-volatile memory block is also required, it would be advantageous to be able to use the same process for fabricating the functional logic blocks, 1T-SRAM system and the non-volatile memory block to realize a low cost SoC. It would further be desirable if all of these elements could be fabricated using a conventional logic process. However, traditional processes for fabricating non-volatile memory, which typically use stacked gate or split-gate memory cells, are not compatible with a conventional logic process.
As used herein, a conventional logic process is defined as a semiconductor process that uses a single layer of polysilicon (i.e., a single conductive gate layer), without the addition of material in the gate-oxide. It would therefore be desirable to implement a single-polysilicon non-volatile memory cell using a conventional logic process, where the possibility of over-erase conditions are minimized, and the detection of over-erase conditions is simplified. It would also be advantageous to be able to program and read the non-volatile memory cell while minimizing disturbances from write, erase and read operations, thereby improving the life and reliability of an array fabricated using the non-volatile memory cell.
Because data stored in a nonvolatile memory can lose charge or gain charge over time, it is expected that single bit errors will arise. To address the possibility of a defective bit, periodic refreshes are performed on flash memory arrays. Consequently, refreshing memory cells in a nonvolatile memory is sometimes performed unnecessarily to ensure that data integrity is preserved. It is therefore further desirable to implement a more precise method of refreshing memory cells.
The present invention improves data retention for an entire flash memory array. The present invention minimizes refresh time by distinguishing between the type of failures that produce an erroneous read out of data from a page of memory.
In accordance with the present invention a method is disclosed for selectively refreshing a memory embedded in a semiconductor integrated circuit. The method involves reading one or more words sequentially out of a page of memory; and then checking the integrity of each word with an error correction code (ECC) to determine the type of error failure that may be present. If the ECC generates a signal indicating that all of the words on a page are error-free then the page will not be refreshed.
Yet another embodiment of the invention provides a method for detecting and selectively refreshing a memory array having n number of data words in a row, each row being organized in a plurality of pages. In accordance with this method, each data word is sequentially read out of a first page and into an external buffer while the integrity of each data word within the first page is checked with ECC. This method involves generating a signal to indicate whether any data word on the first page contains an error; checking each data word on a next page in the memory array to determine if an error exists, and providing a signal if any data word on said next page contains an error. The advantage of this invention is that a refresh operation is skipped for each page of the memory array that is found to be error-free. If a refresh is considered necessary, it is optimized based on the type of error detected. Each subsequent page is read out and checked for errors until all pages of the memory array have been tested.
The various embodiments of the invention will be more fully understood in view of the following description and drawings.
It would be desirable to construct a memory device in which the impact of refresh operations on external memory access is minimized. Accordingly, the present invention provides a memory system in which the majority of refresh operations are carried out within the memory device during idle memory cycles. Idle memory cycles are clock cycles in which there is not an on-going or pending access to the memory device.
The present invention improves data retention for an entire flash memory array. The present invention also improves endurance of digital data over the conventional nonvolatile memory by reducing the amount of time necessary to refresh the memory cells of an array.
The various elements shown in
Once a specific page of data is requested, the page is output to ECC circuit 135 for evaluation, and if necessary, correction. ECC circuit 135 transfers corrected data to storage buffer 140. ECC Circuit 135 includes both error correction code and an ECC generator. The ECC generator produces an ECC syndrome in response to each data word that is read into the ECC circuit. The syndrome in turn is used to produce a corrected word that will be subsequently moved to the storage buffer 140. An additional ECC generator (not shown) may be located as a separate input to storage buffer 140 for purposes of receiving external data. Such external data may be input to storage buffer 140 via a separate data input (not shown) that provides test patterns for performing various types of tests on the data of memory 110. The Storage buffer of
One of the purposes of the present invention is to identify what type of failure caused a specific error within a word of data. In particular, this invention determines whether or not an error constituted a program failure or an erase failure. Accordingly,
The present invention can determine if data bits have flipped polarity from either a logical “0” to a logical “1”, or vice versa. Since each data word is concatenated with ECC check bits, the check bits are also being tested while the data bits are undergoing testing. The error correction code is preferably embedded on the same chip as memory 110 as shown in
The manner in which a word containing a single bit error is refreshed will now be described in reference to
The present invention accordingly takes into account the syndrome data to correct errors that are found before such errors propagate further. By following this approach, the present invention can reduce the number of single-bit errors located in memory array 110. Subsequently, the storage buffer 140 will read out the first page in corrected format into page buffer 130. Then the corrected data from page buffer 130 in turn is written into memory array 110 to the address specified by refresh controller 150. System 100 will normally then be instructed to test the second page of the same row of memory 110.
Testing of the second page proceeds when the refresh controller provides an address to memory 110 for data words [31:16]. Array 110 thus outputs the second page containing words [31:16] into ECC circuit 135 and each word is individually-tested for errors. If no errors are detected in word 16, then, word 16 will simply be loaded into storage buffer 140. Afterwards, the next word on the second page, word 17, is tested to determine if any errors are present in the data bits or check bits. If a single error is detected in word 17, it will be corrected once the entire page has been loaded into page buffer 130, in accordance with the failure type in the manner set forth in
A syndrome that characterizes the status of the bits in word 18 is generated by the ECC generator contained within ECC circuit 135, as in the case with words 16 and 17 above. If the syndrome determines that a single bit error is attributable to an erase failure in word 18, then an ERASE ER flag will be activated by the syndrome. In addition, ECC circuit 135 notifies the refresh controller 150 of an erase failure in word 18 by transmitting an ERASE error signal, EED#, to the refresh controller. Then after the entire page is loaded back into memory 110, the page is either reprogrammed or erased and then reprogrammed. That is, in the case of a program failure, reprogramming of the page of memory occurs within array 110 without any intervening erase operation being performed. Of course, with respect to an erase failure, the erasure that precedes programming will also occur after the page of memory has been completely tested. In which case, the erroneous data will be erased from memory 110 prior to programming the memory 110 with correct data. Testing of the second page will proceed uninterrupted until word 31 has been evaluated by ECC circuit 135. A refresh operation is activated on a positive clock edge. Specifically, when the refresh_start, page_refresh, and page_address pins are activated, the system will commence the page refresh mode shown in
A simplified example for refreshing a data word will now be discussed in reference to
If all words of a given page are determined to be error-free in step 330 then refresh is complete. In actuality, no refresh operation is performed on an error-free page. This method of performing page refresh is selective because if the bits are read out properly, without any errors, then it is unnecessary to subject the page to a refresh operation. Thus, an error-free page will simply be read and no erase or program steps will occur. Once the final word of the error-free page is transferred to storage buffer 140, the entire error-free page will be moved to page buffer 130 to be read back into memory 110.
If on the other hand, the page is determined to include defective words, a signal is generated in step 330 by the ECC circuit to alert the controller as to what action to take. In one embodiment, the signal may be a global flag that changes state when an error is detected. In step 370, if is determined that the error arose from an erased bit, then it becomes essential to erase the entire page in accordance with step 390. This is significant because if single bit errors were inadvertently erased, it is necessary to erase the entire source page together. Subsequent to erasing a page containing an erase failure, programming of correct data can begin immediately. After a correctible page has been erased, the intelligent program step of 395 is executed. At step 395 the corrected page is loaded into the memory array using a self-timed operation. After the intelligent page program 395, the device is free to be accessed for any purpose and the refresh is complete.
ECC circuit generates new check bits for a corrected data word and then reads out the corrected word into storage buffer 140. Almost simultaneously, the ECC circuit 135 provides an error signal to characterize the type of failure that has been detected. Error signals such as SEC#, or EED# of
The data from memory 110 can be evaluated at periodic intervals, although preferably the integrity of the data is checked when the memory array is idle. That is, if the memory is not being accessed for reading or writing, then its data can be evaluated for correctness. In addition, the memory can be checked for integrity at power up.
A second embodiment of the invention, as it applies to refreshing an entire memory, will be explained in conjunction with
To initiate step 460, a specific value for the Address ID storage is set. Next, each row of the ID storage is read out in sequence into a temporary storage buffer. If the ID Storage is error-free then the yes branch of step 470 will be executed and the memory refresh will be complete. Under those circumstances, the error-free Address ID will be reloaded into the ID storage from the temporary storage buffer.
On the other hand, if an error is identified in the ID storage then steps 475 and 480 will be executed to enable an intelligent erase and an intelligent programming of the Address ID data which is written out in corrected form from the temporary storage buffer into the Address ID storage. After step 480 is performed then the memory refresh is complete 490.
As mentioned previously, the present invention optimizes the handling of a defective word depending on the failure type detected.
On the other hand, if the word is incorrect, then it must be determined whether the error is an erase failure or a program failure. If the single error is an erase failure 575, then refresh path 577 will be pursued. Otherwise, if the single error is a program failure 595 then the detected error will be corrected through refresh path 580.
In the case where the single error is an erase failure 575, the word will eventually be erased, once each word in the source page has been fully tested by the ECC circuit. As used herein, a source page refers to the page from which the erase failure 575 originated. Consequently, if the end of the source page has been reached, then the source page is erased. Thereafter, the erased page is programmed with correct data in step 595. Once the source page is correct, it will be reprogrammed, or written, into memory 110 of
Refresh path 580 in conjunction with the detection of a program failure will now be discussed. If a program failure is detected then an erase operation can be skipped or omitted. Omitting the erase operation 590 will substantially improve the refresh time for providing corrected data into the memory array. Consequently, the refresh operation for a program failure 595 is expedited in comparison to the prior method of processing program failures by bypassing an erase operation and proceeding to execute the additional steps of refresh path 580. Specifically, if the end of page 540 has been reached, then it is determined whether there are any additional pages of memory to be evaluated in step 550. If there are additional pages to evaluate, then the page address of memory is incremented in 560 and the entire process of testing each word of a new page is repeated. Only until the end of memory 110 has been reached will the refresh be considered complete 565.
Within ECC circuit 135 is a check bit generator (not shown) which generates check bits S[6:0] in response to the data values read from each word in accordance with parity check matrix 600. Check bit generator produces check bits S[6:0] in response to the read data values RD[31:0] in accordance with parity check matrix 600. The check bits S[6:0] are then passed to ECC generator. The ECC generator uses a decoder to perform a bit-wise comparison between check bits S[6:0] and the read check bits RCB[6:0], wherein each of the check bits S[6:0] is exclusive OR'ed with a corresponding one of the read check bits RCB[6:0], as illustrated in the last 7 columns of parity-matrix 600. The resultant syndrome 7-bit word is decoded to identify whether the 39-bit code read from the memory array is free of error, contains a single-bit error, or contains multiple-bit errors. To illustrate the manner in which the check bits [6:0] are correctly generated, assume that an error bit occurs in data word D[31:0], such that the associated syndrome word SYN[6:0] has a non-zero value. The value of syndrome word SYN[6:0] is governed by the column (bit) location of the error bit in the parity-check matrix of
Based on this information, the system will use syndrome 14 to repair the error at bit location 28 by inverting the bit at location 28. Table 1 thus defines the manner in which each syndrome maps to a specific bit position to thereby identify the position of an error location. The present invention has been described by various examples above. However, the aforementioned examples are illustrative only and are not intended to limit the invention in any way. The skilled artisan would readily appreciate that the examples above are capable of various modifications. Thus, the invention is defined by the claims set forth below.
Number | Name | Date | Kind |
---|---|---|---|
4359772 | Patel | Nov 1982 | A |
5142541 | Kim et al. | Aug 1992 | A |
5574684 | Tomoeda | Nov 1996 | A |
6088268 | Gupta et al. | Jul 2000 | A |
6421757 | Wang et al. | Jul 2002 | B1 |
6496437 | Leung | Dec 2002 | B2 |
6751123 | Katayama et al. | Jun 2004 | B2 |
6751127 | Chou et al. | Jun 2004 | B1 |
6901011 | Micheloni et al. | May 2005 | B2 |
7051264 | Leung et al. | May 2006 | B2 |
7173862 | Futatsuyama et al. | Feb 2007 | B2 |
7283397 | Harari et al. | Oct 2007 | B2 |
7353438 | Leung et al. | Apr 2008 | B2 |
7360136 | Guterman et al. | Apr 2008 | B2 |
7379341 | Shen et al. | May 2008 | B2 |
7386765 | Ellis et al. | Jun 2008 | B2 |
7392456 | Leung et al. | Jun 2008 | B2 |
7426678 | Cory et al. | Sep 2008 | B1 |
20010015905 | Kim et al. | Aug 2001 | A1 |
20040153902 | Machado et al. | Aug 2004 | A1 |
20040170060 | Ishimoto | Sep 2004 | A1 |
20060120156 | Kawai et al. | Jun 2006 | A1 |
20060245259 | Fukuda et al. | Nov 2006 | A1 |
20060291289 | Lee | Dec 2006 | A1 |
20070109858 | Conley et al. | May 2007 | A1 |
20070297236 | Tokiwa | Dec 2007 | A1 |
20080068912 | Lee | Mar 2008 | A1 |
20080109705 | Pawlowski et al. | May 2008 | A1 |
20080301525 | Hirose | Dec 2008 | A1 |
20090024904 | Roohparvar et al. | Jan 2009 | A1 |
20100070801 | Cornwell et al. | Mar 2010 | A1 |
20100199150 | Shalvi et al. | Aug 2010 | A1 |
20110096605 | Kosaki et al. | Apr 2011 | A1 |
20110141812 | Choi et al. | Jun 2011 | A1 |
Number | Date | Country | |
---|---|---|---|
20100205504 A1 | Aug 2010 | US |