Fault-tolerant memory system with graceful degradation

• US 5,278,847 A
• Filed: 12/28/1990
• Issued: 01/11/1994
• Est. Priority Date: 12/28/1990
• Status: Expired due to Term

• Current Assignee: L-3 Communications Corporation (L3Harris Technologies, Inc.)
• Original Assignee: General Electric Company
• Inventors: Anastasia, Thomas, Helbig, Walter A. Sr.
• Primary Examiner(s): Canney, Vincent P.
• Application Number:
  US07/635,068
• Time in Patent Office: 1,110 Days
• Field of Search: 370/40.1, 370/40.2, 370/40.3, 370/40.4, 370/21.6, 370/10.2, 370/10.3, 365/200
• US Class Current: 714/764
• CPC Class Codes: G06F 11/1044 with specific ECC/EDC distribution
• CPC Class Codes: G11C 29/808 using a flexible replacement scheme
• CPC Class Codes: G11C 29/88 with partially good memories

Abstract

A fault-tolerating memory system has a data memory with a large number (M+N) of data storage words each having a length greater than the length of user data to be stored in that word; the extra word length is used for at least an error-detecting-and-correcting (EDAC) code. The user data is stored in a smaller number (N) of the words, with the remaining number (M) of words being used to store a map of which portions, if any, of each word are not usable. The N words of user data storage can include S normal storage words and (N-S) spare words, each for use if one of the normal storage words has too many unusable portions. A portion of each word length can contain at least one spare word portion, to which a block of data can be moved if any bit of a like-sized portion of the normal storage word is unusable. The reliability of storage is greatly improved by extension of each word to add EDAC encoding and spare-bit portions, as well as by extension of depth to allow spare words to be present, along with high-reliability storage of word maps.

Claims

1. A method for tolerating faults during the storage of data words in a gigabit memory system having a multiplicity of individual memory integrated circuits (ICs) and a totality of storage words, each having a multiplicity of bytes to be stored in a plurality of different ICs, comprising the steps of:

(a) dividing the totality of storage words into a first multiplicity of storage words and a second multiplicity of storage words;

(b) storing in the second multiplicity of storage words at least one updatable map of known-good space in each of the storage words in the memory;

(c1) first forming each data word with a first multi-byte portion of received user data, a second portion of at least one byte of EDAC-encoding data for detecting and correcting errors in the multiplicity of bytes of user data of the first portion of the same data word, and a third portion as a spare storage space with a sufficient length to allow a plurality of multiple-bit bursts of unusable storage bits in the user data and EDAC-encoding data portions of that same data word to be tolerated;

(c2) determining if at least one burst of unusable data bits exists in a next available storage word in the system memory;

(c3) then transferring the data bits in each burst of user data and EDAC-encoding data corresponding to an unusable memory burst portion in that word to the third portion of that same data word, prior to storage;

(d) after step (c3), storing each sequentially-received EDAC-encoded data word in that next-available one of the first multiplicity of storage words having sufficient known-good space for storage of a data word, as determined by reference to the at least one map in the second multiplicity of storage words;

(e) retrieving stored data from a sequence of the first multiplicity of storage words determined by reference to the at least one map in the second multiplicity of storage words; and

(f) then removing unusable burst and other errors in that data word, by (1) first transferring back to the proper burst locations within the same data word, as determined by reference to the associated map, the bit bursts from the third portion of that data word, (2) removing the third word portion to obtain a burst-transferred retrieved word, and then (3) utilizing the EDAC coding data of each retrieved burst-transferred word to correct at least one burst of user data error.

2. The method of claim 1, wherein the EDAC-encoding step allows error in a plurality of multiple-bit bursts to be detected.

3. The method of claim 2, wherein each burst is one nybble in length.

4. The method of claim 1, further comprising the steps of:

providing a spare word portion of memory; and

storing an EDAC-encoded data word into the spare memory portion in the event that a storage word selected therefore in the first storage word multiplicity can not faultlessly contain the data word to be stored.

5. A method for storage of gigabytes of data, comprising the steps of:

(A) recording data in a memory by the steps of

(1) receiving user data;

(2) adding EDAC-encoding data, based upon the received user data, to form a storage word;

(3) formatting the storage word by (a) adding a plurality of spare portions;

(b) recognizing any existing burst of unusable data bits in a data word about to be stored;

(c) transferring, prior to storage, the bits occurring in any plurality of unusable bit bursts of any of user data and EDAC-encoding data for a present word, to the spare portion of that word;

(4) determining, by reference to an updatable map, acceptable storage word spaces having sufficient good bits available for storage of any formatted storage word; and

(5) storing the formatted storage word in a next available acceptable space; and

(B) playing data back from the memory by the steps of

(1) retrieving the formatted storage word from its assigned memory storage space;

(2) unformatting the storage word by transferring back to the proper burst locations within the present word the bits previously transferred into the spare portion and then removing all spare portions;

(3) utilizing the EDAC-encoding data to detect if any error has been introduced into the user data, and to correct at least a portion of the detected error in that storage word; and

(4) outputting the corrected user data.

6. The method of claim 5, wherein the EDAC-encoding step (B3) detects a plurality of seperate errors and corrects at least one detected error.

7. The method of claim 5, wherein data from a plurality of unusable burst spaces can be saved in the plurality of spare portions of each storage word.

8. The method of claim 5, further comprising the steps of:

providing a spare word portion of memory; and

storing an EDAC-encoded data word into the spare memory portion in the event that a storage word initially selected therefore can not acceptably contain the data word to be stored.

9. The method of claim 5, further comprising the step of transferring to a spare word storage space any storage word having a greater number of unusable burst spaces than can be overcome by use of the EDAC-encoding and spare data word portions.

10. The method of claim 9, further comprising the steps of:

mapping all storage words to determine the location of unuseable burst spaces; and

storing the maps in a portion of the total memory space.

11. The method of claim 10, wherein each map is stored as at least one data word having its own EDAC-encoding and spare data portions.

12. Data storage apparatus, comprising:

memory means for storing a gigabit multiplicity of multi-byte storage words, each having a user data portion of an integer number of nybbles in length and having a spare word portion;

dynamic means for periodically mapping at least those of the storage words then having at least one burst of a plurality of bad storage bits therein, and for storing the bad-burst map until a next mapping;

means for receiving and outputting user data;

means for EDAC-encoding an assembled word of the received user data prior to storage and for adding a plurality of spare data portions to each data word for receiving up to a like plurality of bursts of data bits identified as having bit positions identical with those positions mapped as bad, and for operating upon a data word retrieved from said memory means first to restore transferred bit bursts to their original positions and then to detect error is the restored retrieved word and correct at least a portion of the detected error before transmittal to the outputting means; and

means for determining, in cooperation with said mapping means, a location within said memory means into which, in a storage operation, to store a data word, including locations in the memory means spare word portion in the event that a storage word location initially selected for a data word can not acceptably contain the data word to be stored, and also for determining from which location, in a retrieval operation, to take a data word.

13. The apparatus of claim 12, wherein the mapping means stores at least one storage word location map within the memory means.

14. The apparatus of claim 12, wherein each storage word has a user data portion of an integer number of nybbles in length, with said EDAC-encoding means providing an additional two nybbles of data, and said formatting means using two spare portions, into each of which an associated nybble of user data and EDAC-encoding data can be transferred.