US 20070061685 A1
A circuit and method for generating an Error Correcting Code (ECC) based on an adjacent symbol codeword that is formed in two clock phases.
1. A method for testing a memory comprising:
generating a plurality of check bits to append to data that is forwarded to the memory;
generating an adjacent symbol codeword based at least in part on data received from the memory;
decoding the adjacent symbol codeword; and
determining whether an error exists in the memory.
2. The method of
3. The method of
4. The method of
classifying the error in the received data; and
correcting the error in the received data.
5. The method of
6. The method of
This application is a division of U.S. patent application Ser. No. 10/624,424, (Attorney Docket No. 042390.P15881), filed Jul. 21, 2003 entitled “A METHOD, SYSTEM, AND APPARATUS FOR ADJACENT-SYMBOL ERROR CORRECTION AND DETECTION” and is related to concurrently filed U.S. Patent Application No. (unassigned), Attorney Docket No. 042390.P15881D2, entitled “A METHOD, SYSTEM, AND APPARATUS FOR ADJACENT-SYMBOL ERROR CORRECTION AND DETECTION.”
The present disclosure pertains to the field of memory and computer memory systems and more specifically to error detection and correction for memory errors.
2. Description of Related Art
Error correcting codes (ECC) have been routinely used for fault tolerance in computer memory subsystems. The most commonly used codes are the single error correcting (SEC) and double error detecting (DED) codes capable of correcting all single errors and detecting all double errors in a code word.
As the trend of chip manufacturing is toward a larger chip capacity, more and more memory subsystems will be configured in b-bits-per-chip. The most appropriate symbol ECC to use on the memory are the single symbol error correcting (SbEC) and double symbol error detecting (DbED) codes, wherein “b” is the width (number of bits in output) of the memory device, that correct all single symbol errors and detect all double symbol errors in a code word. A memory designed with a SbEC-DbED code can continue to function when a memory chip fails, regardless of its failure mode. When there are two failing chips that line up in the same ECC word sometime later, the SbEC-DbED code would provide the necessary error detection and protect the data integrity for the memory.
Existing and imminent memory systems utilize eighteen memory devices. However, the present SbEC-DbED error correcting codes utilize 36 memory devices in order to provide chipfail correction and detection. Thus, the cost increases due to the added expense of 36 memory devices for error correcting purposes and they are inflexible because they do not scale (adapt) to the memory systems with eighteen memory devices. Furthermore, the various circuits for encoding and decoding the errors are complex. Thus, this increases the cost and design of computer systems to insure data integrity.
The present invention is illustrated by way of example and not limitation in the Figures of the accompanying drawings.
The following description provides a method, apparatus, and system for error detection and correction of memory devices. In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. It will be appreciated, however, by one skilled in the art that the invention may be practiced without such specific details. Those of ordinary skill in the art, with the included descriptions, will be able to implement appropriate logic circuits without undue experimentation.
As previously described, typical ECC code utilizes 36 memory devices for chipfail detection and correction that results in increased cost and design of a computer system. Also, with the advent of eighteen memory devices in a system, the present ECC codes do not scale. In contrast, the claimed subject matter facilitates a new ECC code, “adjacent-symbol” code that supports memory systems with 18 memory devices For example, in one embodiment, the claimed subject matter facilitates the ability for decoding and correcting memory errors in systems that utilize 18 memory devices for a memory transaction (memory rank). Furthermore, the claimed subject matter facilitates forming a code word of data with only two clock phases. Also, the adjacent-symbol ECC code corrects any error pattern within the data from one memory device and detects various errors (double device errors) from failures in 2 memory devices.
In one embodiment, the adjacent-symbol ECC code is utilized for a memory system with two channels of Double Data Rate (DDR) memory, wherein each channel is 64 bits wide with eight optional bits for ECC. Also, the memory system may utilize x4 or x8 wide memory devices (x4 and x8 refers to the number of bits that can be output from the memory device). Thus, the claimed subject matter supports various configurations of memory systems. For example, a memory system with x8 devices would utilize 18 memory devices per memory rank if ECC is supported, otherwise, 16 memory devices per memory rank if ECC is not supported. Alternatively, a memory system with x4 devices would utilize 36 memory devices per memory rank if ECC is supported, otherwise, 32 memory devices per memory rank if ECC is not supported.
The scheme illustrated in the block diagram facilitates error detection and improves fault coverage of common mode errors. For example, for an x4 memory device, there is a one to one mapping of nibbles from the x4 memory device to a symbol in the underlying code word. In contrast, for a x8 memory device, there is a one to one mapping of nibbles from half of the x8 memory device to a symbol in the underlying code word Thus, the claimed subject matter facilitates isolating common mode errors across nibbles to the symbol level and results in increased fault coverage. Therefore, for the x8 memory device, the claimed subject matter precludes aliasing for a second device failure. Likewise, device errors in the x4 memory devices are isolated to a single symbol in the codeword 106, thus, there is complete double device coverage for the x4 memory devices.
To further illustrate, there are typically two classes of double device failures, simultaneous and sequence, that occur in the same memory rank.
A simultaneous double device failure has no early sign warning because there is no indication of an error in a previous memory transaction. Typically, the computer system reports an uncorrectable error in the absence of an aliasing. However, the system might incorrectly report a correctable single device failure. This time the aliasing may be discovered in subsequent accesses because an error pattern might change as to preclude the alias.
In contrast, a sequential double device failure is a more typical failure pattern than a simultaneous double device failure. Typically, the first device error is detected as a correctable error. For a second device failure, there may be two outcomes in one embodiment; the error is reported as uncorrectable, otherwise, the error is reported as a correctable error at a new location. In the event of an uncorrectable error for the second device failure, the analysis is complete. Otherwise, the system changes the error location from the first device failure to the second device's failure location. Therefore, the preceding method for detecting the alias is accurate because it is unlikely that the first device failure location resolves itself and even less likely that is does at the simultaneous instant that the second device failure has failed.
A few examples of double device errors that are always detected (no aliasing) are double bit errors, double wire faults, wire faults in one memory device with a single bit error in a second memory device, and a fault that affects only one nibble of each memory device.
In one example of a device error for the x8 memory device, all 16 bits of the codeword (adjacent symbols) may be affected (corrupted) because the failure results in an error for both nibbles and both clock phases of the memory device's data. Thus, the claimed subject matter facilitates the correction of this device failure by first correcting the 16 bits that are in error. However, in the event of a second memory device failure, the code detects the error pattern in two groups of 16 bits which are aligned on 16-bit boundaries in the code word 106.
The apparatus comprises an encoder circuit 202, at least one memory device 204, a decoder circuit 206, an error classification circuit 208, and a correction circuit 210.
The encoder circuit receives data that is to be forwarded to the memory device or memory devices 204. The encoder circuit generates a plurality of check bits based at least in part on the data. Thus, a codeword is formed based at least in part on the plurality of check bits and the data and is forwarded to the memory device or memory devices 204.
In one embodiment, the check bits are generated from the binary form of a G-matrix, wherein the matrix has 32 rows and 256 columns to form 32 check bits. The check bits are computed as follows:
For binary data, the multiply operation becomes an AND function and the sum operation the 1-bit sum or XOR operation. Thus, the resulting encoding circuit comprises 32 XOR, each tree computing one of the 32 check bits.
Subsequently, the memory device or memory devices 104 returns data and the check bits back to the decoder circuit 106. In one embodiment, the decoder circuit generates a 32-bit syndrome based at least in part on a 288-bit code word (as earlier described in connection with
In one embodiment, the syndrome is generated from an H-matrix, wherein the matrix comprises 32 rows and 288 columns. Each syndrome bit is calculated as follows:
As previously described with the encoder circuit, the generation of the syndrome bits is simplified to a XOR operation over the code word bits corresponding to the columns of the H-matrix that have a binary 1 value. Thus, the decoding circuit comprises 32 XOR trees, each tree computing one of the 32 syndrome bits. Therefore, in one embodiment, a 32 bit syndrome is generated by an H matrix receiving a 288 bit codeword. However, the claimed subject matter is not limited to this bit configuration. One skilled in the art appreciates modifications to the size of the syndrome and codeword.
The error classification and error correction are described in connection with
The error classification is based at least in part on the decoding circuit's computation of the syndrome. For example, in one embodiment, if the syndrome (S)=0, then there is NO error. Otherwise, if the syndrome (S)>0, there is an error. Also, it is optional to further classify the error by computing an error location vector L. For example, in one embodiment, the error is uncorrectable if L=0. Otherwise, the error is correctable in an indicated column if L>0. Furthermore, one may further classify the correctable error as whether the error occurs in a data column or check column. For example, if the error is in a check column, the data portion of the code word may bypass the correction logic.
In yet another embodiment, a single device correctable error may be classified based at least in part on a weight of the error value. As depicted in
Thus, the claimed subject matter allows for test coverage of both single and double device errors.
While certain exemplary embodiments have been described and shown in the accompanying drawings, it is to be understood that such embodiments are merely illustrative of and not restrictive on the broad invention, and that this invention not be limited to the specific constructions and arrangements shown and described, since various other modifications may occur to those ordinarily skilled in the art upon studying this disclosure.