ERROR HANDLING APPARATUS

摘要

1513831 Error correction INTERNATIONAL BUSINESS MACHINES CORP 5 Sept 1975 [31 Dec 1974] 36572/75 Heading G4A Data words encoded in an error correcting code are read from a memory 10 and, when found to be in error, are subjected to two different sequences of error correction operations, the resulting "corrected" words being compared for equality before one of them is delivered to an output. First correction sequence. A word W is read from memory 10 into a register A and a register MDR, syndrome bits and byte parity bits being generated at 11, 20. If any syndrome bit is non-zero, an output E from OR gate 16 signifies the presence of an error, and the syndrome bits, decoded by AND gates 13, are used to attempt correction of the word W from MDR in EXOR gates 14, the "corrected" word W1 being then entered in register A, MDR and checked for non- zero syndrome bits (E), syndrome parity (G), and correct byte parity (R). RE indicates no error and allows the corrected word to be read out from register MDR to a data processor. Certain other states of R, E indicate an uncorrectable error and result in an error stop. EG initiates a further sub-sequence of correction operations in which the complement of the "corrected" word W1 is written into the memory 10 and rea9 back to register A for comparison at 18 with the original word W from MDR. The resulting fault indicating word W* has a "1" at each bit position corresponding to a faulty, i.e. "stuck" cell in the memory word location concerned. The word W* is tested for EG and, if this condition is satisfied (other conditions result in an error stop) a further attempt at correction is made by EXOR gates 17 operating on W from MDR and W* from A. Finally the corrected word is re-tested as at the start of the sequence. Second correction sequence. After the initial test of a word W has indicated the presence of an error, the signal G is stored as a signal H and the steps of complementing W, storing in memory, reading back and forming the fault indicating word W* are performed as in the first sequence. If the test on the word W* indicates GEH, the original word W and W* are EXOR'ed at 17, the result is entered in registers A, MDR and an attempt at correction using EXOR gates 14 is made. A final test of the "corrected" word W1 results in delivery of the word W1 if ER = 1, or in an error stop. If the test on the fault indicator word W* indicates GEH or GEH, a similar sub-sequence of operations is performed, but at the final test of the "corrected" word W1, if ER = 1, the corrected word is compared at 19 with the corrected word resulting from the first correction sequence and stored in a register B, and only if equality is detected is the word delivered to the processor. The second correction sequence is more fault secure than the first. A detailed diagram of the system is given in Fig. 2 (not shown).