Identification and mitigation of hard errors in memory systems

摘要

Embodiments provide a method comprising estimating a first set of log-likelihood ratio (LLR) values for a plurality of memory cells of a memory; based on the first set of LLR values, performing a first error correcting code (ECC) decoding operation; in response to determining a failure of the first ECC decoding operation, generating, by adjusting the first set of LLR values, a second set of LLR values for the plurality of memory cells; and based on the second set of LLR values, performing a second ECC decoding operation.

主权项

1. A method comprising:
estimating a first set of log-likelihood ratio (LLR) values for a plurality of memory cells of a memory, wherein each LLR value of the first set of LLR values is assigned to a respective memory cell of the plurality of memory cells; based on the first set of LLR values, performing a first error correcting code (ECC) decoding operation; in response to determining a failure of the first ECC decoding operation, generating, by adjusting the first set of LLR values, a second set of LLR values for the plurality of memory cells, wherein each LLR value of the second set of LLR values is assigned to a respective memory cell of the plurality of memory cells, wherein each LLR value of the second set of LLR values is generated by decreasing a magnitude of a corresponding LLR value of the first set of LLR values by a corresponding amount; and based on the second set of LLR values, performing a second ECC decoding operation, wherein the first set of LLR values comprises at least a first LLR value and a second LLR value, wherein a magnitude of the first LLR value is greater than a magnitude of the second LLR value, and wherein generating the second set of LLR values further comprises
generating the second set of LLR values such that the second set of LLR values comprises at least (i) a third LLR value that is generated by decreasing a magnitude of the first LLR value by a first value, and (ii) a fourth LLR value that is generated by decreasing a magnitude of the second LLR value by a second value, wherein based on the magnitude of the first LLR value being greater than the magnitude of the second LLR value, (i) the first value is greater than the second value, and (ii) a magnitude of the third LLR value is greater than a magnitude of the fourth LLR value.