Machine-learning assisted decoder with asymmetric ratio tracking in non-volatile memory devices

An apparatus for improving a performance of an iterative decoder in a non-volatile memory, comprising:

• a memory controller;
• and
• a neural network communicatively coupled to the memory controller,
• wherein the memory controller is configured to:
• receive a noisy codeword that is based on a transmitted codeword generated from a low-density parity-check (LDPC) code;
• perform, on the noisy codeword for a first number of iterations, a message passing algorithm between a plurality of variable nodes and a plurality of check nodes that represent a parity matrix of the LDPC code, wherein the message passing algorithm uses a set of log-likelihood ratios (LLRs) initialized with symmetric LLRs;
• assign, based on comparing an asymmetric ratio to a threshold, a set of asymmetric LLRs to the set of LLRs, wherein the set of asymmetric LLRs is determined based on the asymmetric ratio, and wherein a neural network is used to determine the asymmetric ratio based on a number of zero-to-one bit errors and a number of one-to-zero bit errors;
• and
• determine, subsequent to performing the message passing algorithm on the noisy codeword for a second number of iterations, a candidate version of the transmitted codeword, wherein a sum of the first number of iterations and the second number of iterations is less than or equal to a maximum number of iterations configured for the iterative decoder.

A non-transitory computer-readable storage medium having instructions stored thereupon for improving a performance of an iterative decoder in a non-volatile memory, comprising:

• instructions for receiving a noisy codeword that is based on a transmitted codeword generated from a low-density parity-check (LDPC) code;
• instructions for performing, on the noisy codeword for a first number of iterations, a message passing algorithm between a plurality of variable nodes and a plurality of check nodes that represent a parity matrix of the LDPC code, wherein the message passing algorithm uses a set of log-likelihood ratios (LLRs) initialized with symmetric LLRs;
• instructions for assigning, based on comparing an asymmetric ratio to a threshold, a set of asymmetric LLRs to the set of LLRs, wherein the set of asymmetric LLRs is determined based on the asymmetric ratio, and wherein a neural network is used to determine the asymmetric ratio based on a number of zero-to-one bit errors and a number of one-to-zero bit errors;
• and
• instructions for determining, subsequent to performing the message passing algorithm on the noisy codeword for a second number of iterations, a candidate version of the transmitted codeword, wherein a sum of the first number of iterations and the second number of iterations is less than or equal to a maximum number of iterations configured for the iterative decoder.