Electronic device and method of operating the same
Devices for using a neural network to choose an optimal error correction algorithm are disclosed. An example device includes a decoding controller inputting at least one of the number of primary unsatisfied check nodes (UCNs), the number of UCNs respectively corresponding to at least one iteration, and the number of correction bits respectively corresponding to the at least one iteration to a trained artificial neural network, and selecting any one of a first error correction decoding algorithm and a second error correction decoding algorithm based on an output of the trained artificial neural network corresponding to the input, and an error correction decoder performing error correction decoding on a read vector using the selected error correction decoding algorithm. The output of the trained artificial neural network may include a first predicted value indicating a possibility that a first error correction decoding using the first error correction decoding algorithm is successful.
1. An electronic device comprising:
a decoding controller configured to
transmit, to a trained artificial neural network, a first input comprising at least one of a number of primary unsatisfied check nodes (UCNs), a number of UCNs corresponding to at least one iteration, and a number of correction bits corresponding to the at least one iteration, and
select, based on an output of the trained artificial neural network corresponding to the first input, either a first error correction decoding algorithm or a second error correction decoding algorithm; and
an error correction decoder configured to perform an error correction decoding operation on a read vector using the selected error correction decoding algorithm,
wherein the output of the trained artificial neural network includes a first predicted value indicative of a probability that the first error correction decoding operation using a first error correction decoding algorithm is successful.
2. The electronic device of claim 1 , wherein the number of primary UCNs is calculated based on the read vector and a parity check matrix.
3. The electronic device of claim 2 , wherein the selected error correction decoding algorithm is the first error correction decoding algorithm,
wherein the decoding controller transmits, to the trained artificial neural network, a second input comprising at least one of the number of UCNs corresponding to an i-th iteration of the error correction decoding operation and the number of correction bits corresponding to the i-th iteration,
wherein i is a natural number,
wherein the decoding controller is further configured to determine whether to perform, based on an output of the trained artificial neural network corresponding to the second input, a second error correction decoding operation using the second error correction decoding algorithm, and
wherein the error correction decoder ends the first error correction decoding operation and performs the second error correction decoding operation based on the determination of the decoding controller.
4. The electronic device of claim 3 , wherein the decoding controller determines whether to transmit the second input to the trained artificial neural network based on the second input.
5. The electronic device of claim 4 , wherein the decoding controller transmits the second input to the trained artificial neural network based on whether the number of correction bits corresponding to the i-th iteration is less than a second threshold value.
6. The electronic device of claim 4 , wherein the decoding controller transmits the second input to the trained artificial neural network based on whether the number of UCNs corresponding to the i-th iteration exceeds a first threshold value.
7. The electronic device of claim 1 , wherein the first error correction decoding algorithm is a bit-flipping algorithm, and wherein the second error correction decoding algorithm is a min-sum algorithm or a sum-product algorithm.
8. The electronic device of claim 1 , wherein the first predicted value is a probability value to which logistic regression is applied.
9. The electronic device of claim 1 , wherein the first error correction decoding algorithm is selected upon a determination that the first predicted value of the output of the trained artificial neural network corresponding to the first input is equal to or greater than a first reference value, and wherein the second error correction decoding algorithm is selected upon a determination that the first predicted value is less than the first reference value.
10. The electronic device of claim 9 , wherein the output of the trained artificial neural network further includes a second predicted value indicative of a probability that the first error correction decoding operation has failed, and wherein the first reference value is the second predicted value.
11. The electronic device of claim 9 , wherein the decoding controller is further configured to select, based on the output of the trained artificial neural network corresponding to the first input, a power mode to be applied to the selected error correction decoding algorithm, wherein the power mode includes either a low power mode and a high power mode, and wherein the error correction decoder performs the error correction decoding operation by applying the selected power mode to the selected error correction decoding algorithm.
12. The electronic device of claim 11 , wherein the decoding controller is configured to:
select the high power mode upon a determination that the first predicted value is equal to or greater than the first reference value and less than a second reference value, the second reference value being greater than the first reference value, and
select the low power mode upon a determination that the first predicted value is equal to or greater than the second reference value.
13. The electronic device of claim 11 , wherein applying the high power mode comprises using an unreliability value of each variable node compared to the low power mode.
14. The electronic device of claim 11 , wherein applying the high power mode comprises using a higher bit precision compared to a bit precision used in the low power mode.
15. The electronic device of claim 1 , further comprising:
a retraining component configured to retrain the trained artificial neural network by labeling a retraining input vector,
wherein the retraining input vector includes at least one of the number of primary UCNs, the number of UCNs corresponding to the at least one iteration, and the number of correction bits corresponding to the at least one iteration,
wherein the retraining input vector is labeled as a failure class upon a determination that the error correction decoding operation using the first error correction decoding algorithm has failed, and
wherein the first error correction decoding algorithm is selected upon a determination that the first predicted value is equal to or greater than a first reference value.
16. The electronic device of claim 15 , wherein retraining component retrains the trained artificial neural network upon a determination that the first predicted value is equal to or greater than a second reference value, and wherein the second reference value is greater than the first reference value.
17. The electronic device of claim 1 , further comprising:
a retraining component configured to retrain the trained artificial neural network by soft labeling a retraining input vector,
wherein the retraining input vector includes at least one of the number of primary UCNs, the number of UCNs corresponding to the at least one iteration, and the number of correction bits respectively corresponding to the at least one iteration,
wherein the retraining input vector is soft labeled as a success class upon a determination that the error correction decoding operation is successful, and
wherein the selected error correction decoding algorithm is the second error correction decoding algorithm.
18. The electronic device of claim 17 , wherein the retraining component retrains the trained artificial neural network upon a determination that the number of iterations for which the second error correction decoding operation is successful is equal to or less than a third reference value.
19. The electronic device of claim 18 , wherein the retraining component sets a label value for the success class higher as a number of iterations in which the second error correction decoding operation is successful is lower.
20. A method of operating an electronic device, the method comprising:
transmitting, to a trained artificial neural network, an input comprising at least one of a number of primary unsatisfied check nodes (UCNs), a number of UCNs corresponding to at least one iteration, and a number of correction bits corresponding to the at least one iteration;
selecting, based on an output of the trained artificial neural network corresponding to the input, either a first error correction decoding algorithm or a second error correction decoding algorithm; and
performing an error correction decoding operation on a read vector using the selected error correction decoding algorithm,
wherein the output of the trained artificial neural network includes a first predicted value indicative of a probability that a first error correction decoding operation using the first error correction decoding algorithm is successful.