Solid-state drive error recovery based on machine learning

Patent metadata

Field	Value
Publication number	US-11275646-B1
Application number	US-202016815753-A
Country	US
Kind code	B1
Filing date	Mar 11, 2020
Priority date	Mar 11, 2019
Publication date	Mar 15, 2022
Grant date	Mar 15, 2022

Abstract

Systems and methods for selecting an optimal error recovery procedure for correcting a read error in a solid-state drive are provided. A machine learning model is trained to forecast which error recovery procedure of a plurality of error recovery procedures is most likely to achieve a predetermined goal given a state of a solid-state drive. The predetermined goal is based on at least one of a read latency and a failure rate of the solid-state drive. A current state of the solid-state drive is determined. An error recovery procedure is selected from among the plurality of error recovery procedures by inputting the current state of the solid-state drive into the trained machine learning model, thereby triggering the trained machine learning model to output the selected error recovery procedure. The selected error recovery procedure is executed to recover data from the solid-state drive.

First claim

What is claimed is:

1. A method for selecting an optimal error recovery procedure for correcting a read error in a solid-state drive, the method comprising:
training a machine learning model to forecast which error recovery procedure among a plurality of error recovery procedures is most likely to achieve a predetermined error recovery goal for a given state of a solid-state drive, wherein the predetermined goal is based on at least one of a read latency and a failure rate of the solid-state drive;
determining a current state of the solid-state drive by reading state data from the solid-state drive;
selecting an error recovery procedure from among the plurality of error recovery procedures by inputting the current state of the solid-state drive into the trained machine learning model, thereby triggering the trained machine learning model to output the selected error recovery procedure; and
executing the selected error recovery procedure to recover data from the solid-state drive.

2. The method as claimed in claim 1 , wherein the training the machine learning model comprises:
observing a state of the solid-state drive;
executing one of the plurality of error recovery procedures to recover data from the solid-state drive;
determining a result indicating whether the executed error recovery procedure successfully recovered the data from the solid-state drive and indicating a latency of the executing of the error recovery procedure;
generating a reward based on the determined result and a predetermined rule indicating a magnitude to be allocated to the reward based on the latency of the executing of the error recovery procedure; and
applying the reward to the machine learning model.

3. The method as claimed in claim 2 , wherein the determining the result of the executed error recovery procedure comprises determining a latency of successfully recovering data from the solid-state drive, and wherein the predetermined rule comprises a rule that dictates an magnitude of the reward that is inversely proportional to the latency.

4. The method as claimed in claim 3 , wherein determining the latency of successfully recovering data from the solid-state drive comprises determining a sum of respective latencies of one or more error recovery procedures that are executed to successfully recover the data from the solid-state drive.

5. The method as claimed in claim 2 , wherein the determining the result of the executed error recovery procedure comprises determining a failure rate of the solid-state drive, and wherein the predetermined rule comprises a rule that dictates a magnitude of the reward that is inversely proportional to the failure rate.

6. The method as claimed in claim 1 , wherein determining the current state of the solid-state drive comprises determining at least one of allocation unit-based decoder statistics, media statistics, media health statistics, and a temperature.

7. The method as claimed in claim 1 , wherein the plurality of error recovery procedures comprise at least one of a one-bit read retry, a two-bit read retry, a k-read deep retry, a voltage reference calibration, a log likelihood ratio calibration, an inter-cell interference cancellation, a hard error mitigation, and a redundant array of independent disks-based procedure.

8. The method as claimed in claim 1 , wherein selecting the error recovery procedure further comprises selecting an error recovery procedure parameter, and wherein the executing the selected error recovery procedure comprises executing the selected error recovery procedure based on the selected error recovery procedure parameter.

9. The method as claimed in claim 8 , wherein the error recovery procedure parameter comprises one or more voltage reference numbers, and wherein the executing the selected error recovery procedure comprises executing the selected error recovery procedure based on the selected one or more voltage reference numbers.

10. The method as claimed in claim 1 , wherein the predetermined goal is based on a combination of the read latency and the failure rate of the solid-state drive.

11. A system for selecting an optimal error recovery procedure for correcting a read error in a solid-state drive, the system comprising:
a machine learning agent that comprises a machine learning model and is communicatively coupled to a solid-state drive, the machine learning agent comprising hardware and being configured to:
train the machine learning model to forecast which error recovery procedure among a plurality of error recovery procedures is most likely to achieve a predetermined error recovery goal for a given state of the solid-state drive, wherein the predetermined goal is based on at least one of a read latency and a failure rate of the solid-state drive;
determine a current state of the solid-state drive by reading state data from the solid-state drive;
select an error recovery procedure from among the plurality of error recovery procedures by inputting the current state of the solid-state drive into the trained machine learning model, thereby triggering the trained machine learning model to output the selected error recovery procedure; and
cause the solid-state drive to execute the selected error recovery procedure to recover data from the solid-state drive.

12. The system as claimed in claim 11 , wherein the machine learning agent is configured to train the machine learning model by:
observing a state of the solid-state drive;
executing one of the plurality of error recovery procedures to recover data from the solid-state drive;
determining a result indicating whether the executed error recovery procedure successfully recovered the data from the solid-state drive and indicating a latency of the executing of the error recovery procedure;
generating a reward based on the determined result and a predetermined rule indicating a magnitude to be allocated to the reward based on the latency of the executing of the error recovery procedure; and
applying the reward to the machine learning model.

13. The system as claimed in claim 12 , wherein the machine learning agent is configured to determine the result of the executed error recovery procedure by determining a latency of successfully recovering data from the solid-state drive, and wherein the predetermined rule comprises a rule that dictates a magnitude of the reward that is inversely proportional to the latency.

14. The system as claimed in claim 13 , wherein the machine learning agent is configured to determine the latency of successfully recovering data from the solid-state drive by determining a sum of respective latencies of one or more error recovery procedures that are executed to successfully recover the data from the solid-state drive.

15. The system as claimed in claim 12 , wherein the machine learning agent is configured to determine the result of the executed error recovery procedure comprises determining a failure rate of the solid-state drive, and wherein the predetermined rule comprises a rule that dictates a magnitude of the reward that is inversely proportional to the failure rate.

16. The system as claimed in claim 11 , wherein the machine learning agent is configured to determine the current state of the solid-state drive by determining at least one of allocation unit-based decoder statistics, media statistics, media health statistics, and a temperature.

17. The system as claimed in claim 11 , wherein the plurality of error recovery procedures comprise at least one of a one-bit read retry, a two-bit read retry, a k-read deep retry, a voltage reference calibrati

Assignees

• Marvell Asia Pte Ltd Registration No 199702379M
• Marvell Asia Pte Ltd

Inventors

• Nguyen Phong Sy
• Nguyen Dung Viet
• Therene Christophe
• Varnica Nedeljko

Classifications

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/088

  Non-supervised learning, e.g. competitive learning · CPC title

• G06N3/006

  based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO] · CPC title

• G06F11/3452

  Performance evaluation by statistical analysis · CPC title

• G06F11/0727

  in a storage system, e.g. in a DASD or network based storage system (drivers for digital recording or reproducing units G06F3/06; circuits for error detection or correction within digital recording or reproducing units G11B20/18; for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS], H04L67/1097) · CPC title