Methods of predicting reliability information of storage devices and methods of operating storage devices

Patent metadata

Field	Value
Publication number	US-12586004-B2
Application number	US-202117497076-A
Country	US
Kind code	B2
Filing date	Oct 8, 2021
Priority date	Feb 5, 2021
Publication date	Mar 24, 2026
Grant date	Mar 24, 2026

Abstract

In a method of operating a storage device including a plurality of nonvolatile memories, reliability information of the storage device is predicted. A read operation on the storage device is performed based on a result of predicting the reliability information. In the predicting the reliability information of the storage device, a model request signal is outputted by selecting one of a plurality of machine learning models as an optimal machine learning model based on deterioration characteristic information and deterioration phase information. The model request signal corresponds to the optimal machine learning model. The plurality of machine learning models are used to generate first reliability information related to the plurality of nonvolatile memories. First parameters of the optimal machine learning model may be received based on the model request signal. The first reliability information is generated based on the deterioration characteristic information and the first parameters.

First claim

What is claimed is:

1 . A method of predicting reliability information of a storage device including a plurality of nonvolatile memories, the method comprising:
outputting a model request signal corresponding to an optimal machine learning model by selecting one of a plurality of machine learning models as the optimal machine learning model, with each of the plurality of machine learning models configured to generate first reliability information related to the plurality of nonvolatile memories and the selecting of the one of the plurality of machine learning models based on deterioration characteristic information obtained by accumulating deterioration information related to the plurality of nonvolatile memories and deterioration phase information representing a degree of deterioration of the plurality of nonvolatile memories; receiving first parameters of the optimal machine learning model based on the model request signal; and generating the first reliability information based on the deterioration characteristic information and the first parameters of the optimal machine learning model.

2 . The method of claim 1 , wherein:
the deterioration phase information represents the degree of the deterioration of the plurality of nonvolatile memories as a plurality of deterioration phases, and wherein selecting the one of the plurality of machine learning models comprises selecting a first machine learning model from among the plurality of machine learning models as the optimal machine learning model based on the deterioration phase information corresponding to an early deterioration phase, and selecting a second machine learning model different from the first machine learning model selected from among the plurality of machine learning models as the optimal machine learning model based on the deterioration phase information corresponding to a middle deterioration phase that is subsequent to the early deterioration phase.

3 . The method of claim 2 , wherein the first machine learning model has a processing speed higher than that of the second machine learning model, and the second machine learning model has an accuracy higher than that of the first machine learning model.

4 . The method of claim 3 , wherein:
each of the plurality of machine learning models is stored in one of a plurality of volatile memories that includes a tightly-coupled memory (TCM), a static random access memory (SRAM) and a dynamic random access memory (DRAM), and the first machine learning model is stored in the TCM, and the second machine learning model is stored in one of the SRAM and the DRAM.

5 . The method of claim 1 , wherein outputting the model request signal includes:
collecting the deterioration characteristic information; and generating the deterioration phase information based on the deterioration characteristic information, the deterioration phase information representing the degree of the deterioration of the plurality of nonvolatile memories as a plurality of deterioration phases.

6 . The method of claim 5 , wherein collecting the deterioration characteristic information includes:
generating the deterioration characteristic information based on at least one of commands issued by a storage controller included in the storage device; and updating the deterioration characteristic information based on at least one of the commands.

7 . The method of claim 6 , wherein the deterioration characteristic information includes at least one of program/erase (P/E) cycles, read counts, a retention time, a number of on-cells, and a number of error bits for the plurality of nonvolatile memories.

8 . The method of claim 7 , wherein the commands include a program command, a read command, an erase command, a retention time generating command, and an on-cell count generating command.

9 . The method of claim 8 , wherein updating the deterioration characteristic information includes:
updating the P/E cycles based on the program command and the erase command; and updating the read counts based on the read command.

10 . The method of claim 8 , wherein updating the deterioration characteristic information includes:
updating the retention time based on the program command and the retention time generating command; updating the number of on-cells based on the on-cell count generating command; and updating the number of error bits based on one of the read command and the program command.

11 . The method of claim 5 , wherein generating the deterioration phase information includes:
retrieving the deterioration characteristic information; retrieving a lookup table representing a relationship between the deterioration characteristic information and the plurality of deterioration phases; and generating the deterioration phase information based on the deterioration characteristic information and the lookup table.

12 . The method of claim 1 , wherein outputting the model request signal includes:
evaluating the plurality of machine learning models according to a plurality of model evaluation criteria; storing a plurality of parameters associated with each of the plurality of machine learning models in different areas of a plurality of volatile memories included in the storage device based on an evaluation result according to the plurality of model evaluation criteria; and selecting one of the plurality of machine learning models based on the deterioration phase information.

13 . The method of claim 12 , wherein the plurality of machine learning models are based on at least one of a convolution neural network (CNN), a recurrent neural network (RNN), a support vector machine (SVM), a linear regression, a logistic regression, a naive bayes classification, a random forest, a decision tree, and/or k-nearest neighbor (KNN) algorithms.

14 . The method of claim 12 , wherein the plurality of model evaluation criteria include first to X-th model evaluation criteria, where X is an integer greater than or equal to two, and
each of the first to X-th model evaluation criteria is a criterion for sorting the plurality of machine learning models based on at least one of a processing speed and an accuracy of the plurality of machine learning models.

15 . The method of claim 14 , wherein:
the plurality of volatile memories include a tightly-coupled memory (TCM), a static random access memory (SRAM) and a dynamic random access memory (DRAM), numbers of the plurality of parameters included in the plurality of machine learning models are different from each other, the plurality of machine learning models are classified into first machine learning models, second machine learning models and third machine learning models based on the numbers of the plurality of parameters, the first machine learning models are stored in the TCM, the second machine learning models are stored in the SRAM, and the third machine learning models are stored in the DRAM.

16 . The method of claim 15 , wherein:
a number of parameters included in the first machine learning models is smaller than a number of the parameters included in the second machine learning models, and the number of parameters included in the second machine learning models is smaller than a number of the parameters included in the third machine learning models.

17 . The method of claim 15 , wherein receiving the first parameters of the optimal machine learning model includes:
when one of the first machine learning models is selected as the optimal machine learning model, receiving the first parameters via a TCM interface; and when one

Assignees

• Samsung Electronics Co Ltd

Inventors

• Kim Chanha
• Kim Youngeun
• Roh Kangho
• Jung Younghoon
• Cho Mijung

Classifications

• G11C16/26

  Sensing or reading circuits; Data output circuits · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N3/08

  Learning methods · CPC title

• G06F3/0679

  Non-volatile semiconductor memory device, e.g. flash memory, one time programmable memory [OTP] · CPC title

• G06F3/0658

  Controller construction arrangements · CPC title