Memory controller and operating method thereof

1. A memory controller controlling a memory device, the memory controller comprising:

an available-time prediction component configured to perform a machine learning operation using a Recurrent Neural Network (RNN) model based on input/output request information about an input/output request input from a host, and to predict an idle time representing a time during which the input/output request is not expected to be input from the host; and

a data compression controller configured to generate, in response to the idle time longer than a set reference time, compressed map data by compressing map data of data having a read count less than a reference number among data stored in the memory device,

wherein the map data indicates a mapping relationship between a logical address provided by the host and a physical address indicating a physical location of a memory block included in the memory device.

2. The memory controller of claim 1 , wherein the available-time prediction component comprises:

a plurality of preprocessing components configured to translate the input/output request information into data used for machine learning; and

a plurality of information processing components configured to perform machine learning for predicting the idle time by using output of the plurality of preprocessing components, and to output available-time prediction information about whether the idle time is enough time to compress the map data, which represents an available time.

3. The memory controller of claim 2 , wherein the plurality of information processing components perform machine learning by the Recurrent Neural Network (RNN) model.

4. The memory controller of claim 3 , wherein the plurality of information processing components perform machine learning by a Long Short Term Memory (LSTM) method which is a type of Recurrent Neural Network (RNN) model designed to have long-term memory or short-term memory.

5. The memory controller of claim 2 , wherein the plurality of information processing components have a Fully Connected Layer (FCL) in which nodes of a previous layer and nodes of a current layer are fully connected.

6. The memory controller of claim 2 , wherein the data compression controller comprises:

a block select component configured to determine, in response to the idle time longer than the set reference time, compression target map data which is map data to be compressed among the map data according to block information of the memory block;

a similarity detector configured to detect a difference among the compression target map data; and

a compression component configured to compress the compression target map data.

7. The memory controller of claim 6 , wherein the compression component compresses the compression target map data by a delta compression method.

8. The memory controller of claim 6 , wherein the compression component compresses the compression target map data by a GZIP method.

9. The memory controller of claim 6 , wherein the block information includes information about a number of valid pages included in the memory block or a read count of pages included in the memory block.

10. The memory controller of claim 1 , wherein the set reference time is an average time spent in compressing the map data.

11. A method of operating a memory controller for controlling a memory device, the method comprising:

receiving input/output request information from a host;

predicting, based on machine learning using a Recurrent Neural Network (RNN) model, an idle time representing a time during which an input/output request is not expected to be input from the host; and

compressing, in response to the idle time longer than a set reference time, map data of data having a read count less than a reference number among data stored in the memory device,

wherein the map data which is mapping information of a logical address provided by the host and a physical address indicating a physical location of a memory block included in the memory device.

12. The method of claim 11 , wherein the predicting of the idle time comprises:

processing the input/output request information to be optimized to machine learning;

performing machine learning for predicting the idle time based on the input/output request information; and

generating available-time prediction information about whether the idle time is enough time to compress the map data, which represents an available time.

13. The method of claim 12 , wherein the performing of the machine learning is executed by a Long Short Term Memory (LSTM) method which is a type of Recurrent Neural Network (RNN) model designed to have long-term memory or short-term memory.

14. The method of claim 12 , wherein the performing of the machine learning is executed in a Fully Connected Layer (FCL) in which nodes of a previous layer and nodes of a current layer are fully connected.

15. The method of claim 12 , wherein the generating of the available-time prediction information is executed in a Fully Connected Layer (FCL) in which nodes of a previous layer and nodes of a current layer are fully connected.

16. The method of claim 11 , wherein the compressing of the map data is executed by a delta compression method in which data may be compressed based on a difference between pieces of sequential data.

17. The method of claim 16 , wherein the compressing of the map data comprises compressing data by a GZIP method.

18. The method of claim 11 , wherein the compressing of the map data comprises determining compression target map data based on block information of blocks storing the map data.

19. The method of claim 18 , wherein the block information includes information about a number of valid pages included in the memory block or a read count of pages included in the memory block.

20. An operating method of a controller for controlling a storage device including blocks, the operating method comprising:

controlling, in response to access requests, the storage device to access the blocks;

accumulating the access requests;

predicting, by applying machine learning to the accumulated access requests, an idle time during which an access request is not expected to be received; and

compressing map information related to select blocks during the idle time,

wherein time required to compress the map information is shorter than the idle time, and

wherein the select blocks include more valid data than other blocks.