STORAGE DEVICE, METHOD AND NON-VOLATILE MEMORY DEVICE PERFORMING GARBAGE COLLECTION USING ESTIMATED NUMBER OF VALID PAGES

• US 20200133845A1
• Filed: 06/19/2019
• Published: 04/30/2020
• Est. Priority Date: 10/25/2018
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: KIM, BYEONG-HUI, SEO, JUNG-MIN, MIN, HYEON-GYU, YANG, SEUNG-JUN, HWANG, JOO-YOUNG
• Granted Patent: US 10,896,126 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 12/0246 in block erasable memory, e.g. flash memory
• CPC Class Codes: G06F 12/0253 Garbage collection, i.e. reclamation of unreferenced memory
• CPC Class Codes: G06F 12/121 using replacement algorithms
• CPC Class Codes: G06F 2212/1036 Life time enhancement
• CPC Class Codes: G06F 2212/1044 Space efficiency improvement
• CPC Class Codes: G06F 2212/7205 Cleaning, compaction, garbage collection, erase control
• CPC Class Codes: G06F 2212/7211 Wear leveling
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

Garbage collection is performed according to an estimated number of valid pages. A storage device estimates a valid page count at a future time based on a valid page count at each of past time steps and a present time step using a neural network model and selects a victim block that undergoes the garbage collection from memory blocks based on an estimated valid page count. A memory block having a lowest estimated valid page count or having an estimated valid page count having a maintaining tendency is selected as the victim block or a memory block having the estimated valid page count having a decreasing tendency is excluded from selecting the victim block.

Claims

1. A method of operating a storage device including a memory controller and a non-volatile memory device storing data in memory blocks, the method comprising:

estimating a valid page count (VPC) for each of memory blocks using a neural network model trained based on the VPC for each of the memory blocks;

selecting a victim block from the memory blocks based on an estimated VPC for each of the memory blocks; and

performing garbage collection on the victim block.

2. The method of claim 1, wherein the estimating of the VPC for each of the memory blocks comprises performing an operation in the neural network model based on the VPC for each of the memory blocks at each of at least one past time step and a present time step.

3. The method of claim 2, further comprising:

obtaining training data by performing the operation using the neural network model; and

providing the estimated VPC for each of the memory blocks at a future time step as output data corresponding to the training data.

4. The method of claim 3, further comprising:

collecting internal information associated with prior data processing operations by the storage device; and

providing the internal information to the neural network model.

5. The method of claim 4, wherein the internal information includes at least one of an average number of read input/output (I/O) operations per second, an average number of write I/O operations per second, an average data size of a read request, an average data size of a write request, read throughput, write throughput, and a free block count.

6. The method of claim 1, wherein the selecting of the victim block from the memory blocks comprises selecting, as the victim block, a memory block having a lowest estimated VPC among the memory blocks.

7. The method of claim 1, wherein the selecting of the victim block from the memory blocks comprises excluding a memory block having the estimated VPC having a decreasing tendency from among the memory blocks.

8. The method of claim 1, wherein the selecting of the victim block from the memory blocks comprises selecting, as the victim block, a memory block having an estimated VPC with a maintaining tendency among the memory blocks.

9. The method of claim 1, further comprising:

determining whether a free block count is less than a threshold value, the free block count indicating a number of free blocks among the memory blocks.

10. The method of claim 9, further comprising:

obtaining the free block count by counting the free blocks among the memory blocks when a write request is received.

11. The method of claim 9, further comprising:

obtaining the free block count by periodically counting the free blocks among the memory blocks.

12. The method of claim 1, wherein the performing of the garbage collection on the victim block comprises:

copying data of valid pages from the victim block to a destination block; and

erasing the victim block.

13. A storage device comprising:

a non-volatile memory device including memory blocks storing data; and

a memory controller configured to estimate a valid page count (VPC) for each of the memory blocks using a neural network model and to select a victim block from the memory blocks based on an estimated VPC for each of the memory blocks,

wherein the neural network model is trained based on the VPC for each of the memory blocks.

14. The storage device of claim 13, wherein the memory controller is further configured to perform the garbage collection on the victim block by copying data of valid pages from the victim block to a destination block and erasing the victim block.

15. The storage device of claim 13, wherein the memory controller is further configured to count free blocks among the memory blocks and to perform the garbage collection when a free block count is less than a threshold value, and

the memory controller counts the free blocks periodically or upon receiving a write request.

16. The storage device of claim 13, wherein the memory controller is further configured to provide the VPC as input data of the neural network model for each of the memory blocks at each one of at least one past time step and a present time step, and to obtain the estimated VPC for each of the memory blocks at a future time step as output data of the neural network model.

17. The storage device of claim 13, wherein the memory controller selects, as the victim block, a memory block having a lowest estimated VPC among the memory blocks.

18. A non-volatile memory device comprising:

a memory cell array including memory blocks storing data; and

a control logic block configured to control garbage collection performed on a victim block, the victim block being selected from the memory blocks based on estimated valid page counts provided from a neural network model trained based on valid page counts of the memory blocks.

19. The non-volatile memory device of claim 18, wherein, when a memory block having a lowest estimated valid page count among the memory blocks is selected as the victim block, the control logic block performs the garbage collection by copying data of valid pages from the victim block to a destination block and erasing the victim block.

20. The non-volatile memory device of claim 19, wherein, when a memory block having an estimated valid page count having a maintaining tendency among the memory blocks is selected as the victim block, the control logic block performs the garbage collection by copying data of valid pages from the victim block to a destination block and erasing the victim block.