Granted Patent B2

US 12,450,010 · App. 17/899,407 · Granted Oct 21, 2025

Inventors: Robert Richard Noel Bielby (Placerville, CA); Poorna Kale (Folsom, CA)

Assignee: Lodestar Licensing Group LLC

G06F3/0659G06F3/0616G06F3/0673G06F12/10G06N3/049G06N3/08G06F2212/657

Abstract

Claims (59)

1. A device, comprising:

memory cells; and

a controller configured to:

determine, as a function of time, operating parameters indicative of a data access pattern;

predict, based on the operating parameters as a function of time, an indicator of write frequency in a region of the memory cells;

determine, based on the indicator of write frequency and based on a performance indicator of a data placement scheme for the data access pattern, an optimized operation of wear leveling for the memory cells, wherein the optimized operation comprises combining at least one wear leveling operation with at least one predicted write operation; and

perform the optimized operation of wear leveling for the memory cells.

2. The device of claim 1 , further comprising:

one or more storage media components containing the memory cells; and

an interface;

wherein the controller is further configured to store data into and retrieve data from the one or more storage media components according to commands received in the interface; and

wherein the controller is configured to use an address map to convert between:

logical addresses specified in the commands received in the interface; and

physical addresses of the memory cells in the one or more storage media components.

3. The device of claim 2 , further comprising:

a neural network accelerator;

wherein the controller is configured to use the neural network accelerator to perform computations of an artificial neural network responsive to the operating parameters as a function of time to predict the indicator of write frequency.

4. The device of claim 3 , wherein the controller is further configured to postpone, in response to the indicator of write frequency indicating that data stored in the region is to be modified within a predetermined period of time, copying data from the region during the optimized operation of wear leveling.

5. The device of claim 3 , wherein the region is a first region; and the controller is further configured to select a second region having wearing higher than the first region and copy data from the first region to the second region during the optimized operation of wear leveling.

6. The device of claim 3 , wherein the artificial neural network includes a spiking neural network; and

wherein the artificial neural network is configured to receive write activities in the commands and identify logical address groups having write operations correlated in time.

7. The device of claim 3 , wherein the controller is configured to predict the data placement scheme that identifies placement of multiple selected logical address groups in an erasure block.

8. The device of claim 3 , wherein the region is a first region; and the controller is configured to write, in response to receiving a write command for first data stored in the first region, into a second region, in accordance with the write command and the first data stored in the first region.

9. The device of claim 3 , wherein the controller is further configured to generate training data during usages of the device during a period and train, using the neural network accelerator, the artificial neural network using the training data; and

wherein the training data includes write activities in commands received in the period, the operating parameters in the period, data placement schemes implemented in the address map in the period, and the performance indicator measured during the period for the data placement schemes.

10. The device of claim 3 , wherein the controller is further configured to train, using the neural network accelerator, the artificial neural network using training data including:

the operating parameters in a time period; and

data placement schemes determined for data access patterns recognized from the operating parameters in the time period and optimized for a combined target of write-amplification reduction and wear-leveling.

11. A method, comprising:

determining, by a controller of a device having memory cells and as a function of time, operating parameters indicative of a data access pattern;

predicting, by the controller and based on the operating parameters as a function of time, an indicator of write frequency in a region of the memory cells;

determining, by the controller and based on the indicator of write frequency and on a performance indicator of a data placement scheme for the data access pattern, an optimized operation of wear leveling for the memory cells, wherein the optimized operation comprises combining at least one wear leveling operation with at least one predicted write operation; and

performing, by the controller, the optimized operation of wear leveling for the memory cells.

12. The method of claim 11 , further comprising:

storing, by the controller, data into and retrieve data from one or more storage media components containing the memory cells according to commands received in an interface of the device; and

converting, by the controller using an address map to convert, between:

logical addresses specified in the commands received in the interface; and

physical addresses of the memory cells in the one or more storage media components.

13. The method of claim 12 , further comprising:

performing, by the controller using a neural network accelerator of the device, computations of an artificial neural network responsive to the operating parameters as a function of time to predict the indicator of write frequency.

14. The method of claim 13 , further comprising:

postponing, by the controller and in response to the indicator of write frequency indicating that data stored in the region is to be modified within a predetermined period of time, copying data from the region during the optimized operation of wear leveling.

15. The method of claim 13 , wherein the region is a first region; and the method further comprises:

selecting, by the controller, a second region having wearing higher than the first region to copy data from the first region to the second region during the optimized operation of wear leveling.

16. The method of claim 13 , wherein determining the optimized operation of wear leveling further comprises identifying the performance indicator of the data placement scheme for the data access pattern indicated by the operating parameters.

17. The method of claim 13 , wherein the region is a first region; and the method further comprises:

writing, by the controller and in response to receiving a write command for first data stored in the first region, into a second region, in accordance with the write command and the first data stored in the first region.

18. A non-transitory computer storage medium storing instructions which, when executed in a device having memory cells and a controller, cause the device to perform a method, comprising:

determining, by the controller of the device having the memory cells and as a function of time, operating parameters indicative of a data access pattern;

predicting, by the controller and based on the operating parameters as a function of time, an indicator of write frequency in a region of the memory cells;

determining, by the controller and based on the indicator of write frequency and on a performance indicator of a data placement scheme for the data access pattern, an optimized operation of wear leveling for the memory cells, wherein the optimized operation comprises combining at least one wear leveling operation with at least one predicted write operation; and

performing, by the controller, the optimized operation of wear leveling for the memory cells.

19. The non-transitory computer storage medium of claim 18 , wherein the method further comprises:

generating, by the controller, training data during usages of the device during a period and train, using a neural network accelerator of the device, an artificial neural network using the training data;

wherein the training data includes write activities in commands received in the period, the operating parameters in the period, data placement schemes implemented in an address map in the period, and the performance indicator measured during the period for the data placement schemes.

20. The non-transitory computer storage medium of claim 18 , wherein the method further comprises:

training, by the controller using a neural network accelerator of the device, the artificial neural network using training data including:

the operating parameters in a time period; and

data placement schemes determined for data access patterns recognized from the operating parameters in the time period and optimized for a combined target of write-amplification reduction and wear-leveling.

Assignments (1)

Continuity (2)

References Cited (245)