Granted Patent B2

US 10,732,871 · App. 15/918,019 · Granted Aug 4, 2020

Abstract

Claims (45)

1. A method of transitioning between a sleep mode for a storage device to reduce power consumption and to increase responsiveness, comprising:

collecting one or more recent parameters related to host-storage device workload, wherein the collected one or more recent parameters comprises a command queue depth;

correlating the host-storage device workload to project a next host idle time; and determining a transition between a storage sleep mode.

2. The method of claim 1 , further comprising collecting one or more historical parameters related to the host-storage device workload.

3. The method of claim 2 , wherein the host-storage device workload is correlated by comparing the collected one or more recent parameters and the collected one or more historical parameters.

4. The method of claim 1 , wherein the collected one or more recent parameters further comprises a command type selected from a group consisting of random read command, random write command, sequential read command, sequential write command, and flush command, and administrative command.

5. The method of claim 1 , wherein the collected one or more recent parameters further comprises an accessed logical block addresses range.

6. A storage device, comprising:

a non-volatile memory; and

a controller coupled to the non-volatile memory, the controller comprising a host interface and a neural network, the neural network operable to:

receive one or more parameter inputs of host storage-device workload; and

output a sleep mode projection for the storage device based upon the one or more parameter inputs.

7. The storage device of claim 6 , further comprising a neural network update module.

8. The storage device of claim 7 , wherein the neural network update module is operable to determine an accuracy of the sleep mode projection and to update the neural network.

9. The storage device of claim 7 , wherein the neural network update module is operable to update the neural network by updating a weight value between nodes of the neural network.

10. The storage device of claim 7 , wherein the neural network is operable to update the neural network during a host idle time.

11. The storage device of claim 7 , wherein the neural network update module is operable to determine an accuracy of the sleep mode projection by determining if the sleep mode projection accurately projected a host idle time.

12. A method of transitioning between a sleep mode for a storage device to reduce power consumption and to increase responsiveness, the storage device comprising a controller having a neural network, the method comprising:

offline training of the neural network;

determining a sleep mode projection using the neural network during online use of the storage device; and

online training of the neural network by determining if the sleep mode projection is accurate and by updating the neural network.

13. The method of claim 12 , wherein offline training comprises supervised learning.

14. The method of claim 12 , wherein online training comprises reinforcement learning.

15. The method of claim 12 , furthering comprising updating the neural network during a host idle time.

16. The method of claim 15 , wherein the host idle time is a projected host idle time.

17. The method of claim 15 , wherein updating the neural network adapts the neural network to the online use of a user.

18. A storage device, comprising:

a non-volatile memory means;

an interface means for interfacing with a host; and

a neural network controller means for determining a sleep mode projection, the neural network controller means coupled to the non-volatile memory means and the interface means.

19. The storage device of claim 18 , wherein the neural network controller means is operable to collect historical parameters related to past host-storage device workload.

20. The storage device of claim 18 , wherein the neural network controller means is operable to collect recent parameters related to recent host-storage device workload.

21. The storage device of claim 18 , wherein the neural network controller means is operable to collect an input from a sensor selected from a group consisting of a storage device sensor, a host sensor, or a sensor connected to the host.

22. The storage device of claim 18 , wherein the neural network controller means is operable to tailor the sleep mode projection based upon an online use of the storage device.

23. The storage device of claim 18 , wherein the neural network controller means is operable to determine a sleep mode projection by correlating a workload between the host and the storage device.

24. A method of transitioning between a sleep mode for a storage device to reduce power consumption and to increase responsiveness, comprising:

correlating one or more parameters of a host-storage device workload to project a next host idle time using a neural network;

determining a sleep mode projection based upon the projected next host idle time;

determining whether the sleep mode projection was accurate; and

modifying the neural network based upon the determination of whether the sleep mode projection was accurate.

25. The method of claim 24 , further comprising determining a window to modify the neural network.

26. The method of claim 24 , wherein the one or more parameters is selected from a group consisting of idle state triggers, sleep mode triggers, command LBA ranges, command types, command queue contents, storage device operational states, storage device register contents; and sensors.

27. The method of claim 26 , wherein correlating the one or more parameters comprises correlating one or more parameters from a historical host-storage device workload and one or more parameters from a recent host-storage device workload.

28. The method of claim 27 , wherein the neural network is further modified based upon the recent host-storage device workload.

29. The method of claim 28 , wherein the recent host-storage device workload is from online use.

Assignments (10)