# US 2022/0035569 A1 · Ordering reads to limit collisions in a non-volatile memory (nvm)

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0035569 A1
- Document ID: US20220035569A1
- Status: Published application
- Kind: A1
- Application number: 17/443,554
- Filed: 2021-07-27
- Published: 2022-02-03
- Earliest claimed: 2020-07-31
- Assignees: Seagate Technology LLC
- Applicants: Seagate Technology LLC
- Inventors: Christopher Smith
- Web page: https://patentgrep.com/patent/US20220035569A1

## CPC

- G06F 3/0659: {Command handling arrangements, e.g. command buffers, queues, command scheduling}
- G06F 3/0611: {in relation to response time}
- G06F 3/0679: {Non-volatile semiconductor memory device, e.g. flash memory, one time programmable memory [OTP]}
- G06F 12/0804: with main memory updating (G06F12/0806 takes precedence)
- G06F 12/0868: Data transfer between cache memory and other subsystems, e.g. storage devices or host systems
- G06F 12/0875: with dedicated cache, e.g. instruction or stack
- G06F 2212/1016: Performance improvement
- G06F 2212/1024: Latency reduction
- G06F 2212/2022: Flash memory
- G06F 2212/214: Solid state disk
- G06F 2212/261: Storage comprising a plurality of storage devices
- G06F 2212/502: using adaptive policy
- G06F 2212/7203: Temporary buffering, e.g. using volatile buffer or dedicated buffer blocks
- G06F 2212/7208: Multiple device management, e.g. distributing data over multiple flash devices
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Method and apparatus for managing data in a storage device, such as a solid-state drive (SSD). In some embodiments, a data storage device includes a main non-volatile memory (NVM), a host command queue that lists pending host read and host write commands, and a write cache which temporarily stores write data sets pending transfer to the NVM responsive to execution of the associated host write commands in the host command queue. A collision prediction circuit predicts a rate of future collisions involving the cached write data sets. A storage manager directs storage of the write data sets to a first target location responsive to the rate of future collisions being at a first level, and directs storage of the write data sets to a different, second target location responsive to the rate of future collisions being at a different, second level.

## Claims

### Claim 1

An apparatus comprising:
a non-volatile memory (NVM);
a host command queue that lists pending host read and host write commands to transfer data between the NVM and a host;
a write cache which temporarily stores write data sets pending transfer to the NVM responsive to execution of the associated host write commands in the host command queue;
a collision prediction circuit configured to predict a rate of future collisions between the pending host read commands in the host command queue; and
a storage manager configured to operate in a first mode to direct storage of the write data sets from the write cache to a first target location responsive to the rate of future collisions being at a first level, and to operation in a second mode to direct storage of the write data sets from the write cache to a different, second target location responsive to the rate of future collisions being at a different, second level.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein the rate of future collisions is based on history data from previous commands issued by the host.

### Claim 3 (depends on 1)

The apparatus of claim 1, wherein the rate of future collisions is based on predicted requests, in the near future, for the write data sets.

### Claim 4 (depends on 1)

The apparatus of claim 1, wherein the rate of future collisions is further based on pending background read requests in a background command queue.

### Claim 5 (depends on 1)

The apparatus of claim 1, wherein the NVM is divided into a plurality of garbage collection units (GCUs), each separately allocated and erased as a unit.

### Claim 6 (depends on 5)

The apparatus with claim 5, wherein the first target location in the NVM comprises a first GCU and the second target location in the NVM comprises a second GCU, the first and second GCUs arranged on different combinations of semiconductor dies and corresponding channels.

### Claim 7 (depends on 1)

The apparatus of claim 1, wherein the NVM comprises a NAND flash memory of a solid-state drive (SSD).

### Claim 8 (depends on 1)

The apparatus of claim 1, wherein the NVM comprises a rotatable data recording medium.

### Claim 9 (depends on 1)

The apparatus of claim 1, wherein the collision prediction circuit comprises a resource manager configured to identify resources required for each of the sets of writeback data in the write queue, a history tracker that maintains, as a data structure in a memory, history data associated with recent host commands, a predictor circuit that generates an estimated read collision rate responsive to the resource manager and the history tracker, and an observer circuit which subsequently examines a rate at which read collisions occur during subsequent accesses of the sets of writeback data written to the NVM.

### Claim 10 (depends on 1)

The apparatus of claim 1, wherein only a single set of each set of writeback data are written to the NVM.

### Claim 11

A solid-state drive (SSD), comprising:
a flash memory arranged as a main memory store for a host device;
a controller circuit configured to manage data transfers between the flash memory and the host device, the controller circuit configured to manage a host command queue that lists pending host read commands and host write commands to transfer data between the flash memory and the host device, and a non-volatile write cache which temporarily stores write data sets pending transfer to the flash memory responsive to execution of the associated host write commands in the host command queue;
a collision prediction circuit configured to predict a rate of future collisions between the pending host read commands in the host command queue; and
a storage manager circuit configured to operate in a first mode to direct storage of the write data sets from the write cache to a first target location responsive to the rate of future collisions being at a first level, and to operation in a second mode to direct storage of the write data sets from the write cache to a different, second target location responsive to the rate of future collisions being at a different, second level.

### Claim 12 (depends on 11)

The SSD of claim 11, wherein the rate of future collisions is based on history data from previous host read commands and host write commands issued by the host.

### Claim 13 (depends on 11)

The SSD of claim 11, wherein the rate of future collisions is based on predicted requests, in the near future, for the write data sets in the write cache.

### Claim 14 (depends on 11)

The SSD of claim 11, wherein the rate of future collisions is based on pending background read requests in a background command queue managed by the controller circuit.

### Claim 15 (depends on 11)

The SSD of claim 11, wherein the collision prediction circuit comprises a resource manager configured to identify resources required for each of the sets of writeback data in the write queue, a history tracker that maintains, as a data structure in a memory, history data associated with recent host commands, a predictor circuit that generates an estimated read collision rate responsive to the resource manager and the history tracker, and an observer circuit which subsequently examines a rate at which read collisions occur during subsequent accesses of the sets of writeback data written to the NVM.

### Claim 16

A method comprising:
accumulating pending host read commands and host write commands in a queue to transfer data between a non-volatile memory (NVM) and a host device;
temporarily storing, in a write cache, write data sets associated with the host write commands;
predicting a rate of future collisions between the pending host read commands in the host command queue; and
directing the write data sets in the write cache to different first or second locations in the NVM responsive to the predicted rate of future collisions.

### Claim 17 (depends on 16)

The method of claim 16, wherein the directing step comprises operating a data storage device having the NVM in a first mode to direct storage of the write data sets from the write cache to a first target location responsive to the rate of future collisions being at a first level, and to operate the data storage device in a second mode to direct storage of the write data sets from the write cache to a different, second target location responsive to the rate of future collisions being at a different, second level.

### Claim 18 (depends on 17)

The method of claim 17, wherein the NVM is divided into a plurality of garbage collection units (GCUs), each separately allocated and erased as a unit, wherein the first target location in the NVM comprises a first GCU and the second target location in the NVM comprises a second GCU, the first and second GCUs arranged on different combinations of semiconductor dies and corresponding channels.

### Claim 19 (depends on 16)

The method of claim 16, wherein the rate of future collisions is based on at least a selected one of history data from previous host read commands and previous host write commands issued by the host, predicted requests in the near future for the write data sets, or pending background read requests in a background command queue.

### Claim 20 (depends on 16)

The method of claim 16, wherein the NVM comprises a NAND flash memory of a solid-state drive (SSD).
