IP Library Granted Patent US 12705255
Granted Patent B2
US 12705255 · App. 18/497,250 · Granted Aug 11, 2026

Systems and methods for grouping embedding vectors in a storage device for improved read latency

Inventors: Usman Sajid (Lawrence, KS); Sai Prahladh Padmanabhan (San Jose, CA); Marie Mai Nguyen (Pittsburgh, PA); Rekha Pitchumani (Oak Hill, VA); Yang Seok Ki (Palo Alto, CA)
Assignee: Samsung Electronics Co., Ltd.
G06F16/285G06F16/24552G06F16/24569
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12705255
App. No.
18/497,250
Granted
Aug 11, 2026
Kind
B2
Abstract

A method of grouping embedding vectors in a storage device includes receiving, at a storage device, an inference query request including a first access index associated with a first embedding vector, determining, with a classification model, a first classification of the first access index, based on the first classification, searching for the first embedding vector in a cache of the storage device, receiving, at the storage device, a second access index associated with a second embedding vector, determining, with the classification model, a second classification of the second access index, and based on the second classification, searching for the second embedding vector in a persistent storage of the storage device.

Claims (57)

1 . A method, comprising:

receiving, at a storage device, an inference query request comprising a first access index to locate a first embedding vector, the first embedding vector representing an object for processing by a machine-learning (ML) model;

determining, from a first output of a classification model, a first location of the first embedding vector;

searching for the first embedding vector in the first location;

receiving, at the storage device, a second access index to locate a second embedding vector;

determining, from a second output of the classification model, a second location of the second embedding vector; and

searching for the second embedding vector in the second location.

2 . The method of claim 1 , wherein:

the first output indicates a first access frequency; and

the second output indicates a second access frequency that is less than the first access frequency.

3 . The method of claim 1 , wherein:

the determining, from the first output of the classification model, comprises comparing an access frequency associated with the first access index with a reference access frequency; and

the determining, from the second output of the classification model, comprises comparing an access frequency associated with the second access index with the reference access frequency.

4 . The method of claim 1 , wherein the classification model comprises a machine-learning (ML) grouping model.

5 . The method of claim 4 , further comprising generating, by the machine-learning (ML) grouping model, the first output based on the first access index.

6 . The method of claim 1 , wherein the storage device is configured to communicate with a host via a cache coherent protocol.

7 . The method of claim 1 , wherein the second embedding vector is prevented from being stored in the first location of the storage device based on a reference access frequency.

8 . The method of claim 1 , wherein:

the first location of the first embedding vector comprises a cache; and

the second location of the second embedding vector comprises a persistent storage.

9 . A hardware storage device, comprising:

a first location; and

a second location,

wherein, the hardware storage device is configured to:

receive an inference query request comprising a first access index to locate a first embedding vector, the first embedding vector representing an object for processing by a machine-learning (ML) model;

determine, from a first output of a classification model, to search the first location for the first embedding vector;

search for the first embedding vector in the first location;

receive a second access index to locate a second embedding vector;

determine, from a second output of the classification model, to search the second location for the second embedding vector; and

search for the second embedding vector in the second location.

10 . The hardware storage device of claim 9 , wherein:

the first output indicates a first access frequency; and

the second output indicates a second access frequency that is less than the first access frequency.

11 . The hardware storage device of claim 9 , wherein:

the determining, from the first output of the classification model, comprises comparing an access frequency associated with the first access index with a reference access frequency; and

the determining, from the second output of the classification model, comprises comparing an access frequency associated with the second access index with the reference access frequency.

12 . The hardware storage device of claim 9 , wherein the classification model comprises a machine-learning (ML) grouping model.

13 . The hardware storage device of claim 12 , wherein the hardware storage device is configured to generate the first output from the machine-learning (ML) grouping model based on the first access index.

14 . The hardware storage device of claim 9 , wherein the second embedding vector is prevented from being stored in the first location based on a reference access frequency.

15 . A system, comprising:

a processor; and

a memory storing instructions, which, when executed by the processor, cause the processor to perform:

receiving an inference query request comprising a first access index to locate a first embedding vector, the first embedding vector representing an object for processing by a machine-learning (ML) model;

determining, from a first output of a classification model, a first location of the first embedding vector;

searching for the first embedding vector in the first location;

receiving a second access index to locate a second embedding vector;

determining, from a second output of the classification model, a second location of the second embedding vector; and

searching for the second embedding vector in the second location.

16 . The system of claim 15 , wherein:

the first output indicates a first access frequency; and

the second output indicates a second access frequency that is less than the first access frequency.

17 . The system of claim 15 , wherein:

the determining, from the first output of the classification model, comprises comparing an access frequency associated with the first access index with a reference access frequency; and

the determining, from the second output of the classification model, comprises comparing an access frequency associated with the second access index with the reference access frequency.

18 . The system of claim 15 , wherein the classification model comprises a machine-learning (ML) grouping model.

19 . The system of claim 18 , wherein the machine-learning (ML) grouping model generates the first output based on the first access index.

20 . The system of claim 15 , wherein the second embedding vector is prevented from being stored in the first location based on a reference access frequency.