Memory lookup computing mechanisms

Patent metadata

Field	Value
Publication number	US-2023101422-A1
Application number	US-202218060276-A
Country	US
Kind code	A1
Filing date	Nov 30, 2022
Priority date	Dec 26, 2017
Publication date	Mar 30, 2023
Grant date	—

Abstract

According to some example embodiments of the present disclosure, in a method for a memory lookup mechanism in a high-bandwidth memory system, the method includes: using a memory die to conduct a multiplication operation using a lookup table (LUT) methodology by accessing a LUT, which includes floating point operation results, stored on the memory die; sending, by the memory die, a result of the multiplication operation to a logic die including a processor and a buffer; and conducting, by the logic die, a matrix multiplication operation using computation units.

First claim

What is claimed is:

1 . A storage system comprising:
a first device configured to store matrix data; and a second device configured to:
receive intermediate matrix multiplication result data based on the matrix data from the first device; and
output a matrix multiplication result based on the intermediate matrix multiplication result data.

2 . The storage system of claim 1 , wherein the first device supports in-memory computing operations.

3 . The storage system of claim 1 , wherein the first device further stores a lookup table, and wherein the first device is further configured to output the intermediate matrix multiplication result data based on the lookup table.

4 . The storage system of claim 1 , wherein the first device corresponds to a first storage device, the storage system further comprising a second storage device distinct from the first storage device.

5 . The storage system of claim 4 , wherein the second storage device comprises dynamic random access memory.

6 . The storage system of claim 1 , wherein the second device is configured to execute an artificial intelligence application based on access to the first device.

7 . A storage device comprising:
storage media configured to store matrix data; and an interface configured to send intermediate matrix multiplication result data, based on the matrix data, to a second device configured to output a matrix multiplication result based on the intermediate matrix multiplication result data.

8 . The storage device of claim 7 , wherein the intermediate matrix data is distinct from the matrix data.

9 . The storage device of claim 7 , wherein the storage media supports in-memory computing operations.

10 . The storage device of claim 7 , wherein the storage media further stores a lookup table, and wherein the storage media is further configured to output the intermediate matrix multiplication result data based on the lookup table.

11 . The storage device of claim 7 , wherein the storage media corresponds to a first storage device, the storage device further comprising a second storage device distinct from the first storage device.

12 . The storage device of claim 11 , wherein the second storage device comprises dynamic random access memory.

13 . The storage device of claim 7 , wherein the second device is configured to execute an artificial intelligence application based on access to the storage media.

14 . A method comprising:
storing matrix data at a storage device; and sending, from the storage device, intermediate matrix multiplication result data, based on the matrix data, to a second device configured to output a matrix multiplication result based on the intermediate matrix multiplication result data.

15 . The method of claim 14 , wherein the storage device supports in-memory computing operations.

16 . The method of claim 14 , further comprising storing at the storage device a lookup table, and wherein the storage device is further configured to output the intermediate matrix multiplication result data based on the lookup table.

17 . The method of claim 14 , wherein the storage device corresponds to a first storage device, and wherein a second storage device is distinct from the first storage device.

18 . The method of claim 17 , wherein the second storage device comprises dynamic random access memory.

19 . The method of claim 14 , wherein the second device is configured to execute an artificial intelligence application based on access to the storage device.

20 . The method of claim 14 , wherein the intermediate matrix data is distinct from the matrix data.

21 . A method comprising:
performing, by a memory device, a first operation using a lookup table (LUT) stored on the memory die; sending, by the memory device, a result of the first operation to a processor device; and performing, by the processor device, a matrix multiplication operation using the result of the first operation performed on the memory device.

22 . The method of claim 21 , wherein the first operation includes a multiplication operation.

Assignees

• Samsung Electronics Co Ltd

Inventors

• Gu Peng
• Malladi Krishna T
• Zheng Hongzhong

Classifications

• G06F9/3004 Primary

  to perform operations on memory · CPC title

• G06F7/4876 Primary

  Multiplying · CPC title

• G06F9/30007

  to perform operations on data operands · CPC title

• G06F17/16

  Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)} · CPC title

• G06F7/00

  Methods or arrangements for processing data by operating upon the order or content of the data handled (logic circuits H03K19/00) · CPC title