IP Library Granted Patent US 12,333,419
Granted Patent B2
US 12,333,419 · App. 18/757,335 · Granted Jun 17, 2025

Neural processing device and transaction tracking method thereof

Inventors: Wongyu Shin (Seongnam-si, KR); Kyeongryeol Bong (Seongnam-si, KR)
Assignee: Rebellions Inc.
G06N3/063G06F12/00
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 12,333,419
App. No.
18/757,335
Granted
Jun 17, 2025
Kind
B2
Abstract

A neural processing device and transaction tracking method thereof are provided. The neural processing device comprises a first set of a plurality of neural cores, a shared memory shared by the first set of the plurality of neural cores, and a programmable hardware transactional memory (PHTM) configured to receive a memory access request directed to the shared memory from the first set of the plurality of neural cores and configured to commit or buffer the memory access request.

Claims (46)

1. A neural processing device comprising:

a plurality of neural cores;

a shared memory shared by the plurality of neural cores; and

at least one transactional memory,

wherein the at least one transactional memory is configured to:

receive a memory access request directed to the shared memory from at least one of the plurality of neural cores,

receive a memory access scenario for a plurality of memory access operation groups, and

perform at least one of buffer the received memory access request and commit the received memory access request, in response to a determination of whether the received memory access request belongs to a current memory access operation group among the plurality of memory access operation groups.

2. The neural processing device of claim 1 , wherein the memory access scenario indicates group numbers, memory access types, service orders, and numbers of memory accesses, each of the group numbers is associated with a respective one of the plurality of memory access operation groups, each of the memory access types is associated with a respective one of the plurality of memory access operation groups, each of the service orders is associated with a respective one of the plurality of memory access operation groups, each of the numbers of memory accesses is associated with a respective one of the plurality of memory access operation groups.

3. The neural processing device of claim 2 , the at least one transactional memory is further configured to determine one of the plurality of memory access operation groups defined by the memory access scenario as the current memory access operation group according to the service orders defined by the memory access scenario.

4. The neural processing device of claim 1 , wherein the plurality of the neural cores comprises:

a first set of a plurality of neural cores; and

a second set of a plurality of neural cores that are different from the first set of the plurality of neural cores, and

the at least one transactional memory comprises:

a first transactional memory configured to receive memory access requests from the first set of the plurality of neural cores; and

a second transactional memory configured to receive memory access requests from the second set of the plurality of neural cores.

5. The neural processing device of claim 4 , wherein the at least one transactional memory further comprises:

a third transactional memory configured to receive memory access requests from neural cores including the first set of the plurality of neural cores and the second set of the plurality of neural cores.

6. The neural processing device of claim 4 , further comprising:

an L2 sync path configured to transmit synchronization signals received from neural cores including the first set of the plurality of neural cores and the second set of the plurality of neural cores.

7. The neural processing device of claim 6 , wherein the L2 sync path performs many-to-many connections among the neural cores including the first set of the plurality of neural cores and the second set of the plurality of neural cores.

8. The neural processing device of claim 6 , wherein the L2 sync path performs a one-to-one connection among the neural cores including the first set of the plurality of neural cores and the second set of the plurality of neural cores.

9. The neural processing device of claim 8 , wherein the L2 sync path is a ring-shaped interconnection.

10. The neural processing device of claim 1 , wherein each transactional memory of the at least one transactional memory comprises:

one or more transaction regions that commits or buffers memory access requests;

and a non-transaction region that does not track memory access requests.

11. The neural processing device of claim 10 , wherein an address of a transaction region of the one or more transaction regions of a first transactional memory of the at least one transactional memory is different from an address of another transaction region of the one or more transaction regions of the first transactional memory of the at least one transactional memory.

12. The neural processing device of claim 11 , wherein a size of a transaction region of the one or more transaction regions is different from a size of another transaction region of the one or more transaction regions.

13. The neural processing device of claim 1 , wherein each transactional memory of the at least one transactional memory is further configured to:

buffer the received memory access request in response to a determination that the received memory access request belongs to one or more memory access operation groups following the current memory access operation group.

14. The neural processing device of claim 13 , wherein each transactional memory of the at least one transactional memory is further configured to:

commit the received memory access request in response to a determination that the received memory access request belongs to the current memory access operation group and another memory access request is not being processed, and

buffer the received memory access request in response to a determination that the received memory access request belongs to the current memory access operation group and the another memory access request is being processed.

15. A transaction tracking method of a neural processing device including at least one transactional memory, comprising:

receiving a memory access scenario for a plurality of memory access operation groups; receiving a memory access request; and

performing at least one of buffering the received memory access request and committing the received memory access request, in response to a determination of whether the received memory access request belongs to a current memory access operation group among the plurality of memory access operation groups.

16. The transaction tracking method of a neural processing device of claim 15 , wherein the memory access scenario indicates group numbers, memory access types, service orders, and numbers of memory accesses, each of the group numbers is associated with a respective one of the plurality of memory access operation groups, each of the memory access types is associated with a respective one of the plurality of memory access operation groups, each of the service orders is associated with a respective one of the plurality of memory access operation groups, each of the numbers of memory accesses is associated with a respective one of the plurality of memory access operation groups.

17. The transaction tracking method of the neural processing device of claim 15 , further comprises:

starting one of the plurality of memory access operation groups defined by the memory access scenario as the current memory access operation group according to the service orders defined by the memory access scenario.

18. The transaction tracking method of the neural processing device of claim 15 , wherein the performing at least one of the buffering the received memory access request and the committing the received memory access request comprises:

determining whether another memory access request is being processed; and

committing the received memory access request in response to the determination that the another memory access request is not being processed.

19. The transaction tracking method of the neural processing device of claim 18 , wherein the performing at least one of the buffering the received memory access request and the committing the received memory access request further comprises:

buffering the received memory access request in response to the determination that the another memory access request is being processed.

20. The transaction tracking method of the neural processing device of claim 15 , further comprising:

buffering the received memory access request in response to the determination that the received memory access request belongs to a memory access operation group following the current memory access operation group.

Assignments (1)
MERGER AND CHANGE OF NAME Recorded May 22, 2025
From: REBELLIONS INC.; SAPEON KOREA INC.
To: REBELLIONS INC.
Reel/Frame 071355/0107 →
Priority Claims (1)
KR 10-2021-0192183 · Dec 30, 2021 · national
Continuity (2)
Continuation 17938027 · Oct 4, 2022
Related Publication 20240346300A1 · Oct 17, 2024
References Cited (14)
US 9720819B2 · Anderson · 2017 [cited by examiner]
US 11593273B2 · Patel · 2023 [cited by examiner]
US 20070198781A1 · Dice · 2007 [cited by examiner]
US 20110055493A1 · Kottapalli et al. · 2011 [cited by applicant]
US 20110179230A1 · Chou · 2011 [cited by examiner]
US 20130160028A1 · Black · 2013 [cited by applicant]
US 20150212851A1 · Busaba · 2015 [cited by examiner]
US 20190243651A1 · Venkataramani et al. · 2019 [cited by applicant]
US 20200073830A1 · Verrilli et al. · 2020 [cited by applicant]
KR 102258566B1 · 2021 [cited by applicant]
Avni, Hillel et al., Hardware Transactions in Nonvolatile Memory, 2015, Springer-Verlag Berlin Heidelberg, Y. Moses (Ed.): DISC 2015, LNCS 9363, pp. 617-630 (Year: 2015). [cited by examiner]
Avni, Hillel et al. (2015). “Hardware Transactions in Nonvolatile Memory”, Lecture Notes in Computer Science. In: Moses, Y. (eds) Distributed Computing. DISC 2015. vol 9363. p. 617-630. Springer, Berlin, Heidelberg. DOI… [cited by applicant]
Extended European Search Report for EP 22213511.3 by European Patent Office dated Apr. 24, 2023. [cited by applicant]
Office Action for KR 10-2021-0192183 by Korean Intellectual Property Office dated Mar. 25, 2024. [cited by applicant]