IP Library Granted Patent US 12,423,568
Granted Patent B2
US 12,423,568 · App. 18/510,755 · Granted Sep 23, 2025

Counter based resistive processing unit for programmable and reconfigurable artificial-neural-networks

Inventors: Siyuranga Koswatta (Carmel, NY); Yulong Li (Westchester, NY); Paul M. Solomon (Westchester, NY)
Assignee: International Business Machines Corporation
G06N3/065G06N3/08H03K19/20
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 12,423,568
App. No.
18/510,755
Granted
Sep 23, 2025
Kind
B2
Abstract

Technical solutions are described for storing weight in a crosspoint device of a resistive processing unit (RPU) array. An example system includes a crosspoint array, wherein each array node represents a connection between neurons of the neural network, and wherein each node stores a weight assigned to the node. The crosspoint array includes a crosspoint device at each node. The crosspoint device includes a counter that has multiple single bit counters, and states of the counters represent the weight to be stored at the crosspoint device. Further, the crosspoint device includes a resistor device that has multiple resistive circuits, and each resistive circuit is associated with a respective counter from the counters. The resistive circuits are activated or deactivated according to a state of the associated counter, and an electrical conductance of the resistor device is adjusted based at least in part on the resistive circuits that are activated.

Claims (17)

1. A method for implementing a neural network, the method comprising:

selecting, by a global controller, a matrix to be loaded in a crosspoint array, the matrix corresponding to a connection between two layers of the neural network;

in response, loading, by a local controller at a crosspoint in the crosspoint array, a weight value in a crosspoint device associated with the crosspoint, the weight value assigned to the crosspoint for the selected matrix; and

in response, adjusting, by the crosspoint device, a conductance of a resistor device associated with the crosspoint, the conductance corresponding to the weight value loaded into the crosspoint device;

wherein the crosspoint device comprises:

a counter comprising a plurality of single bit counters, states of the single bit counters representing the weight to be stored at the crosspoint device; and

a resistor device comprising a plurality of resistive circuits, each resistive circuit associated with a respective single bit counter from the plurality of single bit counters, the resistive circuits activated or deactivated according to a state of the associated single bit counter.

2. The method of claim 1 , wherein an electrical conductance of the resistor device is adjusted based at least in part on the resistive circuits that are activated.

3. The method of claim 1 , wherein the global controller is configured to adjust values stored at each crosspoint device in the crosspoint array.

4. The method of claim 1 , wherein the method further comprises reading, from a local memory, a current state of each single bit counter of a plurality of single bit counters in the crosspoint device.

5. The method of claim 4 , wherein the method further comprises updating the local memory to indicate an updated state of each single bit counter of the plurality of single bit counters.

6. The method of claim 1 , wherein the plurality of resistive circuits in the resistor device of the crosspoint device include a quadratically increasing resistance, a first resistive circuit having a predetermined resistance and each further successive resistive circuit having a resistance that is twice of a previous resistive circuit.

7. The method of claim 1 , wherein the plurality of resistive circuits are field effect transistors (FETs).

8. The method of claim 1 , wherein the plurality of resistive circuits are a resistor ladder, each of the resistive circuit comprising:

a logic gate; and

a series of resistors; and

wherein, the logic gate is activated based at least in part on the state of the corresponding single bit counter.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 16, 2023
From: KOSWATTA, SIYURANGA; LI, YULONG; SOLOMON, PAUL M.
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 065581/0371 →
Continuity (3)
Division 17518629 · Nov 4, 2021
Division 15840322 · Dec 13, 2017
Related Publication 20240086697A1 · Mar 14, 2024
References Cited (56)
US 2715656A · Andrews, Jr. · 1955 [cited by applicant]
US 5202956A · Mashiko · 1993 [cited by applicant]
US 5293457A · Arima · 1994 [cited by applicant]
US 5396581A · Mashiko · 1995 [cited by applicant]
US 6256247B1 · Perner · 2001 [cited by applicant]
US 6876989B2 · Shi et al. · 2005 [cited by applicant]
US 8009455B2 · Lowrey et al. · 2011 [cited by applicant]
US 8542521B2 · Hamada · 2013 [cited by applicant]
US 9412446B1 · Lohn et al. · 2016 [cited by applicant]
US 9466362B2 · Yu et al. · 2016 [cited by applicant]
US 9646243B1 · Gokmen · 2017 [cited by applicant]
US 9659249B1 · Copel · 2017 [cited by applicant]
US 9715656B1 · Gokmen · 2017 [cited by applicant]
US 9779355B1 · Leobandung · 2017 [cited by applicant]
US 9812196B2 · Perner · 2017 [cited by applicant]
US 10127494B1 · Cantin et al. · 2018 [cited by applicant]
US 10482929B2 · Li · 2019 [cited by examiner]
US 11222259B2 · Koswatta et al. · 2022 [cited by applicant]
US 20050078536A1 · Perner et al. · 2005 [cited by applicant]
US 20050083748A1 · Lemus et al. · 2005 [cited by applicant]
US 20050102576A1 · Perner et al. · 2005 [cited by applicant]
US 20080212382A1 · Mouttet · 2008 [cited by examiner]
US 20110119215A1 · Elmegreen · 2011 [cited by examiner]
US 20120173471A1 · Ananthanarayanan et al. · 2012 [cited by applicant]
US 20120259804A1 · Brezzo et al. · 2012 [cited by applicant]
US 20130325775A1 · Sinyavskiy et al. · 2013 [cited by applicant]
US 20140114893A1 · Arthur et al. · 2014 [cited by applicant]
US 20150255157A1 · Ikeda et al. · 2015 [cited by applicant]
US 20150278682A1 · Saxena · 2015 [cited by applicant]
US 20160049195A1 · Yu et al. · 2016 [cited by applicant]
US 20160336064A1 · Seo · 2016 [cited by examiner]
US 20170040054A1 · Friedman et al. · 2017 [cited by applicant]
US 20170091621A1 · Gokmen et al. · 2017 [cited by applicant]
US 20170109626A1 · Gokmen et al. · 2017 [cited by applicant]
US 20170124025A1 · Gokmen · 2017 [cited by applicant]
US 20170243108A1 · Ritter et al. · 2017 [cited by applicant]
US 20190180174A1 · Koswatta et al. · 2019 [cited by applicant]
US 20190325291A1 · Gokmen · 2019 [cited by examiner]
US 20220058474A1 · Koswatta et al. · 2022 [cited by applicant]
CN 106158017A · 2016 [cited by applicant]
JP H02236659A · 1990 [cited by applicant]
JP H0380379A · 1991 [cited by applicant]
JP H03223982A · 1991 [cited by applicant]
JP H06215163A · 1994 [cited by applicant]
JP H0884078A · 1996 [cited by applicant]
WO 2009142828A1 · 2009 [cited by applicant]
WO 2017136100A1 · 2017 [cited by applicant]
WO 2017155544A1 · 2017 [cited by applicant]
Chi et al., “Processing-in-Memory in ReRAM-based Main Memory,” SEAL-lab Technical Report—No. 2015-001, Apr. 29, 2016, pp. 1-12. [cited by applicant]
Chinese Office Action; Application No. 201880077961.9; Date of mailing: Apr. 22, 2023; 9 pages. [cited by applicant]
Chinese Office Action; Application No. 201880077961.9; Date of mailing: Aug. 25, 2023; 6 pages. [cited by applicant]
German Office Action; Application No. 112018005726.7; Filing Date: May 28, 2020; Date of mailing: Dec. 21, 2021; 4 pages. [cited by applicant]
Gokmen et al., “Acceleration of Deep Neural Network Training with Resistive Cross Point Devices: Design Considerations,” Frontiers in Neuroscience, DOI: 10.3389/fnins.2016.00333, Jul. 21, 2016, pp. 1-13. [cited by applicant]
Japanese Office Action; Application No. 2020-531461; Date of Drafting: Jun. 16, 2022; 7 pages. [cited by applicant]
Kim et al., “Analog CMOS-based Resistive Processing Unit for Deep Neural Network Training,” IEEE 60th International Midwest Symposium on Circuits and Systems (MWSCAS), Aug. 6-9, 2017, 4 pages. [cited by applicant]
List of IBM Patents or Patent Applications Treated as Related (Appendix P); Date Filed: Nov. 16, 2023; 2 pages. [cited by applicant]