IP Library › Granted Patent US 10,831,860
Granted Patent B2
US 10,831,860 · App. 16/158,056 · Granted Nov 10, 2020

Alignment techniques to match symmetry point as zero-weight point in analog crosspoint arrays

Inventors: Seyoung Kim (White Plains, NY); Hyungjun Kim (Nam, KR); Tayfun Gokmen (Briarcliff Manor, NY); Malte Rasch (Chappaqua, NY)
Assignee: International Business Machines Corporation
G06F17/16
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 10,831,860
App. No.
16/158,056
Filed
Oct 11, 2018
Granted
Nov 10, 2020
Kind
B2
Art Unit
2182
USPC
708/520
Abstract

Zero-shifting techniques in analog crosspoint arrays are provided. In one aspect, an analog array-based vector-matrix multiplication includes: a weight array connected to a reference array, each including a crossbar array having a set of conductive row wires and a set of conductive column wires intersecting the set of conductive row wires, and optimizable crosspoint devices at intersections of the set of conductive column wires and the set of conductive row wires. A method for analog array-based vector-matrix computing is also provided that includes: applying repeated voltage pulses to the crosspoint devices in the weight array until all of the crosspoint devices in the weight array converge to their own symmetry point; and copying conductance values for each crosspoint device from the weight array to the reference array.

Claims (34)

1. An analog array-based vector-matrix, comprising:

a weight array connected to a reference array, each comprising a crossbar array having a set of conductive row wires and a set of conductive column wires intersecting the set of conductive row wires, and optimizable crosspoint devices at intersections of the set of conductive column wires and the set of conductive row wires in each of the weight array and the reference array, wherein all of the optimizable crosspoint devices in the weight array are at their own symmetry point.

2. The analog array-based vector-matrix of claim 1 , wherein conductance values for each optimizable crosspoint device from the weight array are copied to the reference array.

3. The analog array-based vector-matrix of claim 1 , wherein the optimizable crosspoint devices comprise resistive processing units (RPUs).

4. The analog array-based vector-matrix of claim 1 , further comprising:

a first voltage pulse generator connected to the set of conductive row wires in the weight array; and

a second voltage pulse generator connected to the set of conductive column wires in the weight array.

5. The analog array-based vector-matrix of claim 1 , further comprising:

an analog-to-digital converter (ADC) connected, via a current integrator, to the set of conductive column wires in the weight array.

6. An analog array-based vector-matrix, comprising:

a weight array connected to a reference array, each comprising a crossbar array having a set of conductive row wires and a set of conductive column wires intersecting the set of conductive row wires, and optimizable crosspoint devices at intersections of the set of conductive column wires and the set of conductive row wires in each of the weight array and the reference array,

wherein all of the optimizable crosspoint devices in the weight array are at their own symmetry point, and wherein conductance values for each optimizable crosspoint device from the weight array are copied to the reference array.

7. The analog array-based vector-matrix of claim 6 , wherein the optimizable crosspoint devices comprise RPUs.

8. A method for analog array-based vector-matrix computing, the method comprising the steps of:

providing a weight array connected to a reference array, each comprising a crossbar array having a set of conductive row wires and a set of conductive column wires intersecting the set of conductive row wires, and crosspoint devices comprising RPUs at intersections of the set of conductive column wires and the set of conductive row wires in each of the weight array and the reference array;

applying repeated voltage pulses to the crosspoint devices in the weight array until all of the crosspoint devices in the weight array converge to their own symmetry point; and

copying conductance values for each crosspoint device from the weight array to the reference array.

9. The method of claim 8 , wherein the voltage pulses applied to the crosspoint devices in the weight array comprise up and down voltage pulses.

10. The method of claim 9 , wherein the up and down voltage pulses are applied randomly to the crosspoint devices in the weight array.

11. The method of claim 9 , wherein the up and down voltage pulses are applied to the crosspoint devices in the weight array in an alternating manner.

12. The method of claim 8 , wherein the conductance values copied to the reference array comprise zero-weight values.

13. The method of claim 8 , further comprising the step of:

zero-shifting outputs of the weight array by the zero-weight values from the reference array.

14. The method of claim 13 , further comprising the steps of:

applying a same signal to both the weight array and the reference array; and

subtracting the output of the weight array from output of the reference array.

15. A method for analog array-based vector-matrix computing, the method comprising the steps of:

providing a weight array connected to a reference array, each comprising a crossbar array having a set of conductive row wires and a set of conductive column wires intersecting the set of conductive row wires, and crosspoint devices at intersections of the set of conductive column wires and the set of conductive row wires in each of the weight array and the reference array;

applying repeated voltage pulses to the crosspoint devices in the weight array until all of the crosspoint devices in the weight array converge to their own symmetry point;

copying conductance values for each crosspoint device from the weight array to the reference array; and

zero-shifting outputs of the weight array by zero-weight values from the reference array by applying a same signal to both the weight array and the reference array, and subtracting the output of the weight array from output of the reference array.

16. The method of claim 15 , wherein the voltage pulses applied to the crosspoint devices in the weight array comprise up and down voltage pulses.

17. The method of claim 16 , wherein the up and down voltage pulses are applied randomly to the crosspoint devices in the weight array.

18. The method of claim 16 , wherein the up and down voltage pulses are applied to the crosspoint devices in the weight array in an alternating manner.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 31, 2018
From: KIM, SEYOUNG; KIM, HYUNGJUN; GOKMEN, TAYFUN; RASCH, MALTE
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 047875/0398 →
Continuity (1)
Related Publication 20200117699A1 · Apr 16, 2020
Cited By (2)
US 12,254,945 US 12,744,080