IP Library › Granted Patent US 11,481,611
Granted Patent B2
US 11,481,611 · App. 16/178,627 · Granted Oct 25, 2022

RRAM crossbar array structure for multi-task learning

Inventors: Takashi Ando (Tuckahoe, NY); Reinaldo Vega (Mahopac, NY); Hari Mallela (Poughquag, NY)
Assignee: International Business Machines Corporation
G06N3/063G06N3/04G06N3/08
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 11,481,611
App. No.
16/178,627
Granted
Oct 25, 2022
Kind
B2
Abstract

Provided are embodiments of a multi-task learning system with hardware acceleration that includes a resistive random access memory crossbar array. Aspects of the invention includes an input layer that has one or more input layer nodes for performing one or more tasks of the multi-task learning system, a hidden layer that has one or more hidden layer nodes, and a shared hidden layer that has one or more shared hidden layer nodes which represent a parameter, wherein the shared hidden layer nodes are coupled to each of the one or more hidden layer nodes of the hidden layer.

Claims (7)

1. A resistive random access memory (RRAM) crossbar array in a multi-task learning system with hardware acceleration, the RRAM crossbar array comprising:

one or more cross point devices coupled to a first pair of metal lines, wherein the first pair of metal lines is associated with a first task of the multi-task learning system, wherein a resistivity of each cross point device of the one or more cross point devices represent a parameter of a connection in the multi-task learning system; and

a second pair of metal lines is associated with a second task of the multi-task learning system, wherein the first pair of metal lines is coupled to the second pair of metal lines,

wherein the RRAM crossbar array is coupled to a first top layer and a first bottom layer coupled to the RRAM crossbar array via the first pair of metal lines; and a second top layer and a second bottom layer coupled to the RRAM crossbar array via the second pair of metal lines, wherein the first top layer, the first bottom layer, the second top layer, and the second bottom layer access weights from the RRAM crossbar array.

2. The RRAM crossbar array of claim 1 , wherein the first pair of metal lines is coupled to the second pair of metal lines by one or more vias.

3. The RRAM crossbar array of claim 1 , wherein the parameter is a weight or bias associated with the one or more hidden layer nodes of the one or more hidden layer nodes.

4. The RRAM crossbar array of claim 1 , wherein the RRAM crossbar array is a shared hidden layer, wherein the shared hidden layer comprises the one or more cross point devices, wherein the shared hidden layer is coupled to an input layer and one or more hidden layers comprising one or more nodes.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 2, 2018
From: ANDO, TAKASHI; VEGA, REINALDO; MALLELA, HARI
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 047390/0134 →
Continuity (1)
Related Publication 20200143233A1 · May 7, 2020
Cited By (1)
US 12,610,753