US 11,816,558 B2
Integrated circuit designs for reservoir computing and machine learning
Daniel Lathrop, University Park, MD (US); Itamar Shani, Rockville, MD (US); Peter Megson, Washington, DC (US); Alessandro Restelli, Greenbelt, MD (US); and Anthony Robert Mautino, Damascus, MD (US)
Assigned to UNIVERSITY OF MARYLAND, COLLEGE PARK, College Park, MD (US)
Appl. No. 16/613,990
Filed by UNIVERSITY OF MARYLAND, COLLEGE PARK, College Park, MD (US)
PCT Filed May 16, 2018, PCT No. PCT/US2018/032902
§ 371(c)(1), (2) Date Nov. 15, 2019,
PCT Pub. No. WO2018/213399, PCT Pub. Date Nov. 22, 2018.
Claims priority of provisional application 62/662,119, filed on Apr. 24, 2018.
Claims priority of provisional application 62/555,511, filed on Sep. 7, 2017.
Claims priority of provisional application 62/506,951, filed on May 16, 2017.
Prior Publication US 2021/0406648 A1, Dec. 30, 2021
Int. Cl. G06N 5/00 (2023.01); G06N 7/00 (2023.01); G06N 10/00 (2022.01); G06N 3/08 (2023.01); G06N 3/063 (2023.01); G06N 3/044 (2023.01)
CPC G06N 3/063 (2013.01) [G06N 3/08 (2013.01); G06N 3/044 (2023.01)] 20 Claims
OG exemplary drawing
 
1. A device, comprising:
a set of input channels;
a set of output channels;
a plurality of reservoir nodes, the plurality of reservoir nodes including a first set of input nodes coupled to the set of input channels, and a second set of variable-weight output nodes coupled to the set of output channels, each reservoir node in the of the plurality of reservoir nodes comprising a component configured to perform a non-linear operation on a respective set of input signals and output a set of output signals, the second set of variable weight output nodes each being configured to input an analog input signal and provide an analog output signal;
a set of interconnections, each of which couples together two or more of the reservoir nodes; and
a set of weighting units, each of which is coupled to at least one variable-weight output node in the second set of output nodes and configured for weighting the output signals therefrom with a continuously variable weighting signal based on a present node state of the at least one variable weight output node.