Switched networks
OwnerYannis Tsividisinventor
Life of this patent
Estimating the term…
Abstract
Techniques are described for making electronic neural networks with variable synapse strengths using the variability of transistor conductance or transconductance with gate bias and utilizing capacitors for the temporary storage of the gate bias used for control.
Claims9
A neural network comprising a plurality of neurons each having one or more synapse circuit means each comprising a voltage-variable resistive means, each voltage-variable resistive means including a control element, each of said synapse circuit means further comprising a capacitor one of whose plates is connected both to said control element and by way of a switch to a terminal of a variable control voltage source.
A neural network including a plurality of synapse circuit means, each of which includes a pair of first and second MOS transistors, each MOS transistor having a source, a drain and a gate, the drains of the first transistors of each pair being connected to a first load node and the drains of the second transistors of each pair being connected to a second load node, a plurality of MOS control transistors each having its source/drain branch connected between sources of a respective pair of first and second transistors of each of said synapse circuit means and a point of reference potential, and including a capacitor in its gate/source branch and its gate being connected to a control voltage source by way of a separate switch, and load means connected between a voltage supply bus and the first and second load nodes.
Dependencies are inferred from the claim text.