Spiking neural network

Patent metadata

Field	Value
Publication number	US-10922608-B2
Application number	US-201715452792-A
Country	US
Kind code	B2
Filing date	Mar 8, 2017
Priority date	Mar 8, 2017
Publication date	Feb 16, 2021
Grant date	Feb 16, 2021

Abstract

Broadly speaking, embodiments of the present technique provide a neuron for a spiking neural network, where the neuron is formed of at least one Correlated Electron Random Access Memory (CeRAM) element or Correlated Electron Switch (CES) element.

First claim

What is claimed is:

1. A spiking neuron for a spiking neural network, the spiking neuron comprising:
a correlated electron switch (CES) element for implementing a thresholding function of the spiking neuron;
an accumulator circuit for summing current signals received by the spiking neuron to provide an accumulated current signal;
a further CES element for storing the accumulated current signal as a compliance current;
wherein the CES element stores a threshold current value corresponding to a compliance current, the spiking neuron further including:
a comparator circuit for comparing the accumulated current signal with the threshold current value stored by the CES element and outputting a spike signal if the accumulated current signal is greater than or equal to the threshold current value, the comparator circuit including:
a first mirror circuit for mirroring the accumulated current signal stored in the further CES element; and
a second mirror circuit for mirroring the threshold current value stored by the CES element.

2. The spiking neuron as claimed in claim 1 wherein the CES element is programmed into an initial high impedance state, and the spiking neuron further comprises:
circuitry for applying voltage Vset across the CES element,
wherein the spiking neuron outputs a spike when the accumulated current signal exceeds a threshold current Iset and causes the CES element to switch out of the initial high impedance state.

3. A spiking neuron for a spiking neural network, the spiking neuron comprising:
a correlated electron switch (CES) element for implementing a thresholding function of the spiking neuron
an accumulator circuit for summing current signals received by the spiking neuron to provide an accumulated current signal,
wherein the CES element is programmed into one of a plurality of low impedance states, and the spiking neuron further comprises:
circuitry for applying voltage Vreset across the CES element,
wherein the spiking neuron outputs a spike when the accumulated current signal exceeds a threshold current Ireset, and causes the CES element to switch into a high impedance state upon application of voltage Vreset.

4. The spiking neuron as claimed in claim 3 wherein the circuitry for applying voltage Vreset across the CES element comprises a capacitor provided in a parallel arrangement with the CES element.

5. A spiking neuron for a spiking neural network, the spiking neuron comprising:
a correlated electron switch (CES) element for implementing a thresholding function of the spiking neuron
an accumulator circuit for summing current signals received by the spiking neuron to provide an accumulated current signal,
wherein the accumulator circuit for summing the signals received by the input node comprises a crosspoint array for applying weights to the received signals.

6. A synapse for a spiking neural network, the synapse comprising:
a crosspoint array comprising:
at least one row signal line and at least one column signal line; and
a plurality of programmable CES elements provided at each intersection of a row signal line and a column signal line, wherein each CES element is programmable into a high impedance state or one of a plurality of low impedance states.

7. The synapse as claimed in claim 6 further comprising:
input nodes coupled to each row signal line, for receiving current signals from the spiking neural network; and
output nodes coupled to each column signal line, wherein each output node is couplable to a spiking neuron.

8. The synapse as claimed in claim 6 further comprising:
circuitry for coupling the crosspoint array to calibration circuitry for writing the programmable CES elements into a required impedance state.

9. A method of outputting spike signals from a spiking neuron, the method comprising:
using a correlated electron switch (CES) element to implement a thresholding function of the spiking neuron, wherein the CES element stores a threshold current;
accumulating two or more current signals received from the spiking neural network to provide an accumulated current signal;
outputting a spike signal;
comparing the stored threshold current and the accumulated current signal;
determining if the accumulated current signal is greater than or equal to the stored threshold current; and
outputting, responsive to the determining, the spike signal.

10. The method as claimed in claim 9 further comprising:
programming the CES element into one of a plurality of low impedance states to store a threshold current value; and
applying, subsequent to the programming, a voltage Vreset across the CES element.

11. The method as claimed in claim 10 further comprising:
outputting the spike signal when the accumulated current signal exceeds a threshold current Ireset and causes the CES element to switch out of the low impedance state upon application of voltage Vreset.

12. The method as claimed in claim 9 further comprising:
programming the CES element into a high impedance state to store a threshold current value; and
applying, subsequent to the programming, a voltage Vset across the CES element.

13. The method as claimed in claim 12 further comprising:
outputting the spike signal when the accumulated current signal exceeds a threshold current Iset and causes the CES element to switch out of the high impedance state.

14. The method as claimed in claim 9 further comprising:
resetting the spiking neuron subsequent to outputting a spike signal.

Assignees

• Advanced Risc Mach Ltd

Inventors

• Suda Naveen
• Chandra Vikas
• Cline Brian Tracy
• Sinha Saurabh Pijuskumar
• Das Shidhartha

Classifications

• G06N3/065

  Analogue means · CPC title

• G06N3/049 Primary

  Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G11C13/0002

  using resistive RAM [RRAM] elements · CPC title

• G11C11/54

  using elements simulating biological cells, e.g. neuron · CPC title