Spike neural network circuit
Disclosed is an SNN circuit including a spike generator that receives an input spike signal and converts the input spike signal into a sub-spike signal and a main spike signal, a synaptic circuit that generates an operation signal based on a weight and outputs the operation signal in response to the main spike signal, a membrane capacitor that accumulates the operation signal, and a potential correction circuit that corrects an output terminal voltage of the synaptic circuit based on a voltage of the membrane capacitor.
1. A spike neural network (SNN) circuit comprising:
a spike generator configured to receive an input spike signal and to convert the input spike signal into a sub-spike signal and a main spike signal;
a synaptic circuit configured to generate an operation signal based on a weight and to output the operation signal in response to the main spike signal;
a membrane capacitor configured to accumulate the operation signal; and
a potential correction circuit configured to correct an output terminal voltage of the synaptic circuit based on a voltage of the membrane capacitor in response to the sub-spike signal.
2. The SNN circuit of claim 1 , further comprising:
a transistor including a gate terminal that receives the main spike signal, a source terminal connected to the synaptic circuit and the potential correction circuit, and a drain terminal connected to the membrane capacitor and the potential correction circuit.
3. The SNN circuit of claim 2 , wherein, in a fire section of the main spike signal, the transistor delivers the operation signal to the membrane capacitor, and
wherein, when a current section is not the fire section of the main spike signal, the transistor does not deliver the operation signal to the membrane capacitor.
4. The SNN circuit of claim 1 , wherein, in a fire section of the sub-spike signal, the output terminal voltage of the synaptic circuit is controlled by the potential correction circuit to have a level the same as the voltage of the membrane capacitor.
5. The SNN circuit of claim 1 , wherein the spike generator includes:
a first NAND gate configured to output the sub-spike signal based on a first signal identical to the input spike signal, and a second signal obtained by inverting and delaying the first signal; and
a second NAND gate configured to output the main spike signal based on the first signal and a third signal obtained by inverting the second signal.
6. The SNN circuit of claim 1 , wherein the synaptic circuit includes:
a binary memory configured to store the weight; and
a current-mode digital-to-analog converter (C-DAC) configured to generate the operation signal corresponding to the weight stored in the binary memory.
7. The SNN circuit of claim 1 , wherein the potential correction circuit further includes an oscillation prevention capacitor.
8. The SNN circuit of claim 1 , further comprising:
a neuron circuit configured to fire when the membrane capacitor is charged to have a voltage that is not smaller than a threshold voltage.
9. The SNN circuit of claim 8 , wherein the neuron circuit outputs a spike signal to another SNN circuit.
10. The SNN circuit of claim 1 , further comprising:
a neuron circuit configured to fire when the membrane capacitor is discharged to have a voltage that is not greater than a threshold voltage.
11. The SNN circuit of claim 10 , wherein the neuron circuit outputs a spike signal to another SNN circuit.
12. An operating method of an SNN circuit, the method comprising:
converting an input spike signal into a sub-spike signal and a main spike signal to correct a potential;
matching an output terminal voltage of a synaptic circuit based on a voltage of a membrane capacitor in a fire section of the sub-spike signal;
delivering an operation signal to the membrane capacitor in a fire section of the main spike signal; and
accumulating the operation signal in the membrane capacitor.
13. The method of claim 12 , further comprising:
firing, by a neuron circuit, when the voltage of the membrane capacitor is greater than a threshold voltage.
14. The method of claim 12 , further comprising:
firing, by a neuron circuit, when the voltage of the membrane capacitor is smaller than a threshold voltage.
15. The method of claim 12 , wherein the converting of the input spike signal into the sub-spike signal and the main spike signal includes:
outputting the sub-spike signal by inputting a first signal, which is identical to the input spike signal, and a second signal, which is obtained by inverting and delaying the first signal, into a first NAND gate; and
outputting the main spike signal by inputting the first signal and a third signal, which is obtained by inverting the second signal, into a second NAND gate.