Spike neural network circuit including probabilistic operator
Disclosed is a spike neural network circuit including an axon circuit that generates a first input spike signal, a conversion table that converts a first fire probability of a first neuron corresponding to the first input spike signal into a first threshold value, and a probabilistic operator. The probabilistic operator includes a random number generator that generates a random number value based on an event that the first input spike signal is at a first logic level, a random number comparator that generates a first comparison signal by comparing the first threshold value with the random number value, and a spike generator that generates an output spike signal corresponding to the first neuron based on an event that the first comparison signal is at the first logic level.
1 . A spike neural network circuit comprising:
an axon circuit configured to:
generate a first input spike signal; and
generate a second input spike signal;
a conversion table configured to:
convert a first fire probability of a first neuron corresponding to the first input spike signal into a first threshold value; and
convert a second fire probability of a second neuron corresponding to the second input spike signal into a second threshold value; and
a probabilistic operator,
wherein the probabilistic operator includes:
a random number generator configured to generate a random number value based on an event that at least one of the first input spike signal and the second input spike signal is at a first logic level;
a random number comparator configured to;
generate a first comparison signal by comparing the first threshold value with the random number value; and
generate a second comparison signal by comparing the second threshold value with the random number value; and
a spike generator configured to generate an output spike signal corresponding to the first neuron based on an event that at least one of the first comparison signal and the second comparison signal is at the first logic level,
wherein the first fire probability is identical to a probability that the random number value is greater than or equal to the first threshold value, and
wherein the second fire probability is identical to a probability that the random number value is greater than or equal to the second threshold value.
2 . The spike neural network circuit of claim 1 , wherein the probabilistic operator further includes a spike detector configured to:
generate a first activation signal having the first logic level when the first input spike signal is received from the axon circuit; and
generate the first activation signal having a second logic level when the first input spike signal is not received from the axon circuit.
3 . The spike neural network circuit of claim 2 , wherein the random number comparator is further configured to:
generate the first comparison signal by comparing the first threshold value with the random number value when the first activation signal is at the first logic level; and
not generate the first comparison signal when the first activation signal is at the second logic level.
4 . The spike neural network circuit of claim 2 , wherein the spike detector is further configured to:
generate a second activation signal having the first logic level when the second input spike signal is received; and
generate the second activation signal having the second logic level when the second input spike signal is not received.
5 . The spike neural network circuit of claim 4 , further comprising:
a first logic gate configured to generate a random number generator activation signal corresponding to an OR operation of the first activation signal and the second activation signal; and
a second logic gate configured to generate a spike generator activation signal corresponding to an OR operation of the first comparison signal and the second comparison signal.
6 . The spike neural network circuit of claim 5 , wherein the random number generator is further configured to:
generate the random number value when the random number generator activation signal is at the first logic level,
wherein the spike generator is further configured to:
generate the output spike signal when the spike generator activation signal is at the first logic level, and
wherein the first logic level is at a logic high level.
7 . The spike neural network circuit of claim 1 , wherein the random number comparator is further configured to:
generate the first comparison signal having the first logic level when the random number value is greater than or equal to the first threshold value; and
generate the first comparison signal having a second logic level when the random number value is less than the first threshold value, and
wherein the first logic level is at a logic high level, and the second logic level is at a logic low level.
8 . The spike neural network circuit of claim 1 , wherein the conversion table is further configured to:
convert a first weight value learned for a first synapse corresponding to the first input spike signal and the first neuron into the first fire probability, and
wherein the first fire probability is a probability that the first neuron fires.
9 . The spike neural network circuit of claim 1 , wherein the random number generator is further configured to:
output one value of a lower limit value, an upper limit value, and values between the lower limit value and the upper limit value as the random number value, and
wherein probabilities that each of the lower limit value, the upper limit value, and the values between the lower limit value and the upper limit value is output as the random number value are the same as each other.
10 . A spike neural network circuit comprising:
an axon circuit configured to generate a first input spike signal;
a conversion table configured to convert a first fire probability of a first neuron corresponding to the first input spike signal into a first threshold value and to convert a second fire probability of a second neuron corresponding to the first input spike signal into a second threshold value;
a first probabilistic operator; and
a second probabilistic operator,
wherein the first probabilistic operator includes:
a first random number generator configured to generate a first random number value based on an event that the first input spike signal is at a first logic level;
a first random number comparator configured to generate a first comparison signal by comparing the first threshold value with the first random number value; and
a first spike generator configured to generate a first output spike signal corresponding to the first neuron based on an event that the first comparison signal is at the first logic level,
wherein the second probabilistic operator includes:
a second random number generator configured to generate a second random number value based on an event that the first input spike signal is at the first logic level;
a second random number comparator configured to generate a second comparison signal by comparing the second threshold value with the second random number value; and
a second spike generator configured to generate a second output spike signal corresponding to the second neuron based on an event that the second comparison signal is at the first logic level,
wherein the first fire probability is identical to a probability that the first random number value is greater than or equal to the first threshold value, and
wherein the second fire probability is identical to a probability that the second random number value is greater than or equal to the second threshold value.
11 . The spike neural network circuit of claim 10 , wherein the first probabilistic operator further includes:
a first spike detector configured to:
generate a first activation signal having the first logic level when the first input spike signal is received from the axon circuit; and
generate the first activation signal having a second logic level when the first input spike signal is not received from the axon circuit, and
wherein the second probabilistic operator further includes:
a second spike detector configured to:
generate a second activation signal having the first logic level when the first input spike signal is received from the axon circuit; and
generate the second activation signal having the second logic level when the first input spike signal is not received from the axon circuit.
12 . The spike neural network circuit of claim 10 , wherein the first random number comparator is further configured to:
generate the first comparison signal having the first logic level when the first random number value is greater than or equal to the first threshold value; and
generate the first comparison signal having a second logic level when the first random number value is less than the first threshold value,
wherein the second random number comparator is further configured to:
generate the second comparison signal having the first logic level when the second random number value is greater than or equal to the second threshold value; and
generate the second comparison signal having the second logic level when the second random number value is less than the second threshold value, and
wherein the first logic level is at a logic high level, and the second logic level is at a logic low level.
13 . The spike neural network circuit of claim 10 , wherein the conversion table is further configured to:
convert a first weight value learned for a first synapse corresponding to the first input spike signal and the first neuron into the first fire probability; and
convert a second weight value learned for a second synapse corresponding to the first input spike signal and the second neuron into the second fire probability,
wherein the first fire probability is a probability that the first neuron fires, and
wherein the second fire probability is a probability that the second neuron fires.