Sysnapse circuit for preventing errors in charge calculation and spike neural network circuit including the same
Disclosed is a synaptic circuit including a weight memory that stores a weight value, a current-mode digital-to-analog converter (C-DAC) circuit that receives the weight value from the weight memory and supplies a current based on the weight value, a parasitic capacitor correction circuit that receives the weight value from the weight memory and to correct a value of parasitic capacitance generated by the C-DAC circuit based on the weight value, and a pre-discharge circuit that drains charges accumulated by the parasitic capacitance.
1 . A synaptic circuit comprising:
a weight memory configured to store a weight value;
a current-mode digital-to-analog converter (C-DAC) circuit configured to receive the weight value from the weight memory and to supply a current based on the weight value;
a parasitic capacitor correction circuit configured to receive the weight value from the weight memory and to correct a value of parasitic capacitance generated by the C-DAC circuit based on the weight value; and
a pre-discharge circuit configured to drain charges accumulated by the parasitic capacitance.
2 . The synaptic circuit of claim 1 , wherein the weight value is an n-bit binary number.
3 . The synaptic circuit of claim 2 , wherein the C-DAC circuit includes ‘n’ PMOS transistors and ‘n’ switches, and the parasitic capacitor correction circuit includes one inverter and ‘n’ switches,
wherein the parasitic capacitance is generated based on whether the ‘n’ switches included in the C-DAC circuit are opened or closed, with respect to the C-DAC circuit, and
wherein the parasitic capacitance is generated based on the ‘n’ switches included in the parasitic capacitor correction circuit are opened or closed, with respect to the parasitic capacitor correction circuit.
4 . The synaptic circuit of claim 3 , wherein it is determined, based on the weight value, whether the ‘n’ switches included in the C-DAC circuit are opened or closed, and
wherein it is determined, based on a result of inverting the weight value through the one inverter, whether the ‘n’ switches included in the parasitic capacitor correction circuit are opened or closed.
5 . The synaptic circuit of claim 2 , wherein the ‘n’ bits are 8 bits.
6 . A spike neural network (SNN) circuit comprising:
a synaptic circuit array including a plurality of synaptic circuits arranged in ‘n’ rows and ‘m’ columns;
a plurality of spike generators electrically connected to correspond to the ‘n’ rows, respectively;
a plurality of neuron circuits electrically connected to correspond to the ‘m’ columns, respectively; and
a plurality of membrane capacitors electrically connected to correspond to the ‘m’ columns, respectively,
wherein each of the plurality of synaptic circuits includes:
a weight memory configured to store a weight value;
a C-DAC circuit configured to receive the weight value from the weight memory and to supply a current based on the weight value;
a parasitic capacitor correction circuit configured to receive the weight value from the weight memory and to correct a value of parasitic capacitance generated by the C-DAC circuit based on the weight value; and
a pre-discharge circuit configured to drain charges accumulated by the parasitic capacitance.
7 . The SNN circuit of claim 6 , wherein the weight value is an n-bit binary number.
8 . The SNN circuit of claim 7 , wherein the C-DAC circuit includes ‘n’ PMOS transistors and ‘n’ switches,
wherein the parasitic capacitor correction circuit includes one inverter and ‘n’ switches,
wherein the parasitic capacitance is generated based on whether the ‘n’ switches included in the C-DAC circuit are opened or closed, with respect to the C-DAC circuit, and
wherein the parasitic capacitance is generated based on the ‘n’ switches included in the parasitic capacitor correction circuit are opened or closed, with respect to the parasitic capacitor correction circuit.
9 . The SNN circuit of claim 8 , wherein it is determined, based on the weight value, whether the ‘n’ switches included in the C-DAC circuit are opened or closed, and
wherein it is determined, based on a result of inverting the weight value through the one inverter, whether the ‘n’ switches included in the parasitic capacitor correction circuit are opened or closed.
10 . The SNN circuit of claim 6 , wherein a first neuron circuit among the plurality of neuron circuits receives a signal accumulated from a plurality of synaptic circuits arranged in a first column of the synaptic circuit array, and
wherein, when a potential by the accumulated signal exceeds a fire reference potential of the first neuron circuit, the first neuron circuit outputs a spike voltage.
11 . The SNN circuit of claim 6 , wherein each of the plurality of spike generator includes:
a first NAND operator configured to output a spike sub-voltage; and
a second NAND operator configured to output a spike main voltage.
12 . The SNN circuit of claim 11 , wherein the first NAND operator is configured to:
receive a first voltage, which is an input voltage entered into each of the plurality of spike generator, and a second voltage generated after the input voltage passes through five inverters connected in series; and
generate the spike sub-voltage by performing a NAND operation on the first voltage and the second voltage, and
wherein the second NAND operator is configured to:
receive the first voltage and a third voltage generated after the input voltage passes through six inverters connected in series; and
generate the spike main voltage by performing the NAND operation on the first voltage and the third voltage.
13 . The SNN circuit of claim 7 , wherein the ‘n’ bits are 8 bits.