阅读说明：本技术 类神经电路以及运作方法 (Neural circuit and operation method ) 是由 林仲汉 邱青松 于 2019-11-15 设计创作，主要内容包括：一种类神经电路包含突触电路以及后神经元电路。突触电路包含相变化元件、第一开关以及第二开关。第一开关耦接相变化元件,并用以接收第一脉冲信号。第二开关耦接相变化元件。后神经元电路的输入端耦接切换电路,且其输出端耦接相变化元件。输入端响应于第一脉冲信号而透过切换电路进行充电。后神经元电路用以依据输入端的电压位准与电压门槛值产生激发信号,且更用以依据激发信号产生第一及第二控制信号。后神经元电路依据第一控制信号以关闭切换电路。第二控制信号用以协同第二脉冲信号以控制第二开关,以控制相变化元件的状态,进而决定类神经电路的权重。如此,可利用电路建造出类神经网络系统。(A neural circuit includes a synaptic circuit and a post-neuron circuit. The synapse circuit comprises a phase change element, a first switch and a second switch. The first switch is coupled to the phase change element and is used for receiving a first pulse signal. The second switch is coupled to the phase change element. The input end of the back neuron circuit is coupled with the switching circuit, and the output end of the back neuron circuit is coupled with the phase change element. The input end responds to the first pulse signal and charges through the switching circuit. The back neuron circuit is used for generating a trigger signal according to the voltage level and the voltage threshold value of the input end, and is further used for generating a first control signal and a second control signal according to the trigger signal. The back neuron circuit closes the switching circuit according to the first control signal. The second control signal is used for cooperating with the second pulse signal to control the second switch so as to control the state of the phase change element and further determine the weight of the neural circuit. Thus, the neural network system can be built by using the circuit.)

1. A neural circuit, comprising:

a synaptic electrical circuit comprising:

a phase change element;

a first switch coupled to the phase change element for receiving a first pulse signal; and

a second switch coupled to the phase change element for receiving a second pulse signal; and a posterior neuron circuit comprising:

an input terminal coupled to the phase change element;

a switching circuit coupled to the input terminal;

a capacitor coupled to the switching circuit; and

an output terminal;

the input end responds to the first pulse signal and charges the capacitor through the switching circuit, the back neuron circuit generates an excitation signal at the output end according to a voltage level and a voltage threshold value of the capacitor, and generates a first control signal and a second control signal according to the excitation signal, the first control signal is used for turning off the switching circuit, and the second control signal is used for controlling the second switch by cooperating with the second pulse signal so as to control the state of the phase change element.

2. The neural circuit of claim 1, further comprising a pre-neuron circuit, wherein the pre-neuron circuit is coupled to the synapse circuit and configured to send the first and second pulse signals.

3. The neural circuit of claim 1, wherein the post-neuron circuit further comprises:

a comparator for comparing the voltage level of the capacitor with the voltage threshold to generate the excitation signal.

4. The neural circuit of claim 1, wherein the post-neuron circuit further comprises:

a first controller coupled to the output terminal and a control terminal of the switching circuit, wherein the first controller generates the first control signal to the control terminal according to the excitation signal; and

and the second controller is coupled with the output end and the input end, and generates the second control signal according to the excitation signal.

5. The neural circuit of claim 4, wherein the switching circuit comprises a transistor.

6. The neural circuit of claim 5, wherein the first controller comprises a high pass filter, wherein the high pass filter generates the first control signal to turn off the transistor after filtering the excitation signal.

7. The neural circuit of claim 4, wherein the second controller comprises:

a delay circuit coupled to the output terminal for delaying the excitation signal; and

a pulse signal generator coupled to the delay circuit for generating the second control signal according to the delayed trigger signal.

8. The neural circuit of claim 7, wherein the pulse signal generator sends the second control signal to the phase change element.

9. A method of operating a neural circuit, comprising:

receiving a first pulse signal through a first switch of a synaptic circuit;

in response to the first pulse signal, an input end of a back neuron circuit is charged through a switching circuit;

the rear neuron circuit compares a voltage level of the input end with a voltage threshold value through a comparison unit to generate a trigger signal;

generating a first control signal and a second control signal according to the excitation signal through the rear neuron circuit;

closing the switching circuit by the rear neuron circuit according to the first control signal; and

and cooperatively controlling a second switch of the synaptic circuit according to the second control signal and a second pulse signal to adjust the magnitude of the current flowing through a phase change element of the synaptic circuit, thereby determining a weight of the neural circuit.

10. The method of claim 9, wherein turning off the switch circuit by the post-neuron circuit according to the first control signal comprises:

the switching circuit is turned off by a controller of the back neuron circuit according to the excitation signal to generate the first control signal, wherein the controller is coupled to an output terminal of the back neuron circuit and the switching circuit.

11. The method of claim 10, wherein the switching circuit comprises a transistor and the controller comprises a filter, and wherein the generating the first control signal to turn off the switching circuit according to the excitation signal by the controller of the post-neuron circuit comprises:

the first control signal is generated to turn off the transistor after the excitation signal is filtered by the filter.

12. The method of claim 11, wherein the filter comprises a high pass filter, wherein the generating the first control signal to turn off the transistor after filtering the excitation signal by the high pass filter comprises:

the first control signal is generated by filtering the excitation signal through the high-pass filter to turn off the transistor.

13. The method of claim 12, further comprising:

delaying the fire signal by a delay circuit of the post neuron circuit;

generating the second control signal by a pulse signal generator of the rear neuron circuit according to the delayed excitation signal; and

and transmitting the second control signal to the phase change element through the pulse signal generator.

Technical Field

Embodiments described herein relate generally to circuit technology, and more particularly, to a neural circuit and method of operation.

Background

A neural network system is included in a living organism. The neural network system contains many neurons (neurons). Neurons were proposed by Heinrich Wilhelm Gottfriend von Waldehyer-Hartz in 1891. Neurons are the processing units that acquire discrete information from the brain. In 1897, Charles Sherrington referred to the interface (junction) between two neurons as a "synapse" (synapse). Discrete information flows through the synapse in one direction. According to this direction, a distinction is made between "presynaptic (presynaptic) neurons" and "postsynaptic (postsynaptic) neurons". Neurons fire when they receive enough input to emit a "spike".

Theoretically, the captured experience is as conduction of synapses in the brain (conductance). Synaptic conduction may vary over time according to the relative spike times of the pre-synaptic neuron circuit and the post-synaptic neuron. If a post-synaptic neuron fires before a pre-synaptic neuron circuit fires (fire), synaptic conductance may increase. If the order of the two excitations is reversed, the synaptic conductance will decrease. In addition, such changes may depend on the delay between two events. The more delay, the smaller the magnitude of the change.

Artificial neural networks allow electronic systems to function in a manner similar to that of biological brains. The neuron system may include various electronic circuits that model biological neurons.

The neural network system affects perception, selection, decision or other various behaviors of the living body, and thus plays a very important role in the living body. If a neural network system in a similar organism can be built by using the circuit, the circuit has a critical influence on many fields.

For example, U.S. Pat. No. 9,830,981 or Chinese patent No. 107111783 disclose that neural network-like systems can be constructed using phase change elements and other elements.

Disclosure of Invention

Some embodiments of the present disclosure relate to a neural circuit including a synaptic circuit and a post-neuron circuit. The synapse circuit comprises a phase change element, a first switch and a second switch. The first switch is coupled to the phase change element and is used for receiving a first pulse signal. The second switch is coupled to the phase change element and is used for receiving a second pulse signal. The back neuron circuit comprises an input end, a switching circuit, a capacitor and an output end. The input terminal is coupled to the phase change element. The switching circuit is coupled to the input terminal. The capacitor is coupled to the switching circuit. The input end responds to the first pulse signal and charges the capacitor through the switching circuit, the back neuron circuit generates an excitation signal at the output end according to the voltage level and the voltage threshold value of the capacitor, and generates a first control signal and a second control signal according to the excitation signal, the first control signal is used for turning off the switching circuit, and the second control signal is used for controlling the second switch in cooperation with the second pulse signal so as to control the state of the phase change element.

In some embodiments, the neuron-like circuit further comprises a pre-neuron circuit, wherein the pre-neuron circuit is coupled to the synapse circuit and configured to send the first pulse signal and the second pulse signal.

In some embodiments, the post-neuron circuit further comprises a comparator for comparing a voltage level of the capacitor with a voltage threshold to generate the excitation signal.

In some embodiments, the post-neuron circuit also includes a first controller and a second controller. The first controller is coupled to the output end and the control end of the switching circuit, wherein the first controller generates a first control signal to the control end according to the excitation signal. The second controller is coupled to the output end and the input end, wherein the second controller generates a second control signal according to the excitation signal.

In some embodiments, the switching circuit includes a transistor.

In some embodiments, the first controller includes a high pass filter, wherein the high pass filter generates the first control signal to turn off the transistor after filtering the excitation signal.

In some embodiments, the second controller includes a delay circuit and a pulse signal generator. The delay circuit is coupled to the output terminal and is used for delaying the excitation signal. The pulse signal generator is coupled to the delay circuit and is used for generating a second control signal according to the delayed excitation signal.

In some embodiments, the pulse signal generator sends the second control signal to the phase change element.

Some embodiments of the present disclosure relate to a method of operating a neural circuit. The operation method comprises the following steps: receiving, by a first switch of a synaptic electrical circuit, a first pulse signal; in response to the first pulse signal, the input end of the back neuron circuit is charged through the switching circuit; the back neuron circuit compares the voltage level of the input end with a voltage threshold value through a comparison unit to generate an excitation signal; generating a first control signal and a second control signal according to the excitation signal through a back neuron circuit; closing the switching circuit through the rear neuron circuit according to the first control signal; and cooperatively controlling a second switch of the synaptic circuit according to the second control signal and the second pulse signal to adjust the magnitude of the current flowing through the phase change element of the synaptic circuit, thereby determining the weight of the neural circuit.

In some embodiments, the step of turning off the switching circuit by the post-neuron circuit according to the first control signal comprises: the switching circuit is turned off by a first control signal generated by a controller of the back neuron circuit according to the excitation signal, wherein the controller is coupled to the output end of the back neuron circuit and the switching circuit.

In some embodiments, the switching circuit comprises a transistor, the controller comprises a filter, and the step of generating the first control signal by the controller of the post-neuron circuit according to the excitation signal to turn off the switching circuit comprises: the excitation signal is filtered by a filter to generate a first control signal to turn off the transistor.

In some embodiments, the filter comprises a high pass filter, and the step of generating the first control signal to turn off the transistor after filtering the excitation signal by the high pass filter comprises: the first control signal generated after filtering the excitation signal by the high pass filter turns off the transistor.

In some embodiments, the method further comprises: delaying the firing signal by a delay circuit of the post neuron circuit; generating a second control signal by a pulse signal generator of the rear neuron circuit according to the delayed excitation signal; and transmitting a second control signal to the phase change element through the pulse signal generator.

In summary, the neural network system can be built by using the neural circuit and the operation method of the present disclosure.

Drawings

The foregoing and other objects, features, advantages and embodiments of the disclosure will be apparent from the following more particular description of the embodiments, as illustrated in the accompanying drawings in which:

FIG. 1 is a schematic diagram of a type of neural circuit according to some embodiments of the present disclosure;

FIG. 2 is a schematic diagram of a type of neural circuit according to some embodiments of the present disclosure;

FIG. 3 is a waveform diagram of a plurality of signals according to some embodiments of the present disclosure;

FIG. 4 is a schematic diagram of a type of neural circuit according to some embodiments of the present disclosure;

FIG. 5 is a schematic diagram of a type of neural circuit according to some embodiments of the present disclosure;

FIG. 6 is a schematic diagram of a type of neural circuit according to some embodiments of the present disclosure;

FIG. 7 is a schematic diagram of a type of neural circuit according to some embodiments of the present disclosure; and

fig. 8 is a flow chart of a method of operating a type of neural circuit according to some embodiments of the present disclosure.

Detailed Description

The term "coupled", as used herein, may also mean "electrically coupled", and the term "connected", as used herein, may also mean "electrically connected". "coupled" and "connected" may also mean that two or more elements co-operate or interact with each other.

Please refer to fig. 1 and fig. 2. Fig. 1 and 2 are schematic diagrams of a neural circuit 1000 according to some embodiments of the disclosure.

For example, as shown in fig. 1, the neuron-like circuit 1000 includes a synaptic circuit (synapse circuit)1200, a pre-synaptic neuron circuit 1300 (hereinafter referred to as "pre-neuron" 1300), and a post-synaptic neuron circuit 1400 (hereinafter referred to as "post-neuron" 1400), wherein the pre-neuron 1300 includes an axon driver 1310, the axon driver 1310 includes two pulse generators G1G2, and the post-neuron 1400 includes a dendrite (dendrite) for receiving signals. In some embodiments, the axon driver 1310 of the anterior neuron 1300 sends spike signals (spike) to the dendrites (dendrites) of the posterior neuron 1400 via the synaptic electrical circuit 1200 to stimulate the posterior neuron 1400, thus achieving an efficacy similar to signal delivery in a neural network system.

In one embodiment, synaptic electrical circuit 1200 includes phase change element PCM, switch D1, and switch SW 2. The phase change element PCM comprises a phase change material. Phase change materials have different phases based on the magnitude of the current. The information may be stored in the corresponding phase. For example, when the phase change element PCM is a crystalline phase (crystal) or a polycrystalline phase (polycrystalline), the resistance value is small. When the phase change element PCM is in an amorphous phase (amorphous phase), the resistance is large. The phase change element PCM may store a logic value 1 or 0 based on the magnitude of the resistance value of the phase change element PCM.

In one embodiment, switch D1 is implemented as a diode. The switch SW2 is implemented as a transistor. In some other embodiments, the switch D1 may be implemented by a transistor. Switch D1 includes a first terminal and a second terminal. The first terminal is an anode terminal and the second terminal is a cathode terminal. The switch D1 has a first terminal coupled to the pulse signal generator G1 for receiving the pulse signal PS 1. The control terminal of the switch SW2 is coupled to the pulse signal generator G2 to receive the pulse signal PS 2. The second terminal of the switch D1 and the second terminal of the switch SW2 are coupled to the first terminal of the phase change element PCM. The second terminal of the phase change element PCM is coupled to the posterior neuron 1400.

In one embodiment, the rear neuron 1400 includes a switch SW3, a capacitor C1, a resistor R1, a comparator COM, a delay circuit TD, a wave shaping circuit G3, and a controller CTR 1. The capacitor C1 and the resistor R1 have a first terminal coupled to a voltage terminal VL, which may be a low voltage terminal, for exampleAnd a ground terminal GND. The comparator COM includes a positive input terminal, a negative input terminal, and an output terminal. The second terminal of the capacitor C1 and the second terminal of the resistor R1 are coupled to the positive input terminal of the comparator COM, and the second terminal of the phase change element PCM is coupled to the positive input terminal of the comparator COM through the switch SW3 of the back neuron 1400. The negative input terminal of the comparator COM is used for receiving a voltage threshold value V_th. The output terminal of the comparator COM is coupled to the delay circuit TD and the output terminal OUT. The delay circuit TD is coupled to the wave shaping circuit G3. The wave shaping circuit G3 is coupled to the second terminal of the phase change element PCM. The controller CTR1 is coupled to the output terminal OUT and the switch SW 3.

The capacitance C1 in the posterior neuron 1400 simulates the membrane potential of the neuron cell, and there are many charged ions inside and outside the neuron cell membrane, and the voltage difference Vp between the inside and outside of the cell membrane (also called membrane potential Vp) is reflected by the capacitance C1 due to the difference between the types and the charge amount of the charged ions inside and outside the cell membrane. The neuron cell membranes are provided with channels (channels) which are different in size and can control the charged ions to enter and exit, the charged ions inside and outside the cell membranes can pass through the channels to cause the change of the Vp value of the membrane potential, and the resistor R1 is an electrical effect for simulating the charged ions to pass through the channels back and forth. The pulse signal sent from the axon (axon) of the pre-synaptic neuron is received by the dendrite (dendrite) of the post-neuron to change the membrane potential Vp of the post-neuron cell membrane, which corresponds to the behavioral effect of the post-neuron 1400, i.e., to charge the capacitor C1.

If the pulse signal is strong enough, the membrane potential Vp of the capacitor C1 exceeds the voltage threshold V_thThe posterior neuron 1400 outputs a FIRE signal (FIRE). On the contrary, if the pulse signal strength is not large enough, the voltage on the capacitor C1 will rise, but will not exceed the voltage threshold V_thThe posterior neuron 1400 does not output the FIRE signal (FIRE), and the rising membrane potential Vp thereof gradually decreases through the leakage of the resistor R1. The behavior of the neuron on neuron cells is that the concentration of charged ions inside and outside a cell membrane is instantaneously changed by a back neuron due to an excitation signal of a front neuron, and then the charged ions are diffused and balanced through a channel on the cell membrane, so that the membrane potential Vp of the cell membrane of the back neuron is restored to a balanced value. Thus, a pulse signal is sent from the anterior neuron to the posterior neuronThe behavior of this path on the capacitance C1 is electrically called leakage sum and Fire (LIF), and the neuronal cell membrane potential Vp is a function of LIF (Vp ═ F (LIF)).

Although the excitation signal of each pre-neuron affects the cell membrane potential of the post-neuron 1400 via synapses (including axons of the pre-neuron and dendrites of the post-neuron), even with the same excitation signal, different pre-neurons affect the cell membrane potential of the post-neuron in different magnitudes, so that the magnitude of a synaptic weight (W) between the pre-neuron and the post-neuron differs, the synaptic weight (W) is plastic (or adaptable), and the magnitude of the weight change (Δ W) is the pre-neuron excitation time (t) t_pre) And post-neuron excitation time (t)_post) Δ W ═ F (t)_post-t_pre). In other words, the magnitude of the synaptic weight change (Δ W) and t_pre、t_postThe time difference between the two is related, and the synaptic weight W is adaptively adjusted according to the value of the time difference. Therefore, synaptic weight W is related to the causal relationship index between neurons, and a characteristic index representing the change of synapse (synapse) in weight W due to the relative relationship between the excitation times of the neurons before and after is defined, which is called Spike Timing Dependent Plasticity (STDP). Synaptic STDP is also indirectly associated with LIF, since LIF determines the firing time (t) of the posterior neuron_post). In one embodiment, the STDP of a synapse represents the plasticity of the synaptic current conductivity, and more specifically, in one embodiment, the STDP of a synapse represents the magnitude of the synaptic resistance.

Referring to fig. 1 and 3, before the pre-neuron 1300 sends a spike, the membrane potential Vp of the capacitor C1 in the post-neuron 140 gradually moves toward a balance potential (VL) through the resistor R1. In one embodiment, the balance potential is a ground potential, but not limited thereto. In operation, switch SW3 of the posterior neuron 1400 is turned on, the anterior neuron 1300 sends a spike to the posterior neuron 1400, and the pulse generator G1 in the axon driver 131 sends a pulse signal PS1 at time T1, with the pulse duration from T1 to T2, during which time (T1) switch D1 is turned on and switch SW2 is turned off. In one embodiment, the pulse signal PS1 is also called "axon pulse LIF" and has a width of 0.1ms, but not limited thereto. The axon pulse LIF passes through the phase change element PCM of the synaptic circuit 1200 and charges the second terminal of the capacitor C1 (the positive input terminal of the comparator COM) through the switch SW3, so that the membrane potential of the cell membrane has the voltage level Vp.

As shown in FIG. 3, the membrane potential Vp gradually rises during the time interval T1. If the voltage level Vp of the capacitor C1 is higher than the voltage threshold V of the negative input terminal of the comparator COM before the time point t2 (including the time point t 2)_thAt this time, the output of the comparator COM immediately sends the FIRE signal FIRE. When the membrane potential Vp is smaller than the voltage threshold value V_thThe comparator COM does not output the FIRE signal FIRE. Therefore, the magnitude of the PCM resistance value of the phase change element can control the charging speed of the capacitor C1.

At time t2, the pulse signal generator G2 of the axon drive 131 generates a pulse signal PS2, in one embodiment the pulse signal PS2 is also referred to as "axon pulse STDP". In one embodiment, the pulse time T of the axon pulse STDP is 100ms, but not limited thereto. The axon pulse STDP is divided into two time zones with equal time (T/2) in the front and back, and the pulse in the front time zone (T/2) is increased by a voltage value (not shown) instantaneously and then gradually increased. The pulse in the later time zone (T/2) is momentarily decreased by a voltage value (not shown) and then gradually decreased. Specifically, the highest voltage value in the latter time zone of the axon pulse STDP is smaller than the lowest voltage value in the latter time zone. After the axon pulse LIF ends, the membrane potential of the capacitor C1 in the posterior neuron 1400 gradually discharges back to the equilibrium value of the cell membrane potential via the resistor R1.

With continued reference to fig. 2 and fig. 3, the FIRE signal FIRE from the COM output of the comparator passes through the control circuit CTR1 to immediately generate the control signal CS1 to turn off the switch SW3, at which time the signal of the synaptic electrical circuit 1200 can no longer affect the voltage of the capacitor C1 of the neuron 1400. Meanwhile, the FIRE signal FIRE passes through the delay circuit TD and the wave shaping circuit G3 to output the control signal CS2 to the second terminal of the phase change element PCM. The delay circuit TD adds a delay time TD to the FIRE signal FIRE. In one embodiment the delay time td is 50 ms. After the delay time td, the control signal CS2 is output to the second terminal of the phase change element PCM at time t 3. The pulse interval of the control signal CS is from a time point t3 to a time point t 4. In one embodiment, the control signal CS2 is also called Post-Neuron STDP Trigger (Post-synthetic Neuron STDP Trigger), and the pulse time is 0.1 ms.

Control circuit CTR1 sends a control signal CS1 to turn off switch SW3 and maintain the on-off state for a pulse time T greater than (and including equal to) axon pulse STDP (PS 2). The capacitor C1 will begin to discharge through the resistor R1, and the membrane potential Vp will gradually decrease, during which time the neuron will no longer receive signals from other synaptic circuits. The pulse duration of the control signal CS2, the current flowing through the phase change element PCM in the synaptic electrical circuit 1200 is determined by the cooperation of the pulse signal PS2 (axon pulse STDP) and the control signal CS2 (post neuron STDP trigger). In other words, the axon pulse STDP (PS2) and the post-neuron STDP trigger (CS2) signal determine the magnitude of the current flowing through the phase change element PCM.

The axon pulse STDP (PS2) controls the gate of the switch SW2 of the synaptic circuit 1200, and in the previous time zone (T/2) of the axon pulse STDP (PS2), the voltage is high, and the current that can flow through the switch SW2 is large. Conversely, in the latter time zone (T/2) of the axon pulse STDP, the voltage is low and the current that can flow through switch SW2 is small. For example, during the pulse of the post-neuron STDP trigger (CS2), the switch D1 in the synaptic electrical circuit 1200 is not conductive, the post-neuron STDP trigger (CS2) only flows through the phase change element and the switch SW2, and the level of the voltage of the axon pulse STDP (PS2) can control the amount of current flowing through the switch SW 2.

In one embodiment, if the firing signal (FIRE) of the posterior neuron 1400 is caused by the axon pulse LIF (PS1) of the anterior neuron 1200, the firing time (t) of the posterior neuron 1400 is_post) Will be later than the firing time (t) of the anterior neuron 1200_pre) At this time, the pulse duration of the post-neuron STDP trigger (CS2) falls in the posterior time zone of the axon pulse STDP (PS 2). Therefore, the current flowing through the switch SW2 is small instantaneously, which means that the current flowing through the phase change element PCM is small instantaneously. The phase change element PCM has a smaller resistance value. In other words, the synaptic electrical circuit 1200 has better conductivityTherefore, it is called Synapse Long Term Potentiation (Synapse Long Term Potentiation). The FIRE representing the posterior neuron 1400 is related to the pulse signal PS1 (axon pulse LIF) of the anterior neuron 1300, and the FIRE of the posterior neuron 1400 is related to the FIRE of the anterior neuron 1300, so that the weight W of the synaptic circuit 1200 connecting the two neurons is increased.

In one embodiment, the firing time (t) of the posterior neuron 1400 is determined if the firing signal (FIRE) of the posterior neuron 1400 is not caused by the axonal pulse LIF (PS1) of the anterior neuron 1200_post) Will be earlier than the firing time (t) of the pre-neuron 1200_pre) At this time, the pulse duration of the posterior neuron STDP trigger (CS2) falls in the anterior time zone of the axon pulse STDP (PS 2). Therefore, the current flowing through the switch SW2 is larger at the moment, which indicates that the current flowing through the phase change element PCM is larger at the moment. Therefore, the phase change element PCM has a larger resistance value. In other words, synaptic electrical circuit 1200 has poor conductivity, referred to as Synapse Long Term suppression (Synapse Long Term suppression). The representation of FIRE associated with the posterior neuron 1400 has no causal relationship with the FIRE of the anterior neuron 1300, and thus the weight W of the synaptic electrical circuit 1200 connecting these two neurons is adjusted down.

The neural circuit 1000 can perform learning actions using the above operations to implement a neural network system in a similar organism.

Please refer to fig. 4. Fig. 4 is a schematic diagram of a neural circuit 2000 according to some embodiments of the disclosure. The neural circuit 2000 of fig. 4 is different from the neural circuit 1000 of fig. 1 in that the controller of the rear neuron 2400 of the neural circuit 2000 is a filter HP, and the switch SW3 is a transistor.

For example, the filter HP is coupled to the output OUT and the gate of the transistor SW 3. When the membrane potential Vp is larger than the voltage threshold value V_thThe comparator COM outputs a positive FIRE signal FIRE. Meanwhile, upon receiving the FIRE signal FIRE, the filter HP generates the control signal CS1 to turn off the transistor SW 3. The connections and operations of the other components of the neural circuit 2000 are similar to those of the neural circuit 1000 in FIG. 1, and thus are not described herein again.

Please refer to fig. 5. Fig. 5 is a schematic diagram of a neural circuit 3000 according to some embodiments of the present disclosure. The difference between the neural circuit 3000 of fig. 5 and the neural circuit 2000 of fig. 4 is that fig. 5 shows in detail the filter HP of the post neuron 2400 of the neural circuit 3000.

For example, the Filter HP is a High Pass Filter (High Pass Filter), the High Pass Filter HP includes a capacitor C2 and a resistor R2, and the transistor SW3 is a PMOS. When the membrane potential Vp is larger than the voltage threshold value V_thThe comparator COM outputs a positive peak excitation signal FIRE. Meanwhile, the high pass filter HP generates the control signal CS1 with a high voltage level to turn off the PMOS (SW3) immediately after receiving the FIRE signal FIRE. It should be noted that the turn-off time of the PMOS (SW3) can be determined by adjusting the time constant τ (time constant τ) of the capacitor C2 and the resistor R2 of the high pass filter HP. The connections and operations of the other components of the neural circuit 3000 are similar to those of the neural circuit 1000 shown in fig. 1, and thus are not described herein again.

Please refer to fig. 6. Fig. 6 is a schematic diagram of a neural circuit 4000 according to some embodiments of the disclosure. The difference between the neural circuit 4000 of fig. 6 and the neural circuit 1000 of fig. 1 is that the controller of the posterior neuron 4400 of the neural circuit 4000 is a filter HP. In fig. 6, the neuron-like circuit 4000 further includes an inverter INV1, and the switch SW3 is an NMOS.

For example, the filter HP is coupled to the gate of the NMOS (SW3) and coupled to the output terminal OUT through the inverter INV 1. When the voltage level Vp of the positive input terminal is larger than the voltage threshold value V_thAt this time, the comparator COM outputs a FIRE signal FIRE having a high voltage level, and the inverter INV1 receives the FIRE signal FIRE to generate an inverted signal. Upon receiving the inverted signal, the filter HP generates the control signal CS1 with a low voltage level to turn off the NMOS (SW 3). The connections and operations of the other components of the neural circuit 4000 are similar to those of the neural circuit 1000 shown in fig. 1, and therefore are not described herein again.

Please refer to fig. 7. Fig. 7 is a schematic diagram of a neural circuit 5000 according to some embodiments of the disclosure. The neural circuit 5000 of fig. 7 is different from the neural circuit 1000 of fig. 1 in that the controller of the rear neuron 5400 of the neural circuit 5000 can be implemented by the level latch LA and the delay circuit TD2, but the present invention is not limited thereto. The quasi-latch LA and the delay circuit TD2 are used to control the off and on time of the switch SW 3. For example, after the firing signal (FIRE) of the posterior neuron is sent, the quasi-latch LA generates a high voltage to turn off the PMOS (SW3) immediately, and the delay circuit TD2 delays the high voltage signal for a period of time to maintain the off state of the PMOS. The connections and operation of the remaining elements of the neural circuit 7000 are similar to those of the neural circuit 1000 of fig. 1, and therefore will not be described herein.

Please refer to fig. 8. Fig. 8 is a flow chart 8000 of a method of operating a neural circuit according to some embodiments of the present disclosure. For example, in fig. 8, the operation method 8000 includes an operation S8100, an operation S8200, an operation S8300, an operation S8400, an operation S8500, and an operation S8600. In some embodiments, the operation method 8000 is applied to the neural circuit 1000 of fig. 1, but the disclosure is not limited thereto. For ease of understanding, the following discussion will be in conjunction with FIG. 1.

In operation S8100, a pulsed signal PS1 is received through a switch D1 of the synaptic electrical circuit 1200. In some embodiments, synaptic circuit 1200 acts as an axon of a pre-synaptic neuron to send a spike to post-synaptic neuron 1400.

In operation S8200, the positive input terminal of the rear neuron 1400 (comparator COM) is charged through the switching circuit SW3 in response to the pulse signal PS 1. In some embodiments, the post-neuron 1400 acts as a dendrite of the post-synaptic neuron to receive a signal from the synaptic electrical circuit 1200.

In operation S8300, the pass-back neuron 1400 determines the voltage level Vp and the voltage threshold V according to the positive input terminal_thThe value generates the FIRE signal FIRE. In some embodiments, when the voltage level Vp is greater than the voltage threshold V_thThe comparator COM outputs the FIRE signal FIRE.

In operation S8400, a control signal CS1 and a control signal CS2 are generated by the posterior neuron 1400 according to the FIRE signal FIRE. In some embodiments, the controller CTR1 of the rear neuron 1400 generates the control signal CS1 according to the FIRE signal FIRE. The delay circuit TD of the rear neuron 1400 introduces a delay time to the FIRE signal FIRE to output the control signal CS 2.

In operation S8500, the switching circuit SW3 is turned off by the rear neuron 1400 according to the control signal CS 1. In one embodiment, the switch circuit SW3 is turned off for a period of time, and the amount of time the switch circuit SW3 is turned off can be set according to actual requirements.

In operation S8600, a switch SW2 of the synaptic circuit 1200 is controlled according to the control signal CS2 and the pulse signal PS2 to control the state of the phase change element PCM of the synaptic circuit 1200. Accordingly, the weight of the neural circuit 1000 can be determined according to the state of the phase change element PCM. In one embodiment, the control signal CS2 flows through the switch SW2, and the pulse signal PS2 controls the amount of current that can flow through the switch SW. In some embodiments, the phase change element PCM comprises a phase change material. The different resistance values correspond to different phase change materials.

The description of method 8000 above includes exemplary operations, but the operations of method 8000 need not be performed in the order shown. It is within the spirit and scope of the present disclosure that the order of the operations of method 8000 may be altered, or that the operations may be performed concurrently, partially concurrently, or partially omitted, as appropriate.

In summary, the neural network system can be built by using the neural circuit and the operation method of the present disclosure.

Although the present disclosure has been described with reference to particular embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present disclosure, and therefore the scope of the present disclosure should be limited only by the terms of the appended claims.