Low-power-consumption neuron circuit supporting multiple coding modes
A low-power neuron circuit with a two-stage sensitive amplifier and logic control circuit addresses high power and design complexity issues, enabling flexible encoding and efficient network training.
Patent Information
- Application Number
- CN202310662172.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-06
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2043-06-06
AI Technical Summary
Current spiking neural networks face challenges in high power consumption and design complexity due to the use of digital and analog circuits, while analog neurons suffer from increased design difficulty and circuit overhead.
A low-power neuron circuit with a two-stage sensitive amplifier structure and logic control circuit that supports multiple coding modes, including frequency and time encoding, and adjusts leaky behavior through external signals, reducing power consumption and enabling flexible encoding methods.
The circuit maintains low power consumption and supports flexible encoding, enhancing energy efficiency and network training flexibility, particularly in sparse event networks.
Smart Images

Figure CN119089952A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of neuron circuits, and in particular to a low-power neuron circuit supporting multiple coding modes. Background Art
[0002] Deep learning based on artificial neural networks has shown wide applications in many fields. However, high-performance learning algorithms rely on extremely large network scales and huge storage and computing consumption. Therefore, while pursuing high-performance artificial intelligence chips, the industry has proposed pulse neural networks that are closer to biological behavior characteristics. Pulse neural networks use relatively sparse pulse signals for transmission and calculation, and have greater development potential in improving the energy efficiency of algorithms.
[0003] As a key component of spiking neural networks, neurons respond differently to input pulses according to different dynamic models. Current spiking neural networks generally use a simple LIF model (leakage, integration, pulse emission) as the behavioral characteristics of neurons. In terms of implementation, although neurons based on digital circuit design can be compatible with flexible and diverse behavioral models, they are higher than analog neurons in terms of circuit area overhead and circuit control complexity.
[0004] Neurons implemented based on pure analog circuits have problems such as increased design difficulty and circuit power consumption as the complexity of neuron functions increases. Summary of the invention
[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a low-power neuron circuit supporting multiple coding modes.
[0006] The purpose of the present invention can be achieved by the following technical solutions:
[0007] The present invention provides a low-power neuron circuit supporting multiple coding modes, the neuron circuit comprising a two-stage sensitive amplifier structure and a corresponding logic control circuit, specifically:
[0008] The input pulse signal generates an enable signal of the first-stage sensitive amplifier through delay control. The first-stage sensitive amplifier compares the potential Vmem of the neuron membrane capacitance with the threshold voltage Vrefs of the first-stage sensitive amplifier to determine whether to generate an enable signal of the second-stage sensitive amplifier.
[0009] When the enable signal of the second-stage sensitive amplifier is valid, the second-stage sensitive amplifier compares the potential Vmem of the neuron membrane capacitance with the second-stage threshold voltage Vref. The output of the second-stage sensitive amplifier is stored through the SR latch and determines whether to generate an output pulse signal. The output pulse signal and the input configuration signal control the reset behavior of the neuron through the logic control circuit.
[0010] Preferably, the threshold voltage Vrefs of the first stage sense amplifier is lower than the threshold voltage Vref of the second stage sense amplifier.
[0011] Preferably, the neuron circuit also includes an input configuration signal to control the leaky behavior of the neuron through a logic control circuit; wherein the leaky behavior of the neuron is characterized by multiple groups of transmission gates with adjustable gate voltages, and under the action of an external power supply, the time constant for the neuron membrane potential to recover to Vb can be flexibly configured; the enabling and shielding of the leaky behavior of the neuron is regulated by the input configuration signal.
[0012] Preferably, the control logic of the neuron circuit is specifically as follows:
[0013] The interface between the neuron circuit and the external circuit includes an input pulse terminal, a potential terminal of the membrane capacitor and an output pulse terminal;
[0014] When a pulse signal is generated at the input pulse end, the external synaptic circuit integrates the potential node of the membrane capacitor, causing it to produce a certain potential rise and fall; after the integration operation, the sensitive amplifier is triggered to compare the membrane potential with the threshold voltage and determine whether the neuron should emit a pulse.
[0015] Preferably, the input configuration signal includes a global reset signal Reset, an on-chip training enable signal Train and an encoding control signal Type; the output signal of the logic control circuit includes OnR and Leak_en signals;
[0016] When the global reset signal Reset is valid, the membrane potential is reset to Vb, and the output pulse signal is output at a low level;
[0017] When the on-chip training enable signal Train is valid, the Leak_en signal turns off the Leaky circuit M0 to shield the Leaky behavior of the neuron;
[0018] Coding control signal Type, which determines the encoding method of the current neural network.
[0019] Preferably, the encoding method includes a frequency encoding method and a time encoding method.
[0020] Preferably, the frequency encoding method is specifically as follows: under the frequency encoding system, after the neuron performs a pulse firing operation, its membrane potential will be reset immediately and wait for the next input pulse signal to arrive.
[0021] Preferably, the time coding method is specifically as follows: under the time coding system, after the neuron performs the pulse firing operation, it will shield the input pulse signal for a subsequent period of time and wait for the global reset signal Reset to wake up.
[0022] Preferably, the input pulse signal is shielded during the subsequent period of time, which is specifically controlled by the switch W0 and the pull-down tube N0.
[0023] Compared with the prior art, the present invention has the following advantages:
[0024] 1) The two-stage sensitive amplifier structure enables the neuron circuit to maintain low power consumption in networks containing sparse events;
[0025] 2) Through the configuration of external signals, the neuron circuit can respond to pulse signals with two different encoding methods;
[0026] 3) When the network is trained on-chip, the control logic circuit inside the neuron will adjust the LIF behavior and assist the external circuit to complete the weight update. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 The structure diagram of the LIF neuron circuit based on a two-stage sensitive amplifier;
[0028] Figure 2 It is a schematic diagram of key signal waveforms under frequency coding configuration;
[0029] Figure 3 This is a schematic diagram of key signal waveforms under time coding configuration;
[0030] Figure 4 The figure is a schematic diagram comparing the power consumption of traditional neuron circuits and this circuit under different weight sparsity. DETAILED DESCRIPTION
[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0032] Example
[0033] This embodiment provides a low-power neuron circuit that supports multiple coding methods. Through the configuration of external signals, the neuron circuit can respond to pulse signals of two different coding methods; the two-stage sensitive amplifier structure enables the neuron circuit to maintain low power consumption in a network containing sparse events; in addition, when the network is trained on-chip, the control logic circuit inside the neuron will adjust the LIF behavior to assist the external circuit in completing the weight update.
[0034] The neuron circuit is composed of a two-stage sensitive amplifier S0-S1 structure and a corresponding logic control circuit M1. The input pulse ispike signal generates an enable signal CLK of the first-stage sensitive amplifier S0 through delay control, and S0 compares the potential Vmem of the neuron membrane capacitor Cmem with a smaller threshold voltage Vrefs to determine whether to generate an enable signal for the second-stage sensitive amplifier S1.
[0035] The output of the second-stage sensitive amplifier S1 generates an output pulse oSpike through the SR latch M2. Finally, the oSpike signal and the input configuration signals Train, Type and Reset generate an OnR signal through the logic control circuit M1 to control the reset of the neuron (pull-up tube P0).
[0036] Among them, the size of the transistor constituting the sensitive amplifier S1 is larger than S0; the offset voltage of S1 is lower than S0; and the power consumption of S1 is greater than S0. The characterization of the neuron leaky behavior is composed of multiple sets of transmission gates T0-T1 with adjustable gate voltages. The external input configuration signals Train and Type can enable and shield the neuron's leaky behavior. In addition, under the action of external power supplies LP and LN, the time constant for the neuron membrane potential to recover to Vb can be flexibly configured.
[0037] Control method: The interface between the neuron circuit and the external circuit is reflected in the input pulse iSpike terminal, the Vmem terminal of the Cmem membrane capacitor, and the output pulse oSpike terminal. When iSpike generates a pulse signal, the external synaptic circuit performs an Integrate operation on the Vmem node of the membrane capacitor Cmem, causing it to produce a certain potential rise and fall. After the Integrate operation is completed, the CLK signal is enabled to control the two-stage sensitive amplifier to compare the membrane potential Vmem with the threshold voltage, and to determine whether the neuron fires a pulse (fire operation).
[0038] Among them, the input configuration signal Reset is a global reset signal. When this signal is valid, the membrane potential Vmem is reset to Vb, and the oSpike output is low level; the input configuration signal Train is an on-chip training enable signal. When this signal is valid, the Leak_en signal turns off the Leaky circuit M0 and shields the Leaky behavior of the neuron; the input configuration signal Type is a 1-bit coding control signal, which determines the coding method of the current neural network: under the frequency coding system, after the neuron fires, its membrane potential Vmem will be reset immediately and wait for the next input pulse iSpike to arrive; under the time coding system, after the neuron fires, it will shield the input pulse iSpike (controlled by switch W0 and pull-down tube N0) for a period of time afterwards and wait for the Reset signal to wake up.
[0039] The two-stage sensitive amplifier structure of the neuron circuit can meet the low power consumption requirements of sparse network operation. By setting the value of Vrefs to be less than the threshold voltage Vref of the neuron pulse, the circuit only turns on the low-power sensitive amplifier S0 when most of the pulses are not emitted (the membrane potential Vmem does not reach Vrefs). Compared with the neuron circuit composed of a single-stage S1, this structure has higher energy efficiency under the overall network operation. In addition, compared with the analog neuron circuit composed of operational amplifiers, the low static power consumption of the sensitive amplifier also gives this neuron circuit a clear advantage.
[0040] The introduction of input configuration signals increases the flexibility of the neuron circuit in network applications. While supporting two different encoding methods, it also provides a certain reference value for the future development of on-chip learning of pulse neural networks.
[0041] Figure 1 The structural diagram of the LIF neuron circuit based on a two-stage sensitive amplifier is given; Figure 2 The waveform diagram of key signals under frequency coding is given; Figure 3 The waveform diagram of key signals under time coding configuration is given; Figure 4 The power consumption of the traditional neuron circuit is compared with that of the proposed circuit under different weight sparsity.
[0042] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
Claims
1. A low-power neuron circuit supporting multiple coding modes, characterized in that: The neuron circuit includes a two-stage sensitive amplifier structure and a corresponding logic control circuit, specifically: The input pulse signal generates an enable signal of the first-stage sensitive amplifier through delay control. The first-stage sensitive amplifier compares the potential Vmem of the neuron membrane capacitance with the first-stage threshold voltage Vrefs to determine whether to generate an enable signal of the second-stage sensitive amplifier; When the enable signal of the second-stage sensitive amplifier is valid, the second-stage sensitive amplifier compares the potential Vmem of the neuron membrane capacitance with the second-stage threshold voltage Vref. The output of the second-stage sensitive amplifier is stored through the SR latch and determines whether to generate an output pulse signal. The output pulse signal and the input configuration signal control the reset behavior of the neuron through the logic control circuit.
2. A low-power neuron circuit supporting multiple coding modes according to claim 1, characterized in that: The transistor size of the second-stage sensitive amplifier is larger than that of the first-stage sensitive amplifier, and the power consumption of the second-stage sensitive amplifier is greater than that of the first-stage sensitive amplifier; the offset voltage of the second-stage sensitive amplifier is lower than that of the first-stage sensitive amplifier.
3. The low-power neuron circuit supporting multiple coding modes according to claim 1, characterized in that: The threshold voltage Vrefs of the first sense amplifier is lower than the threshold voltage Vref of the second stage sense amplifier.
4. The low-power neuron circuit supporting multiple coding modes according to claim 1, characterized in that: The neuron circuit also includes an input configuration signal that controls the leaky behavior of the neuron through a logic control circuit; wherein the leaky behavior of the neuron is characterized by multiple groups of transmission gates with adjustable gate voltages, and under the action of an external power supply, the time constant for the neuron membrane potential to recover to Vb can be flexibly configured; the enabling and shielding of the leaky behavior of the neuron is regulated by the input configuration signal.
5. The low-power neuron circuit supporting multiple coding modes according to claim 1, characterized in that: The control logic of the neuron circuit is specifically as follows: The interface between the neuron circuit and the external circuit includes an input pulse terminal, a potential terminal of the membrane capacitor and an output pulse terminal; When a pulse signal is generated at the input pulse end, the external synaptic circuit integrates the potential node of the membrane capacitor, causing it to produce a certain potential rise and fall; After the integration operation is completed, the sensitive amplifier is triggered to compare the membrane potential with the threshold voltage and decide whether the neuron should fire a pulse.
6. The low-power neuron circuit supporting multiple coding modes according to claim 5, characterized in that: The input configuration signal includes a global reset signal Reset, an on-chip training enable signal Train and an encoding control signal Type; the output signal of the logic control circuit includes OnR and Leak_en signals; When the global reset signal Reset is valid, the membrane potential is reset to Vb, and the output pulse signal is output at a low level; When the on-chip training enable signal Train is valid, the Leak_en signal turns off the Leaky circuit M0 to shield the Leaky behavior of the neuron; Coding control signal Type, which determines the encoding method of the current neural network.
7. The low-power neuron circuit supporting multiple coding modes according to claim 5, characterized in that: The encoding method includes a frequency encoding method and a time encoding method.
8. The low-power neuron circuit supporting multiple coding modes according to claim 7, characterized in that: The frequency coding method is specifically as follows: under the frequency coding system, after the neuron performs a pulse firing operation, its membrane potential will be reset immediately and wait for the next input pulse signal to arrive.
9. The low-power neuron circuit supporting multiple coding modes according to claim 7, characterized in that: The time coding method is specifically as follows: under the time coding system, after the neuron performs the pulse firing operation, it will shield the input pulse signal for a subsequent period of time and wait for the global reset signal Reset to wake up.
10. The low-power neuron circuit supporting multiple coding modes according to claim 9, characterized in that: The input pulse signal is shielded during the subsequent period of time, which is specifically controlled by the switch W0 and the pull-down tube N0.
Citation Information
Patent Citations
Novel high-sensitivity sensitiveness amplifier with output edges symmetrical
CN105788624A
A hardware impulse neural network system
CN108985447A
Memristor-based low-power-consumption pulse convolutional neural network hardware architecture
CN112183739A
Neuron and neuromorphic system comprising same
CN113196307A
Pulse neural network chip architecture, chip and system
CN115392445A
