Programmable lif-type somatic cell circuit, superconducting neuron circuit and electronic product

Abstract

The application provides a programmable LIF type somatic cell circuit, a superconducting neuron circuit and an electronic product. The programmable LIF type somatic cell circuit comprises a first Josephson junction, a first inductor, a first resistor and a second Josephson junction connected in series. The first Josephson junction receives a pulse signal of an input end and generates a magnetic flux quantum. When the cumulative current generated by the magnetic flux quantum and the first bias current applied to the first Josephson junction is greater than the critical current of the first Josephson junction, the first Josephson junction generates a phase flip and releases the magnetic flux quantum, and stores the magnetic flux quantum into a superconducting loop. When the cumulative current generated by the loop current and the second bias current applied to the second Josephson junction is greater than the critical current of the second Josephson junction, the second Josephson junction generates a phase flip and outputs a pulse signal. The application externally controls the threshold value of the LIF type somatic cell circuit through the bias current, can externally control the superconducting neuron circuit parameters, and has programmability.

Description

Technical Field

[0001] This application relates to the field of superconducting electronics technology, and in particular to a programmable LIF-type somatic cell circuit, a superconducting neuron circuit, and electronic products. Background Technology

[0002] With the continuous development of deep learning, the scientific community is constantly proposing new architectures for brain-inspired neural networks. Spiking Neural Networks (SNNs) are the third generation of neural networks. Compared with Artificial Neural Networks (ANNs), they can more deeply mimic the behavioral patterns of biological neurons, such as potential accumulation and pulse firing. Furthermore, due to the use of an event-driven triggering mechanism for pulse information processing similar to that of biological neurons, SNNs can achieve a simpler, more efficient, and lower-power hardware structure than traditional ANNs.

[0003] Superconducting single-flux quantum circuits (SFQs) are considered a powerful way to solve post-Moore's Law problems due to their ultra-high speed and low power consumption. Furthermore, SFQ circuits can be naturally matched with the working mode of spiking neurons. The threshold triggering characteristics of Josephson nodes and the near-lossless transmission of pulse signals perfectly simulate the firing and electrical signal propagation process of biological neurons, which makes the hardware design of SNNs simpler and more efficient.

[0004] Figure 1 This diagram illustrates a biological neuron system comprising a single neuron. This neuron includes synapses connecting the neurons and responsible for signal transmission and weight adjustment; dendrites that receive and integrate pulse signals from multiple synaptic inputs and transmit them to somatic cells; somatic cells that integrate the pulse signals transmitted from dendrites and determine whether to generate output pulses; and synapses that transmit the pulse signals generated and output by the somatic cells to other neurons. These four parts correspond to weight multiplication, accumulation, activation functions, and output in the hardware computation process, respectively.

[0005] The spiking neuron is the basic unit in a spiking neural network. Currently, there are two main design approaches for superconducting spiking neurons: digital and analog. Digital neuron design faces the challenge of high hardware resource consumption, and its ability to form large networks is limited by process integration. Analog neurons are primarily based on novel superconducting electronic devices, such as superconducting quantum phase-slip junctions and magnetic Josephson junctions. However, this approach suffers from high manufacturing difficulty and is not yet practical. Furthermore, analog designs exhibit weak anti-interference capabilities and poor resolution. In addition, superconducting neuron circuits generally suffer from poor programmability and low precision, and no feasible design scheme suitable for large-scale spiking neural networks has been provided. Summary of the Invention

[0006] The purpose of this application is to provide a programmable LIF-type somatic cell circuit to solve the technical problems of poor programmability and low precision that are common in superconducting neuron circuits.

[0007] To achieve the above and other related objectives, a first aspect of this application provides a programmable LIF-type somatic cell circuit. The programmable LIF-type somatic cell circuit includes: a first Josephson junction, a first inductor, a first resistor, and a second Josephson junction connected in series; wherein the first Josephson junction receives a pulse signal from an input terminal; when the accumulated current generated by the received pulse signal and a first bias current applied to the input terminal of the first Josephson junction exceeds a critical current of the first Josephson junction, the first Josephson junction undergoes a phase flip and releases a magnetic flux quantum, which is stored in a superconducting loop formed by the first inductor, the first Josephson junction, and the second Josephson junction; the first inductor induces a loop current in the magnetic flux quantum stored in the superconducting loop, and the loop current is consumed by the first resistor; when the accumulated current generated by the loop current and a second bias current applied to the input terminal of the second Josephson junction exceeds a critical current of the second Josephson junction, the second Josephson junction undergoes a phase flip and outputs a pulse signal.

[0008] In some embodiments of the first aspect of this application, the first bias current and the second bias current are adjustable currents, and the first bias current and the second bias current are adjusted according to a preset somatic cell threshold; the somatic cell threshold is the number of pulse signals at the input terminal that trigger the second Josephson junction to generate a phase flip.

[0009] In some embodiments of the first aspect of this application, adjusting the first bias current and the second bias current according to a preset somatic cell threshold includes the following formula:

[0010]

[0011] Wherein, N-1 represents the (N-1)th pulse signal received at the input terminal of the first Josephson junction, N represents the Nth pulse signal received at the input terminal of the first Josephson junction, i.e., the preset somatic cell threshold, R represents the resistance value of the first resistor, L represents the inductance of the first inductor, f represents the input frequency of the pulse signal, and i B1 Indicates the first bias current, i B2 Indicates the second bias current, i C2 This represents the critical current of the second Josephson junction.

[0012] In some embodiments of the first aspect of this application, the relationship between the first bias current, the second bias current, the critical current of the first Josephson junction, and the critical current of the second Josephson junction is as follows:

[0013] i B1 i B2 C1 C2

[0014] Among them, i B1 Indicates the first bias current, i B2 Indicates the second bias current, i C1 i represents the critical current of the first Josephson junction. C2 This represents the critical current of the second Josephson junction.

[0015] In some embodiments of the first aspect of this application, the loop current is consumed by the first resistor, and the loop current decays over time, as expressed by the formula:

[0016]

[0017] in, This refers to the magnetic flux quantum stored in the first inductor. This represents the loop current when the initial time is 0 within one frequency cycle. It represents the retention rate of the loop current at time t within one frequency cycle.

[0018] In some embodiments of the first aspect of this application, the loop current is a piecewise function, wherein the loop current generates a step increase every one frequency cycle, and the amount of the step increase in the loop current is the magnitude of the first bias current.

[0019] To achieve the above and other related objectives, a second aspect of this application provides a superconducting neuron circuit. The superconducting neuron circuit includes: an SFQ pulse signal unit, a plurality of synaptic units, dendritic units, axonal units, and the LIF-type somatic cell circuit described in any one of the first aspects of this application; wherein the synaptic units receive pulse signals output by the SFQ pulse signal unit; the dendritic units integrate the pulse signals received by the synaptic units; the LIF-type somatic cell circuit responds to the integrated pulse signals and outputs pulse signals; and the axonal units conduct and transmit the pulse signals output by the LIF-type somatic cell circuit units.

[0020] ​​In some embodiments of the second aspect of this application, each synaptic unit includes: a non-destructive readout unit and the LIF-type somatic cell unit; the LIF-type somatic cell unit is the LIF-type somatic cell circuit described in any one of the first aspects of this application; wherein, the non-destructive readout unit multiplies the received pulse signal and outputs it to the LIF-type somatic cell unit for accumulation; when the pulse signal received by the LIF-type somatic cell unit exceeds the somatic cell threshold, a feedback signal is generated to the non-destructive readout unit to stop the multiplication.

[0021] In some embodiments of the second aspect of this application, the somatic threshold of each synaptic unit is not the same; the weight of each synaptic unit is obtained and adjusted based on the different somatic thresholds.

[0022] To achieve the above and other related objectives, a third aspect of this application provides an electronic product. The electronic product includes at least the programmable LIF-type somatic cell circuit as described in any of the first aspects of this application and / or the superconducting neuron circuit as described in any of the second aspects of this application.

[0023] As described above, the programmable LIF-type somatic cell circuit, superconducting neuron circuit, and electronic product of this application have the following beneficial effects:

[0024] First, this application externally regulates the somatic cell threshold and synaptic weight of LIF-type somatic cells by using bias current, so that the parameters of the proposed superconducting neuron can be externally regulated, thereby making the superconducting neuron circuit programmable. Furthermore, by using externally set bias current to regulate internal parameters, the problem of decreased accuracy caused by simulation implementation methods is alleviated.

[0025] Second, this application utilizes a biomimetic neuron design scheme, making full use of the existing basic units in the current superconducting SFQ circuit, without the need to develop new devices or process fabrication conditions, thus reducing development costs.

[0026] Third, this application implements a LIF-type somatic cell circuit through simulation and a synaptic unit through a mixed-signal approach. It uses quantized magnetic flux in a superconducting loop to represent multiple neuron states, thus avoiding the use of complex digital logic gates and reducing hardware overhead and circuit power consumption. Through the design of circuit parameters, the superconducting loop can receive multiple magnetic flux inputs, thereby enabling the neuron to have high parameter accuracy. Therefore, the superconducting neuron circuit of this application has significant advantages in performance, power consumption, and footprint. Attached Figure Description

[0027] Figure 1 The diagram shows a simple biological nervous system and its artificial abstract model.

[0028] Figure 2 The diagram shown is a schematic diagram of the circuit structure of a programmable LIF-type somatic cell circuit according to an embodiment of this application.

[0029] Figure 3 The diagram shown is a schematic representation of a superconducting neuron circuit in one embodiment of this application.

[0030] Figure 4 The diagram shown is a schematic diagram of the circuit structure of a synaptic unit in one embodiment of this application.

[0031] Figure 5 The diagram shown is a schematic diagram of the circuit structure of a 2-input-2-output superconducting neuron circuit according to an embodiment of this application.

[0032] Figure 6 The diagram shown is a schematic diagram of the circuit structure of a 4-input-4-output superconducting neuron circuit according to an embodiment of this application.

[0033] Component designation explanation

[0034] 10 LIF-type somatic cell circuits

[0035] 300 Superconducting Neuron Circuit

[0036] 310 SFQ pulse signal unit

[0037] 320 synaptic units

[0038] 330 dendritic units

[0039] 340 axon units Detailed Implementation

[0040] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0041] Before providing a further detailed description of the present invention, the nouns and terms used in the embodiments of the present invention are explained, and the nouns and terms used in the embodiments of the present invention are subject to the following interpretations:

[0042] Superconducting neuron circuits are circuits that simulate the behavior of biological neurons, utilizing the properties of superconducting materials to achieve efficient information processing. Their core is the Josephson junction (JJ), a structure consisting of two layers of superconductor sandwiching a layer of insulator (SIS). It possesses threshold characteristics similar to neurons, and the Josephson transmission circuit formed by it has the capability for lossless pulse signal transmission. By designing the circuit, the action potentials of neurons and the weighted summation operations of synapses can be simulated.

[0043] The confluence buffer (CB) of the superconducting SFQ circuit can be used to achieve the convergence of multiple pulses and complete the dendritic accumulation function. The Josephson transmission line (JTL) and splitter (SPL) can be used to achieve the transmission and output of pulses and realize the output function of axons. Based on the implementation of the LIF-type somatic cell circuit unit and synaptic unit provided in the embodiments of this application, in addition to realizing the activation function and weight multiplication, the somatic cell threshold and synaptic weight of the LIF-type somatic cell circuit unit can be externally adjusted based on the external bias current, so that the parameters of the proposed superconducting neuron circuit can be externally adjusted, thus possessing programmability.

[0044] Please see Figures 2-6 It should be noted that the illustrations provided in this embodiment are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0045] Since simulated neurons are primarily based on novel superconducting electronic devices, such as superconducting quantum phase-slip junctions and magnetic Josephson junctions, this approach faces challenges in manufacturing and is currently not practical. Furthermore, the simulation designs suffer from weak anti-interference capabilities and poor resolution. In addition, superconducting neuron circuits generally suffer from poor programmability and low precision, and no feasible design scheme suitable for large-scale spiking neural networks has been provided.

[0046] For the reasons mentioned above, this application provides a programmable LIF somatic cell circuit, a superconducting neuron circuit, and an electronic product. By externally controlling the somatic cell threshold and synaptic weights of LIF somatic cells through bias current, the parameters of the proposed superconducting neuron can be externally controlled, thereby enabling the superconducting neuron circuit to have programmability. Furthermore, by controlling the internal parameters through an externally set bias current, the problem of decreased accuracy caused by the simulation implementation method is alleviated.

[0047] To facilitate understanding of the embodiments of this application, firstly, in conjunction with Figure 2 Detailed explanation. Figure 2 A schematic diagram of the circuit structure of a programmable LIF-type somatic cell circuit according to an embodiment of this application is shown.

[0048] like Figure 2 As shown, the programmable LIF-type somatic cell circuit 10 includes: a first Josephson junction J1, a first inductor L, a first resistor R, and a second Josephson junction J2 connected in series.

[0049] The first Josephson junction J1 receives a pulse signal from its input terminal. When the received pulse signal and the first bias current i applied to the input terminal of the first Josephson junction J1... B1 When the cumulative current generated by the combination exceeds the critical current of the first Josephson junction J1, the first Josephson junction J1 undergoes a phase flip and releases a magnetic flux quantum, which is stored in the superconducting loop formed by the first inductor L, the first Josephson junction J1, and the second Josephson junction J2. The first bias current i B1 The first bias current i is applied to the input terminal of the first Josephson junction J1 and grounded through the first Josephson junction J1. B1 The ground can be established through the first inductor L, the first resistor R, and the second Josephson junction J2.

[0050] Wherein, the first Josephson junction J1 serves as an accumulation junction, having a relatively small critical current, and the first Josephson junction J1 is biased by the first bias current i B1 The bias is applied so that the first Josephson junction J1 can respond quickly to each input pulse signal and generate a phase flip to inject the magnetic flux quantum into the loop of the programmable LIF somatic circuit 10 to increase the loop current.

[0051] The first inductor L induces the magnetic flux quantum stored in the superconducting loop into a loop current, which is consumed by the first resistor R.

[0052] When the loop current and the second bias current i applied to the input terminal of the second Josephson junction J2 B2 When the cumulative current generated by both exceeds the critical current of the second Josephson junction J2, the second Josephson junction J2 undergoes a phase flip and outputs a pulse signal. The second bias current i B2 It is applied to the input terminal of the second Josephson junction J2 and grounded through the second Josephson junction J2.

[0053] Among them, the second Josephson junction J2 serves as the trigger junction, possessing a relatively large critical current. The second Josephson junction J2 is the direct component for pulse delivery, and when the loop current and the second bias current i... B2 When the cumulative current generated by the second Josephson junction J2 exceeds its critical current, the second Josephson junction J2 is triggered, thereby generating a pulse output.

[0054] Thus, the programmable LIF-type somatic cell circuit 10 constitutes a typical leaky-integrate-and-fire (LIF) somatic cell model in SNN, wherein the loop current is characterized by a characteristic time... Attenuation is performed.

[0055] In practical applications, key parameters of the LIF-type somatic cell circuit 10 include the threshold and leakage rate. The threshold is defined as the number of pulse signals received at the input terminal of the first Josephson junction required to trigger the second Josephson junction J2, and can be adjusted by changing the first bias current i. B1 and the second bias current i B2 The number of input pulse signals required to trigger the second Josephson junction J2 is changed, and the somatic cell threshold is adjusted to achieve a programmable function; the leakage rate of the somatic cell is determined by the characteristic time of loop current decay. By definition, the leakage rate also determines the frequency response of the LIF-type somatic circuit 10. If the interval between the input pulse signals is greater than the decay time, the loop current decays to 0 within the time interval between two inputs, and the current cannot be effectively accumulated, so that the second Josephson junction J2 cannot be triggered. If the interval between the input pulse signals is less than the decay time, the loop will accumulate current quickly, further accelerating the triggering of the second Josephson junction J2.

[0056] In some implementations, the first bias current i B1 and the second bias current i B2 To provide an adjustable current, the first bias current i is adjusted according to a preset somatic cell threshold. B1 and the second bias current i B2 The somatic cell threshold is the number of pulse signals at the input terminal that trigger the second Josephson junction J2 to generate a phase flip.

[0057] In this embodiment, the first bias current i is adjusted according to a preset somatic cell threshold. B1 The second bias current includes the following formula:

[0058]

[0059] Wherein, N-1 represents the (N-1)th pulse signal received at the input terminal of the first Josephson junction, N represents the Nth pulse signal received at the input terminal of the first Josephson junction, i.e., the preset somatic cell threshold, R represents the resistance value of the first resistor, L represents the inductance of the first inductor, f represents the input frequency of the pulse signal, and i B1 Indicates the first bias current, i B2 Indicates the second bias current, i C2 This represents the critical current of the second Josephson junction.

[0060] In this embodiment, the relationship between the first bias current, the second bias current, the critical current of the first Josephson junction, and the critical current of the second Josephson junction is as follows:

[0061] i B1 i B2 C1 C2 (2)

[0062] Among them, i B1 Indicates the first bias current, i B2 Indicates the second bias current, i C1 i represents the critical current of the first Josephson junction. C2 This represents the critical current of the second Josephson junction.

[0063] In some implementations, the loop current is consumed by the first resistor, and the loop current decays over time. The formula for the decay of the loop current over time is:

[0064]

[0065] in, This refers to the magnetic flux quantum stored in the first inductor. This represents the loop current when the initial time is 0 within one frequency cycle. It represents the retention rate of the loop current at time t within one frequency cycle.

[0066] In this embodiment, the loop current is a piecewise function, and the loop current generates a step increase every one frequency cycle. The amount of the step increase in the loop current is the magnitude of the first bias current.

[0067] The design of the parameters for the somatic cell threshold and the leakage rate will be described in detail below.

[0068] First, a detailed introduction to the somatic cell threshold will be given.

[0069] ​​The LIF-type somatic cell circuit 10 can be controlled by the first bias current i B1 and the second bias current i B2 To regulate the somatic cell threshold N. Figure 2 For example, the LIF-type somatic cell circuit 10 is formed by the first Josephson junction - the first inductor - the first resistor - the first Josephson structure forming a loop. In order to complete the work of the somatic cell, the current relationship between the first bias current and the second bias current and the critical current of the first Josephson junction and the critical current of the second Josephson junction is designed as shown in formula (2).

[0070] According to Kirchhoff's current law and Josephson's equation, we can obtain:

[0071]

[0072]

[0073] in, The phase difference between the first Josephson junction J1 and the second Josephson junction J2 is i. LR This is the loop current in the LR branch. When the loop current does not reach the critical current of the second Josephson junction J2, the second Josephson junction J2 does not flip. Therefore, we obtain information about i. LR The inequality is:

[0074] i LR C2 -i B2 (6)

[0075] When magnetic flux is added to the loop through the first Josephson junction J1, the first Josephson junction J1 undergoes a phase reversal, and the current flowing through the first Josephson junction J1 is minimized. Therefore, the maximum value of the loop current during magnetic flux injection is obtained as follows:

[0076] Max (i LR ) = i B1 (7)

[0077] Because of the resistance in the loop, a portion of the magnetic flux escapes from the loop through the first resistor R. Therefore, the loop no longer possesses the continuity of the superconducting loop wave function, and the loop magnetic flux Φ L With the loop current i LR It decays over time. The derivation of the loop current formula is explained below:

[0078] According to Ohm's law and Josephson's equation, the current flowing through the resistor can be obtained as:

[0079] ​

[0080] When the loop is stable With a fixed phase difference of 2nπ, n = 0, ±1, ±2, ..., and n does not change with time, therefore formula (8) can be written as:

[0081]

[0082] For the loop magnetic flux Φ L The resulting phase difference is therefore equivalent to equation (9):

[0083]

[0084] The induced current generated in the inductor by the loop magnetic flux is:

[0085]

[0086] Since the first inductor L and the first resistor R are on the same branch, therefore according to i L =i R =i LR We can obtain:

[0087]

[0088] Solving the differential equation yields:

[0089]

[0090] This formula represents the loop flux with characteristic time. The loop current decays over time as shown in formula (3) by attenuation.

[0091] Based on formula (3), when the LIF-type somatic cell circuit 10 is in operation, a magnetic flux is injected into the loop at a frequency f. Due to the resistance of the loop, the current decays with time according to formula (3), therefore the loop current is a piecewise function. Every [time period]... A periodic magnetic flux is injected into the loop, causing a step jump in the loop current, with the current increment being:

[0092]

[0093] Subsequently During the time interval, the loop current decays according to formula (3), and the current decay amount is:

[0094]

[0095] Therefore, when the (N-1)th pulse is input, the loop current is:

[0096]

[0097] Combining this expression with formula (6), we obtain the following inequality:

[0098]

[0099] This formula indicates the condition under which the second Josephson junction J2 is not triggered after the (N-1)th magnetic flux is input into the loop. Based on this relationship, the condition under which the second Josephson junction J2 is triggered after the Nth magnetic flux is input can be obtained as follows:

[0100]

[0101] The first bias current i can be obtained from formulas (17) and (18). B1 The second bias current i B2 The relationship with the somatic cell threshold N is shown in Equation (1).

[0102] As can be seen, when the circuit element parameters and the frequency f of the input pulse signal are fixed, the somatic cell threshold N is only related to the external first bias current i. B1 The second bias current i B2 Related to increasing the first bias current i B1 The second bias current i B2 The somatic cell threshold decreases. The range of bias current values ​​corresponding to the required somatic cell threshold is determined by the above inequality (1).

[0103] Furthermore, the leakage rate is described in detail.

[0104] From formula (3), it can be seen that the characteristic time of somatic cell current decay is: Therefore, the desired somatic cell leakage rate can be obtained by designing the first inductor L and the first resistor R in the LIF-type somatic cell circuit 10. Furthermore, as shown in equation (15), the input pulse frequency f affects the loop current attenuation, which in turn affects the threshold performance of the somatic cell, as shown in equation (1).

[0105] This application also provides a superconducting neuron circuit 300. Please refer to [link / reference needed]. Figure 3 The diagram shown is a schematic diagram of the structure of a superconducting neuron circuit 300 according to an embodiment of this application.

[0106] like Figure 3 As shown, the superconducting neuron circuit 300 includes: an SFQ pulse signal unit 310, several synaptic units 320, dendritic units 330, axonal units 340, and a LIF-type somatic cell circuit 10.

[0107] In this embodiment, the synaptic unit 320 receives the pulse signal output by the SFQ pulse signal unit 310; the dendritic unit 330 integrates the pulse signal received by the synaptic unit 320; the LIF somatic cell circuit 10 responds to the integrated pulse signal and outputs a pulse signal; the axon unit 340 conducts and transmits the pulse signal output by the LIF somatic cell circuit 10.

[0108] In some implementations, each of the synaptic units 320 includes: a non-destructive readout unit and a LIF-type somatic cell unit, wherein the LIF-type somatic cell unit employs the LIF-type somatic cell circuit 10. The non-destructive readout unit multiplies the received pulse signal and outputs it to the LIF-type somatic cell unit for accumulation; when the pulse signal received by the LIF-type somatic cell unit exceeds the somatic cell threshold, a feedback signal is generated to the non-destructive readout unit to stop the multiplication.

[0109] In this embodiment, each synaptic unit 320 corresponds to the output of one SFQ pulse signal unit 310. Since the somatic cell thresholds of the LIF-type somatic cell circuit 10 units in each synaptic unit 320 are not the same, the weights of each synaptic unit 320 are obtained and adjusted based on the different somatic cell thresholds.

[0110] like Figure 4 As shown, the synaptic unit 320 includes: a non-destructive readout (NDRO) unit and a LIF-type somatic cell unit.

[0111] The non-destructive readout unit uses the T1 signal at the clk terminal as a multiplication reference to multiply the IN signal input to the set terminal. The output pulse signal is accumulated by the LIF-type somatic cell unit. When the number of input pulse signals to the LIF-type somatic cell unit exceeds the somatic cell threshold, an output is generated and fed back to the reset terminal of the non-destructive readout unit to stop the multiplication.

[0112] In this embodiment, the parameter design of the LIF-type somatic cell unit is the same as that of the LIF-type somatic cell circuit 10 described above, so it will not be described in detail here.

[0113] In the above manner, the somatic cell threshold is converted into the weight of the synaptic unit 320. Similar to the LIF-type somatic cell circuit 10, the weight can be programmed by the bias current. When a large bias current is applied to the synaptic unit 320, the synaptic unit 320 has a smaller weight performance; conversely, the synaptic unit 320 has a larger weight performance.

[0114] The following example, a 2-input-2-output neuron, illustrates how to construct a superconducting neuron circuit 300.

[0115] Please see Figure 5 The diagram shows a schematic of the circuit structure of the 2-input-2-output superconducting neuron circuit 300 described in this embodiment of the application.

[0116] like Figure 5 As shown, the SFQ pulse signal unit 310 includes one SPL unit and two JTL units to input two pulse signals through the two JTL units;

[0117] Synaptic unit 320: The 2-input-2-output superconducting neuron circuit 300 includes two synaptic units 320, each connected to a pulse signal;

[0118] Dendritic unit 330: includes a CB unit to ensure the transmission of pulse signals;

[0119] LIF-type somatic cell circuit 10: includes a LIF-type somatic cell circuit 10 to integrate pulse signals transmitted by dendritic unit 330 and determine whether to generate output pulses;

[0120] Axon unit 340: includes 3 JTL units and 1 SPL unit to conduct and transmit the pulse signal output by LIF somatic circuit 10.

[0121] In this embodiment, a 2-input-2-output superconducting neuron circuit 300 can be constructed by cascading the SFQ pulse signal unit 310, synaptic unit 320, dendritic unit 330, LIF-type somatic cell circuit 10, and axonal unit 340.

[0122] Please see Figure 6 The diagram shows a schematic of the circuit structure of the 4-input-4-output superconducting neuron circuit 300 described in this embodiment of the application.

[0123] like Figure 6 As shown, the SFQ pulse signal unit 310 includes 3 SPL units and 5 JTL units to input four pulse signals through the JIL units;

[0124] Synaptic unit 320: The 4-input-4-output superconducting neuron circuit 300 includes 4 synaptic units 320, each connected to a pulse signal;

[0125] Dendritic unit 330: includes a CB unit to ensure the transmission of pulse signals;

[0126] LIF-type somatic cell circuit 10: includes a LIF-type somatic cell circuit 10 to integrate pulse signals transmitted by dendritic unit 330 and determine whether to generate output pulses;

[0127] Axon unit 340: includes one JTL unit and three SPL units to conduct and transmit the pulse signals output by the LIF somatic cell circuit 10.

[0128] In the embodiments of this application, terms such as "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, "first Josephson node" and "second Josephson node" are only used to distinguish Josephson nodes in different positions and do not limit their order. Those skilled in the art will understand that terms such as "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply that they are different.

[0129] It should be noted that, in the embodiments of this application, the words "exemplary" or "for example" indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0130] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0131] This application also provides an electronic product, which includes the programmable LIF-type somatic cell circuit or superconducting neuron circuit of this application. The programmable LIF-type somatic cell circuit and superconducting neuron circuit of this application have the advantages of high efficiency and programmability.

[0132] In summary, this application provides a programmable LIF-type somatic cell circuit 10, a superconducting neuron circuit 300, and an electronic product. The application externally controls the somatic cell threshold and synaptic weights of the LIF-type somatic cells via bias current, enabling external control of the proposed superconducting neuron parameters. This makes the superconducting neuron circuit 300 programmable. Furthermore, the method of externally controlling internal parameters via bias current alleviates the accuracy degradation problem caused by analog implementation methods. This application utilizes a biomimetic neuron design, fully leveraging existing basic units in current superconducting SFQ circuits, eliminating the need to develop new devices or fabrication processes, thus reducing development costs. This application implements the LIF-type somatic cell circuit 10 using an analog-digital hybrid approach, and the synaptic unit 320 using quantized magnetic flux in the superconducting loop to express multiple neuron states, thereby avoiding the use of complex digital logic gates and reducing hardware overhead and circuit power consumption. Through circuit parameter design, the superconducting loop can receive multiple magnetic flux inputs, resulting in high parameter accuracy for the neuron. Therefore, the superconducting neuron circuit 300 of this application has significant advantages in performance, power consumption, and footprint. Therefore, this application effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0133] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A programmable LIF-type somatic cell circuit, characterized in that, The programmable LIF-type somatic cell circuit includes: a first Josephson junction, a first inductor, a first resistor, and a second Josephson junction connected in series; wherein... The first Josephson junction receives a pulse signal from the input terminal. When the cumulative current generated by the received pulse signal and the first bias current applied to the input terminal of the first Josephson junction exceeds the critical current of the first Josephson junction, the first Josephson junction generates a phase flip and releases a magnetic flux quantum. The magnetic flux quantum is stored in the superconducting loop formed by the first inductor, the first Josephson junction and the second Josephson junction. The first inductor induces the magnetic flux quantum stored in the superconducting loop into a loop current, which is consumed by the first resistor; When the cumulative current generated by the loop current and the second bias current applied to the input terminal of the second Josephson junction exceeds the critical current of the second Josephson junction, the second Josephson junction generates a phase flip and outputs a pulse signal.

2. The programmable LIF-type somatic cell circuit according to claim 1, characterized in that, The first bias current and the second bias current are adjustable currents, and the first bias current and the second bias current are adjusted according to a preset somatic cell threshold. The somatic cell threshold is the number of pulse signals at the input terminal that trigger the second Josephson junction to generate a phase flip.

3. The programmable LIF-type somatic cell circuit according to claim 2, characterized in that, Adjusting the first bias current and the second bias current according to a preset somatic cell threshold includes the following formula: Wherein, N-1 represents the (N-1)th pulse signal received at the input terminal of the first Josephson junction, N represents the Nth pulse signal received at the input terminal of the first Josephson junction, i.e., the preset somatic cell threshold, R represents the resistance value of the first resistor, L represents the inductance of the first inductor, f represents the input frequency of the pulse signal, and i B1 Indicates the first bias current, i B2 Indicates the second bias current, i C2 This represents the critical current of the second Josephson junction.

4. The programmable LIF-type somatic cell circuit according to claim 3, characterized in that, The relationship between the first bias current, the second bias current, the critical current of the first Josephson junction, and the critical current of the second Josephson junction is as follows: i B1 ,i B2 <i C1 <i C2 Among them, i B1 Indicates the first bias current, i B2 Indicates the second bias current, i C1 i represents the critical current of the first Josephson junction. C2 This represents the critical current of the second Josephson junction.

5. The programmable LIF-type somatic cell circuit according to claim 1, characterized in that, The loop current is consumed by the first resistor, so the loop current decays over time. The formula for the decay of the loop current over time is: in, This refers to the magnetic flux quantum stored in the first inductor. This represents the loop current when the initial time is 0 within one frequency cycle. It represents the retention rate of the loop current at time t within one frequency cycle.

6. The programmable LIF-type somatic cell circuit according to claim 5, characterized in that, The loop current is a piecewise function, and the loop current generates a step increase at every frequency cycle. The amount of the step increase in the loop current is equal to the magnitude of the first bias current.

7. A superconducting neuron circuit, characterized in that, The superconducting neuron circuit includes: an SFQ pulse signal unit, several synaptic units, dendritic units, axonal units, and the LIF-type somatic cell circuit according to any one of claims 1 to 6; wherein The synaptic unit receives the pulse signal output by the SFQ pulse signal unit; The dendritic unit integrates the pulse signal received by the synaptic unit; The LIF-type somatic cell circuit responds to the integrated pulse signal and outputs a pulse signal. The axon unit conducts and transmits the pulse signal output by the LIF-type somatic cell circuit.

8. The superconducting neuron circuit according to claim 7, characterized in that, Each of the synaptic units comprises: a non-destructive readout unit and a LIF-type somatic cell unit; the LIF-type somatic cell unit is the LIF-type somatic cell circuit according to any one of claims 1 to 6; wherein, The non-destructive readout unit amplifies the received pulse signal and outputs it to the LIF somatic cell unit for accumulation; When the pulse signal received by the LIF somatic cell unit exceeds the somatic cell threshold, a feedback signal is generated to the non-destructive readout unit so that the non-destructive readout unit stops multiplying.

9. The superconducting neuron circuit according to claim 8, characterized in that, The somatic cell thresholds of each synaptic unit are not the same; the weights of each synaptic unit are obtained and adjusted based on the different somatic cell thresholds.

10. An electronic product, characterized in that, The electronic product includes at least the programmable LIF-type somatic cell circuit as described in any one of claims 1 to 6 and / or the superconducting neuron circuit as described in any one of claims 7 to 9.

Images