A method for realizing photon pulse neurons based on side-mode light injection DFB

Abstract

The present invention discloses a method for realizing photon pulse neurons based on side mode light injection DFB, including: setting the DFB to work at a target bias current; injecting a light modulation signal of target power into the DFB, so that the DFB is stimulated to perform a nonlinear response operation on the light modulation signal by neuron characteristics, and the DFB outputs the operation result of the photon pulse neuron; the target bias current is greater than the excitation threshold current of the DFB; the carrier of the light modulation signal is a CW signal of the target wavelength, and the modulation information carried by the light modulation signal is represented by the instantaneous power drop of the CW signal; the distance between the target wavelength and the target side mode wavelength is 0nm to 0.2nm, and the target side mode wavelength is selected from a plurality of side mode wavelengths output by the DFB in a free running state; the target power at least ensures that the CW signal can make the DFB enter an injection locked state. The present invention provides a better solution for realizing nonlinear calculations in the optical domain.

Description

Technical Field

[0001] The present invention belongs to the field of optical computing, and in particular relates to a method for realizing a photon pulse neuron based on side mode light injection DFB (distributed feedback semiconductor laser). Background Art

[0002] The rapid development of machine learning and artificial intelligence has generated a large amount of data that needs to be processed in real time. Processors based on the traditional von Neumann architecture are unable to meet the growing demand. Neuromorphic computing has become a non-von Neumann computing paradigm and has made significant progress. Photonic neuromorphic computing has shown significant advantages in speed and energy efficiency, but it is still in its infancy. In recent years, the successful demonstration of photonic neuromorphic chips has been mainly limited to the implementation of weights and matrix multiplication. Among them, optical linear computing has made significant progress, but optical nonlinear computing remains a major challenge.

[0003] In the prior art, when the photon pulse neural network based on the integrated DFB performs nonlinear operations on data, it is necessary to adjust the bias current of the DFB to load the information to the photon pulse neural network through electro-optical modulation for calculation. However, this introduces electro-optical modulation into the calculation process, which does not belong to the nonlinear calculation performed in the optical domain, and will cause the processing speed of the system where the photon pulse neural network is located to be delayed, and introduce additional power consumption. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a method for realizing photon pulse neurons based on side mode light injection DFB.

[0005] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0006] A method for realizing a photon pulse neuron based on side mode light injection DFB, comprising:

[0007] Set DFB to operate at the target bias current;

[0008] Injecting a light modulation signal of target power into the DFB, so that the DFB is stimulated to perform a nonlinear response operation on the light modulation signal with neuron characteristics, and the DFB outputs the operation result of the photon pulse neuron;

[0009] in,

[0010] The target bias current is greater than the excitation threshold current of the DFB;

[0011] The carrier of the optical modulation signal is a CW signal of a target wavelength, and the modulation information carried by the optical modulation signal is represented by the instantaneous power drop of the CW signal; the distance between the target wavelength and the target side mode wavelength is 0nm to 0.2nm, and the target side mode wavelength is selected from a plurality of side mode wavelengths output by the DFB in a free running state;

[0012] The target power at least ensures that the CW signal can enable the DFB to enter an injection locked state.

[0013] Optionally, the neuron characteristics include: threshold characteristics, time domain accumulation characteristics and refractory period characteristics of photon pulse neurons.

[0014] Optionally, the calibration method of the target bias current includes:

[0015] Set the bias current of the DFB from small to large and detect the output power of the DFB at the same time;

[0016] When the DFB is excited to output power, the corresponding bias current is the excitation threshold current;

[0017] The target bias current is calibrated according to the excitation threshold current.

[0018] Optionally, the target wavelength calibration method includes:

[0019] Setting the bias current of the DFB to the target bias current;

[0020] Make the DFB work in a free-running state and detect the output spectrum of the DFB;

[0021] From the output spectrum, one of the four side mode wavelengths on the left side of the main peak wavelength and the four side mode wavelengths on the right side of the main peak wavelength is selected as the target wavelength.

[0022] Optionally, selecting one of four side mode wavelengths on the left side of the main peak wavelength and four side mode wavelengths on the right side of the main peak wavelength from the output spectrum as the target wavelength comprises:

[0023] From the four side mode wavelengths on the left side of the main peak wavelength and the four side mode wavelengths on the right side of the main peak wavelength, the side mode wavelength with the strongest power is selected as the target wavelength.

[0024] Optionally, the target power calibration method includes:

[0025] Setting the bias current of the DFB to the target bias current;

[0026] Inputting a CW signal of the target wavelength into the DFB, and adjusting the power of the CW signal from small to large until the DFB reaches a homoclinic bifurcation point; the homoclinic bifurcation point is: the injection power boundary point when the DFB switches from an excitatory state to an injection locked state or from an injection locked state to an excitatory state; the DFB outputs periodic pulses in the excitatory state;

[0027] Above the homoclinic bifurcation point, calibrating the target power according to a desired DFB spike pulse response threshold;

[0028] The peak pulse response threshold is: the lowest instantaneous input power at which the DFB can be in an injection locked state.

[0029] Optionally, the greater the power of the CW signal input to the DFB, the greater the spike pulse response threshold; and the amplitude of the spike pulse output by the DFB in response to the instantaneous power drop of the CW signal is independent of the spike pulse response threshold.

[0030] Optionally, the method of injecting an optical modulation signal of target power into the DFB includes:

[0031] Inputting a CW signal of a target wavelength into an intensity modulator, and using the intensity modulator to modulate an electrical signal containing modulation information onto the CW signal to obtain an optical modulated signal;

[0032] The optical modulation signal of target power is input to the DFB.

[0033] Optionally, the intensity modulator comprises: a Mach-Zehnder modulator.

[0034] Optionally, injecting a light modulation signal of target power into the DFB so that the characteristics of the neurons excited by the DFB perform a nonlinear response operation on the light modulation signal and the DFB outputs the operation result of the photon pulse neuron, comprises:

[0035] An optical modulated signal of target power is injected into the DFB through an optical circulator, so that the neuron characteristics excited by the DFB perform nonlinear response operations on the optical modulated signal, and the optical circulator is used to derive the operation results of the photon pulse neurons output by the DFB.

[0036] In the photon pulse neuron realization method based on side mode light injection DFB provided by the present invention, a CW signal is used as a carrier to carry modulation information to form an optical modulation signal; by injecting an optical modulation signal of target power into the DFB and setting the DFB to work at a target bias current, the DFB can be stimulated to exhibit neuron characteristics, thereby using the DFB to realize nonlinear response calculation of the optical modulation signal and realize the function of the photon pulse neuron. Among them, the present invention does not need to realize information modulation by adjusting the bias current of the DFB, but directly modulates the information to the input of the photon pulse neuron, so that the nonlinear calculation process of the photon pulse neuron does not introduce optoelectronic / electro-optical (O / E / O) modulation, realizes nonlinear calculation in the optical domain, does not cause a delay in the processing speed of the system where the photon pulse neural network is located, and does not introduce additional power consumption to the DFB itself.

[0037] The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of a method for realizing a photon pulse neuron based on side mode light injection DFB provided by an embodiment of the present invention;

[0039] Figure 2 is a schematic diagram of inputting an optical modulated signal to a DFB using an optical circulator in an embodiment of the present invention;

[0040] Figure 3 The experimental device used to verify the characteristics of photon pulse neurons in an embodiment of the present invention is shown in FIG.

[0041] Figure 4 is a PI curve of a photon pulse neuron formed based on side mode light injection DFB in an embodiment of the present invention;

[0042] Figure 5 is the spectrum of the photon pulse neuron formed based on the side mode light injection DFB in the free running state in the embodiment of the present invention;

[0043] Figure 6 shows the pulse sequence randomly output by the DFB near the homoclinic bifurcation point;

[0044] Figure 7 shows the periodic pulses output by the DFB below the homoclinic bifurcation point;

[0045] Figure 8 FIG shows the frequency distribution of periodic pulses output by the DFB under different injection power conditions;

[0046] Fig. 9 It is an experimental result verifying the threshold characteristics of the photon pulse neuron formed in the embodiment of the present invention;

[0047] Fig.10 It is an experimental result verifying the time domain accumulation characteristics of the photon pulse neuron formed in the embodiment of the present invention;

[0048] Fig.11 It is an experimental result verifying the refractory period characteristics of the photon pulse neuron formed in the embodiment of the present invention;

[0049] Fig.12 This is an experimental result that verifies that the photon pulse neuron formed in the embodiment of the present invention has both threshold characteristics and time domain accumulation characteristics. DETAILED DESCRIPTION

[0050] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0051] In order to solve the problem of realizing nonlinear computing in the optical domain, many scholars have conducted extensive research on nonlinear computing of photonic neural networks. Among them, integrated devices based on vertical cavity surface emitting lasers (VCSELs), VCSELs with built-in saturable absorbers, micro-column lasers, DFBs, micro-rings or micro-resonators, and phase change materials can all be used to realize the basic components of photonic pulse neural networks, namely photonic pulse neurons.

[0052] Among them, the output power of photon pulse neurons based on microcolumn lasers or VCSELs is relatively low. When applied to multi-layer or deep optical pulse neural networks, additional amplification is required to compensate for the loss. Photon pulse neurons based on phase change materials lack the time domain accumulation capability that is critical for pulse processing. The existing photon pulse neurons based on DFB introduce photoelectric conversion into the nonlinear computing process, which will cause the processing speed of the system where the photon pulse neural network is located to be delayed and introduce additional power consumption.

[0053] Therefore, in order to solve the many problems existing in the existing photon pulse neurons, an embodiment of the present invention provides a method for realizing photon pulse neurons based on side mode light injection DFB, which provides a better solution for realizing nonlinear calculations in the optical domain.

[0054] See also Figure 1 As shown, the photon pulse neuron realization method based on side mode light injection DFB provided in an embodiment of the present invention includes the following steps:

[0055] Step 1, setting the DFB to operate at a target bias current I;

[0056] Step 2: Inject a light modulation signal of target power into the DFB, so that the DFB is stimulated to have neuron characteristics to perform nonlinear response operations on the light modulation signal, and the DFB outputs the operation results of the photon pulse neuron.

[0057] Among them, the target bias current I is greater than the DFB excitation threshold current I th ; The carrier of the optical modulation signal is a CW signal of the target wavelength, and the modulation information carried by the optical modulation signal is represented by the instantaneous power drop of the CW signal; the target wavelength λ CW and the target side mode wavelength λ s The distance between them is 0nm~0.2nm, and the target side mode wavelength λ s It is selected from several side mode wavelengths output by the DFB in a free-running state; the target power at least ensures that the CW signal can make the DFB enter an injection-locked state.

[0058] Exemplarily, in one implementation, a method of injecting an optical modulation signal of target power into a DFB may include:

[0059] A CW signal of a target wavelength is input to the intensity modulator, and an electrical signal containing modulation information is modulated onto the CW signal using the intensity modulator to obtain an optical modulated signal; then, the optical modulated signal of a target power is injected into the DFB.

[0060] The intensity modulator may include a Mach-Zehnder modulator, but is certainly not limited thereto.

[0061] In another implementation, the method of injecting the optical modulation signal of the target power into the DFB may include: inputting the optical modulation signal from the optical synapse device into the DFB.

[0062] In one implementation, Figure 2 As shown, step 2 may specifically include: injecting an optical modulation signal of target power into the DFB through an optical circulator, so that the neuron characteristics excited by the DFB perform nonlinear response operations on the optical modulation signal, and using the optical circulator to derive the operation results of the photon pulse neurons output by the DFB. Port 1 of the optical circulator is an optical input port, port 2 is an optical output port, and port 3 is a feedback port, and the operation results of the photon pulse neurons are obtained from the feedback port.

[0063] In the embodiment of the present invention, when the DFB is configured to work in the manner shown in step 1 and step 2, the neuron characteristics excited by the DFB include the threshold characteristics, time domain accumulation characteristics and refractory period characteristics of the photon pulse neuron.

[0064] Specifically, by building Figure 3The experimental device shown can verify that the photon pulse neuron implemented in the embodiment of the present invention has the above-mentioned multiple characteristics. In the experimental device, a current source (Bias) and a temperature controller (Tec) are used to set the bias current and operating temperature for the DFB; a tunable laser (TL) generates a CW (continuous wave) signal; the CW signal is injected into the intensity modulator (MZM); an arbitrary waveform generator (AWG) generates a custom external electrical stimulus as the information to be modulated, and the information is modulated onto the CW signal through an electrical amplifier (EA) and MZM. Two polarization controllers (PC) are used before and after the MZM to match the polarization state of the CW signal with that of the MZM and DFB, and the MZM is configured with a bias power supply VS. The adjustable optical attenuator (VOA) is used to adjust the optical power injected into the DFB. An optical coupler (OC) is used to split the modulated optical signal into two paths, one of which is injected into the DFB through a three-port optical circulator, and the other enters the photodetector PD1 for photoelectric conversion and then enters the real-time oscilloscope (OSC) for analysis; in addition, another optical coupler (OC) is used to split the output of the DFB into two paths, one of which enters the photodetector PD2 for photoelectric conversion and then enters the real-time oscilloscope (OSC) for comparison and analysis with the previous analysis signal, and the other is sent to the spectrometer (OSA) for spectral analysis. The splitting ratio of the above optical couplers is 50:50. The models of the various instruments used can be seen as follows:

[0065] meter model AWG Tektronix AWG70001A OSC Keysight DSOV334A or DSOZ592A OSA Advantest Q8384

[0066] Based on the above experimental device, a DFB model is selected and the power current (PI) curve of the DFB is first tested. Figure 4 As shown. Figure 4 As can be seen from the figure, as the bias current gradually increases, when the bias current I increases to 44mA, the DFB is excited to output power, so the excitation threshold current of the DFB is determined to be 44mA. Among them, when the bias current exceeds the excitation threshold current, the DFB spectrum of the DFB in the free running state is as follows: Figure 5 As shown. It can be seen that in the free-running state, the output spectrum of the DFB includes the main peak wavelength and several side mode wavelengths dispersedly arranged on both sides of the main peak wavelength, where λ max Indicates the side mode wavelength that is closest to the main peak wavelength and has the strongest power.

[0067] Then, the nonlinear dynamic effect of DFB under the condition of side mode injection is tested. Specifically, a side mode wavelength is selected, and a CW signal of the side mode wavelength is injected into the DFB. The initial injection power is above the homoclinic bifurcation point of the DFB, and then the injection power is gradually reduced. The homoclinic bifurcation point is defined as: the injection power dividing point when the DFB switches from the excitatory state to the injection locked state or from the injection locked state to the excitatory state. Among them, the DFB outputs periodic pulses in the excitatory state.

[0068] In the process of injecting a CW signal of the side mode wavelength into the DFB, as the injected power gradually decreases, the DFB will enter the excitatory state from the injection locked state and output periodic pulses. At this time, the output spectrum of the DFB is tested using a spectrometer, and the experimental results are as follows: Figure 6 As shown in the figure, sub-figures (a) to (c) show the pulses excited by the DFB under different injection power conditions. It can be seen that the number of pulses is different, which explains that under the side mode injection condition, due to the influence of the system inherent noise of the experimental device, the pulses excited by the DFB under different injection power conditions near the homoclinic bifurcation point are randomly generated, which may be one or more, and there is no regularity.

[0069] Then, the injected power is further reduced near the homoclinic bifurcation point, and periodic pulse output can be observed at the output of the DFB, such as Figure 7 For example, when the injection power is fixed and the bias current of the DFB is adjusted, it is found that the pulse frequency changes slightly with the bias current and the frequency is not fixed. Multiple frequency values ​​can be observed (such as Figure 8 This may also be related to the inherent noise of the experimental setup.

[0070] According to the above experiments, the following conclusion can be drawn: if the wavelength of the CW signal injected into the DFB is set near the side mode wavelength or equal to the side mode wavelength, then when the power of the CW signal injected into the DFB is below the homoclinic bifurcation point, the DFB will output periodic pulses. If the power of the CW signal injected into the DFB is set near the homoclinic bifurcation point, the DFB will randomly output pulses; if the power of the CW signal injected into the DFB is set above the homoclinic bifurcation point, the DFB enters the injection locked state and outputs a CW signal with stable power. Therefore, when the injected power is above the homoclinic bifurcation point, once the injected CW signal has an instantaneous power drop and drops below the homoclinic bifurcation point, the DFB will return to the state of outputting periodic pulses during the power drop period. The above experimental conclusions illustrate that the DFB can have the threshold characteristics of a photon pulse neuron. The following experiment is used to specifically verify the threshold characteristics of the photon pulse neuron possessed by the photon pulse neuron implemented in the embodiment of the present invention.

[0071] Specifically, the power of the CW signal injected into the DFB is set above the homoclinic bifurcation point. Then, the RF output of the AWG is turned on to modulate the custom external stimulus signal onto the optical carrier to form a modulated optical signal. In order to clearly demonstrate the threshold characteristics of the photon pulse neuron, the experiment defines different degrees of instantaneous power drop for the modulated optical signal (such as Fig. 9 The DFB is stimulated by using the stimulation intensity Kp = (V CW -V S ) / V CW , where V CW represents the amplitude of the CW signal, and Vs represents the power amplitude of the CW signal at the moment of instantaneous power drop.

[0072] Among them, when the power P of the modulated optical signal injected into the DFB inj =360.5μW, due to P inj The amplitude of the CW signal and the power amplitude of the power reduction stimulus are not enough to make the DFB enter the injection locking state. inj It is below the homoclinic bifurcation point, so DFB outputs periodic pulses at this time, such as Fig. 9 As shown in sub-figure (b1) in .

[0073] When P inj =380.5μW, the amplitude of the CW signal can make the DFB operate in the injection locked state, while the power amplitude of all power reduction stimuli cannot make the DFB reach the injection locked state. Therefore, for each power reduction stimulus, it will cause the DFB to generate a corresponding spike pulse, such as Fig. 9 As shown in sub-figure (b2)

[0074] When P inj When the power is further increased, since the stimulus intensity remains unchanged, the power amplitude of some power-down stimulus is large enough to keep the DFB in the injection-locked state and thus not output spike pulses. inj As the value of DFB increases, the number of spike pulses output by DFB decreases gradually. Fig. 9 As shown in the sub-graphs (b3) to (b7) in FIG. 1 , where sub-graphs (b3) to (b7) correspond to P inj is 411.2μw, 441.3μw, 460.4μw, 591.8μw; it can be seen that when P inj =591.8μW, at this time, the power amplitudes of all power reduction stimuli are relatively large, which can make the DFB reach the injection locking state. Therefore, the DFB does not output any spike pulses. This is because the DFB is completely injection locked by the input light stimulus.

[0075] Fig. 9The subgraphs (c1) to (c3) in P inj The output spectra of the DFB corresponding to sub-figures (b1), (b4), and (b7) during the increase process show that the mode suppression ratio (the difference between the two peaks in the spectrum) increases with P. inj increases with the increase of injection intensity.

[0076] according to Fig. 9 The experimental results show that, under the condition of a given bias current, the spike pulse response threshold of the DFB can be controlled by controlling the optical power of the CW signal injected into the DFB. The spike pulse response threshold is: the lowest instantaneous input power at which the DFB can be in the injection locked state. In other words, the photon pulse neuron implemented by the embodiment of the present invention not only has the threshold characteristics of the photon pulse neuron, but also has the characteristic of adjustable threshold.

[0077] Then, the time domain accumulation characteristics of the photon pulse neuron implemented in the embodiment of the present invention are verified. Specifically, Fig.10 As shown in the sub-figure (a) in Figure 1, three power reduction stimuli are modulated on the basis of the CW signal. The stimulus intensity is relatively low, so a single power reduction stimulus cannot trigger the spike pulse response of the DFB. However, two closely spaced power reduction stimuli and three closely spaced power reduction stimuli accumulated together can trigger the DFB output spike pulse response, as shown in Figure 1. Fig.10 As shown in sub-figure (b) in the figure, the injection power of DFB is 452.1μW at this time. That is to say, even if the Kp of a single power reduction stimulus is not large enough, two closely spaced power reduction stimuli can accumulate in time and exceed the spike pulse response threshold, thereby triggering the DFB to output a spike pulse; that is, within a certain time range, the power reduction stimulus can be accumulated in the time domain. When the power of the CW signal is increased, the spike pulse response threshold increases accordingly. Fig.10 In the sub-figure (a) of FIG, the temporal cumulative effect of two closely spaced power-down stimuli cannot reach the spike pulse response threshold, so only three closely spaced power-down stimuli can cause the DFB to produce a spike pulse response, as shown in FIG. Fig.10 As shown in sub-figure (c) in FIG, the injection power of DFB is 531.8 μW at this time.

[0078] In addition, according to Fig.10 The experimental method of sub-figures (a) to (c) in Figure 1 was used for 50 consecutive experiments. The time domain folding diagram drawn based on the experimental results is shown in the figure below. Fig.10 As shown in sub-figures (d) to (f) in , sub-figures (d) to (f) correspond to the results of 50 consecutive experiments in sub-figures (a) to (c), respectively.

[0079] based on Fig.10It can be seen from the experimental results that the photon pulse neuron realized based on the side mode light injection DFB in the embodiment of the present invention has a time domain accumulation characteristic.

[0080] Then, the refractory period characteristics of the photon pulse neuron implemented in the embodiment of the present invention are verified. Specifically, Fig.11 As shown in (a), 10 pairs of power reduction stimuli are defined based on the CW signal. The paired power reduction stimuli are referred to as pulse pairs here. The inter-spike-interval (ISI) of these pulse pairs gradually increases, which are 0.3ns, 0.5ns, 0.7ns, 0.9ns, 1.1ns, 1.3ns, 1.5ns, 1.7ns, 1.9ns, and 2.1ns. Under the condition of an injection power of 452.1μW, as shown in 11(b), the first 5 pairs of pulse pairs each trigger a spike pulse, and the last 5 pairs of pulse pairs each trigger two spike pulse responses. Among them, for each of the first 5 pairs of pulse pairs, the first power reduction stimulus triggers the DFB to output a spike pulse. Since the internal carriers of the DFB are consumed and need time to recover, the second power reduction stimulus cannot trigger the DFB to output another spike pulse in a short time. At this point, we know that the absolute refractory period of the photon pulse neuron is about 1.1ns, which is approximately equal to the reciprocal of the period of the periodic pulse output by the DFB in the sensitive state. For the last five pulse pairs, since the ISI is large enough and greater than the absolute refractory period, the two power reduction stimuli contained in each pulse pair can trigger spike pulses separately. In addition, from Fig.11 As can be seen in (b), the amplitude of the second spike pulse of the 6th, 7th and 8th pulse pairs is relatively small, because the ISI of the 6th, 7th and 8th pulse pairs is within the relative refractory period; the amplitudes of the two spike pulses triggered by the 9th and 10th pulse pairs are similar, because the ISI of the 9th and 10th pulse pairs is large and is already outside the relative refractory period.

[0081] In addition, according to Fig.11 The experimental method of sub-figures (a) to (b) in FIG. 50 was continuously carried out, and the time domain folding diagram drawn according to the experimental results is shown in FIG. Fig.11 As shown in sub-figures (c) to (d) in , sub-figures (c) to (d) correspond to the results of 50 consecutive experiments in sub-figures (a) to (b), respectively.

[0082] based on Fig.11 It can be seen from the experimental results that the photon pulse neuron realized based on the side mode light injection DFB in the embodiment of the present invention has a refractory period characteristic.

[0083] In addition, it is verified that the time domain accumulation characteristics and threshold characteristics of the photon pulse neuron realized by the side mode light injection DFB in the embodiment of the present invention can exist simultaneously. Specifically, three types of power reduction stimulation are modulated on the basis of CW signal, such as Fig.12 As shown in the sub-figure (a) in Figure 1, the first is a single power-down stimulus with a smaller Kp, the second is three closely spaced power-down stimuli with a smaller Kp, and the third is a single power-down stimulus with a larger Kp. Under the condition of an injected power of 446.9 μw, the DFB's response output to the three power-down stimuli is shown in Figure 1. Fig.12 As shown in sub-figure (b) in the figure, it can be seen that the first stimulus has a low stimulus intensity and does not reach the peak pulse response threshold of the DFB, so the DFB does not output a spike pulse; the second stimulus stimulates the DFB to output a spike pulse due to the time domain accumulation characteristics of the DFB as a photon pulse neuron; the third stimulus also stimulates the DFB to output a spike pulse due to its high stimulus intensity.

[0084] In addition, according to Fig.12 The experimental method of sub-figures (a) to (b) in FIG. 50 was continuously carried out, and the time domain folding diagram drawn according to the experimental results is shown in FIG. Fig.12 As shown in sub-figures (c) to (d) in , sub-figures (c) to (d) correspond to the results of 50 consecutive experiments in sub-figures (a) to (b), respectively.

[0085] based on Fig.12 It can be seen from the experimental results that the time domain accumulation characteristic and the threshold characteristic of the photon pulse neuron realized based on the side mode light injection DFB in the embodiment of the present invention exist simultaneously.

[0086] In the photon pulse neuron realization method based on side mode light injection DFB provided in the embodiment of the present invention, a CW signal is used as a carrier to carry modulation information to form an optical modulation signal; by injecting an optical modulation signal of target power into the DFB and setting the DFB to work at a target bias current, the DFB can be stimulated to exhibit neuron characteristics, thereby using the DFB to realize nonlinear response calculation of the optical modulation signal and realize the function of the photon pulse neuron. Among them, the embodiment of the present invention does not need to realize information modulation by adjusting the bias current of the DFB, but directly modulates the information as the input of the DFB, so that the nonlinear calculation process of the photon pulse neuron does not introduce optoelectronic / electro-optical (O / E / O) modulation, realizes nonlinear calculation in the optical domain, does not cause a delay in the processing speed of the system where the photon pulse neural network is located, and does not introduce additional power consumption to the DFB itself.

[0087] DFB is one of the most common light sources in optical communication systems. Its technology is relatively mature and easy to achieve large-scale integration. It can realize multi-wavelength DFB arrays with up to 64 channels. Therefore, it is easy to realize large-scale photon pulse neuron arrays based on DFB; in addition, the photon pulse neurons realized based on DFB can be planar integrated with the mainstream silicon photonic weight-based network; the output power of DFB is relatively high, and no additional power amplification compensation is required when applied to multi-layer or deep optical pulse neural networks. In summary, the photon pulse neurons realized based on DFB in the embodiment of the present invention are of great significance to the development of integrated photon pulse neural network chips, and provide some feasible solutions for the development of integrated photon pulse neural network chips.

[0088] In practical applications, the target bias current, target wavelength and target power mentioned above need to be calibrated in advance. The calibration methods of the target bias current, target wavelength and target power mentioned above are respectively described below.

[0089] The target bias current calibration method may include the following steps:

[0090] (I-1) setting the bias current of the DFB from small to large and detecting the output power of the DFB at the same time;

[0091] (I-2) When the DFB is excited to output power, the corresponding bias current is the excitation threshold current;

[0092] (I-3) Calibrate the target bias current according to the excitation threshold current.

[0093] Specifically, according to the excitation threshold current, a current value greater than the excitation threshold current is selected as the target bias current.

[0094] The target wavelength calibration method may include:

[0095] (F-1) Setting the bias current of the DFB to the target bias current;

[0096] (F-2) operating the DFB in a free-running state and detecting an output spectrum of the DFB;

[0097] (F-3) From the output spectrum, select one of the four side mode wavelengths on the left side of the main peak wavelength and one of the four side mode wavelengths on the right side of the main peak wavelength as the target wavelength.

[0098] In a preferred implementation, in step (F-3), the side mode wavelength with the strongest power can be selected from the four side mode wavelengths on the left side of the main peak wavelength and the four side mode wavelengths on the right side of the main peak wavelength (e.g. Figure 6 λ in inj ) as the target wavelength.

[0099] The target power calibration methods include:

[0100] (P-1) Setting the bias current of the DFB to the target bias current;

[0101] (P-2) Inputting a CW signal of a target wavelength to the DFB, and adjusting the power of the CW signal from small to large until the DFB reaches a homoclinic bifurcation point; the homoclinic bifurcation point is: the injection power boundary point when the DFB switches from an excitatory state to an injection locked state or from an injection locked state to an excitatory state; the DFB outputs periodic pulses in the excitatory state;

[0102] (P-3) Above the homoclinic bifurcation point, the target power is calibrated according to the desired peak pulse response threshold of the DFB; wherein the peak pulse response threshold is: the minimum instantaneous input power at which the DFB can be in the injection locked state.

[0103] In practice, the greater the power of the CW signal input to the DFB, the greater the peak pulse response threshold. Therefore, if a higher peak pulse response threshold is expected, the target power can be set larger in step (P-3). If a very high peak pulse response threshold is not expected, the target power can be set slightly larger than the homoclinic bifurcation point of the DFB in step (P-3).

[0104] In addition, when the DFB is actually used as a photon pulse neuron, the amplitude of the spike pulse output by the DFB in response to the instantaneous power drop of the CW signal has nothing to do with the spike pulse response threshold. Therefore, when selecting the desired spike pulse response threshold, it is only necessary to consider whether the instantaneous power drop of the CW signal can trigger the DFB to output a periodic pulse. As long as the DFB can enter the sensitive state to output a periodic pulse, the amplitude of the spike pulse output by the DFB is the same.

[0105] In summary, the photon pulse neuron based on side mode light injection DFB implemented by the embodiment of the present invention realizes nonlinear calculation in the optical domain, can be applied to multi-layer or deep optical pulse neural networks, is easy to achieve large-scale integration, and provides a partially feasible solution for the development of integrated photon pulse neural network chips.

[0106] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0107] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification.

[0108] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings and the disclosed content.

[0109] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the scope of protection of the present invention.

Claims

1. A method for realizing photon pulse neurons based on side mode light injection DFB, It is characterized in that include: Set DFB to operate at the target bias current; Injecting a light modulation signal of target power into the DFB, so that the DFB is stimulated to perform a nonlinear response operation on the light modulation signal with neuron characteristics, and the DFB outputs the operation result of the photon pulse neuron; in, The target bias current is greater than the excitation threshold current of the DFB; The carrier of the optical modulation signal is a CW signal of a target wavelength, and the modulation information carried by the optical modulation signal is represented by the instantaneous power drop of the CW signal; the distance between the target wavelength and the target side mode wavelength is 0nm to 0.2nm, and the target side mode wavelength is selected from a plurality of side mode wavelengths output by the DFB in a free running state; The target power at least ensures that the CW signal can enable the DFB to enter an injection locked state.

2. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 1, It is characterized in that The neuron characteristics include: threshold characteristics, time domain accumulation characteristics and refractory period characteristics of photon pulse neurons.

3. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 1, It is characterized in that The target bias current calibration method includes: Set the bias current of the DFB from small to large and detect the output power of the DFB at the same time; When the DFB is excited to output power, the corresponding bias current is the excitation threshold current; The target bias current is calibrated according to the excitation threshold current.

4. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 1, It is characterized in that The calibration method of the target wavelength includes: Setting the bias current of the DFB to the target bias current; Make the DFB work in a free-running state and detect the output spectrum of the DFB; From the output spectrum, one of the four side mode wavelengths on the left side of the main peak wavelength and the four side mode wavelengths on the right side of the main peak wavelength is selected as the target wavelength.

5. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 4, It is characterized in that Selecting one of four side mode wavelengths on the left side of the main peak wavelength and four side mode wavelengths on the right side of the main peak wavelength from the output spectrum as the target wavelength includes: From the four side mode wavelengths on the left side of the main peak wavelength and the four side mode wavelengths on the right side of the main peak wavelength, the side mode wavelength with the strongest power is selected as the target wavelength.

6. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 1, It is characterized in that The target power calibration method includes: Setting the bias current of the DFB to the target bias current; Inputting a CW signal of the target wavelength into the DFB, and adjusting the power of the CW signal from small to large until the DFB reaches a homoclinic bifurcation point; the homoclinic bifurcation point is: the injection power boundary point when the DFB switches from an excitatory state to an injection locked state or from an injection locked state to an excitatory state; the DFB outputs periodic pulses in the excitatory state; Above the homoclinic bifurcation point, calibrating the target power according to a desired DFB spike pulse response threshold; The peak pulse response threshold is: the lowest instantaneous input power at which the DFB can be in an injection locked state.

7. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 6, It is characterized in that The greater the power of the CW signal input to the DFB, the greater the spike pulse response threshold; and the amplitude of the spike pulse output by the DFB in response to the instantaneous power drop of the CW signal has nothing to do with the spike pulse response threshold.

8. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 1, It is characterized in that Methods of injecting an optical modulated signal of target power into the DFB include: Inputting a CW signal of a target wavelength into an intensity modulator, and using the intensity modulator to modulate an electrical signal containing modulation information onto the CW signal to obtain an optical modulated signal; The optical modulation signal of target power is input to the DFB.

9. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 8, It is characterized in that The intensity modulator includes: a Mach-Zehnder modulator.

10. The method for realizing photon pulse neurons based on side mode light injection DFB according to claim 9, It is characterized in that Injecting a light modulation signal of target power into the DFB so that the neuron characteristics excited by the DFB perform a nonlinear response operation on the light modulation signal and the DFB outputs the operation result of the photon pulse neuron, including: An optical modulated signal of target power is injected into the DFB through an optical circulator, so that the neuron characteristics excited by the DFB perform nonlinear response operations on the optical modulated signal, and the optical circulator is used to derive the operation results of the photon pulse neurons output by the DFB.

Images