Realization method of monolithic integrated photon pulse neuron based on DFB-SA laser

Abstract

The present invention discloses a DFB-SA-based monolithic integrated photon pulse neuron implementation method, including: performing power testing on a DFB-SA laser to determine the operating voltage and bias current range of the DFB-SA laser; performing spectrum testing on the DFB-SA laser to determine the wavelength range of the DFB-SA laser itself; determining the wavelength range of the optical signal to be injected according to the wavelength range of the DFB-SA laser itself; applying a bias current to the gain region of the DFB-SA laser based on the determined bias current range, and applying a reverse bias voltage to the SA region of the DFB-SA laser based on the determined operating voltage, and under the above conditions, injecting an optical signal into the DFB-SA laser according to the determined injection optical signal wavelength range, observing the output spectrum of the DFB-SA laser, to determine the injected optical signal waveform corresponding to the target to be processed, thereby realizing the function of the photon pulse neuron. The method is simple to implement, reduces system complexity and power consumption, and has a high output power.

Description

Technical Field

[0001] The present invention belongs to the technical field of photon pulse neural networks, and in particular relates to a method for realizing a monolithic integrated photon pulse neuron based on DFB-SA. Background Art

[0002] Artificial neural networks (ANNs) have achieved great success in performing artificial intelligence tasks such as image recognition, object detection, and object tracking. However, running ANN algorithms on traditional von Neumann computers suffers from huge power consumption and large processing delays. Spiking neural networks (SNNs), generally considered as the third generation of neural networks, have the characteristics of low power consumption and low latency when running on neuromorphic hardware platforms. As a non-von Neumann paradigm, neuromorphic computing has made significant progress in recent years. Compared with electronic computing, photonic neuromorphic computing has obvious advantages in terms of fast speed and high energy efficiency, but it is still in its infancy.

[0003] Linear computing and nonlinear computing are two basic functions of photonic neuromorphic computing. Wavelength division multiplexing architecture based on microring resonators (MRRs) and coherent architecture based on Mach-Zehnder interferometers (MZIs) are two mainstream methods to implement linear matrix-vector multiplication. Semiconductor lasers have become the main devices for photonic nonlinear neuromorphic computing due to their biologically nonlinear neuron-like functions.

[0004] However, existing schemes for realizing photon pulse neurons based on semiconductor lasers still have some defects. For example, the output power of microcolumn laser neurons is relatively low, and when applied to multi-layer or deep optical pulse neural networks, additional amplification may be required to compensate for losses. Photon pulse neurons based on integrated distributed feedback semiconductor lasers require photoelectric conversion, which increases the complexity and power consumption of the system. Multi-longitudinal mode photon pulse neurons based on built-in saturated absorbers are compatible with broadband weighted devices, but are not suitable for narrow bandwidth weighted devices. Lasers based on DFB are not easy to achieve cascade transmission. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a method for realizing a monolithic integrated photon pulse neuron based on DFB-SA. The technical problem to be solved by the present invention is achieved by the following technical solutions:

[0006] A method for realizing a monolithic integrated photon pulse neuron based on a DFB-SA laser, comprising:

[0007] Step 1: Perform a power test on the DFB-SA laser to determine the operating voltage and bias current range of the DFB-SA laser;

[0008] Step 2: Perform spectrum test on the DFB-SA laser to determine the wavelength range of the DFB-SA laser.

[0009] Step 3: Determine the wavelength range of the optical signal to be injected according to the wavelength range of the DFB-SA laser itself;

[0010] Step 4: Based on the determined bias current range, a bias current is applied to the gain region of the DFB-SA laser, and based on the determined operating voltage, a reverse bias voltage is applied to the SA region of the DFB-SA laser. Under the above conditions, an optical signal is injected into the DFB-SA laser according to the determined injection optical signal wavelength range, and the output spectrum of the DFB-SA laser is observed to determine the injected optical signal waveform corresponding to the target to be processed, thereby realizing the function of the photon pulse neuron.

[0011] In one embodiment of the present invention, step 1 comprises:

[0012] For a certain DFB-SA laser, a reverse bias voltage and a bias current are applied to its SA region and gain region respectively, and the operating voltage of the DFB-SA laser is obtained by traversal;

[0013] The reverse bias voltage of the DFB-SA laser is set to a fixed value based on the working voltage, and the bias current of the DFB-SA laser is set to change from small to large, and the corresponding output power is detected at the same time; when the DFB-SA laser is excited to output power, the corresponding bias current is recorded as the excitation threshold current;

[0014] Continue to increase the current and use an oscilloscope to observe the self-pulsation state of the DFB-SA laser to find the bias current that can make the laser emit self-pulsation, which is recorded as the self-pulsation threshold current;

[0015] A bias current range is determined according to the excitation threshold current and the self-pulse threshold current.

[0016] In one embodiment of the present invention, the operating voltage is -6V to 0V.

[0017] In one embodiment of the present invention, determining the bias current range according to the excitation threshold current and the self-pulse threshold current includes:

[0018] The bias current is set between the excitation threshold current and the self-pulsation threshold current to enable the DFB-SA laser to operate in an excitation state.

[0019] In one embodiment of the present invention, step 2 comprises:

[0020] The reverse bias voltage of the DFB-SA laser is set to a fixed value, and the bias current of the DFB-SA laser is changed, and the corresponding output spectrum is detected at the same time to detect the wavelength range of the DFB-SA laser itself.

[0021] In one embodiment of the present invention, in step 4, the information carried by the injected optical signal is represented by an instantaneous power rise, and the injected optical signal is proportional to the intensity of the target signal to be processed.

[0022] In one embodiment of the present invention, after step 4, the method further includes:

[0023] Step 5: Test the threshold characteristics, time domain accumulation characteristics, and refractory period characteristics of the photon pulse neuron implemented based on the DFB-SA laser to verify the characteristics of the monolithic integrated photon pulse neuron.

[0024] In one embodiment of the present invention, step 5 comprises:

[0025] The electrical stimulation signals with the same pulse interval but different intensities were modulated into optical signals using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its threshold characteristics.

[0026] The electrical stimulation signal with low intensity and different pulse intervals was modulated into an optical signal by using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its time domain accumulation characteristics.

[0027] The electrical stimulation signal with high intensity and different pulse intervals was modulated into an optical signal using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its refractory period characteristics.

[0028] Beneficial effects of the present invention:

[0029] 1. The method provided by the present invention can make the DFB-SA laser excited to have neuron characteristics by applying a suitable bias current, carrier wavelength and frequency of the injected DFB-SA optical signal to the DFB-SA laser, so as to respond to and process the input optical signal to realize the function of the photon pulse neuron. The photon pulse neuron realized by the method has a pure optical path from its input to its output, and does not require an additional photoelectric conversion structure. The implementation is simple, reduces system complexity and power consumption, and has a higher output power. At the same time, the method can realize cascade transmission, thereby forming a multi-layer or deep neural network;

[0030] 2. The method provided by the present invention is not only compatible with broadband weighted devices, but also applicable to narrow-bandwidth weighted devices, and is compatible with existing production processes, and has high application prospects and practical value.

[0031] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a schematic diagram of a method for realizing a monolithic integrated photon pulse neuron based on a DFB-SA laser provided in an embodiment of the present invention;

[0033] Figure 2 It is an experimental device and test result diagram of DFB-SA laser photon pulse neuron based on the embodiment of the present invention;

[0034] Figure 3 is a spectrum diagram of pulse outputs with different frequency periods provided by an embodiment of the present invention;

[0035] Figure 4 The pulse frequency provided by the embodiment of the present invention is about I G With V SA Function graph of ;

[0036] Figure 5 Schematic diagram of DFB-SA photon pulse neuron threshold characteristics provided by an embodiment of the present invention;

[0037] Figure 6 Schematic diagram of the cumulative characteristics of DFB-SA photon pulse neurons provided by an embodiment of the present invention;

[0038] Figure 7 It is a schematic diagram of the refractory period characteristics of DFB-SA photon pulse neurons provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] 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.

[0040] Embodiment 1

[0041] DFB (Distributed Feedback Laser) laser, that is, distributed feedback laser, has a built-in Bragg grating (Bragg Grating) and is a side-emitting semiconductor laser. In the DFB laser, the Bragg grating provides distributed feedback. This grating structure can replace electron beam lithography with a method combining traditional holographic exposure with micron-level lithography. Anti-reflection (AR) and high reflection (HR) coatings are applied to the two laser surfaces respectively to increase the laser emission power. On this basis, this embodiment performs a butterfly package on the chip without an isolator, and designs and manufactures a single longitudinal mode photon pulse neuron based on a built-in saturated absorber (SA) DFB laser.

[0042] For details, see Figure 1 , Figure 1 : is a schematic diagram of a method for realizing a monolithic integrated photon pulse neuron based on a DFB-SA laser provided by an embodiment of the present invention. The method for realizing a monolithic integrated photon pulse neuron based on a DFB-SA laser provided by the present invention comprises:

[0043] Step 1: Perform a power test on the DFB-SA laser to determine the operating voltage and bias current range of the DFB-SA laser.

[0044] Firstly, for a certain DFB-SA laser, a reverse bias voltage and a bias current are applied to its SA region and gain region respectively, and the operating voltage of the DFB-SA laser is obtained by traversal.

[0045] Specifically, in this embodiment, by selecting a DFB-SA laser and testing it, it can be found that the operating voltage of the DFB-SA laser is generally -6V to 0V.

[0046] Then, within the above-mentioned operating voltage range, the reverse bias voltage of the DFB-SA laser is set to a fixed value, and the bias current of the DFB-SA laser is set to change from small to large, and the corresponding output power is detected at the same time; and when the DFB-SA laser is excited to output power, the corresponding bias current is recorded as the excitation threshold current.

[0047] For details, see Figure 2 , Figure 2 1 is an experimental device and test result diagram of a DFB-SA laser photon pulse neuron provided by an embodiment of the present invention. Figure 2 Figure (a) shows the experimental setup for testing the DFB-SA. A tunable laser (TL) generates a continuous optical carrier (CW). The continuous optical carrier is injected into an intensity modulator (MZM). An arbitrary waveform generator (AWG, Tektronix AWG70001A) generates a defined external electrical stimulation signal. Two polarization controllers are used before and after the MZM to match the polarization state. The modulated optical signal is then injected into the isolator-free DFB-SA through a three-port optical circulator (CIRC). In the experiment, an optical spectrum analyzer (OSA, Advantest Q8384) is used to measure the spectrum. Two electro-optical detectors (pd) are used to achieve photoelectric conversion, and a real-time oscilloscope (OSC, Keysight DSOV334A, DSOZ592A) is used to measure the time series. The gain region of the DFB-SA is driven by a current source, while the SA region is reversely driven by a voltage source. The gain region bias current is denoted as the gain current I G , the reverse bias voltage of the SA region is recorded as V SA .

[0048] Figure 2 Figure (b) shows the power current (PI) curves measured under two representative reverse bias voltages. SA =0V, the threshold current of DFB-SA is I G =86mA; when V SA =-0.4V, the excitation threshold current of DFB-SA is I G = 94 mA. The results show that when a reverse bias voltage is applied, the threshold moves to a larger injection current and the output power decreases to some extent due to the absorption effect of the SA region.

[0049] Next, continue to increase the current and use an oscilloscope to observe the self-pulsation state of the DFB-SA laser to find the bias current that can make the laser emit self-pulsation, which is recorded as the self-pulsation threshold current.

[0050] Specifically, when the bias current changes, periodic spike pulse output can be observed. The reverse voltage of SA is fixed to V SA = -0.4V, change the gain current. Consider three gain current situations. The time domain output and the corresponding spectrum are shown in Figure 3 As shown. Figure 3 Available, when I G =115mA, the self-pulsation frequency is 0.703GHz. G =120mA and I G =130mA, the frequencies are 0.938GHz and 1.133GHz respectively. In addition, the pulse amplitude increases for larger gain currents.

[0051] To further reveal the self-pulsation dynamics, this embodiment also tests different V SA Under this condition, the peak pulse frequency and gain current I G The relationship between Figure 4 As shown. Figure 4 It can be obtained that for different V SA , the gain current range of the self-pulse output is different. SA = -0.4V, the gain current is 115mA ~ 132mA, and the self-pulse output can be achieved with a frequency of about 0.703GHz ~ 1.25GHz. SA = -0.8V, the gain current range of the self-pulse output is 123mA to 152mA, and the frequency range is 0.879GHz to 1.78GHz. SA= -1.4V, in the range of gain current from 141mA to 174mA, DFB-SA operates in a periodic self-pulsation state, and the frequency increases from 1.33GHz to 2.25GHz. Note that this self-pulsation characteristic simulates the frequency coding of biological neurons, and the coding rate is much higher than the biological coding rate.

[0052] Thus, the corresponding self-pulse threshold current range for different operating voltages can be obtained.

[0053] Finally, a bias current range is determined according to the excitation threshold current and the self-pulse threshold current.

[0054] Optionally, as an implementation manner, the bias current may be set between the excitation threshold current and the self-pulsation threshold current to force the DFB-SA laser to operate in an excitation state.

[0055] Step 2: Perform spectrum test on the DFB-SA laser to determine the carrier wavelength range injected into the DFB-SA laser.

[0056] The reverse bias voltage of the DFB-SA laser is set to a fixed value, and the bias current of the DFB-SA laser is changed, and the corresponding output spectrum is detected at the same time to detect the wavelength range of the DFB-SA laser itself. The spectrum of the free-running DFB-SA is as follows: Figure 2 As shown in (c), it can be seen that the wavelength of DFB-SA at this time is 1548.725nm.

[0057] Step 3: Determine the wavelength range of the optical signal to be injected according to the wavelength range of the DFB-SA laser itself.

[0058] It can be understood that, in this embodiment, the external wavelength to be injected must be consistent with the inherent wavelength of the laser so that the laser can work normally.

[0059] Step 4: Based on the determined bias current range, a bias current is applied to the gain region of the DFB-SA laser, and based on the determined operating voltage, a reverse bias voltage is applied to the SA region of the DFB-SA laser. Under the above conditions, an optical signal is injected into the DFB-SA laser according to the determined injection optical signal wavelength range, and the output spectrum of the DFB-SA laser is observed to determine the injected optical signal waveform corresponding to the target to be processed, thereby realizing the function of the photon pulse neuron.

[0060] Specifically, this embodiment determines the operating voltage, bias current and wavelength range of the optical signal to be injected of the DFB-SA laser through the operations of step 1, step 2 and step 3. According to the result, a bias current, a reverse bias voltage and an optical signal in a certain carrier wavelength range are applied to the DFB-SA laser, thereby realizing the DFB-SA photon pulse neuron.

[0061] It should be noted that the information carried by the injected light signal is represented by an instantaneous power rise, and the injected light signal is proportional to the intensity of the target signal to be processed, so that the formed photon pulse neurons can propagate in cascade, thereby realizing a multi-layer or deep neural network.

[0062] In another embodiment of the present invention, after step 4, the method further includes:

[0063] Step 5: Test the threshold characteristics, time domain accumulation characteristics, and refractory period characteristics of the photon pulse neuron implemented based on the DFB-SA laser to verify the characteristics of the monolithic integrated photon pulse neuron.

[0064] Specifically, step 5 includes:

[0065] The electrical stimulation signals with the same pulse interval but different intensities were modulated into optical signals using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its threshold characteristics.

[0066] The electrical stimulation signal with low intensity and different pulse intervals was modulated into an optical signal by using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its time domain accumulation characteristics.

[0067] The electrical stimulation signal with high intensity and different pulse intervals was modulated into an optical signal using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its refractory period characteristics.

[0068] The verification process of the above three neuron characteristics is introduced in detail below.

[0069] 1. Verify the DFB-SA photon pulse neuron threshold characteristics.

[0070] Specifically, in this experiment, the SA drive voltage is set to be fixed at V SA = -0.4V, the gain current is biased below the self-pulse threshold. The excitability threshold characteristics of light pulse neurons are as follows Figure 5 As shown, (a) is the external stimulation signal, and (b) is the corresponding DFB-SA pulse output under the corresponding stimulation.

[0071] from Figure 5 It can be seen that for 5 input stimulus pulses of different intensities, DFB-SA generates 4 spike pulses of almost the same intensity. Figure 5 Figure (c) shows the temporal folding plot of the response to 100 consecutive stimuli. Four distinguishable spike pulses can be observed in the two-dimensional plot, indicating that the excitatory threshold properties are reproducible across experiments.

[0072] 2. Verify the time domain accumulation characteristics of DFB-SA.

[0073] See also Figure 6 , Figure 6 is the cumulative characteristic of the DFB-SA photon pulse neuron provided by the embodiment of the present invention. Figure 6 The response of DFB-SA under the stimulation of pulse pairs with different pulse intervals (ISIs) shown in (a) is shown in Figure (b). In Figure (a), the first single weak pulse and the last single strong pulse are reference pulses. The state of modulated light injection into DFB-SA makes the single weak pulse below the excitation threshold, while the single strong pulse exceeds the excitation threshold. It can be seen from Figure (b) that a single weak stimulation pulse cannot trigger DFB-SA to produce a spike pulse response, while the last strong stimulation pulse can trigger DFB-SA to produce a spike pulse response.

[0074] In addition, the first four pairs of pulses with relatively small ISIs and closely spaced pulses triggered DFB-SA to produce four spike pulse responses. For the last three pairs of pulses with large ISIs, the two stimulation pulses could not be accumulated in the time domain and could not exceed the excitability threshold. Therefore, DFB-SA simulated the time domain accumulation characteristics of neurons.

[0075] 3. Verify the refractory period characteristics of DFB-SA photon pulse neurons.

[0076] To simulate the refractory period characteristics, this embodiment is designed as follows Figure 7 (a) shows the external stimulus pulse pair. Similarly, a single pulse is introduced as a reference pulse. The selection criteria of the stimulus signal and stimulus intensity are to ensure that the first single pulse can trigger the DFB-SA to generate a spike pulse response. Figure 7 As shown in (b)-(f), different gain currents lead to different refractory periods. Figure 7 In (b), when I G =98.7mA, each pulse pair only generates one spike pulse. Figure 7 (c) When I G=104.4mA, the last pulse pair with ISI=1.24ns can generate two spike pulses. As the gain current increases, more pulse pairs can generate two spike pulses. That is, the refractory period decreases with the increase of gain current. Note that the information processing speed is limited by the refractory period. In other words, the gain current can be optimized in practical applications.

[0077] In this embodiment, the DFB-SA chip can simulate the frequency encoding mechanism and the neuron-like response of LIF neurons. Therefore, the DFB-SA chip can be used as a basic component of photonic SNN.

[0078] The method provided by the present invention can make the DFB-SA laser excited to have neuron characteristics by applying a suitable bias current, carrier wavelength and frequency of the injected DFB-SA optical signal to the DFB-SA laser, so as to respond to and process the input optical signal to realize the function of the photon pulse neuron. The photon pulse neuron realized by the method has a pure optical path from its input to its output, does not require an additional photoelectric conversion structure, is simple to implement, reduces system complexity and power consumption, and has a higher output power; at the same time, the method can realize cascade transmission, thereby forming a multi-layer or deep neural network. In addition, the method is not only compatible with broadband weighted devices, but also suitable for narrow bandwidth weighted devices, and is compatible with existing production processes, and has a high prospect of use and practical value.

[0079] 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 protection scope of the present invention.

Claims

1. A method for realizing monolithic integrated photon pulse neurons based on DFB-SA lasers, It is characterized in that include: Step 1: Perform a power test on the DFB-SA laser to determine the operating voltage and bias current range of the DFB-SA laser; including: For a certain DFB-SA laser, a reverse bias voltage and a bias current are applied to its SA region and gain region respectively, and the operating voltage of the DFB-SA laser is obtained by traversal; The reverse bias voltage of the DFB-SA laser is set to a fixed value based on the working voltage, and the bias current of the DFB-SA laser is set to change from small to large, and the corresponding output power is detected at the same time; when the DFB-SA laser is excited to output power, the corresponding bias current is recorded as the excitation threshold current; Continue to increase the current and use an oscilloscope to observe the self-pulsation state of the DFB-SA laser to find the bias current that can make the laser emit self-pulsation, which is recorded as the self-pulsation threshold current; Setting the bias current between the excitation threshold current and the self-pulsation threshold current to cause the DFB-SA laser to operate in an excitation state; Step 2: Perform spectrum test on the DFB-SA laser to determine the wavelength range of the DFB-SA laser. Step 3: Determine the wavelength range of the optical signal to be injected according to the wavelength range of the DFB-SA laser itself; Step 4: applying a bias current to the gain region of the DFB-SA laser based on the determined bias current range, and applying a reverse bias voltage to the SA region of the DFB-SA laser based on the determined operating voltage, and under the above conditions, injecting an optical signal into the DFB-SA laser according to the determined injection optical signal wavelength range, observing the output spectrum of the DFB-SA laser to determine the injected optical signal waveform corresponding to the target to be processed, thereby realizing the function of the photon pulse neuron; Step 5: Test the threshold characteristics, time domain accumulation characteristics, and refractory period characteristics of the photon pulse neuron implemented based on the DFB-SA laser to verify the characteristics of the monolithic integrated photon pulse neuron.

2. The method for realizing a monolithic integrated photon pulse neuron based on a DFB-SA laser according to claim 1, It is characterized in that The working voltage is -6V to 0V.

3. The method for realizing a monolithic integrated photon pulse neuron based on a DFB-SA laser according to claim 1, It is characterized in that Step 2 includes: The reverse bias voltage of the DFB-SA laser is set to a fixed value, and the bias current of the DFB-SA laser is changed, and the corresponding output spectrum is detected at the same time to detect the wavelength range of the DFB-SA laser itself.

4. The method for realizing a monolithic integrated photon pulse neuron based on a DFB-SA laser according to claim 1, It is characterized in that In step 4, the information carried by the injected optical signal is represented by an instantaneous power rise, and the injected optical signal is proportional to the intensity of the target signal to be processed.

5. The method for realizing a monolithic integrated photon pulse neuron based on a DFB-SA laser according to claim 1, step 5 include: The electrical stimulation signals with the same pulse interval but different intensities were modulated into optical signals using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its threshold characteristics. The electrical stimulation signal with low intensity and different pulse intervals was modulated into an optical signal by using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its time domain accumulation characteristics. The electrical stimulation signal with high intensity and different pulse intervals was modulated into an optical signal using an optical waveform generator and input into the DFB-SA laser. The output waveform of the DFB-SA laser was observed to verify its refractory period characteristics.

Images