Method for implementing a photonic pulse neuron based on a two-section Fabry-Perot laser

Abstract

The present invention discloses a method for implementing a photonic pulse neuron based on a two-section Fabry-Perot laser, comprising: applying a bias electrical signal to the FP-SA and inputting an incoherent photonic pulse signal to the FP-SA, so as to utilize the neuron characteristics excited by the FP-SA to perform response processing on the incoherent photonic pulse signal and realize the function of the photonic pulse neuron; wherein, the bias electrical signal includes: a gain region current applied to the gain region of the FP-SA and a reverse bias voltage applied to the saturable absorption region of the FP-SA; the gain region current does not exceed the Q-switching pulse threshold corresponding to the reverse bias voltage and is sufficient to prompt the FP-SA to operate in an excited state; the Q-switching pulse threshold is defined as: under the conditions of no external light input and a given reverse bias voltage, the minimum gain region current at which the FP-SA is excited to the Q-switching pulse state. The photonic pulse neuron implemented by the present invention exhibits better performance in all aspects.

Description

Technical Field

[0001] The present invention belongs to the technical field of photon pulse neural networks, and particularly relates to a method for realizing photon pulse neurons based on a two-section Fabry-Perot laser. Background Art

[0002] Currently, it is increasingly difficult for traditional electronic processors based on the von Neumann architecture to maintain Moore's Law. Inspired by the network structure and principle of the human brain, neuromorphic computing has become one of the methods to overcome the von Neumann bottleneck in the post-Moore era.

[0003] Compared with traditional continuous-value artificial neural networks (ANNs) and convolutional neural networks (CNNs), spiking neural networks (SNNs) provide a more biologically plausible way to implement neuromorphic computing. It takes into account the influence of time information and provides a more biologically meaningful method for implementing neuromorphic computing.

[0004] Photon technology has advantages such as high speed, wide bandwidth, parallel processing, and low power consumption, and is one of the candidates for the next-generation neuromorphic processors. In neuromorphic computing, linear and nonlinear computations are indispensable basic components and are equally important. Currently, linear computations have been successfully implemented optically using Mach–Zehnder interferometers (MZIs), micro-ring resonators (MRRs) weight libraries, waveguide-integrated phase change materials (PCMs), and semiconductor optical amplifiers (SOAs). However, in terms of nonlinear computations, the vast majority of nonlinear computations in photon neural network chips are implemented electronically rather than optically. Therefore, the optical implementation of nonlinearity remains one of the most challenging problems in optical neural networks.

[0005] In the closest prior art, there are methods for realizing photon pulse neurons based on PCMs, methods for realizing photon pulse neurons based on microcavity lasers, and methods for realizing photon pulse neurons based on integrated distributed feedback semiconductor lasers (DFBs).

[0006] Among them, the photon pulse neuron implemented based on PCM lacks the time integration ability that is crucial for optical pulse processing; the output power of the photon pulse neuron implemented based on a microcolumn laser is in the microwatt level, and the output power is relatively low, which is not conducive to the cascading of neurons; the photon pulse neuron implemented based on an integrated DFB requires optoelectronic conversion on the optical path between DFBs inside a single neuron, and the implementation complexity is relatively high.

[0007] It can be seen from this that there is no method for implementing a photon pulse neuron with excellent performance in all aspects in the prior art. Summary of the Invention

[0008] In order to solve the above problems existing in the prior art, the present invention provides a method for implementing a photon pulse neuron based on a two-section Fabry-Perot laser.

[0009] The technical problem to be solved by the present invention is realized through the following technical solutions:

[0010] A method for implementing a photon pulse neuron based on a two-section Fabry-Perot laser, wherein the two-section Fabry-Perot laser is a two-section FP-SA laser formed by introducing a saturable absorber SA into a Fabry-Perot FP cavity;

[0011] The method includes:

[0012] Applying a bias electrical signal to the FP-SA, and inputting an incoherent photon pulse signal to the FP-SA under the condition of maintaining the bias electrical signal, so as to perform response processing on the incoherent photon pulse signal by using the neuron characteristics excited by the FP-SA, and realize the function of the photon pulse neuron;

[0013] Among them, the bias electrical signal includes: a gain region current applied to the gain region of the FP-SA, and a reverse bias voltage applied to the saturable absorption region of the FP-SA; the gain region current does not exceed the Q-switching pulse threshold corresponding to the reverse bias voltage and is sufficient to cause the FP-SA to enter the excited state for operation;

[0014] The Q-switching pulse threshold is defined as: under the condition of no external light input and a given reverse bias voltage, the minimum gain region current at which the FP-SA is excited to the Q-switching pulse state.

[0015] Preferably, the neuron characteristics include: threshold characteristics, time-domain cumulative characteristics, and cascading characteristics.

[0016] Preferably, the value range of the reverse bias voltage is -6V to 0V.

[0017] Preferably, the wavelength difference between the wavelength of the incoherent photon pulse signal and the central operating wavelength of the FP-SA is 0nm ± 15nm.

[0018] Preferably, in the bias electrical signal, the current in the gain region is equal to the Q-switching pulse threshold corresponding to the reverse bias voltage × A%, where A = 80 - 95.

[0019] Preferably, the method of applying the current in the gain region to the FP-SA includes:

[0020] Applying the current in the gain region to the FP-SA using a laser diode controller, and simultaneously controlling the operating temperature of the FP-SA using the laser diode controller.

[0021] Preferably, the length of the saturable absorption region is equal to the length of the gain region × B%, where B = 1.6 - 7.

[0022] Preferably, the FP-SA is a PIN structure grown based on AlGaInAs / InP materials.

[0023] In the method for realizing a photonic pulse neuron based on a two-section Fabry-Perot laser provided by the present invention, by applying an appropriate bias electrical signal to the FP-SA laser, the FP-SA laser can be excited to exhibit neuron characteristics, so as to perform response processing on an incoherent photonic pulse signal and realize the function of a photonic pulse neuron. Among them, the neuron characteristics include time-domain cumulative characteristics. Therefore, the photonic pulse neuron realized by using the present invention has time integration ability; moreover, since the output power of the FP-SA laser is in the milliwatt level, even cascading can meet the power requirement; and, the photonic pulse neuron realized by using the present invention has a pure optical path from its input to its output, without inserting an optoelectronic conversion structure, and the implementation method only needs to apply an appropriate bias to the FP-SA laser, and the implementation method is simple. Thus, it can be seen that the photonic pulse neuron realized by using the method provided by the present invention has better performance in all aspects. And, since the FP-SA laser is already a device in production, when using the method provided by the present invention to realize neurons and then based on these neurons combined with an optical synapse device to realize a photonic pulse neural network, it can be compatible with the existing production process and has high application prospects and practical value.

[0024] The following will further elaborate on the present invention in conjunction with the drawings. Description of the Drawings

[0025] Figure 1 The chip micrograph of the FP-SA used in the embodiment of the present invention is shown;

[0026] Figure 2 It is a schematic diagram of the experimental platform used to verify the neuron characteristics of the FP-SA during the process of implementing the embodiment of the present invention;

[0027] Figure 3It is a timing diagram that verifies the cascade characteristic and time-domain accumulation characteristic of FP-SA during the implementation of the embodiments of the present invention;

[0028] Figure 4 It is a timing diagram that verifies the cascade characteristic and threshold characteristic of FP-SA during the implementation of the embodiments of the present invention;

[0029] Figure 5 It is the power-current-voltage curve of FP-SA tested during the implementation of the embodiments of the present invention;

[0030] Figure 6 It is the spectrogram when FP-SA is excited to a multimode state during the implementation of the embodiments of the present invention;

[0031] Figure 7 It is the timing diagram when FP-SA is excited to a self-pulsing state during the implementation of the embodiments of the present invention;

[0032] Figure 8 It is a schematic diagram of the method for implementing a photonic pulse neuron based on a two-section Fabry-Perot laser provided by the embodiments of the present invention. Detailed implementation manners

[0033] The following further describes the present invention in detail with reference to specific embodiments, but the implementation manners of the present invention are not limited thereto.

[0034] In order to implement a photonic pulse neuron with better performance in all aspects, the embodiments of the present invention provide a method for implementing a photonic pulse neuron based on a two-section Fabry-Perot laser. The two-section Fabry-Perot laser mentioned here is a two-section FP-SA laser formed by introducing a saturable absorber (SA) into a Fabry-Perot (FP) cavity, abbreviated as FP-SA.

[0035] Among them, the FP-SA laser is a PIN structure grown based on AlGaInAs (aluminum gallium indium arsenide) / InP (indium phosphide) materials, and its chip microstructure is as Figure 1 shown, including a gain region and a saturable absorption region. There is an electrical isolation region between the two regions, and electrodes are respectively provided on the two regions. The gain region electrode is externally connected to a bias current (gain region current), and the saturable absorption region electrode is externally connected to a reverse-biased voltage (reverse bias voltage). External light enters the laser, and is excited to generate laser light and output outward inside the laser. Figure 1 The square icon is used to indicate the tail of the laser, and the light reflectivity of the end face of the laser tail is about 95%. The triangular icon is used to indicate the light output direction of the laser, and the light reflectivity of the end face of the laser light output port is about 30%.

[0036] Preferably, the length of the saturable absorption region is equal to the length of the gain region × B%, where B = 1.6 - 7.

[0037] Exemplarily, under the condition that the total length of the FP-SA is 1500 μm, the lengths of the SA can be 25 μm, 30 μm, 60 μm, 75 μm, 90 μm, 105 μm, 120 μm respectively, but are not limited thereto.

[0038] In the process of implementing the present invention, a series of tests were conducted on the above-mentioned FP-SA laser, verifying the complex non-linear neuron-like dynamics of the FP-SA laser as a photonic pulse neuron, and determining the specific implementation method for realizing the photonic pulse neuron based on the FP-SA laser.

[0039] First, the verification process of the complex non-linear neuron-like dynamics of the FP-SA laser as a photonic pulse neuron will be described in detail.

[0040] Specifically, an FP-SA with a total length of 1500 μm and a saturated absorption region length of 60 μm was selected as an example for the experiment, and a photonic pulse neural network experimental platform as shown in Figure 2 was built. The details of each module are as follows:

[0041]

[0042]

[0043] Among them, the AWG is used to generate electrical signals; the TL is used to provide an optical carrier (wavelength 1561.48 nm); after the optical carrier passes through PC1, it enters the MZM; here, PC1 can control the polarization state of the externally injected light to match the working state of the MZM. The optical carrier carrying the signal is injected into the gain region of FP-SA1 through the EDFA, VOA, OC1, PC2, OC2, CIRC. The gain region current of each FP-SA is provided by the LDC, the reverse bias voltage of the saturated absorption region is adjusted by the VS, and the working temperature of each FP-SA is controlled by the LDC. The optical power of the externally injected light into FP-SA1 can be adjusted by the EDFA and VOA, and can be detected by the PM. PC2 can adjust the polarization state of the externally injected light entering FP-SA1 to match FP-SA1 to achieve the characteristics of the photonic pulse neuron. The output of FP-SA1 is divided into 3 paths through OC3. One path can be used for spectral analysis through the OSA, one path is converted into an electrical signal by the PD and then enters the OSC for timing analysis and recording, and the other path is injected into FP-SA2 through PC3, VOA, OC4 and the optical circulator. Similarly, the VOA mentioned here can adjust the magnitude of the optical power injected into FP-SA2, and PC3 can adjust the polarization state of the injected light to match the working state of FP-SA2. Thus, two FP-SAs are cascaded to test the cascading of the photonic pulse neuron characteristics.

[0044] First, configure the AWG to generate an external stimulus signal (the optical signal entering FP-SA1) as shown in Figure 3 subfigure (a), which includes three consecutive small pulses (with a pulse interval of 500 ps) and a small pulse with the same power. Figure 3 Subfigure (b) is the pulse response generated by FP-SA1. It can be seen that for the three consecutive small pulses, FP-SA1 generates pulses, while for a single small pulse with the same power, FP-SA1 does not generate pulses. This shows that a single sub-threshold pulse cannot cause FP-SA1 to generate a pulse, while three closely spaced sub-threshold pulses are integrated in the time domain, thus exceeding the threshold for FP-SA1 to generate a pulse and triggering FP-SA1 to generate a pulse, which illustrates that FP-SA has the time-domain cumulative characteristic. Figure 3 Subfigure (c) is the pulse response generated by FP-SA2. It can be seen that FP-SA2 has almost the same pulse output as FP-SA1. It is worth emphasizing that the response of FP-SA2 to the above three consecutive small pulses does not significantly weaken due to its later cascaded position, and FP-SA2 has almost no response to the above single small pulse, showing a larger extinction ratio, which indicates that the performance of the photon pulse neural network formed by cascading FP-SA will be better. The inventor analyzed and believed that this is because cascading FP-SA can produce a combined saturable absorption effect.

[0045] Then, configure the AWG to generate a perturbation signal as shown in Figure 4 subfigure (a) of, which includes three pulses with different powers. After this perturbation signal is injected into FP-SA1, the response of FP-SA1 is as shown in Figure 4 figure (b) of. It can be seen that FP-SA1 hardly generates a response to the 1st and 2nd perturbation pulses, and only generates an excitatory response to the 3rd perturbation pulse with a higher power, which indicates that FP-SA has a threshold characteristic. Correspondingly, the pulse response of FP-SA2 cascaded after FP-SA1 corresponds to the output of FP-SA1, which indicates that the output response of the previous FP-SA1 can be propagated to the subsequent FP-SA2 through cascading, that is, it verifies that FP-SA has a cascading characteristic.

[0046] In summary, FP-SA can simulate the cumulative, threshold, and cascading characteristics of biological neurons and has the performance and functions of a photon pulse neuron.

[0047] Then, the determination process of the implementation method for realizing a photon pulse neuron based on an FP-SA laser is described in detail.

[0048] First, use the LDC to apply a gain region current to the FP-SA, use the VS to apply a reverse bias voltage to the saturable absorption region of the FP-SA, and regulate the operating temperature of the FP-SA. Adjust the gain region current, reverse bias voltage, and operating temperature to obtain the power-current-voltage curves at different temperatures and different reverse bias voltages as shown in Figure 5 shown. Here, the purpose of obtaining this curve is to determine the threshold (current threshold) for the excitation of the FP-SA under various temperatures and reverse biases; as can be seen from Figure 5 , the greater the reverse bias voltage, the greater the threshold for the excitation of the FP-SA; in addition, an increase in the operating temperature will also increase the threshold of the FP-SA laser.

[0049] Among them, during the process of testing the power-current-voltage curve with a reverse bias voltage of 0V, when the gain region current is greater than the threshold current of 45mA at 0V, the spectrum observed from the output of the FP-SA is as shown in Figure 6 shown. At this time, the FP-SA is excited to a multimode state. According to f = c / (2n g ×L cavity ), it can be known that the mode-locked pulse frequency of the FP-SA in the multimode state is 28.9GHz, and the spacing between adjacent two modes is 0.24nm; where c is the speed of light in vacuum, and n g = 3.46 is the group refractive index of the ridge waveguide of the FP-SA.

[0050] During the process of testing the power-current-voltage curve at a larger reverse bias voltage, it is found that when the gain region current is increased to a certain condition, the FP-SA will appear in the self-pulsing state as shown in Figure 7 shown. For example, when the reverse bias voltage is -4.49V, when the gain region current increases from 100mA to 120mA, the FP-SA can appear in the self-pulsing state, and the corresponding pulse frequency increases from 1.49GHz to 1.95GHz.

[0051] Thus, the critical point of the gain region current corresponding to the self-pulsing state of the FP-SA under various reverse bias voltages can be obtained, which is used as the Q-switching pulse threshold corresponding to each reverse bias voltage. By setting the gain region current of the FP-SA slightly lower than the Q-switching pulse threshold, the FP-SA can perform a non-linear response to the incoherent light input into it, realizing the non-linear operation function of the photonic pulse neuron.

[0052] Specifically, as shown in Figure 8 , the method for realizing a photonic pulse neuron based on a two-section Fabry-Perot laser provided by the embodiment of the present invention includes:

[0053] A bias electrical signal is applied to the FP-SA laser, and an incoherent photon pulse signal is input to the FP-SA laser while maintaining the bias electrical signal, so as to utilize the characteristics of the neurons excited by the FP-SA laser to respond to the incoherent photon pulse signal (output a response pulse) and realize the function of a photon pulse neuron.

[0054] The bias electrical signal includes: a gain region current I applied to the gain region of the FP-SA a , and the reverse bias voltage V applied to the saturation absorption region of the FP-SA laser s ; Gain region current I a Less than the reverse bias voltage V s The corresponding Q-switched pulse threshold I Q The Q-switched pulse threshold is defined as the minimum gain region current at which the FP-SA laser is excited to operate in the Q-switched pulse state without external light input and with a given reverse bias voltage.

[0055] Preferably, in the above bias electrical signal, the gain region current is equal to the Q-switching pulse threshold corresponding to the reverse bias voltage × A%, A = 80 to 95. In this way, the gain region current is both below the Q-switching pulse threshold and not too far from the Q-switching pulse threshold, so that even a relatively low-power incoherent photon pulse signal can excite the FP-SA to generate a pulse response.

[0056] Preferably, the wavelength difference between the wavelength of the incoherent photon pulse signal and the central operating wavelength of the FP-SA is 0 nm±15 nm.

[0057] Preferably, the reverse bias voltage V s The value range is -6V to 0V.

[0058] In an optional implementation, the method of applying the gain region current to the FP-SA may include:

[0059] A laser diode controller is used to apply a gain region current to the FP-SA, and the laser diode controller is used to control the operating temperature of the FP-SA, so as to ensure that the Q-switched pulse threshold of the FP-SA is stable and does not drift.

[0060] Of course, if the working environment temperature of FP-SA is relatively strictly controllable, then a common current source can also be used to apply gain region current to FP-SA. Alternatively, if the power of the incoherent photon pulse signal is large enough, the influence of Q-switched pulse threshold drift caused by working temperature change on the normal operation of photon pulse neurons can be ignored.

[0061] In the method for implementing a photonic pulse neuron based on a two-section Fabry-Perot laser provided by an embodiment of the present invention, by applying an appropriate bias electrical signal to the FP-SA laser, the FP-SA laser can be excited to exhibit neuron characteristics, so as to respond to an incoherent photonic pulse signal and realize the function of a photonic pulse neuron. Among them, the neuron characteristics include time-domain cumulative characteristics. Therefore, the photonic pulse neuron implemented by using the present invention has the ability of time integration; moreover, since the output power of the FP-SA laser is in the milliwatt level, even cascading can meet the requirements of power magnitude; furthermore, the photonic pulse neuron implemented by using the present invention has a pure optical path from its input to its output, without inserting an optoelectronic conversion structure, and the implementation method only needs to apply an appropriate bias to the FP-SA laser, and the implementation method is simple. Thus, it can be seen that the photonic pulse neuron implemented by using the method provided by the embodiment of the present invention has better performance in all aspects. Moreover, since the FP-SA laser is already a mass-produced device, when using the method provided by the embodiment of the present invention to implement neurons and then implementing a photonic pulse neural network based on the neurons combined with an optical synapse device, it can be compatible with the existing production process and has high application prospects and practical value.

[0062] Therefore, by integrating the photonic pulse neuron implemented by using the embodiment of the present invention with a silicon photonics-based weight device (such as a weight device implemented by using an MRR or MZI network) or an InP-based weight device (such as a weight device implemented by using an SOA), a photonic pulse neural network can be realized.

[0063] It should be noted that for the embodiment of the photonic pulse neuron chip, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment.

[0064] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification.

[0065] Although the present application has been described in conjunction with various embodiments herein, however, in the process of implementing the claimed present application, those skilled in the art can understand and realize other variations of the disclosed embodiments by viewing the accompanying drawings, the disclosure content, and the appended claims.

[0066] The above content is a further detailed description of the present invention in combination with specific preferred embodiments. It cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can still be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for implementing a photonic pulse neuron based on a two - section Fabry - Perot laser, characterized in that, the two - section Fabry - Perot laser is a two - section FP - SA laser formed by introducing a saturable absorber (SA) into a Fabry - Perot (FP) cavity; the method includes: applying a bias electrical signal to the FP - SA, and inputting an incoherent photonic pulse signal to the FP - SA under the condition of maintaining the bias electrical signal, so as to perform response processing on the incoherent photonic pulse signal by using the neuron characteristics excited in the FP - SA, and realizing the function of the photonic pulse neuron; wherein, the bias electrical signal includes: a gain - region current applied to the gain region of the FP - SA, and a reverse bias voltage applied to the saturable absorption region of the FP - SA; the gain - region current does not exceed the Q - switching pulse threshold corresponding to the reverse bias voltage and is sufficient to cause the FP - SA to operate in an excited state; the Q - switching pulse threshold is defined as: under the condition of no external light input and a given reverse bias voltage, the minimum gain - region current at which the FP - SA is excited to the Q - switching pulse state.

2. The method according to claim 1, characterized in that, the neuron characteristics include: threshold characteristics, time - domain cumulative characteristics, and cascade characteristics.

3. The method according to claim 1, characterized in that, the value range of the reverse bias voltage is - 6V to 0V.

4. The method according to claim 1, characterized in that, the wavelength difference between the wavelength of the incoherent photonic pulse signal and the central operating wavelength of the FP - SA is 0nm ± 15nm.

5. The method according to claim 1, characterized in that, in the bias electrical signal, the gain - region current is equal to the Q - switching pulse threshold corresponding to the reverse bias voltage × A%, where A = 80 - 95.

6. The method according to claim 1, characterized in that, the method of applying the gain - region current to the FP - SA includes: using a laser diode controller to apply the gain - region current to the FP - SA, and at the same time using the laser diode controller to control the operating temperature of the FP - SA.

7. The method according to claim 1, characterized in that, the length of the saturable absorption region is equal to the length of the gain region × B%, where B = 1.6 - 7.

8. The method according to claim 1, characterized in that, the FP - SA is a PIN structure grown based on AlGaInAs / InP materials.

Images