CN115456158A - Method for realizing photon pulse neuron based on two-section Fabry-Perot laser - Google Patents

Abstract

The invention discloses a method for realizing photon pulse neurons based on a two-section Fabry-Perot laser, which comprises the following steps: applying a bias electric signal to the FP-SA, and inputting a non-coherent photon pulse signal to the FP-SA so as to respond and process the non-coherent photon pulse signal by utilizing the neuron characteristic excited by the FP-SA to realize the function of a photon pulse neuron; wherein the bias electrical signal comprises: a gain region current applied to a gain region of the FP-SA, and a reverse bias voltage applied to a saturable absorption region of the FP-SA; the current of the gain area does not exceed the Q-switched pulse threshold corresponding to the reverse bias voltage and is enough to promote the FP-SA to work in an excitation state; the Q-switched pulse threshold is defined as: the FP-SA is excited to emit the minimum gain region current of the Q-switched pulse state under the condition of no external light input and given reverse bias voltage. The photon pulse neuron realized by the invention has better performance in all aspects.

Description

Method for realizing photon pulse neuron based on two-section Fabry-Perot laser

Technical Field

The invention belongs to the technical field of photon pulse neural networks, and particularly relates to a method for realizing a photon pulse neuron based on a two-section Fabry-Perot laser.

Background

Currently, it is increasingly difficult for traditional electronic processors based on von neumann architectures to maintain moore's law. Inspired by the structure and principles of human brain networks, neuromorphic computing has become one of the methods to overcome the bottleneck of von-niemann in post-molar times.

Compared with the traditional Artificial Neural Network (ANN) and Convolutional Neural Network (CNN) based on continuous values, the impulse Neural Network (SNN) provides a more biologically reasonable way to realize the neuromorphic calculation, takes the influence of time information into consideration, and provides a method for realizing the neuromorphic calculation with more biological significance.

Photon technology has the advantages of high speed, wide bandwidth, parallel processing, low power consumption and the like, and is one of candidates of next-generation neuromorphic processors. In the neuromorphic calculation, linear and nonlinear calculations are indispensable basic components and have equal importance. At present, linear calculation is successfully achieved optically by using a Mach-Zehnder interferometer (MZI), a Micro Ring Resonator (MRR) weight library, a waveguide integrated Phase Change Material (PCM), and a Semiconductor Optical Amplifier (SOA); in terms of nonlinear computation, nonlinear computation in most photonic neural network chips is realized electronically, not optically, so that nonlinear optical realization remains one of the most challenging problems of the optical neural network.

In the closest prior art, there are a method for implementing a photon pulse neuron based on PCM, a method for implementing a photon pulse neuron based on a micro-column Laser, and a method for implementing a photon pulse neuron based on an integrated Distributed Feedback semiconductor Laser (DFB).

Among them, PCM-based implementation of photon-pulsed neurons lacks the time-integration capability critical for light-pulse processing; the photon pulse neuron output power realized based on the micro-column laser is at a microwatt level, and the output power is low, so that the cascade of neurons is not facilitated; photon pulse neurons realized based on integrated DFBs need to perform photoelectric conversion on light paths among the DFBs inside a single neuron, and the realization complexity is high.

Therefore, the prior art has not provided a photon pulse neuron implementation method with better performance in all aspects.

Disclosure of Invention

In order to solve the problems in the prior art, the invention provides a method for realizing a photon pulse neuron based on a two-segment Fabry-Perot laser.

The technical problem to be solved by the invention is realized by the following technical scheme:

a method for realizing photon pulse neurons based on a two-segment Fabry-Perot laser is disclosed, wherein the two-segment Fabry-Perot laser is a two-segment FP-SA laser formed by introducing a saturable absorber SA into a Fabry-Perot FP cavity;

the method comprises the following steps:

applying a bias electric signal to the FP-SA, and inputting a non-coherent photon pulse signal to the FP-SA under the condition of maintaining the bias electric signal so as to utilize the neuron characteristics excited by the FP-SA to perform response processing on the non-coherent photon pulse signal and realize the function of a photon pulse neuron;

wherein the bias electrical signal comprises: a gain region current applied to a gain region of the FP-SA, and a reverse bias voltage applied to a saturable absorption region of the FP-SA; the current of the gain region does not exceed a Q-switched pulse threshold corresponding to the reverse bias voltage and is enough to cause the FP-SA to enter an excitation state to work;

the Q-switched pulse threshold is defined as: under the condition of no external light input and given reverse bias voltage, the FP-SA is excited out of the minimum gain region current of the Q-switched pulse state.

Preferably, the neuron characteristics include: a threshold characteristic, a time domain accumulation characteristic, and a cascadable characteristic.

Preferably, the value range of the reverse bias voltage is-6V-0V.

Preferably, the wavelength difference between the wavelength of the incoherent photon pulse signal and the central operating wavelength of the FP-SA is 0nm +/-15 nm.

Preferably, in the bias electrical signal, the gain section current is equal to a Q-switched pulse threshold × a% corresponding to the reverse bias voltage, and a =80 to 95.

Preferably, the manner of applying the gain region current to the FP-SA includes:

applying a gain region current to the FP-SA with a laser diode controller while controlling an operating temperature of the FP-SA with the laser diode controller.

Preferably, the length of the saturable absorption region is equal to the length of the gain region × B%, B =1.6 to 7.

Preferably, the FP-SA is a PIN structure grown on the basis of AlGaInAs/InP material.

In the method for realizing the photon pulse neuron based on the two-section Fabry-Perot laser, the FP-SA laser can be excited to generate neuron characteristics by applying a proper bias electric signal to the FP-SA laser, so that the incoherent photon pulse signal is subjected to response processing, and the function of the photon pulse neuron is realized. The neuron characteristics comprise time domain accumulation characteristics, so that the photon pulse neuron realized by the invention has time integration capability; moreover, as the output power of the FP-SA laser is in the megawatt level, the requirement of the power can be met even if cascade connection is carried out; in addition, the photon pulse neuron realized by the invention is a pure optical path from the input to the output of the photon pulse neuron, a photoelectric conversion structure is not inserted, the realization mode only needs to apply proper bias to the FP-SA laser, and the realization mode is simple. Therefore, the photon pulse neuron realized by the method provided by the invention has better performance in all aspects. Moreover, as the FP-SA laser is already produced, the method provided by the invention can be compatible with the existing production process when the neuron is realized by utilizing the method and then the photon pulse neural network is realized by combining the neuron with the optical synapse device, thereby having higher use prospect and practical value.

The present invention will be described in further detail with reference to the accompanying drawings.

Drawings

A micrograph of a chip of FP-SA used in an embodiment of the invention is shown in FIG. 1;

FIG. 2 is a schematic illustration of an experimental platform used in validating neuronal characteristics of FP-SA in the course of implementing an embodiment of the invention;

FIG. 3 is a timing diagram illustrating the verification that the FP-SA has the cascading characteristic and the time-domain accumulation characteristic in the implementation of an embodiment of the present invention;

FIG. 4 is a timing diagram illustrating the verification that the FP-SA has the cascade characteristic and the threshold characteristic in the implementation of an embodiment of the present invention;

FIG. 5 is a power current voltage curve of a FP-SA tested in the course of implementing an embodiment of the present invention;

FIG. 6 is a spectral diagram of a FP-SA excited multi-mode state in a process implementing an embodiment of the present invention;

FIG. 7 is a timing diagram of the FP-SA being excited out of a pulse state in implementing an embodiment of the present invention;

fig. 8 is a schematic diagram of a method for implementing a photon pulse neuron based on a two-segment fabry-perot laser according to an embodiment of the present invention.

Detailed Description

The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.

In order to realize photon pulse neurons with better performance in all aspects, the embodiment of the invention provides a method for realizing photon pulse neurons based on a two-section Fabry-Perot laser. The two-section Fabry-Perot laser is a two-section Fabry-Perot (FP-SA) laser formed by introducing a Saturable Absorber (SA) into a Fabry-Perot (FP) cavity, and is called FP-SA for short.

The FP-SA laser is a PIN structure based on AlGaInAs (aluminum gallium indium arsenide)/InP (indium phosphide) material growth, a chip microstructure of the FP-SA laser is shown as figure 1 and comprises a gain region and a saturation absorption region, an electric isolation region is arranged between the two regions, electrodes are respectively arranged on the two regions, the gain region electrode is externally connected with a bias current (gain region current), and the saturation absorption region electrode is externally connected with a reverse bias voltage (reverse bias voltage). The external light enters the laser, and is excited in the laser to generate laser output outwards. The block diagram in fig. 1 is used for indicating the tail of the laser, the light reflectivity of the end face of the tail of the laser is about 95%, the triangular diagram is used for indicating the light emitting direction of the laser, and the light reflectivity of the end face of the light emitting port of the laser is about 30%.

Preferably, the length of the saturable absorption region is equal to the length of the gain region × B%, B =1.6 to 7.

Illustratively, the total length of FP-SA is 1500 μm, and the lengths of SA may be 25 μm, 30 μm, 60 μm, 75 μm, 90 μm, 105 μm, and 120 μm, respectively, but is not limited thereto.

In the process of realizing the invention, a series of tests are carried out on the FP-SA laser, the complex nonlinear neuron dynamics of the FP-SA laser as a photon pulse neuron is verified, and a specific realization method for realizing the photon pulse neuron based on the FP-SA laser is determined.

Firstly, the verification process of the complex nonlinear neuron dynamics of the FP-SA laser as a photon pulse neuron is explained in detail.

Specifically, an FP-SA with a total length of 1500 μm and a saturable absorption region with a length of 60 μm is selected as an example for experiment, and a photon pulse neural network experiment platform as shown in fig. 2 is built, wherein details of each module are as follows:

wherein, the AWG is used for generating an electric signal; TL is used to provide an optical carrier (wavelength 1561.48 nm); after passing through the PC1, the optical carrier enters the MZM; here, PC1 can control the polarization state of the external-injection light to match the operating state of the MZM. An optical carrier carrying a signal is injected into a gain region of the FP-SA1 through the EDFA, the VOA, the OC1, the PC2, the OC2 and the CIRC. The gain region current of each FP-SA is provided by LDC, the reverse bias of the saturable absorption region is adjusted by VS, and the working temperature of each FP-SA is controlled by LDC. The optical power of the external light injected into the FP-SA1 can be adjusted by the EDFA and the VOA and can be detected by the PM. The PC2 can adjust the polarization state of external light entering the FP-SA1, so that the FP-SA1 is matched to realize the characteristics of the optical pulse neuron. The output of FP-SA1 is divided into 3 paths through OC3, one path can be subjected to spectrum analysis through OSA, the other path enters OSC for time sequence analysis and recording after being converted into an electric signal by PD, and the other path is injected into FP-SA2 through PC3, VOA, OC4 and an optical circulator. Similarly, the VOA mentioned here can adjust the amount of optical power injected into the FP-SA2, and the PC3 can adjust the polarization state of the injected light, thereby matching the operating state of the FP-SA2. Therefore, two FP-SA cascades can test the cascade of the characteristics of the optical pulse neurons.

First, an external stimulus signal (optical signal into FP-SA 1) as shown in sub-diagram (a) in fig. 3 is generated by configuring AWG to include three consecutive (pulse interval is 500 ps) small pulses and one small pulse of the same power. Sub-graph (b) in FIG. 3 is the impulse response generated by FP-SA 1; it can be seen that for three consecutive small pulses, there is a pulse generation of FP-SA1, while for a single small pulse of the same power, there is no pulse generation of FP-SA 1. This indicates that a single sub-threshold pulse cannot cause the FP-SA1 to reach the generation pulse, while three closely spaced sub-threshold pulses integrate in the time domain, thereby exceeding the threshold of the FP-SA1 generation pulse, triggering the FP-SA1 generation pulse, which indicates that the FP-SA has a time domain accumulation characteristic. Sub-diagram (c) in FIG. 3 is the impulse response generated by FP-SA 2; it can be seen that FP-SA2 has almost the same pulse output as FP-SA 1; it is worth emphasizing that the response of the FP-SA2 to the three continuous small pulses is not greatly weakened by the cascade position, and the FP-SA2 has almost no response to the single small pulse, and shows a larger extinction ratio, which indicates that the performance of the photonic pulse neural network formed by using the FP-SA cascade is better. The inventors have analyzed that this is because the FP-SA cascade can produce a combined saturable absorption effect.

Then, a perturbation signal as shown in sub-graph (a) in fig. 4 is generated by configuring the AWG to include three pulses of different powers. After the perturbation signal is injected into the FP-SA1, the response of the FP-SA1 is shown as a graph (b) in FIG. 4; it can be seen that FP-SA1 has little response to the 1 st and 2 nd perturbation pulses, and only excitatory response to the 3 rd perturbation pulse with higher power, which indicates that FP-SA has a threshold characteristic. Correspondingly, the impulse response of the FP-SA2 cascaded after the FP-SA1 corresponds to the output of the FP-SA1, which shows that the output response of the prior FP-SA1 can be propagated to the subsequent FP-SA2 through the cascade, namely, the FP-SA is verified to have the cascade characteristic.

In conclusion, FP-SA can simulate the accumulation, threshold and cascade characteristics of biological neurons, and has the performance and function of serving as photon pulse neurons.

Next, a determination process of the implementation method for implementing the photon pulse neuron based on the FP-SA laser will be described in detail.

Firstly, LDC is used for applying gain region current to FP-SA, VS is used for applying reverse bias voltage to the saturable absorption region of FP-SA, and the working temperature of FP-SA is regulated and controlled. The current, the reverse bias voltage and the operating temperature of the gain region are adjusted to obtain power current-voltage curves at different temperatures and different reverse bias voltages, as shown in fig. 5. Here, the purpose of obtaining this curve is to determine the threshold (current threshold) for FP-SA excitation at various temperatures and reverse bias pressures; as can be seen in FIG. 5, the larger the reverse bias voltage, the larger the threshold for FP-SA excitation; in addition, an increase in operating temperature also increases the threshold of the FP-SA laser.

Wherein, in the process of testing the power current-voltage curve with the reverse bias voltage of 0V, when the current of the gain region is larger than the threshold current 45mA at 0V, the light observed from the output of the FP-SAThe spectra are shown in FIG. 6. At this time, the FP-SA is excited to a multimode state. According to f = c/(2 n) _g ×L _cavity ) It can be known that the mode-locked pulse frequency of the FP-SA in the multimode state is 28.9GHz, wherein the distance between two adjacent modes is 0.24nm; where c is the speed of light in vacuum, n _g =3.46 is the group refractive index of the ridge waveguide of FP-SA.

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

Therefore, the critical point of the gain region current corresponding to the FP-SA self-pulse state under various reverse bias voltages can be obtained and used as the Q-switched pulse threshold corresponding to each reverse bias voltage. By setting the current of the gain region of the FP-SA to be slightly lower than the Q-switched pulse threshold, the FP-SA can carry out nonlinear response on incoherent light input into the FP-SA, and the nonlinear operation function of the photon pulse neuron is realized.

Specifically, referring to fig. 8, a method for implementing a photon pulse neuron based on a two-segment fabry-perot laser according to an embodiment of the present invention includes:

applying a bias electric signal to the FP-SA laser, and inputting an incoherent photon pulse signal to the FP-SA laser under the condition of maintaining the bias electric signal so as to utilize the neuron characteristics excited by the FP-SA laser to perform response processing (output response pulse) on the incoherent photon pulse signal and realize the function of a photon pulse neuron.

Wherein the bias electrical signal comprises: gain region current I applied to gain region of FP-SA _a And a reverse bias voltage V applied to a saturable absorption region of the FP-SA laser _s (ii) a Gain region current I _a Less than reverse bias voltage V _s Corresponding Q-switched pulse threshold I _Q And is sufficient to cause the FP-SA to enter into an excited state of operation. The Q-switched pulse threshold is defined as: FP-S without external light input and given reverse bias voltageThe a laser is excited out of the minimum gain region current for the Q-switched pulse state.

Preferably, in the above-mentioned bias electrical signal, the gain section current is equal to the Q-switched pulse threshold × a% corresponding to the reverse bias voltage, and a =80 to 95. In this way, the gain region current is not too far away from the Q-switched pulse threshold, and even a less-powered incoherent photon pulse signal can excite the FP-SA to generate an impulse response.

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

Preferably, the reverse bias voltage V of the saturable absorption region _s The value range of (a) is-6V to 0V.

In an alternative implementation, the manner of applying the gain region current to the FP-SA may include:

a gain section current is applied to the FP-SA using a laser diode controller, while the operating temperature of the FP-SA is controlled using the laser diode controller. Thus, the Q-switched pulse threshold of the FP-SA can be ensured to be stable and not to drift.

Of course, if the operating environment temperature of the FP-SA is relatively tightly controlled, a common current source may be used to apply the gain section current to the FP-SA. Alternatively, if the power of the incoherent photon pulse signal is large enough, the influence of the shift of the Q-switched pulse threshold value caused by the change of the working temperature on the normal work of the photon pulse neuron can be ignored.

In the method for realizing the photon pulse neuron based on the two-segment Fabry-Perot laser, provided by the embodiment of the invention, the FP-SA laser can be excited to generate neuron characteristics by applying a proper bias electric signal to the FP-SA laser, so that the incoherent photon pulse signal is subjected to response processing, and the function of the photon pulse neuron is realized. The neuron characteristics comprise time domain accumulation characteristics, so that the photon pulse neuron realized by the invention has time integration capability; moreover, as the output power of the FP-SA laser is in the megawatt level, the requirement of the power can be met even if cascade connection is carried out; in addition, the photon pulse neuron realized by the invention is a pure optical path from input to output thereof, a photoelectric conversion structure is not inserted, the realization mode only needs to apply proper bias to the FP-SA laser, and the realization mode is simple. Therefore, the photon pulse neuron realized by the method provided by the embodiment of the invention has better performance in all aspects. Moreover, as the FP-SA laser is already produced, the method provided by the embodiment of the invention can be compatible with the existing production process when the neuron is realized and then the photonic pulse neural network is realized by combining the neuron with the optical synapse device, thereby having higher application prospect and practical value.

Thus, photonic pulse neural networks can be implemented by integrating photonic pulse neurons implemented using embodiments of the present invention with silicon-photon-based weighting devices (e.g., weighting devices implemented using MRR or MZI networks) or InP-based weighting devices (e.g., weighting devices implemented using SOAs), etc.

It should be noted that, for the embodiment of the photonic pulse neuron chip, since it is substantially similar to the embodiment of the method, the description is simple, and the relevant points can be referred to the partial description of the embodiment of the method

In the description of the specification, references to descriptions of the terms "one embodiment," "some embodiments," "an example," "a specific example," or "some examples" or the like are intended to mean that a particular feature or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples described in this specification can be combined and combined by those skilled in the art.

While the present application has been described in connection with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed application, from a review of the drawings, the disclosure, and the appended claims.

The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.

Claims (8)

1. A method for realizing photon pulse neurons based on a two-section Fabry-Perot laser is characterized in that the two-section Fabry-Perot laser is a two-section Fabry-Perot-SA laser formed by introducing a saturable absorber SA into a Fabry-Perot FP cavity;

the method comprises the following steps:

applying a bias electric signal to the FP-SA, and inputting a non-coherent photon pulse signal to the FP-SA under the condition of maintaining the bias electric signal so as to respond and process the non-coherent photon pulse signal by utilizing the neuron characteristics excited by the FP-SA to realize the function of a photon pulse neuron;

wherein the bias electrical signal comprises: a gain region current applied to a gain region of the FP-SA, and a reverse bias voltage applied to a saturable absorption region of the FP-SA; the current of the gain region does not exceed a Q-switched pulse threshold corresponding to the reverse bias voltage and is enough to cause the FP-SA to enter an excitation state to work;

the Q-switched pulse threshold is defined as: the FP-SA is excited to emit the minimum gain region current of the Q-switched pulse state under the condition of no external light input and given reverse bias voltage.

2. The method of claim 1, wherein the neuron properties comprise: threshold characteristics, time domain accumulation characteristics, and cascadable characteristics.

3. The method of claim 1, wherein the reverse bias voltage is in a range of-6V to 0V.

4. The method of claim 1, wherein the wavelength difference between the wavelength of the incoherent photon pulse signal and the central operating wavelength of the FP-SA is 0nm ± 15nm.

5. The method of claim 1, wherein the gain section current in the bias electrical signal is equal to the threshold of the Q-switched pulse corresponding to the reverse bias voltage x A%, A = 80-95.

6. The method of claim 1, wherein applying the gain region current to the FP-SA comprises:

applying a gain region current to the FP-SA with a laser diode controller while controlling an operating temperature of the FP-SA with the laser diode controller.

7. The method of claim 1, wherein the length of the saturable absorption region is equal to the length of the gain region x B%, B = 1.6-7.

8. The method of claim 1, wherein the FP-SA is a PIN structure grown on the basis of AlGaInAs/InP material.

Images