A cross-array photon synaptic weight matrix device and its weight adjustment method

Abstract

The present invention provides a crossbar array-based photonic synaptic weight matrix device and a method for adjusting photoelectric signal weights. The network includes a tunable light source and an electro-optical modulator for acquiring at least two photoelectric signals, each comprising a first photoelectric signal and a second photoelectric signal. A photonic synaptic crossbar array, connected to the tunable light source and the electro-optical modulator, adjusts the pulse intensity of each of the two photoelectric signals to obtain adjusted photoelectric output signals. By adjusting the pulse intensity of the two photoelectric signals using the crossbar array, the data processing speed of an artificial intelligence neural network is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence neural networks, and in particular to a cross-array photon synaptic weight matrix device and a weight adjustment method thereof. Background Art

[0002] In the era of big data, people have proposed artificial intelligence neural networks, which can simulate the neuromorphic operations of the human brain, abstract the intelligent behavior of the human brain into the information processing process of artificial nerves on the computer side, and analyze and reason about large amounts of data to obtain accurate results.

[0003] With the development of technology, based on the optical domain, people have studied the use of phase change materials (PCM) and microring resonators (MRR) as synaptic weight matrices to realize artificial intelligence neural networks.

[0004] However, in the optical domain, the PCM speed adopted by existing technologies is relatively slow, resulting in slow data processing speed of existing artificial intelligence neural networks. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention provides a cross-array photon synaptic weight matrix device and a weight adjustment method thereof to improve the speed of data processing of artificial intelligence neural networks.

[0006] The present invention is achieved through the following technical solutions:

[0007] In one aspect, the present invention provides a cross-array photon synaptic weight matrix device, comprising:

[0008] A tunable light source and an electro-optical modulator are used to obtain at least two photoelectric signals, wherein the two photoelectric signals include a first photoelectric signal and a second photoelectric signal;

[0009] The photon synapse cross array connected to the tunable light source and the electro-optical modulator is used to adjust the pulse intensity of the two photoelectric signals respectively to obtain adjusted photoelectric output signals.

[0010] Further, it includes: a first optical fiber coupler and a second optical fiber coupler;

[0011] The tunable light source and the electro-optical modulator include at least a first channel and a second channel;

[0012] The first optical fiber coupler connected to the first channel is used to split the first photoelectric signal of the first channel to obtain a first photoelectric signal and a second photoelectric signal;

[0013] The second optical fiber coupler connected to the second channel is used to split the second photoelectric signal of the second channel to obtain a third photoelectric signal and a fourth photoelectric signal.

[0014] Furthermore, the photon synapse crossbar array comprises at least a first vertical cavity surface semiconductor emitting laser VCSEL, a second VCSEL, a third VCSEL, and a fourth VCSEL;

[0015] The first VCSEL is used to receive the first photoelectric signal and adjust the pulse intensity of the first photoelectric signal to obtain a first photoelectric output signal;

[0016] The second VCSEL is used to receive the second photoelectric signal and adjust the pulse intensity of the second photoelectric signal to obtain a second photoelectric output signal;

[0017] The third VCSEL is used to receive the third photoelectric signal and adjust the pulse intensity of the third photoelectric signal to obtain a third photoelectric output signal;

[0018] The fourth VCSEL is used to receive the fourth photoelectric signal and adjust the pulse intensity of the fourth photoelectric signal to obtain a fourth photoelectric output signal.

[0019] Furthermore, it also includes: a waveform generator AWG, which is used to be connected to the first VCSEL, the second VCSEL, the third VCSEL, and the fourth VCSEL, respectively, and to apply bias current to the first VCSEL, the second VCSEL, the third VCSEL, and the fourth VCSEL, and to adjust the pulse intensity of the first photoelectric signal, the second photoelectric signal, the third photoelectric signal, and the fourth photoelectric signal, respectively, to obtain adjusted photoelectric output signals.

[0020] Furthermore, it also includes:

[0021] a third optical fiber coupler, a fourth optical fiber coupler, a first photodetector, and a second photodetector;

[0022] The third optical fiber coupler is connected to the first VCSEL and the third VCSEL respectively, and is used to output a third coupled signal coupled by the third optical fiber coupler, wherein the third coupled signal includes a signal obtained by weighting the first photoelectric output signal and the third photoelectric output signal, and the first photodetector detects the third coupled signal;

[0023] The fourth optical fiber coupler is connected to the second VCSEL and the fourth VCSEL, respectively, and is configured to output a fourth coupled signal coupled by the fourth optical fiber coupler, wherein the fourth coupled signal includes a signal obtained by weighting the second photoelectric output signal and the fourth photoelectric output signal, respectively. The second photodetector detects the fourth coupled signal.

[0024] In another aspect, the present invention provides a method for adjusting photoelectric signal weights of a crossbar array-based photonic synaptic weight matrix device, comprising:

[0025] Acquire at least two photoelectric signals, the two photoelectric signals comprising: a first photoelectric signal and a second photoelectric signal;

[0026] The pulse intensities of the two photoelectric signals are adjusted respectively to obtain adjusted photoelectric output signals.

[0027] Furthermore, the step of obtaining at least two photoelectric signals further includes:

[0028] Splitting the first photoelectric signal generated by the first channel to obtain a first photoelectric signal and a second photoelectric signal;

[0029] The second photoelectric signal generated by the second channel is split to obtain a third photoelectric signal and a fourth photoelectric signal.

[0030] Furthermore, the pulse intensities of the two photoelectric signals are adjusted respectively to obtain adjusted photoelectric output signals, including:

[0031] Adjusting the pulse intensity of the first photoelectric signal to obtain a first photoelectric output signal;

[0032] Adjusting the pulse intensity of the second photoelectric signal to obtain a second photoelectric output signal;

[0033] Adjusting the pulse intensity of the third photoelectric signal to obtain a third photoelectric output signal;

[0034] The pulse intensity of the fourth photoelectric signal is adjusted to obtain a fourth photoelectric output signal.

[0035] Furthermore, the step of adjusting the pulse intensities of the two photoelectric signals to obtain adjusted photoelectric output signals further includes:

[0036] Get the bias current output by the waveform generator AWG;

[0037] respectively adjusting the pulse intensities of the first photoelectric signal, the second photoelectric signal, the third photoelectric signal, and the fourth photoelectric signal;

[0038] Obtain the adjusted photoelectric output signal.

[0039] Furthermore, the step of adjusting the pulse intensities of the two photoelectric signals to obtain adjusted photoelectric output signals further includes:

[0040] Outputting a third coupled signal coupled by the third optical fiber coupler, wherein the third coupled signal includes a signal obtained by weighting the first optoelectronic output signal and the third optoelectronic output signal;

[0041] A fourth coupled signal coupled by the fourth optical fiber coupler is output, where the fourth coupled signal includes a signal of a weighted sum of the second optoelectronic output signal and the fourth optoelectronic output signal.

[0042] Compared with the prior art, the present invention has the following beneficial technical effects:

[0043] An embodiment of the present invention provides a crossbar array-based photonic synaptic weight matrix device and a weight adjustment method thereof. The crossbar array photonic synaptic weight matrix device includes: a tunable light source and an electro-optical modulator for acquiring at least two photoelectric signals, the two photoelectric signals comprising a first photoelectric signal and a second photoelectric signal; and a photonic synaptic crossbar array connected to the tunable light source and the electro-optical modulator for adjusting the pulse intensity of the two photoelectric signals to obtain adjusted photoelectric output signals. By adjusting the pulse intensity of the two photoelectric signals using the crossbar array, the data processing speed of the artificial intelligence neural network is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic structural diagram of a cross-array photon synapse weight matrix device according to an embodiment of the present invention;

[0045] Figure 2 2 is a schematic structural diagram of a cross-array photon synapse weight matrix device according to another embodiment of the present invention;

[0046] Figure 3 This is a simplified structural diagram of a 2×2 crossbar array photon synaptic weight matrix device according to an embodiment of the present invention;

[0047] Figure 4 2 is a schematic structural diagram of a cross-array photon synapse weight matrix device according to another embodiment of the present invention;

[0048] Figure 5 It is a flowchart of a method for adjusting photoelectric signal weights in a cross-array-based photonic synaptic weight matrix device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described in detail below with reference to specific embodiments, which are intended to explain the present invention rather than to limit it.

[0050] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a cross-array photon synapse weight matrix device, comprising:

[0051] The tunable light source and the electro-optical modulator 11 are used to obtain at least two photoelectric signals, wherein the two photoelectric signals include a first photoelectric signal and a second photoelectric signal;

[0052] The photon synapse crossbar array 12 connected to the tunable light source and the electro-optical modulator is used to adjust the pulse intensity of the two photoelectric signals respectively to obtain adjusted photoelectric output signals.

[0053] For example, a first fiber coupler and a second fiber coupler are provided between the tunable light source and electro-optic modulator 11 and the photonic synapse crossbar array 12;

[0054] The tunable light source and electro-optical modulator 11 includes at least a first channel 111 and a second channel 112;

[0055] The first optical fiber coupler 131 connected to the first channel 111 is used to split the first photoelectric signal generated by the first channel 111 to obtain a first photoelectric signal and a second photoelectric signal;

[0056] The second optical fiber coupler 132 connected to the second channel 112 is used to split the second optical signal of the second channel to obtain a third optical signal and a fourth optical signal.

[0057] It should be noted that in this embodiment, a vertical cavity semiconductor surface emitting laser (VCSEL) biased below its laser threshold is used as a vertical cavity semiconductor amplifier (VCSOA), which is capable of controlled weighting of optical pulses with a length of 150 ps and a peak power of 1 μW. Since the VCSOA has nonlinear gain characteristics when injected with external light, full weight adjustability of sub-nanosecond input optical pulses can be achieved by simply changing the external bias current of the VCSOA statically or dynamically. The VCSOA-based synapse can not only adjust the intensity of the incoming optical pulse, but also provide gain, that is, the weighting factor is greater than 1. The experimental scheme is based on commercial VCSELs and optical fiber components at the key communication wavelength of 1550 nm. The method is compatible with optical network and data center technologies.

[0058] Specifically, the photon synapse crossbar array 12 includes at least a first vertical cavity semiconductor surface emitting laser VCSEL 121 , a second VCSEL 122 , a third VCSEL 123 , and a fourth VCSEL 124 ;

[0059] The first VCSEL 121 is used to receive the first photoelectric signal and adjust the pulse intensity of the first photoelectric signal to obtain a first photoelectric output signal;

[0060] The second VCSEL 122 is used to receive the second photoelectric signal and adjust the pulse intensity of the second photoelectric signal to obtain a second photoelectric output signal;

[0061] The third VCSEL 123 is used to receive the third photoelectric signal and adjust the pulse intensity of the third photoelectric signal to obtain a third photoelectric output signal;

[0062] The fourth VCSEL 124 is configured to receive the fourth photoelectric signal and adjust the pulse intensity of the fourth photoelectric signal to obtain a fourth photoelectric output signal.

[0063] The crossbar array photon synaptic weight matrix device of this embodiment includes a tunable light source and an electro-optical modulator for acquiring at least two photoelectric signals, the two photoelectric signals comprising a first photoelectric signal and a second photoelectric signal; and a photonic crossbar array connected to the tunable light source and the electro-optical modulator for adjusting the pulse intensity of the two photoelectric signals to obtain adjusted photoelectric output signals. By adjusting the pulse intensity of the two photoelectric signals using the crossbar array, the data processing speed of the artificial intelligence neural network is improved.

[0064] like Figure 3 and Figure 4 As shown, another embodiment of the present invention further provides a cross-array photon synaptic weight matrix device, which, based on the above embodiment, includes: a first VCSEL, a second VCSEL, a third VCSEL, and a fourth VCSEL respectively adjusting the pulse intensity of the first photoelectric signal, the second photoelectric signal, the third photoelectric signal, and the fourth photoelectric signal, including at least two implementation methods.

[0065] The first implementation method: static adjustment;

[0066] By using a fixed bias current for each VCSEL, it is possible to adjust each photoelectric signal using a fixed amplification weight.

[0067] The second implementation method: dynamic adjustment;

[0068] A waveform generator (AWG) is configured to apply bias currents to the first, second, third, and fourth VCSELs connected to the AWG, thereby adjusting the pulse intensities of the first, second, third, and fourth photoelectric signals, respectively, to obtain adjusted photoelectric output signals. The AWG outputs different bias currents to each VCSEL, thereby dynamically adjusting the photoelectric signals using amplification weights.

[0069] Furthermore, the cross-array photon synaptic weight matrix device is characterized in that it also includes: a third fiber coupler 133, a fourth fiber coupler 134, a first photodetector PD1 (141) and a second photodetector PD2 (142);

[0070] The third optical fiber coupler 133 is connected to the first VCSEL 121 and the third VCSEL 123 respectively, and is used to output a third coupled signal coupled by the third optical fiber coupler 133, wherein the third coupled signal includes a signal obtained by weighting the first photoelectric output signal and the third photoelectric output signal, and the first photodetector PD1 (141) detects the third coupled signal;

[0071] The fourth optical fiber coupler 134 is connected to the second VCSEL 122 and the fourth VCSEL 124 respectively, and is used to output a fourth coupled signal coupled by the fourth optical fiber coupler, wherein the fourth coupled signal includes a signal obtained by weighting the second photoelectric output signal and the fourth photoelectric output signal respectively, and the second photodetector PD2 (142) detects the fourth coupled signal.

[0072] Specifically, the first photoelectric signal generated by the first tunable light source TL1 and the second photoelectric signal generated by the second tunable light source TL2 first pass through the optical isolator ISO and then are output through the variable optical attenuator VOA. The optical isolator ISO is used to avoid unnecessary reflections of the transmitted optical signal, and the variable optical attenuator VOA can control the output power of the optical signal. Then, the optical signal is polarized by the polarization controller PC and introduced into the Mach-Zehnder modulator MZM. At this time, the 15MHz pulse generator (PG) is used together with MZM1 to modulate the injection of TL1, and together with MZM2 to modulate the injection of TL2. Then, the polarization controller PC is used to match the polarization of the optical signal to the parallel resonant mode of the VCSEL, and two 50:50 fiber couplers are used to split each modulated light into two paths. The optical signal generated by TL1 is injected into VCSOA11 and VCSOA12 respectively through the optical circulator, and the optical signal generated by TL2 is injected into VCSOA21 and VCSOA22 respectively through the optical circulator. The incident light pulses are weighted according to the VCSOA operating point and then combined in a fast 9 GHz amplified photodetector before analysis. The outputs of VCSOA11 and VCSOA21 pass through a 50:50 coupler and then photodetector PD1, which is connected to oscilloscope channel 1 for testing. The outputs of VCSOA12 and VCSOA22 pass through a 50:50 coupler and then PD2, and the post 2 results are observed on the oscilloscope channel 2.

[0073] In a 2×2 VCSOA crossbar array, an arbitrary waveform generator (AWG) is used to generate four-channel electrical signals, and a tunable laser (TL) provides an optical carrier.

[0074] Static adjustment, i.e. static weighting: Four VCSOAs (operating at a wavelength of 1550 nm) are used to measure the static weights of the 150 ps long pulses injected into the device at their respective wavelengths. These weight values ​​provide the maximum amplification of the light pulse injected into each device. Appropriate bias current settings w11, w21, w12, and w22 are selected in VCSOA11, VCSOA21, VCSOA12, and VCSOA22, respectively. Each VCSOA generates a corresponding amplification factor. The electrical detector collects the output pulse obtained after the synaptic combination of the two VCSOAs at the output of the system. This value is related to the sum of the two weighted pulses w11 and w21, thus realizing the multiplication and addition operation of the synapse. The weight matrix is For VCSOA21 and VCSOA22, the same operation obtains the output of post2, post2=(w12×pre1)+(w22×pre2).

[0075] The static weights applied to the VCSOA synapse can be used to perform synaptic multiplication-addition operations at low input power (pulse peak power is only tens of μW) and with very small changes in applied bias current (tens of μA). The limit on input speed is set by the recovery time of the VCSOA and determined by the carrier lifetime, which is typically ~1 ns for the VCSEL used in this scheme.

[0076] Dynamic adjustment, or dynamic weighting: To dynamically adjust the synaptic weights of the VCSOA at high speed, the AWG controls the VCSOA-based synaptic system to quickly adapt to new tasks or training sets without the need for manual parameter setting. In the proposed scheme, a VCSOA crossbar array with dynamic (time-varying) weights is used. The input pulse-encoded optical signal generated by the PG is injected into the VCSOA crossbar array. The AWG is used to generate four different signals to modulate the bias current of each VCSOA. The pulse amplitude levels generated at the output of the four VCSOAs are consistent with the level signals generated by the corresponding AWGs, corresponding to weights w11, w12, w21, and w22, respectively. Before all VCSOA branches are added to the photodetector (regardless of their wavelength), the input pulses are multiplied by the configurable weights. In other words, the combined weighting of the output pulses completes the synaptic multiplication and addition operation.

[0077] In summary, weighted sum operations of fast optical pulses can be performed using VCSOA crossbar array synapses with high-speed and user-defined weight control. Furthermore, in each configuration studied, we dynamically adjust the weights at high speed by adjusting the applied bias current, thus allowing for rapid and remote tuning of synaptic weights. This allows our proposed photonic crossbar array synapses to be easily retuned for different network tasks or training cycles. Furthermore, this scheme is built using commercially available VCSEL and fiber components at a critical communications wavelength (1550nm), making our approach fully compatible with optical communication networks and data center technologies. Thus, this scheme leverages the advantages of VCSELs in photonic synapses in terms of hardware friendliness, low power consumption, high speed, and fast weight tuning for future neuromorphic photonic spike processing platforms.

[0078] like Figure 5 As shown, an embodiment of the present invention provides a method for adjusting the photoelectric signal weight of a photonic synaptic weight matrix device based on a crossbar array, comprising the following steps:

[0079] Step 501: Acquire at least two photoelectric signals.

[0080] The two optoelectronic signals of this embodiment include: a first optoelectronic signal and a second optoelectronic signal;

[0081] Step 502: Adjust the pulse intensities of the two photoelectric signals respectively to obtain adjusted photoelectric output signals.

[0082] The photoelectric signal weight adjustment method of a crossbar array photonic synaptic weight matrix device of this embodiment includes: a tunable light source and an electro-optical modulator for obtaining at least two photoelectric signals, the two photoelectric signals comprising a first photoelectric signal and a second photoelectric signal; and a photonic synaptic crossbar array connected to the tunable light source and the electro-optical modulator for adjusting the pulse intensity of each of the two photoelectric signals to obtain an adjusted photoelectric output signal. By adjusting the pulse intensity of the two photoelectric signals using the photonic synaptic crossbar array, the data processing speed of the artificial intelligence neural network is improved.

[0083] Based on the above embodiment, the acquiring of at least two optoelectronic signals further includes:

[0084] Splitting the first photoelectric signal of the first channel to obtain a first photoelectric signal and a second photoelectric signal;

[0085] The second photoelectric signal of the second channel is split to obtain a third photoelectric signal and a fourth photoelectric signal.

[0086] Furthermore, based on the above embodiment, the pulse intensities of the two photoelectric signals are adjusted respectively to obtain adjusted photoelectric output signals, including:

[0087] Adjusting the pulse intensity of the first photoelectric signal to obtain a first photoelectric output signal;

[0088] Adjusting the pulse intensity of the second photoelectric signal to obtain a second photoelectric output signal;

[0089] Adjusting the pulse intensity of the third photoelectric signal to obtain a third photoelectric output signal;

[0090] The pulse intensity of the fourth photoelectric signal is adjusted to obtain a fourth photoelectric output signal.

[0091] Furthermore, based on the above embodiment, the pulse intensities of the two photoelectric signals are adjusted respectively to obtain adjusted photoelectric output signals, further comprising:

[0092] Get the bias current output by the waveform generator AWG;

[0093] respectively adjusting the pulse intensities of the first photoelectric signal, the second photoelectric signal, the third photoelectric signal, and the fourth photoelectric signal;

[0094] Obtain the adjusted photoelectric output signal.

[0095] Furthermore, based on the above embodiment, the pulse intensities of the two photoelectric signals are adjusted respectively to obtain adjusted photoelectric output signals, further comprising:

[0096] Outputting a third coupled signal, the third coupled signal comprising a signal obtained by weighting the first photoelectric output signal and the third photoelectric output signal;

[0097] A fourth coupled signal is output, where the fourth coupled signal includes a signal obtained by weighting the second optoelectronic output signal and the fourth optoelectronic output signal.

[0098] The working principle and technical effect of the photoelectric signal weight adjustment method of the cross array photon synapse weight matrix device provided in this embodiment are similar to those described above. Figure 4 The working principles and technical effects are similar and will not be described in detail here.

[0099] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0100] In the description of this specification, reference to the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" means that the specific features, structures, materials, or special features 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 representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or special features described can be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0101] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A cross-array photon synaptic weight matrix device, characterized in that: include: A tunable light source and an electro-optical modulator are used to obtain at least two photoelectric signals, wherein the two photoelectric signals include a first photoelectric signal and a second photoelectric signal; A photonic synaptic crossbar array connected to the tunable light source and the electro-optical modulator, for adjusting the pulse intensities of the two photoelectric signals respectively to obtain adjusted photoelectric output signals; The photon synapse crossbar array comprises at least a first VCSEL, a second VCSEL, a third VCSEL, and a fourth VCSEL; The first VCSEL is used to receive a first photoelectric signal and adjust the pulse intensity of the first photoelectric signal to obtain a first photoelectric output signal; The second VCSEL is used to receive a second photoelectric signal and adjust the pulse intensity of the second photoelectric signal to obtain a second photoelectric output signal; The third VCSEL is used to receive a third photoelectric signal and adjust the pulse intensity of the third photoelectric signal to obtain a third photoelectric output signal; The fourth VCSEL is used to receive a fourth photoelectric signal and adjust the pulse intensity of the fourth photoelectric signal to obtain a fourth photoelectric output signal; The first photoelectric signal and the second photoelectric signal are signals obtained by branching the first photoelectric signal, and the third photoelectric signal and the fourth photoelectric signal are signals obtained by branching the second photoelectric signal. The device further includes: a waveform generator AWG, configured to be connected to the first VCSEL, the second VCSEL, the third VCSEL, and the fourth VCSEL, respectively, and to apply bias currents to the first VCSEL, the second VCSEL, the third VCSEL, and the fourth VCSEL, and to adjust the pulse intensities of the first photoelectric signal, the second photoelectric signal, the third photoelectric signal, and the fourth photoelectric signal, respectively, to obtain adjusted photoelectric output signals; Also included: a third optical fiber coupler, a fourth optical fiber coupler, a first photodetector, and a second photodetector; The third optical fiber coupler is connected to the first VCSEL and the third VCSEL respectively, and is used to output a third coupled signal coupled by the third optical fiber coupler, wherein the third coupled signal includes a signal obtained by weighting the first photoelectric output signal and the third photoelectric output signal, and the first photodetector detects the third coupled signal; The fourth optical fiber coupler is connected to the second VCSEL and the fourth VCSEL, respectively, and is configured to output a fourth coupled signal coupled by the fourth optical fiber coupler, wherein the fourth coupled signal includes a signal obtained by weighting the second photoelectric output signal and the fourth photoelectric output signal, respectively. The second photodetector detects the fourth coupled signal.

2. The crossbar array photon synaptic weight matrix device according to claim 1, characterized in that: Also includes: a first fiber optic coupler and a second fiber optic coupler; The tunable light source comprises at least a first channel and a second channel; The electro-optical modulator comprises: a first electro-optical modulator and a second electro-optical modulator; The first optical fiber coupler connected to the first channel is used to split the first photoelectric signal of the first channel to obtain a first photoelectric signal and a second photoelectric signal; The second optical fiber coupler connected to the second channel is used to split the second photoelectric signal of the second channel to obtain a third photoelectric signal and a fourth photoelectric signal.

3. A method for adjusting the photoelectric signal weight of a cross-array-based photonic synaptic weight matrix device, characterized in that: include: Acquire at least two photoelectric signals, the two photoelectric signals comprising: a first photoelectric signal and a second photoelectric signal; respectively adjusting the pulse intensities of the two photoelectric signals to obtain adjusted photoelectric output signals; The pulse intensities of the two photoelectric signals are adjusted respectively to obtain adjusted photoelectric output signals, further comprising: Outputting a third coupled signal, the third coupled signal comprising a signal obtained by weighting the first photoelectric output signal and the third photoelectric output signal; Outputting a fourth coupled signal, the fourth coupled signal comprising a signal obtained by weighting the second photoelectric output signal and the fourth photoelectric output signal; The first photoelectric output signal and the second photoelectric output signal are signals obtained by performing branching processing and adjusting the pulse intensity of the first photoelectric signal; The third photoelectric output signal and the fourth photoelectric output signal are signals obtained by performing branching processing and pulse intensity adjustment on the second photoelectric signal.

4. The photoelectric signal weight adjustment method of the crossbar array-based photonic synaptic weight matrix device according to claim 3, characterized in that: The first photoelectric signal is located in a first channel, the second photoelectric signal is located in a second channel, and acquiring at least two photoelectric signals further includes: Splitting the first photoelectric signal of the first channel to obtain a first photoelectric signal and a second photoelectric signal; The second photoelectric signal of the second channel is split to obtain a third photoelectric signal and a fourth photoelectric signal.

5. The photoelectric signal weight adjustment method of the crossbar array-based photonic synaptic weight matrix device according to claim 4, characterized in that: The pulse intensities of the two photoelectric signals are adjusted respectively to obtain adjusted photoelectric output signals, including: Adjusting the pulse intensity of the first photoelectric signal to obtain a first photoelectric output signal; Adjusting the pulse intensity of the second photoelectric signal to obtain a second photoelectric output signal; Adjusting the pulse intensity of the third photoelectric signal to obtain a third photoelectric output signal; The pulse intensity of the fourth photoelectric signal is adjusted to obtain a fourth photoelectric output signal.

6. The photoelectric signal weight adjustment method of the crossbar array-based photonic synaptic weight matrix device according to claim 5, characterized in that: The pulse intensities of the two photoelectric signals are adjusted respectively to obtain adjusted photoelectric output signals, further comprising: Get the bias current output by the waveform generator AWG; respectively adjusting the pulse intensities of the first photoelectric signal, the second photoelectric signal, the third photoelectric signal, and the fourth photoelectric signal; Obtain the adjusted photoelectric output signal.

Images