CN116258189A - Photon synapse weight matrix device of cross array and weight adjustment method thereof - Google Patents
Photon synapse weight matrix device of cross array and weight adjustment method thereof Download PDFInfo
- Publication number
- CN116258189A CN116258189A CN202310248817.XA CN202310248817A CN116258189A CN 116258189 A CN116258189 A CN 116258189A CN 202310248817 A CN202310248817 A CN 202310248817A CN 116258189 A CN116258189 A CN 116258189A
- Authority
- CN
- China
- Prior art keywords
- photoelectric
- path
- signal
- signals
- vcsel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/067—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means
- G06N3/0675—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means using electro-optical, acousto-optical or opto-electronic means
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Neurology (AREA)
- Optical Communication System (AREA)
Abstract
The invention provides a photon synapse weight matrix device based on a cross array and a photoelectric signal weight adjustment method, wherein the network comprises the following components: the tunable light source and the electro-optic modulator are used for acquiring at least two paths of photoelectric signals, wherein the two paths of photoelectric signals comprise: a first photoelectric signal and a second photoelectric signal; and the photon synapse crossing array is connected with the tunable light source and the electro-optic modulator and is used for respectively adjusting the pulse intensity of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal. The pulse intensity of the two paths of photoelectric signals is adjusted through the photon synapse crossing array, so that the data processing speed of the artificial intelligent neural network is improved.
Description
Technical Field
The invention relates to the field of artificial intelligent neural networks, in particular to a photon synapse weight matrix device of a cross array and a weight adjusting method thereof.
Background
In the big data age, artificial intelligent neural networks are proposed, which can simulate the neuromorphic operation of the human brain, abstract the intelligent behavior of the human brain into the information processing process of the artificial nerve at the computer end, analyze and infer a large amount of data, and further obtain an accurate result.
With the development of technology, based on the optical domain, phase change materials (Phase Change Materials, PCM) and micro-ring resonators (Microring Resonator, MRR) are adopted as synaptic weight matrices for realizing artificial intelligent neural networks.
However, in the optical domain, PCM speeds employed by the prior art are relatively slow, resulting in slower data processing speeds for existing artificial intelligence neural networks.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a photon synapse weight matrix device of a cross array and a weight adjusting method thereof, which improve the data processing speed of an artificial intelligent neural network.
The invention is realized by the following technical scheme:
in one aspect, the present invention provides a photonic synapse weight matrix device of a cross array, comprising:
the tunable light source and the electro-optic modulator are used for acquiring at least two paths of photoelectric signals, wherein the two paths of photoelectric signals comprise: a first photoelectric signal and a second photoelectric signal;
and the photon synapse crossing array is connected with the tunable light source and the electro-optic modulator and is used for respectively adjusting the pulse intensity of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal.
Further, the method comprises the steps of: a first optical fiber coupler and a second optical fiber coupler;
the tunable light source and the electro-optic modulator comprise at least a first channel and a second channel;
the first optical fiber coupler is connected with the first channel and is used for carrying out branching treatment on the first photoelectric signal of the first channel to obtain a first path of photoelectric signal and a second path of photoelectric signal;
and the second optical fiber coupler is connected with the second channel and is used for carrying out branching treatment on the second photoelectric signal of the second channel to obtain a third path of photoelectric signal and a fourth path of photoelectric signal.
Further, the photonic synapse crossing array comprises at least a first vertical cavity surface semiconductor emitting laser VCSEL, a second VCSEL, a third VCSEL and a fourth VCSEL;
the first VCSEL is used for receiving the first path of photoelectric signals and adjusting the pulse intensity of the first path of photoelectric signals to obtain a first path of photoelectric output signals;
the second VCSEL is used for receiving the second path of photoelectric signals and adjusting the pulse intensity of the second path of photoelectric signals to obtain a second path of photoelectric output signals;
the third VCSEL is used for receiving the third path of photoelectric signals and adjusting the pulse intensity of the third path of photoelectric signals to obtain a third path of photoelectric output signals;
the fourth VCSEL is used for receiving the fourth path of photoelectric signals and adjusting the pulse intensity of the fourth path of photoelectric signals to obtain fourth path of photoelectric output signals.
Further, the method further comprises the following steps: the waveform generator AWG is used for being respectively connected with the first VCSEL, the second VCSEL, the third VCSEL and the fourth VCSEL, applying bias current to the first VCSEL, the second VCSEL, the third VCSEL and the fourth VCSEL, and respectively adjusting pulse intensities of the first path of photoelectric signal, the second path of photoelectric signal, the third path of photoelectric signal and the fourth path of photoelectric signal to obtain adjusted photoelectric output signals.
Further, the method further comprises the following steps:
the device comprises a third optical fiber coupler, a fourth optical fiber coupler, a first photoelectric detector and a second photoelectric detector;
the third optical fiber coupler is respectively connected with the first VCSEL and the third VCSEL and is used for outputting a third coupling signal coupled by the third optical fiber coupler, the third coupling signal comprises a signal obtained by respectively weighting the first path of photoelectric output signal and the third path of photoelectric output signal, and the first photoelectric detector detects the third coupling signal;
the fourth optical fiber coupler is respectively connected with the second VCSEL and the fourth VCSEL and is used for outputting a fourth coupling signal coupled by the fourth optical fiber coupler, the fourth coupling signal comprises a signal obtained by respectively weighting a second path of photoelectric output signal and a fourth path of photoelectric output signal, and the second photoelectric detector detects the fourth coupling signal.
In another aspect, the present invention provides a method for adjusting the weight of an optical signal of a photonic synapse weight matrix device based on a cross array, including:
acquiring at least two paths of photoelectric signals, wherein the two paths of photoelectric signals comprise: a first photoelectric signal and a second photoelectric signal;
and respectively adjusting the pulse intensity of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal.
Further, the acquiring at least two paths of photoelectric signals further includes:
carrying out branching treatment on the first photoelectric signal generated by the first channel to obtain a first path of photoelectric signal and a second path of photoelectric signal;
and carrying out branching treatment on the second photoelectric signal generated by the second channel to obtain a third path of photoelectric signal and a fourth path of photoelectric signal.
Further, the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain adjusted photoelectric output signals includes:
the pulse intensity of the first path of photoelectric signals is adjusted to obtain first path of photoelectric output signals;
adjusting the pulse intensity of the second path of photoelectric signals to obtain second path of photoelectric output signals;
the pulse intensity of the third path of photoelectric signals is adjusted to obtain third path of photoelectric output signals;
and adjusting the pulse intensity of the fourth path of photoelectric signals to obtain fourth path of photoelectric output signals.
Further, the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain adjusted photoelectric output signals further includes:
obtaining bias current output by a waveform generator AWG;
respectively adjusting the pulse intensities of the first path of photoelectric signal, the second path of photoelectric signal, the third path of photoelectric signal and the fourth path of photoelectric signal;
and obtaining an adjusted photoelectric output signal.
Further, the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain adjusted photoelectric output signals further includes:
outputting a third coupling signal coupled by a third optical fiber coupler, wherein the third coupling signal comprises a signal obtained by respectively weighting the first path of photoelectric output signal and the third path of photoelectric output signal;
and outputting a fourth coupling signal coupled by a fourth optical fiber coupler, wherein the fourth coupling signal comprises a signal obtained by respectively weighting the second path of photoelectric output signals and the fourth path of photoelectric output signals.
Compared with the prior art, the invention has the following beneficial technical effects:
the embodiment of the invention provides a photon synapse weight matrix device based on a cross array and a weight adjusting method thereof, wherein the photon synapse weight matrix device of the cross array comprises: the tunable light source and the electro-optic modulator are used for acquiring at least two paths of photoelectric signals, wherein the two paths of photoelectric signals comprise: a first photoelectric signal and a second photoelectric signal; and the photon synapse crossing array is connected with the tunable light source and the electro-optic modulator and is used for respectively adjusting the pulse intensity of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal. The pulse intensity of the two paths of photoelectric signals is adjusted through the photon synapse crossing array, so that the data processing speed of the artificial intelligent neural network is improved.
Drawings
FIG. 1 is a schematic diagram of a photonic synaptic weight matrix device of a cross-array according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a photonic synaptic weight matrix device of a cross-array according to another embodiment of the present invention;
FIG. 3 is a simplified schematic diagram of a photonic synaptic weight matrix device of a2×2 cross array according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a photonic synapse weight matrix device for a cross-array in accordance with yet another embodiment of the invention;
fig. 5 is a flow chart illustrating a method for adjusting the weight of an optoelectronic signal of a photonic synapse weight matrix device based on a cross array in accordance with an embodiment of the present invention.
Detailed Description
The invention will now be described in further detail with reference to specific examples, which are intended to illustrate, but not to limit, the invention.
As shown in fig. 1 and 2, an embodiment of the present invention provides a photonic synapse weight matrix device of a cross array, including:
a tunable light source and electro-optical modulator 11 for acquiring at least two optical-electrical signals, the two optical-electrical signals comprising: a first photoelectric signal and a second photoelectric signal;
and the photon synapse crossing array 12 is connected with the tunable light source and the electro-optic modulator and is used for respectively adjusting the pulse intensities of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal.
For example, a first fiber coupler and a second fiber coupler are disposed between the tunable light source and the electro-optic modulator 11 and the photonic synapse crossing array 12;
the tunable light source and electro-optical modulator 11 comprises at least a first channel 111 and a second channel 112;
the first optical fiber coupler 131 connected to the first channel 111 is configured to split the first optical signal generated by the first channel 111 to obtain a first optical signal and a second optical signal;
the second optical fiber coupler 132 connected to the second channel 112 is configured to split the second optical signal of the second channel to obtain a third optical signal and a fourth optical signal.
In this embodiment, the VCSELs of the VCSOA are used as vertical cavity semiconductor optical amplifiers (vertical cavity semiconductor amplifier, VCSOA) which are capable of controlled weighting of optical pulses of 150ps long and with peak power of 1 μw. Because the VCSOA has nonlinear gain characteristics when being injected by external light, the full-weight adjustability of the input optical pulse of subnanoseconds can be realized by only changing the external bias current of the VCSOA in a static or dynamic mode. The VCSOA-based synapses are not only able to adjust the intensity of the incoming light pulses, but may also provide gain, i.e. a weighting factor > 1. The experimental scheme is built based on commercial VCSELs and optical fiber assemblies at a key communication wavelength of 1550nm, and the method is compatible with optical network and data center technologies.
Specifically, the photonic synapse crossing array 12 includes at least a first vertical cavity surface semiconductor emitting laser VCSEL121, a second VCSEL122, a third VCSEL123, a fourth VCSEL124;
the first VCSEL121 is configured to receive the first optical signal, and adjust a pulse intensity of the first optical signal to obtain a first optical output signal;
the second VCSEL122 is configured to receive the second optical signal, and adjust a pulse intensity of the second optical signal to obtain a second optical output signal;
the third VCSEL123 is configured to receive the third optical signal, and adjust pulse intensity of the third optical signal to obtain a third optical output signal;
the fourth VCSEL124 is configured to receive the fourth optical signal, and adjust a pulse intensity of the fourth optical signal to obtain a fourth optical output signal.
The photonic synapse weight matrix device of the cross array of the embodiment comprises: the tunable light source and the electro-optic modulator are used for acquiring at least two paths of photoelectric signals, wherein the two paths of photoelectric signals comprise: a first photoelectric signal and a second photoelectric signal; and the photon synapse crossing array is connected with the tunable light source and the electro-optic modulator and is used for respectively adjusting the pulse intensity of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal. The pulse intensity of the two paths of photoelectric signals is adjusted through the photon synapse crossing array, so that the data processing speed of the artificial intelligent neural network is improved.
As shown in fig. 3 and 4, another embodiment of the present invention further provides a photonic synapse weight matrix device of a cross array, which includes: the first VCSEL, the second VCSEL, the third VCSEL, and the fourth VCSEL respectively adjust pulse intensities of the first optical signal, the second optical signal, the third optical signal, and the fourth optical signal, including at least 2 implementations.
The first implementation mode: static adjustment;
the fixed bias current of each VCSEL is adopted to realize the adjustment of each path of photoelectric signal by adopting fixed amplification weight.
The second implementation mode: dynamically adjusting;
the waveform generator AWG is used for connecting the first VCSEL, the second VCSEL, the third VCSEL and the fourth VCSEL with the AWG, adding bias current to the first VCSEL, the second VCSEL, the third VCSEL and the fourth VCSEL, and respectively adjusting the pulse intensities of the first path of photoelectric signal, the second path of photoelectric signal, the third path of photoelectric signal and the fourth path of photoelectric signal to obtain adjusted photoelectric output signals. The AWG is adopted to output different bias currents to each VCSEL, and the adjustment of each path of photoelectric signals by adopting dynamic amplification weights is realized.
Further, the photonic synapse weight matrix device of the cross array is characterized by further comprising: a third optical fiber coupler 133, a fourth optical fiber coupler 134, a first photodetector PD1 (141), and a second photodetector PD2 (142);
the third optical fiber coupler 133 is connected to the first VCSEL121 and the third VCSEL123, respectively, and is configured to output a third coupling signal coupled to the third optical fiber coupler 133, where the third coupling signal includes a sum signal obtained by weighting the first optical output signal and the third optical output signal, respectively, and the first photodetector PD1 (141) detects the third coupling signal;
the fourth optical fiber coupler 134 is connected to the second VCSEL122 and the fourth VCSEL124, respectively, and is configured to output a fourth coupled signal coupled by the fourth optical fiber coupler, where the fourth coupled signal includes a sum signal obtained by weighting the second optical output signal and the fourth optical output signal, respectively, and the second photodetector detection PD2 (142) detects the fourth coupled signal.
Specifically, the first optical-electrical signal generated by the first tunable optical source TL1 and the second optical-electrical signal generated by the second tunable optical source TL2 are output through an optical isolator ISO and then a variable optical attenuator VOA, where the optical isolator ISO is used to avoid unnecessary reflection of the transmitted optical signal, the variable optical attenuator VOA may control the output power of the optical signal, and then the optical signal is polarization-processed by a polarization controller PC and then is introduced into a mach-zehnder modulator MZM, where a 15MHz Pulse Generator (PG) is used to modulate injection of TL1 together with MZM1, modulate injection of TL2 together with MZM2, and then use a polarization controller PC to match the polarization of the optical signal to a parallel resonant mode of the VCSEL, and use two 50: the optical fiber coupler 50 splits each path of modulated light into two paths, the optical signals generated by TL1 are respectively injected into the VCSOA11 and the VCSOA12 through the optical circulator, and the optical signals generated by TL2 are respectively injected into the VCSOA21 and the VCSOA22 through the optical circulator. The incident light pulses are weighted according to the operating point of the VCSOA and then combined in a fast 9GHz amplified photodetector before analysis, the output of the VCSOA11 and the VCSOA21 passing through 50: the coupler 50 is then connected to the oscilloscope 1 channel via the photodetector PD1 for testing, and the outputs of the VCSOA12 and VCSOA22 are then passed through the coupler 50: the 50 coupler was passed through PD2 again and the results of post2 were observed on an oscilloscope in 2 channels.
For the generation of 4 channel electrical signals in a2 x 2VCSOA crossover array using an arbitrary waveform generator AWG, the tunable laser TL provides an optical carrier.
Static adjustment, i.e. static weights: the static weights were measured at the respective wavelengths for 150ps long pulses injected into the devices using 4 VCSOAs (operating wavelength 1550 nm), these weight values providing the maximum amplification of the light pulses injected into each device, and appropriate bias current settings w11, w21, w12, w22 were selected in the VCSOAs 11, 21, 12, 22, respectively, each VCSOA yielding a corresponding amplification factor. The electrical detector collects the output pulses obtained after the combination of the two VCSOA synapses at the system output. This value is related to the sum of the two weighted pulses w11 and w21, thus effecting a multiplication-addition of the synapses, the weight matrix being
For VCSOA21 and VCSOA22, the same operation yields the output of post2, post 2= (w12×pre1) + (w22×pre2).
The static weights applied to VCSOA synapses can be used to perform the multiply-add operation of the synapses with low input power (pulse peak power of only a few tens of μw) and very small applied bias current variations (tens of μa), the limits of input speed being set by the recovery time of the VCSOA and determined by the carrier lifetime, which is typically-1 ns for VCSELs used in this scheme.
Dynamic adjustment, i.e. dynamic weight: for high-speed dynamic adjustment of the synaptic weight of the VCSOA, the AWG controls the synaptic system based on the VCSOA to quickly adapt to a new task or training set without manually setting parameters. In the proposed scheme, a VCSOA cross array with dynamic (time varying) weights is used. The input pulse coded optical signals generated by PG are injected into the VCSOA cross array, 4 paths of different signals are generated by using the AWG to modulate the bias current of each VCSOA, the amplitude levels of the pulses generated by the four VCSOA output ends are consistent with the level signals generated by the corresponding AWG, the weights w11, w12, w21 and w22 respectively correspond to the weights, and the input pulses are multiplied by the configurable weights before all VCSOA branches are added in the photoelectric detector (without considering the wavelengths of the weights), namely, the combined weights of the output pulses finish the multiplication and addition operation of synapses.
In general, the weighted sum operation of fast light pulses can be performed by VCSOA cross array synapses with high speed and user-defined weighted control. Furthermore, in each configuration studied, we dynamically weighted it at high speed by adjusting the applied bias current; thus allowing for quick and remote adjustment of synaptic weights. This allows our proposed photonic cross-array synapses to be easily readjusted for different network tasks or training periods. Furthermore, this scheme was built based on commercial VCSEL and fiber optic assemblies at critical communication wavelengths (1550 nm); thus making our method fully compatible with optical communication network and data center technologies. Therefore, the scheme utilizes the advantages of the VCSEL in terms of hardware friendliness, low power consumption, high speed and fast tuning weight in photon synapse, and is used for a future neuromorphic photon spike processing platform.
As shown in fig. 5, an embodiment of the present invention provides a method for adjusting the weight of an optical signal of a photonic synapse weight matrix device based on a cross array, which includes the following steps:
The two-path photoelectric signal of the embodiment includes: a first photoelectric signal and a second photoelectric signal;
The method for adjusting the weight of the photoelectric signal of the photon synapse weight matrix device of the cross array in the embodiment comprises the following steps: the tunable light source and the electro-optic modulator are used for acquiring at least two paths of photoelectric signals, wherein the two paths of photoelectric signals comprise: a first photoelectric signal and a second photoelectric signal; and the photon synapse crossing array is connected with the tunable light source and the electro-optic modulator and is used for respectively adjusting the pulse intensity of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal. The pulse intensity of the two paths of photoelectric signals is adjusted through the photon synapse crossing array, so that the data processing speed of the artificial intelligent neural network is improved.
On the basis of the foregoing embodiment, the acquiring at least two paths of photoelectric signals further includes:
branching the first photoelectric signal of the first channel to obtain a first path of photoelectric signal and a second path of photoelectric signal;
and carrying out branching treatment on the second photoelectric signal of the second channel to obtain a third path of photoelectric signal and a fourth path of photoelectric signal.
Further, on the basis of the foregoing embodiment, the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain adjusted photoelectric output signals includes:
the pulse intensity of the first path of photoelectric signals is adjusted to obtain first path of photoelectric output signals;
adjusting the pulse intensity of the second path of photoelectric signals to obtain second path of photoelectric output signals;
the pulse intensity of the third path of photoelectric signals is adjusted to obtain third path of photoelectric output signals;
and adjusting the pulse intensity of the fourth path of photoelectric signals to obtain fourth path of photoelectric output signals.
Further, on the basis of the foregoing embodiment, the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain adjusted photoelectric output signals further includes:
obtaining bias current output by a waveform generator AWG;
respectively adjusting the pulse intensities of the first path of photoelectric signal, the second path of photoelectric signal, the third path of photoelectric signal and the fourth path of photoelectric signal;
and obtaining an adjusted photoelectric output signal.
Further, on the basis of the foregoing embodiment, the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain adjusted photoelectric output signals further includes:
outputting a third coupling signal, wherein the third coupling signal comprises a signal obtained by respectively weighting the first path of photoelectric output signal and the third path of photoelectric output signal;
and outputting a fourth coupling signal, wherein the fourth coupling signal comprises a signal obtained by respectively weighting the second path of photoelectric output signals and the fourth path of photoelectric output signals.
The working principle and technical effects of the photoelectric signal weight adjustment method of the photon synapse weight matrix device of the cross array provided in this embodiment are similar to those described in fig. 4, and are not described here again.
In the description of the present invention, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic point described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristic data points described may be combined in any suitable manner in any one or more embodiments or examples. Further, one skilled in the art can engage and combine the different embodiments or examples described in this specification.
The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.
Claims (10)
1. A photonic synapse weight matrix device of a cross-array, comprising:
the tunable light source and the electro-optic modulator are used for acquiring at least two paths of photoelectric signals, wherein the two paths of photoelectric signals comprise: a first photoelectric signal and a second photoelectric signal;
and the photon synapse crossing array is connected with the tunable light source and the electro-optic modulator and is used for respectively adjusting the pulse intensity of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal.
2. The photonic synaptic weight matrix device of claim 1, further comprising: a first optical fiber coupler and a second optical fiber coupler;
the tunable light source includes at least a first channel and a second channel;
the electro-optic modulator includes: a first electro-optic modulator and a second electro-optic modulator;
the first optical fiber coupler is connected with the first channel and is used for carrying out branching treatment on the first photoelectric signal of the first channel to obtain a first path of photoelectric signal and a second path of photoelectric signal;
and the second optical fiber coupler is connected with the second channel and is used for carrying out branching treatment on the second photoelectric signal of the second channel to obtain a third path of photoelectric signal and a fourth path of photoelectric signal.
3. The photonic synaptic weight matrix device of claim 2, the photonic synaptic crossover array comprising at least a first vertical cavity surface semiconductor emitting laser VCSEL, a second VCSEL, a third VCSEL, a fourth VCSEL;
the first VCSEL is used for receiving the first path of photoelectric signals and adjusting the pulse intensity of the first path of photoelectric signals to obtain a first path of photoelectric output signals;
the second VCSEL is used for receiving the second path of photoelectric signals and adjusting the pulse intensity of the second path of photoelectric signals to obtain a second path of photoelectric output signals;
the third VCSEL is used for receiving the third path of photoelectric signals and adjusting the pulse intensity of the third path of photoelectric signals to obtain a third path of photoelectric output signals;
the fourth VCSEL is used for receiving the fourth path of photoelectric signals and adjusting the pulse intensity of the fourth path of photoelectric signals to obtain fourth path of photoelectric output signals.
4. The photonic synaptic weight matrix device of claim 3, further comprising: the waveform generator AWG is used for being respectively connected with the first VCSEL, the second VCSEL, the third VCSEL and the fourth VCSEL, applying bias current to the first VCSEL, the second VCSEL, the third VCSEL and the fourth VCSEL, and respectively adjusting pulse intensities of the first path of photoelectric signal, the second path of photoelectric signal, the third path of photoelectric signal and the fourth path of photoelectric signal to obtain adjusted photoelectric output signals.
5. The photonic synaptic weight matrix device of claim 4, further comprising: the device comprises a third optical fiber coupler, a fourth optical fiber coupler, a first photoelectric detector and a second photoelectric detector;
the third optical fiber coupler is respectively connected with the first VCSEL and the third VCSEL and is used for outputting a third coupling signal coupled by the third optical fiber coupler, the third coupling signal comprises a signal obtained by respectively weighting the first path of photoelectric output signal and the third path of photoelectric output signal, and the first photoelectric detector detects the third coupling signal;
the fourth optical fiber coupler is respectively connected with the second VCSEL and the fourth VCSEL and is used for outputting a fourth coupling signal coupled by the fourth optical fiber coupler, the fourth coupling signal comprises a signal obtained by respectively weighting a second path of photoelectric output signal and a fourth path of photoelectric output signal, and the second photoelectric detector detects the fourth coupling signal.
6. A method for adjusting the weight of an optical signal of a photonic synapse weight matrix device based on a cross array, comprising:
acquiring at least two paths of photoelectric signals, wherein the two paths of photoelectric signals comprise: a first photoelectric signal and a second photoelectric signal;
and respectively adjusting the pulse intensity of the two paths of photoelectric signals to obtain an adjusted photoelectric output signal.
7. The method for adjusting the weight of the optical-electrical signal of the photonic synapse weight matrix device based on the cross array according to claim 6, wherein the acquiring at least two paths of optical-electrical signals further comprises:
branching the first photoelectric signal of the first channel to obtain a first path of photoelectric signal and a second path of photoelectric signal;
and carrying out branching treatment on the second photoelectric signal of the second channel to obtain a third path of photoelectric signal and a fourth path of photoelectric signal.
8. The method for adjusting the weights of the photoelectric signals of the photonic synapse weight matrix device based on the cross array according to claim 7, wherein the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain the adjusted photoelectric output signals comprises:
the pulse intensity of the first path of photoelectric signals is adjusted to obtain first path of photoelectric output signals;
adjusting the pulse intensity of the second path of photoelectric signals to obtain second path of photoelectric output signals;
the pulse intensity of the third path of photoelectric signals is adjusted to obtain third path of photoelectric output signals;
and adjusting the pulse intensity of the fourth path of photoelectric signals to obtain fourth path of photoelectric output signals.
9. The method for adjusting the weights of the photoelectric signals of the photonic synapse weight matrix device based on the cross array according to claim 8, wherein the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain adjusted photoelectric output signals further comprises:
obtaining bias current output by a waveform generator AWG;
respectively adjusting the pulse intensities of the first path of photoelectric signal, the second path of photoelectric signal, the third path of photoelectric signal and the fourth path of photoelectric signal;
and obtaining an adjusted photoelectric output signal.
10. The method for adjusting the weights of the photoelectric signals of the photonic synapse weight matrix device based on the cross array according to claim 9, wherein the adjusting the pulse intensities of the two paths of photoelectric signals respectively to obtain adjusted photoelectric output signals further comprises:
outputting a third coupling signal, wherein the third coupling signal comprises a signal obtained by respectively weighting the first path of photoelectric output signal and the third path of photoelectric output signal;
and outputting a fourth coupling signal, wherein the fourth coupling signal comprises a signal obtained by respectively weighting the second path of photoelectric output signals and the fourth path of photoelectric output signals.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310248817.XA CN116258189A (en) | 2023-03-15 | 2023-03-15 | Photon synapse weight matrix device of cross array and weight adjustment method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310248817.XA CN116258189A (en) | 2023-03-15 | 2023-03-15 | Photon synapse weight matrix device of cross array and weight adjustment method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116258189A true CN116258189A (en) | 2023-06-13 |
Family
ID=86684221
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310248817.XA Pending CN116258189A (en) | 2023-03-15 | 2023-03-15 | Photon synapse weight matrix device of cross array and weight adjustment method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116258189A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116756475A (en) * | 2023-06-27 | 2023-09-15 | 北京信息科技大学 | Jitter control method and system for micro-ring resonator weight matrix |
-
2023
- 2023-03-15 CN CN202310248817.XA patent/CN116258189A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116756475A (en) * | 2023-06-27 | 2023-09-15 | 北京信息科技大学 | Jitter control method and system for micro-ring resonator weight matrix |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112001487A (en) | Photon neural network | |
| CN116258189A (en) | Photon synapse weight matrix device of cross array and weight adjustment method thereof | |
| CN115222035B (en) | Photon neural network convolution acceleration chip | |
| Alanis et al. | Weight adjustable photonic synapse by nonlinear gain in a vertical cavity semiconductor optical amplifier | |
| Yu et al. | Reconfigurable low-threshold all-optical nonlinear activation functions based on an add-drop silicon microring resonator | |
| Moss | Microcombs for ultrahigh bandwidth optical data transmission and neural networks | |
| US6804047B2 (en) | All-optical or gate embodied by using semiconductor optical amplifiers | |
| Xiang et al. | Experimental demonstration of a photonic spiking neuron based on a DFB laser subject to side-mode optical pulse injection | |
| Gao et al. | Cascadable excitability and inhibition in DFB laser-based photonic spiking neurons | |
| Tan et al. | Advanced applications of microcombs: From optical neural networks to data transmission | |
| Ohno et al. | Demonstration of classification task using optical neural network based on Si microring resonator crossbar array | |
| Capmany et al. | Reconfigurable fiber-optic delay line filters incorporating electrooptic and electroabsorption modulators | |
| Kan et al. | Silicon photonic delay-based reservoir computing incorporated with a multimode interference structure | |
| Cheng et al. | Measurement-based model for cross-absorption modulation in an MQW electroabsorption modulator | |
| Thakur et al. | Optical wavelength converters based on cross gain modulation and cross phase modulation in SOA | |
| Li et al. | SOA-based all-optical neuron with reconfigurable nonlinear activation functions | |
| Mourgias-Alexandris et al. | Experimental demonstration of an optical neuron with a logistic Sigmoid activation function | |
| Dionísio et al. | Experimental study of a phase modulator using an active interferometric device | |
| Chrysostomidis et al. | 10 GHz Optoelectronic Circuit with Variable Sigmoid Shaped Activation Function for Photonic Neural Networks | |
| Lupo et al. | Parallel and deep reservoir computing based on frequency multiplexing | |
| Rashed | Ultra high speed LiNbO3 and polymer electrooptic modulators in lightwave optical access communication networks | |
| Stephanie et al. | Photonic Neuron with on Frequency-Domain ReLU Activation Function | |
| Tan et al. | Microcombs for ultrahigh bandwidth optical data transmission and neural networks | |
| EP4160976A1 (en) | Low loss routing for quantum communication | |
| CN115775016A (en) | Device for realizing nonlinear neural computation based on dual-wavelength optical pulse injection FP-SA |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |