CN-116011538-B - Photonic nerve synaptic device with double micro-ring structure and convolution operation network model

CN116011538BCN 116011538 BCN116011538 BCN 116011538BCN-116011538-B

Abstract

The invention discloses a photonic nerve synaptic device with a double-micro-ring structure and a convolution operation network model, wherein the device comprises a front resonant ring, a rear resonant ring and two coupling phase change material film layers, the two coupling phase change material film layers are matched and correspond to the front resonant ring and the rear resonant ring respectively, synaptic weight adjustment, namely synaptic weight, is completed by controlling the coupling phase change material film layers, and the front resonant ring and the rear resonant ring are used for 50% -50% of light splitting of an input optical signal in a bus waveguide and outputting differential signal current through a balanced photoelectric detector. The photonic nerve synapse has the advantages of high bandwidth and high speed, and is a passive device after the synapse weight is determined, so that the theoretical power consumption is zero. In addition, the invention designs a double micro-ring structure, the method does not influence the resonant wavelength during the light modulation, improves the modulation precision, reduces the loss of light transmission and widens the range of synaptic weights. The invention can be widely applied to the technical field of micro-nano photoelectrons.

Inventors

LI JINGXI
ZHOU SHAOLIN
HU BO
ZOU YI
CHEN ZHIJIAN
LI BIN

Assignees

华南理工大学

Dates

Publication Date: 20260505
Application Date: 20221226

Claims (8)

1. The photonic nerve synaptic device with the double-micro-ring structure is characterized by comprising a front resonant ring, a rear resonant ring and two coupling phase change material film layers, wherein the two coupling phase change material film layers are matched and correspond to the front resonant ring and the rear resonant ring respectively, and synaptic weight adjustment, namely synaptic weight adjustment, is completed by controlling the coupling phase change material film layers; The front resonant ring and the rear resonant ring are used for 50 percent to 50 percent of light splitting of input optical signals in the bus waveguide, and the light splitting is carried out through the balance photoelectric detector to output differential signal current; the coupling phase change material film layer structure is a multi-stage structure, n different phase change material units are arranged in the multi-stage structure, and each phase change material unit can be switched between a crystalline state and an amorphous state; realizing the graded change of the waveguide transmittance between 0 and 1 through a multi-level structure; The ring radius, the coupling gap and the number of the front resonant ring and the rear resonant ring are adjusted according to the requirements of the working wavelength, the bandwidth and the light splitting coefficient.
2. The photonic nerve synapse device with double-microring structure as in claim 1, wherein the phase change material selected for the coupling phase change material film layer is 。
3. The photonic nerve synapse device of double-microring structure of claim 1 in which the coupled phase change material film layer is used to achieve single-stage modulation.
4. A convolution operation network model, which is characterized by comprising a bus waveguide, a balanced photoelectric detector, a transimpedance amplifier and k photonic nerve synapse devices according to any one of claims 1-3, wherein k is a positive integer greater than 1; Front resonant rings and corresponding coupling phase change material film layers in the k photonic nerve synaptic devices form an upper waveguide, and rear resonant rings and corresponding coupling phase change material film layers in the k photonic nerve synaptic devices form a lower waveguide; The bus waveguide comprises all wavelength input optical signals, synaptic weighting is respectively realized after the input optical signals are modulated by a coupling phase change material film layer in the photon nerve synapse, namely, the optical signal intensities of different wavelengths are multiplied by corresponding synaptic weights, and finally, incoherent summation of all product signals is finished by the balanced photoelectric detector, so that the function of matrix operation is realized, and the operation result is represented by the magnitude of current signals; the transimpedance amplifier is used for amplifying signals output by the balanced photoelectric detector.
5. The convolution operation network model according to claim 4, wherein the coupling phase change material film layer can realize single-stage regulation, and further the balanced photoelectric detector is respectively input and regulated through the upper waveguide and the lower waveguide, so that the difference signal output of the upper waveguide and the lower waveguide is realized, namely, the weighted signal is output between-1, 0 and +1.
6. The convolution operation network model according to claim 5, wherein when the coupled phase change material film layers corresponding to the upper and lower waveguides are crystalline or amorphous, the output of the balanced photodetector is 0, namely, the weighted value is 0; when the coupling phase change material film layer corresponding to the upper waveguide is amorphous and the coupling phase change material film layer corresponding to the lower waveguide is crystalline, balancing the output forward current of the photoelectric detector, wherein the current represents a weight value of 1; when the coupling phase change material film layer corresponding to the upper waveguide is crystalline and the coupling phase change material film layer corresponding to the lower waveguide is amorphous, the reverse current output by the photoelectric detector is balanced, and the current represents a weight value of-1.
7. The convolution operation network model according to claim 4, wherein the coupled phase change material film layer structure is a multi-stage structure, n different phase change material units are arranged in the multi-stage structure, and multi-stage regulation and control are realized based on the plurality of phase change material units; When all the phase change material units are in an amorphous state, the optical signal transmittance in the waveguide is close to 1, when all the phase change material units are in a crystalline state, the optical signal transmittance in the waveguide is close to 0, when one part of the phase change material units are in the crystalline state and the other part of the phase change material units are in the amorphous state, the optical signal transmittance in the waveguide is between 0 and 1, and the optical transmittance value can be designed and adjusted according to the number, the structure and the size of the phase change material film units.
8. The convolutional network model of claim 7, wherein in the multi-level modulation mode of the photonic neurite device, the optical transmittance of the upper waveguide is The light transmittance of the lower waveguide is The balanced photoelectric detector converts the optical signals of the upper waveguide and the lower waveguide into electric signals and subtracts the electric signals, and the current output by the balanced photoelectric detector is And (3) with Linear correlation of Maximum value (maximum value), The minimum value is normalized as the synaptic weight 1 to achieve quasi-continuous regulation of multiple state values between-1 and 1.

Description

Photonic nerve synaptic device with double micro-ring structure and convolution operation network model Technical Field The invention relates to the field of multi-disciplinary crossing leading edge research of optics, integrated optoelectronics, neural network computation, nanomaterial, micro-nano processing and the like, in particular to a photonic nerve synapse device with a double-micro-ring structure and a convolution operation network model. Background In recent years, an Artificial Neural Network (ANN) has received a great deal of attention due to the wide application of machine learning, and with the rapid development of applications such as artificial intelligence and big data, the demands for the size and computational power of the neural network have been rapidly increased. Convolutional Neural Networks (CNNs) are powerful and widely used tools that extract features from large datasets, with an unparalleled position in computer vision and natural language processing. The convolutional neural network needs to perform a large amount of matrix operations in the training process, however, although the traditional von neumann architecture can also simulate the architecture of the neural network, the mode of separating the operations from data storage is not essentially suitable for realizing the neural network, the convolutional operation occupies more than 80% of the operation time of a computer, and a large amount of energy consumption is brought. Compared with the prior art, photons have unique advantages in energy consumption, bandwidth and speed, particularly in matrix calculation and interconnection, and the optical method has great potential, is naturally co-shooting with a neural network, and can support various large-scale neuron connections. The combination of ANN and optics provides a new computational architecture, the Optical Neural Network (ONN). ONN a neural network is seen as a collection of computing units and connections between computing units, a photonic neuron contains two important tasks, one is a weighted sum and the other is a nonlinear activation. The weighted summation comprises two methods of micro-ring weight group and MZI, wherein the micro-ring weight group transmits a plurality of optical signals in parallel on a main waveguide by utilizing WMD, micro-rings with different radiuses couple the optical signals with different wavelengths, and the output optical signals can be subjected to power adjustment by changing the coupling efficiency of the micro-rings, so that the loading of weight is realized, then the loaded optical signals are input into a pair of balanced photoelectric detectors PD, the summation of the plurality of optical signals is completed, and the multiply-accumulate operation MAC is realized. Changing the coupling efficiency of the micro-ring requires changing the effective refractive index of the micro-ring by methods such as thermo-optical modulation, electro-optical modulation, etc. However, since the effective refractive index of the material is changed little by the electro-optic effect and thermo-optic effect, to achieve a significant change in the effective refractive index, a large-sized modulator needs to be added, which is disadvantageous for integration. In addition, both of these methods are not nonvolatile, and require maintenance of heat or current to maintain efficient operation of the device, increasing power consumption. The phase change material has the characteristics of non-volatility, small volume, large change of effective refractive index between crystalline state and amorphous state, long state holding time and the like, and is an ideal material applied to micro-ring weight sets to reduce the integration size. In addition, for a particular single wavelength coupled single micro-ring structure, having pass-through and download ports, the weight distribution for the subtraction operation is achieved by feeding the outputs of the two ports into the balanced photodiode. The modulation of the effective refractive index by this method must be applied directly to the ring, which changes the resonant wavelength of the micro-ring, affecting the effect of wavelength division multiplexing. And a complete synaptic weight of-1 to 1 cannot be achieved due to the coupling loss of the waveguide to the micro-ring and the loss of the waveguide. Disclosure of Invention In order to solve at least one of the technical problems existing in the prior art to a certain extent, the invention aims to provide a photonic nerve synaptic device with a double-micro-ring structure and a convolution operation network model. The technical scheme adopted by the invention is as follows: The photonic nerve synaptic device with the double-micro-ring structure comprises a front resonant ring, a rear resonant ring and two coupling phase change material film layers, wherein the two coupling phase change material film layers are matched and correspond to the front resonant r