A method for manufacturing a micro-ring resonator-based activation function device

By fabricating a microring resonator and a coupled ring PN junction, setting up an auxiliary light source and output device, and fitting the ReLU activation function, the problem of realizing the activation function of the microring resonator in the optical neural network was solved, thus improving the computational efficiency and accuracy.

CN116151346BActive Publication Date: 2026-03-24HARBIN ENG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-02-28
Publication Date
2026-03-24

AI Technical Summary

Technical Problem

How to use microring resonators to implement activation functions to improve the computational efficiency and accuracy of optical neural networks.

Method used

By fabricating a microring resonator, coupling a ring PN junction to the microring, setting up an auxiliary light source and output signal measurement equipment, and using a TIA transimpedance amplifier and bias unit, the ReLU activation function is fitted to realize the activation function of the optical neural network.

Benefits of technology

Optical devices that implement activation functions can achieve efficient fitting under low power consumption, improving the computational efficiency and accuracy of optical neural networks, and the device is small in size and easy to integrate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116151346B_ABST
    Figure CN116151346B_ABST
Patent Text Reader

Abstract

The application aims to provide a kind of micro-ring resonator-based activation function device manufacturing method, comprising the following steps: determining the standard nonlinear function to be fitted by optical device;Preparation of micro-ring resonator;Coupling ring PN junction and micro-ring;Set up auxiliary light source and output signal measurement equipment;Set up TIA transimpedance amplifier and bias unit.The application can realize the function of activation function by optical device, and it is relatively easy to combine with other optical elements that can realize the summation of neural unit weight, realize the complete function of single neuron, thereby improve the operation efficiency and operation accuracy of optical neural network.The device is small in size and easy to integrate, and the operation efficiency and operation accuracy of the optical neural network integrated by the neural unit integrated by the device can be significantly improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to an activation function device, and more specifically, to a method for manufacturing an activation function device. Background Technology

[0002] In recent years, the demand for optical computing technology has increased rapidly, mainly due to two factors: First, with the gradual deterioration of Moore's Law and the ever-increasing demands for power consumption and speed in the era of big data, the high-speed and low-power characteristics of optical computing technology are receiving increasing attention. Second, the parallel computing capabilities of optical computing technology, along with the development of algorithms and hardware architectures such as optical neural networks, provide the most promising solutions for the computing power requirements of artificial intelligence technologies such as image recognition, speech recognition, and virtual reality. Optical computing can be divided into analog optical computing and digital optical computing. A typical example of analog optical computing is Fourier transform computation, which is required in fields such as image processing for calculations related to Fourier transforms, such as convolution. Calculating Fourier transforms using traditional computers is computationally very demanding, while the process of light passing through a lens is itself a Fourier transform, a process that takes almost no time. Digital optical computing utilizes light and optical devices to form classic logic gates, constructing a computing system similar to the principles of traditional digital electronic computing, and achieving computation through complex combinations of logic gates.

[0003] Microring resonators (MRRs) possess numerous advantages such as small area and low power consumption, making them an important integrated optoelectronic device with wide applications in optical communication, optical computing, and photobiological sensing. In the field of photonic neural networks, utilizing MRRs to implement neural network computations is currently a common method. However, using optical devices such as MRRs to implement nonlinear operations such as activation functions can further realize optical neural networks, thereby significantly improving computational efficiency and accuracy. Therefore, how to implement activation functions using optical devices such as MRRs is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide a method for fabricating an activation function device based on a microring resonator, which can significantly improve computational efficiency and accuracy.

[0005] The objective of this invention is achieved as follows:

[0006] This invention discloses a method for fabricating an activation function device based on a microring resonator, characterized by:

[0007] (1) Determine the standard nonlinear function to be fitted by the optical device;

[0008] (2) Fabrication of micro-ring resonators;

[0009] (3) Couple the ring PN junction to the micro-ring;

[0010] (4) Set up auxiliary light sources and output signal measurement equipment;

[0011] (5) Set up the TIA transimpedance amplifier and bias unit.

[0012] The present invention may also include:

[0013] 1. Step (1) is as follows:

[0014] The ReLU function was chosen as the standard nonlinear function to be implemented.

[0015] A target fitting function corresponding to the standard nonlinear function is determined, and the target fitting function is realized through optical devices;

[0016] The target fitting function is used as the target activation function of the photonic neural network.

[0017] 2. Step (2) is as follows:

[0018] Electron beam lithography is used to spin a layer on the SOI substrate to form a mask layer with a thickness of 110 nm.

[0019] The pattern of a micro-ring resonator with a radius of 50 μm was written onto photoresist using an electron beam lithography machine. After development, the pattern was transferred onto the photoresist.

[0020] After inductively coupled plasma reactive ion etching, the microring resonator is transferred to the upper silicon layer of the substrate. The coupling length of the microring resonator is 10μm to 15μm.

[0021] 3. Step (3) is as follows:

[0022] The fabricated fiber microring is fixed on a cleaned glass slide. A uniform ring PN junction with a radius larger than that of the microring is taken. The thickness of the ring PN junction is 10μm, with the P pole on the outside and the N pole on the inside. It is made into an arc that occupies 2 / 3 and 1 / 6 of the circumference of the ring. The microring is placed at the junction of the P pole and N pole in the 2 / 3 arc PN junction and embedded therein.

[0023] Take the N-terminal of a 1 / 6 circular arc PN junction and place it directly below the inner side of the micro-ring and close to the micro-ring. Take the N-terminal of a 1 / 6 circular arc PN junction and place it directly below the outer side of the micro-ring, with a distance of 10-12μm from the micro-ring. The two circular arc PN junctions should not touch each other.

[0024] The four arc-shaped PN junctions are connected to the negative terminal of the external voltage, with the N-terminal PN junction inside the micro-ring connected to the right end of the arc and the P-terminal PN junction outside the micro-ring connected to the left end of the arc.

[0025] 4. Step (4) is as follows:

[0026] A 1560nm continuous laser was selected as the auxiliary light source for the final signal output.

[0027] After connecting the optical fiber to a 1560nm wavelength laser, the middle part of the optical fiber is coupled to the bottom of the fixed micro-ring. This part of the optical fiber is located between the bottom of the micro-ring and the 1 / 6 outer P-pole PN junction, and the remaining part extends to the optical signal power measurement end.

[0028] 5. Step (5) is as follows:

[0029] Select a transimpedance amplifier and apply a bias voltage Vb to its output. After applying the bias voltage, the circuit extends to the center of the inner side of the micro-ring.

[0030] The circuit in the inner center of the microring includes four branches, two of which are connected to the left end of the PN junction inside the microring, and the other two are connected to the right end of the PN junction outside the microring.

[0031] The advantages of this invention are:

[0032] (1) The activation function can be fitted by optical devices, and it is relatively easy to combine with other optical elements that can realize the summation of neural unit weights to realize the complete function of a single neuron, thereby improving the computational efficiency and accuracy of optical neural networks.

[0033] (2) This device is small in size and easy to integrate. The optical neural network obtained by integrating the neural units obtained by combining this device can have its computational efficiency and accuracy significantly improved. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of a neural network;

[0035] Figure 2 This is a schematic diagram of a mathematical model of a neural unit;

[0036] Figure 3 This is a schematic diagram of the ReLU activation function (the target fitting function for this device);

[0037] Figure 4 This is a schematic diagram of an activation function device based on a microring resonator.

[0038] Figure 5 The signal output diagram of the resonant region of the micro-ring before PN junction modulation is shown.

[0039] Figure 6 The signal output diagram is shown after the resonant region of the micro-ring is modulated by the PN junction.

[0040] Figure 7This is a schematic diagram of the signal output fitted by the activation function device based on a microring resonator. Detailed Implementation

[0041] The invention will now be described in more detail with reference to the accompanying drawings:

[0042] Combination Figure 1-7 This invention provides a method for fabricating an activation function device based on a microring resonator, the scheme of which includes:

[0043] 1. Determine the standard nonlinear function to be fitted to the optical device.

[0044] (1) Determine whether to use the ReLU function as the standard nonlinear function to be implemented by this device.

[0045] (2) Determine the target fitting function corresponding to the standard nonlinear function. The target fitting function is realized through optical devices.

[0046] (3) Use the target fitting function as the target activation function of the photonic neural network.

[0047] 2. Fabrication of microring resonators.

[0048] (1) A mask layer with a thickness of 110 nm is formed by electron beam lithography on the SOI substrate.

[0049] (2) The pattern of a micro-ring resonator with a radius of 50 μm is written on the photoresist using an electron beam exposure machine. After development, the pattern is transferred to the photoresist.

[0050] (3) After inductively coupled plasma reactive ion etching (ICP-RIE), the microring resonator is transferred to the upper silicon layer of the substrate. The coupling length of the microring resonator is 10 μm to 15 μm.

[0051] 3. Coupling between the toroidal PN junction and the microring.

[0052] (1) Fix the fabricated fiber microring onto the cleaned glass slide. Take a uniform ring PN junction with a radius larger than the microring radius (50μm). The ring PN junction has a thickness of 10μm, with the P pole on the outside and the N pole on the inside. Fabricate it into an arc that occupies 2 / 3 and 1 / 6 of the circumference of the ring. Place the microring at the junction of the P pole and N pole in the 2 / 3 arc PN junction and embed it therein.

[0053] (2) Take the N pole of the 1 / 6 circular arc PN junction and place it directly below the inner side of the micro ring and close to the micro ring. Take the P pole of the 1 / 6 circular arc PN junction and place it directly below the outer side of the micro ring, with a distance of 10-12μm from the micro ring. The two circular arc PN junctions do not contact each other.

[0054] (3) Connect the four arc PN junctions to the negative terminal of the external voltage respectively, wherein the N-terminal PN junction inside the micro-ring is connected to the right end of the arc, and the P-terminal PN junction outside the micro-ring is connected to the left end of the arc.

[0055] 4. Setting up auxiliary light sources and output signal measurement equipment.

[0056] (1) The auxiliary light source is used as the final signal output. The original resonant wavelength of the micro-ring is 1560nm, so a continuous laser of 1560nm is selected as the auxiliary light source.

[0057] (2) After connecting a sufficiently long optical fiber to a 1560nm wavelength laser, couple a middle portion of the optical fiber to the bottom of the fixed microring. This portion of the optical fiber is located between the bottom of the microring and the 1 / 6 outer P-polar PN junction. The remaining portion extends to the optical signal power measurement end.

[0058] 5. TIA transimpedance amplifier and bias unit settings.

[0059] (1) Select a transimpedance amplifier with a gain of 3dB (which can be adjusted as needed), and add a bias voltage Vb to its output terminal. After the bias voltage is applied, the circuit extends to the center of the inner side of the micro-ring.

[0060] (2) The circuit in the inner center of the micro-ring is divided into 4 branches, of which 2 are connected to the left end of the PN junction (N-pole part) inside the micro-ring, and the other 2 are connected to the right end of the PN junction (P-pole part) outside the micro-ring.

[0061] This invention, as a device capable of implementing activation functions in neural units, can be combined with other necessary devices to form neural units, and further, neural networks. A neural network is an algorithmic mathematical model that mimics the behavioral characteristics of animal neural networks, performing distributed parallel information processing. This type of network, depending on the complexity of the system, achieves its information processing purpose by adjusting the interconnections between a large number of internal nodes. See also... Figure 1 This is a schematic diagram of a simple three-layer neural network, consisting of an input layer, hidden layers (also called intermediate layers), and an output layer. Activation functions play a crucial role in enabling neural network models to learn and understand highly complex and nonlinear functions, introducing nonlinear characteristics into the neural network. See also... Figure 2 This is a schematic diagram of a neural unit mathematical model, such as... Figure 2 As shown, in a neuron, the input Xn is weighted, summed, and then subjected to a function, which is the activation function. The activation function is introduced to increase the non-linearity of the neural network model. Without an activation function, each layer would be equivalent to matrix multiplication, meaning that any number of hidden layers without activation functions could be replaced by a single hidden layer.

[0062] Since most problems encountered in artificial intelligence computing scenarios are nonlinear, and introducing nonlinear activation functions into neural networks can enhance their computational power, nonlinearity is the most prominent characteristic of activation functions. The activation function fitted in this invention is described below:

[0063] The formula for the corrected linear unit function ReLU can be expressed as:

[0064] ReLU = max(0, x). See also Figure 3 The figure shows the function curve of the ReLU activation function. The characteristics of ReLU as an activation function are: 1. ReLU is simple to compute, improving computation speed; 2. The effective derivative of ReLU is a constant 1, solving the gradient vanishing problem; 3. It is easy to obtain better models, so ReLU is often used in CNN classification operations.

[0065] The technical solutions in the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] See Figure 4 The present invention provides a schematic diagram of an activation function device based on a microring resonator. The device includes: a TIA transimpedance amplifier 1, a bias voltage 2, a circuit 3, a microring resonator 4, a parallel positive voltage circuit 5, a parallel negative voltage circuit 6 and 7, a P-type PN junction 8 and 9, an N-type PN junction 10 and 11, a 1560nm continuous laser 12, a single-mode optical fiber 13, and a power meter 14.

[0067] Specifically:

[0068] (1) After the input of a neuron is weighted and summed, it will be activated by an activation function. This process is reflected in the following process in this device: all input optical signals in the photonic neuron device are summed and converted into a current signal I. sum The current signal I sum Enter the activation function device.

[0069] (2) Current signal I sum After being processed by the TIA (Trans-Impedance Amplifier) ​​of the activation function device, a voltage signal V is generated. TIA V TIA After superimposing an external voltage signal V b This generates a voltage signal V.m The voltage signal V m It is divided into 4 parallel branches, which are connected to the 4 parts of the PN junction respectively.

[0070] (3) Each PN junction has a built-in threshold voltage V. th When the external voltage V m <V th At this time, the charge carriers inside the PN junction do not change, thus not modulating the microring. The resonant wavelength λ of the microring... r No movement occurred, see Figure 5 This is the signal output diagram of the resonant region of the micro-ring before modulation by the PN junction. At this time, the auxiliary light source λ... L Completely coupled and absorbed by microrings (corresponding to) Figure 5 (a)) No signal is generated at the output terminal, that is, the signal power is 0 (corresponding to Figure 5 (b)).

[0071] (4) When the external voltage V m >V th At this time, the charge carriers inside the PN junction change, generating current, which modulates the micro-ring, V m The larger the value, the more pronounced the effect; modulation of the PN junction leads to the resonant wavelength λ of the microring. r Movement occurred, see Figure 6 This is a diagram showing the signal output of the micro-ring modulated by a PN junction in its resonant region. The resonant wavelength λ... r The movement of the auxiliary light source λ, coupled and absorbed within the microring. L Fewer and fewer (corresponding) Figure 6 (a)) The corresponding output signal becomes larger and larger, that is, the signal power becomes larger and larger (corresponding to Figure 6 (b)).

[0072] (5) By combining processes (3) and (4), this device can achieve fitting of the ReLU activation function with relatively low power consumption. See [link to relevant documentation]. Figure 7 This is a schematic diagram of the signal output of a microring resonator-based activation function device fitting the ReLU function. The location of the point where the output signal is generated can be adjusted by changing the external bias voltage V. b Modulation based on size: V b If Vb increases, the point shifts to the left; if Vb decreases, the point shifts to the right.

Claims

1. A method for fabricating an activation function device based on a microring resonator, characterized in that: (1) Determine the standard nonlinear function to be fitted to the optical device; (2) Fabrication of a microring resonator; (3) Coupled to a ring PN junction and a microring; The fabricated fiber microring is fixed on a cleaned glass slide. A uniform ring PN junction with a radius larger than that of the microring is taken. The thickness of the ring PN junction is 10μm, with the P pole on the outside and the N pole on the inside. It is made into an arc that occupies 2 / 3 and 1 / 6 of the circumference of the ring. The microring is placed at the junction of the P pole and N pole in the 2 / 3 arc PN junction and embedded therein. Take the N-terminal of a 1 / 6 circular arc PN junction and place it directly below the inner side of the micro-ring and close to the micro-ring. Take the N-terminal of a 1 / 6 circular arc PN junction and place it directly below the outer side of the micro-ring, with a distance of 10-12μm from the micro-ring. The two circular arc PN junctions should not touch each other. The four arc PN junctions are connected to the negative terminal of the external voltage, with the N-terminal PN junction inside the micro-ring connected to the right end of the arc and the P-terminal PN junction outside the micro-ring connected to the left end of the arc. (4) Set up auxiliary light sources and output signal measurement equipment; (5) Set up the TIA transimpedance amplifier and bias unit.

2. The method for fabricating an activation function device based on a microring resonator according to claim 1, characterized in that: Step (1) is as follows: The ReLU function was chosen as the standard nonlinear function to be implemented. A target fitting function corresponding to the standard nonlinear function is determined, and the target fitting function is realized through optical devices; The target fitting function is used as the target activation function of the photonic neural network.

3. The method for fabricating an activation function device based on a microring resonator according to claim 1, characterized in that: Step (2) is as follows: Electron beam lithography is used to spin a layer on the SOI substrate to form a mask layer with a thickness of 110 nm. The pattern of a micro-ring resonator with a radius of 50 μm was written onto photoresist using an electron beam lithography machine. After development, the pattern was transferred onto the photoresist. After inductively coupled plasma reactive ion etching, the microring resonator is transferred to the upper silicon layer of the substrate. The coupling length of the microring resonator is 10μm to 15μm.

4. The method for fabricating an activation function device based on a microring resonator according to claim 1, characterized in that: Step (4) is as follows: A 1560nm continuous laser was selected as the auxiliary light source for the final signal output. After connecting the optical fiber to a 1560nm wavelength laser, the middle part of the optical fiber is coupled to the bottom of the fixed micro-ring. This part of the optical fiber is located between the bottom of the micro-ring and the 1 / 6 outer P-pole PN junction, and the remaining part extends to the optical signal power measurement end.

5. The method for fabricating an activation function device based on a microring resonator according to claim 1, characterized in that: Step (5) is as follows: Select a transimpedance amplifier and apply a bias voltage Vb to its output. After applying the bias voltage, the circuit extends to the center of the inner side of the micro-ring. The circuit in the inner center of the microring includes four branches, two of which are connected to the left end of the PN junction inside the microring, and the other two are connected to the right end of the PN junction outside the microring.

Citation Information

Patent Citations

  • Method and device for achieving all-optical nonlinear activation function of optical neural network

    CN111860822A

  • Non-linear activation function RELU implementation method, equipment and medium

    CN113673677A