Photoelectric synaptic device based on oxygen plasma assistance and application of photoelectric synaptic device in neuromorphic calculation
By adopting oxygen plasma assist technology and multi-channel photosynthesis device architecture in photoelectric synaptic devices, the existing photoelectric synaptic devices lack the light modulation capability and nonlinearity of synaptic weights is solved, and efficient optical information processing and recognition accuracy are achieved.
Patent Information
- Application Number
- CN202510325914.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-19
- Publication Date
- 2025-06-27
AI Technical Summary
Existing photoelectric synaptic devices lack the optical modulation capability and cannot achieve the coordinated processing of optical information perception and storage. The nonlinearity of synaptic weight changes may lead to the accumulation of errors in neural network training, affecting the calculation accuracy.
Using oxygen plasma-assisted photoelectric synaptic devices, the oxygen plasma treatment is performed on the surface of the two-dimensional tellurene nanosheet layer to achieve dynamic conversion of short-term plasticity and long-term plasticity under light stimulation, and feature extraction and classification are performed through the convolutional neural network architecture of multi-channel photosynthesis devices.
The hyperlinear synaptic weight change under light stimulation was achieved, the dynamic range was significantly better than the existing two-dimensional material synaptic devices, and the recognition accuracy was improved through a multi-channel architecture.
Smart Images
Figure CN120224798A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of semiconductors, and specifically to an oxygen plasma-assisted optoelectronic synaptic device and its application in neuromorphic computing. Background Art
[0002] In recent years, neuromorphic computing technology has developed rapidly, aiming to achieve efficient artificial intelligence computing by simulating biological nervous systems. Among them, optoelectronic synaptic devices based on two-dimensional materials have become a research hotspot due to their high integration, low power consumption, and parallel processing capabilities. The closest prior art is a three-terminal electrical stimulation synaptic device based on two-dimensional tellurene (2D-Te) proposed in the reference (Lee et al., 2023). This device achieved 100 effective multi-level states through electrical signal regulation and demonstrated good linearity. Its core solution is to utilize the high carrier mobility and environmental stability of two-dimensional tellurene to simulate the weight change of biological synapses through electrical pulse stimulation to support neuromorphic computing tasks.
[0003] The prior art has the following problems in optoelectronic synaptic devices:
[0004] 1. Lack of optical modulation ability: This technology only relies on electrical signal regulation and does not integrate optical sensing functions, so it cannot achieve the collaborative processing of optical information perception and storage, which limits its application in bionic vision systems;
[0005] 2. Nonlinear problem: Although the linearity is good, the nonlinearity of its synaptic weight change may lead to error accumulation in neural network training, affecting the calculation accuracy;
[0006] 3. Single signal channel: The device only supports single electrical signal input and cannot utilize multi-wavelength optical signals to process information in parallel, which limits the feature extraction ability and classification performance of neural networks. Summary of the Invention
[0007] The purpose of the present invention is to provide an oxygen plasma-assisted optoelectronic synaptic device and its application in neuromorphic computing to solve the problems mentioned in the above background art.
[0008] To achieve the above purpose, the present invention provides the following technical solution: An oxygen plasma-assisted optoelectronic synaptic device, the optoelectronic synaptic device includes a substrate layer, an optoelectronic sensitive layer, and an electrode layer arranged from bottom to top. The substrate layer includes an Si layer and an SiO2 layer, the SiO2 layer is located above the Si layer, and the optoelectronic sensitive layer includes a two-dimensional tellurene nanosheet layer deposited on the surface of the substrate layer. The surface of the two-dimensional tellurene nanosheet layer is treated with oxygen plasma, and the electrode layer is deposited on the surface of the optoelectronic sensitive layer. The electrode layer includes a first electrode and a second electrode arranged in parallel.
[0009] Preferably, the thickness of the SiO2 layer is 300 nm.
[0010] Preferably, the two-dimensional tellurene nanosheets of the two-dimensional tellurene nanosheet layer are synthesized by a hydrothermal method, and the thickness of the two-dimensional tellurene nanosheets is 1 - 3 nm and the width is 50 - 200 nm.
[0011] Preferably, both the first electrode and the second electrode are bilayer structures, and the bottom layer of the first electrode and the second electrode is a Ni or Ti layer, and the top layer is an Au layer.
[0012] Preferably, the first electrode and the second electrode are deposited on the surface of the photosensitive layer by DC magnetron sputtering.
[0013] Preferably, the oxygen plasma treatment parameters are treatment for 60 s at a power of 60 W.
[0014] According to the above application of an optoelectronic synaptic device assisted by oxygen plasma in neuromorphic computing, based on the above optoelectronic synaptic device, a convolutional neural network architecture of a multi-channel optical synaptic device and its application in an image recognition task.
[0015] Preferably, the application of the convolutional neural network architecture of the multi-channel optical synaptic device in the image recognition task includes the following steps:
[0016] Step 1, Image encoding: Encoding the MNIST handwritten digit image into a grayscale image of 28×28 pixels;
[0017] Step 2, Synaptic weight initialization: Initializing the synaptic weights by optical pulses with wavelengths of 405 nm, 450 nm, and 520 nm;
[0018] Step 3, Convolutional calculation: Using a three-layer convolutional neural network (32, 64, 128 filters) for feature extraction;
[0019] Step 4, Fully connected layer calculation: Classifying through fully connected layers with 256, 512, and 10 neurons respectively;
[0020] Step 5, Result output: Outputting the recognition results of 10 classes of handwritten digits.
[0021] Compared with the prior art, the beneficial effects of the present invention are:
[0022] Under the stimulation of optical pulses with wavelengths of 405 nm, 450 nm, and 520 nm, the structure of the present invention captures photo-generated electrons by oxygen vacancies and delays recombination, realizing the dynamic conversion of short-term plasticity (STP) and long-term plasticity (LTP) under optical stimulation; at the same time, the change of synaptic weights shows superlinearity (linear goodness of fit R 2> 0.99), with a dynamic range (DR) of 1.52 - 1.64, significantly superior to existing two-dimensional material synaptic devices;
[0023] The structure of the present invention improves the recognition accuracy through single-channel (405 nm), dual-channel (405 + 450 nm), and triple-channel (405 + 450 + 520 nm) optical synaptic architectures respectively. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 It is a schematic structural diagram of the optoelectronic synaptic device of the present invention;
[0025] Figure 2 It is the dynamic conversion of short-term plasticity (STP) and long-term plasticity (LTP) of the device under optical stimulation of the present invention, (a) optical wavelength 405 nm; (b) optical wavelength 450 nm; (c) optical wavelength 520 nm;
[0026] Figure 3 It is the high linearity weight regulation of the device under optical stimulation of the present invention, (a) optical wavelength 405 nm; (b) optical wavelength 450 nm; (c) optical wavelength 520 nm;
[0027] Figure 4 It is a flowchart of the method for constructing a convolutional neural network based on optical synapses of the present invention;
[0028] Figure 5 It is the application result of three optical coding methods in the present invention's embodiment for handwritten digit recognition.
[0029] In the figure: 1. Substrate layer; 2. Optoelectronic sensitive layer; 3. Electrode layer. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
[0031] Embodiment 1
[0032] Please refer to Figures 1 - 3, As shown in the figure, a photoelectric synaptic device based on oxygen plasma assistance. The photoelectric synaptic device includes a substrate layer 1, a photosensitive layer 2, and an electrode layer 3 arranged from bottom to top. The substrate layer 1 includes a Si layer and a SiO2 layer, and the SiO2 layer is located above the Si layer. The photosensitive layer 2 includes a two-dimensional tellurene nanosheet layer deposited on the surface of the substrate layer 1. The surface of the two-dimensional tellurene nanosheet layer is treated with oxygen plasma, and the electrode layer 3 is deposited on the surface of the photosensitive layer 2. The electrode layer 3 includes a first electrode and a second electrode arranged in parallel.
[0033] In this embodiment, the thickness of the SiO2 layer is 300 nm. The two-dimensional tellurene nanosheets of the two-dimensional tellurene nanosheet layer are synthesized by a hydrothermal method. The thickness of the two-dimensional tellurene nanosheets is 1 - 3 nm, and the width is 50 - 200 nm. Both the first electrode and the second electrode are bilayer structures. The bottom layer of the first electrode and the second electrode is a Ni or Ti layer, and the top layer is an Au layer. The first electrode and the second electrode are deposited on the surface of the photosensitive layer 2 by DC magnetron sputtering. The parameters of the oxygen plasma treatment are 60 s under a power of 60 W.
[0034] An application of a photoelectric synaptic device based on oxygen plasma assistance in neuromorphic computing, a convolutional neural network architecture of a multi-channel optical synaptic device constructed based on the above photoelectric synaptic device, and its application in image recognition tasks.
[0035] Furthermore, referring to Figure 4 , the application of the convolutional neural network architecture of the multi-channel optical synaptic device in image recognition tasks includes the following steps:
[0036] Step 1, Image encoding: Encode the MNIST handwritten digit image into a grayscale image of 28×28 pixels;
[0037] Step 2, Synaptic weight initialization: Initialize the synaptic weights with light pulses of wavelengths 405 nm, 450 nm, and 520 nm;
[0038] Step 3, Convolution calculation: Use a three-layer convolutional neural network (32, 64, 128 filters) for feature extraction;
[0039] Step 4, Fully connected layer calculation: Classify through fully connected layers with 256, 512, and 10 neurons respectively;
[0040] Step 5, Result output: Output the recognition results of 10 categories of handwritten digits.
[0041] Under the stimulation of light pulses with wavelengths of 405 nm, 450 nm, and 520 nm, the oxygen vacancies in the structure of the present invention capture photo-generated electrons and delay recombination, realizing the dynamic conversion of short-term plasticity (STP) and long-term plasticity (LTP) under light stimulation. Refer to Figure 2 ; Refer toFigure 3 , while the synaptic weight change shows superlinearity (linear goodness of fit R 2 > 0.99), and the dynamic range (DR) reaches 1.52 - 1.64, significantly better than existing two-dimensional material synaptic devices.
[0042] Refer to Figure 5 , the structure of the present invention realizes recognition accuracies of 95%, 97% and 98.5% respectively through single-channel (405nm), dual-channel (405 + 450nm) and triple-channel (405 + 450 + 520nm) optical synaptic architectures.
[0043] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.
[0044] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. An oxygen plasma-assisted optoelectronic synaptic device, characterized in that: The photoelectric synapse device comprises a substrate layer (1), a photoelectric sensitive layer (2) and an electrode layer (3) arranged from bottom to top, wherein the substrate layer (1) comprises a Si layer and a SiO2 layer, wherein the SiO2 layer is located above the Si layer, and the photoelectric sensitive layer (2) comprises a two-dimensional tellurene nanosheet layer deposited on the surface of the substrate layer (1), wherein the surface of the two-dimensional tellurene nanosheet layer is treated with oxygen plasma, and the electrode layer (3) is deposited on the surface of the photoelectric sensitive layer (2), wherein the electrode layer (3) comprises a first electrode and a second electrode arranged in parallel.
2. The oxygen plasma-assisted optoelectronic synaptic device according to claim 1, characterized in that: The thickness of the SiO2 layer is 300nm.
3. The oxygen plasma-assisted optoelectronic synaptic device according to claim 2, characterized in that: The two-dimensional tellurene nanosheets of the two-dimensional tellurene nanosheet layer are synthesized by a hydrothermal method, and the two-dimensional tellurene nanosheets have a thickness of 1-3 nm and a width of 50-200 nm.
4. The oxygen plasma-assisted optoelectronic synaptic device according to claim 3, characterized in that: The first electrode and the second electrode are both double-layer structures, and the bottom layers of the first electrode and the second electrode are Ni or Ti layers, and the top layers are Au layers.
5. The oxygen plasma-assisted optoelectronic synaptic device according to claim 4, characterized in that: The first electrode and the second electrode are deposited on the surface of the photoelectric sensitive layer (2) by direct current magnetron sputtering.
6. The oxygen plasma-assisted optoelectronic synaptic device according to claim 5, characterized in that: The oxygen plasma treatment parameters are 60 W power and 60 seconds.
7. The application of an oxygen plasma-assisted optoelectronic synaptic device in neuromorphic computing according to any one of claims 1 to 6, characterized in that: Based on the above-mentioned optoelectronic synaptic devices, a convolutional neural network architecture of a multi-channel optical synaptic device is constructed and its application in image recognition tasks.
8. The application of an oxygen plasma-assisted optoelectronic synaptic device in neuromorphic computing according to claim 7, characterized in that: The steps include: Step 1: Image encoding: Encode the MNIST handwritten digit image into a 28×28 pixel grayscale image; Step 2: Initialize synaptic weights: Initialize synaptic weights using 405nm, 450nm, and 520nm wavelength light pulses. Step 3, convolution calculation: use a three-layer convolutional neural network (32, 64, and 128 filters) for feature extraction; Step 4: Fully connected layer calculation: Classification is performed through fully connected layers of 256, 512, and 10 neurons respectively; Step 5: Output the results of the recognition of 10 types of handwritten digits.