Photoelectric memristor for wafer detection field and preparation method and application thereof

Abstract

The invention belongs to the field of organic / polymer neuromorphic devices, and particularly provides a photoelectric memristor used in the field of wafer detection, preparation of the photoelectric memristor and application of the photoelectric memristor in the field of wafer detection. The structure comprises the following components from bottom to top: (1) a glass substrate; (2) an indium tin oxide electrode (ITO); (3) preparing an ODAE-COF thin film active layer; and (4) an ITO electrode. The device shows photoelectric response performance, shows an excellent history dependence memristive effect, and promotes biological stimulation similar to a learning / forgetting / re-learning process observed in human memory. The invention develops an innovative integrated edge computing system which comprises a sensing unit and a data processing network for detecting anomalies in a wafer. The system can effectively identify cavity defects on the surfaces of the wafers and classify the wafers according to different types of surface roughness.

Description

Technical Field

[0001] The present invention belongs to the field of organic / polymeric neuromorphic devices, and provides a photoelectric memristor with a covalent organic framework (COF) as an active layer and its application in the field of wafer inspection. Specifically, it is a photoelectric memristor for the field of wafer inspection, its preparation method and application. Background Art

[0002] Wafer inspection is crucial in the field of integrated circuits (ICs) and has a significant impact on chip quality, yield, and cost control. It is a key process to ensure the success of semiconductor manufacturing. Surface defects on wafers, such as roughness, voids, scratches, and particle contamination, may lead to serious defects, including electrical short circuits, open circuits, and degraded device performance. Therefore, evaluating the surface quality of wafers is crucial for ensuring the reliability and functionality of the final products. Researchers have developed various methods to evaluate wafer surfaces, including stylus profilometry, optical profilometry, atomic force microscopy, scanning electron microscopy, capacitance measurement, and laser interferometry. Although these techniques are effective, they are still hindered by factors such as resolution and range limitations, low time efficiency, high error rates, excessive costs, and potential damage to samples. Therefore, there is an increasing need for more advanced, efficient, and reliable evaluation techniques to ensure higher production yields and better chip performance. Given its significant advantages in real-time processing, reduced latency, scalability, and flexibility, edge computing based on photoelectric memristors is considered an alternative for evaluating wafer surface quality.

[0003] Both electrical pulses and optical pulses can induce artificial synaptic behavior. In previous studies, a large number of studies focused on fully electrically tunable artificial synaptic devices. Compared with electrical signals, optical signals have multiple advantages, including minimal crosstalk, high bandwidth, low power consumption, and excellent stability. Recently, photoelectric memristors have attracted much attention as novel multifunctional devices capable of exhibiting electrically and optically tunable synaptic plasticity. These devices are expected to be applied in photoelectric neuromorphic computing systems, which can use optical stimuli to rapidly manipulate memory activities in artificial neural networks with low power consumption. Notably, brain-inspired photoelectric memristors with photoelectric sensing capabilities are particularly good at perceiving and memorizing visual information. They have significant advantages, including wide optical signal bandwidth, high-speed data processing, and low power consumption requirements, making them ideal for visual automation and integrated edge computing applications.

[0004] So far, various optoelectronic memristive materials, including graphene, carbon nanotubes, metal oxides, halide perovskites, and ferroelectric materials, have been explored to enhance the photosensing function of traditional electrical synaptic memristors. However, the potential of highly structured organic materials such as covalent organic frameworks (COFs) in optoelectronic memristors remains underexplored. Due to the tunability of the molecular framework, novel functionalized COFs can respond to external stimuli (such as light, electricity, and heat) by incorporating various functional groups into their conjugated structures, thereby generating controllable physical properties. The inherent high crystallinity, large specific surface area, and high porosity of COFs, combined with their adaptable response behavior, endow them with excellent performance in various functional devices. Among the numerous groups responsive to external stimuli, photochromic molecules with significant color changes have become an important class of light-triggered organic switches due to their photo-isomerization ability. Over the years, various photo-responsive units, including azobenzene, spirooxazine, fulgide, and diarylethene, have been extensively studied. Notably, diarylethene analogs have attracted much attention due to their excellent photo-switching performance, thermal stability, and anti-fatigue properties, becoming the focus of multiple research fields. Currently, it has been reported that diarylethene units are incorporated into the COF framework as photo-responsive groups. However, the application of these COFs in the development of optoelectronic memristive devices has not been deeply studied. Summary of the Invention

[0005] An object of the present invention is to propose an optoelectronic memristor with a COF thin film as the active layer.

[0006] A second object of the present invention is to provide a synthetic route for the COF used as the active layer of the optoelectronic memristor.

[0007] A third object of the present invention is to provide a preparation method for an optoelectronic memristor with a COF thin film material as the active layer.

[0008] A fourth object of the present invention is to provide an application of the COF optoelectronic memristor in the field of wafer detection.

[0009] Technical solution of the present invention:

[0010] An optoelectronic memristor with a COF thin film material as the active layer, that is, an optoelectronic memristor for the field of wafer detection, the device structure from bottom to top is:

[0011] 1. Glass substrate

[0012] 2. Indium tin oxide electrode (ITO)

[0013] 3. ODAE-COF thin film active layer

[0014] 4. ITO electrode

[0015] The present invention also provides a method for preparing the optoelectronic memristor as described above, comprising the following steps:

[0016] 1) Clean the glass substrate coated with ITO and dry it.

[0017] 2) Deposit the ODAE-COF thin film on the pre-cleaned ITO plate.

[0018] 3) Vacuum dry the obtained device overnight to remove the excess solution.

[0019] 4) Finally, deposit the ITO top electrode on the active layer to obtain the device ITO / ODAE-COF / ITO.

[0020] Further preferably, the specific steps of the preparation method include:

[0021] 1) Wash the glass substrate coated with ITO (preferably 1.5 cm × 1.5 cm) with dishwashing liquid and deionized water, and then wash it in ethanol, acetone and isopropanol respectively (preferably for 15 minutes) and dry it.

[0022] 2) Cover a COF layer with a thickness of 50 nm - 100 nm on the ITO glass, and vacuum dry it overnight to remove the excess solution.

[0023] 3) Use the magnetron sputtering method to deposit the ITO top electrode (preferably with a thickness of 100 - 150 nm and an area of about 0.4 × 0.4 mm2) on the active layer through a mask template, and finally obtain the device ITO / ODAE-COF / ITO.

[0024] The ODAE-COF structure in the active layer thin film provided by the present invention is shown by the following formula:

[0025]

[0026] The thin film ODAE-COF is prepared by the following steps:

[0027] Synthesis of 1,5-bis-3-(2-chloro-5-dimethylthiophenyl)-1,5-pentanedione (1):

[0028] Under an Ar atmosphere, add anhydrous AlCl3 and anhydrous CH2Cl2 to the reaction flask, and slowly stir at 0 °C. Then slowly add 2-chloro-5-methylthiophene and glutaroyl chloride, and react at room temperature for 4 h; then pour the mixture into ice water, add hydrochloric acid until no gas is generated; after post-treatment, a white solid is obtained.

[0029] Synthesis of 1,2-bis(5-chloro-2-methylthiophen-3-yl)cyclopent-1-ene (2)

[0030] Under an Ar atmosphere, Zn and anhydrous THF were added to a reaction flask, and the reaction was carried out at -5 °C; stirred; then TiCl4 was added to the reaction flask, and the mixture was refluxed for 3 h; after cooling to 0 °C, anhydrous THF was slowly added to dissolve 1; then the liquid was reheated to reflux; then a saturated K2CO3 solution was added, filtered and collected; after post-treatment, a colorless crystal was obtained.

[0031] Synthesis of 1,2-bis(5-formyl-2-methylthiophen-3-yl)cyclopentene (ODAE)

[0032] Under argon protection, 2 and anhydrous THF were added to a reaction flask, stirred and cooled to -78 °C, then n-BuLi was slowly added. During this process, the mixture turned dark pink. After addition, the reaction was stirred; subsequently, tributyl borate was added, and the mixture was stirred at room temperature; the resulting orange mixture was used directly as a raw material without purification; meanwhile, under argon protection, Pd(PPh)4 and THF were added to another reaction flask; after stirring and refluxing for 30 min, a 2.5 M aqueous K2CO3 solution, ethylene glycol, 4-bromobenzaldehyde and the above orange mixture were added to the reaction flask, and the mixture was refluxed; after cooling, post-treatment gave a yellow solid.

[0033] Synthesis of ODAE-COF: ODAE was dissolved in CH2Cl2 as the organic phase; subsequently, pure water was slowly added to the top layer of the CH2Cl2 solution to ensure the stability of the CH2Cl2 / H2O interface; TAPA and TsOH·H2O solution were added to CH3CN and H2O and slowly dropped; after reacting at room temperature for 72 h, a transparent and smooth film was obtained at the interface of the two immiscible phases.

[0034] The preparation method of the ODAE-COF active layer described in the present invention is specifically as follows:

[0035] Synthesis of 1,5-bis-3-(2-chloro-5-dimethylsulfanylphenyl)-1,5-pentanedione (1): Under an Ar atmosphere, anhydrous AlCl3 (16 g, 120.00 mmol) and anhydrous CH2Cl2 (100 mL) were added to a reaction flask, and the mixture was slowly stirred at 0 °C. Then 2-chloro-5-methylthiophene (8.10 mL, 75.56 mmol) and glutaroyl chloride (6 g, 35.50 mmol) were slowly added, and the reaction was carried out at room temperature for 4 h. Then the mixture was poured into ice water, and hydrochloric acid was added until no gas was generated. The mixture was extracted three times with CH2Cl2, the organic phases were combined, and washed with brine. Then it was dried over anhydrous Na2SO4 and filtered. The filtrate was concentrated by rotary evaporation and purified by column chromatography (silica gel; petroleum ether:CH2Cl2 = 3:1), and finally a white solid (6.91 g, 54%) was obtained.

[0036] Synthesis of 1,2-bis(5-chloro-2-methylthiophen-3-yl)cyclopent-1-ene (2): Under an Ar atmosphere, Zn (20 g) and anhydrous THF (80 mL) were added to the reaction flask and reacted slowly at -5 °C. Stir slowly. Then

[0037] TiCl4 (19.43 mL, 128.71 mmol) was added to the reaction flask and refluxed at 65 °C for 3 h. After cooling to 0 °C, 30 mL of anhydrous THF was slowly added to dissolve 1 (7 g, 19.37 mmol). Then the liquid was reheated to 65 °C and refluxed for 12 h. Then saturated K2CO3 solution was added, filtered and collected. The obtained filtrate was extracted three times with CH2Cl2, the organic phases were combined, washed with brine, then dried over anhydrous Na2SO4 and filtered. The filtrate was concentrated by rotary evaporation and purified by column chromatography (silica gel; petroleum ether) to obtain a colorless crystal (5.35 g, 84%).

[0038] Synthesis of 1,2-bis(5-formyl-2-methylthiophen-3-yl)cyclopentene (ODAE): Under argon protection, 2 (5.50 g, 16.76 mmol) and anhydrous THF (30 mL) were added to the reaction flask. After stirring and cooling to -78 °C, n-BuLi (16.72 mL, 2.5 M) was slowly added. During this process, the mixture turned dark pink. After addition, the reaction was stirred for 1 h. Subsequently, tributyl borate (13.56 mL, 50.17 mmol) was added and stirred at room temperature for 2 h. The obtained orange mixture was used directly as the raw material without purification. Meanwhile, under argon protection, Pd(PPh)4 (500 mg, 0.43 mmol) and THF (25 mL) were added to another reaction flask. After stirring and refluxing at 70 °C for 30 min, 2.5 M aqueous K2CO3 solution (50 mL), ethylene glycol (10 drops), 4-bromobenzaldehyde (7.4 g, 40.00 mmol) and the above orange mixture were added to the reaction flask and refluxed at 70 °C for 12 h. After cooling, deionized water (25 mL) was added and extracted three times with CH2Cl2. The organic phases were combined, washed with brine, dried over anhydrous Na2SO4, and filtered. The filtrate was evaporated to dryness and purified by column chromatography (silica gel; CH2Cl2) to obtain a yellow solid (5.22 g, 66%). 1 1H NMR (CDCl3, 400 MHz): δ / ppm = 9.96 (s, 2H); 7.83 (d, 4H); 7.63 (d, 4H); 7.19 (s, 2H); 2.86 (t, 4H); 2.15 - 2.08 (m, 2H); 2.04 (s, 6H).

[0039] Synthesis of ODAE-COF: ODAE was dissolved in CH2Cl2 as the organic phase. Subsequently, pure water was slowly added to the top layer of the CH2Cl2 solution to ensure the stability of the CH2Cl2 / H2O interface. The TAPA and TsOH·H2O solutions were added dropwise slowly to CH3CN and H2O. The beaker was covered to maintain stability and prevent air flow. After reacting for 72 hours at room temperature, a transparent and smooth film was clearly observed at the interface of the two immiscible phases.

[0040] The specific reaction route is as follows:

[0041]

[0042] Fabrication of ITO / ODAE-COF / ITO device:

[0043] A commercially available indium tin oxide (ITO) coated glass substrate (size 1 cm × 1 cm) was carefully cleaned in an ultrasonic bath with detergent, deionized water, ethanol, acetone, and 2-propanol, with each cleaning time being 15 minutes in sequence. To transfer the ODAE-COF film onto the ITO / glass substrate, a liquid-liquid interface reaction system was constructed. Previously, the ITO / glass substrate was placed at the bottom of the beaker. After the appearance of the ODAE-COF film, the organic phase and the ODAE-COF film completely covered the surface of the ITO / glass substrate in the aqueous phase. The ITO top electrode (thickness 120 nm, area 0.4 × 0.4 mm 2 ) was thermally deposited onto the surface through a shadow mask at 10 -7 Torr. All optoelectronic property tests of the fabricated devices in this work were carried out on a Keithley 4200 semiconductor parameter analyzer equipped with a pulse measurement device under atmospheric conditions without further protection

[0044] The present invention also provides an application of the optoelectronic memristor with COF as the active layer in the field of wafer detection.

[0045] Furthermore, under ultraviolet and blue light illumination, the device exhibits significantly different resistance states, which can be defined as on and off states and used to store data as "0" and "1" in binary, and the data can still be preserved after power-off.

[0046] The present invention provides a photoelectric memristor with a covalent organic framework (COF) as the active layer and its application in the field of wafer detection. The fabricated device exhibits photoelectric response performance, which is due to the reversible conversion of diarylethene between its closed-loop and open-loop forms under ultraviolet and visible light irradiation. In addition, benefiting from the redox characteristics of the triphenylamine unit, the device exhibits excellent history-dependent memristive effect, promoting bio-stimulation similar to the "learning / forgetting / re-learning" process observed in human memory. Based on the unique memristive performance, we have developed a novel integrated edge computing system that combines a sensing unit and a data processing model. The system has two different data processing models, which are respectively used for void defect detection and surface roughness classification. The first model integrates an image enhancement network that utilizes the light responsiveness of the photoelectric memristor for denoising and feature extraction, followed by a neural network based on the COF memristor, which consists of input, encoder, decoder, and output modules for precise defect localization. Tests on 400 X-ray images show that this model can effectively detect void defects. The model exhibits strong convergence and enhanced robustness against noise and complex patterns, thus enabling reliable and stable defect detection. The second model consists of an image attention module, a feature convolution module, and a probability mapping module, all constructed from 64 conductivity states. This model performs excellently in surface roughness classification, with an average accuracy of over 90%, thus confirming its effectiveness for precise and consistent classification.

[0047] Advantageous technical effects of the present invention:

[0048] The present invention proposes to use COF materials as the active layer of neuromorphic devices, and designs and synthesizes a photoelectric dual-responsive COF. The present invention selects ODAE and TAPA as reaction monomers, and synthesizes a two-dimensional conjugated polymer film ODAE-COF with light response characteristics through Schiff base reaction.

[0049] The present invention has developed an innovative integrated edge computing system, which includes a sensing unit and a data processing network for detecting anomalies in wafers. The system can effectively identify void defects on the wafer surface and classify wafers according to different types of surface roughness. Description of the Drawings

[0050] Figure 1 (a) is the X-ray image of the chip. Figure 1 (b) is the image enhancement network based on the COF memristor. Figure 1 (c) is the filtered X-ray image of the chip. Figure 1 (d) is the deep neural network based on the COF memristor. Figure 1 (e) is the reconstructed void image.

[0051] Figure 2It is the change of F1 score with the number of conductance states.

[0052] Figure 3 (a) is the AFM image of the chip. Figure 3 (b) is the deep learning model based on the COF memristor. Figure 3 (c) are four typical AFM images with different roughness grades. Figure 3 (d) is the confusion matrix of the deep learning model.

[0053] Figure 4 It is the evolution of prediction accuracy with the number of conductance states.

[0054] Figure 5 (a) are the current-voltage characteristics of the ITO / ODAE-COF / ITO device. Figure 5 (b) is the 64-state modulation of the device conductance. Figure 5 (c) is the retention performance of the 64-state conductance. Figure 5 (d) are the conductance-voltage characteristics of 64 states recorded between 0V and 0.5V.

[0055] Figure 6 (a) are the current-voltage characteristics of the ITO / ODAE-COF / ITO device. Figure 6 (b) is the 64-state modulation of the device conductance. Figure 6 (c) is the retention performance of the 64-state conductance. Figure 6 (d) are the conductance-voltage characteristics of 64 states recorded between 0V and 0.5V.

[0056] Figure 7 It is the device diagram of the optoelectronic memristor and its application in wafer detection. Specific implementation manners

[0057] Combined with the following drawings, the embodiments of the present invention are described in detail so that the features and advantages of the present invention can be more obvious and understandable.

[0058] Preparation Example 1: Preparation method of the ODAE-COF active layer:

[0059] Synthesis of 1,5-bis-3-(2-chloro-5-dimethylsulfanyl phenyl)-1,5-pentanedione (1): Under an Ar atmosphere, anhydrous AlCl3 (16 g, 120.00 mmol) and anhydrous CH2Cl2 (100 mL) were added to the reaction flask and stirred slowly at 0 °C. Then, 2-chloro-5-methylthiophene (8.10 mL, 75.56 mmol) and glutaroyl chloride (6 g, 35.50 mmol) were added slowly, and the reaction was carried out at room temperature for 4 h. Then, the mixture was poured into ice water, and hydrochloric acid was added until no gas was generated. The mixture was extracted three times with CH2Cl2, the organic phases were combined, washed with brine, dried over anhydrous Na2SO4, and filtered. The filtrate was concentrated by rotary evaporation and purified by column chromatography (silica gel; petroleum ether:CH2Cl2 = 3:1) to obtain a white solid (6.91 g, 54%). 1 H NMR (CDCl3, 400 MHz): δ / ppm = 7.19 (s, 2H); 2.87 (m, 4H); 2.66 (s, 6H); 2.10 - 2.03 (m, 2H).

[0060] Synthesis of 1,2-bis(5-chloro-2-methylthiophen-3-yl)cyclopent-1-ene (2): Under an Ar atmosphere, Zn (20 g) and anhydrous THF (80 mL) were added to the reaction flask and reacted slowly at -5 °C with slow stirring. Then,

[0061] TiCl4 (19.43 mL, 128.71 mmol) was added to the reaction flask and refluxed at 65 °C for 3 h. After cooling to 0 °C, 30 mL of anhydrous THF was slowly added to dissolve 1 (7 g, 19.37 mmol). Then, the liquid was reheated to 65 °C and refluxed for 12 h. Then, saturated K2CO3 solution was added, filtered, and collected. The resulting filtrate was extracted three times with CH2Cl2, the organic phases were combined, washed with brine, dried over anhydrous Na2SO4, and filtered. The filtrate was concentrated by rotary evaporation and purified by column chromatography (silica gel; petroleum ether) to obtain colorless crystals (5.35 g, 84%). 1 H NMR (CDCl3, 400 MHz): δ / ppm = 6.58 (s, 1H); 2.73 (m, 2H); 2.08 - 1.99 (m, 1H); 1.88 (s, 3H).

[0062] Synthesis of 1,2-bis(5-formyl-2-methylthiophen-3-yl) cyclopentene (ODAE): Under argon protection, 2 (5.50 g, 16.76 mmol) and anhydrous THF (30 mL) were added to a reaction flask. After stirring and cooling to -78 °C, n-BuLi (16.72 mL, 2.5 M) was slowly added. During this process, the mixture turned dark pink. After addition, the reaction was stirred for 1 h. Subsequently, tributyl borate (13.56 mL, 50.17 mmol) was added and stirred at room temperature for 2 h. The resulting orange mixture was used directly as a raw material without purification. Meanwhile, under argon protection, Pd(PPh)4 (500 mg, 0.43 mmol) and THF (25 mL) were added to another reaction flask. After stirring and refluxing at 70 °C for 30 min, an aqueous solution of 2.5 M K2CO3 (50 mL), ethylene glycol (10 drops), 4-bromobenzaldehyde (7.4 g, 40.00 mmol), and the above orange mixture were added to the reaction flask and refluxed at 70 °C for 12 h. After cooling, deionized water (25 mL) was added, and the mixture was extracted three times with CH2Cl2. The combined organic phases were washed with brine, dried over anhydrous Na2SO4, and filtered. The filtrate was concentrated by rotary evaporation and purified by column chromatography (silica gel; CH2Cl2) to obtain a yellow solid (5.22 g, 66%). 1 1H NMR (CDCl3, 400 MHz): δ / ppm = 9.96 (s, 2H); 7.83 (d, 4H); 7.63 (d, 4H); 7.19 (s, 2H); 2.86 (t, 4H); 2.15 - 2.08 (m, 2H); 2.04 (s, 6H).

[0063] Synthesis of ODAE-COF: ODAE was dissolved in CH2Cl2 as the organic phase. Subsequently, pure water was slowly added to the top layer of the CH2Cl2 solution to ensure the stability of the CH2Cl2 / H2O interface. The TAPA and TsOH·H2O solutions were slowly added dropwise to CH3CN and H2O. The beaker was covered to maintain stability and prevent air flow. After reacting at room temperature for 72 h, a transparent and smooth film was clearly observed at the interface of the two immiscible phases.

[0064] The specific reaction route is as follows:

[0065]

[0066] Preparation method of a diarylethene-based covalent organic framework optoelectronic memristor:

[0067] 1) Wash the glass substrate (1.5 cm × 1.5 cm) coated with ITO with dishwashing liquid and deionized water, and then wash it in ethanol, acetone, and isopropanol for 15 min each and dry.

[0068] 2) A 50-nm-thick COF thin film was deposited on the ITO glass by a deposition method, and the excess solution was removed by vacuum drying overnight.

[0069] 3) The ITO top electrode (thickness = 100 nm) with an area of about 0.4 × 0.4 mm was deposited by magnetron sputtering. 2 .

[0070] 4) It was deposited on the active layer through a mask, and finally the device ITO / TAPB-DCA-2DP / ITO was obtained.

[0071] The commercially available indium tin oxide (ITO)-coated glass substrate was carefully cleaned in an ultrasonic bath with a cleaner, deionized water, ethanol, acetone, and 2-propanol, with each cleaning time being 15 minutes in turn. To transfer the ODAE-COF thin film onto the ITO / glass substrate, a liquid-liquid interface reaction system was constructed. Previously, the ITO / glass substrate was placed at the bottom of the beaker. After the ODAE-COF thin film appeared, the organic phase and the ODAE-COF thin film completely covered the surface of the ITO / glass substrate in the aqueous phase. The ITO top electrode (with a thickness of 120 nm and an area of 0.4 × 0.4 mm) was thermally deposited onto the surface through a shadow mask at 10 2 Torr. All optoelectronic property tests of the fabricated devices in this work were carried out on a Keithley 4200 semiconductor parameter analyzer equipped with a pulse measurement device under atmospheric conditions without further protection. -7

[0072] Effect Example 1

[0073] As Figure 1 , to evaluate the potential of the ITO / ODAE-COF / ITO optoelectronic memristor with 64 conductive states in evaluating the surface quality of wafers, an edge computing system was constructed to detect void defects on the wafer surface by analyzing X-ray images. As Figure 1 shown in a, the X-ray image was obtained using a Y.Cheetah X-ray detection system. The edge computing system includes an image enhancement network and a deep neural network, both constructed from the fabricated memristors ( Figure 1 b). Before inputting the X-ray image into the deep neural network, the image must be denoised. The fabricated optoelectronic memristor shows an enhancement effect under ultraviolet irradiation and an inhibition effect under visible light, fully meeting the requirements of the denoising process. In the image enhancement network, the captured image is first segmented row by row. Each row undergoes preprocessing, including normalization and image filtering operations, to generate a preprocessed and denoised image ( Figure 1c). We utilize optoelectronic memristors to facilitate this preprocessing and design a memristor network to perform normalization and Gaussian filtering. This process enhances image features and achieves effective denoising. The deep neural network based on COF memristors consists of four modules: an input module, an encoder module, a decoder module, and an output module ( Figure 1 d and 1e). The input module prepares the data for further processing by extracting initial features from the X-ray images. The encoder module captures and encodes important features in a compact representation by learning global dependencies and local features. The decoder module refines the encoded features using an attention mechanism while preserving the order of the data for accurate defect detection. Finally, the output module converts the processed feature vectors into a 2D image format to generate the visualization of the final defect prediction.

[0074] Effect Example 2

[0075] As Figure 2 , we use 400 X-ray images to evaluate the effectiveness of the proposed void detection method. First, the original X-ray images are denoised using the above image enhancement network to obtain a set of processed images. Subsequently, these denoised images are analyzed using a deep neural network to obtain the final reconstructed images, in which void defects are successfully identified. We consider the influence of the number of conductive states on the prediction accuracy by training and validating with different numbers of conductive states (2, 4, 8, 16, 32, and 64) while keeping all other parameters unchanged. As the number of conductive states increases, the F1 score of the model improves significantly, reaching the highest accuracy at 64 conductive states. More conductive states enable the model to capture more detailed information in the data, enhancing its robustness to noise and complex patterns, thus making the performance of overall void defect detection more stable and reliable.

[0076] Effect Example 3

[0077] As Figure 3 , we utilize a COF-memristor-based edge computing system to measure the surface roughness of the wafer by analyzing AFM images ( Figure 3a). We constructed a database containing 200 AFM images to train and validate the memristor-based edge computing system. In this database, the roughness grade quantifies the wafer flatness: grade A indicates a roughness below 0.5 nanometers, grade B corresponds to 0.5 nanometers to 2 nanometers, grade C ranges from 2 nanometers to 5 nanometers, and grade D applies to roughness greater than 5 nanometers. Each grade contains 50 images, resulting in a balanced dataset. Next, we constructed a deep learning model using COF memristors, aiming to classify the surface roughness of wafers at different grades. This deep learning model based on COF memristors mainly includes three modules: an image attention module, a feature convolution module, and a probability mapping module. At the input layer, the two-dimensional image is flattened into a one-dimensional vector along its width and then processed by the image attention module. The position information is integrated through the embedding layer, and the multi-head dot product attention mechanism generates the output, which is then mapped through a linear layer. The feature convolution module consists of a 1D convolutional layer, batch normalization, and a Leaky ReLU activation layer to alleviate the gradient explosion, and then a pooling layer is used for downsampling. The flattened features are then passed to the probability mapping module, which consists of three fully connected layers. The final softmax layer outputs the probability distribution of various surface roughness categories. This study constructed a deep neural network using memristors with 64 conductance states and verified the model on the test set, and the results are presented in the form of a confusion matrix, as Figure 3 shown in b. The confusion matrix indicates that the proposed deep learning model based on COF memristors performs well in all categories, and the correct classification rates for categories A, B, C, and D are 0.94, 0.98, 0.97, and 0.96, respectively ( Figure 3 c and 3d). Overall, the average accuracy of the model in all categories exceeds 90%.

[0078] Effect Example 4

[0079] As Figure 4 , the convergence curve shows that the loss function gradually converges during the training process, while the accuracy steadily improves. At the beginning of training, the loss drops rapidly and then gradually levels off, finally reaching a high accuracy level, indicating that the model has converged sufficiently to complete this task. To further study the influence of the number of conductivity states on the prediction accuracy, we trained and validated different numbers of conductivity states (2, 4, 8, 16, 32, and 64) while keeping the same parameters. As shown in the figure, as the number of conductivity input states increases, the prediction performance of the model improves, and the accuracy rises from approximately 50% to over 90%, reaching the highest accuracy at 64 conductivity states.

[0080] Effect Example 5

[0081] As Figure 5, in neuromorphic computing, memristors with multiple conductive states can improve the efficiency of neural system training because multiple states are generally associated with excellent learning ability. Our experiments show that during continuous voltage scanning with an increasing positive voltage on the device, when the voltage increases from 0 V to 0.5 V, the current increases from 8.4×10 -5 A to 9.1×10 -4 ( Figure 5 a). In this voltage range, applying 32 pulses enables the memristor to exhibit 32 stable conductive states, thus facilitating effective synaptic weight updates and multifunctional modulation of the device conductance ( Figure 5 b). These conductance states remain relatively stable within 10,000 seconds at room temperature ( Figure 5 c). In addition, the current-voltage characteristics of these 32 conductive states show a linear relationship in the low voltage range from 0 to 0.25 V, indicating that the corresponding synaptic weights can be kept consistent ( Figure 5 d).

[0082] Effect Example 6

[0083] As Figure 6 , similarly, during the negative voltage scan from 0 V to -0.5 V, the device successfully modulated 32 states in a similar manner ( Figure 6 a, Figure 6 b, Figure 6 c and Figure 6 d).

[0084] The above embodiments are used to explain the present invention rather than limit the present invention. Modifications and changes made to the present invention within the spirit and scope of the claims of the present invention all fall within the protection scope of the present invention.

Claims

1. An optoelectronic memristor for the field of wafer detection, characterized in that: Its structural composition from bottom to top is as follows: (1) Glass substrate (2) Indium tin oxide electrode (ITO); (3) ODAE-COF thin film active layer; (4) ITO electrode.

2. The optoelectronic memristor for the field of wafer detection according to claim 1, characterized in that: The ODAE-COF thin film in the active layer has the following structure: ODAE-COF.

3. The optoelectronic memristor for the field of wafer detection according to claim 2, characterized in that: The thin film ODAE-COF is prepared by the following steps: Synthesis of 1,5-bis-3-(2-chloro-5-dimethylthiophenyl)-1,5-pentanedione (1): Under Ar atmosphere, anhydrous AlCl3 and anhydrous CH2Cl2 were added to the reaction flask and slowly stirred at 0 °C. Then 2-chloro-5-methylthiophene and glutaroyl chloride were slowly added and reacted at room temperature for 4 h; then the mixture was poured into ice water and hydrochloric acid was added until no gas was generated; after post-treatment, a white solid was obtained; Synthesis of 1,2-bis(5-chloro-2-methylthiophen-3-yl)cyclopent-1-ene (2) Under Ar atmosphere, Zn and anhydrous THF were added to the reaction flask and reacted at -5 °C; stirred; then TiCl4 was added to the reaction flask and refluxed for 3 h; after cooling to 0 °C, anhydrous THF was slowly added to dissolve 1; then the liquid was reheated to reflux; then saturated K2CO3 solution was added, filtered and collected; after post-treatment, colorless crystals were obtained; Synthesis of 1,2-bis(5-formyl-2-methylthiophen-3-yl)cyclopentene (ODAE) Under argon protection, 2 and anhydrous THF were added to the reaction flask, stirred and cooled to -78 °C, then n-BuLi was slowly added. During this process, the mixture turned dark pink. After addition, it was stirred and reacted; then tributyl borate was added and stirred at room temperature; the resulting orange mixture was used directly as a raw material without purification; at the same time, under argon protection, Pd(PPh)4 and THF were added to another reaction flask; after stirring and refluxing for 30 min, 2.5 M aqueous K2CO3 solution, ethylene glycol, 4-bromobenzaldehyde and the above orange mixture were added to the reaction flask and refluxed; after cooling, post-treatment gave a yellow solid; Synthesis of ODAE-COF: ODAE was dissolved in CH2Cl2 as the organic phase; subsequently, pure water was slowly added to the top layer of the CH2Cl2 solution to ensure the stability of the CH2Cl2 / H2O interface; TAPA and TsOH·H2O solution were slowly added dropwise to CH3CN and H2O; after reacting at room temperature for 72 h, a transparent and smooth film was obtained at the interface of the two immiscible phases.

4. The optoelectronic memristor for the field of wafer detection according to claim 1, characterized in that: The thickness of the COF layer is 50 nm - 100 nm.

5. A method for preparing an optoelectronic memristor for the field of wafer detection according to any one of claims 1-4, characterized in that, Including the following steps: 1) Clean the ITO-coated glass substrate and dry it; 2) Deposit the ODAE-COF thin film on the pre-cleaned ITO plate; 3) Vacuum dry the obtained device overnight to remove the excess solution; 4) Finally, deposit the ITO top electrode on the active layer to obtain the device ITO / ODAE-COF / ITO.

6. The manufacturing method of the optoelectronic memristor for wafer detection field according to claim 5, characterized in that, Specifically, it includes the following steps: 1) Wash the ITO-coated glass substrate of 1 cm × 1 cm with dishwashing liquid and deionized water, and then wash it in ethanol, acetone and isopropanol for 15 minutes respectively and dry it; 2) A COF layer with a thickness of 50 nm - 100 nm is coated on the ITO glass, and the excess solution is removed by vacuum drying overnight. 3) Using the magnetron sputtering method, deposit an ITO top electrode with a thickness of 100 - 150 nm and an area of approximately 0.4×0.4 mm 2 onto the active layer through a mask, and finally obtain the device ITO / ODAE-COF / ITO.

7. Use of the optoelectronic memristor according to any one of claims 1 - 4 in the field of wafer detection for wafer detection.

Images