Cleaning optimization device and method for power semiconductor discrete device furnace tube cleaning machine

Abstract

The invention provides a cleaning optimization device of a power semiconductor discrete device furnace tube cleaning machine, which comprises at least one liquid storage device, the liquid storage device is connected with a cleaning device through a first pump and a pipeline assembly in sequence, the cleaning device is cleaning tanks with different volumes of the furnace tube cleaning machine, and an ultrasonic generator is arranged in each cleaning tank. A control unit is arranged at an inlet of the cleaning tank and comprises an edge calculation module, the edge calculation module is connected with a visual sensor and an infrared spectrum sensor, material and size parameters are arranged in the edge calculation module, and the visual sensor and the infrared spectrum sensor are located at a feeding port of the cleaning equipment; the control unit is connected with the first pump and the ultrasonic generator. According to the method, the visual sensor and the infrared spectrum sensor are adopted, a material-size database is constructed, cleaning parameters are matched in a self-adaptive mode, efficiency is improved for enterprises, and cost and expenditure are reduced, energy is saved, and consumption is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power semiconductor wet cleaning, and in particular to a cleaning optimization device and method for a power semiconductor discrete device furnace tube cleaning machine based on multi-stage dynamic regulation. Background Art

[0002] Wet cleaning of fixtures for power semiconductor discrete devices is widely used in semiconductor manufacturing, such as fixture cleaning cabinets, flushing cabinets, and furnace tube cleaning machines. However, with the continuous advancement of semiconductor technology, the requirements for cleaning processes are becoming increasingly stringent. The accelerated development of new products, new processes, and new technologies has made manual dedicated cleaning cabinets no longer able to meet these requirements.

[0003] With the intensification of market competition in recent years, chip sales have placed higher demands on product yield, reliability, and surface finish. Chip factories are also in urgent need of particle removal equipment for fixture and furnace cleaning used in semiconductor chip manufacturing. However, the fixture and furnace cleaning equipment currently used by most chip factories has various shortcomings.

[0004] Commonly used fixture and furnace tube cleaning devices suitable for semiconductor chip manufacturing have the following technical defects: 1) Poor cleaning uniformity: Traditional spray systems have blind spots, resulting in residual particles or chemical films on the device surface; 2) Solidified process parameters: It is impossible to adjust the cleaning agent ratio and temperature in real time according to the type of pollutants (such as metal ions, particles); 3) Excessive energy consumption: The single high-pressure flushing mode leads to excessive consumption of deionized water and chemical reagents; 4) Risk of secondary pollution: Incomplete waste liquid treatment can easily cause corrosion of the internal pipes of the equipment; 5) Low particle requirements after cleaning: Traditionally, there is basically no particle requirement, which can easily cause unqualified cleaning of fixtures and furnace tubes, resulting in production capacity congestion; 6) Lack of material adaptability: Traditional devices cannot dynamically adjust the cleaning agent ratio and process parameters according to the material (such as quartz, silicon carbide, quartz + deposition products, TFL) and size differences of the fixture, resulting in low cleaning efficiency or waste of chemical reagents.

[0005] Therefore, how to design a cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine that can adaptively match cleaning parameters of materials, and how to design a cleaning optimization method for a power semiconductor discrete device furnace tube cleaning machine that can intelligently identify material material and size have become issues that need to be urgently addressed. Summary of the Invention

[0006] In response to the problems existing in the prior art, the present invention provides a cleaning optimization device and method for a power semiconductor discrete device furnace tube cleaning machine to solve at least one of the above technical problems.

[0007] The technical solution of the present invention is: a cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine, comprising at least one liquid storage device, which is connected to the cleaning equipment in sequence through a first pump and a pipeline assembly. The cleaning equipment is a cleaning tank of different sizes for the furnace tube cleaning machine, an ultrasonic generator is provided in the cleaning tank, and a control unit is provided at the entrance of the cleaning tank. The control unit includes an edge computing module, which is respectively connected to a visual sensor and an infrared spectrum sensor. The edge computing module has built-in material and size parameters, and the visual sensor and the infrared spectrum sensor are located at the feed port of the cleaning equipment; the control unit is connected to the first pump and the ultrasonic generator.

[0008] The present invention adopts a multimodal sensing and dynamic proportioning system, and uses visual sensors and infrared spectrum sensors to build a material-size database. By supporting classification and identification of quartz / silicon carbide / quartz + deposition products (TOES, POLY), it realizes intelligent identification of material materials and sizes, and adaptively matches cleaning parameters. It can effectively control and clean the equipment and production spare parts involved in the chip manufacturing production process. The multi-level cleaning space, recycled waste liquid recovery design, and multi-layer filtration reduce the presence of various types of particles, increase efficiency for the enterprise, and achieve cost and expenditure reduction, energy saving and consumption reduction; it solves the technical defects of the existing technology such as poor cleaning uniformity, solidified process parameters, excessive energy consumption, secondary pollution risk, low requirements for particles after cleaning, and lack of material adaptability.

[0009] The technical solution of the present invention is: a cleaning optimization method for a power semiconductor discrete device furnace tube cleaning machine, comprising the following operating steps:

[0010] S1. Material preparation,

[0011] The material to be cleaned is placed on a conveyor belt at the entrance of the cleaning tank;

[0012] S2. Image acquisition,

[0013] ①The visual sensor collects material surface image data in real time and transmits it to the control unit;

[0014] ②The infrared spectrum sensor collects the reflected spectrum data in real time and transmits it to the control unit;

[0015] S3. Contour extraction,

[0016] The control unit has a built-in FPGA chip that uses a convolutional neural network algorithm to extract the contours of the material surface image data and obtain the three-dimensional dimensions of the material;

[0017] S4. Material classification,

[0018] The FPGA chip uses a convolutional neural network algorithm to analyze the reflectance spectrum data and obtain the material type;

[0019] S5. Ratio decision,

[0020] ①Calculate the preset ratio based on the material type and the chemical reagents stored in the liquid storage tank;

[0021] ② Select the corresponding cleaning tank according to the size data, and preset the ultrasonic frequency, micro jet pressure, and cleaning time;

[0022] S6. Parameters are sent.

[0023] ①The control unit transfers the material to be cleaned to the corresponding cleaning tank;

[0024] ② The control unit controls the opening of the first solenoid valve and the first pump corresponding to the liquid storage tank, and the opening of the second solenoid valve corresponding to the cleaning tank, and outputs the chemical reagent to the corresponding cleaning tank according to the preset ratio;

[0025] ③ According to the preset data, dynamically adjust the ultrasonic frequency. The ultrasonic frequency adjustment range is ±5kHz, the microjet pressure is 0.5~3.0MPa, and the cleaning time is ±10% of the baseline value.

[0026] The present invention integrates multimodal sensing technology, realizes intelligent identification of material size and material through visual and infrared spectrum sensors, and adaptively matches cleaning parameters. It integrates an ultrasonic cavitation cleaning layer (adjustable from 20-80kHz) and a rotary micro-jet spray layer, and realizes the synchronous cleaning of microscopic pores and macroscopic surfaces through the coordinated action of the bottom matrix array and the top multi-angle nozzle group. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is the front view of the present invention.

[0028] Figure 2 It is a schematic structural diagram of the liquid storage tank of the present invention.

[0029] Figure 3 This is a schematic diagram of the installation structure of the detection unit assembly of the present invention.

[0030] Figure 4 It is a process flow chart of the present invention.

[0031] In the figure: 1. Cleaning equipment; 2. Liquid storage equipment; 3. Pipeline assembly; 4. First pump; 5. Circulation system; 7. Detection unit assembly; 201. Liquid level sensor; 202. First outlet pipeline; 203. Second outlet pipeline; 204. Filtering device; 701. Visual sensor; 702. Infrared spectrum sensor; 703. Edge computing module; 704. Fixed platform; 705. Dynamic adjustment guide rail. DETAILED DESCRIPTION

[0032] The present invention will be further described below with reference to the accompanying drawings.

[0033] See Figure 1-4 The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read. They are not used to limit the conditions for the implementation of the present invention, so they have no substantial technical significance. Any modification of the structure, change in the proportional relationship or adjustment of the size should still fall within the scope of the technical content disclosed by the present invention without affecting the efficacy and purpose that can be achieved by the present invention. At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" quoted in this specification are only for the convenience of description and are not used to limit the scope of the implementation of the present invention. Changes or adjustments in their relative relationships should also be regarded as the scope of the implementation of the present invention without substantially changing the technical content.

[0034] Example 1: A cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine, referring to Figure 1 、 Figure 3 , including at least one liquid storage device 2, which is connected to the cleaning device 1 in turn through the first pump 4 and the pipeline assembly 3, and the cleaning device 1 is used to communicate with the main chemical pipeline. A switching valve is also provided between the cleaning device 1 and the main chemical pipeline, and the switching valve is logically interlocked with the control unit; the cleaning device 1 is a cleaning tank of different sizes of the furnace tube cleaning machine, an ultrasonic generator is provided in the cleaning tank, and a control unit is provided at the entrance of the cleaning tank. The control unit includes a multimodal sensing and dynamic proportioning system, and the multimodal sensing and dynamic proportioning system includes an edge computing module 703. The edge computing module 703 is respectively connected to the visual sensor 701 and the infrared spectrum sensor 702. The visual sensor 701 is 20 million pixels and contains an AI image processing unit. The wavelength range of the infrared spectrum sensor is 2.5-25μm. The edge computing module 703 has built-in material and size parameters. The visual sensor 701 and the infrared spectrum sensor 702 are located at the feed inlet of the cleaning device 1; the control unit is connected to the first pump 4 and the ultrasonic generator. The present invention adopts a multimodal sensing and dynamic proportioning system, and uses visual sensors and infrared spectrum sensors to build a material-size database. By supporting classification and identification of quartz / silicon carbide / quartz + deposition products (TOES, POLY), it realizes intelligent identification of material materials and sizes, and adaptively matches cleaning parameters. It can effectively control and clean the equipment and production spare parts involved in the chip manufacturing production process. The multi-level cleaning space, recycled waste liquid recovery design, and multi-layer filtration reduce the presence of various types of particles, increase efficiency for the enterprise, and achieve cost and expenditure reduction, energy saving and consumption reduction; it solves the technical defects of the existing technology such as poor cleaning uniformity, solidified process parameters, excessive energy consumption, secondary pollution risk, low requirements for particles after cleaning, and lack of material adaptability.

[0035] Example 2: Based on Example 1, the visual sensor 701 and infrared spectrum sensor 702 are mounted on a fixed platform 704 via a dynamically adjustable guide rail 705. Fixed platform 704 is located at the guide mechanism at the entrance of the cleaning tank. The visual sensor 701 is used to collect material surface images, and the infrared spectrum sensor 702 is used to collect material surface spectral data. This invention uses a visual sensor and infrared spectrum sensor installed at a 45° angle at the inlet of the cleaning tank. The sensor data is transmitted to the edge computing module via the RS485 bus.

[0036] Example 3: Based on Example 2, the control unit has a built-in FPGA chip, which uses the convolutional neural network CNN algorithm to analyze the material type and three-dimensional size of the material, and matches the material and size with the built-in material and size matching algorithm of the edge computing module 703.

[0037] Example 4. Based on Example 2, the liquid storage device 2 is a liquid storage tank of the furnace tube cleaning machine. There are three liquid storage tanks, each of which is an independent liquid storage tank. The three liquid storage tanks correspond to tank A, tank B, and tank C respectively, and each liquid storage tank stores chemical reagents with different pH values.

[0038] Example 5: Based on Example 4, refer to Figure 2 Each of the liquid storage tanks is provided with a liquid level sensor 201. Separate channels are provided at the outlets of the three liquid storage tanks. Each channel is connected to the pipeline assembly 3 via the detection unit assembly 7, the first solenoid valve, and the first pump 4. The liquid level sensor 201, the first pump 4, and the first solenoid valve are electrically connected to the control unit. The present invention employs a liquid level sensor for real-time detection of the liquid level within the liquid storage tank, enabling the control unit to prioritize replenishment of the current liquid level.

[0039] Example 6: Based on Example 4, the pipeline assembly 3 includes a Teflon tube, one end of which is connected to a corresponding liquid storage tank via three first pumps 4. The Teflon tube and the liquid storage tank are connected and controlled by a diaphragm valve. The other end of the Teflon tube is connected to a corresponding cleaning tank via several second solenoid valves, all of which are electrically connected to a control unit. The cleaning tank of the present invention is divided into a medium-long cleaning tank and a small inward-facing cleaning tank, ensuring that a single machine can perform multiple cleaning operations and can accommodate a wider range of workpiece cleaning tools. The cleaning tank is equipped with a vapor phase cleaning module that uses an atomized organic solvent (such as IPA) under an inert gas carrier to remove electrostatically adsorbed particles. The cleaning tank is equipped with an online thermoelectric temperature control sensor to monitor the temperature and resistivity of the pure water in the cleaning liquid in real time, dynamically adjust the pH value (with an accuracy of ±0.2) and temperature (with a control of ±1°C) of the cleaning agent to ensure the corrosion rate of the chemical reagent and improve cleaning efficiency. The cleaning tank is equipped with a built-in osmotic filtration composite waste liquid treatment unit to remove metal ions and microparticles from the cleaning agent, achieving a reagent reuse rate of ≥85%.

[0040] Embodiment 7: Based on embodiment 6, a drainage pump is provided at the outlet of each cleaning tank, and the drainage pump is communicated with the discharge pipeline.

[0041] Example 8. On the basis of Example 6, each of the liquid storage tanks includes two outlet pipes on the upper part, which correspond to the first outlet pipe 202 and the second outlet pipe 203 respectively. The second outlet pipe 203 is connected to the pipe assembly 3 through the channel. A filter device 204 is provided in the liquid storage tank, and the filter device 204 is arranged corresponding to the second outlet pipe 203; the first outlet pipe 202 is connected to the circulation system 5, and the circulation system 5 is connected in two ways, one way is connected to the cleaning equipment 1 through two third solenoid valves connected in series, and the other way is connected to the discharge liquid detection component, the second pump, and the fourth solenoid valve in sequence; the discharge liquid detection component is also connected to the discharge pipe through the liquid detection delay component, the drainage pump control component, the drainage pump leakage detection component, and the original equipment concentrated fluorine discharge solenoid valve component.

[0042] Example 9: Based on Example 8, the liquid storage device 2 further includes a liquid inlet and a liquid discharge port at a lower portion. The liquid discharge port is connected to a fourth solenoid valve, and the liquid inlet is connected to the liquid inlet solenoid valve. The liquid inlet solenoid valve is electrically connected to the control unit. The present invention uses a liquid level sensor to detect the liquid level inside the liquid storage tank in real time, so that the control unit can close and stop the liquid inlet solenoid valve connected to the liquid storage tank, thereby achieving automatic control of the pipeline on and off.

[0043] Example 10: A cleaning optimization method for a power semiconductor discrete device furnace tube cleaning machine, comprising the following steps:

[0044] S1. Material preparation,

[0045] The material to be cleaned is placed on a conveyor belt at the entrance of the cleaning tank;

[0046] S2. Image acquisition,

[0047] ① The visual sensor 701 collects material surface image data in real time and transmits it to the control unit;

[0048] ② The infrared spectrum sensor 702 collects the reflected spectrum data in real time and transmits it to the control unit;

[0049] S3. Contour extraction,

[0050] The control unit has a built-in FPGA chip, which uses the convolutional neural network (CNN) algorithm to extract the contours of the material surface image data and obtain the three-dimensional size of the material.

[0051] S4. Material classification,

[0052] The FPGA chip uses the convolutional neural network (CNN) algorithm to analyze the reflectance spectrum data and obtain the material type.

[0053] S5. Ratio decision,

[0054] ① Calculate the preset ratio based on the material type and the chemical reagents stored in the storage tank (A / B / C tank), and preset the reagent ratio table (e.g., quartz → HF / H2O2, 1:4 mixture; silicon carbide → HF / HCL / H2O2, 1:1:4 mixture);

[0055] ② Select the corresponding cleaning tank according to the size data, and preset the ultrasonic frequency, micro jet pressure, and cleaning time;

[0056] S6. Parameters are sent.

[0057] ①The control unit transfers the material to be cleaned to the corresponding cleaning tank;

[0058] ② The control unit controls the opening of the first solenoid valve and the first pump 4 corresponding to the liquid storage tank (tank A / B / C), and opens the second solenoid valve corresponding to the cleaning tank (switching accuracy ±0.01s) to control the reagent mixing ratio, and outputs the chemical reagent to the corresponding cleaning tank according to the preset ratio;

[0059] ③ Dynamically adjust the ultrasonic frequency according to preset data, with an adjustment range of ±5kHz, a microjet pressure of 0.5-3.0MPa, and a cleaning duration of ±10% of the baseline value. This invention integrates multimodal sensing technology, using visual and infrared spectroscopy sensors to intelligently identify material size and material, and adaptively match cleaning parameters. It integrates an ultrasonic cavitation cleaning layer (adjustable from 20-80kHz) and a rotating microjet spray layer. Through the synergistic action of a bottom matrix array and a top multi-angle nozzle group, it achieves simultaneous cleaning of microscopic pores and macroscopic surfaces.

[0060] Example 11. Based on Example 10, when a quartz workpiece is identified, the system automatically calls liquid tank A (HF / H2O2, 1:4 mixture) and sets the ultrasonic frequency to 40kHz; when silicon carbide material is detected and the maximum size is greater than 235mm, the extended cleaning mode is enabled, and the total cleaning time is increased by 15%.

[0061] The above embodiments of the present invention are merely preferred embodiments of the present invention. It should be noted that a person skilled in the art may make several improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine, comprising at least one liquid storage device (2), characterized in that: The liquid storage device (2) is connected to the cleaning device (1) in sequence through the first pump (4) and the pipeline assembly (3). The cleaning device (1) is a cleaning tank of different sizes of a furnace tube cleaning machine. An ultrasonic generator is provided in the cleaning tank. A control unit is provided at the entrance of the cleaning tank. The control unit includes an edge computing module (703). The edge computing module (703) is connected to the visual sensor (701) and the infrared spectrum sensor (702) respectively. The edge computing module (703) has built-in material and size parameters. The visual sensor (701) and the infrared spectrum sensor (702) are located at the feed port of the cleaning device (1); the control unit is connected to the first pump (4) and the ultrasonic generator.

2. The cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine according to claim 1, characterized in that: The visual sensor (701) and the infrared spectrum sensor (702) are mounted on a fixed platform (704) via a dynamically adjustable guide rail (705). The fixed platform (704) is located at the guide mechanism at the entrance of the cleaning tank. The visual sensor (701) is used to collect images of the material surface, and the infrared spectrum sensor (702) is used to collect spectral data of the material surface.

3. The cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine according to claim 2, characterized in that: The control unit has a built-in FPGA chip, which analyzes the material type and three-dimensional size of the material through a convolutional neural network (CNN) algorithm, and has a built-in material and size matching algorithm with the edge computing module (703).

4. The cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine according to claim 2, characterized in that: The liquid storage device (2) is a liquid storage tank of the furnace tube cleaning machine. There are three liquid storage tanks, each of which is an independent liquid storage tank. The three liquid storage tanks correspond to tank A, tank B, and tank C respectively, and each liquid storage tank stores chemical reagents with different pH values.

5. The cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine according to claim 4, characterized in that: A liquid level sensor (201) is provided in any one of the liquid storage tanks, and separate channels are provided at the outlets of the three liquid storage tanks respectively. Any one of the channels is connected to the pipeline assembly (3) through the detection unit assembly (7), the first solenoid valve, and the first pump (4). The liquid level sensor (201), the first pump (4), and the first solenoid valve are electrically connected to the control unit.

6. The cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine according to claim 4, characterized in that: The pipeline assembly (3) includes a Teflon tube, one end of which is connected to a corresponding liquid storage tank via three first pumps (4), and the Teflon tube and the liquid storage tank are connected and controlled via a diaphragm valve; the other end of the Teflon tube is connected to a corresponding cleaning tank via several second solenoid valves, and all the second solenoid valves are electrically connected to the control unit.

7. The cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine according to claim 6, characterized in that: The outlet of each cleaning tank is provided with a drainage pump, which is communicated with a discharge pipeline.

8. The cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine according to claim 6, characterized in that: Each of the liquid storage tanks comprises two outlet pipelines at the top, the two outlet pipelines corresponding to the first outlet pipeline (202) and the second outlet pipeline (203), respectively. The second outlet pipeline (203) is connected to the pipeline assembly (3) through a channel, and the first outlet pipeline (202) is connected to the circulation system (5). The circulation system (5) is connected in two ways, one way is connected to the cleaning device (1) through two third electromagnetic valves connected in series, and the other way is connected to the discharge liquid detection assembly, the second pump, and the fourth electromagnetic valve in sequence; the discharge liquid detection assembly is also connected to the discharge pipeline through the liquid detection delay assembly, the discharge pump control assembly, and the discharge pump leakage detection assembly.

9. The cleaning optimization device for a power semiconductor discrete device furnace tube cleaning machine according to claim 8, characterized in that: The liquid storage device (2) further comprises a liquid inlet and a liquid discharge port at the bottom, the liquid discharge port is connected to the fourth solenoid valve, the liquid inlet is connected to the liquid inlet solenoid valve, and the liquid inlet solenoid valve is electrically connected to the control unit.

10. A cleaning optimization method for a power semiconductor discrete device furnace tube cleaning machine, characterized in that: The following steps are included: S1. Material preparation, The material to be cleaned is placed on a conveyor belt at the entrance of the cleaning tank; S2. Image acquisition, ① The visual sensor (701) collects material surface image data in real time and transmits it to the control unit; ② The infrared spectrum sensor (702) collects the reflected spectrum data in real time and transmits it to the control unit; S3. Contour extraction, The control unit has a built-in FPGA chip, which uses the convolutional neural network (CNN) algorithm to extract the contours of the material surface image data and obtain the three-dimensional size of the material; S4. Material classification, The FPGA chip uses the convolutional neural network (CNN) algorithm to analyze the reflectance spectrum data and obtain the material type; S5. Ratio decision, ① Calculate the preset ratio based on the material type and the chemical reagents stored in the three liquid storage tanks; ② Select the corresponding cleaning tank according to the size data, and preset the ultrasonic frequency, micro jet pressure, and cleaning time; S6. Parameters are sent. ①The control unit transfers the material to be cleaned to the corresponding cleaning tank; ② The control unit controls the opening of the first solenoid valve and the first pump (4) corresponding to the liquid storage tank (tank A / B / C), and the opening of the second solenoid valve corresponding to the cleaning tank, and outputs the chemical reagent to the corresponding cleaning tank according to a preset ratio; ③ According to the preset data, dynamically adjust the ultrasonic frequency. The ultrasonic frequency adjustment range is ±5kHz, the microjet pressure is 0.5~3.0MPa, and the cleaning time is ±10% of the baseline value.

Images