An accurate temperature control method based on machine learning

Abstract

The present invention provides an accurate temperature control method based on machine learning. The temperature control method is used for growing silicon carbide crystals by a liquid phase method, and includes the following steps: (1) Machine learning stage: placing growth raw materials in a crucible, setting a first temperature-time curve for heating up to a target temperature, heating the crucible through a heating device to achieve the first temperature-time curve, so that the growth raw materials are melted into a melt; wherein, collecting the power of the heating device to obtain the temperature-power relationship of the heating device; (2) Growth stage: setting a second temperature-time curve required for growing silicon carbide crystals, and controlling the power of the heating device according to the temperature-power relationship to achieve the second temperature-time curve, and completing the growth of silicon carbide crystals. The temperature control method of the present invention can accurately control the growth temperature of growing silicon carbide by a liquid phase method, can adapt to different working conditions and parameters, has a high degree of automation, good repeatability, and is easy to adjust and improve.

Description

Technical Field

[0001] The present invention belongs to the technical field of silicon carbide. Specifically, the present invention relates to an accurate temperature control method based on machine learning. Background Art

[0002] Silicon carbide (SiC) is one of the wide-bandgap semiconductor materials that have received extensive attention. It has the advantages of low density, large bandgap (at room temperature, the bandgap of 4H-SiC is 3.2 eV), high breakdown field strength (about 10 times that of Si), high saturated electron mobility (about 2 times that of Si), high thermal conductivity (3 times that of Si, 10 times that of GaAs), and good chemical stability. It is an ideal substrate material for making high-frequency, high-voltage, high-power devices and blue light-emitting diodes. It has important application potential in the fields of electric vehicles, rail transit, high-voltage power transmission and transformation, photovoltaics, 5G communication, etc.

[0003] The current main growth method of SiC is physical vapor transport method, but it has problems such as high defect density, difficult diameter expansion and p-type doping. The liquid-phase method has a low growth temperature, a relatively stable growth environment, a growth process close to thermodynamic equilibrium conditions, good crystal quality, and good prospects in terms of diameter expansion and p-type doping. In recent years, it has received widespread attention from the academic and industrial circles.

[0004] The basic principle of growing silicon carbide by the liquid-phase method is as follows: Put the metal raw material containing Si into a high-purity graphite crucible, heat the crucible by induction heating or resistance heating to melt the raw material, bring the silicon carbide seed crystal into contact with the melt, the melt containing Si provides Si, C on the inner wall and bottom of the graphite crucible dissolves into the melt, and Si and C are transported to the seed crystal by diffusion and convection to realize the growth of SiC crystals. During the growth process, generally, the temperature at the seed crystal is collected as the growth temperature. The growth temperature affects the dissolution, transport and precipitation of solutes, and thus affects the quality of silicon carbide crystals. Therefore, it is crucial to accurately control the growth temperature during the crystal growth process.

[0005] The existing process generally uses the existing experience of technicians to directly control the power of the heating power supply so that the growth temperature reaches the ideal crystal growth temperature, and by controlling the change of power, the growth temperature changes according to the expected trend. This method relies heavily on experience, the production process cannot be standardized, and it increases time and labor costs. In actual production, the thermal field configuration is constantly changing. Once the top insulation changes, such as the number of layers or the size of the opening of the top graphite insulation felt, the corresponding relationship between the heating power and the growth temperature that can be achieved under this power will change. The technicians need to re-explore the relationship between power and temperature, which increases the difficulty in the crystal growth process. In large-scale production, there are a large number of crystal growth furnaces, and the heating system (induction coil or resistance heater) of each crystal growth furnace is more or less different. In addition, the effect of the artificial insulation structure is not completely consistent. The heating rules of each furnace during crystal growth are different, and the insulation material in the furnace will be ablated after each growth, which degrades the insulation effect. The heater of the resistance furnace will also have problems such as powdering and ablation after long-term use, which changes the heating effect. This increases the difficulty of power control.

[0006] Using an automatic temperature controller to control the temperature to change according to the ideal crystal growth temperature curve seems to solve this problem. In theory, the actual temperature can be compared with the set temperature, and some computer algorithms (such as PID algorithm) can be used to control the heater to output a certain heating power, thereby controlling the actual temperature to be close to the set temperature. However, this commonly used method is not suitable for the liquid phase growth of silicon carbide crystals. First, the metal will volatilize after melting, and the volatiles will adhere to the surface of the quartz window, affecting the transmittance of the window to infrared light, significantly reducing the temperature measurement accuracy, and the measured temperature will be lower than the actual temperature. In addition, the high-speed rotation of the seed crystal causes the temperature measurement point to change continuously, and the air flow movement in the crystal growth furnace will also affect the temperature measurement. In this case, the temperature control accuracy is poor; second, the temperature fluctuations during the crystal growth process have a great influence on the stability of the growth interface, and the thermal field of the crystal growth furnace is large and complex in structure. It is a system with serious hysteresis. The use of an automatic temperature controller for control will inevitably cause the growth temperature and heater power to fluctuate in a wave-like manner, which is extremely unfavorable for the growth of high-quality crystals.

[0007] Therefore, there is an urgent need for a method that can accurately control the growth temperature of liquid phase silicon carbide growth, while reducing manual operations, achieving high automation and high repeatability to adapt to different working conditions and parameters. Summary of the invention

[0008] The object of the present invention is to provide a precise temperature control method based on machine learning, which is used for growing silicon carbide crystals by the liquid phase method. The temperature control method of the present invention can precisely control the growth temperature of silicon carbide by the liquid phase method, can adapt to different working conditions and parameters, has a high degree of automation, good repeatability, and is easy to adjust and improve.

[0009] The above object of the present invention is achieved by the following technical solutions.

[0010] The present invention provides a precise temperature control method based on machine learning, which is used for growing silicon carbide crystals by the liquid phase method, and includes the following steps:

[0011] (1) Machine learning stage: Place the growth raw material in a crucible, set the first temperature-time curve for heating up to the target temperature, and heat the crucible through a heating device to achieve the first temperature-time curve, so that the growth raw material melts into a melt; wherein, collect the power of the heating device to obtain the temperature-power curve of the heating device;

[0012] (2) Growth stage: Set the second temperature-time curve required for growing silicon carbide crystals, and control the power of the heating device according to the temperature-power relationship to achieve the second temperature-time curve and complete the growth of silicon carbide crystals.

[0013] In some embodiments of the present invention, the first temperature-time curve contains several constant temperature zones, and the constant temperature zone is a region where the temperature does not change with time; preferably, collect the power of the constant temperature zone and fit to obtain the temperature-power curve.

[0014] In some embodiments of the present invention, the first temperature-time curve contains 4 to 10 constant temperature zones.

[0015] In some embodiments of the present invention, the interval between each constant temperature zone is 50 to 200 °C, and the time maintained for each constant temperature zone is 5 to 20 minutes.

[0016] In some embodiments of the present invention, according to the power of the heating device collected, the temperature-power curve is obtained by fitting using a high-order polynomial, an exponential function, the least squares method, the cubic spline interpolation method or the Bezier curve method.

[0017] In the present invention, a constant temperature zone is set, the power change of the constant temperature zone is sampled on average, and the power value required to stably maintain the current temperature point is calculated, so as to fit the temperature-power curve of the heating device.

[0018] In the present invention, in the step of heating up to the target temperature in the machine learning stage of step (1), the target temperature is determined by the melting point of the selected raw material (such as 2000 °C), and usually, after placing the growth raw material in the crucible, it is heated from room temperature to the target temperature.

[0019] In some embodiments of the present invention, in the machine learning stage, a PID algorithm is used to control the power of the heating device to achieve the first temperature-time curve. When the heating device is heating up, an infrared thermometer is usually used to measure the temperature inside the device, and the PID algorithm is used to make the measured temperature by the infrared thermometer during the heating process close to the set temperature-time curve.

[0020] In some embodiments of the present invention, the PID algorithm is a self-tuning algorithm or a fuzzy PID algorithm. The self-tuning algorithm or the fuzzy PID algorithm can be applied to different temperature field arrangements and working conditions and is suitable for silicon carbide single crystal furnaces.

[0021] In some embodiments of the present invention, the heating device is an induction heater or a resistance heater. The silicon carbide single crystal furnace is an induction heating single crystal furnace or a resistance heating single crystal furnace.

[0022] In some embodiments of the present invention, according to the cumulative experimental data set of the heating device and in combination with the power of the heating device collected, the temperature-power curve is obtained by extrapolation fitting.

[0023] The cumulative experimental data set in the present invention is the corresponding relationship between relevant factors and temperature obtained by analyzing factors such as the thermal field size, thermal field structure, thermal field ablation, crucible and seed crystal holder structure, etc. after the heating device has completed crystal growth (such as the cumulative crystal growth experiments of the heating device in the past). Single-factor variable experiments can also be carried out to determine the influence of the above factors on the temperature and obtain the cumulative experimental data set. Combining the cumulative experimental data set with the power of the heating device collected can make the temperature-power curve obtained by extrapolation fitting more accurate.

[0024] In some embodiments of the present invention, in the growth stage, after the silicon carbide seed crystal contacts the raw material liquid surface, according to the temperature-power relationship, the power of the heating device is controlled to achieve the second temperature-time curve.

[0025] In some embodiments of the present invention, in step (2), when controlling the power of the heating device, the power of the heating device changes continuously.

[0026] In the present invention, continuous change means that the change in the power of the heater during crystal growth does not exceed 500 W / min.

[0027] In the present invention, in the growth stage, the power of the heating device is controlled so that the change does not exceed 500 W / min.

[0028] In some embodiments of the present invention, the method further includes the following steps:

[0029] (3) Cooling stage: Separate the silicon carbide crystal, set the third temperature-time curve for reducing the temperature to the target temperature, and control the power of the heating device to achieve the third temperature-time curve.

[0030] In the present invention, after the crystal growth in step (2) is completed, the crystal is lifted and separated from the raw material liquid surface to complete the cooling.

[0031] In some embodiments of the present invention, in the cooling stage of step (3), the third temperature-time curve can be achieved by the temperature-power curve obtained in the machine learning stage to reduce the temperature to the target temperature (such as room temperature), or the cooling can also be completed by directly controlling the power of the heating device or through the temperature control of the PID algorithm.

[0032] Compared with the prior art, the present invention has at least the following beneficial effects:

[0033] (1) The present invention provides a temperature control method for growing silicon carbide crystals by the liquid phase method based on machine learning. In the temperature control method of the present invention, during the melting of the raw materials (acquisition stage), temperature control heating is adopted, and the stable holding power at multiple temperature points is collected, and then the temperature-power curve is fitted. According to the set growth process, the power of the heating device is controlled based on the temperature-power curve; during the cooling / annealing after the crystal growth is completed, the heating power is controlled according to the set cooling curve until the temperature drops to room temperature or the set temperature. The present invention can achieve precise control of the growth temperature during the growth of silicon carbide by the liquid phase method, reduce manual operation, and achieve high automation and high repeatability to adapt to different working conditions and parameters.

[0034] (2) Compared with the simple power control heating method, the temperature control method of the present invention has high temperature control accuracy, does not require manual power control, does not rely on experience, reduces the process difficulty of growing silicon carbide by the liquid phase method, and is not affected by the change of the thermal field configuration, the difference of the heating systems of different heating devices, and the ablation degradation of the thermal insulation material and the heater material of the heating device, reducing the time and labor costs, and can realize the standardization of the growth process, which is beneficial to process research and development and large-scale production.

[0035] (3) Compared with the temperature control heating method, the temperature control method of the present invention adopts power control during the growth of silicon carbide crystals. There will be no power fluctuations during the crystal growth process, and it is not affected by volatiles, seed rotation, and gas convection in the furnace, with higher temperature control accuracy and better quality of the grown silicon carbide crystals.

[0036] (4) The temperature control method of the present invention only adopts temperature control heating during the process of melting raw materials before crystal growth. The power fluctuation during the temperature control heating process will not affect crystal growth. This stage is short compared to the entire growth process, and the influence of volatiles can be basically ignored. Moreover, the seed crystal does not rotate, the temperature measurement point is constant, and the temperature measurement accuracy is high. The obtained temperature-power relationship is more accurate. Description of the Drawings

[0037] Hereinafter, the embodiments of the present invention will be described in detail with reference to the drawings, wherein:

[0038] Figure 1 is the temperature-power curve obtained by fitting in Embodiment 1 of the present invention;

[0039] Figure 2 is the time-temperature-power change diagram of Embodiment 1 of the present invention;

[0040] Figure 3 is the appearance diagram of the silicon carbide crystal obtained in Embodiment 1 of the present invention;

[0041] Figure 4 is the time-temperature-power relationship diagram of Comparative Example 1 of the present invention;

[0042] Figure 5 is the appearance diagram of the silicon carbide crystal obtained in Comparative Example 1 of the present invention. Detailed Embodiments

[0043] The present invention will be further described in detail below in conjunction with the specific embodiments. The embodiments given are only for clarifying the present invention, rather than limiting the scope of the present invention.

[0044] In the following examples and comparative examples, silicon and chromium were purchased from Zhongnuoxin Materials (Beijing) Technology Co., Ltd., with purities of 99.999% and 99.95% respectively; 4-inch semi-insulating 4H-SiC was purchased from Beijing Tiankeheda Semiconductor Co., Ltd.

[0045] Embodiment 1

[0046] In this embodiment, an induction heating crystal growth furnace was used. Silicon and chromium were used as growth raw materials and placed in a graphite crucible. The graphite crucible was placed in a graphite heating cylinder, and a graphite soft felt was wrapped outside the graphite heating cylinder as a heat insulation layer. 4-inch 4H semi-insulating silicon carbide was used as the seed crystal.

[0047] The liquid-phase method for growing silicon carbide crystals in Embodiment 1 includes the following steps:

[0048] (1) After loading the growth raw materials into the silicon carbide single crystal furnace, the power of the heating power supply is controlled by the limit cycle self-tuning PID algorithm to make the measured temperature of the infrared thermometer close to the target temperature until the raw materials melt. The preset temperature-time curve for heating to the target temperature is to increase the temperature from 20°C to 1800°C in 5 hours, and maintain for 10 minutes at 1000°C, 1200°C, 1400°C, 1600°C and 1800°C. During this period, average value sampling is carried out to calculate the power value required to maintain at the current temperature point, and the temperature-power curve in the range of 200 - 2000°C is obtained by extrapolation fitting using the least square method. The result is as Figure 1 shown;

[0049] (2) Lower the silicon carbide seed crystal to contact the liquid surface of the melted raw materials to start growing silicon carbide crystals. Referring to the fitted power-temperature curve, according to the set growth process, directly control the power of the silicon carbide single crystal furnace to change the temperature, so that the growth temperature decreases by 100°C within 100 hours, and crystal growth ends after 100 hours;

[0050] (3) After growth ends, lift the silicon carbide crystal to separate from the liquid surface, control the power of the silicon carbide single crystal furnace to make the temperature drop to 400°C within 20 hours, and then reduce the power to 0.

[0051] The time-temperature-power relationship diagram of growing silicon carbide crystals in Example 1 is as Figure 2 shown, and the external shape diagram of the silicon carbide crystal grown in Example 1 is as Figure 3 shown.

[0052] Comparative Example 1

[0053] In this comparative example, an induction heating crystal growth furnace is used. Silicon and chromium are used as growth raw materials and placed in a graphite crucible. The graphite crucible is placed in a graphite heating cylinder, and a graphite soft felt is wrapped outside the graphite heating cylinder as a heat insulation layer. Semi-insulating 4-inch 4H-SiC is used as the seed crystal. The difference is that temperature control is adopted throughout the entire crystal growth process, and the power of the heating power supply is controlled by the limit cycle self-tuning PID algorithm.

[0054] The liquid phase method for growing silicon carbide crystals in Comparative Example 1 includes the following steps:

[0055] (1) After loading the growth raw materials into the silicon carbide single crystal furnace, the preset temperature-time curve for heating to the target temperature is to increase the temperature from 20°C to 1800°C in 5 hours until the raw materials melt;

[0056] (2) Lower the silicon carbide seed crystal to contact the liquid surface of the melted raw materials to start growing silicon carbide crystals. According to the set growth process, the preset target temperature curve is to decrease by 100°C within 100 hours, that is, decrease to 1700°C, and crystal growth ends after 100 hours;

[0057] (3) After the growth is completed, the crystal is lifted and separated from the liquid surface. The preset target temperature curve is to drop to 400 °C within 20 hours, and then the power is reduced to 0.

[0058] The time-temperature-power relationship diagram of growing silicon carbide crystals in Comparative Example 1 is as Figure 4 shown, and the external shape diagram of the silicon carbide crystal grown from Comparative Example 1 is as Figure 5 shown.

[0059] Comparing Figure 3 the silicon carbide crystal obtained in Example 1 in Figure 5 with the silicon carbide crystal obtained in Comparative Example 1 in

[0060] It should be noted finally that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A precise temperature control method based on machine learning, which is used for growing silicon carbide crystals by the liquid phase method, comprising the following steps: (1) Machine learning stage: Place the growth raw material in a crucible, set the first temperature-time curve for heating up to the target temperature, and heat the crucible by controlling the heating device to achieve the first temperature-time curve, so that the growth raw material melts into a melt. In the machine learning stage, the PID algorithm is used to control the power of the heating device to achieve the first temperature-time curve; wherein, collect the power of the heating device to obtain the temperature-power curve of the heating device, and use a high-order polynomial, exponential function, least squares method, cubic spline interpolation method or Bezier curve method to fit the temperature-power curve according to the collected power of the heating device; (2) Growth stage: Set the second temperature-time curve required for growing silicon carbide crystals, and control the power of the heating device according to the temperature-power relationship to achieve the second temperature-time curve and complete the growth of silicon carbide crystals.

2. The method according to claim 1, wherein, The first temperature-time curve contains several constant temperature zones, and the constant temperature zone is a region where the temperature does not change with time.

3. The method according to claim 2, wherein, Collect the power of the constant temperature zone to obtain the temperature-power curve.

4. The method according to claim 2, wherein, The first temperature-time curve contains 4 to 10 constant temperature zones.

5. The method according to claim 2, wherein, The interval between each constant temperature zone is 50 to 200 °C, and the holding time of each constant temperature zone is 5 to 20 minutes.

6. The method according to any one of claims 1 to 5, wherein, The PID algorithm is a self-tuning algorithm or a fuzzy PID algorithm.

7. The method according to any one of claims 1 to 5, wherein The heating device is an induction heater or a resistance heater.

8. The method according to any one of claims 1 to 5, wherein In the machine learning stage, according to the cumulative experimental data set of the heating device, combined with the collected power of the heating device, extrapolate and fit to obtain the temperature-power curve.

9. The method according to any one of claims 1 to 5, wherein In the growth stage, when the silicon carbide seed crystal contacts the liquid surface of the melt, control the power of the heating device to achieve the second temperature-time curve.

10. The method according to any one of claims 1 to 5, wherein, In the growth stage, control the power of the heating device so that the change does not exceed 500 W / min.

11. The method according to any one of claims 1 to 5, wherein, The method further includes the following steps: (3) Cooling stage: Separate the silicon carbide crystal, set the third temperature-time curve for cooling to the target temperature, and control the power of the heating device to achieve the third temperature-time curve.

Images