E3S Web of Conferences 124, 05013 (2019) https://doi.org/10.1051/e3sconf/201912405013 SES-2019

The problem of creating an automated system to control growth of single from melt as a problem of control and monitoring of a complex nonlinear and dynamic system

V A Petrosyan1,*, A V Belousov2, A G Grebenik2 and Yu A Koshlich2

1LLC "Techsapphire", Belgorod, Russia 2BSTU named after V.G. Shukhov, Belgorod, Russia

Abstract. The paper mentions some problems of automated control system development for growth of large (150 kg and above) sapphires. We obtain an analytical equation for the temperature distribution and thermal stresses along the crystal axis during the growth phase. An analysis was carried out and numerical estimates were obtained for the axial distribution of components of thermoelastic stresses depending on the physical, optical, and geometric parameters of the crystal. It is shown that the cause of thermal stresses and blocks during is the nonlinear temperature dependence of thermal conductivity and thermal expansion coefficients.

1 Introduction a forced convention and with different density of melt saturation with gas inclusions in these zones. As a result, Based on the general concepts of the theory of normal in the places of melting, more favorable conditions arise directional from the melt, it is known that for the capture of micron-sized bubbles and their further the crystallization rate 푣푐푟 is proportional to the growth into a single crystal. supercooling of the melt ∆푇 or the temperature gradient It follows from the above that an important at the phase interface. That is, 푣푐푟 ~ ∆푇 or 푣푐푟 ~ 푔푟푎푑푇. conclusion is that the magnitude of the supercooling at In the harmonic approximation for the growth rate, the crystallization front determines not only the the expression: conditions for the growth of a single crystal, but also its perfection.

vvvcr =++ sint01() (1) We need to note one more important fact for the controlling systems of the growth rate and maintaining stability at the crystallization front. The fact is that for where 푣0 is an average crystal growth rate, 푣1 is an amplitude of change in growth rate due to temperature the melting point and the crystallization fluctuations in the melt near the phase transition temperature do not coincide. The melting point is well boundary, 휔 is a temperature oscillation frequency, 푡 is reproduced at different heating rates, and the time, 휑 is some phase. Based on equation (1), we can crystallization temperature depends on the cooling rate determine the magnitude of the dynamic supercooling or of the melt [1]. Therefore, in formula (2), ∆푇0 should be overheating of the melt at the crystallization front: understood as the average value of "physical supercooling". =TTBsint ++ () (2) It follows from the above that fluctuations in the 0 magnitude of dynamic supercooling affect the change in the rate of crystallization and lead to stricter where ∆푇 is a fixed value of "physical supercooling" 0 requirements for the system for stabilizing gradients at (the difference between the melting point and the the crystallization front of 퐴푙 푂 , than, for example, for crystallization temperature), 퐵 is a supercooling change 2 3 silicon. amplitude. We will now define two cases of dynamic It can be seen from equation (2) that, depending on supercooling. The first is small fluctuations in the amplitude and phase of the change in the supercooling, within the zone of "physical supercooling, periodic localized melting of the crystal at supercooling", which are stochastic in nature and depend the crystallization front can occur with re-crystallization on many factors. These processes are described by of the melted layer. methods of nonlinear dynamics, in particular, the theory It should be noted that temperature fluctuation in of dynamic or deterministic chaos. individual local areas on the crystallization front may The second case is high-amplitude supercooling occur due to mixing of hotter and less hot melt layers at oscillations that go beyond the limits of "physical

* Corresponding author: [email protected] © The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 (http://creativecommons.org/licenses/by/4.0/). E3S Web of Conferences 124, 05013 (2019) https://doi.org/10.1051/e3sconf/201912405013 SES-2019

supercooling". These fluctuations can occur during This model is known from studies of plasma physics growth control using an automated system that operates and the theory of controlled thermonuclear fusion and is on a feedback signal from a weight sensor and regulates called the model of complex thermal structures. The the power at the heater. complex structures in this model are unstable and can Since dynamic weighing is an indirect method of controlling crystal growth, it becomes necessary to use exist only with the correct specification of the initial the so-called observation equation, which relates the data. In [5], we have investigated a similar model when force acting on the weight sensor with the required calculating the temperature distribution along the crystal crystal mass and the rate of its crystallization. axis during growth and optimized some process parameters based on identification with experimental 2 The problem of controlling the crystal data. growth using the feedback signal from In this case, we were guided by the fact that the the weight sensor parameters 푄(푇) and 퐾(푇) in the equation (3) have a different physical interpretation depending on the optical When are grown by the modified Kyropoulos thickness of the layer 휏 = 훽0퐿 (where 훽0 is the method (GOI method), a significant part of the crystal is attenuation coefficient, 퐿 is the thickness of the layer), in the melt and a pushing force acts on the crystal. The which spreads the flow of heat. magnitude of this force depends on the fluctuation of the melt density depending on the temperature, variations in 3 The role of sapphire crystal the crystallization rate, changes in the angle of growth. It is also necessary to add the dependence of the weight transparency and the effect of nonlinear sensor readings on fluctuations in the temperature of its temperature parameters on the quality environment, on various kinds of vibrations, etc. Of all of crystals these parameters, we distinguish two main ones—the linear crystallization rate (the vertical component of the In the general case, the nonstationary problem of joint rate of front advancement into the melt during growth) radiation-conductive heat transfer in the direction of the and the shape of the crystallization front, depending on X axis in a layer of a semi-transparent medium in terms the angle of the growth cone. The weight sensor signal is of dimensionless parameters can be written as [6]: a function of these two parameters and the cause of the ( ,,, )  22(  ) 1 Q ( ) change in weight gain can be either a change in the =−+ r H (, ) (4) linear crystallization rate or a change in the angle at the  2*2 N apex of the growth cone. In view of this, the control object is not observable in a fairly wide "dead zone" [2- where 휃(휏, 휉) is a temperature, 푄푟(휃, 휉) is a radiation 4]. flux density, 퐻(휏, 휉) is a bulk power of internal sources, With this in mind, as well as the inertia of heat 푁∗ is a conductive-radiation parameter. transfer from the heater to the melt, there is a loss of In this equation, the parameter 푁∗ characterizes the controllability and the control system can enter the self- relative contribution of thermal conductivity compared oscillatory mode. During the growth process, turbulent to radiation. This parameter has large values when flows in the melt, vortex flows at the crystallization thermal conductivity prevails, and small ones when front, etc. can occur. radiation predominates. That is, when the medium is There are two main ways to stabilize a chaotic only scattering (푁∗ → ∞) and the radiation term in system: without feedback and with feedback. The first is equation (4) disappears, then it turns into the usual heat called chaos suppression, and the second is chaos conduction equation. control. The second method which includes active For a medium with a high optical thickness, the control will be described in future research. attenuation coefficient 훽0 is replaced by the absorption The task of searching for controlling external coefficient. That is, for a sapphire crystal, 퐾(푇) is the influences, which make it possible to minimize radiative thermal conductivity coefficient in the diffusion temperature fluctuations at the crystallization front, can approximation (Rosseland approximation [6]). The be formulated in a general form. Suppose that in the expression for this coefficient is [6-8]: system under consideration there are two competing processes: a nonlinear source 푄(푇), which reflects 16NT23 K = SL (5) feedback, and a dissipative source, nonlinearity of which SL 3 is determined by the coefficient 퐾(푇). If these functions have a power form, then the system can be described by where 푇 is a crystallization temperature, 휎 is a Stefan- the equation: 푆퐿 Boltzmann constant, 푁 is a refraction coefficient. If we calculate the conducted-radiation parameter TT   (3) ∗ 퐾퐹훽0 푊 =+KTQT( ) ( ) 푁 = 2 3 for sapphire, then with 퐾퐹 = 3 , 훽0 = t  x  x 4푁 휏푇푆퐿 푚∙퐾  1 0.35 , 푁 = 1.79, 푇 = 2308퐾° and 휎 = 5.67 ∙ 푐푚 푆퐿 푊 10−8 will get 푁∗ = 0.0118. 푚2퐾4

2 E3S Web of Conferences 124, 05013 (2019) https://doi.org/10.1051/e3sconf/201912405013 SES-2019

Thus, in solving the problem of temperature where 훼푇 is a thermal expansion coefficient, 휇 is a distribution along the axis of a growing crystal, the Poisson's ratio, 퐸 is Young's modulus. nonlinear parameter 퐾푆퐿(푇) plays a significant role. Substituting 푇푆퐿(푥) from [5] into (8) can be obtained: In [5], we obtained expressions for the temperature distribution along the axis of a growing crystal 2  depending on the geometric and nonlinear parameters TExsinSLTeffSL(0)   푅 2 푄1(푇) [9] and 퐾푆퐿(푇). Based on this, the formulas for Σx =2 calculating the temperature gradient at the crystallization 21ND( −  ) S front and the maximum crystallization rate were (9) obtained:   2effSL sin −x 2  exp 3 sin  effSL  2NDS TSL 2 (6)  gradTxS ( ) = x→0  2NDS 

훼 The calculation results are shown as graph 1 ( =  2 4 3eff sin 훼 8NT 2 90°) and graph 2 ( = 45°) in Figure 1, taking into v =SL (7) 2 max account the numerical values of the parameters 푁 = 3SSLSZD ° 5 1.79, 푇푆퐿 = 2308퐾 , 휇 = 0.27, 퐸 = 4.34 ∙ 10 푀푃푎 1 훼 훼 Taking into account the numerical values for the [10], 훽 = 24.4 , = 90°, = 45°, 훼 = 5.58 ∙ 푆퐿 푚 2 2 푇 −8 푊 Stefan-Boltzmann constant 휎 = 5.67 ∙ 10 , the −6 1 푚2퐾4 10 [1]. 푊∙ℎ 퐾° specific latent heat of fusion 푍 = 683.2 , the melt 푘𝑔 1 absorption coefficient 훽 = 0.3 ÷ 0.9 , with a fixed 푆퐿 푐푚 crystal diameter 퐷푆 = 0.3푚 and a corner of the growth cone 훼 = 180° numerical values can be obtained for temperature gradients at the crystallization front and maximum growth rate. The results of the calculations are given in tables 1 and 2: Taking into account the results obtained in [5], it is possible to conduct a qualitative analysis and compare the role of various factors in the formation of the thermally stressed state of a growing crystal. Taking into account the formula for temperature distribution along the X axis [5], it is possible to calculate the axial component of thermal stresses using the formula [8]:

 Ex2 dTx2 T SL ( ) (8) Σx =2 (1− ) dx Fig. 1. Distribution of components of thermal stress tensor.

Table 1. Numerical values for temperature gradients at the crystallization front.

℃ Melt absorption coefficient 훽 Gradient magnitude [ ] 푆퐿 Angle of growth cone α 푐푚 [푐푚−1] 4.02 0.3 |푔푟푎푑푇 (푇)| 푆 5.2 0.5 180° 5.6 0.6 10 0.9

Table 2. Numerical values of the maximum growth rate.

푚푚 Melt absorption coefficient 훽 Growth rate [ ] 푆퐿 Angle of growth cone α ℎ [푐푚−1] 6.08 0.3 푣 푚푎푥 4.65 0.5 180° 4.25 0.6 3.0 0.9

3 E3S Web of Conferences 124, 05013 (2019) https://doi.org/10.1051/e3sconf/201912405013 SES-2019

In the same figure, the dashed line (graph 3) shows certain growth conditions can lead to the appearance of the critical shear stress for the sapphire easy slip system blocks in the crystal and even to its cracking. [10] (extrapolation of data from [11]). In turn, the Poynting vector [15, 16] is responsible In [10], it was shown that in the case of a nonlinear for the optical symmetry of the crystal (a cruciform quadratic temperature dependence of the thermal conoscopic picture). Poynting vector is associated with expansion coefficient 훼푇(푇), the normal components of electromagnetic parameters and components of the the thermoelastic stress tensor turn out to be linear dielectric constant tensor, which depend on the functions of the axial coordinate. components of the thermal stress tensor. Based on this, it Graph 4 in Figure 1 shows the linear variation of was concluded [15] that the growth of crystals from the stresses for the case of sapphire growth with length 퐿 = melt using the GOI method, when the crystal is hanging 100푚푚 and diameter 퐷푆 = 10푚푚 in a linear on a moving rod, has a fundamental limitation on the temperature field [10]. size and mass of the crystal. That is, the larger the mass To preserve the scale and display in one picture, we of the grown crystal, the more the conoscopic picture is used the same parameters when calculating using distorted and its optical symmetry is broken. expression (9). From Figure 1 it can be seen that the values of thermal stresses (graphs 1, 2 and 3) exceed or are very References close to critical values in the range up to x = 58 mm. 1. K.L. Lobatsevich, Improving the stability of the rate This means that defects in the form of blocks, twins and of crystallization of monocrystals of leucosapphire cracks are possible in this range. Thus, the nonlinear by the Kyropoulos method by introducing predictive nature of the dependence of the thermal expansion control on the rate of change of mass of the single coefficient causes blocks to appear in the upper part of crystal, Thesis for the degree of candidate of the crystal as it grows to a predetermined diameter after technical sciences (2010) seeding, that is, it ultimately affects the quality of the 2. A.V. Borodin, Equipment for growing sapphire crystal. This is especially important when growing large single crystals by the Kyropoulos method with a crystals using the GOI method. The appearance of device for dynamic crystal weighting and automatic blocks in the upper part of the crystal was observed in control with feedback, Scientific instrument experiments performed at Techsapphire LLC (Belgorod), engineering, 24(3), 92-98 (2014) when growing crystals weighing 100 kg at different 3. A.V. Belousov, Y.A. Koshlich, A.G. Grebenik, temperatures of growing the cone of growth. About one approach to the automation of the process of seeding the synthetic sapphire single 4 Conclusion crystals by the Kyropoulos method, The Bulletin of BSTU named after V.G. Shukhov, 3, 128-133 Based on the analysis of work on the development of (2017) GOI and Czochralski [12-14] methods for growing 4. A.G. Grebenik, M.V. Panchenko, S.A. Karatach, single crystals from the melt, we can conclude that, to Improving the efficiency of the process of seeding date, there is no fully automated, strictly replicable single crystals of artificial sapphire on the basis of technology for the production of defect-free single technical vision using the method of selection of crystals. For single crystal sapphires, this is primarily active contours, News of South-West State optical symmetry, the absence of residual thermal University. Management series, computer stresses and blocks, the absence of bubbles and technology, computer science. Medical Instrument, impurities. This is not related to the technical capabilities 8-3(28), 41-49 (2018) of creating automated growth process control systems. 5. V.A. Petrosyan, A.V. Belousov, A.G. Grebenik, One of the reasons for the appearance of defects is Solution of the Stefan problem during radiation- that when the growth is controlled by a feedback signal conductive heat transfer in the process of growing from a weight sensor, a zone of uncontrollability appears sapphire single crystals by a modified Kyropoulos and a random change in the crystallization temperature method, International Seminar on Electron Devices regime occurs. In addition, at high temperatures, the Design and Production (SED) (2019) melts of oxides and semiconductors have a different 6. M.N. Oscisk, Complicated heat transfer, 606 (1976) character of chemical interaction with molybdenum or 7. Kh.S. Bagdasarov, High temperature crystallization tungsten screens, crucible material, etc., and since the from melt, 160 (2004) growth of single crystals can occur both in vacuum and 8. S.P. Malyukov, Yu.A. Klunnikova, The calculated in a reducing or neutral gas atmosphere, there is a need model of the distribution of temperature fields in to work out new technological solutions and temperature sapphire single crystals, Proceedings of the IV regimes in the process of growth. International Scientific Conference "JCSS-2011", However, the main reason is of a fundamental nature 259-264 (2011) and, as noted in this article, is associated with a non- 9. I.I. Ignatov, Temperature growth conditions of linear temperature dependence of some physical and corundum as related to crystal perfection, thermodynamic parameters of sapphire. In particular, Theoretical calculation Journal of Crystal Growth, residual thermal stresses arise in the crystal, which under 52, 411-416 (1981)

4 E3S Web of Conferences 124, 05013 (2019) https://doi.org/10.1051/e3sconf/201912405013 SES-2019

10. S.I. Bakholdin, E.V. Galaktionov, E.A. Tropp, Calculation of thermal stresses in profiled single crystals, taking into account the temperature dependence of thermal expansion, Technical Physics Journal, 60, 11 (2010) 11. K. Peter, D. Lagerlof, A.H. Heuer, Slip and Twinning in Sapphire (α‐Al2O3) J.Am, Ceram, 77(2), 385 (1994) 12. V.A. Tatarchenko, V.A. Antonovich, Sustained crystal growth, 240 (1988) 13. E.R. Dobrovinskaya, L.A. Litvinov, V.V. Pishik, Encyclopedia of Sapphire, 508 (2004) 14. W. Bardsley, G. Green, C. Holliday, D. Hurle, Automatic control of Czochralski crystal growth J, Crystal Growth, 16, 277-279 (1972) 15. A.A. Abgaryan, Modeling of temperature and thermoelastic fields in sapphire in three- dimensional curvilinear coordinates, Mat Modeling, 13(8), 20-30 (2001) 16. L.D. Landau, E.M. Lifshits, Electrodynamics of Continuous Media, A Course of Theoretical Physics, 8, 620 (1982)

5