SU1745780A1 - Method and apparatus for controlling single crystal growth process under protective liquid - Google Patents

Abstract

The use1 of the invention relates to the field of automatic growth of single crystals in plants with a gravity control method. SUBSTANCE: method consists in regulating the diameter of a grown grow. The invention relates to growing single crystals under a protective Czochralski method and can be used to control the crystallization process on growth plants with a weighted method of control. is to improve the quality of a grown single crystal by controlling temperatures the melt and the rate of mono-crystal drawing by the deviation of the rate of change in the weight of the crystal from a given value — at the single-crystal growing section, and by state — at the single-crystal diameter stabilization section, which is used to restore variable states using the crystallization process model . In addition, during the whole process, the correction of the influence of the protective fluid on the weight signal is made. The device performing the control method provides for the restoration of variable states — the single crystal radius and the height of the melt meniscus, in the observer of Lüenborger; using parametric controllers and state controllers. 2 sec. and 1 z. p. f-ly, 5 silt, lov due to an increase in the accuracy of the diameter control. FIG. Figure 1 shows a diagram explaining the features of the process of growing single crystals under a protective liquid by the Czochralski method in the cone formation mode; in fig. 2 - the same, in the mode of stabilization of diameter; in fig. 3 shows a control device; FIG. 4 shows a device for two-channel control of the temperature and the rate of crystal drawing; in fig. 5 - functional block diagram of the observer. 2 SP XI 00 About

Description

FIG. 1-2 the following notation is accepted: r (t) is the current radius of the crystal, r0 is the specified diameter of the cylindrical crystal, R is the radius of the crucibles, rh is the radius of the seed, V0 is the rate of crystal extrusion, M (t), m (t) , / and (t) are the melt masses in the crucible with density p, crystal density p and meniscus density p, respectively, at the time of measurement, Mf is constant flux mass with density pr, Hf (t) is the thickness of the flux, measured from the crystallization front to the upper boundary of the flux, h (t) is the current height of the meniscus, h0 is the stationary height of the meniscus corresponding to the cylindrical for a crystal with a diameter of 2r0, H (t) is the height of the melt in the crucible from the bottom of the crucible to the base of the meniscus, / (t), and (m) the angles of inclination of the side surface and meniscus relative to the vertical, Ј is the constant angle of growth, T3 ( t) is a variable lag time characterizing the formation time of the crystal cross section leaving the flux layer at the time of measurement, f (t) is the current length of the crystal being drawn

The method is implemented on an automated growth facility equipped with a weight sensor, setters and controls for heating power and draw rate, a computer together with an interface module. The adjustment of the crystal diameter according to the method is carried out by controlling the heating temperature and the drawing speed of the single crystal according to the deviation of the rate of change of the measured weight of the single crystal and the meniscus associated with it from a known predetermined value during expansion. Upon completion of the formation of the initial cone, the stabilization of the diameter of the grown single crystal is carried out by adjusting the state using a state variable, such as the deviation of the radius of the crystal and the height of the meniscus, from the known stationary values, using the observer.

An initial charge, for example, gallium arsenide and boron oxide powder is placed in the crucible of the growth installation. After the melting and formation of a layer of protective boron oxide liquid, manual seeding is carried out and the automatic mode of adjustment of the diameter of the grown single crystal is activated.

The weight is fixed by the sensor at any time.

P (t) (t) + / (t) - Fa (t) + const, (1) where g is a gravitational constant;

The constant characterizes the constant weight of the snap-holder crystal holder.

ten

15

20

25

thirty

35

40

45

50

Archimedes' buoyancy is defined by

Fa (t) -gi / (mfft) + / "(t)), (2)

where u const;

mt (t) is the mass of the crystal under the flux layer by the time t.

If m (t) mt (t), in other words, the crystal has not left the flux, then (1) and (2) imply

P (t) g (1-J7) M (t), (3)

where M (t) mf (t) +/ and (t) in accordance with the law of conservation of mass. If m (t) is greater than m $ (t), i.e. the crystal comes out of the flux, from (1) it follows

P (t) (t) + / g (t) -1, (mf (t) + p (t))

- -g (1 - J) M (t) + g 17 m {t - r3) (4) where M (t) m (t) + / i (t);

mf (t) m (t) - m (t - Гз).

Equations (G), (4) characterize the structure of the real signal P (t) of the weight sensor used for comparison with the program task, differentiable, and also remembered for the time Tc.

With the weight method of monitoring the current state of a dynamic system, both the observed value and the program value are the derivative of the weight. Thus, the software task must be calculated as

Kt) -gd-) M (t), (5)

P (t) -g (1 - rj) M (t) + g t) - r3 (t) (6)

Hereinafter, the symbol l denotes program values.

Formula (5) is correct until the cone leaves the flux, and (6) more generally, until the cone finally emerges from the flux

Extrusion with a constant crystal speed of any shape other than a straight circular cylinder leads to a non-linear change in time of its radius r (t) and weight. When calculating the software task, you can directly use the results based on the use of a suitable growth angle function / 3 (g). The final formulas for the variable crystallization rate Vc (t) and the rate of change of the melt level in crucible X (1), included in the expressions m (t) lr r2 (t) Vc (t) and M (t) - pr R2H (t) ,

v, .. frpft V.

V ry rzgCH gm Mm IDGCR

H (t) - V0 - V0 - (hr1 + h1 fr) Vctg D (8) The algorithm of the program assignment during the transition to the discrete time T0 can be represented by the following recurrent calculation.

At the first step, we take p gz, I 1, k 1 (the definition of the counter k is given below), at the l-th step, by the given dependence / 3 (g) we find / 3i / (n), using the known formulas 4.5 calculate the current height meniscus: hi (n,), calculate the current crystallization rate of HGs Vc (n,), calculate the next radius of the crystal:

P + 1-GZ + That At Vcitg #,

the length of the grown crystal h T0 2 Vci,

i

current mass crystallization rate MI ptg / e | R Hi, the total increase in the weighed mass MI T0 Y Mj, the current mass

meniscus / l pr n g hi (n,) + pr a2 cos- (Ј + $) calculate and memorize the mass of the grown crystal on m3: (n Mi - fi.

Since in this method of crystallization, it is fundamentally important to determine the delay time TL (T.), depending on the variable height of the flux Hf (t) with a known initial load Mf, before completing the description of the calculation of the program assignment, we give the formula for calculating the delay time.

To find 1 Hf (counted from the crystallization front), consider the total volume in the crucible of a known radius R under the flux (Fig. 1). Since before the release of the seed crystal from under the flux, the ratio

l (t) + h (t) W, (t) + VMt) +

+ Wfo + W3, (9)

where W fi (t) // oi is the volume of the meniscus:

Wmf ITU / PS is the volume of a crystal grown by the value of l (t) / Vc (t) dt;

 about

Wfo Mf // 9f const - flux volume;

Wa - (t) - the volume of the seed of the radius gz,

under the flux, equation (9) taking into account the condition of the mass balance gives

Hf (t)

one

(P2gs3) 1

mЈ) + 11 g l

I A / 01 L Pb

+ Wfo-Jrr32l (t) - ГС2Ь. (10) -J R - gz

At the moment when the cone apex enters the gas phase, l (t) Hf (t) and L / z 0, from where

n, (,) - 1-G +2 G1-fi.il jrR2 L L -IJ

+ Hf0-h (t), (11)

+ Hfo - h (t), where Hf0 Mf / jrR2 pi.

Hf (t)

No

With the withdrawal of a part of the crystal from under the flux (Fig. 1) in (6), it is necessary to take into account the previously calculated calculated values of mtft - G3):

1 Hm (0-tgE-ta), (0,

tfR2LL

j + Hfc-hft). (12)

Thus, having taken the moment of chasing in the reference point, it is necessary to calculate and memorize the length l (t) and the mass

growing crystal Ch. J m (t) dt. If on

about

In the next calculation step, the condition i (t) Hf (t) is fulfilled; in other words, the crystal begins to emerge from the flux required by the memorized values and Eq. (12) to restore the rh (t). The condition for the correct determination of rh in the program specification (6) is l (t) - l (t - t3) Hf (t).

We continue the algorithm of recurrent calculation of the program task.

Calculate the thickness of the flux Nl:

Ng, a-for Hfi I, - according to the formula; (101 Oe - for Hfi 1 | according to the formula: (11) B - for Hp h according to the formula (12)

0

five

0

five

0

five

The software task before the condition is satisfied, in other words, before the cone leaves the flux, is calculated by the formula f./h

Pi g (1 - q) Mi

pr the condition in the formula

P, g (1) Mi + g J / mi-k, (13)

where k gz / then

Here, the counter k of discrete samples of time T0 determines the length of the memorized data array. If the value of Gz found is not equal to an integer number of sampling cycles T0, then in our memorized array there is no value m (t - gz |) corresponding to the calculated delay time. Therefore, it is necessary to find the two nearest memorized values of the crystal mass, corresponding to the whole of the delay time rail and 1, and then to find some average value between m () and m (-1). The way of finding the average in the general case can be different, but in the approximation of the linear variation of m (t) between adjacent discretization cycles, the desired value is defined as mi (t - gsO (1 - c) m () + Е m), where Ј I li- Hf,, / VoTo - determines the weights when calculating the average value of m, (t-gzO.

Equation (1) defines the structure of the real signal of the weight sensor. Accordingly, the nonlinear program reference (13) calculated using the dependences given above must coincide with the real weight signal for time t t IT0. By subtracting the software non-linear signal (13) from the real signal of the weighing sensor and subsequent differentiation, an output deviation signal is used to regulate the current crystal diameter using a parametrically optimized PID controller during spreading to reach the specified crystal diameter.

The choice of the optimal structure and tuning of the PID controller (in the general case of a two-channel one) is determined by the crystallization dynamics model and the observation model with weight control. The small-scale mathematical model for the Czochralski method with liquid sealing is based on the use of three conservation laws: i) heat balance at the crystallization front, I) save the mass of the substance to crystallize, and Hi) save the growth angle1

AiGi (t) -AsGs (t) psl-vc (t); M (t) + m (t) + «(t) 0; r ((t) tg / 5 (t),

(14)

where A (, Ae - thermal conductivity of the melt and crystal;

G |, Gs are the temperature gradients in the meniscus and the crystal near the interface;

PS is the density of the solid phase;

L is the heat of the azov transition;

Vc (t) V0 - H (t) - h (t) is the crystallization rate;

M (t), rh (t), / i (t) are the rates of change of mass of the crystal, the meniscus of the melt and the main part of the melt in the crucible, respectively.

f (t) is the rate of change of the crystal radius;

/ a - e is the angle of inclination of the isotropic side surface of the crystal with the vertical;

e is the growth angle (Fig. 1).

Linearization of these three conservation laws (14), taking into account the variable angle of inclination of the lateral surface of the crystal / 3 (g) O (Fig. 1), makes it possible to describe the dynamics of non-stationary crystallization by a system of non-autonomous differential equations:

0

d g (t) / Vr (t) J r (t) + An (h (t) + Afi (t) 6 h (t) + A1H (t) d H (t) + Av (t) 6 V0 ; 5 ft (0 / W (t) 6 r (t) + Ai (t) d h (t) - -6 H (t) + 6V0 (t) + AM (t) dT (t); 0H (t ) / Wft) dr (t) + AHr (t) ar (t) + + AHfi (t) (5h (t) + AHV (t) 6Vo (t). F15)

As natural state variables, the variations (deviations) of the crystal radius r r (t), the height of the meniscus d h (t) and the melt level in the crucible H (t) with respect to the known program values are considered. Only two state variables xi 5g and xi oh can be

5 selected as independent, a b T # 1 and (5 Vi - as components of the input perturbation vector Q (t), so that in the matrix notation the model of the crystallization process has the form

GHL, Hz 012 wil / w xaj + b21 b22J uuj {16)

or more compact

X (t) A (t) X (t) + (B (t) Q (t). (17)

The effect of control on the dynamics of transient processes is taken into account by adding in (17) the control vector U Ui, which has the required

control Ut AT (t) and Ua AV0 (t) in T- and V-channels of feedback.

X (t) A (t) X (t) + (B (t) (Q (t) - U (t)) (18) A non-autonomic model of small dimension, defined by equations (18), characterizes the dynamics of transients during small deviations from the initial state during non-stationary crystallization, for example, with an increase in the initial or final cones. In this case, the formation mode of the required crystal geometry can be called the non-linear software task tracking mode.

An autonomous model of the crystallization process with constant coefficients

ass arises in the study of the problem of stabilizing the diameter of a cylindrical crystal with an initial circular cross section of radius G0. In this case, the slope of the side surface / 8 (m.) 0 and the steady-state crystallization rate is equal to VCo V0 - H0 (t), where H (t) -ps r02V0 / o R2 - / 9s r02); R is the crucible radius. A linear autonomous model for stabilization mode is written

Ghl Gai ai fxil, fb O 1Gil. .

X2j 312322 X2J 021 D22J UJ2J

or more compact in the form

X (t) AX (t) + B u (t). (20)

l

Output deviation Y (t) P (t) - P (t)

can be represented in a standard matrix form using state variables

Y (t) Yi (t) + Y2 (t-gz) CiX (t) + D i Q (t) + C2X (t-Gz) + D2 Q (t-Tz) (21) This equation is a model monitoring with weight control.

Parametrically optimized PID controllers use information only about the deviations of the sensor output signal from the program task. The simplicity of introducing integral components of the law of regulation allows one to ensure the required degree of astatism with such regulators, the stability margin of a closed system in the usual Czochralian method. Increasing the degree of astatism is primarily required to ensure accurate tracking of the non-linear program task . Obviously, the control synthesis problem for the conventional Czochralski method almost coincides for the Czochralski method with liquid sealing if the formation of a cone occurs completely under flux. In this case, the observation equation (21) is simplified and is defined as

Y (t) Yi (t) Ci (t) X (t) + Di (t) Q (t). (22) In equations (21) and (22), in contrast to the usual Czochralski method, the previously known multiplier (1 - /) appears. When analyzing a closed state, the vector of input actions Q (t) must be replaced by the vector Q (t) - U (t) where U (t) is the control vector defined above Due to the slow change in the model parameters (16), (22) with The analysis of the crystallization dynamics applies the approximation of the frozen coefficients, which makes it possible to assume that the components of the matrices are constant over sufficiently long times for the formation of transition regions of the crystal. Going over to the Laplace images in the model control (16) and in the observation equation (22), we obtain the matrix transfer functions of the object in a closed state;

X (s) (sl-Ay1B Q (s) - U (s) + (sl-Ay O); Y (s) C (sl-A) TB + (s) -U (s) + C (s1- A) 1X (0), (23)

where I is the identity matrix (2 2).

Non-zero initial conditions X (0) O and also Q & 0 can be considered as initial deviations (disturbances) and input disturbances of the object. As before, the thermal inertia of the response of the object to controlled changes in heating power

AN (t) (inertia of temperature T-channeling) will be taken into account by introducing an additional transfer function

Ui (s) WT (s) AN (s) (24)

For example, for a typical temperature response of an object to a step change in power, D N (t) AN01 (t), approximated by the function AT (t) ku AM0 (1-exp (-at)), where a is the decay damping, has WT (S) aku / (s + a). Further, under the control of ui (t), it is necessary to consider the function defined by equation (24)

The correction of thermal inertia is reduced to the introduction of short forcing effects on the heating power of the crucible within one control cycle.

In dual-channel control, the control of the diameter of the grown single crystal during the spreading is carried out by controlling the melt temperature and the stretching speed of the single crystal, and the control effect on the draw rate both at the growing stage and at the diameter stabilization stage is proportional to the deviation of the rate of change of the drawable single crystal weight from the given, without using in the law of regulation the weight deviation and the second derivative of this measured deviation by weight. The optimally adjusted parametric regulator allows to accurately reproduce the specified geometry of the conical part of the crystal and to achieve a shock-free transition to control of the state in the mode of stabilization of the diameter of the crystal.

Until the final formation of the formed cone from under the flux, control is still carried out using a parametric regulator, with the current variations in the crystal geometry, characterized by the quantities dt (t) and dji (t), which are uniquely associated with the introduced state variables, the known values of the controls U (t) will correspond. The possibility of storing these values, eliminating the error of inaccurate determination of a program task with the beginning of the cylinder exit from the flux, as well as the constancy of the parameters of the control object, allow determining its current state, used to calculate the control for this state.

Control synthesis at the stage of diameter stabilization is based on the use of an equivalent Luenberger type observer that restores current state variables while simultaneously

the use of a memory of finite length, required for the storage and subsequent use of the restored state variables, as well as known control actions. The solution of the problem of synthesizing the optimal regulator according to the state can be found under the following conditions: I) - the autonomous model, defined by equations (19), (21), adequately describes the real object in the diameter stabilization mode, II) - is accurate information about the amount of control that coincides at the entrance to the real object and the observer. The second condition | can be satisfied according to the results of the analysis and correction of the thermal inertia of the object described above.

Consider also the implementation of the method in discrete time, implemented in the control computer. For the sampling time, To neglect the delay for removal and processing of information in a computer, the discrete model of the process is represented as

X (k 4-1) FHM + f (Q (k) - U (k)),

Y (k) СХ (k) + D (Q (k) - U (k)) + (25)

+ Ci X (k - n) 4- Di () - U (k-n))

where n tz (t) / To;

00

F (To) exp (AT0) (ATo) Vv;

f (To) (F-I) 1AB — known numerical values for the selected control object, including the parameters of the growing modes and known material constants;

Claims (3)

I is the identity matrix of the corresponding dimension. Deviations of the system from the specified stationary growth conditions are interpreted by the observer as deviations of the initial conditions X (0) 0 in the next step of information retrieval. When adjusting the controller, these initial perturbations are represented in the class of deterministic perturbations approximated by the step functions X (0) X (k) ConstnpHkT0 t (k-M) T0,. The equivalent observer is represented by the following system of equations: i (k + 1) (k) + fU (k) + UDEG Y (k) С%) ypU (k) × G х (к-тУ + О and (к-m Ae (k)) - y (K), l 4 (26) where X (k), Y (k) are the recovered values state variables and observables: - 0 5- 0 five 0 five Q 5 0 five H is the matrix of the connection between an object and an observer. The delay time in the observer m does not coincide with the real delay time n and turns into a tuned parameter. Considering this fact, the error in restoring the state variables Dx- (k + 1) X (k + 1) - X (k + 1) should be defined as AX (k + 1) -HC) AX (k) t-HCi (X (k-n) - X (k-m)) + HDi (U (k-n) - U (k-m)) (27). The recovery error (27) is determined by the rate of convergence of the observer without accounting delay in the output signal lim ( NA) AX (k) 0 and of the error associated with the choice of discrete delay time m relative to the actual value of discrete time n. The rate of convergence of the observer is determined by the choice of the roots of the characteristic equation det (zl- Ф + НС) 0. (28) According to the separation theorem, the dynamics of a closed state is determined by the set of poles of the characteristic equation of the system det () det (zl-Ф + НС) 0. Moreover, the desired matrices of the observer H and the state regulator U (k) -KX (k) can be found separately . The numerical values of the coefficient matrix H are predetermined from the solution of equation (28). The values of the matrix K, which are further refined during the operation of the regulator, are estimated similarly. The reconstructed values of the variable states are calculated from the known input U (k) and output Y (k) signals of the control object according to the formula A I (k + 1) (F-NA) X (Yu + (f-HD) U (k) - -HCiX (k-m) -HDiU (k-m) + HY (k). X (29) The analysis of the measurement error D e (k) Aei (k) + Ae2 (km) indicates the need to minimize its component D 62 (km) by selecting from the stored values of X (km) and U (km), and the search depth in arrays X and U may be different. Let us consider in more detail the determination of the delay time m as applied to the regime of stabilization of the diameter of a crystal. During the stabilization of the diameter of a single crystal, due to perturbations of the crystallization process, its shape can change randomly and cannot be predicted in advance, unlike the spreading mode, in which its shape and latency are calculated in accordance with the program task. In the proposed method, the time The lag gz and, accordingly, the lag component associated with it in the measured weight signal can be taken into account by using a real-world process observer and memorized variable variables of the state. In the first approximation, the lag time after the transition to state control, i.e. after the cone completely leaves the flux, it can be estimated using the formula TZO Hfo / Vco, where Hfo is the flux thickness during stationary crystallization of a strictly cylindrical crystal with a diameter of 2r0. The calculated stationary height of the flux mass of Mr const, obtained from the solution of the system we have two equations Hf0 Hf0 + h0 and / cf1 Mf  n R 2 Hp - r2, h0 + a2 r0 cos k - l r0Hf0 is equal to Hfo Mf / g / Of (R 2 - r0 2) - a2 g0 cos e / (R2 - - r02) - h0. (30) In the general case, taking into account variations in the flux thickness Hf (Fig. 1.2), counted from the three-phase line to its upper level Hf, caused by perturbations of the meniscus volumes in W / and crystal in Wc, as well as the crystallization rate 5 Vc gz (t) tzo + b gz (t). Variations in the height of the flux about Hf and the delay time of the DTs at the current time can be determined by considering the total volume enclosed between the melt and flux boundaries: W WM + Wmf + Wf0, (31) where Np c / p is the current volume of the meniscus; Wmf mf / PS is the crystal volume under the flux; Wfo Mf // 0f is the constant physical volume of the flux. Insofar as W - W0 + 5 W (t0) W0 + JlR2 (d Hf (to) + + 5h (t0); VV-VW 3w (to) + PTV 5r (to) + + // hdh (to)); (32) for the amendment to the fixed for the given conditions of crystallization of the delay time of the GSO 3Ho) 4p ((to -) (t0-) (Y, °° 3 (34) where A2 1 / tg Vco (R2- r02). Equation (34) can be rewritten using the variable states xi d g, X2 d h and the draw rate control U2 d V0: ЈM4H2c, x, (, y2 (V) -1TR1 J ° U, (35) GD6S1 - 1I9 C2 pi fih - n R - known numerical values. Thus, the variation of the latency is determined by the values of the memorized previous state variables X (i.e. Tc) and the control Ui in the case of using a speed channel, and the reconstructed values of the state variables are used. The transition to the discrete time t T0, where T0 is the sampling time used in the digital control system, can be represented by an algorithm for calculating the current value of g3 (I) as follows. Assume that at the nd time point the values of the variables are known. states and controls, where (i-1), must be calculated for some k value b gz (I) A2 ci Xi (m) + A2 C2 X2 (t) -l R 2 That U2 (36) Oh oh Wmt-PeJ ™ W XVpsJ W. (t) dt VЈ to + J A, 4 ° C (33) where to is the current moment of time of measurement, and to tz (to); Vc Vco + b Vc - variable crystallization rate. From the joint solution (31), (32) it is not difficult to get the final expression 0 five and the whole part of the estimated latency n (k) ((0) / T0 (37) In the general case, the search in k continues until the condition n m (n + 1) is fulfilled. Obviously, this search must be carried out in the vicinity of the values of k i - m0, where m0 is tzo / T0. The refined values of the state variables in (36) must be calculated by the formula 5 (i-k) (1 - B) I (t) + e & (t-1) (38) where e is the real remainder calculated by formula (37). After finding the next value of k, there is no need for further storage of the used values of the variable states and controls. It should also be noted that in the case of using single-channel temperature control, the controlled variations in the draw rate ui 0 and formula (36) are simplified. The control device contains series-connected crystal weight sensor 1, differentiator 2, comparison unit 3, 4 mode switch, parametric controller 5, thermal inertia corrector b, heating power setting device 7, heater 8. crystal weight setting device 9 connected to the second block input comparison, the state regulator 10 connected to the second output of the mode switch parallel to the parametric regulator, a circuit of the series-connected observers 11, the stack memory unit 12, the determination unit 13 rear The inputs of the observer are connected to the outputs of the comparator and regulators 5 and 10, and the output is additionally connected to the second input of the 4 mode switch and to the second input of the delay determining unit, the output of which is in turn connected to the first input of the observer 11. Except In addition, the device may contain additional regulator 14, a parametric regulator 15 connected in parallel with the regulators 5,10. and in series with them are connected a speed setting device 16 with a controlled drive 17 of the extrusion, and the outputs of the controllers 14, 15 are also connected to the input of the observer 11 The device works as follows. In the process of growing, the weight sensor 1 records the weight of the crystal and the associated meniscus of the melt which is together with part of the crystal in the layer of protective liquid. The derivative of the signal from the weight sensor 1 defined in the differentiator 2 enters the comparison unit 3 at the same time as the task generated by the weight setting 9. The deviation signal produced by the comparison unit 3 passes to the observer 11 and the 4 mode switch. By the deviation signal and the current control coming from the outputs of the regulators 5 or 10, the observer 11 restores the current values of the variable states. In block 11, the observer includes adders, well-known matrix elements A, B, C, D, Ci, Di, H, scaling input and output signals and a delay line 18, for one clock cycle of the basic sampling time of the output information from the growth installation (analog of integrator in continuous time in the case of analog execution of the observer) reproduces the dynamic transient process corresponding to the actual transient process in the growth unit. The signals determined by the reconstructed variable state variables, together with the control signal, scaled by the coefficients of the matrix Di, arrive at the input of the stack memory unit 12, where the necessary time, determined by the number of bits of the registers of this block and clock frequency, is stored management The signals from the output of the observer 11 also enter the delay determination unit 13, where signals are generated that are proportional to 0 the length of the grown single crystal and the current thickness of the protective fluid. As a result of their comparison, the register bits of the stack memory 12 are determined, which contain the delayed (stored) 5, the previous values of the variable state and control currently required to correct the effect of the protective fluid. The proportional signal from the output of the block 13 together with the current output signal of the block 3 is fed to the first input of the observer 11. The signals corrected in this way, corresponding to the current values of the variable states, are fed to the second input of the 4 mode switch. The 4 mode switch transmits the deflection signals to it or the signals corresponding to the current state variables either to the parametric controller 5 or, 0 when the crystal reaches the specified diameter, the regulator is 10, respectively. The control formed at the outputs of the regulators 5 or 10 goes to the observer 11 for the subsequent cycle 5 and in the thermal inertia corrector 6. After the corrector 6, the control passes to the heating power setting device 7, which accordingly changes the temperature of the heater 8 to the control signal. In response to the temperature change, the diameter of the single crystal changes in the right direction. Similarly, the change in the rate of drawing of the crystal is carried out in the device, which contains additional controllers 14, 15, the speed setting device 16 and the controlled drawing drive 17. Fo rumula and z oo n i 1. Method of controlling the process of growing single crystals under protective liquid by the Czochralski method, including regulation of the single crystal diameter by controlling the melt temperature by the magnitude of the rate of change of the weight of the grown single crystal from the given one with simultaneous correction of the melt temperature response and stabilization of the single crystal diameter by thermal inertia, and stabilization of the single crystal, characterized in that, in order to improve the quality of the grown single crystals by increasing In addition, during the expansion and stabilization of the diameter, the current values of the single crystal diameter and the height of the meniscus at the crystallization front are calculated, their deviations from the specified values are determined, the control influences are formed on them, the values of the deflections and control influences are corrected, the influence of the protective fluid is corrected the weight of the single crystal is proportional to the deviations obtained and to the values of the control actions and the deviations stored in the previous steps. 2. A method according to claim 2, characterized in that the diameter of the single crystal is adjusted at the stages of expansion and stabilization, and by varying the melt temperature and the drawing rate is proportional to the rate of change of the single crystal weight from its predetermined value. Flux crystal 3. A device for controlling the growth of single crystals under a protective liquid by the Czochralski method, containing a series-connected crystal weight sensor, a differentiator, a comparator unit, a parametric controller, as well as a series of thermal inertia corrector, a generator and a controller of heating power, a generator of crystal weight, connected to the second input of the comparison unit, which differs from the fact that, in order to improve the quality of the grown single crystals by increasing the control accuracy, A state controller, a variable recovery unit (observer), a lag time determining unit, a stack memory unit, and a mode switch are introduced, while the inputs of the variable recovery unit are connected to the outputs of the comparison unit and the regulators, the outputs of the recovery unit of variables are connected the inputs of the stack memory unit, the delay time detection unit and the first input of the mode switch, which is also associated with the output of the delay detection unit, the second input of which is connected with the output of the stack unit memory unit output of comparator coupled to the second input breakers transferred modes, the first and second exits through which the state controller and the feature controller, respectively coupled to the input of the corrector thermal inertia. & RA) .-, / 1745780 6P-0 to a bunch of astanabhv Yabluaate "H Editor N. Gunko FIG. five Compiled by G.Satunkin Tehred M. Morgenthal Order 2366 Circulation VNIIPI State Committee for Inventions and Discoveries at the State Committee on Science and Technology of the USSR 113035, Moscow. Zh-35, Raushska nab., 4/5  -Tr. STSBAY USTHYUBKI BUT Proofreader O. Kundrik Subscription