A method for dynamic compensation of grinding flow rate

Abstract

This invention provides a method for dynamic compensation of grinding fluid flow rate. By establishing a predictive model and using historical processing data as a dataset for supervised learning, the parameters of the predictive model are continuously optimized. As the number of training samples increases, the model becomes increasingly accurate, thereby achieving precise control of the grinding fluid flow rate.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, and more specifically to a method for dynamic compensation of grinding flow. Background Technology

[0002] Chemical mechanical polishing (CMP) equipment is used to planarize semiconductor wafers using abrasive slurry and mechanical polishing. With the development of integrated circuit manufacturing technology, CMP equipment needs to be able to control the amount of thin film removed from the wafer surface. During the polishing process, the polishing head picks up the wafer and applies downward pressure, ensuring full contact between the wafer and the polishing pad attached to the polishing disk. Simultaneously, abrasive slurry is injected between the wafer and the polishing disk. The combined effect of etching and polishing removes the thin film from the wafer surface, thus planarizing the wafer.

[0003] Polishing slurry is a crucial element in chemical mechanical planarization (CMP) wafer polishing, directly impacting the wafer's performance after polishing. Different slurry flow rates need to be set for different wafer characteristics to achieve the required removal rate during polishing. The slurry flow rate is a significant factor affecting the removal rate; therefore, ensuring its stability during polishing is essential. In actual polishing processes, the slurry flow rate fluctuates, and while these fluctuations are monitored and trigger alarms, they are not controlled in real time. Summary of the Invention

[0004] This invention provides a dynamic compensation method for grinding flow rate to overcome the problem of poor stability of grinding fluid flow rate in the prior art.

[0005] This invention provides a method for dynamic compensation of grinding flow rate, comprising: step S1: during the process of controlling the flow rate, setting the first training sample point to the Mth training sample point, and any mth training sample point X m Let X be an N-dimensional vector. m =[x m1 x m2 x m3 …x mN ], N is an integer greater than or equal to 2, m is an integer greater than or equal to 1 and less than or equal to M, M is an integer greater than or equal to 2, X m X represents the m-th training sample point; any m-th training sample point X m x in mn Y represents the value of the nth feature parameter that affects the flow rate in the mth training sample point; the actual flow rate output vector corresponding to the first training sample point to the Mth training sample point is Y. o ,Y o =[y o1 y o2 …y oM ], y oiLet i be the actual flow output value of the i-th training sample point, where i is an integer greater than or equal to 1 and less than or equal to M;

[0006] Step S2: Establish the prediction model Y p =Θ×G T Θ=[θ1θ2θ3…θ N The predicted flow output vector of the prediction model is Y. p =[y p1 y p2 …y pM ], y pi Let i be the predicted flow output value for the i-th training sample point, where i is an integer greater than or equal to 1 and less than or equal to M;

[0007] Step S3: Establish the cost function D(Θ), where D(Θ) represents the difference between the predicted flow output vector and the actual flow output vector of the prediction model; the cost function established for the observation point has the following first expression: The cost function established for non-observation points has a second expression as D(Θ)=H(Θ); H(Θ) is the estimated value of D(Θ);

[0008] Step S4: In the actual flow settings, set the first observation point to the Rth observation point, as well as non-observation points. Any rth observation point is represented as (Θ r J r (Θ r ), r is an integer greater than or equal to 1 and less than or equal to R, R is an integer greater than or equal to 2, and any r-th observation point satisfies the first expression; non-observation points (Θ g H(Θ) g The expression satisfies the second expression.

[0009] Step S5: Obtain the covariance kernel function K corresponding to the first observation point to the Rth observation point. R , k(Θ j ,Θ s ) represents Θ j and Θ s The distance weights are defined as follows: j is an integer greater than or equal to 1 and less than or equal to R, and s is an integer greater than or equal to 1 and less than or equal to R; w = {σ f , l} are hyperparameters;

[0010] Step S6: Obtain The value of w corresponding to the maximum value of ln(p(G, w)) is w'={σ f ',l'};

[0011] Step S7: The value of w is w' = {σ}f When ',l'}, obtain H(Θ) g The estimated value is

[0012]

[0013] Step S8: Obtain H(Θ) g After the estimated value of ), at the non-observation point (Θ) g H(Θ) g Set up several observation points near H(Θ), and repeat steps S4 to S7 until H(Θ) is obtained. g The Θ corresponding to the minimum estimated value of ) g Θ g =[θ g1 θ g2 θ g3 …θ gN ];

[0014] Step S9: Establish actual sample points X' = [x'1, x'2, x'3…x' N ],x' n The value of the nth feature parameter in the actual sample points is represented by x'1, which is set as the feature value for controlling the flow input.

[0015] Step S10: According to the formula x'1=[S-(θ) g2 *x'2…+θ gN *x' N )] / θ g1 Get x'1, S is the target flow value, x'2, x'3…x' N These are the values ​​of the characteristic parameters collected during the grinding process.

[0016] Optionally, the method for obtaining the first training sample point to the Mth training sample point includes: obtaining the first initial training sample point to the Mth initial training sample point, and any mth initial training sample point X. cm =[x cm1 x cm2 x cm3 …x cmN N is an integer greater than or equal to 2, and m is an integer greater than or equal to 1 and less than or equal to M; any m-th initial training sample point X cm x in cmn X represents the value of the nth feature parameter in the mth initial training sample point; for any mth initial training sample point X cm Normalization is performed to form the m-th training sample point. For any m-th initial training sample point X... cm The normalization process is as follows Min(x cm1 ....x cmN) represents X cm The minimum value among the first to Nth feature parameters, Max(x) cm1 ....x cmN ) represents X cm The maximum value among the first to Nth feature parameters.

[0017] Optionally, obtain the first to the Mth initial training sample points, and any mth initial training sample point X. cm =[x cm1 x cm2 x cm3 …x cmN ], N is an integer greater than or equal to 2, m is an integer greater than or equal to 1 and less than or equal to M; set x mn Set to x cmn .

[0018] Optionally, x'1 is updated after each feature grinding time.

[0019] Optionally, the value of M can be between 9999 and 999999.

[0020] The technical solution of the present invention has the following beneficial effects:

[0021] The grinding fluid flow dynamic compensation method provided by this invention establishes a predictive model and uses historical processing data as a dataset for supervised learning, thereby continuously optimizing the parameters of the predictive model. As the number of training samples increases, the model becomes increasingly accurate, thus achieving the goal of precisely controlling the grinding fluid flow rate. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a dynamic compensation method for grinding flow. Detailed Implementation

[0024] The surface of a wafer is uneven during manufacturing, thus requiring chemical mechanical polishing (CMP) for planarization. CMP slurry is a key factor in ensuring polishing performance; it is a mixture of polishing materials and chemical additives, typically composed of ultrafine solid particles (such as nano-sized SiO2 and Al2O3 particles), surfactants, stabilizers, and oxidants. The solid particles provide the polishing action, while the oxidant particles provide the etching and dissolving action. Factors affecting the removal rate of the slurry include: flow rate, chemical composition, concentration, viscosity, and the type, size, and shape of the polishing particles. During the polishing process, the slurry sits between the wafer and the polishing pad, combined with the oscillating and rolling action of the dresser, ensuring wafer removal rate and intra-wafer uniformity. With advancements in chip manufacturing technology, the requirements for the performance of polishing slurries are becoming increasingly stringent. The flow rate stability of the slurry is crucial for ensuring repeatability and consistency between wafers and between batches.

[0025] In CMP polishing, the polishing slurry flow rate is controlled by a flow valve to supply the slurry to the pump. The actual slurry flow rate fluctuates significantly during polishing, necessitating real-time monitoring of these fluctuations and triggering an alarm when the fluctuation exceeds a set value. Currently, the fluctuation range of the polishing slurry cannot be reduced in real time; it can only be passively monitored. With the development of smaller feature size processes, more stringent requirements have been placed on the control of polishing slurry flow rate. Therefore, traditional flow control methods are insufficient for manufacturing processes, necessitating a more refined control method to improve flow stability.

[0026] This invention provides a dynamic compensation method for grinding fluid flow rate, aiming to reduce the variance of the actual grinding fluid flow rate, making the actual flow rate more stably close to the set value with smaller fluctuations. This invention establishes a model and uses historical processing data as a dataset for supervised learning, thereby continuously optimizing the parameters of the control model. As the number of training samples increases, the model becomes increasingly accurate, thus achieving precise control of the grinding fluid flow rate.

[0027] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0029] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can also refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0030] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0031] One embodiment of the present invention provides a method for dynamic compensation of grinding flow rate, referencing Figure 1 This includes the following steps:

[0032] Step S1: During the flow control process, set the first training sample point to the Mth training sample point, and any mth training sample point X m Let X be an N-dimensional vector. m =[x m1 x m2 x m3 …x mN ], N is an integer greater than or equal to 2, m is an integer greater than or equal to 1 and less than or equal to M, M is an integer greater than or equal to 2, X m X represents the m-th training sample point; any m-th training sample point X m x in mn Y represents the value of the nth feature parameter that affects the flow rate in the mth training sample point; the actual flow rate output vector corresponding to the first training sample point to the Mth training sample point is Y. o ,Y o =[y o1 y o2 …y oM ], y oiLet i be the actual flow output value of the i-th training sample point, where i is an integer greater than or equal to 1 and less than or equal to M;

[0033] Step S2: Establish the prediction model Y p =Θ×G T Θ=[θ1θ2θ3…θ N The predicted flow output vector of the prediction model is Y. p =[y p1 y p2 …y pM ], y pi Let i be the predicted flow output value for the i-th training sample point, where i is an integer greater than or equal to 1 and less than or equal to M;

[0034] Step S3: Establish the cost function D(Θ), where D(Θ) represents the difference between the predicted flow output vector and the actual flow output vector of the prediction model; the cost function established for the observation point has the following first expression: The cost function established for non-observation points has a second expression as D(Θ)=H(Θ); H(Θ) is the estimated value of D(Θ);

[0035] Step S4: In the actual flow settings, set the first observation point to the Rth observation point, as well as non-observation points. Any rth observation point is represented as (Θ r J r (Θ r R is an integer greater than or equal to 2, r is an integer greater than or equal to 1 and less than or equal to R, and any r-th observation point satisfies the first expression; non-observation points (Θ) g H(Θ) g The expression satisfies the second expression.

[0036] Step S5: Obtain the covariance kernel function K corresponding to the first observation point to the Rth observation point. R , k(Θ j ,Θ s ) represents Θ j and Θ s The distance weights are defined as follows: j is an integer greater than or equal to 1 and less than or equal to R, and s is an integer greater than or equal to 1 and less than or equal to R; w = {σ f , l} are hyperparameters;

[0037] Step S6: Obtain The value of w corresponding to the maximum value of ln(p(G, w)) is w'={σ f ',l'};

[0038] Step S7: The value of w is w' = {σ}f When ',l'}, obtain H(Θ) g The estimated value is

[0039] Step S8: Obtain H(Θ) g After the estimated value of ), at the non-observation point (Θ) g H(Θ) g Set up several observation points near H(Θ), and repeat steps S4 to S7 until H(Θ) is obtained. g The Θ corresponding to the minimum estimated value of ) g Θ g =[θ g1 θ g2 θ g3 …θ gN ];

[0040] Step S9: Establish actual sample points X' = [x'1, x'2, x'3…x' N ],x' n The value of the nth feature parameter in the actual sample points is represented by x'1, which is set as the feature value for controlling the flow input.

[0041] Step S10: According to the formula x'1=[S-(θ) g2 *x'2…+θ gN *x' N )] / θ g1 Get x'1, S is the target flow value, x'2, x'3…x' N These are the values ​​of the characteristic parameters collected during the grinding process.

[0042] Many factors influence the flow rate of the polishing slurry, including the response speed of the flow valve, the stability of the flow pump, the set value of the flow recipe, and the volume of the polishing slurry. In controlling the flow rate of the polishing slurry, several characteristic parameters are selected. These characteristic parameters can be, for example, the target flow rate value, the difference between the actual flow rate and the target flow rate value, or (actual flow rate - target flow rate value) / actual flow rate. These characteristic parameters include the first to the Nth characteristic parameter; N is an integer greater than or equal to 2.

[0043] During the flow control process, the first training sample point to the Mth training sample point are set, and any mth training sample point X... m Let X be an N-dimensional vector. m =[x m1 x m2 x m3 …x mN ], N is an integer greater than or equal to 2, m is an integer greater than or equal to 1 and less than or equal to M, M is an integer greater than or equal to 2, X mLet X represent the m-th training sample point. Any m-th training sample point X... m x in mn This represents the value of the nth feature parameter that affects the flow rate in the mth training sample point.

[0044] A larger value for M is more beneficial for establishing the prediction model. However, if M is too large, there will be too much data and too much computation. In a preferred embodiment, the value of M is between 9999 and 999999.

[0045] The first training sample point to the Mth training sample point are represented by the sample matrix G, which is an M-row * N-column matrix.

[0046] Sample matrix

[0047] The actual flow output vector corresponding to the first training sample point to the Mth training sample point is Y. o ,Y o =[y o1 y o2 …y oM ], where Y o Let y be an M-dimensional vector. oi Let i be the actual flow output value of the i-th training sample point, where i is an integer greater than or equal to 1 and less than or equal to M.

[0048] Establish a prediction model Y p =Θ×G T Θ = [θ1θ2θ3…θ] N The predicted flow output vector of the prediction model is Y. p Y p Y is an M-dimensional vector. p =[y p1 y p2 …y pM ], y pi G represents the predicted flow output value for the i-th training sample point. T This represents the transpose of G. Y in the prediction model. p and G T The relationship is linear.

[0049] A cost function D(Θ) is established, where D(Θ) represents the difference between the predicted flow output vector and the actual flow output vector of the prediction model. This function is used to train Θ, minimizing the difference between the predicted and actual flow output vectors. In one embodiment, the cost function established for the observation point has the following first expression: Where i is an integer greater than or equal to 1 and less than or equal to M. For ||y 0i -ΘG TThe 2-norm of ||. The independent variable is Θ, and the dependent variable is J(Θ). J(Θ) is the difference between the predicted flow output vector and the actual flow output vector of the prediction model for the observation point.

[0050] The methods for obtaining the first initial training sample point to the Mth initial training sample point include: obtaining the first initial training sample point to the Mth initial training sample point, and any mth initial training sample point X. cm =[x cm1 x cm2 x cm3 …x cmN N is an integer greater than or equal to 2, and m is an integer greater than or equal to 1 and less than or equal to M; any m-th initial training sample point X cm x in cmn X represents the value of the nth feature parameter in the mth initial training sample point; for any mth initial training sample point X cm Normalization is performed to form the m-th training sample point. For any m-th initial training sample point X... cm The normalization process is as follows Min(x cm1 ....x cmN ) represents X cm The minimum value among the first to Nth feature parameters. Max(x) cm1 ....x cmN ) represents X cm The maximum value among the first to Nth feature parameters.

[0051] In another embodiment, the first initial training sample points to the Mth initial training sample points are obtained, and any mth initial training sample point X cm =[x cm1 x cm2 x cm3 …x cmN N is an integer greater than or equal to 2, and m is an integer greater than or equal to 1 and less than or equal to M; it is not necessary to initialize any m-th initial training sample point X. cm Normalization is performed on x mn =x cmn .

[0052] In the actual flow settings, R observation points are set, including the first observation point to the Rth observation point. Any r-th observation point is represented as (Θ). r J r (Θ r )), where r is an integer greater than or equal to 1 and less than or equal to R. Where Θ r and J r (Θ r)satisfy That is, when Θ = Θ r When J(Θ) equals J r (Θ r ). Θ r =[θ r1 θ r2 θ r3 …θ rN ].

[0053] For unobserved points, the cost function has a second expression: D(Θ) = H(Θ). Let the unobserved point (Θ) be defined. g H(Θ) g When Θ = Θ g At that time, D(Θ)=H(Θ) g ). Θ g =[θ g1 θ g2 θ g3 …θ gN ]

[0054] Both the unobserved points and the corresponding values ​​of D(Θ) conform to a Gaussian distribution. The Gaussian distribution for the unobserved points is expressed as f = N(u f , σ f ), u f H(Θ) g The mean of D(Θ) when it conforms to a Gaussian distribution, H(Θ) is the estimated value of D(Θ), and σ f It is H(Θ) g The variance of the Gaussian curve when it conforms to a Gaussian distribution.

[0055] Obtain the covariance kernel function K corresponding to the first observation point to the Rth observation point. R Represented as:

[0056]

[0057] K R Let be a matrix, where, k(Θ j ,Θ s ) represents Θ j and Θ s The distance weights are defined as follows: j is an integer greater than or equal to 1 and less than or equal to R, and s is an integer greater than or equal to 1 and less than or equal to R. w = {σ} f , l} are hyperparameters.

[0058] For the J1(Θ1) corresponding to the first observation point to the J1(Θ1) corresponding to the Rth observation point R (Θ R ), and H(Θ) corresponding to any non-observed point gThe conditional probability density of ) follows a Gaussian distribution, and the conditional probability density p(H(Θ) is given by the following formula. g )|Q)=N(K * K -1 Q,K ** -K * K -1 K T * ), Q is a column vector.

[0059] Solve for w using the maximum likelihood estimation. The formula is used to calculate the value of w corresponding to the maximum value of ln(p(G, w)). The value of w corresponding to the maximum value of ln(p(G, w)) is w'={σ f The methods for finding w include quadratic programming or grid search. The value of w is w'={σ}. f When ',l'}, obtain H(Θ) g The estimated value is K. * K -1 Q.

[0060] Get H(Θ) g After the estimated value of ), at the non-observation point (Θ) g H(Θ) g Set up several observation points near the observation point (Θ), repeat steps S4 to S7, and continue to update the non-observation points (Θ). g H(Θ) g )), until H(Θ) is obtained g The Θ corresponding to the minimum estimated value of ) g Θ g =[θ g1 θ g2 θ g3 …θ gN ].

[0061] In actual flow control, establish actual sample points X' = [x'1, x'2, x'3…x'] N ], where x' n This represents the value of the nth feature parameter in the actual sample points. x'1 is set as the feature value of the control flow input, meaning x'1 represents the control flow input value.

[0062] Correspondingly, at the m-th training sample point X m In the middle, x m1 This represents the feature value of the control flow input in the m-th training sample point.

[0063] x'1=[S-(θ g2 *x'2…+θ gN *x' N )] / θg1 Set the target flow value to S. x'2, x'3…x' N These are the values ​​of the characteristic parameters collected during the grinding process.

[0064] x'1 is updated every feature grinding time. The grinding time can be set as needed, for example, from 8 to 12 seconds, such as 10 seconds.

[0065] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for dynamic compensation of grinding flow rate, characterized in that, include: Step S1: During the flow control process, set the first training sample point to the Mth training sample point, and any mth training sample point X m Let X be an N-dimensional vector. m =[x m1 x m2 x m3 … x mN ], N is an integer greater than or equal to 2, m is an integer greater than or equal to 1 and less than or equal to M, M is an integer greater than or equal to 2, X m Represents the m-th training sample point; Any m-th training sample point X m x in mn Y represents the value of the nth feature parameter that affects the flow rate in the mth training sample point; the actual flow rate output vector corresponding to the first training sample point to the Mth training sample point is Y. o ,Y o =[y o1 y o2 … y oM ], y oi Let i be the actual flow output value of the i-th training sample point, where i is an integer greater than or equal to 1 and less than or equal to M; Step S2: Build a prediction model Θ=[θ1 θ2 θ3 …θ N The predicted flow output vector of the prediction model is Y. p = [y p1 y p2 … y pM ], y pi Let be the predicted flow output value for the i-th training sample point, where i is an integer greater than or equal to 1 and less than or equal to M. Represents the sample matrix Transpose of; Step S3: Establish the cost function D(Θ), where D(Θ) represents the difference between the predicted flow output vector and the actual flow output vector of the prediction model; The cost function established for the observation points has the following first expression: The cost function established for non-observation points has a second expression: D(Θ) = H(Θ); H(Θ) is the estimated value of D(Θ). This represents the difference between the predicted flow output vector and the actual flow output vector of the prediction model for the observation point. Step S4: In the actual flow settings, set the first observation point to the Rth observation point, as well as non-observation points. Any rth observation point is represented as (Θ... r J r (Θ) r (), r is an integer greater than or equal to 1 and less than or equal to R, and any r-th observation point satisfies the first expression; non-observation points (Θ) g H(Θ) g The expression satisfies the second expression. Represents the g-th non-observation point, when hour, ; Step S5: Obtain the covariance kernel function corresponding to the first observation point to the Rth observation point. , , , express and The distance weights are defined as follows: j is an integer greater than or equal to 1 and less than or equal to R, and s is an integer greater than or equal to 1 and less than or equal to R. For hyperparameters, This is expressed as the length-scale hyperparameter of the covariance kernel function. The signal variance hyperparameter represents the covariance kernel function; Step S6: Obtain middle The value of w corresponding to the maximum value. The value of w corresponding to the maximum value is ; Step S7: The value of w is... At that time, obtain The estimated value is , , Represents matrix multiplication; Step S8: Obtain After the estimated value, at the non-observation point (Θ g H(Θ) g Set up several observation points near the location, and repeat steps S4 to S7 until the data is obtained. The Θ corresponding to the minimum estimated value g Θ g =[θ g1 θ g2 θ g3 …θ gN ] ; Step S9: Establish actual sample point X ’ =[x ’ 1,x ’ 2,x ’ 3 … x ’ N ], x ’ n The value of the nth feature parameter in the actual sample points is represented by x. ’ 1 represents the characteristic value for controlling the flow input; Step S10: According to the formula Get x ’ 1. S represents the target flow rate, x ’ 2,x ’ 3 … x ’ N These are the values ​​of the characteristic parameters collected during the grinding process. This represents matrix multiplication.

2. The dynamic compensation method for grinding flow rate according to claim 1, characterized in that, The methods for obtaining the first initial training sample point to the Mth initial training sample point include: obtaining the first initial training sample point to the Mth initial training sample point, and any mth initial training sample point X. cm =[x c m1 x c m2 x cm3 … x cmN N is an integer greater than or equal to 2, and m is an integer greater than or equal to 1 and less than or equal to M; any m-th initial training sample point X cm x in cmn X represents the value of the nth feature parameter in the mth initial training sample point; for any mth initial training sample point X cm Normalization is performed to form the m-th training sample point. For any m-th initial training sample point X... cm The normalization process is as follows , Represents X cm The minimum value among the first to Nth characteristic parameters. Represents X cm The maximum value among the first to Nth feature parameters.

3. The dynamic compensation method for grinding flow rate according to claim 1, characterized in that, Obtain the first to the Mth initial training sample points, and any mth initial training sample point X cm =[x c m1 x c m2 x cm3 … x cmN ], N is an integer greater than or equal to 2, m is an integer greater than or equal to 1 and less than or equal to M; x mn Set to x cmn .

4. The dynamic compensation method for grinding flow rate according to claim 1, characterized in that, Update x after each feature grinding time. ’ 1.

5. The dynamic compensation method for grinding flow rate according to claim 1, characterized in that, The value of M ranges from 9999 to 999999.

Images