Semiconductor polishing method for reducing TTV
By establishing a dynamic model of the polishing machine and regressing moving average model and optimizing the polishing parameters, the problem that the polishing process in the existing technology cannot flexibly cope with the changes in the characteristics and state of different workpieces is achieved, more accurate polishing control is achieved, and semiconductor products are improved Quality and stability.
Patent Information
- Application Number
- CN202411707702.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-27
- Publication Date
- 2024-12-31
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing semiconductor polishing process cannot flexibly cope with changes in different workpiece characteristics and polishing machine state, resulting in uncertainty and waste of polishing process, affecting product quality and stability.
By establishing a dynamic model and regression moving average model of the polishing machine, the polishing parameters are predicted, and the polishing parameters are optimized based on the behavior description function and reward mechanism, the adjustment of the optimal polishing parameters is achieved.
Accurately control polishing parameters, reduce semiconductor thickness changes, improve polishing efficiency and stability, and improve product quality and reliability.
Smart Images
Figure CN119217156A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of semiconductor polishing, and in particular relates to a semiconductor polishing method for reducing TTV. Background Art
[0002] Semiconductor polishing is a very important process step in the manufacture of semiconductor devices. Its main purpose is to improve the flatness and smoothness of the silicon wafer surface, thereby providing a better foundation for subsequent lithography, etching and other processes.
[0003] In the semiconductor manufacturing process, TTV refers to the variation in wafer thickness, which can be controlled and reduced through the polishing process. Through a precise polishing process, the thickness variation of the silicon wafer can be made more uniform, thereby reducing TTV. During the polishing process, controlling parameters such as polishing time, choice of polishing agent, and polishing pressure can effectively reduce the thickness variation of the silicon wafer. In addition, the use of high-quality polishing equipment and process flow can also effectively control TTV.
[0004] However, for the existing semiconductor polishing process, the polishing operation is completed by setting fixed polishing parameters. This method has the problem of being unable to flexibly cope with the characteristics of different workpieces and the changes in the state of the polishing machine to a certain extent, resulting in some uncertainty and waste in the polishing process. Therefore, the existing technology often cannot effectively control the polishing parameters to reduce TTV, thus affecting the quality and stability of semiconductor products. Summary of the invention
[0005] In view of the above-mentioned defects of the prior art, the present invention proposes a semiconductor polishing method for reducing TTV. The technical solution steps designed by the present invention include:
[0006] S10: defining polishing parameters of the polishing machine, and establishing a dynamic model of the polishing machine based on the polishing parameters;
[0007] S20: establishing a state description function of the polishing machine based on the dynamic model;
[0008] S30: adopting a regression moving average model, training the regression moving average model based on the state description function and the preset loss function, and outputting the predicted polishing parameters of the polishing machine;
[0009] S40: establishing a behavior description function of the polishing machine based on the predicted polishing parameters, calculating the behavior utility based on the state description and the behavior description function, defining a random polishing parameter scheme, and optimizing the random polishing parameter scheme based on the behavior utility to obtain an adjustment scheme for the polishing parameters;
[0010] S50: defining a reward mechanism, and constructing a polishing parameter optimization function based on the reward mechanism and the polishing parameter adjustment scheme;
[0011] S60: Outputting optimal polishing parameters based on the optimized polishing parameter function, and inputting the optimal polishing parameters into a polishing machine to perform a polishing operation on the semiconductor.
[0012] Preferably, the dynamic model formula for establishing the polishing machine in S10 is as follows:
[0013]
[0014] In the formula, is the amount of semiconductor removed, is the polishing liquid ratio coefficient, The electromagnetic force generated by the polishing machine surface, is the relative speed between the semiconductor and the polishing machine during the polishing process, is the amount of semiconductor dissolved during the polishing process, is the dissolution rate of the semiconductor, is the penetration value of the abrasive into the semiconductor surface, is the rotation speed of the polishing machine, is the area of contact between the polisher and the semiconductor, is the diameter of the abrasive, is the density of the abrasive, is a constant for the material concentration.
[0015] Preferably, the state description function formula of the polishing machine established in S20 is as follows:
[0016] .
[0017] Preferably, the preset loss function formula in S30 is as follows:
[0018]
[0019] In the formula, is the sample size, is the predicted polishing parameters of the polishing machine, is the actual polishing parameter of the polishing machine.
[0020] Preferably, the include , and , one-to-one correspondence ;
[0021] Said correspond , the corresponding parameter is the polishing liquid ratio coefficient , Electromagnetic force generated on the polishing machine surface and the relative speed of the semiconductor and the polisher during polishing ;
[0022] Said correspond , the corresponding parameter is the indentation value of the abrasive into the semiconductor surface , the polishing machine's rotation speed Electromagnetic force generated by the polishing machine surface ;
[0023] Said correspond , the corresponding parameter is the diameter of the abrasive and the density of the abrasive .
[0024] Preferably, the behavior description function formula of the polishing machine in S40 is as follows:
[0025]
[0026] In the formula, is the weight matrix, ReLU is the activation function, is the bias term, is the weight of the output layer, is the bias term of the output layer.
[0027] Preferably, the polishing parameter adjustment formula in S40 is as follows:
[0028]
[0029] In the formula, is the similarity, is the random polishing parameter scheme, For behavioral utility, is the normalization function.
[0030] Preferably, the reward mechanism formula in S50 is as follows:
[0031]
[0032] In the formula, and For balance and The weight parameters between For balance Parameters that influence the polishing effect.
[0033] Preferably, the formula for constructing the optimized polishing parameter function in S50 is as follows:
[0034]
[0035] In the formula, is a tuple of state description function and behavior description function, To adjust the plan Next , is the mathematical expectation operator, is the entropy coefficient, is the i-th adjustment plan The state description function under the behavior entropy.
[0036] Preferably, the outputting of the optimal polishing parameters in S60 includes:
[0037] Initialize the polishing parameters of the polishing machine and calculate the optimized polishing parameter function The gradient of the polishing parameters is adjusted gradually until the maximum target optimization function is achieved. And output the optimal polishing parameters.
[0038] Beneficial effect: The present invention establishes a dynamic model of the polishing machine, then uses the regression moving average model for training, outputs predicted polishing parameters, and finally optimizes the polishing parameters to achieve the effect of reducing TTV. Through this method, the polishing parameters can be controlled more accurately, the efficiency and stability of polishing can be improved, thereby reducing the thickness variation of the semiconductor, and further improving the quality and reliability of the product. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 It is a flow chart of a preferred embodiment of the present invention. DETAILED DESCRIPTION
[0040] The embodiments of the present invention are described in detail below. The following embodiments are implemented on the premise of the technical solution of the present invention, and detailed implementation methods and specific operation processes are given, but the protection scope of the present invention is not limited to the following embodiments.
[0041] The present invention provides a semiconductor polishing method for reducing TTV, such as Figure 1 As shown, the technical solution includes the following steps, specifically including:
[0042] S10: defining polishing parameters of the polishing machine, and establishing a dynamic model of the polishing machine based on the polishing parameters;
[0043] S20: Establishing a state description function of the polishing machine based on the dynamic model;
[0044] S30: adopting a regression moving average model, training the regression moving average model based on a state description function and a preset loss function, and outputting a predicted polishing parameter of the polishing machine;
[0045] S40: establishing a behavior description function of the polishing machine based on the predicted polishing parameters, calculating the behavior utility based on the state description and the behavior description function, defining a random polishing parameter scheme, and optimizing the random polishing parameter scheme based on the behavior utility to obtain an adjustment scheme for the polishing parameters;
[0046] S50: defining a reward mechanism, and constructing a polishing parameter optimization function based on the reward mechanism and the polishing parameter adjustment scheme;
[0047] S60: Outputting the optimal polishing parameters based on the optimized polishing parameter function, and inputting the optimal polishing parameters into the polishing machine to perform a polishing operation on the semiconductor.
[0048] Preferably, the dynamic model formula for establishing the polishing machine in S10 is as follows:
[0049]
[0050] In the formula, is the amount of semiconductor removed, is the polishing liquid ratio coefficient, The electromagnetic force generated by the polishing machine surface, is the relative speed between the semiconductor and the polishing machine during the polishing process, is the amount of semiconductor dissolved during the polishing process, is the dissolution rate of the semiconductor, is the penetration value of the abrasive into the semiconductor surface, is the rotation speed of the polishing machine, is the area of contact between the polisher and the semiconductor, is the diameter of the abrasive, is the density of the abrasive, is a constant for the material concentration.
[0051] Preferably, the state description function formula for establishing the polishing machine in S20 is as follows:
[0052] .
[0053] Preferably, the preset loss function formula in S30 is as follows:
[0054]
[0055] In the formula, is the sample size, is the predicted polishing parameters of the polishing machine, is the actual polishing parameter of the polishing machine.
[0056] Preferably, include , and , one-to-one correspondence ;
[0057] correspond , the corresponding parameter is the polishing liquid ratio coefficient , Electromagnetic force generated on the polishing machine surface and the relative speed of the semiconductor and the polisher during polishing ;
[0058] correspond , the corresponding parameter is the indentation value of the abrasive into the semiconductor surface , the polishing machine's rotation speed Electromagnetic force generated by the polishing machine surface ;
[0059] correspond , the corresponding parameter is the diameter of the abrasive and the density of the abrasive .
[0060] Preferably, the behavior description function formula of the polishing machine in S40 is as follows:
[0061]
[0062] In the formula, is the weight matrix, ReLU is the activation function, is the bias term, is the weight of the output layer, is the bias term of the output layer.
[0063] Specifically, the behavior description function of the polishing machine is output by a neural network. First, the input layer contains three nodes, corresponding to , and , the formula is as follows:
[0064]
[0065] Set one or more hidden layers, each of which contains several neurons. Each neuron undergoes nonlinear transformation through the activation function ReLU, and the formula is as follows:
[0066]
[0067]
[0068] In the formula, is the weight matrix, is the bias term.
[0069] The output layer formula is as follows:
[0070]
[0071] In the formula, is the weight of the output layer, is the bias term of the output layer.
[0072] Preferably, the polishing parameter adjustment formula in S40 is as follows:
[0073]
[0074] In the formula, is the similarity, is the random polishing parameter scheme, For behavioral utility, is the normalization function.
[0075] Specifically, is a normalization function used to normalize the probability distribution. To obtain the function with the smallest independent variable.
[0076] Preferably, the reward mechanism formula in S50 is as follows:
[0077]
[0078] In the formula, and For balance and The weight parameters between For balance Parameters that influence the polishing effect.
[0079] Specifically, by maximizing and minimize , can better control and reduce the impact of TTV, considering , can more comprehensively evaluate the polishing effect and maximize and minimize At the same time, the contact area is also optimized. Such a reward mechanism can help achieve more comprehensive optimization of polishing parameters.
[0080] Preferably, the formula for constructing the optimized polishing parameter function in S50 is as follows:
[0081]
[0082] In the formula, is a tuple of state description function and behavior description function, To adjust the plan Next , is the mathematical expectation operator, is the entropy coefficient, is the i-th adjustment plan The state description function under the behavior entropy.
[0083] Preferably, the outputting of the optimal polishing parameters in S60 includes:
[0084] Initialize the polishing parameters of the polishing machine and calculate the optimized polishing parameter function The gradient of the polishing parameters is adjusted gradually until the maximum target optimization function is achieved. And output the optimal polishing parameters.
[0085] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.
Claims
1. A semiconductor polishing method for reducing TTV, characterized in that: include: S10: defining polishing parameters of the polishing machine, and establishing a dynamic model of the polishing machine based on the polishing parameters; S20: establishing a state description function of the polishing machine based on the dynamic model; S30: adopting a regression moving average model, training the regression moving average model based on the state description function and the preset loss function, and outputting the predicted polishing parameters of the polishing machine; S40: establishing a behavior description function of the polishing machine based on the predicted polishing parameters, calculating the behavior utility based on the state description and the behavior description function, defining a random polishing parameter scheme, and optimizing the random polishing parameter scheme based on the behavior utility to obtain an adjustment scheme for the polishing parameters; S50: defining a reward mechanism, and constructing a polishing parameter optimization function based on the reward mechanism and the polishing parameter adjustment scheme; S60: Outputting optimal polishing parameters based on the optimized polishing parameter function, and inputting the optimal polishing parameters into a polishing machine to perform a polishing operation on the semiconductor.
2. A semiconductor polishing method for reducing TTV according to claim 1, characterized in that: The formula for establishing the dynamic model of the polishing machine in S10 is as follows: ; In the formula, is the amount of semiconductor removed, is the polishing liquid ratio coefficient, The electromagnetic force generated by the polishing machine surface, is the relative speed between the semiconductor and the polishing machine during the polishing process, is the amount of semiconductor dissolved during the polishing process, is the dissolution rate of the semiconductor, is the penetration value of the abrasive into the semiconductor surface, is the rotation speed of the polishing machine, is the area of contact between the polisher and the semiconductor, is the diameter of the abrasive, is the density of the abrasive, is a constant for the material concentration.
3. A semiconductor polishing method for reducing TTV according to claim 2, characterized in that: The state description function formula of the polishing machine in S20 is as follows: 。 4. A semiconductor polishing method for reducing TTV according to claim 3, characterized in that: The preset loss function formula in S30 is as follows: ; In the formula, is the sample size, is the predicted polishing parameters of the polishing machine, is the actual polishing parameter of the polishing machine.
5. A semiconductor polishing method for reducing TTV according to claim 4, characterized in that: Said include , and , one-to-one correspondence ; Said correspond , the corresponding parameter is the polishing liquid ratio coefficient , Electromagnetic force generated on the polishing machine surface and the relative speed of the semiconductor and the polisher during polishing ; Said correspond , the corresponding parameter is the indentation value of the abrasive into the semiconductor surface , the polishing machine's rotation speed Electromagnetic force generated by the polishing machine surface ; Said correspond , the corresponding parameter is the diameter of the abrasive and the density of the abrasive .
6. A semiconductor polishing method for reducing TTV according to claim 5, characterized in that: The behavior description function formula of the polishing machine in S40 is as follows: ; In the formula, is the weight matrix, ReLU is the activation function, is the bias term, is the weight of the output layer, is the bias term of the output layer.
7. A semiconductor polishing method for reducing TTV according to claim 6, characterized in that: The formula for adjusting the polishing parameters in S40 is as follows: ; In the formula, is the similarity, is the random polishing parameter scheme, For behavioral utility, is the normalization function.
8. A semiconductor polishing method for reducing TTV according to claim 7, characterized in that: The reward mechanism formula in S50 is as follows: ; In the formula, and For balance and The weight parameters between For balance Parameters that influence the polishing effect.
9. A semiconductor polishing method for reducing TTV according to claim 8, characterized in that: The formula for constructing the optimized polishing parameter function in S50 is as follows: ; In the formula, is a tuple of state description function and behavior description function, To adjust the plan Next , is the mathematical expectation operator, is the entropy coefficient, is the i-th adjustment plan The state description function under the behavior entropy.
10. A semiconductor polishing method for reducing TTV according to claim 9, characterized in that: The output optimal polishing parameters in S60 include: Initialize the polishing parameters of the polishing machine and calculate the optimized polishing parameter function The gradient of the polishing parameters is adjusted gradually until the maximum target optimization function is achieved. And output the optimal polishing parameters.