Study On The Operating State Monitoring Method Of Semiconductor Silicon Single Crystal Growth Process

Monocrystalline silicon material is the foremost basic material for integrated circuit chips.The continuous evolution of chip manufacturing processes has led to the progress of monocrystalline silicon materials in the orientation of large size,high quality,and low cost.Therefore,it is necessary to optimize and upgrade all aspects of crystal growth equipment and growth processes.During this process,monitoring the crystal growth process is an essential means and method for achieving the growth of high-quality monocrystalline silicon materials.In the environment of multi field,multi-phase coupling,vacuum,and high temperature monocrystalline silicon growth,how to establish a multi-state,multi output mono crystalline silicon growth model,analyze the data of process parameters and variables,extract the hidden information and features within the data,and achieve intelligent monitoring of the state of the crystal growth process is the key to improve the crystal growth efficiency and crystal quality.This topic takes the modeling of the monocrystalline silicon growth simulation system as the starting point,establishes state monitoring frameworks based on deep learning by obtaining the system process operation data,and completes the research on the monocrystalline silicon growth process operation state monitoring methods.The main research contents of this thesis are as follows:1.Based on the growth equipment and technology of monocrystalline silicon,the relationships between the growth process variables and quality of monocrystalline silicon are analyzed in detail,and the state monitoring problem and strategy of silicon single crystal growth process are put forward.Among the numerous process variables,heater temperature,crucible temperature,melt temperature,crystal radius,crystal growth velocity,crystal rotation rate,crucible rotation rate,and melt height,which are closely related to crystal growth,are selected as the monitoring variables for the monocrystalline silicon growth state.2.Aiming at the complex physical and chemical changes and nonlinear processes of Czochralski(CZ)monocrystalline silicon growth,and based on the internal characteristics of the monocrystalline silicon growth process and the transfer mechanism of internal material flow,Firstly,a lumped parameter model of monocrystalline silicon growth based on energy balance model and geometric model is established;Secondly,a simulation control system for CZ monocrystalline silicon growth is established,and real-time simulation data of various variables during the crystal growth process are obtained;Finally,the effectiveness of the model is analyzed according to the characteristics of semiconductor industrial production process.The experimental results indicate that the model not only has high accuracy,but also can better depict the dynamic behavior of actual monocrystalline silicon growth systems.3.The growth process of monocrystalline silicon has the characteristics of large amount of data and high requirement for real-time and accuracy of process status monitoring.This thesis utilizes the ability of Deep Belief Network(DBN)to process data and extract data features,a condition monitoring method based on DBN(Condition Monitor-DBN,CM-DBN)is proposed.By using monocrystalline silicon growth data as input for the state monitoring model,selecting appropriate state monitoring indicators and calculating their control limits.A CM-DBN state monitoring model is trained.The simulation results show that the model is effective and accurate in monitoring the growth state of silicon single crystal.4.There are various noises and other interferences in the actual growth process of monocrystalline silicon,in order to further improve the accuracy of state monitoring,this thesis combines the Wavelet Packet Decomposition(WPD)method with DBN to propose a condition monitoring method based on WPD(WPD-CM-DBN).This method performs wavelet packet decomposition on various process data generated in a monocrystalline silicon growth system,extracts the low-frequency and high-frequency subsequences of each variable,and inputs them into the corresponding sub DBN network.Finally,the outputs of all sub DBN networks are combined and input into the total DBN network.The status monitoring index values are calculated and compared with the monitoring index control limits of the normal state WPD-CM-DBN model,to determine the operation status of the monocrystalline silicon growth system.Through simulation experiments,it has been verified that this model has a stronger ability to mine data features than the CM-DBN model,and has a higher accuracy in monitoring the growth status of monocrystalline silicon.