Run-to-Run Controller Design And Performance Assessment Using The Disturbance State Estimation

Run-to-run(R2R)control has become more widely used throughout the semiconductor industry.Nowadays it has become common that the high mix of products made in a single ASIC fab or foundry and different products run on the same processing tool.However,the data feedback is not timely and the information of the low-frequent product is poor with the single-threaded control.They all result in the increasement of the product defective rate.So the non-threaded control is introduced into the semiconductor industry and some of the algorithms are verified effective.Therefore,this thesis studies the state estimation method and the controller design in the framework of the non-threaded R2 R control for the high-mix manufacturing process.The main work can be summarized as follows:(1)It discusses the state space model for the non-threaded control.Aiming at the rank deficiency problem of the state matrix,a new method is proposed based on the ANOVA model and the Bayesian estimation theory.It takes advantage of the IMA(1,1)disturbance model and the output residuals to construct a context state matrix.Then it uses the Bayesian estimation theory to update the context matrix and adjusts the input to obtain the optimal output results.The calculation of the inverse of the context matrix is avoided and the unobservability problem is addressed with this method.And it also reduces the computational complexity compared with the Kalman filter method.The simulation results and the reverse engineering data both validate that the proposed algorithm is a highly suitable method for high-mix product processes.(2)Aiming at the metrology delay caused by the characteristics of the R2 R control and semiconductor fabrication,a novel algorithm based on the expectation maximization(EM)algorithm and exponentially weighted moving average(EWMA)filter is developed to predict the metrology delay.First,the likelihood function of the metrology delay is established.With the iteration in the EM algorithm,the estimated metrology delay is obtained when the convergence condition is met.Meanwhile,it analyzes the stability of this system and the static offset existing in the output is compensated by varying the estimated item.The simulation results verify the effectiveness of the algorithm.(3)In the semiconductor manufacturing,the process parameters may be unstable because of the uncertainty factors of the production environment,the metrology delay,distribution of the noise and so on which are hard to model.In this work,a novel method is proposed based on the idea of the extended state observer(ESO)for the disturbance estimation.It lumps all the uncertain disturbance together to be the only one based on the theory of active disturbance rejection controller(ADRC).Then the ESO is established to track the changes of the state and the disturbance.Finally,the effectiveness of the proposed algorithm is evaluated in different simulation cases.(4)Control performance assessment(CPA)is an important tool to realize high performance R2 R control systems in manufacturing plants.A CPA index is discussed in this thesis based on the existing CAP methods.First,the output error source of the processes with a R2 R controller is analyzed based on closed-loop parameter estimation.Then a state space model is constructed.The ARMAX regression method is applied for the estimated disturbance data.The simulation results show that this CPA index is sensitive to the performance degradation caused by plant model mismatch,improper controller tuning,metrology delay and high-mix process parameters.Finally,this method is applied to the non-threaded R2 R control which is based on the state estimation and Kalman filter.Overall process control performance with the plant model mismatch is assessed.