Fig 1 - uploaded by Takeshi Nishida
Content may be subject to copyright.
Source publication
For cleaning silicon wafers via the RCA clean, temperature control is important for stable cleaning performance, but difficult owing that the RCA solutions expose nonlinear and time-varying exothermic chemical reactions. So far, the MSPC (model switching predictive controller) using the CAN2 has been developed and the effectiveness has been validat...
Context in source publication
Context 1
... where the temperature control of chemical solutions is important for a stable cleaning performance. Since the RCA clean uses the mixture of corrosive and hazardous chemical solutions such as SPM (sulfuric acid H 2 SO 4 and hydrogen peroxide H 2 O 2 mixture) and so on, several special equipment are arranged for heating solutions as shown in Fig. 1, where there are solution bath, a bellows pump, an infrared heater and a cleaning filter which are connected by anti-corrosive recirculation pipes, and the thermal sensor is covered by an anti-corrosive glass tube (see [2], [3] for details of the system and the thermal model). Thus, this system involves long and fluctuating time lags ...
Similar publications
This paper illustrates and analyzes the competi-tive associative net called CAN2 applied to model-switching control of the temperature of RCA clean-ing solutions, where the RCA clean is the in-dustry standard way to clean silicon wafers and the temperature control is important for a stable cleaning performance. Since the control is dii-cult owing t...
Citations
... Thus, this system involves long and fluctuating time lags and delays, and the mixture of the solutions exposes several exothermic reactions which are nonlinear and time-varying. In order to control such nonlinear and time-varying plants involving time lags and delays, we have developed a control method called MSPC (model-switching predictive controller) using the CAN2 [3][10]. Precisely, the CAN2 in the MSPC learns multiple linear models of the plant dynamics from the input and output data of the plant, and then selects an appropriate linear model at each time of the control phase in order for the GPC (generalized predictive controller) to use the selected linear model. ...
... data depending on the previous control trajectory, and the CAN2 may overlearn the data. One of the methods for avoiding overlearning is the cross-validation, with which we have obtained a certain level of improvement [10]. As another method for avoiding overlearning, we in this article try to apply the ensemble method, where the ensemble of the CAN2s may have a higher prediction ability than the single CAN2 because there are a number of researches showing that an ensemble prediction is (not average but) often more accurate than any of the single predictions in the ensemble [14]. ...
For cleaning silicon wafers via the RCA clean, temperature control is important in order to obtain a stable performance, but it is difficult mainly because the RCA solutions expose nonlinear and time- varying exothermic chemical reactions. So far, the MSPC (model switch- ing predictive controller) using the CAN2 (competitive associative net 2) has been developed and the effectiveness has been validated. However, we have observed that the control performance, such as overshoot and settling time, does not always improve as the number of learning itera- tions increases when using multiple units of the CAN2. So we apply the ensemble learning scheme to the CAN2 for stable control over learning iterations, and we examine the effectiveness of the present method by means of computer simulation.
The RCA cleaning method is the industry standard way to clean silicon wafers, where temperature control is important for a stable cleaning performance. However, it is difficult mainly because the RCA solutions cause nonlinear and time-varying exothermic chemical reactions. So far, the MSPC (model switching predictive controller) using the CAN2 (competitive associative net 2) has been developed and the effectiveness has been validated. However, we have observed that the control performance, such as the settling time and the overshoot, does not always improve with the increase of the number of learning iterations of the CAN2. To solve this problem, we introduce the bagging method for the CAN2 and first-difference signals for the MSPC. The effectiveness of the present method is shown by means of computer simulation.
