随着大规模集成电路集成度与运行速度的提高，芯片制造的关键技术之一——硅片的平坦化技术，已从传统的化学机械抛光（Chemical Mechanical Polishing，CMP）发展为超低压CMP。由此要求控制系统必须对新型CMP装备的多区压力进行高精度快响应协同控制，才能满足抛光均匀度等日益提高的技术指标要求。为此，本文根据国家科技重大专项02专项项目“超低下压力CMP系统及工艺开发”的需求，围绕超低压CMP装备多区压力协同控制的关键问题和控制系统实现技术开展研究工作。首先，对超低压CMP多区压力协同控制的需求进行分析，并据此对多区压力协同控制系统进行设计。然后，对超低压CMP装备的运行机理进行研究，建立了多区压力控制系统的数学模型，并进一步研究闭环辨识方法，实现了模型参数的准确辨识，为后续的分析与控制提供了依据。在此基础上，对CMP多区压力协同控制的三方面关键问题进行了研究。第一，针对CMP装备多区压力控制中的耦合问题，提出了定性解耦和定量解耦控制方法，有效抑制了多区加压过程中的相互影响，保证各区压力满足工艺要求，误差小于专项项目要求。第二，为抑制生产环境干扰对CMP装备运行的影响，对多区协同加压过程中的抗扰问题进行了研究，提出一种基于DRNN在线辨识+神经元解耦+分段变参数复合控制策略。该策略利用神经网络的非线性逼近和学习能力，对系统的逆模型进行在线实时精确辨识，从而避免离线求解复杂非线性系统的逆模型，为基于自适应逆控制的多区压力协同控制系统的实现提供了保证。第三，对多区压力协同控制装卸片原理进行了研究，提出一种通过多区压力主动协同控制实现可靠装卸片的方法，并给出了多区协同控制装片的实现工艺。根据工艺需求建立了装卸片系统数学模型，进行了工艺参数优化，给出了可靠运行的边界条件。运行结果表明，该方法可有效提高装卸片的可靠性，满足应用需求。最后，在以上工作基础上，先后完成了超低压CMP装备原理样机和α机控制系统开发。并在原理样机上进行了实验验证，在α机上进行了应用验证。验证结果表明，所开发的控制系统各项技术指标均达到专项项目要求。

With the improvement of the integration degree and operating speed of large scale integrated circuit, one of the key technologies in chip manufacturing, and the global planarization technology for wafer, has developed from traditional Chemical Mechanical Polish(CMP) into Ultra Low Downforce CMP(ULDCMP). This requires a new control system which has the cooperative control ability of high precision and fast response in order to meet the much more strict technical specifications such as polishing uniformity.Therefore, based on the demand of the national science and technology major projects (e.g. 02 Special Project) “The development of Ultra Low Downforce CMP system and process”, the research on the key technical issues of ULDCMP equipment and achievement of control system has been carried out.First of all, a multi-zones’ cooperative pressure control system (MCPCS) is built after the system need is analyzed. Then the research on operating mechanism of ULDCMP equipment is implemented and the mathematical model of the control system is modeled. Furthermore, an identification method of closed-loop is studied and model parameters are accurately achieved, which provides reference for following analysis and control.Based on the above mathematical model, this paper mainly focuses on three key questions of the MCPCS.First, as to the coupling problem of MCPCS, both qualitative decoupling and quantitative decoupling methods are proposed, which effectively inhibit the multi-zones’ influence. The result of experiments has validated that the above methods meet the requirement of project.Second, in order to weak the impact on the operation of the equipment in the production process, anti-disturbance work is done and a hybrid control strategy based on DRNN decoupling& neuron decoupling&segmentation control with variable parameters is proposed. The strategy uses the learning and nonlinear approximation ability of neural network to identify the system’s inverse model real-time and online, thus avoids solving the inverse model of complex nonlinear system off-line. The strategy provides a guarantee to realize MCPCS based on the adaptive inverse control theory (AIC).Third，the principle of load-wafer&unload-wafer process is studied in detail and puts forward a method which can realize the process stably and reliably. In addition, the load-wafer process is realized. The mathematical model of load-wafer&unload-wafer process is bulit, the process parameters are optimized and boundary conditions that ensure reliable operation of equipment are obtained. The results prove that the above method can improve the reliablity and meet the requirements of the project.Finally, based on the above studies, the control systems of ULDCMP prototype and α-machine have been developed. Experimental verification is conducted on the prototype and application verification is performed on the α-machine. The results demonstrate that specifications of the whole control systems meet the requirements of the project.