Modeling and Control of Single Wafer Microelectronics Processes

单晶片快速热化学气相沉积（RTCVD）系统是一项新兴技术。在这些系统中，单个晶圆片被装入一个冷却的不锈钢室，在那里低压气体在辐射加热的晶圆片表面发生反应。与多晶圆系统相比，RTCVD系统在初始资本和所需晶圆制造空间方面的成本都要低得多。此外，RTCVD系统中的冷壁操作消除了多晶圆系统热壁上产生的颗粒。在CVD反应器（PECVD）中加入等离子体增强，通过允许较低的操作温度来防止晶圆的热损伤。单晶片反应器的吞吐量需要最大化（例如快速热处理），以与多晶片系统竞争。本研究将集中于利用单晶片CVD和具有等离子体增强和快速热处理能力的蚀刻反应器的沉积和蚀刻技术。对于这些单元操作，将开发新的基本（理论）数学模型，将操作和设备设计变量与反应堆性能联系起来。这些模型将使用来自两种反应堆配置的数据进行验证；其中一个反应器是德州仪器公司的柔性制造反应器，将用于研究氮化硅沉积。另一种是使用四氟化碳CF4和氢H2的反应离子蚀刻剂。该项目还将研究新的控制策略，允许为每个反应堆设计基于模型的控制器。特别感兴趣的将是一种基于非线性规划的技术，称为非线性模型预测控制。与德州仪器和SEMATECH人员的积极合作将促进这项技术最终转移到半导体行业。

An emerging technology is the single wafer rapid thermal chemical vapor deposition (RTCVD) system. In these systems a single wafer is loaded into a cooled stainless steel chamber where low pressure gases react on the radiatively heated wafer surface. Compared to the multiwafer system, the RTCVD system represents considerably less cost both in initial capital and in required wafer fabrication floor space. In addition, cold wall operation in the RTCVD system eliminates the particulate generation occurring on the hot walls of the multiwafer system. Adding plasma enhancement to a CVD reactor (PECVD) prevents thermal damage to the wafer by permitting lower temperatures of operation. The throughput for single wafer reactors needs to be maximized (e.g. rapid thermal processing) to be competitive with multiwafer systems. This study will focus on deposition and etch technology using single wafer CVD and etching reactors with plasma enhancement and rapid thermal processing capability. For these unit operations, new fundamental (theoretical) mathematical models will be developed that relate the operating and equipment design variables to reactor performance. These models will be verified using data from two reactor configurations; one of these reactors is a flexible manufacturing reactor from Texas Instruments, which will be used to study silicon nitride deposition. The other is a reactive-ion etcher using carbon tetrafluoride CF4 and hydrogen H2. This project will also research new control strategies that permit design of model- based controllers for each reactor. Of particular interest will be a nonlinear programming-based technique called nonlinear model predictive control. Active collaboration with personnel at Texas Instruments and SEMATECH will facilitate ultimate transfer of this technology to the semiconductor industry.