AVS1997 Session MS-FrM: Real-Time Process Control

Friday, October 24, 1997 8:20 AM in Room J1/4

Friday Morning

Start	Invited?	Item
8:20 AM	Invited	MS-FrM-1 Trends in Real-Time Process Control S.W. Butler (Texas Instruments) Control has slowly evolved from the first applications of automatic real-time pressure control and auto-shutdown based on setpoint guardbands. However, the recent challenges of increasingly smaller geometry, productivity demands, and increasing wafer size are driving more rapid introduction of advanced controls. Simultaneous advances in in-situ sensor technology, such as commercially available rf sensors, full wafer interferometry, and in-line FTIR, is making advanced control possible. Environmental, Safety, and Health concerns are producing further applications to meet the need for "green processing." The complexity of these new controllers varies from elaborate algorithms to detect a change in a single signal to multiple input-multiple output controllers. Some examples of emerging technologies in etch include neural networks for endpointing small percent open area to multizoned rf electrodes. Control of multizoned lamps for rapid thermal processing has become common place, and now similar control technology is being considered for bake plates on lithography tracks. Utilization of SECS equipment signals for automatic real-time interdiction (shut-down) is also becoming more common. However, considerable challenges remain. The integration of the sensors, the algorithms, the equipment, and the manufacturing enterprise system still remains the single largest hurdle. The evolution of the controller to match changes in hardware state is a particularly vexing problem. Introduction of new devices and processes also creates the need for rapid optimization of the control model. This talk will analyze the drivers of control and the challenges that are faced by the semiconductor industry as the level of control moves beyond a state of infancy. Examples of real-time control will also be given to illustrate these drivers and challenges.
9:00 AM	Invited	MS-FrM-3 The Software Controls Challenge: Delivering Solutions to OEMs and End-Users M. Hanssmann (Brooks Automation) Brooks Automation Canada (formerly Techware Systems) has been supplying control solutions to the thin film industry for 14 years. Throughout this time the company has worked with OEMs, end users, as well as instrumentation suppliers. This presentation will discuss some of the challenges that a second tier supplier faces in working with the first tier suppliers (OEMs), and/or working with the end users directly. These challenges include sales issues, technical issues, working with standards organizations, and business considerations. Specifically, the evolution of software controls in this industry will be reviewed, and the role of a controls supplier today discussed. As the level of sophistication of processing equipment continue to rise, the demands on software have soared from almost non-existent 15 years ago, to a level where they can represent over 30 % of the engineering effort today. With process equipment representing up to 70% of the investment in a fab today, this software component is very large. The result is an industry with some very significant software problems. In an effort to address these problems, the industry has, in the past few years, started to make greater use of third party controls suppliers. Some of the advantages and pitfalls of this approach will be reviewed. The rapid evolution of software in controlling process equipment has also resulted in a corresponding evolution in the controls architecture. As almost every actuator and sensor today must be interfaced to a control system, there is a need to integrate products from a large number of different suppliers. To accommodate the increasing sophistication of these sensors, and to reduce some of the integration effort, as well as equipment manufacturing costs, sensor bus technology is being deployed. At the other end of the control architecture spectrum, computers are also continuing to change rapidly. The effect of these changes, and where the controls architecture can be expected to evolve to will be discussed.
9:40 AM		MS-FrM-5 Sensor-based RF Power Monitoring and Control in an Inductively Coupled Plasma Etcher N. Williams, C.G. Lee, R. Patrick (Lam Research Corporation) A RF sensor has been installed in the power delivery circuit between the matching network and the lower wafer-biasing electrode of a commercial inductively coupled plasma etching system. The sensor has been used to characterize the variation of delivered power to the load and the load impedance as process conditions such as generator set-point, process pressure, gas flow and composition were varied. It was found that the efficiency of power delivery was significantly less than unity and that it varied according to the impedance of the load. Since the plasma impedance is a function of both top, inductive power and bottom power, the magnitude of the deposited power at the lower electrode depends not only on the bottom power setting but also on the top power setting. This makes true decoupling of plasma generation and wafer biasing difficult. It was also found that the magnitude of stray capacitances and inductances can have a significant impact on the power delivery system, particularly when comparing one etching tool with another. Comparison has been made between etching results for polysilicon and oxide films and RF parameters. It was found, for instance, that oxide etch rates were sensitive to the magnitude of RF voltage whereas polysilicon etch rates were more dependent on the RF current. In addition a feedback loop was implemented where the output of the probe, such as delivered power, voltage or current, was used as the control variable rather than relying solely on the generator power output for the process setpoint. Operating in this mode was found to compensate for interactions of top and bottom power allowing for a more decoupled plasma generation and wafer biasing. This technique has also been investigated from the viewpoint of minimizing the variability in power delivery caused by uncontrolled strays in the RF circuit. The impact of the direct feedback loop method of power control on etching results such as uniformity and repeatability has also been studied.
10:00 AM		MS-FrM-6 Feedback Control of Inductively Coupled Plasma Reactors¹ S. Rauf, M.J. Kushner (University of Illinois, Urbana) To maintain high yields during device fabrication, feedback control of microelectronics manufacturing equipment has become an important consideration. This is particularly true for plasma processing steps. To investigate and optimally select feedback control strategies, comprehensive plasma equipment models linked to control algorithms are desirable. Such a numerical simulation tool, called the Virtual Plasma Equipment Model (VPEM), has recently been developed. The VPEM is an extension of the Hybrid Plasma Equipment Model (HPEM) in which sensor, actuator and controller modules have been added to address real-time-control (RTC) issues. The sensor module emulates measurements from experimental diagnostics. Data from the sensor module is used by the controller to specify changes in process parameters, enacted by the actuator module, to bring simulated reactor properties to specified criteria. In this paper, we use the VPEM to investigate RTC for inductively coupled plasma sources operating in Ar/Cl₂ and other gas mixtures. The sensors emulate optical emission spectroscopy, ion current and etch rate diagnostics. The actuators include inductive power, gas pressure, rf bias voltage and gas flow rates. Examples of RTC strategies will be discussed to nullify drifts in actuators, compensate for gas leaks and to correct for long term drifts in wall conditions. We also investigate RTC techniques for improving etch uniformity. ¹Work supported by ARPA/AFOSR, NIST, SRC and NSF
10:20 AM		MS-FrM-7 Automatic Matching Algorithms with Improved Convergence for Inductively Coupled Plasma Sources R.H. Goulding, L.A. Berry, J.B.O. Caughman, M.S. Hileman (Oak Ridge National Laboratory) Automatic matching algorithms have been developed and tested on the Large Area Source Experiment at Oak Ridge National Laboratory (ORNL) which are convergent over a significantly greater range of initial matching element values than commonly used algorithms. This eliminates the need for carefully selected presets dependent on the final plasma load, or the use of separate pre-strike and post-strike algorithms. Unlike the standard "phase-mag" algorithm, which uses error signals derived from the phase and magnitude of the input impedance on the generator side of the matching network, the method presented here incorporates information from a probe (current or voltage) on the load side of the network, as well as probes on the generator side. Error signals are derived which are monotonic for variations in matching element values of 20% or greater about either side of the the match point, in contrast to phase-mag error signals. Convergence over a greatly increased range of initial matching element settings has been observed, allowing a single matching algorithm to be used to track between near perfect pre-strike and post-strike matches (<< 0.1% reflected power). Tests have been conducted using a capacitive tee matching network on the ORNL Large Area Source with automatic control of the series load side capacitor and either the series generator side or shunt capacitor. Final matches with << 0.1% reflected power have been achieved with initial post-strike mismatches having up to 94% reflected power, and initial matching element values differing from the final values by > 100%. These tests were run using an Ar plasma load at a pressure of 10 mtorr. In addition to plasma tests with a tee network, simulations have been performed using the commercially available SIMULINK control system modeling package, indicating that a similar algorithm can be used successfully with transformer coupled matching networks. Research sponsored by the Laboratory Directed Research and Development Program of Oak Ridge National Laboratory, managed by Lockheed Martin Energy Research Corp. for the U. S. Department of Energy under Contract No. DE-AC05-96OR22464.
10:40 AM		MS-FrM-8 Run-to-Run Process Control and Endoint Detection on a Dual-Coil TCP with FWI and OES M.S. Le, T.H. Smith, D.S. Boning, H.H. Sawin (Massachusetts Institute of Technology) A technique for improving the etch rate uniformity in a high density polysilicon process as well as a technique to detect endpoint is presented. A Transformer Coupled Plasma (TCP) etcher was outfitted with two concentric independent spiral antennae. A full wafer interferometry (FWI) system is used to measure the etch rate at up to 121 locations on the wafer surface, and a spatially resolved optical emission spectroscopy (OES) in-situ sensor system provides a measure of the plasma state. An Artificial Neural Network EWMA controller was incorporated to control the wafer etch rate profile and plasma state condition, while minimizing the change in the process recipe, via a multi-objective optimization function. A design of experiments was initially performed and artificial neural network models were trained to provide a functional relationship between the process recipe and the system response, as measured by FWI and OES. This run-to-run process control configuration was found to control the etching uniformity across blanket polysilicon 150 mm wafers and maintain a consistent plasma state in the face of a disturbance applied to the inner and outer coil powers. The above process control system can be coupled with an endpoint detection system. Several works have considered the use of single or multiple wavelengths of OES trace for endpoint detection in plasma etching. Many of these methods have encountered difficulties, particularly for low open-area etching. We outline a method for using the entire OES spectra for reliable and sensitive endpoint detection. Analysis of historical OES spectra taken from etching of production wafers demonstrate that this method provides excellent endpoint prediction capability, even for low open-area cases. An integrated process control system can incorporate both the run-to-run model based controller and an endpoint detection system to provide not only control of the etching uniformity and plasma chemistry but also endpoint.
11:00 AM		MS-FrM-9 Non-Periodic Lot Processing, Random Measurement Delays, and Intermittent Lot Processing with an Extended Predictor Corrector Controller T.H. Smith, D.S. Boning (Massachusetts Institute of Technology) Run by Run (RbR) process control is gaining recognition in the semiconductor manufacturing industry as a means to compensate for errors caused by drifting process equipment. Of particular interest is the ability for these controllers to handle wafer processing when feedback measurements may be taken at possibly random intervals with possibly random measurement delays. These effects are of key concern to the RbR control of processes such as metal sputter deposition, where various desired film thicknesses cause data sampling at random kilowatt-hour locations (over which the deposition rate is known to drift linearly) and wafers need to be continually processed in-between possibly inconsistent (with respect to KWH) measurements having possibly random delays. We present experimental evidence that such additions to RbR control methods are necessary. Modifications have been made to the Predictor Corrector Control (PCC) algorithm in order to address situations where non-periodic lot processing (e.g. different lot size) occurs, wafer processing takes place between possibly non-periodic measurements, and measurement delays occur in a possibly non-periodic fashion. The effects of measurement delay errors and intermittent processing errors are illustrated for a sputter deposition simulation. Additional simulations demonstrate the benefits obtained from the new generalized PCC algorithm.