A new strategy based algorithm to optimize process parameter uniformity (e.g.sheet resistance, oxide thickness) and temperature uniformity on wafers in a commercially available Rapid Thermal Processing (RTP) system with independent lamp control is described. The computational algorithm uses an effective strategy to minimize the standard deviation of the considered parameter distribution. It is based on simulation software which is able to calculate the temperature and resulting parameter distribution on the wafer for a given lamp correction table. A cyclical variation of the correction values of all lamps is done while minimizing the standard deviation of the considered process parameter. After the input of experimentally obtained wafer maps the optimization can be done within a few minutes. This technique is an effective tool for the process engineer to use to quickly optimize the homogeneity of the RTP tool for particular process requirements. The methodology will be shown on the basis of three typical RTP applications (Rapid Thermal Oxidation, Titanium Silicidation and Implant Annealing). The impact of variations of correction values for single lamps on the resulting process uniformity for different applications will be discussed.

Temperature measurements and processing uniformity continue to be major issues in Rapid Thermal Processing. Spatial and temporal variations in thermal radiative properties of the wafer surface are sources of non-uniformities and dynamic variations. These effects are due to changes in spectral distribution (wafer or heat source), oxidation, epitaxy, silicidation, and other microstructural transformations. Additionally, other variations are induced by the underlying (before processing) and developing (during processing) patterns on the wafer. Numerical simulations of Co silicidation that account for these factors are conducted to determine the radiative properties, heat transfer dynamics, and resultant processing uniformity.

This work aims to systematically gain an understanding of the effects of multilayer patterns on wafer temperature uniformity during rapid thermal processing, and explore possible solutions to the problem. Steady state and transient wafer temperature distributions are simulated by combining a detailed reactor transport model with multilayer electromagnetic theory to predict wafer radiative properties. A generic axisymmetric RTP system with single-side illumination is used as a testbed to explore pattern effects for a simulated source/drain implant anneal.

The directional reflectance and approximate emissivity of rough silicon wafers were measured by reflection measurements using a single point detector and a broad area illumination source. Experiments were also performed to determine the cone angle of the incident light required to properly measure the emissivity of rough backsides. Based on surface roughness parameters acquired with an Atomic Force Microscope, reflectance calculations were performed within the framework of the Beckmann-Spizzichino model. The results are qualitatively consistent with experimental observations.

A highly flexible Rapid Thermal Multiprocessing (RTM) reactor is described. This flexibility is the result of several new innovations: a lamp system, an acoustic thermometer and a real-time control system. The new lamp has been optimally designed through the use of a “virtual reactor” methodology to obtain the best possible wafer temperature uniformity. It consists of multiple concentric rings composed of light bulbs with horizontal filaments. Each ring is independently and dynamically controlled providing better control over the spatial and temporal optical flux profile resulting in excellent temperature uniformity over a wide range of process conditions. An acoustic thermometer non-invasively allows complete wafer temperature tomography under all process conditions - a critically important measurement never obtained before. For real-time equipment and process control a model based multivariable control system has been developed. Extensive integration of computers and related technology for specification, communication, execution, monitoring, control, and diagnosis demonstrates the programmability of the RTM.

Measurement of resistance in-situ during rapid thermal annealing is a powerful technique for process characterization and optimization. A major advantage of in-situ resistance measurements is the very rapid process learning. With silicides, in-situ resistance measurements can quickly determine an appropriate thermal process in which a low resistance silicide phase is formed without the agglomeration or inversion of silicide/polycrystalline silicon structures. One example is an optimized two step anneal for CoSi2 formation which was developed in less than one day. Examples of process characterization include determining the phase formation kinetics of TiSi2 (C49 and C54), Co2Si, and CoSi2 using in-situ ramped resistance measurements. The stability of TiSi2 or CoSi2/poly-Si structures has also been characterized by isothermal measurements. Resistance measurements have been made at heating rates from 1 to 100°C/s and temperatures up to 1000°C. The sample temperature was calibrated by melting Ag, Al, or Au/Si eutectics.

Starting with basic physical radiation principles and with Beers law formulas for the radiation characteristics of arrays of pointlike radiation sources and tubular halogen lamps are analytically derived. Different arrangements of tubular halogen lamps are considered and compared with each other. Homogeneity criteria are worked out and it is demonstrated that the minimum homogeneity distance between a plane in which all filaments are arranged parallel to each other, and an object depends only on the distance between two neighbouring filaments and thus can be given by a simple number. Interesting aspects for RTP equipment manufacturers as well as for users are demonstrated by comparing different lamp arrangements with each other: In crossed lamp fields latticelike inhomogeneity knots arise in the object plane at points where the lamp filaments of the upper and the lower lamp plane cross. If the filaments are arranged parallel to each other in two filament planes, very homogeneous results are obtained when the filaments of the two planes interlace.

The cost of a fabrication line, such as one in a semiconductor house, has increased dramatically over the years, and it is possibly already past the point that some new start-up company can have sufficient capital to build a new fabrication line. Such capital-intensive manufacturing needs better utilization of resources and management of equipment to maximize its productivity. In order to maximize the return from such a capital-intensive manufacturing line, we need to work on the following: 1) increasing the yield, 2) enhancing the flexibility of the fabrication line, 3) improving quality, and finally 4) minimizing the down time of the processing equipment. Because of the significant advances now made in the fields of artificial neural networks, fuzzy logic, machine learning and genetic algorithms, we advocate the use of these new tools in manufacturing. We term the applications to manufacturing of these and other such tools that mimic human intelligence neural manufacturing. This paper describes the effort at the Lawrence Livermore National Laboratory (LLNL) [1] to use artificial neural networks to address certain semiconductor process modeling, monitoring and control questions.

A new approach to semiconductor manufacturing is necessary to decouple end product cost from factory volume. The economy of scale model for manufacturing semiconductor integrated circuits is leading to factories that cost in excess of $1 billion and with costs anticipated to increase nonlinearly. Perhaps of greater concern is that process development and integration costs are of the same magnitude. The Advanced Research Projects Agency (ARPA) has funded several stand-alone research programs to explore new ways of designing and fabricating leading-edge semiconductors. The next steps are to tightly couple design and manufacturing through powerful new frameworks and embedded intelligence with tools. This new manufacturing capability based on flexible, intelligent equipment coupled to a powerful design framework could lead to a leading-edge manufacturing capability, but at a fraction of the cost of a conventional factory. Furthermore, this manufacturing model supports innovation and could lead to a new generation of ICs and systems. The key elements necessary to enable this new semiconductor fabrication environment is described.

A concurrent-engineering approach is applied to the development of an axisymmetric rapidthermal- processing (RTP) reactor and its associated temperature controller. Using a detailed finite-element thermal model as a surrogate for actual hardware, we have developed and tested a multiinput multi-output (MIMO) controller. Closed-loop simulations are performed by linking the control algorithm with the finite-element code. Simulations show that good temperature uniformity is maintained on the wafer during both steady and transient conditions. A numerical study shows the effect of ramp rate, feedback gain, sensor placement, and wafer-emissivity patterns on system performance.

This paper outlines our current approach to utilize infrared reflectance spectroscopy for thin film measurement in the semiconductor industry. The multi-layer thickness and doping concentration of IC wafers can be determined by a single angle, unpolarized infrared reflectance measurement performed using Fourier transform infrared spectrometer. A computer algorithm, which matches theoretical tc measured infrared reflectance spectra, was successfully employed to determine multiple thin film properties.

The Magnetron Sputter Ion Plating (MSIP) technology has been applied in a wide range of modern technological fields, due to the virtually unlimited potential of coating and substrate materials. However, up to now, the development of optimization has been depending mainly on empiricism and experiments.

Energies and the directions of the particles sputtered with various ejection angles and initial energies from the target surface are changed due to collisions with gas atoms during their movement within the sputtering chamber. Therefore the particles have a distribution of incident energies and angles which have a great influence on structure and properties of thin film coatings. This paper describes computer investigations to simulate the motions of the particles from target to substrate using a Monte-Carlo method and a binary collision approximation. The applied model takes into account non-uniform sputtering effects of the target in the MSIP process due to the magnetic field and the erosion zone.

The Direct Simulation Monte Carlo (DSMC) method is applied to the problem of pulsed laser deposition in both 1D and 2D axisymmetric simulations. A source of target atoms expands into a cylindrical volume containing a background gas at room temperature and pressures up to 100 mTorr in the presence of a diffusely reflecting substrate. At regular intervals, the density and temperature of each species are computed. Particle flux and energy per particle incident on the substrate are also monitored as functions of time. The simulation results qualitatively compare well with experimental plume diagnostics, with measured film growth rates as a function of background gas pressure, and with measured changes in film growth stoichiometry resulting from the introduction of a background gas into a two-component PLD system.

Collimated sputtering has been successful in providing good contact barriers for sub-half micron contacts with aspect ratios of 3 and greater. This approach does present drawbacks however, particularly in terms of reduced deposition rates and degraded film uniformity. The flux collected on the collimator leads to closing off of the cells, further reducing deposition rate on the wafer and limiting the life of the collimator. This paper demonstrates simulation of the filling of the collimator with different system configurations and pressures using the SIMSPUD vapor transport and SIMBAD thin film growth simulators. The model can determine collimator filling uniformity, blanket film uniformity, angular distribution of collimated sputter flux, and lifetime of the collimator. Given the target erosion profile, system geometry, and deposition rate, collimator lifetime can be predicted. The model indicates that for a 300 mm diameter source a drop in operating pressure from 0.67 Pa to 0.27 Pa has little effect on collimator life in terms of kWh, while increasing collimator life in terms of wafers by about 50%. The increase in the number of wafers processed comes at the expense of a small loss of uniformity.

Experimental and simulation studies were conducted in an attempt to understand the effects of collimator life time on the Ti and TiN film growth rates and conformalities in sputter deposition processes. The Ti and TiN films were deposited with and without collimation. The hexagonal cells of the collimator used in this study have a 1:1 aspect ratio. A Monte Carlo based simulator was used to calculate the angular distributions of species exiting from a collimator cell and the percentage decrease in the rate of film growth as a function of the collimator life time. Then, a low pressure deposition process simulator, EVOLVE, was used to predict the conformalities of deposited films in contacts or vias, assuming that the films were uniformly deposited on the side-walls of collimator cells. We conclude that the loss in growth rate is largely due to the shrinkage in the cross sectional area of the collimator cell inlets. We arrive at this conclusion after comparing an estimated film thickness on the collimator side-walls with experimental measurements. With extended collimator usage, the predicted and experimental film profiles in contacts or vias show increasing bottom coverage and decreasing side-wall coverages.

This paper presents the application of ripple pyrometry on AG Associates Integra systems which are three phase powered and multi-zone controlled rapid thermal CVD cluster tools. Two types of sensor configurations were explored for optimum emissivity compensated temperature measurement performance. It was found that it is very difficult to measure temperature when a two sensor configuration is used due primarily to the chamber and reflector architecture. Good temperature measurement was obtained at high temperatures when zone ratios between the wafer and lamp sensors were aligned. It was also found that the temperature signal to noise ratio is very sensitive to ripple quality and zone ratio alignment accuracy.

A new speckle technique, sub-feature speckle interferometry, is introduced that relies on the amplitude interference of two independent speckle patterns, originating from coherent illumination, using an optical system that produces interferometric quality interference fringes on a scale comparable to the speckle correlation length. Examples are given for in-plane translation, sample tilt, and temperature measurement (strain). A temperature measurement accuracy σ = 0.92°C is realized. In contrast to traditional full-field speckle cross-correlation techniques, this technique requires only a small number of detector elements with minimal signal processing and is compatible with many real-time sensor applications. Measurements of the optical phase across a speckle feature are presented.

Process control for rapid thermal annealing (RTA) steps in silicon integrated circuit manufacturing is highly dependent on the repeatability of wafer annealing temperatures. The heating lamps in RTA ovens are controlled by an infrared pyrometer in a feedback-control loop. Methods currently in production allow some variability in annealing temperature, owing to the variability of wafer backside emissivities and the fixed emissivity value used for each process recipe. To provide an independent in-situ temperature sensor with potentially improved accuracy, an RTA oven equipped with a ripple pyrometer was used for passive data collection. Results obtained over extensive wafer processing indicate that the ripple temperatures have a correlation with electrical tester data and that the ripple pyrometer repeatability is better than 3°C at 1 standard deviation.

In this work we have examined the experimental low-temperature limits of 1.1μm pyrometry for the measurement of the temperature of silicon wafers and aluminumcoated silicon wafers at temperatures under 700°C in RTP chambers. In-situ emissivity correction in the same range has also been demonstrated with a single detector for radiation and reflection measurements. Temperatures as low as 450°C have been measured on metallized surfaces with an accuracy of better than 10°C without any a priori knowledge of the wafer emissivity.

We present simulation and experimental results for determining the distribution of radiation throughout an RTP system and evaluating the design of optical temperature sensors. The simulations are performed with a general purpose, three dimensional Monte Carlo method. The simulation models internal reflection, absorption, and transmission within participating media explicitly, and includes wavelength, temperature, and material dependent properties. Simulation results are compared to measurements of the lamp intensity profile within the lamphouse and at possible sensor locations. Simulations are used to examine the effect of several optical probe designs. The predicted responses of potential sensor designs resulting from these studies are presented.