A Monte Carlo program named SEEL (Simulation of Electron Energy Loss) was developed to simulate electron trajectories in arbitrary line geometries. Submicron scale electronic structures were fabricated in order to compare the simulation with experimental results. SEEL was used to determine the radial exposure distribution for electron energy dissipated in resist during electron beam lithography. A companion program called PRESTO (PRoximity Effect Simulation TOO was developed to predict the exposure of submicron scale patterns. Exposures were made for the purpose of comparing the experimental patterns with simulations.

Critical dimension metrology, the measurement of the dimensions of features on an integrated circuit, is a vital part of device fabrication technology. As the feature size requirements of ULSI devices continue to decrease below the practical limits of optical metrology, scanning electron microscopy (SEM) inspection and measurement will be utilized on a more routine basis during device fabrication. In semiconductor fabrication applications, metrology techniques are typically used for groundrule verification, in-line inspection of critical layers after photoresist development and after etch, pattern definition and sidewall profile of critical patterns, and determination of etch undercut. With the introduction of submicron design rules, the more established techniques of optical metrology I are inadequate and are rapidly being supplanted by metrology performed on the SEM2. The recent availability of SEM tools which can be operated in a low accelerating voltage mode for inspection of uncleaved device wafers without conductive coatings enables implementation of SEM metrology in a manufacturing environment. Much work is being undertaken to understand the parameters involved in measurement accuracy and precision. However, since radiation exposure to devices occurs during SEM metrology, there is also cause for concern that there could be resultant degradation of the electrical properties of the devices. Degradation has been generally related to generation of fixed positive charge, neutral electron traps, and fixed negative charge in the Si02 due to exposure to ionizing radiation.

Two techniques suited for high speed SEM wafer inspection are presented. The dependancies of beam focus variations on Linewidth Measurements (LWM) precision are reviewed. It is reasoned that high LWM precisions and high-speed operator-independent wafer inspection are mutually exclusive requirements. A more aggressive inference is made : existing SEMs may not satisfy the inspection and LWM production environment requirements of lines 0.5 micron or less. Revolutionary, new inspection techniques are needed in the SEM to make them as easy to use as today's light optics instrumentation. One such technique is presented, which is a laser-based Automatic Working Distance Control (AWDC) system. AWDC eliminates the need to re-focus the electron beam during wafer inspection. It also creates optimum scenarios for precise LWM and inspection in future sub-half micron production lines of 200 mm wafers. The second technique is an in-vacuum wafer transfer and staging mechanism. The mechanism is suited for very high speed SEM inspection, and integrates the technique described before. The design was implemented with judicious consideration of the sub-micron production requirements: high reliability, low contamination, high speed inspection, and low cost. The novel components of the design are described. Emphasis is made on a minimum volume and surface area vacuum lock, with no internal moving parts. Three commercially available vacuum pumps are evaluated. The outgassing load of 200 mm wafers with recently coated photoresist is quantified. Computer simulations show that high throughputs are feasible using standard vacuum engineering techniques. Throughput figures are given for "single-wafer" and "wafer-queueing" modes of operation. Operational speeds comparable to light optics tools are achievable.

A negative electron affinity (NEA) GaAs photocathode has been used to generate continuous and pulsed electron beams for electron microscopy and time resolved electron beam metrology. Short electron pulses are emitted from the photocathode when excited by a mode locked laser. Laser pulsing the cathode eliminates the need for complex electron beam blanking optics when performing time domain metrology. The GaAs photocathode exhibits a brightness comparable to LaB6 and is stable enough to display real time TV images in both the continuous and pulsed modes of operation. In the time resolved mode, the temporal resolution of the cathode is currently limited by the laser pulse duration, which is approximately 1 ns.

The National Institute of Standards and Technology has, for several years, been developing a metrological electron microscope system traceable to national standards of length. This metrology instrument will certify standards for the calibration of the magnification of scanning electron microscopes (SEM) and for the certification of artifacts for SEM linewidth measurement. These artifacts are not only directed to instruments used in the semiconductor community but will also be useful for the various other applications to which the SEM is currently being used. The SEM-based metrology system now operational at the Institute will be described as well as its design criteria and procedures for its characterization. The design and criteria for a new lithographically produced SEM low-accelerating-voltage magnification standard to be calibrated on this system will also be presented.

The magnification calibration method employed by the Nanolab series of instruments (U.S. patent application #117,741) uses a fast Fourier transform (FFT) of the line profile obtained from a calibration standard to accurately determine the system magnification. The standard used is a diffraction grating replica with certification traceable to the National Institute of Standards and Technology (NIST). The calibration routine is fast and simple, allowing the system to be easily recalibrated in less than 60 seconds. After calibration, the Nanolab maintains accurate magnification without the risk of errors associated with methods using compensation factors. This paper reviews factors affecting the accuracy and stability of electron optical measurement system calibration, and different approaches to system calibration. The Nanolab calibration procedure for critical dimension (CD) measurement is then described and the performance evaluated.

The advantages of electron-beam based metrology over photon-optical have been well established. Those advantages include (1) the ability to truly resolve feature ulgo of 0.1 μm or less, (2) the ability to accurately measure features with aspect ragas above 1 to 1, and (3) negligible sensitivity to thickness variationsl . A further advancement of e-beam based metrology is a novel method called. Spatial Metrology, introduced at this conference last year. Spatial Metrology is not based on the e-beam image but rather on a unique secondary electron signal. The major advantage of Spatial Metrology is accurate top and bottom linewidth measurement, without the need to cross section the wafer.

Application of the nondestructive Photocleave Technique for determining the shape and size of sub-micron contact windows defined in photoresist is described. Stepper-based printing in positive photoresist is assumed. Following conventional exposure of the contact windows in pass 1, the stepper is programmed to immediately execute pass 2, with an appropriate pass-shift, to expose a linear feature that sections the latent images of the contact windows. After development, a fast turn around SEM is used to determine the contact window parameters with an enhanced level of certainty as a result of using the Photocleave Technique.

Electrical linewidth measurement is well-known for high precision and throughput. However, the standard four-point probe testsite is only useful for measuring the width of an isolated conducting line. Line-and-space and isolated spaces can be simulated satisfactorily by adding dummy lines parallel to the active line but weak links or potential electrical shorting situations often prematurely cause these structures to fail before their true limits are reached. In this paper, fully wrapped proximity- and astigmatism-tolerant designs for line-and-space and isolated spaces are shown. They have been successfully demonstrated with printed images. An application in evaluating the exposure-defocus window of a one-layer i-line resist using the proximity-tolerant testsites is given.

The use of electrical defect monitoring for process control is described. Details are presented of the use of fast cycle time short flow snake lots for metal and poly processes in an advanced CMOS pilot production line. Contributions to defectivity from diffusion, CVD, and etch processes are described briefly. The nature and origin of three different types of photo process defects are discussed together with methods of eliminating these defects; a track develop system gave lower defect density than a batch develop system, and a higher numerical aperture stepper led to the reduction of a micro bridging defect mechanism. An application of snake processing to give improvements in a contrast enhancement layer photo technique showed that contrast enhancement layer strip time was a key factor in improving yields. The need for an integrated photo monitoring system consists of snake patterns and other forms of inspection is discussed.

The use of electrical test patterns is examined as an analysis tool for electron beam (e-beam) pattern writing systems. The method makes extensive use of electrical linewidth measurement structures and pattern redundancy to determine pattern placement accuracy and fidelity on a single reticle layer. Significant improvement in resolution and repeatability over conventional optical based measurement systems is apparent. The quantity of data that was collected quickly and automatically provides an accurate and large-scale view of subtle distortions caused by both the e-beam and reticle processing.

This paper describes how a technique was developed to make precise and accurate submicron linewidth measurements on photomasks which takes advantage of the inherent benefits of both optical and electrical measurement techniques. We present the theory of operation for both the optical and electrical equipment used, and the theory behind the electrical test structure design. Test mask development includes the materials (types of chrome and glass), processing, and test structure design. The optical and electrical test procedures are discussed, with emphasis placed on set-up and sample preparation. Test results are summarized and conclusions are drawn based on data analysis. Finally, we show how this integrated technique of submicron measurement of photomasks is applied in a production environment.

This paper presents results from an experimental study of the effects of resist side-wall slope variations on the linewidth measurement accuracy of a Confocal Scanning Laser Microscope (CSLM), as compared with a Scanning Electron Microscope (SEM). Both the top and bottom dimensions of patterned resist features from 0.51 μm to 2.0 μm wide were measured using CSLM, cross-sectional SEM and on-axis SEM. Substrates tested included bare silicon and three thicknesses of oxide on silicon. The resist was patterned with a range of deliberate stepper defocus which provided a 15° variation in side-wall angle. Data for the three methods is compared and shows that in this case the CSLM demonstrated good measurement accuracy for independent top and bottom width measurements. Errors due to resist sidewall slope variation were small, random and consistent with the experimental noise level.

The present trend in submicron critical dimension process control is toward in-line SEM systems. Recently several new optical tools have been developed for application in the submicron environment. Optical metrology offers many well publicized operational advantages and in general higher throughput than an SEM. This paper reports on the application of a new optical metrology technique based on coherence probe imaging using a Linnik interferometer with incoherent, broad band illumination. Linewidth measurement is performed on three dimensional images produced by determining the degree of mutual coherence in the reference and object planes of the interferometer. The mutual coherence function is evaluated on a pixel by pixel basis as the sample is moved in the z direction. A commonly encountered problem of coherent optical metrology arises from interference produced by the interaction of the coherent light wave with material of varying optical path length. Incoherent systems offer a higher degree of immunity to material induced phase interference but suffer from poor resolution. The use of coherence probe imaging increases the resolution while preserving the positive aspects of incoherent illumination. The ability of the coherence probe microscope to perform accurate edge detection in a submicron IC fabrication process subject to a wide degree of process variance is compared with the in-line SEM. The effect of material thin film variations have been evaluated with respect to their influence on measurement accuracy and precision. Response surface models are described which illustrate the effect of thin film material factors on coherence probe measurement accuracy and precision. These factors include relative film reflectance, thin film thickness, and feature line width. The effects of resist slope variations and proximity of measured feature to other features is also evaluated. Measurement precision has been evaluated using component variance analysis to isolate the sources of measurement error relative to a 0.3 μm process tolerance.

The suitability of the Real-time Confocal Scanning Optical Microscope (RSOM) for semiconductor process metrology had not previously been examined. Recent improvements in the RSOM have enabled us to make precise measurements on weakly reflecting sub-micron geometries. Data will be presented to demonstrate the lateral and depth resolution of the improved microscope, which will be compared to theoretically predicted results. Preliminary data from critical dimension measurements of photoresist on silicon, and photoresist on silicon dioxide wafers, for linewidths in the range of 0.7 to 2.6μm will also be presented.

Accurate CD linewidth and overlay measurement are dependent upon the ability to correctly find edges of line objects. Software algorithms used to find edges often have difficulty distinguishing the edges of features to be measured from surrounding noise and coherent edge ringing. The problem becomes even more difficult when an optical scanning microscope operates in a highly coherent mode and enhanced edge ringing and interference effects are present. Simple edge detection algorithms such as thresholding and steepest slope become confused and may select erroneous artifacts. As a result, optical measurement systems have leaned toward high NA illumination systems and broad spectral bandwidth, both of which reduce coherent edge ringing but are accompanied by a loss in accuracy. Alternative techniques such as correlation and adaptive signal processing also have difficulties finding true edges due to the variation in image profiles resulting from changes in materials, layer thickness, edge slope, etc. This paper describes an alternative approach based on a combination of mean-square integration (MSI) and adaptive techniques with function parameters determined from the digitized waveform. This approach has been successfully utilized on experimental data and appears to work for a wide variety of signal waveforms. Theoretical and experimental results are given.

Much has been written in the literature concerning the importance of linewidth measurement in controlling the lithographic process. Numerous sources of variability in both the process and the measurement system itself complicate linewidth measurement and control. For this reason, linewidth measurement standards have been developed for use in calibrating systems to provide information as to the precision and variability of the systems in terms of 3-sigma values. In a manufacturing environment however, the measurement system used to determine the final critical dimension is not necessarily the same system that was used to measure the particular dimension during processing. This raises a question concerning the compatibility of the measurement systems. It is therefore desirable to acquire an understanding of the significant differences, if present, between different measurement systems. This can be accomplished by examining the statistical correlation of measurements and by employing statistical models to identify, quantify and compare sources of system variability independent of process variability. Statistical correlation analysis offers a method for assessing the relationship of linewidth measurements and repeatability of different systems. In addition, system variation on factors influencing CD measurements can be modeled using variance components analysis, quantified and evaluated as to significance. Properties to be modeled include repeatability, agreement on accuracy, and precision for various feature types and sizes. The statistical models developed may also be used for calibration purposes. Statistical properties for both standard and alternative calibration procedures will be discussed. Briefly discussed will be Response Surface Methodology for obtaining combinations of factors which tend to minimize linewidth variability.

An automated system has been developed at the National Institute of Standards and Technology (NIST), formerly the National Bureau of Standards, for calibrating optical photo-mask linewidth standards. This system, controlled by a desktop computer, locates each feature to be measured in the field of view of the microscope, centers and focuses the image, scans the image, and calculates the optical linewidth from the scan data. The results are checked for errors and the process repeated until every feature on the photomask has been calibrated. If statistical tests are passed, a calibration certificate is printed.

With some companies now ramping up production of the 4 Mbit DRAM, the era of submicron device technology has begun. To produce each new device with an acceptable yield, there is a requirement for a decrease in defect density in comparison to previous generations of technology as shown in the figure 1.

As defect densities decrease, ever increasing sample areas will be required for statistically valid estimates of defect densities. Several tools exist, such as electrical defect monitors (serpentines and combs), that can test a large area but provide little information about the defects detected. At the same time, automated visual inspection techniques (KLA 2028) exist that inspect smaller areas with greater detail. We propose a strategy for analyzing large areas with great detail by combining both of these tools. The necessity of obtaining visual information for process evaluations is discussed. Experimental results are described in support of this strategy.

As IC groundrules shrink, manual optical inspection of multilevel patterned wafers becomes ineffective if not impossible, and efforts to develop automatic wafer inspection systems have expanded. This paper describes one successful approach, the P300 Automatic Wafer Inspection System[1], which uses a greylevel reference comparison of adjacent cells to locate defects on periodic pattern. The defects may be either pattern anomalies or particulates. Experiments demonstrate that the P300, scanning at a rate significantly faster than a human inspector, finds over ninety percent of half micron defects and over ninety-five percent of defects one micron or larger. By basing the inspection algorithm on a cell-to-cell comparison within a frame, as opposed to the conventional chip to chip or chip to CAD database reference, the system avoids detecting false alarms caused by acceptable variations in reflectivity, film thickness, critical dimensions and overlay registration over the surface of the wafer. A simple cell-to-cell comparison, however, would he prone to detecting false alarms due to electronic and digitization noise, aliasing, vibration, and illumination non-uniformity, as well as small scale acceptable process variation. By adding a statistical test to filter out noise and an edge detector to reduce sensitivity on edges, the false positive rate has been kept below a fraction of a percent of the frames inspected. The paper will discribe the system architecture and inspection algorithms and discuss specific inspection applications.

The implementation of full wafer inspection at each point in the fabrication process has resulted in a new methodology for tracking and then eliminating defects. The three-dimensional inspection, made possible by use of holography and optical spatial frequency filtering, means that a recorded defect is never out of focus. Full wafer inspection is therefore a volume, rather than an area, technology. Now, wafer inspection is not limited to the less complex topography of some levels, but includes levels such as trench, contact, via, metal, and passivation. A defect partitioning technique has been developed for the purpose of: (1) pinpointing the source of each defect and, (2) determining if the defect results in permanent damage to the circuit and subsequent yield loss. Wafers are inspected at all critical points in the process, with and without photoresist, and the defect maps are stacked. Special algorithms have been developed to process data at each level and to subtract from it defects from all previous levels. This procedure isolates the defect maps that are unique to each layer with the precise X, Y coordinates for each defect. Classified defects may be assigned a color and mapped from the printed paper report, identifying clusters as they appear in a single layer and tracking them through partitioning studies. Using this method of layer subtraction, and by using color for the study, defect clusters and individual defects can be tracked throughout the process. A variety of analysis programs will be presented including the results from partitioning studies completed with fabricated VLSI circuits.

Currently, wafer steppers are capable of routinely meeting a single machine overlay accuracy of 150 nm or less. However, this figure is relatively meaningless in a production environment equipped with multiple machines. The question is how to exploit single machine accuracy in a multiple stepper production environment such that production logistics will be independent of machine allocation. This paper describes the use of a matching management system supported by an overlay simulator and a reference wafer set. Every stepper is measured on a regular basis with respect to a reference wafer to determine possible deviation from the nominal grid. The reference wafers are resistant to multiple use. The simulator can be used to estimate the matching performance between non-measured machine pairs so that the need for particular adjustments can be determined. The theory is completed by an extended overlay budget justified by experimental results and a ten machine overlay experiment.

Step and repeat camera optical systems today exhibit greater resolution, numerical aperture and field sizes than in the past. It has become necessary to control lens distortion and matching errors to less than one-tenth micron across the entire usable field. The quality of the optics has risen to the point that fifth order distortion modeling has become a necessary component of the stepper matching procedure. Methods of data gathering and analysis vary significantly in the industry. In this paper we investigate four methods of overlay measurement including electrical, optical coherence probe, automated optical and optical vernier techniques. Precision to tolerance ratios and throughput for the measurement methods are discussed. Three methods of data analysis are compared including KLASS II for KLA 2020* data, EM1** for electrical data and SASO*** for both. A novel multi-substrate calibration technique is presented. In any matching situation the required sample size for accurate estimation of the lens components is important. Our study reviews the results of analyses of variances due to daily repeatability, wafer films, the number of measurement sites on each wafer and stepped field. Two methods of artifact generation were investigated, that of stage referenced matching and matching to a 'golden standard'. Finally, the expansion of the model to include seventh order distortions and the significance of this for the now emerging high numerical aperture, large field g-line lens designs is discussed.

A procedure has been developed for the analysis of the performance of photolithography systems. This procedure uses statistical methods to characterize the system. This is done by using small number of process observations to determine the values of coefficients of an empirical model. The predictions of the model can then be used to determine such factors as the full process volume and anticipated process stability. The method is demonstrated by a case study of an evaluation of a multi-dimensional process space.

A computer aided engineering workstation is described which provides for the comprehensive characterization and control of registration performance over an entire process, from stepper set-up through layer-to-layer inspection. The workstation accepts data entered manually and from a number of automated metrology tools. The measurement layout is user defmed and supports the diverse sampling needs of set-up and production. Layouts can be defmed to obtain grid and intrafield placement data simultaneously. Data obtained from all sources can be displayed, manipulated, and analyzed in a user-friendly workstation environment that has been optimized for direct visual and numeric comparison of lithography data. Registration modeling, grid and intrafield, is performed to relate the measured quantities to assignable causes, predict worst case overlay, and to allow for the trending of error components. The workstation can be connected to many steppers and metrology tools at once to provide a registration control center. The application of the workstation to stepper set-up and production line monitoring is demonstrated. Use of the workstation as the registration control center for a production line is discussed.

An advanced manufacturing photolithographic process has been developed and implemented. Utilizing conventional processing techniques the process demonstrates high levels of consistency and uniformity while maintaining simplicity. The 1.5 micron process as monitored by electrical metrology techniques has demonstrated long term consistency of exposure latitude and critical dimension uniformity. All aspects of processing including coat, expose, develop and etch contribute to a total variation of less than 10% of nominal linewidth. Statistical process control combined with equipment matching has facilitated process standardization and allows for the elimination of all pilots and set-up wafers bringing high throughput ASIC manufacturing to a reality.

A framework for process optimization and characterization was discussed and applied to the development of an 0.8μm i-line lithography process for CMOS manufacturing. Response surface and characterization data were presented. Metrics for process optimization were discussed. Depth of focus windows for i-line and g-line lithography were compared for resist materials of similar capability run with optimized processes. Depth of focus data on 3 different stepper types were used to draw conclusions: a 0.38 NA g-line and 0.40 NA i-line from Manufacturer 1, and a 0.48 NA g-line from Manufacturer 2. I-line resists from 2 different manufacturers were seen to have similar depth of focus characteristics. Maintaining acceptable wall angle for the resist profile was found to be a more severe constraint on the depth of focus than maintaining critical dimension control. The i-line resist offered better wall angle than g-line resist, but less global process stability. At center of field, i-line lenses and currently available i-line resists have effectively 10 to 20% more depth of focus than g-line lenses and g-line resists at comparable resolution.

A computer program has been developed for the purpose of estimating process sensitivity and improving process robustness. In this paper the linewidth of polysilicon gates as a function of the photolithographic process parameters is analyzed. First, a model of the linewidth is obtained using the method of Statistical Design of Experiments and the Response Surface Methodology (RSM). Using such models, the program computes the breadth of linewidth variations that can result from simultaneous variations in the process parameters. This method of doing the sensitivity analysis offers significant advantages over one-parameter at a time studies, Taguchi methods, and contour plots of desirability regions.

This paper describes a technique used to determine an optimized microlithographic process using statistical methods which included a statistically designed experiment (SDE); a desirability function, d(θ*); and a rigorous daily statistical process control program, (SPC).

Photoresist linewidths are affected by many factors such as resist film thickness, solvent content, exposure dose, etc. As design rules shrink below 1 μm, quick and accurate determination of linewidths not only becomes more critical, it becomes more difficult. Fiber optic based reflectivity measurements during spray or spray/puddle development provide a process sensitive method of wafer inspection which may be used to optimize equipment and process setup. Critical dimension uniformity can be optimized by studying the uniformity of development in the inspection area. The correlation between changes in an end point uniformity measurement and the final linewidth is shown as a function of exposure dose variation, development spin speed, and developer dispense rate. A methodology is presented for determining optimum developer usage and equipment setup for a spray process. Reflectivity measurements provide a diagnostic capability as it provides a window into the resist film's dissolution profile. A study of the differences between development signals can reveal such processing problems such as resist lifting, scumming, and overbaking. Examples of several of these cases are given.

A novel methodology for development endpoint detection is presented which utilizes a polarizing optical head (Tritec Industries) to monitor dissolution of exposed photoresist by interferometry. We have found significant improvement in the interferometric signal modulation when using circularly polarized illumination and detection. This result is attributed to a reduction in ambient light effects and to rejection of depolarized light due to scattering from resist sidewalls, aerosol droplets, suspended particles, and bubbles in the developer. Additional improvement in the interferometric signal modulation has been obtained by monitoring an infrared wavelength at which the developer absorption is minimized. This approach tends to ameliorate the problem of "red cloud" opacity that sometimes plagues single-wavelength laser endpoint detection systems. Interferometric data at 660, 830, and 890nm wavelengths are compared using robust fast Fourier transform and finite difference algorithms.

End point detect during photoresist develop has the potential to reduce the variation of line width caused by variations in coat, bake, develop, wafer reflectivity, air temperature etc. This paper will describe a model for the reflectivity signals used to detect the end point during develop. Examples will be given of signal variation caused by a number of different process variations.

Resist and chromium features on electron beam generated photomasks ranging from 0.4 to 10. μm were measured with a classical scanning slit optical microscope, a confocal scanning laser microscope and a low -voltage scanning electron microscope. The scanning slit microscope was used in a calibrated way. The results obtained with those instruments are compared mutually and with the designed values. The accuracy of the measurements is evaluated using a model for electron beam exposure of resist in order to explain trends observed in the measurements. The results show that these systems can be used for mask metrology of sub -micron features with mutual differences of about 10 nm. The resolution limit of the optical systems appeared to be 0.4 μm.

Recent studies show that white light, or broadband, metrology systems continue to be used as the "work-horse" method for measuring features at 1.0 micron and below in production manufacturing environments. With the advances in the area of broadband optical metrology using sophisticated digital-image enhancement techniques, production quality measurements are now possible down to 0.5 micron and below. The purpose of this paper is to show substantial evidence that reliable, fully automatic, high speed, dynamic mode measurements can be made using optical methods on substrates in the range of 0.5 to 0.3 microns, with excellent repeatability and correlation to a SEM reference.

We have proposed a novel phase-shifting and Fourier transform technique for linewidth measurement. The novelty of the technique is the transfer of linewidth information into the zero-order spatial frequency component of the image. The method we proposed incorporates phase-cancellation and spatial Fourier transform to achieve resolution beyond the Rayleigh diffraction limit and should be applicable for metrology of sub-micron features. Theoretical analyses have shown that this technique will be viable well into the submicron range. Experimental measurements have been made on different features having relatively large dimensions (~ 100 μm) to verify the theory. Experimental results have demonstrated the validity of the concept and given the directions of practical limitations and requirements of the approach in order to obtain true submicron linewidth measurement. Such limitations include consideration of the precision of lateral translation of phase mask with respect to substrate, and of the mask to substrate distance. Further, extensive simulations of the robustness of this technique with respect to material composition, contrast and feature morphology (slope and height of the linewidth features) have been carried out for both individual and periodic features.

The predominant method used in the past for the measurement of overlay has been manual reading of the "optical vernier." This method can be reasonably precise and has been sufficient for most semiconductor products made up until a few short years ago. The ever-increasing number of masking levels below 1.5-micron Minimum Feature Size (MFS) requires large statistical bases of overlay measurements with quick turnaround. No longer are 4 to 10 sites per wafer sufficient to accurately judge overlay, nor can we afford to wait 20 minutes for an operator to manually read these verniers. For years, Perkin-Elmer has used a unique and proprietary electrical probe system custom-built by Perkin-Elmer prior to the introduction of the Micralign Model 500. Capable of gathering large amounts of data and performing statistical analysis, it became a standard for overlay evaluation within Perkin-Elmer. An alternative to electrical probe is automated optical measurement. One such system is the Perkin-Elmer OMSTM. This system has the advantage of being "non destructive" and can be used to measure actual product wafers in process. This paper will provide a performance comparison of both techniques, optical and electrical. Using a mask with both optical and electrical probe patterns, a series of wafers was exposed. The evaluation compares accuracy, precision, speed, and statistical capabilities.

When registration requirements exceed conventional optical inspection capabilities, the need arises for a more precise method of inspection. The Perkin-Elmer OMSTM overlay measurement system has a specified precision of 0.05 micron, 3 sigma. After exposure and development, a production wafer can be quickly and precisely measured, utilizing non-destructive methods employed by the OMS instrument. After the wafers are measured, corrective inputs can be easily entered into lithographic tools to improve overlay and yield. A statistical approach to data analysis was used to establish the actual level of dynamic precision or repeatability and the reproducibility of the overlay measurement tool. For this study, X and Y distortion measurements taken on production wafers were analyzed. Wafers were printed and developed by normal production methods, using a Micralign M544 projection printer and an AEBLE 150 direct-write E-beam system. Samples from these lots were then used to establish the intra-tool repeatability by repeated measurements on 45 sites on each of three wafers for five days. Another sample was used to establish the inter-tool reproducibility by measuring 441 sites on each of two overlay measurement systems and statistically comparing the data. The results of these analyses will be presented, as well as the methodology.

The rapid technological change in the VLSI industry has resulted in a constant upgrading of measurement equipment. One question to be asked is whether the upgrades recommended really improve the measurement system. Precise measurement equipment is one of the most important components in the next generation of VLSI technology. A systematic approach to measurement equipment upgrades in one micron technology can save much grief and remove uncertainty. In order to compare three optical CD measurement systems simultaneously, a statistically designed systematic approach was employed. The major contributors of variation were identified and quantified. The precision of each optical CD system was then compared. Findings from the study showed the upgraded system reduced variability associated with machine repeatability by a third, but only reduced overall measurement variation by a tenth. The same methods used here can apply in most cases where one piece of equipment is evaluated or several are compared. Vendor claims can be easily tested through the approach described. Reductions in measurement variation associated with an upgrade can be actually quantified allowing management to weigh benefits against costs.

Recently several groups have reported about the production of half micron devices using SOR X-ray lithography /1,2,3/. The device performance was compared to those devices produced by using other lithographic techniques (e.g. electron beam direct writing). However, for future applications it is necessary to fulfill the more stringent subhalf micron design rules. One important challenge is the linewidth control within several nm and the impact on yield, device quality and overlay accuracy.

Measurement of submicron device structures on semiconductors requires increased accuracy from metrology equipment. Scanning Tunneling Microscopy (STM) techniques are being used to meet these needs. Operating in air, these devices generate 3-D images of surface features with a resolution similar to that of a Scanning Electron Microscope (SEM) operating in vacuum.

A method for measuring the flatness of mounted stage mirrors is presented. This is the first part of an effort to map stage mirrors mounted on an X-Y stage for the purpose of subtracting mirror flatness errors from distance measurements. The absolute surface flatness of an optical straight edge is determined using an interferometric three flat test. This calibrated mirror is measured against the stage mirror using a displacement measuring interferometer (DMI) system. A sample stage mirror is measured using a three flat test. The measured flatness is compared against that obtained using the DMI system. Results to date are presented. The accuracy of the system is currently limited by the accuracy of the three flat test, environmental factors, and second order rotation sensitivity of the interferometer.

The continuing trend toward smaller geometries in IC lithography is placing increased demands on the displacement transducers used in wafer steppers and inspection machines. Among the needs are improved accuracy, repeatability and resolution. This paper describes a new displacement measuring interferometer which has resolution twice that of conventional plane mirror interferometers as well as improved accuracy and repeatability. These improvements are achieved by double passing the measurement arm and optically compensating the thermal drift error inherent in other plane mirror interferometer designs. This drift arises from small changes in ambient temperature and if not eliminated can be significant even in a temperature controlled environment. The origin of thermal drift error in plane mirror interferometers is analyzed and comparative test results are discussed.

Optical measurement of contacts and vias is possible because when a microscope is focused at a plane near to the bottom of a contact, light is reflected from the bottom and forms a characteristic image: a dark ring surrounding a brighter center. The dark ring image is common to a variety of substrates, including resist on poly, etched metal, nitride, resist on doped glass on grainy metal, etc.. Using image processing and dedicated measurement algorithms, it is possible to separate the dark ring from the brighter center by a "contour" or edge, and to measure the diameters (in x or y) and the area inside the contour. Also, by comparing these measurements against reference measurements it is possible to determine if the contact is open or closed. These measurements can be repeated with precision between .01 and .02 microns (depending on materials) and can be calibrated to a standard such as a SEM or an electrical test.

An automated patterned wafer inspection system has been developed with submicron sensitivity which is capable of inspecting a complete 150mm wafer in less than three and a half minutes. The system was designed to be used on-line in a production environment to assist the process engineer in identifying critical yield limiting process steps. The system can be interfaced to an automated off-line defect review microscope which allows for the classification of defects. An overview of the optical system and unique signal processing techniques will be presented. Performance results on a variety of different process levels on various types of patterned wafers will be discussed.

Increasing requirements for quality assurance and process control dictate the need for non-contact, non-destructive film thickness measurements in semiconductor manufacturing. Accurate and repeatable film thickness measurements, ranging from <10 nanometers to 50 micrometers can be achieved with a high resolution reflection spectrophotometer on both patterned and un-patterned wafers. Dedicated measurement techniques are required for thin layer research and Integrated Circuit manufacturing (automatic patterned wafer measurements), coatings on optical elements (layers with minimal differences in refractive index), Compact Discs, Liquid Crystal Displays (measurements on transparent substrates), as well as thick layer measurements on polyimide, resists, etc. Correlation of the measured interference intensity spectrum with the spectrum theoretically calculated from the assumed dispersive optical data of the layer materials can be used to investigate the effect of dispersive changes in the refractive and absorption indices on the measured film thickness.

An apparatus designed to monitor the development rate and thus control the feature size of electron beam exposed positive-acting resists on photomask substrates is presented. Development rate monitoring and development end-point identification is accomplished using real time optical interferometric techniques. This system is interfaced with an Applied Process Technology (APT)0 resist spray-spin development processor. The interferometric signal is collected and processed by an automated computer system, and is used to determine the required total development time necessary to obtain the correct critical dimensions on photomask substrates. Initial testing of the apparatus consisted of monitoring and controlling the development of electron beam exposed poly(butene-l-sulfone) (PBS) films on different reflectivity photomask substrates. For specific sets of processing conditions, calibration plots are generated by using the penultimate extrema of the interferometric development trace of an exposed center pad as a time reference point. Results have indicated that due to exposure proximity effects, it is necessary to generate separate calibration plots for a feature when it is isolated vs. when present in a periodic array. These calibration plots were tested for their validity using a series of developer concentrations (70/30 to 90/10 v/v % 5- methy1-2-hexanone/2-pentanone) and under varying relative humidity (12 to 40%) conditions. For previously defined geometry types, features in 1.25 to 4.0 μm range were within ± 0.05 μm of their coded sizes. These results represent a clear indication that stringent control PBS processing conditions and the need for iterative development can be eliminated, thereby providing a certain degree of automation in the photomask fabrication process.

Recently developed statistical inference methods permit detailed process information to be developed from experiments without relying on particular physical process models. Generalized Cross Validation is a powerful calculational method that can be used to find the smooth function that best approximates noisy, scattered experimental data. Already employed in applications in such fields as geology, biology, and meteorology, its three principal benefits for process engineering are: - The method does not require a parameterized model of the experiment, so it can be used in situations where no model, or no sufficiently detailed model, is available. - The results of the calculation are expressed in terms of piecewise polynomial or polylogarithm functions, rather than the global polynomials, which are used by traditional Response Surface Methods, that can fail if the experimental data cannot be described well by a polynomial. - The method can be used effectively in situations where the errors in the experimental results are not known precisely. These qualities make Generalized Cross Validation (GCV) a very attractive tool for analyzing and optimizing processes. We present an overview of the GCV method with two examples that show how it can be applied to solve problems in process modeling and optimization. In our first example, we efficiently determine the best values for the GHOST proximity-effect correction parameters (defocused-beam diameter and dose) by using the GCV method to fit and interpolate experimental data. For the 50-kV electron-beam system we studied, the best values were predicted to be a defocused beam diameter of 20 μm and a dose of 34 per cent of the pattern dose. These results are in good agreement with earlier work that employed a much more time-consuming method. In our second example, the GCV method is applied to interpolate experimental data to predict the developer concentration and prebake temperature that produce maximum contrast for a particular deep-UV resist.

The purpose of this paper is to present a statistically designed approach to identify and characterize the observed interactions between hardbake and softbake on thermal softening and the ensuing post-hardbake feature deformities. A statistically designed experiment (SDE) was employed for the simplicity of experimentation and maximal results obtained with such an approach. The processing parameters evaluated and optimized in this SDE included softbake temperature (Ts), softbake time (ts), and hardbake temperature, (Th). The magnitude of the interaction between softbake and hardbake processing parameters was calculated according to a quadratic fit of the primary response attribute. Softbake and hardbake process steps were characterized and found to be synergistic with respect to their effects on the resulting post-hardbake microlithographic feature shape.

Post Exposure Bake (PEB) is one of the simple method for minimizing the influences of reflection from a substrate to a resist film. Although various references have been made to the heating process ,its mechanism has not been fully clarified yet and it has been difficult to forecast its effects accurately. To grasp PEB effects upon the resist performance exactely ,the next two kinds of change in the development rate by this process must be measured. (1) Change by PEB in the development rate curve for the remain of photoactive compounds. (2) Smoothing by PEB of the local development rate distribution. These data can be acquired by Development Rate Monitor. The first one is translated into the sets of the fitting parameters of the development rate curve in the lithography simulator. The second one will be represented by means of the thermal diffusion model. The change (2) has significant effects upon the lithography on highly reflective substrates (silicon, aluminum etc.), for these substrates generate large amount of standing wave in a resist film. Because the smoothing effect enhances the development contrast of photoresist, PEB process contributes to improvement of the lithography on these substrates. On the contrary, for low reflective substrates (silicon-oxide etc.) which give only small amount of standing wave the effects of (1) is relatively given more weight than the smoothing effect. Because the change (1) decreases the development contrast, PEB on low-reflective layers depresses lithographic performance. This report will simulate PEB effects, compare them with experimental data and clarify the relation between PEB performance and reflection of substrates.

An effective repeating defect detection program is essential for a successful stepper-based photolithography process. Repeating defects caused by problems with the reticle or stepper optics have a direct and dramatic impact on wafer yield. As die sizes increase and the number of die per field decrease, the importance of this inspection program increases to a critical level in manufacturing. To protect against yield-impacting repeating defects, a variety of automated inspection strategies have been implemented in lithography processes, including particle inspection of the reticle, metal on glass wafer inspection, and single layer resist on silicon inspection. Each of these techniques have limitations in the manufacturing area. A new technology, resist on quartz inspection, has recently emerged for the detection of repeating defects in high volume wafer fabrication. Production reticles are stepped onto resist-coated, transparent quartz wafers. After development, a repeating defect inspection can be automatically performed using transmitted light which provides optimum defect detection sensitivity and throughput. This novel technique uses standard wafer fabrication processes to produce the resist images. This paper discusses the implementation of resist on quartz inspection in a manufacturing environment. Focus/exposure matrices were run to determine the process window for generating resist images on quartz substrates and to confirm inspection performance on typical production resist images. The resist on quartz inspection strategy implemented in the production area is explained. Examples of repeating defects detected by the inspection program are shown. To confirm the capability of processing quartz wafers with an existing production lithography process, a 5X VeriMaskTm 1045 test reticle was used to generate resist images. The reticle has programmed defects of varying sizes, types and locations on one micron geometries. Test images were created with various exposure and focus conditions, and the programmed defects measured using a low voltage SEM. The resist images were inspected on a KLA-259 Resist Image Inspection System. Capture rate curves generated from the inspection and measurement data are shown.

There is a lot of inspection in the manufacturing of semiconductor devices. Generally, the more important a manufacturing step, the higher is the level of inspection. In some cases 100% of the wafers are inspected after certain steps. Inspection is a non-value added and expensive activity. It requires an army of "inspectors," often times expensive equipment and becomes a "bottle neck" when the level of inspection is high. Although inspection helps identify quality problems, it hurts productivity. The new management, quality and productivity philosophies recommend against over inspection. [Point #3 in Dr. Deming's 14 Points for Management (1)] 100% inspection is quite unnecessary . Often the nature of a process allows us to reduce inspection drastically and still maintain a high level of confidence in quality. In section 2, we discuss such situations and show that some elementary probability theory allows us to determine sample sizes and measure the chances of catching a bad "lot" and accepting a good lot. In section 3, we provide an example and application of the theory, and make a few comments on money and time saved because of this work. Finally, in section 4, we draw some conclusions about the new quality and productivity philosophies and how applied statisticians and engineers should study every situation individually and avoid blindly using methods and tables given in books.

A multi-stroboscopic sampling (MSS) technique was devised for logic state measurement with the electron beam tester. Electron beam pulses are shot and secondary electron signals are sampled m times each repetition period of LSI operation. In addition, an interpolated s-curve (IPS) method was introduced in the MSS technique for quantitative voltage measurement. Using this technique, the measurement time required for 1024 logic states was reduced by 1/70 compared to the stroboscopic waveform measurement technique.

The paper presents a machine vision architecture which can address a wide range of machine vision applications in the manufacture of integrated circuits. A brief review of current applications and the reasons for the success of machine vision in the industry are also presented.

As processes advance into production of submicron devices, reducing defect density to an acceptable level is becoming a more difficult task. To deal with this problem, new wafer inspection technologies have been developed. The new systems can inspect dense patterned wafers to identify particles and process defects. An improvement over manual inspection is realized in defect sensitivity, inspection speed, and consistency of results. The technologies available for automatic wafer inspection have different capabilities. Therefore, to take advantage of each technology, the methods for system utilization must be considered. The methods involve identification of killer defects and determining the problem cause and required corrective action. This paper will focus on typical defects found in a submicron manufacturing facility. An evaluation of new wafer inspection technology will be described. Examples will be given to illustrate how inspection technology can be applied to solve problems in a production line.