One of the goals of the Substrate Process & Technology group at IBM in Rochester, Minnesota is to provide disk programs with consistently high quality substrates. A substrate inspection system that uses laser light has been designed, built, and is being used at Rochester. This paper explains the optics and detection scheme of the inspection tool called the Substrate Defect Detector (SDD). The SDD has been useful in studies to improve disk quality and to solve substrate problems in both manufacturing and development environments. Some results of the studies are reported and explained. Several examples of how the inspection system has been used to help solve substrate manufacturing process problems are discussed. The design objectives, operation, and use of the SDD are explained.

This analysis examines several sources of random and systematic errors present in most interferometer systems. Optical cavity errors are typically the primary limitation on measurement accuracy. Secondary sources of error include imaging distortion, ray-mapping errors, and detector noise. When necessary, differences between phase and fringe measuring systems are addressed. The analysis is kept as generic as possible for both fringe and phase measurement, though quantitative analysis of some error sources is instrument dependent. Such analysis is directed toward the Zygo MARK interferometers and Zygo ZAPP processor. The use of reference subtraction, distortion calibration, and averaging for improving both accuracy and precision are also discussed. Ray-mapping analysis indicates practical limits on the amount of tilt, power/focus shift, and asphericity in the test piece or aberrations in the wavefront of the light source, which may be tolerated without significantly degrading measurement accuracy. The role of absolute testing techniques is also discussed.

Small topographic variations of a spinning surface (e.g. disc, wafer) cause a reflected laser beam to change in two fundamentally distinct ways, in pathlength and in angle. This paper describes an experiment which compares surface topography scanning using each of these aspects. A laser beam heterodyne scanner was built to profile pathlength variations along a track. This instrument was modified to incorporate a beam position sensing capability using the same beam that is projected to and reflected from the surface. This per-mitted comparison of the two approaches using the same track topography. The topographic velocity of a beam spot on a spinning surface is given by pathlength variations as a function of time. The signal processor of the laser beam heterodyne interferometer has an output that is directly proportional to this velocity. This voltage relates directly to the circumferential slope along the track for a given radius. A quadrant sensor, indicating position of this same reflected beam, outputs a pair of voltages which indicate the orthogonal components of absolute slope. The quadrant sensor is oriented so that one of these components is circumferentially aligned with the track and can be directly compared with the heterodyne slope scan. Both approaches demonstrate excellent sensitivity for the surface waviness encountered, and within certain limits are not affected by reflectivity variations of the surfaces tested. Each approach has several features and advantages. The heterodyne approach is calibrated both to the laser wavelength and the flat electronic transfer function of the frequency-to-voltage converter used. It is capable of sensing topographic velocity over a broad dynamic range and wide frequency span. The reflected beam deflection approach has a simple optical design, has a second channel for radial slope, is immune to vibration, and is independent of the traversing speed.

In situ mirror distortion measurements were made with a lateral shearing interferometer on three mirrors in beam line X17T at the National Synchrotron Light Source. Lateral shearing interference is insensitive to vibrational motion in five of the six degrees of freedom, so it is well-suited for investigations in the synchrotron radiation (SR) environment. No distortion was seen in an uncooled silicon carbide mirror and in a cooled copper alloy mirror on X17TB, but a change in the radius of an uncooled electroless nickel-plated aluminum cylinder mirror of about 6.2% was observed on X17TA. Angular vibrations in the 2-3 arc second range were easily observed on one of the beam lines, as was an overall mirror rotation in the arc second range.

The fabrication of an R:1 demagnifying ellipsoidal mirror to be used for an x-ray microprobe at the National Synchrotron Light Source X-26 beam port. The design aim was to produce a mirror that could be used over the photon energy range from about I to 17 keV. The 300-mm long mirror was required to operate at a grazing angle of 5 mr. The semimaior axis was 4500 mm and the semiminor axis 14.142 mm. Surface roughness of 1 nm or less and slope errors of 1 arc second parallel to the long axis and 2nn arc seconds parallel to the short direction were specified. Production of the first electroless nickel-coated aluminum mirror using a diamond-turning technique has been completed. The mirror meets the 1 arc sec surface figure specification except for areas near the ends of the mirror. The reasons for these deviations arise from subtle details of the diamond-turning process which have not been fully incorporated into the computer program that controls the diamond-turning machines. Further work in computer correction of repeatable errors of the diamond-turning machine can eliminate the waviness at the ends of the mirror. The diamond-turned mirror surface was not fully polished under this effort and therefore does not meet the roughness specification; however, surface smoothness of a fully polished cylindrical mirror manufactured using the same techniques does meet the specification. It can be concluded that it is now technically feasible to meet the required specifications for the mirror and that the x-ray microprobe based on its use can be achieved.

The light scattered by optical components is a sensitive indication of component surface and bulk quality. Angular scatter measurement as an indication of component quality has been reported by a number of investigators since the early 1970's [1-4]. Most of these measurements are limited to angles outside a near angle cone centered about the reflected (or transmitted) specular beam. Measurements are more difficult at near angles because the instrument (scatterometer) source optics create scatter near the specular input beam that is superimposed on the sample scatter. This source scatter is often referred to as instrument signature and its presence must be considered when near angle scatter measurements are taken. The minimum measurable angle depends on the difference between instrument signature and sample scatter. Usually these two are of comparable size somewhere in the region of 0.50 to 50 from the specular beam. Inside this limit, sample scatter is referred to as near, small or low angle scatter. Outside this limit, sample scatter is referred to as far, large or high angle scatter. Because the scattered light density increases rapidly as the specular beam is approached, reduction of near angle scatter is a design concern for many modern optical systems. At least four investigators have reported scatter measurements in the near angle region [5-8], using various techniques to reduce or account for instrument signature.

The technique of scanning tunneling microscopy has been applied to topographic mapping of two optical surfaces -- a ruled grating replica and a diamond-turned gold mirror. By taking measurements with both a scanning tunneling microscope and a conventional stylus instrument, we have compared profiles and power spectral density (PSD) functions calculated from the profiles of a grating replica. Furthermore, surface structure was observed and PSD's were calculated for a diamond-turned surface measured with a STM. No structure was detected by the stylus instrument due to the spacing of the grooves on the diamond-turned sample. These measurements yield information necessary to gaining a better understanding of the diamond-turning process.

The design of a long-trace surface profiler for the non-contact measurement of surface profile, slope error and curvature on cylindrical synchrotron radiation (SR) mirror is pre-sented here. The optical system is based upon the concept of a pencil-beam interferometer with an inherent large depth-of-field. The key feature of the optical system is the zero-path-difference beam splitter, which separates the laser beam into two colinear, variable-separation probe beams. A linear array detector is used to record the interference fringe in the image, and analysis of the fringe location as a function of scan position allows one to reconstruct the surface profile. The optical head is mounted on an air bearing slide with the capability to measure 38" long aspheric optics, typical of those encountered in SR applications. A novel feature of the optical system is the use of a transverse "outrigger" beam which provides information on the relative alignment of the scan axis to the cylinder optic symmetry axis.

Profile and area measurements of the roughness of a given surface will generally be different since the two measurement techniques are sensitive to different areas of surface-frequency space. We explore the magni-tudes of these differences by calculating the ratio of rms roughness values, σ(TIS)/ σ(Wyko), using strawman models of a Wyko profiling microscope and a Talandic integrating scatterometer applied to surfaces having different roughness power spectra. As expected, the results show that this ratio can vary widely about unity, with values depending on the magnification of the objective used in the Wyko microscope and the "color" of the surface spectrum. An amazing counter example appears to occur for surfaces having an approximately "1/82" BRDF, or equivalently, a 1/f profile power spectrum -- shapes which are frequently observed for non-metallic mirror surfaces. In this case the predicted TIS and Wyko roughness values are essentially identical and inde-pendent of the Wyko magnification. This equality, however, comes from a curious mathematical-numerical coinci-dence and does not mean that these apparently "universal" values represent any intrinsic finish parameters of the surface being measured. In fact, if the Wyko data are filtered to remove the contributions from surface wavelengths longer than those included in the TIS measurements in order to more nearly match the instrumental bandwidths, the calculated ratio of measured rms roughness values increases to 1.5 to 5, depending on the Wyko parameters used. These results illustrate the fact that any realistic comparison of profile and area measure-ments of surface finish requires a knowledge of both the instrumental transfer functions and the form of the power spectrum of the surface being measured. The present paper discusses these issues and provides analytic machinery for the detailed quantitative comparison of profile, TIS and BRDF measurements.

Cochran and Wyant have recently proposed the extension of profile measurements by overlapping a number of successive colinear traces to generate a composite trace which is an order of magnitude longer than any individual measurement. This paper presents an error analysis of this method as a guide for its application and improvement. Expressions are derived for: a) the cumulative rms error of the composite profile, b) the trade-off between this error, the number of traces required, and the degree of overlap of successive traces, and c) the power spectral density of the cumulative error. This last is important in the application of the method since -- as expected -- it predicts that most of the error is concentrated in long surface wavelengths, in precisely the region the method was designed to illuminate. The present analysis suggests that the performance of the method may be improved by using a global fitting procedure rather than the serial method originally considered.

Frequency-domain analysis offers several advantages when considering surface profile measurements. Vibration of the sample during the measurement is often clearly defined in the profile frequency spectrum and its effect on the measured data can be reduced by an appropriate filter. The finish of surfaces with periodic height variations (such as diffraction gratings) can be measured if the periodic structure is first removed. This paper examines the application of frequency filtering techniques to these problems.

We describe two ways which we have developed for obtaining phase information in Type 2 (confocal) microscopes: heterodyne interferometry using a Bragg cell for both scanning and frequency shifting, and a novel "heterodyne phase contrast" technique using a time-variable electro-optic phase plate, somewhat similar to a Zernike phase plate with mechanical scanning. The heterodyne interferometer can measure amplitude to 0.2% and phase to the 0.1° level at a rate of 30,000 to 50,000 points per second, with exel -lent vibration immunity. The heterodyne phase contrast system is slower, but is a single-beam system and could be retrofitted to a standard Type 2 system without affecting normal operation. If scan stage vibration can be neglected, for example in acoustic wave detection, it should be able to measure optical phase to around 0.1 millidegree; however, this accuracy is not yet demonstrated.

The use of a 2-frequency laser allows precise interferometric measurements to be made by observing the phase of the detected beat frequency, rather than the intensity of a conventional interference pattern. This allows both improved resolution and insensitivity to low-frequency source intensity fluctuations. This paper describes two different implementations of 2-frequency interferometric micro-profilometry: the first is based on a conventional Michelson interferometer configuration, and the second is a common-path technique which senses the spatial derivative of the surface profile and which is largely insensitive to vertical displacements of the surface relative to the optical components. These instruments can be used to measure either profiles of reflective surfaces, or optical thickness profiles of transparent objects placed on a plane mirror.

A technique for evaluating surface microtopography is described which is based on the image analysis of optical micrographs yielded by Nomarski DIC microscopy. The technique evaluates slopes of surface asperities by calculating an intensity ratio of two Nomarski images of the same surface area. The images differ only in the intensity distribution across the microtopographical features as produced by different phase contrast conditions in the Nomarski optics. The image formed by calculating the intensity ratio on a pixel by pixel basis from the two corresponding DIC images is shown to be independent to first order of variations in sample reflectivity and sample illumination. Topographical data yielded by the present technique are compared with data obtained by commercially-available contact and optical profilometers.

This paper presents surface roughness measurements for a variety of surfaces using an automated Nomarski type profiling instrument. The instrument provides a quantitative measure of the surface slope and surface roughness of a surface. The vertical resolution of the Nomarski technique allows surface height evaluation with a resolution of one Angstrom and a lateral resolution of approximately one micron. A scan length of up to 100 millimeters can be used. The profiling instrument measures slope and calculates the surface profile. Both slope and profile descriptions of the surface are useful in examining the topography of the surface. Examples of the surfaces presented include a laser gyro mirror and a diamond turned mirror. Statistical surface parameters are shown for these surfaces, including the power spectrum and the autocovariance function. The statistical parameters are calculated from the both the slope and profile data with a comparison of the results.

This paper concerns the properties of a set of ellipsoidal x-ray mirrors manufactured for use at the National Synchrotron Light Source at Brookhaven. The objective is to compare the results of functional tests made at x-ray wavelengths and at glancing incidence with predictions based on laboratory measurements of their surface shapes made with a Wyko profiling microscope. The functional tests of the fully-illuminated mirrors indicated unacceptable image widths of roughly 300 μrad. Subaperture tests involving 1.5-mm-long segments of the mirror surface gave images which consisted of 2 to 5 sharp sub-images with separations of 17 - 50 μrad and individual widths comparable with the measurement point-spread function. Profile measurements indicated that the surfaces had a strong periodic ripple with an rms amplitude of about 85 Angstroms and a period of about 1.7 millimeters. When these ripple parameters are fed into the Fresnel-Kirchhoff diffraction integral we pre-dict that the idealized image should break up into fine structure consisting of approximately 7 lines, uniformly spaced 17 pm apart; in excellent agreement with the results of the subaperture functional tests. The wash-out of the sub-image fine structure in the case of the fully-illuminated mirrors is attributed to the presence of longer-wavelength surface errors than are included within the bandwidth of the Wyko measurements. The present analysis is unusual in that it involves the prediction of the effects of shape errors with amplitudes which lie between the smooth-surface limit, where the intensity in the image plane is a mapping of the power spectral density of the error, and the rough-surface limit, where it is a mapping of its slope distribution function.