RESULTS AND DISCUSSIONS

Secondly, the purchase and use of optical equipment without either specific optical test data furnished by the instrument manufacturer or optical acceptance tests by the user is common practice. Consequently, optical instruments are used for fabricating small dimensional geometries on photomasks and IC devices without adequate information about the performance capabilities of the optical system. The absence of optical testing is not limited to the IC community; although, the large percentage of capital investment in optical equipment and its extensive use by this industry dramatizes this deficiency. One reason for this absence of test data is that optical components mounted in photolithographic instruments do not usually lend themselves to isolated testing. Optical performance is governed by such fundamental properties as the pupil function (wavefront error), the optical transfer function (OTF), resolution, depth of focus, and the presence of aberrations. These properties can be obtained by interferometric lens testing [21, 22] prior to mounting the lens in the system. In the absence of such testing, the quality of photomasks and IC devices is limited by an unknown factor of optical performance.

Finally, it should be noted that some U.S. and foreign optical companies have responded to the needs of the IC industry by providing optical systems specifically designed for photomask and IC processing. Such systems include minimum image distortion over large fields, lens-aberration corrections for the wavelengths of the light used to expose emulsions and photoresists, and materials which are dimensionally stable within the operating environment. It should be noted that small-volume IC equipment manufacturers generally do not have the financial means to have these special lenses designed and cannot guarantee a sufficient market to assure the lens manufacturer the recovery of design and manufacturing cost. Therefore, optical components in some photomask and IC processing equipment are sometimes off-the-shelf items that were designed for other purposes and constitute a strong compromise in performance and cost.

4.2. Dimensional Measurements

Photomask and IC producers are generally operating within the same technology for dimensional micromeasurements that has existed since the early development of IC's around 1960. Basically, this approach is a human operator collecting visual readings through a microscope fitted with a micrometer stage or measurement eyepiece. As photomask and IC patterns have approached 1-um line widths, these optical measurements have involved larger discrepancies and errors. As in the past, most photomask and IC producers have continued to rely on commercially available

RESULTS AND DISCUSSION

instruments to satisfy their measurement requirements. Only a few IC facilities with research and development capabilities have developed any in-depth programs to address their own particular measurement needs.

Generally, the approach adopted by the instrument manufacturer to the increasing micromeasurement problem is to improve the mechanical repeatability, or precision, of the equipment and to replace visual observations and manual operations with automation and electronics. This approach is necessary for improved measurement precision, but is not sufficient to obtain the accuracy that is needed for measurements approaching 1 um. However, it is important to note that existing optical measurement equipment could be used satisfactorily for accurate measurements down to about 1 um under the following conditions: (1) knowing the operation limitations and error values with different types of illumination and samples; and (2) calibrating the system with a material length standard, or artifact, that is directly related to the standard unit of length.

One attitude adopted by some IC producers has been to assume that dimensionalmeasurement problems can be tolerated provided that the final IC device performs satisfactorily. This approach may be costly and time consuming by requiring the producer to test each IC device and to select as acceptable only those devices meeting performance specifications. Since the manufacture, inspection, and use of photomasks occur early in semiconductor processing, it appears far more desirable to detect defects, including dimensional errors which would lead to device failure, in these front-end processes rather than at the final test station. With early inspection techniques, the unit cost of the surviving devices can be much lower; moreover their reliability will be improved, and the ultimate yield can be increased significantly since the photomask is a pattern for many individual devices. Of course, photomask inspection is routinely performed in current IC processing in order to achieve these desirable objectives. However, such inspection is of little value if existing measurement difficulties or limitations lead to dimensional-measurement inaccuracies and eventual device rejection or failure.

The major dimensional-measurement problems associated with photomask technology in the IC industry can be grouped into the following categories: (1) measurement of small dimensions in the micrometer and submicrometer range; (2) edge definition or location of a physical edge for a line; and (3) registration, or relative alignment, of several photomasks used to form successive patterns on a single wafer. The common manifestation of these problems appears in the optical instruments used to fabricate, inspect, and register photomasks. Therefore, as mentioned earlier, the following discussion of these specific measurement problems will include the related optical problems.

RESULTS AND DISCUSSION

4.2.1. Measurement of Small Dimensions

As noted earlier, dimensional measurements are routinely made on photomasks to determine if design dimensions have been held within tolerances during mask processing. These measurements are generally line-width measurements made on several different parts of the circuit geometry.. Selection of these measurement areas is made during preparation of the initial photomask artwork. It is only necessary to make these measurements at specified locations over the entire mask because measurement errors are essentially not randomly distributed. These measurements should be made on different line widths at different locations to account for variations in optical quality over the exposure field of the instruments used to print the mask.

Line-width or oxide-window measurements are a length measurement from one edge of the line or window to the other edge of the same line or window. This type of measurement is clearly different from such linear measurements as scale calibrations which are basically line-spacing measurements. To further illustrate this difference, it is noted that the present NBS line standard interferometer [23] for linear-scale calibration provides accurate measurements between centers of adjacent lines or from the edge of one line to the corresponding edge of the adjacent line, but not from one edge to the other edge of the same line. This important measurement difference is generally recognized by the IC community even though linear scales or other line-spacing artifacts, which have been calibrated from line center to line center or from left (right) edge to left (right) edge, are used routinely for line-width measurement calibrations.

Typical critical dimensions for photomask and IC device geometry using conventional photolithographic processing presently range from 2 um + 0.25 um for lowvolume production to 5 pm +0.25 um for high-volume production. Special devices such as microwave circuitry include 1 um and submicrometer geometry. These devices are generally produced on a best-effort basis, and the measurement inaccuracy is either unknown, unquoted, or so large that it approaches the nominal dimensions of the patterns. These measurement inaccuracies of ± 0.25 μm are often desired values rather than values that can be substantiated by relating the measurement to the standard unit of length. Also, these measurement tolerances sometimes include precision tolerances. Precision is usually defined [24] as a measure of the ability to repeat a given measurement, while accuracy is a measure of the difference between a given measurement and the actual or standard value.

It is often suggested that if all masks used to print a wafer are generated in the same system, the precision of mask measurements is all that is required; whereas, if several systems are used, then accuracy, or at least dimensional correlation, is required. As discussed earlier, this conclusion does not stand in view of the desir

RESULTS AND DISCUSSION

of many IC producers to predict final device performance by comparing circuit dimensions with design specifications. For this comparison, measurement accuracy must be known.

In another recent study supported by ARPA [25], it was reported that automatic photomask systems commercially available in the U.S. provide the capability for efficient production of photomasks with the levels of accuracy and precision needed for highly reliable devices and yields. However, this report does not point out that this accuracy is attributed to the equipment and not necessarily the generated photomask. The report also does not mention that the equipment accuracy can be assured only if the manufacturer calibrates the instrument by techniques, such as interferometry, that relate to the standard unit of length. In any event, the equipment user is not guaranteed that the accuracy quoted by the equipment manufacturer can be transferred directly to the photomask. Accurate dimensional measurements on the photomask must be provided by an independent method.

The majority of photomask suppliers and users rely chiefly on the optical microscope with the filar or image-shearing eyepiece to perform small dimensional measurements down to about 2 μm. Other microscope systems that are used less frequently include coordinate-measuring systems with translating stages and scanning microdensitometers. More recent developments to reduce human reading errors, measurement time, and visual strain employ electronic interfacing with the optical microscope; these system include automated dimensional readout or a video display with fiducial lines for positioning over the magnified image of the object to be measured. Advertised values of accuracy for these systems range from 0.1 um to 0.01 μm. It should be stressed that very often precision and accuracy are advertised as being the same value. The quoted error is termed an instrument error or accuracy by the manufacturer when, in fact, the values quoted are the precision. In some cases, the measuring-equipment manufacturer states that furnished linear scales, or micrometer stages, are traceable to NBS dimensional calibration. Other instrument manufacturers quote only the precision and accuracy of mechanical stages or other moving parts, but do not list the accuracy and precision expected with the actual dimensional measurements. Still other manufacturers quote only the instrument sensitivity which is usually the finest divisions on dials, verniers, or gauges.

The filar or image-shearing eyepiece replaces the normal eyepiece in an optical microscope to permit dimensional measurement of objects placed on the specimen stage. The filar eyepiece is basically a crosshair which is moved across the field of view by turning a micrometer drum, and the difference between two drum readings is related to the linear dimension of the object. A shearing eyepiece produces two identical images that are sheared or split by turning a micrometer drum; these two images are often filtered to give a red and green image. The amount of shear, or

RESULTS AND DISCUSSION

difference between the drum reading for zero shear with images superimposed and the reading for shear with the two image edges (left edge from one image and right edge from other image) just touching or slightly overlapping, is related to the object length in the direction of shear. Both of these measurement systems require visual judgement to position the crosshair and sheared images even though the readout of the drum settings may be automated as mentioned earlier.

One of these two measurement systems is usually preferred by users at each photomask and IC facility based on their requirements and experience. At some facilities, both systems are used routinely, and the choice for a particular measurement depends on the type of photomask and the measurement conditions. Sample types and measurement conditions used with both the filar and image-shearing systems include emulsion masks with low reflectance, highly reflective chromium masks, semitransparent ironoxide masks, opaque wafers, transmitted and reflected illumination, and filtered and polychromatic illumination.

Differences between filar and image-shearing measurements of the same geometrical feature on a mask vary from a few percent to over twenty percent throughout the IC industry. These differences most likely stem from the variable measurement conditions and sample types already mentioned as well as other factors such as optical differences and illumination coherence. The effort of some IC facilities to control tightly the measurement conditions results primarily in improved precision on a particular measurement system, but does not eliminate differences between the measurement systems. A preliminary comparison of filar and image-shearing measurements made at the NBS on a nominally 15-um diameter wire and a 10-μm diameter pinhole was included in the present study. A comparison of line-width measurements on a nominally 12-um width chromium line on a glass substrate was also included. The results of these measurements are reported in appendix A, and they represent some of the measurement differences typically found between the filar and image-shearing systems.

The primary difficulties associated with dimensional measurements of mask patterns appear to be line-width measurements, unknown accuracies, and operating limitations of the optical-microscope systems used to make the measurements. Linewidth measurements require locating the left and right edges of the line. Unless the physical profiles of both edges are identical, differences in edge location can be expected in measurement systems ranging from relatively simple microscopes to sophisticated interferometers. This fundamental measurement problem of edge detection is discussed further in the following section.

Measurement accuracy requires that the measurement be traceable to a defined or standard unit of measurement. This link is normally provided in the laboratory by secondary or working standards that are, in turn, related to the defining standard by a primary or reference standard. The two types of material standards in common use