A technique for varying the isotropic magnification of a scanning ring field projection mask aligner has been developed. This makes it possible to completely compensate for scale changes that occur as a consequence of the integrated circuit manufacturing process and temperature changes from one masking step to another. Since magnification errors are often a major source of overlay error in full field projection systems, correction of magnification errors on a wafer-by-wafer basis substantially improves the overlay accuracy. The Micralign Model 500 instrument contains a number of optically weak fused silica refractive components. The main purpose of these elements is to extend the degree of optical correction over a much wider annulus than could be obtained in an all-reflective design. An additional property of the design is that a pair of the refractive elements known as strong shells, which are symmetrically disposed about an intermediate image plane, can also provide correction for a variety of optical characteristics, one of which is magnification. Small motions of the strong shell elements in the same axial direction can readily provide magnification changes on the order of 25 parts per million. In order to produce an isotropic magnification change across the wafer, any departure from unity magnification must also be accompanied by a corresponding linear shift in the relative positions of the mask and wafer as they are scanned through the annular field of the projection system. This is provided by a precise, pneumatically-controlled micropositioning stage which also provides accurate registration control during alignment. Design characteristics of the shell adjustment mechanisms, the microstage, and the control algorithms are presented along with test data showing a +25 parts per million isotropic scale change.

As alignment tolerances and the associated overlay characteristics become more important to microlithography, automatic alignment has proven to be necessary for high yields and consistent throughput. This paper describes an Automatic Alignment System designed for, but not limited to, projection printers. In this scheme a vidicon is used to detect the images projected from special targets on the mask and wafer. The video information is procesed by a microcomputer system that performs a pattern recognition algorithm to determine target position and the necessary alignment correction. The system closely resembles operator alignment in concept, yet gives greater accuracy and more consistent performance. Operational principles and actual performance are discussed.

As feature sizes and overlay tolerances shrink to keep pace with the increase in integrated circuit complexity, there is a concomitant need to advance the methods used to characterize and compare the performance of lithographic instruments. The electrical probe measurement technique has emerged as the most accurate, most effective approach to lithographic evaluation. We demonstrate the versatility of the technique as applied to state-of-the-art 1:1 projection printers. Among the characteristics of projection printers discussed are linewidth uniformity, focus and exposure latitude, pellicle mask protection, and distortion and alignment contributions to overlay accuracy.

The factors influencing registration accuracy in a scanning 1:1 optical projection system are investigated. The overall registration accuracy in several versions of this system is evaluated. Data was gathered either optically, using Vernier patterns, or electrically, using an automated measurement technique.

As microlithography progresses to smaller geometries, the tolerance available for exposure variation is rapidly diminishing. Due to this requirement, it is advantageous to set the exposure more accurately by using an exposure meter that has its spectral response matched to the photoresist spectral response. A new exposure meter has been developed that contains a complete UV spectrometer in a package the size of a 100mm wafer chuck. This meter can be adjusted to match the response of any photoresist over the wavelength range of 254 to 436 nanometers.

The change in the size of a feature due to the proximity of large features is explored experimentally and theoretically, and the effects of nonuniform light source intensity are examined theoretically. For proximity effects, a test mask was designed to provide test patterns for two different systems: the Perkin-Elmer "Micralign" model 220 and the Canon FPA-141 4x reduction projection printer. The mask consists of isolated features of various widths which pass parallel to large features at various spacings. Exposures made on the Perkin-Elmer showed proximity effects dependent on the amount of partial coherence present in the light. For relatively incoherent light, proximity effects are intuitive and cause linewidth and spacewidth changes for small features. For relatively coherent light, the proximity effects can be counterintuitive in that the width of a small feature can be reduced due to the presence of adjacent features. Proximity effects are particularly significant for small geometries in the range of 0.5w/(NA) and below, where w is the wavelength. Some mask design guidelines are offered, suggesting that better overall image quality may be obtained not with coherent illumination, but with a partial coherence factor of about 0.5. Several types of condenser aperture illumination sources are explored. Central obscuration is shown to be disadvantageous for traditional optical lithography where the fundamental image spatial frequency is below that of coherent cutoff. A four spot Gaussian source is shown to give image quality similar to that of a uniform source of the same diameter. It is shown that source asymmetry cannot contribute asymmetrical resist profiles in a diffraction limited system, and thus any resist asymmetry must be attributed to proximity effects.

A brief characterization of the Ultratech Model 900 wafer stepper is presented. Excellent control critical dimensions for 1-gum minimum features is accomplished by using a broad spectral bandwidth for exposure which minimizes standing wave effects and by using numerical aperture of 0.315 illuminated with coherence factor of 0.45. Minimal variation in linewidth is seen over 5000A to 8000A poly and metal steps with little evidence of standing wave patterns in the resist profiles. A large depth of focus is obtained with a highly corrected 1:1 lens design which keeps astigmatism and field curvature below 0.5um. The automatic site-by-site alignment system on the Model 900 has proven extremely reliable at all wafer levels with a repeatability better than 0.16um (2 sigma). Lens-to-lens distortion below 0.2um (2 sigma) results from the inherent symmetry in the folded 1:1 design and from careful lens fabrication. A precision lenedistortion test is described with a 2 sigma error below 0.04um and the overlay distortion for the three Ultratech lenses is presented.

Photomask limited yields in LSI and VLSI processes are examined in this paper. Mask defects are classified into two catagories i.e., soft and hard mask defects. Theoretical modelling indicates a substantial yield improvement with pellicle protected masks. In 1:1 projection technology soft mask defects are the predominent cause of mask limited yield. Use of pellicles eliminates the effect of soft defects up to 100 microns in size, does not cause a degradation of image quality or dimensional control, prolongs mask life, and saves considerable labour and cost in maintaining high quality masks. Pellicle mounting, inspection, and handling techniques used are described. Very large die with 3 micron, NMOS, Si-gate technology are used to determine the actual yield improvement. Lots were processed using identical sets of masks of which one set was pellicle protected. Defect density at each process step and final probe yield are statistically analysed to show the individual contributions of hard mask defects, soft mask defects, and random process defects to the overall device yield. Actual yield increase data is presented. This pellicle technology is directly applicable to 10:1 stepping exposure systems where high soft mask defect density could be a more severe problem than in 1:1 projection systems.

A simplified but general model of wafer throughput for wafer steppers has been developed. The model shows clearly the throughput dependence on full-wafer overhead time, area of exposure field, alignment times, exposure time, and stage stepping time; trends are quite clear as to which parameters carry the most leverage, and which should therefore receive the resources for development. The model is used to compare the relative merits of two Zeiss lens series, a 5X wide-field lens and the standard 10X lens with regard to the anticipated IC production demands over the next five years. A case is made for the 5X lens, which can then later be replaced by the 10X lens.

Design criteria for wafer stepper alignment systems are discussed. Several examples are presented. A glossary of terms is included.

A simple model of the deviations in position between an ideal grid and a corresponding grid imaged through a photolithographic projection lens is presented. The model describes the dependence within the image field of origin translation, rotation, lens magnification, lens trapezoid, and lens distortion. The model has been used to analyze the image placement behavior of a Zeiss reduction lens mounted in a David Mann wafer stepper. The method of measurement of image field placement deviations relies upon the overlay of optical position verniers. One vernier half is arrayed by step-and-repeat to form an ideal grid in the image plane, while the other vernier half derives from an ideal vernier array in the object plane projected through the reduction lens. Vernier readings then yield placement deviations throughout the image field of the lens. A single reticle is used to generate this data, and its design will be briefly discussed. These measurements have been made utilizing a GCA 4800 DSWtm wafer stepper to construct the ideal grid. The lens evaluated was a Zeiss 10 78 06 5X reduction lens. The results of these measurements are then fit to the model in a least squares fashion. This allows separation of the mechanisms leading to placement deviations and assesses their relative importance. The overall measurement and fitting procedure is capable of detecting changes in the model coefficients which correspond to image field deviations as small as .03pm at the image field edge. Image field deviation values for a single lens will be shown along with the data fit coefficients and residuals. Systematic errors relating to generation of the ideal grid, as well as lens behavior outside the scope of the model will be discussed.

EUROSTEP 2000 is a new wafer-stepper oriented towards three increasing requirements coming from integrated circuits manufacturing : throughput : in order to be faster than modified step and repeat cameras, EUROSTEP 2000 uses a stage made of a sandwich of aluminum honeycomb and carbon fiber, at least as stiff as metallic stages but much lighter. This allows high accelerations (0,3 g) and a maximum speed of 200 millimeters/second. Joined to a short exposure time (<150 ms) and a short alignment time, the high stepping speed provides the same throughput as 1 to 1 projection aligners. repeatability : For mix and match requirements, all variations of metrology between two identical or different equipments must be avoided. For positioning, EUROSTEP 2000 uses HOLOGRID, a proprietary X - Y holographic grating interferometry. The holographic grating is a stiff replica of a master standard, with short term and long term repeatability, not affected at all by pressure, humidity, and fast variations of temperature. resolution : EUROSTEP 2000 is oriented towards manufacturing of micron and submicron devices. The present lens is able to print 0.7 micrometer lines and spaces everywhere in the 1 cm2 field. This lens could be replaced by another one, according to wanted field-resolution trade-offs, without any other change in the machine, including the die to reticle automatic alignment device. In order to allow submicrometer printing in 1 cm2 field, die by die autoleveling compensates for wafer defects of flatness, when necessary.

The operation of the Philips Wafer stepper and the automated measurement techniques used to characterize it are reviewed. Characterization data in terms of registration accuracy, lens performance and throughput are presented. Finally, our processing experience with this stepper is described.

A process control vehicle is described which allows the characterization and comparison of wafer steppers with respect to distortion, resolution, uniformity, and misregistration. A block of test structures consisting of optical resolution patterns, verniers, and electrical line width and misalignment resistors is arrayed on an 11 x 11 grid which fills the entire available field of a 10X reticle. Fach block also contains a pair of targets for the THE laser-interferometric auto-alignment system. The ability of the auto-aligner to acquire such targets to within 500 is exploited as a metrology tool whereby the measured coordinates at each site are compared to the ideal (theoretical) coordinates to generate a vector distortion map across the field. Subsequent reduction of misregistration data is accomplished via application of the six parameter model developed by Perloff and co-workers. It is shown that these diagnostic tools permit the rapid characterization of distortion anisotropy for a given stepper and can be used to optimize and monitor level-to-level regis-tration. Further applications are suggested.

Wafer stepper technology is now accepted as an important tool in the generation of VLSI circuits. More than with most products, the evolutionary nature of steppers is clear. The early systems were based on mask-making equipment, lacked automation and had few features suitable for the VLSI production environment. Systems have changed substantially since then and are now more compatible with manufacturing requirements. This paper will discuss one of the more advanced automated systems available, the TRE 800 SLR, and its engineering evaluation in the fabrication of VLSI circuits. Emphasis will be placed on overlay accuracy including field-by-field registration, matching of steppers and mixing with other aligners. Finally, an automated track process for this stepper will be described.

We report here the use of automatic die-by-die alignment to overlay 10:1 wafer-stepper images on patterns delineated by 1:1 proximity and projection systems. The motivation for these programs arises from considerations of cost-effectiveness and productivity while upgrading existing IC fab lines; 1:1 lithographic systems already in place can be used to generate non-critical layers, while the wafer steppers can be used to align and delineate the patterns of critical layers. Excellent intra-die overlay match is demonstrated, mixing with some of the more prevalent 1:1 systems. Practical issues relating to net wafer throughput are also illuminated.

The ability to view an area of a wafer and 10X stepper reticle, simultaneously on axis, has been used to match different 10X reduction wafer steppers. This paper defines the problem and discusses test results. Operational parameters, global alignment, and intra-field die alignment are discussed. Wafer distortion and field distortion are examined. Registration data gathered from visual and computer diagnostics, which support the conclusions, are discussed.

A wafer step-and-repeat alignment system, presented in 1980, is briefly reviewed 1, followed by a discussion emphasizing the aspects of cross-matching to wafer global alignment instruments. Issues relevant for cross-matching (wafer pre-alignment and exposure field fine-alignment) are covered in detail. Basic performance data for resolution, alignment accuracy, and throughput are given and first cross-matching results are reported.

The mixing of 10:1 direct step on wafer aligner to a 1:1 full mask projection aligner has often been proposed.1 This hybrid alignment scheme offers the advantages of excellent registration and high resolution found in the current steppers in combination with the established throughput and lower cost of the 1:1 projection aligners. This paper reports on fabrication of HMOS circuits using the hybrid technique at one critical layer. Full integration of the automatic global alignment features, field-by-field alignment, process control drop-ins and automatic framing has been accomplished. A technique for comparison of the registration between the conventional and hybrid process will be presented. Considerations of mask and reticle layout will be discussed. The influence of the additional time required for field-by-field alignment in a theoretical model will be presented.

Plasma etch equipment has made impressive headway in recent years, both technically andin the market place. The trend to ever finer lines and spaces in semiconductor manufacturing brought out the limitations of wet processing and has all but necessitated the development of dry methods. Indeed, the requirement for plasma processes arose so rapidly that the equipment andrrocess industries were not completely prepared. At the same time, the strong need and rapid market growth attracted a large number of companies to enter the business with a bewildering assortment of equipment and processes. These conditions have posed some difficult problems for the user, the wafer fabricator. On the one hand, due to lithographic resolution sliding toward 2 um and below, the semiconductor manufacturer finds being forced to purchase dry-etch equipment in growing volume. One the other hand, equipment and processes are still in development and new ones are being announced almost daily. The purpose of this paper is to improve communication between suppliers and users by surveying the user community and reflecting its unfilled needs. What is that the semiconductor manufacturer really requires in plasma equipment? How can the equipment maker improve its position? What are the problems with equipment already installed? What are the critical process parameters? What should the next generation equipment do and be like?

As the Semiconductor industry moves toward 1 micron and submicron lithography, it is critical that anisotropic dry-etching reproduce the pattern linewidth control of the lithographic tools generating the patterns. The strength of custom processing individual wafers as opposed to batch treatments lies in creating an identical chemical and physical environment for each wafer (absence of loading effects) and in individual process monitoring and control to compensate for variations among the wafers themselves: the result is exceptional process-consistency and pattern-transfer control. This paper will discuss the correlation between real-time optical process signature monitoring and post-etch patttern-transfer analysis. Elimination of contamination via load-locked reactor design and contaminant monitoring will be discussed, as well as new methods for precision control and maintaining consistent accuracy of process parameters such as pressure, gas-flow rates, residence time and RF power.

The requirements for high-fidelity, anisotropic etching are discussed. A brief discussion on the various dry etch techniques available is followed by a description of a batch reactive sputter etch system that fulfills all patterning requirements. Examples of results obtained with this system are given and demonstrated with SEM photographs.

Measurement systems for determining the exposure and development model are coupled with SAMPLE resist profile simulations to explore the impact of positive photoresist types, processing conditions, developer types, and resist lots on line edge profiles. The exposure system uses the optical transmission of a resist layer on a nonreflecting glass substrate as a function of exposure. A full spectrum, high intensity "hard" bleach is used to completely expose the photoactive compound. The companion development system uses optical interference through a transparent substrate to measure the development rate. The reproducibility of these systems and their accuracy compared to independent measurements are considered. These systems are applied to the problems of determining the critical prebake temperature at which thermal degradation occurs, exploring differences between developers, and making lot-to-lot comparisons of resists. Parameter values for several resists are reported. The importance of the exposure and development parameters is illustrated using the SAMPLE program.

At the Semiconductor Microlithography VI Conference, March 30-31, 1981 a new high speed positive resist, WX118 was described with particular reference to its "properties" (i.e.; 1.) absorption spectrum, 2.) film thickness vs. spin speed, and 3.) optimum softbake and hardbake conditions), and was compared to Hunt's HPR-204 in "conventional" UV exposure, using both contact and 1:1 projection systems. Recent studies with WX-118 resist imaged with a GCA 4800 DSW Wafer Stepper have shown, with different developer systems (i.e., LSI Developer, MIF Developer, WX-ll6 Developer, etc.) structures as small as 0.4 microns wide with near vertical sidewalls. SEMs are presented to demonstrate the superior resolution capabilities of WX-118 resist Plus high photospeed in step and repeat systems. Data are also presented on the modification of WX-118 image profiles from vertical sidewalls to undercut sidewalls. Preliminary information is supplied on the increase in WX-118 softening point by the employment of a post-development UV exposure.

This paper describes the use of computer profile simulation and scanning electron microscopic analyses in a study of the combined performance of the mid UV Perkin Elmer exposure tool and AZ2400 resist. New generations of UV resists, including AZ2400, demonstrate a development phenomenon termed the "induction effect" which tends to impart improved performance. The induction effect has been characterized by a new development model that imparts a depth dependence to the rate function. Computer simulated resist profiles generated by SAMPLE, modified to include the induction phenomenon, have been compared with SEM studies of experimental data. The extent of agreement between simulation and experiment over the variables of dose, developer strength, aperture and linewidth is good and is shown to provide a predictive capability which allows rapid and convenient establishment of the effect of process variables on performance.

This presentation describes a newly developed and very innovative reticle inspection system, the purpose of which is to make a comparison between the reticle pattern and the data base stored on the Mag Tape. Also, this paper will include the operational report from a couple of semi-conductor companies where the systems are already installed and are operating very successfully. To verify the reticle pattern as compared with the data base is becomming extremely important for the subsequent process; especially in the upcomming Direct Step on the Wafer era. The system consists of three parts: X-Y stage unit, Control unit, and Video conversion unit. With respect to the inspection principle, the actual reticle pattern is converted into a Video signal by utilizing the combination of the mercury lamp and a 1024 bits linear photo diode array. This signal is shown on the monitor #1 of the Control unit. The data stored on the PG tape is converted into its own data format which is fed into the Video conversion unit for the conversion into a video signal and then displayed on the monitor #2 of the Control unit. The system processes and compares the two video signals electronically, and any descrepancy between two images is identified as the defect and displayed on the monitor #3. The defect detectability is approx. 2 to 3 microns at present and it requires approx. 5 to 6 min. to inspect a 50 mm2 pattern. The system is capable of providing all the necessary information regarding the detected defects, such as its Location and kind to the plotter, repair system (Zapper) and other peripherals by either On-line or Off-line methods. One of the key features of this system is that all the defects detected electronically can be reviewed and confirmed by the human eye. The use of this system is expected to result in a higher device yield.

Automatic photomask inspection systems have been commercially available for a number of years. KLA Instruments has manufactured systems which utilize die-comparison to detect photomask defects. This paper describes some of the technical modifications and enhancements which augment the basic photomask inspection capability of a die-comparison system by adding the capability to inspect a single-die reticle against the data base which generated it.

It is becoming apparent that optical lithography will remain the dominant imaging technique in the production of semiconductor devices for some time to come. Production devices with a minimum feature size of 1.5 microns have been manufactured using lx, 5X and 10X wafer steppers. Lenses are presently being designed with higher numerical apertures and promise a resolution of one micron or less. Full field exposure instruments employing mid to deep ultraviolet illumination also promise a resolution down to one micron or less. It has taken the better part of a decade to perfect technologies in the 3-5 micron region, and it will probably take the better part of the present decade to perfect technologies in the 1-2 micron region. If optical lithography can serve our requirements for a 1-2 micron technology, photomasks will remain an important part of the overall photolithographic process. In order to achieve a 1-2 micron technology, some considerable demands will be made of the photomask manufacturer, especially in the area of overlay accuracy.

This paper describes techniques to detect submicron pattern defects on optical photomasks with an enhanced direct-write, electron-beam lithographic tool. EL-3 is a third generation, shaped spot, electron-beam lithography tool developed by IBM to fabricate semiconductor devices and masks. This tool is being upgraded to provide 100% inspection of optical photomasks for submicron pattern defects, which are subsequently repaired. Fixed-size overlapped spots are stepped over the mask patterns while a signal derived from the back-scattered electrons is monitored to detect pattern defects. Inspection does not require pattern recognition because the inspection scan patterns are derived from the original design data. The inspection spot is square and larger than the minimum defect to be detected, to improve throughput. A new registration technique provides the beam-to-pattern overlay required to locate submicron defects. The 'guard banding" of inspection shapes prevents mask and system tolerances from producing false alarms that would occur should the spots be mispositioned such that they only partially covered a shape being inspected. A rescanning technique eliminates noise-related false alarms and significantly improves throughput. Data is accumulated during inspection and processed offline, as required for defect repair. EL-3 will detect 0.5 um pattern defects at throughputs compatible with mask manufacturing.

An electrodeless microwave powered deep UV source has been developed. This source pro-duces a unique intense spectrum between 190 and 260nm. Deep UV radiated power output efficiencies exceed 10% at microwave power input levels of 1500 watts. The lack of elec-trodes avoids output attenuation due to electrode sputtering that can cause deep UV fall-off and early bulb failure with compact arcs. Optical systems have been designed to utilize this source for effective flood irradiation of wafers with deep UV having good intensity, uniformity and collimation characteristics. One application of this source is to the portable conformable mask (PCM) technique for meeting submicrometer resolution requirements over profiled surfaces. In this application the exposure time to clear a 2 micrometer PMMA layer using the microwave powered deep UV source is less than 30 seconds.

New laser ellipso-interferometry has been developed on the basis of the principle that spatially and temporally coherent light reflected from the two surfaces of a thin solid film will interfere to form a two-dimensional fringe pattern which is directly related to the spatial distribution of the film thickness if the optical constants are uniformly distributed. This technique differs from the conventional ellipsometry and interferometry in that the new laser ellipso-interferometer can be used to map the thickness distribution of a whole thin film without involving the use of any scanning technique, and that its spatial resolution is high and can reach the diffraction-limited resolution under certain conditions. In this paper the principle and the applications of this technique are described and some experimental results are presented to demon-strate the use of this technique for checking the thickness uniformity of oxide layers in semiconductor devices. This new technique is non-destructive and the time required for this kind of measurements is much less than that by any other conventional techniques. Furthermore, no computation is involved for checking the film thickness uniformity if both the film and the substrate are transparent or low-absorbing to the laser light used for the measurements.

Linewidth control is one of the most important requirements in integrated circuit fabrication. Usually, linewidth variations after each processing step are measured with a microscope containing an eyepiece in which a scale is engraved. The precision obtained by this method is very low. To achieve a high precision, electronic image processing of optical microscopic images offers fast, non-destructive, and repeatable readings. However, these tools are expensive. They require frequent calibration to ascertain consistency of the absolute reading. A correction increment has to be established and added to the apparent linewidth for each line thickness, the composition of layers, and microscope settings such as illumination coherence, focussing condition, numerical aperture, lens aberrations, and the intensity threshold selected to define the electronic linewidth.

A new technique for fine-line, high-speed photolithography using ultraviolet excimer lasers is proposed and demonstrated. Absence of speckle and resolution down to 1000 line-pairs/mm are experimentally demonstrated. Using a XeCl laser at 308 nm and a KrF laser at 249 nm, excellent quality images are obtained by contact printing in two positive photoresists. These images are comparable to state-of-the-art lithography done with conventional lamps, the major difference being that the excimer laser technique is -2 orders of magnitude faster. Preliminary results on reciprocity behavior in several resists are also presented.