Conventional techniques of process control are inadequate for the production of modern hard surface photomasks because they do not accommodate all the variables and because, even utilizing the tightest controls practical, the total variability of the process exceeds the allowable variability of the final product. A system of interactive process control has been developed and put into operation. This system uses the characteristics of the outgoing product in a negative feed-back loop to keep the process stabilized and properly adjusted. A mathematical model of the process is used to convert product data into changes in process parameters. A computerized reporting system keeps the process engineer informed of changes in mask quality and yield.

The aim of this work is to determine different types of contributions to dimensional errors so that appropriate and selective improvements can be undertaken toward optimization of the photolithographic process. Preliminary results are given on contributions by the resist, lens, and errors due to focus and exposure. Systematic variations below O.lμm were found in many instances.

This paper was prepared for presentation at the SPIE meeting, "Developments in Semiconductor Microlithography II", held in San Jose, California, April 4-5, 1977. Most photolithographic masks have defects, and one of the most difficult tasks of the mask maker is to locate and identify these defects. A knowledge of the size, type, and location of defects is needed not only to tell the wafer process area how good, or bad, the delivered masks are but also to help the mask shop make better masks. An Automatic Mask Inspection System (AMIS) has been in routine operation for measuring the defect count on various types of masks at Bell Laboratories, Murray Hill, N.J., since February, 1975. This talk will give a brief description of AMIS and will describe its use in photomask manufacture.

Semiconductor yield can be greatly affected by defective photomasks. Various methods of detection of photomask defects have been investigated in the past. This paper discusses some methods available for the detection of random defects and describes a newly developed automatic mask inspection system.

In the current linewidth-measurement program at the National Bureau of Standards, the primary measurement of micrometer-wide lines on black-chromium artifacts is made with an interferometer located in a scanning electron microscope (SEM). The data output consists of a line-image profile from the electron detector and a fringe pattern from the interferometer. A correlation between edge location and fringe location is made for both line edges to give the linewidth in units of the wavelength of a He-Ne laser. A model has been developed to describe the interaction of the electrons with the material line and thereby relate a threshold value on the SEM image profile to a selected point on the material line. An optical linewidth-measuring microscope is used to transfer the primary measurements to secondary measurement artifacts; these artifacts will be used to transfer the linewidth measurements to the integrated-circuit industry. Linewidth measurements from the SEM/interferometer system and the optical linewidth-measuring microscope are compared, and the level of measurement uncertainty for each system is discussed.

A method for increasing the functional speed of positive photoresist is discussed. Optical linewidth measurement and SEM analysis are used to illustrate the effects of photoresist thickness, exposure time and developer concentration on photomask image reproduction. Finally, photoresist is related to other process parameters, including mask geometry reproduction, and a method for balancing these factors is suggested.

We have studied the solubilities of photodegraded positive photo resist systems such as AZ*1350, AZ*2400 and AZ*111 resists by using in situ resist film thickness measurement techniques. The photo resist films were exposed at the sensitive wavelength of 0.4047 pm at several exposure dosage levels. The aqueous developer systems used with the above resists were AZ* developer, undiluted; AZ*2401, diluted with DI water (1:4); and AZ*303, diluted with DI water, (1:5), respectively. Published analytical dissolution rate models describe a linear relation between film thickness lost versus development time for unexposed and uniformly exposed resist films. We present results that demonstrate the prebaked resist film dissolution are not, in general, linear in time. Also, the dissolution behaviors are different with the different resist-developer systems we studied. Furthermore, the optical exposure effects are different from resist to resist in a manner that can be explained by the photoresponse of the resists at the exposure wavelength.

The chemical nature of positive working photoresists is examined in the choice of various components and their effect on the practical performance characteristics. The contribution of several potential resist components, alone and in combination, is described.

Direct measurement of lateral distortion in (100) silicon wafers reveals random shifts as large as 0.5 μm resulting from high-temperature processes commonly used during the manufacture of integrated circuits. Such shifts are commensurate in size with the dimensional tolerances required for high-performance integrated circuits, and therefore pose a serious problem in the manufacture of devices requiring submicrometer lines or overlay accuracy of less than 1 μm. The effect has been studied after processing in various conditions; the lateral or in-plane distortions appear to increase as wafers undergo multiple-processing steps.

The continued demand by the integrated circuit industry for larger wafers and higher device densities has placed requirements in the photolithographic manufacturing system for "superflat" substrate glass. Superflat substrates are flatter than current master grades and are primarily used in projection printing. The accurate and consistent measurement of this glass as it relates to both the vendor/purchaser and the test/use situations is examined. This is accomplished through both an overview of the industry's status as well as specific measurement and computer generated data. The study concludes that while the industry objective of much flatter substrate glass is clearly established, the procedures and equipment used to measure this flatness have not, heretofore, been well defined.

Projection printing eliminates the problems introduced by hard-contacting the mask to the wafer while maintaining the high resolution requirements of the integrated circuit manufacturer. Certain advantages of projection printing are obvious: the wafers are not contaminated or damaged by contact with the mask, and the mask will not wear out. Less obvious are the steps required to maximize the advantages of projection printing, thereby reducing costs and increasing yields. This paper considers photoresist choices, suggests some procedures to determine process variables, and discusses mask costs and quality effects.

The automatic production of reticles for semiconductor masks has become a well-established production technique using optical pattern generators. The important characteristics of these generators are defined and a comparison made of the types in general use today. The advantages and disadvantages of the optical and electron beam generators as related to reticle-making are defined and discussed. In a general overview of mask requirements, the characteristics of masks currently in use are related to the trends of the future.

Optomechanical systems and photographic processes impose conditions like mirroring and changes of polarity of images, and limitations due to transfer characteristics of optical systems, quality of photographic materials, etching problems. Conventional optical systems will keep on running for some years and it is worthwhile improving their possibilities by compensating some of their limitations. These corrections can be introduced through a computer program at the level of the control tape of the pattern generator. This paper describes two examples of applications showing how to adapt the data to the process. The first one deals with the optimization of the yield and capacity of the principal mask making machines as a function of the size of circuits, by grouping or on the contrary partitionning the chip on the reticle. The second one describes a method for computation of the polarity, orientation and background of a reticle as a function of the polarity and number of steps of a process.

Present contact masking techniques for integrated circuit manufacturing causes progressive deterioration of the mask through physical damage and particle contamination. The deterioration of the mask during the contact printing operation makes it necessary to generate many mask replications, a process which is both expensive and error prone. The comparatively recent introduction of the off-contact or proximity printing technique offers a solution under certain circumstances to the undesirable aspects of contact printing and the high cost of frequent mask replacement. Off-contact printing, as the name implies, is a procedure requiring the mask and wafer to be separated by an amount just sufficient to avoid physical contact. With the mask and wafer so positioned, the wafer is photo-etched by transilluminating the mask with an optical system having rather special characteristics. In this report, the technical aspects of the off-contact printing operation are explored, particularly as they relate to the optical design of the illumination system. Two candidate illumination systems are contrasted and the results of a trade-off analysis given.

After a survey of the present state of optics in microelectronics future possibilities especially of dioptric lenses are discussed. Those lenses are marked mainly by the numerical aperture providing the resolution, and the diameter of image field. The numerical aper-ture seems to have a usefull limit between 0.4 and 0.5, providing minimum line widths of 0.4 to 0.7 microns. A future limit of field sizes cannot be given today. The possibility to develop new lenses is not only given by the capability of the optical designer, but also by other parameters, such as mechanical size limitations and lens manufactoring problems.

Currently electron beam exposure systems with performance characteristics suitable for the rapid fabrication of high resolution photo master masks are coming onto the market. These machines also have the capability to write patterns directly onto silicon wafers with high accuracy and defect densities better than obtainable by optical techniques to date. Due to the high initial costs of these systems, the costs per mask at low throughput rates (< 1000 master masks per year) are high when compared to the costs of fabricating master masks by conventional optical techniques. However, at higher throughput (> 1000 master masks per year), electron beam generated masks can be fabricated more economically than by conventional techniques. Additional benefits which also have an economic impact can be derived through the use of E-beam lithographic techniques due to the fast turn around of masks, and pattern flexibility on the same substrate. Also, the very low defect densities obtained using E-Beam techniques, substantially affect the economic viability of the very large scale integrated circuits (VLSI).

Measurement of linewidths on silicon and iron-oxide photomasks is hampered by the dark banding which occurs along the edges. It is shown that this banding arises from the combination of low contrast and optical path difference introduced by the silicon or iron-oxide in conjunction with the partial coherence of the illumination. As previously shown in work with chromium-oxide photomasks, when the condenser numerical aperture is sufficiently less than that of the objective, an expression can be derived for the proper transmittance threshold for determining edge location. An expression is also derived for the linewidth error which would result from locating the edge at the center of the dark band. Theoretical and experimental results are compared.

The topic of this presentation is the photolithographic aspect of fabricating CCD arrays. CCD stands for Charge-Coupled Device, which is a relatively new type of semiconductor device. The basic CCD device is an analog shift register. In the same manner a digital shift register is used to delay or buffer a stream of digital data, a CCD can be used to delay an analog signal.

Thin Film Transistor technology permits the generation. af extremely large area active networks, which currently find their principal application in the Large Scale Integration of solid state, flat panel displays. This paper reviews some of the problems encountered in carrying a design for a large area active matrix through the stages of layout, pattern-generation and stencil mask preparation, the latter being used in the final fabrication of the thin film circuits. An alternative approach, used in thl laboratory, which has great versatility, and which has yielded the largest circuits so far generated (36 in), will also be touched upon.

This work shows how the economics of comparing electron beam direct wafer exposure vs. replication is affected by (1) the time required for the electron beam machine to expose a wafer, (2) wafer distortion, and (3) linewidth control. It is shown that time per wafer tw < 500 sec is attractive economically for fabricating chips with an area > 0.3 cm², that machines with such values of tw have been demonstrated, and that such low exposure times are largely determined by the speed of the machine step and repeat operation for beam spot size > 0.5 μm. For very small spot sizes time per wafer is dominated by the resist exposure and there is a need for brighter electron sources or more sensitive high resolution electron resists than are in use today. Evidence is presented showing 1 part in 105 wafer contraction for conventional MOS wafer processing. Such values of wafer distortion will narrow the cost difference between replication and direct writing. Linewidth control with electron beam direct writing was found to be a few tenths of a micron better than available numbers from photolithography, but that this was not a strong differentiating factor economically because the wafer-to-wafer distribution of linewidth values is strongly dependent upon the reproducibility of thin film thickness, and etch endpoint detection.

Optical waveguides having cross-sectional dimensions on the order of a few micrometers on a side and lengths on the order of a few centimeters have been fabricated using photo-lithography and electron beam lithography. These guides can be produced in single crystal materials suitable for optical switches, lasers and detectors. The waveguide devices are often more efficient and potentially less expensive to construct than their bulk counterparts. Moreover, when the demand is sufficiently strong, means may be found to design complex integrated optical circuits consisting of networks of these wave-guide elements connected on a single substrate. The theory and techniques of optical waveguide devices, which have been reviewed elsewhere,^1,2 will be discussed briefly. An especially simple method of producing waveguide switches is by diffusion of metal into LiNb0^3,4 The photomask may be conveniently made using the EBES (electorn beam exposure system) designed for integrated electronic circuits.⁵ The high resolution metal pattern required for preparing the diffused waveguides is made by a positive-resist lift-off method.⁶ Waveguide patterns can also be fabricated using a positive electron-beam resist and a scanning electron microscope.⁷

The progress and prospects for fabrication of integrated circuits with x ray lithography are reviewed. Present technology of masks, alignment, resist, and x-ray sources is examined and found to be adequate to produce IC's with 1-to 5 μm design rules. The exposure time is under 5 minutes and the result is reproducible resolution and high yield due to the off-contact exposure process. Fabrication of IC's with sub-micron dimensions is more difficult because precision alignment to 0.1 μm is not yet available and there is not a ready source of intense collimated x rays. High-resolution lithography can be accomplished using special mask-wafer contact methods with some sacrifice in yield. Synchrotron radiation shows the greatest promise for volume production of IC's with submicron patterns, but the application must be proved to justify investment in the costly facilities.