As the difficulties of producing more complicated integrated circuits increased, the pressures on process designers caused manufacturers of supporting technologies to develop the equipment necessary to generate reproducible product at reasonable yields. A group which proved highly successful in developing higher-performance equipment was in the optical measurement area. In fact, the generation of new measurement systems which were now highly reliable, repeatable, and traceable to NBS gave the device manufacturers a set of "measurement standards" which were rapidly accepted because of proven results. With the results being so positive, impetus was given toward the development of the optically-controlled systems needed to ensure highly-repeatable exposure cycles. A highly-significant advancement in process control, Constant Intensity, was developed. This system uses optical sensors to track the degradation of the source (in respect to specific photosensitive materials) and compensate for the wear by controlling power to the source, main-taining the desired intensity from exposure to exposure. The Constant Intensity capability took what was a variable (intensity and time), replacing them with fixed intensity and time.
Single parameter irradiance measurements commonly used for the determination of photo-resist exposure times are incapable of giving unique values of either the broad band irradiance from a mercury-arc lamp, or the photoresist exposure times. Resolution of these measurement incapabilities through a series of irradiance measurements of narrow, approximately monochromatic radiation bands is discussed. A basis for establishing photoresist sensitivity indices that can be used to calculate. unambiguous exposure times by applying Van Kreveld's additivity law to these narrow band irradiance measurements is suggested. The thickness variation of photoresist films is included in the method of calculating exposure time. Conditions that may simplify the calculations of exposure times or reduce the number of irradiance measurements when a mercury lamp is used as the exposure source are discussed. Suggestions are made for the use of these indices as photoresist specifications, and for the real-time automatic monitoring and control of exposure times in wafer fabrication where optical density variation in the photomasks and thickness variations in the photoresist films are automatically incorporated into the control by a microprocessor.
In practical microcircuitry a two-dimensional source array is necessary to cater for various types and directional orientation of optical features at the image plane. Configurations successfully applied are the annulus and a distribution of sources at the points of a hexagon and an octagon. In each case a single concentrated mercury arc lamp is used as prime source and effectively multiplied or extended by optical means.
The geometries of semiconductor photomasks often incorporate complex regions in the definition of diffusion windows. This paper presents a technique for the computer generation of the master artwork for such photomask patterns. Given the perimeter definition of a convex mask region, a general technique for filling in the region is developed. This technique is then extended to include any convex/concave shaped region. An algorithm that permits optimal photoplotter aperture selection is then introduced. Combining this algorithm with this general technique provides an effective method of automatic mask generation.
A 10x reticle must meet quality criteria completely because one defect, if its size exceeds the resolution of the photorepeater lens, may degrade the performance of all the related IC chips, or even cause their complete failure. The mask surfaces, therefore, must be cleaned thoroughly. The usual method, air-cleaning with a hand-held pressure nozzle, does not remove all the small particles. Liquid cleaning is more effective, but is expensive and usually involves toxic materials. The results of dry cleaning have been improved substantially by a system in which suction is applied instead of blowing. The basic system consists of a hollow metal tube connected to a vacuum pump. A long slit in the tube is brought close to the contaminated surface, which is moved past it on a linear stage. Some particles are not relloved by the air flow alone; mechanical action applied by a slowly vibrating brush dramatically improves the particle removal rate. The cleaning system effectively removes particles of glass, plastic, dandruff, and unknown materials. A few remarks on adhesion forces are given in an appendix.
The copy camera resolves 320 lines/mm on high-resolution film with less than 0.01-mm distortion throughout a 30-mm-diameter image made at 20:1 reduction in white or green light. This performance is the result of five special features: (1) an f/4 precision copy lens, which makes diffraction-limited images throughout the image field; (2) a precision focusing mount; (3) a pressure plate that maintains sharp focus for step-and-repeat exposures; (4) a simple indexing jig that locates the images precisely on 101.6- by 127.0-mm film; and (5) a film holder with registration pins to position the prepunched film precisely in relation to the image positions. The camera is used routinely to make single- and stepped-image photofabrication masks, high-contrast resolution test charts, and special reticles.
Experimental work on a new photographic material and associated process technology. are described with reference to the requirements of the manufacturers of advanced micro electronic devices with high, packing density. The microphotographic process is demonstrated practically for contact printing in a negative situation as well as in a positive/positive situation. The capacity of the new microphotographic process to control accurately the initial dimensions on the micro-image is of great significance in order to define the geolletry of the device to be designed.
Several viable photomask materials and processing systems are evaluated with regard to their defect contribution to the overall photolithographic yield. Practical results are compared with theoretical predictions.
The problem of measuring critical dimensions on photolithographic masks is becoming more acute as production linewidths dip below 2 μm. Visual techniques based on filar and image shearing eyepieces are limited by instrument resolution and operator acuity and consistency. An automatic system, based on diffraction pattern analysis, has been developed. Subjective operator judgment is not required to complete a measurement. Repeatability of 1% has been demonstrated for lines and gaps as small as 1.5 μm. The technique and device will be described and detailed experimental results will be presented. An analysis of the technique limitations shows that the current technique can easily be extended to measure the 0.5 to 1 μm widths. The ultimate potential may well be 1 microinch resolution at 1 microinch widths. Thus, diffraction pattern analysis may be the only viable alternative to the scanning electron microscope for submicrometer production line testing.
A current micrometrology program at the National Bureau of Standards is concerned with the development of calibrated artifacts and accurate methods for line-width measurements in the 1 to 10 μm region with application to measurements of critical dimensions on IC photomasks. A major program objective is to develop improved theory and experimental verification for line-width measurements made with commonly used optical-microscope systems. Most line-width measurement techniques utilizing an optical microscope apply some edge detection criterion to the image profile. The measurement is therefore affected by system parameters which affect the image profile, such as defocus, spherical aberration and coherence of the illumination. The effects of some of these parameters have been treated previously in the literature. As examples, image profiles for incoherent, coherent, and partially coherent illumination are presented. Image profiles are also given for oaring amounts of defocus with and without spherical aberration. However, these curves only indicate trends and do not represent image profiles that would result with realistic values of the system parameters.
As part of the effort conducted at NBS to solve some of the fundamental problems associated with width measurement of very small (l-5-µm) lines and spaces, the performance of an optical microscope with coherent illumination is investigated. From these studies, the theoretical basis for a new method of accurate width measurements is developed and explored. The new method, in effect, produces an optical transformation in which the image no longer resembles the original line but in which the location of the line-edges is marked by two narrow, dark lines within a bright surround. The correct line-width is then given by the distance between these two lines, a measurement that eliminates the orientation problems normally associated with filar eyepieces and sidesteps the coherence problem that affects shearing eyepieces. Suggestions are made about implementing the technique. Available microscope objectives are not suitable for such a system, and a redesign is recommended.
The emphasis today in wafer fabrication is to continually improve device yields. This is usually accomplished at the expense of the mask maker, who is consistently being requested to provide masks of higher and higher quality. Higher mask quality is also desired due to the long-life masks, especially when used with a proximity or projection printing system. A mask inspection and usage scheme has been developed, based on an age-old inspection technique, whereby higher device yields can be realized without paying the price of increasing mask quality.
Attention is directed to the need for pattern-to-pattern overlay precision commensurate with micron lithography for integrated circuit fabrication and an error target is suggested. Several sources of overlay error are identified, including: Position of pattern edges in processed layers Mask/wafer bending distortions Alignment mark detection Pattern generation, and Thermal expansion of masks and wafers Pattern replication The magnitude of parameter control needed to maintain the desired total precision is discussed for several cases. Individual errors need to be less than a few tenths of a pm for the total error objective to be met. Although this magnitude of error approaches the limits of optical measurement precision, the resulting circuit pattern can be adequately measured with vernier patterns. Reported achievements in E-Beam lithography show successful error containment for small fields. High precision overlay for full wafer exposure using masks will be significantly more difficult to achieve (e.g., in optical scanning cameras, photo-emissive E-Beam replicators, and X-ray proximity printers). However, there is justification for the effort in the increased wafers/unit time (thruput) that can be realized for lithographic tools using full wafer exposure.
The edge profiles of semiconductor devices and developed resist images are usually determined with a scanning electron microscope. This paper presents a fast, nondestructive optical method by which edge profiles can be taken at any point on a wafer of any size. The edge profile is deduced from the angular distribution of the light scattered by the edge. The angular distribution is a signature of the edge profile, and contains information about such characteristics as double slopes, hooks or lips, rounding of corners, very low-angle slopes, and even step heights of the order of 500 Å. Such complex edge structure can be evaluated by simply calibrating the angular distribution with SEM observations. By applying a simple model of light diffraction from an edge to simple slopes, we are able to deduce edge slope, slope length, and step heights. This analysis can be applied to slope angles from 5° to 80° and step heights from 0.3 to 2.5 μm. The values thus deduced agree well with those obtained with the SEM. The method can be applied routinely at critical steps in production to monitor process variations across a wafer or from wafer to wafer.
We present results of our evaluation of the X-ray lithographic technique for replication of submicron patterns generated by electron beam. Mylar membranes have largely been employed as substrates for X-ray masks, although silicon substrates have also been used. Absorber patterns with high aspect ratio have been obtained by electroplating and by ion milling. Circuit patterns have been replicated with 0.25μm features in both positive and negative X-ray resists.
The Viking Project required a space-hardened, sterilizeable camera which would survive external temperature, 200 MPH sand storms and a landing on Mars. The approach used to satisfy these constraints will be discussed. The resulting camera was a facsimile design with mechanical scanning and an array of photosensors which were electrically selectable to provide focussing, spectral range selection and resolution selection. The camera provided PCM digital output of the video data and interfaced with an on-board computer to select azimuth and elevation pointing angles data rates, selection of black and white or color imaging, etc. The camera has resolution equivalent to a human observer, positional accuracy equivalent to a metric camera and radiometric accuracy sufficient to relate intensity changes to material and topographical characteristics of the planet.
A facsimile camera was selected over several other types of imaging systems that had been proposed for the Viking lander mission to Mars primarily because it could best meet stringent power and weight constraints. The reason is that the optical-mechanical scanning mechanism of this device provides both the image raster and the field of view. In addition, the Viking lander cameras feature electronic focus selection for high-resolution (0.04 instantaneous field of view) monospectral imaging, six survey channels between 0.4 to 1.0μm for lower resolution (0.12°) multispectral imaging, and a (0.12°) survey imaging mode for rapidly viewing the scene (for example, 60° x 180° in 5.5 minutes). Absolute radiometric accuracies of about 10 percent can be attained with the aid of an internal light source and external reference test charts. The cameras are the most versatile imaging systems ever developed for a planetary mission. However, their performance is not without peculiarities that require meticulous care in reducing spatial and spectral data.
A special presentation will be made to describe the Viking Ground Reconstruction Equipment (GRE) which was designed to compliment the Viking facsimile camera. Together the camera and the GRE make up the Viking Imaging System. The GRE is a computer controlled laser beam recorder that uses a spinning scanning mechanism to write picture information onto film. The geometric accuracy of the GRE is ten (10) times better than the camera. The equipment can produce high resolution black and white negatives, composite color negatives, color separation negatives, and quick look Poloroid pictures, both in color and black and white. For high resolution black and white photographs the GRE is capable of generating sixty-four (64) density levels on film in a linear or logarithmic fashion. The presentation will describe the design requirements of the GRE, the technical implementation, and show some photographs taken by a Viking camera and reproduced on the GRE.
The laser interferometer is attractive as a position-sensing device for microlithography because of its high resolution, high accuracy, ease of installation, and non-contacting measurement capability. The usual confiqu ration for x-y coordinate sensing is the plane mirror interferometer which minimizes Abbe offset errors. The laser's disadvantages are its relatively high cost, susceptibility to environmental disturbances, non-standard output units, and incremental rather than absolute measurement. However, all of these disadvantages may be reduced or avoided by proper design and installation techniques. There are several approaches to electronically interfacing the laser interferometer to a machine controller, and each approach has its own tradeoffs among cost, speed, and hardware and software complexity.
The impact of mask quality on wafer yield can hardly be debated. Photomask quality can seriously limit wafer yield, but that limitation can be minimized by effective mask inspection and process control. Techniques for predicting mask limited yield will be presented in conjunction with a discussion of methods for maximizing yields through appropriate process controls.
This paper discusses the results of an experiment which quantifies the impact of photo-mask materials - both glass substrate and thin film - on mask life and wafer yield in a contact printing process. Chromium, DC triode sputtered iron oxide and RF sputtered iron oxide were deposited on soda lime, alumina soda lime and borosilicate glass substrates making up a matrix of seven mask types. Mask degradation was caused by bringing test wafers into contact with each mask and then exposing them. Masks were inspected at specific intervals as printing progressed; defects were classified and tabulated at each inspection. The results are presented graphically for comparison of each photomask system. The experiment was statistically designed to provide a 95% confidence level in the overall results.
A step repeater and a mask measuring machine have been designed and tested. One of the important goals of the development for both machines was realising small repeatibility errors. For this purpose special principles in measuring techniques,in positioning systems and in design principles have been adheared which are being described, together with the concept of computer controll. For the measuring machine a newly developed two-coordinate photoelectric-microscope is being used as pick up. Repeatibility errors are for the step repeater 0,1 μm measured on different mask plates, and for the mask measuring machine o,o8 μm, 1 sigma value.
How many times have you discussed the need for increased I.C. complexity, C.D.'s with tighter tolerances, submicron C.D.'s, zero defects, larger dies . . . ? Since each one of these needs will require additional mask inspection effort, I suggest that you should ask, "What is it worth to you and when does the cost of mask inspection approach the critical point of a depreciating return on investment (i.e. tooling cost versus die yield)?". It has been my experience that approximately one half of the labor used to fabricate masks is devoted to mask inspection. A survey of mask fabricators and mask users indicated the following ranking, in order of importance, of four mask quality parameters: (1) pattern registration (2) critical dimensions (measured with precision) (3) fatal defects (4) critical dimensions (measured with accuracy)
Photomasks used to produce high frequency microwave transistors have unique and different requirements than do integrated circuit photomasks. Critical dimensions have a tolerance of ± 0.2 micrometers. In most cases the critical geometry is 1.0 micrometer or less. Defect densities are very lenient as compared to integrated circuits. We inspect to a 4.0 A.Q.L. In a two-inch square array we have on the order of 8,000 to 15,000 discrete device patterns. Inspection is made on a random sample as shown on the following table.
Variables are listed that must be considered and included in mask inspection specification when a mask set is obtained from a mask maker. There is nothing more discouraging and abstract than obtaining a mask set from a mask maker to a certain criteria. Put yourself in the mask users shoes. You thought you had written a thorough and concise mask specification. Careful consideration was given to the requirements that would insure good die yield. Attention was paid to the masking processes in use to fabricate your device. Considerations were made for type of resist, exposure equipment, develop and etch techniques and other process variables which will affect the geometries which are transferred from mask to wafer. Then the mask maker informs you of what is actually achievable on fabricated plates and the compromises begin.