The history of the mask industry has three main periods, which I call the Classical Period, the Dark Ages, and the Renaissance, by analogy with those periods in the history of Western Europe. During the Classical Period, people developed 1X masks and the technology to make them. In the Dark Ages, people exploited the equipment developed during the Classical Period to make 5X reduction reticle, ending the nobility of mask making. In today's Renaissance of mask making, a proliferation of mask types is requiring a rebirth of innovation and creativity. The Renaissance resembles the Classical Period: masks are once again strategic, and technological capability is once again the driver. Meanwhile, the mask industry is carrying forward the productivity and efficiency gains it achieved during the Dark Ages. We must create a new business and economic model to support these changes in the characteristics of the marketplace.

Multiphase printing (MPP^TM) has been shown to improve registration and CD performance by reducing butting errors, averaging scan linearity errors and improving stripe boundary CD performance. In the MEBES 4500, improvements in the data path allow MPP to be used with a tolerable increase in write time making it an attractive option. A major concern is determining if MPP printing requires modification of subsequent mask processing steps. The use of larger spot sizes and multiple voting in MPP printing may alter CD performance during mask processing. Additionally, the multipass strategy inherent to MPP could result in a different CD signature due to time based effects. A further concern is to determine if MPP CD's impact pattern fidelity since this will affect inspectibility and wafer performance. Several experiments were conducted to determine how well MPP integrates with processing, how it is impacted by temporal averaging of errors and if it significantly alters pattern fidelity. A range of input addresses and different resist systems were investigated.

SEMATECH has conducted a mask writer benchmark study in order to provide a description of the lithographic capability of the mask writer industry. Systems claimed by their manufacturers to have the ability to support 0.35 um or better lithography groundrules in a manufacturing environment were the focus of this study. The study describes the average performance and identifies any significant differences in capability among the various mask writer tools. The benchmarking was conducted as a blind study. Participant identity was protected by using the services of Arthur Anderson & Co. The parameters described in this paper include critical dimension (CD) control, registration, overlay (two plates), and throughput. Results of the study show that no participant consistently demonstrated the 50 nm CD uniformity performance required by the 1995 SIA roadmap for 0.35 um lithography. However, half the participants were able to demonstrate registration performance consistent with the Roadmap requirement of 70 nm 3 sigma. While throughput results varied greatly, one participant demonstrated superior capability.

Mask makers generally specify Critical Dimension (CD) Uniformity for customer product plates as a composite value. This must include the contributions of machine systematics such as deflection field distortions and stripe butting and external error sources such as resist processing and measurement. This broad definition requires that a line at any location within the quality area must be within spec. Manufacturers of electron beam lithography equipment have traditionally taken a more component-based approach: separating CD Uniformity from other lithography error sources by using a test pattern in which features are written at the center of the writing stripe. This approach removes the effect of field distortion and stripe butting error. These intrastripe and interstripe errors are evaluated in terms of their impact on placement accuracy but not CD control. Another issue that clouds the evaluation of CD Uniformity is the definition of the specification itself. Some in the industry represented CD Uniformity in terms of range while others use a 3(sigma) value. Rarely does a specification include the data sample size. Lepton has adopted a new approach to the evaluation of CD Uniformity for EBES4: CD Uniformity is evaluated in terms of both global and local performance. Global CD Uniformity incorporates effects of beam stability, depth of focus and materials effects while Local CD Uniformity addresses the impacts of deflection distortion and stripe butting. This paper focuses on Local CD Uniformity and the contribution of Stripe Butting to that specification. A methodology will be presented which makes use of a test cell containing features not fractured by stripe or cell boundaries (off-boundary) as well as features bisected by these boundaries (on-boundary). The data for a number of plates are analyzed in terms of range and 3(sigma) as well as being evaluated as a function of sample size. In addition, traditional butting evaluation techniques such as the placement based, `railroad track' method and the `point by point' CD method are used for purposes of comparison.

At Lucent Technologies, lithography tool performance metric measurements are required on every plate. The QA cell patterning is designed to monitor the lithography tool, and the data is captured within the Mask Information Management System. It allows the manufacturer to use actual production data to determine machine performance trends, and it yields an extremely large sample of plates for SPC purposes. Closure, X vs. Y uniformity, butting, and registration data is reported in this way. The usage of on-product data for machine metrology is demonstrated, and actual production performance of the MEBES 4 TFE is shown.

An improved PBS process has been developed and tested for the photomask production. Significant improvements in CD uniformity (3(sigma) < 40 nm over 6' plates) and CD meant- to-target control (< 30 nm) have been achieved through a specially treated puddle development. A low-temperature bake (95 degree(s)C, 30 minutes) was utilized for better linearity while the resolution was 0.5 micrometers for the regular process and as low as 0.35 micrometers for the ghosting process. Defect densities on production plates have been greatly improved to a level lower than our normal production process. The new process is not limited by any specific type of processor, therefore it can be implemented into current production lines easily. It is expected that this new process will meet advanced specs for 256 M DRAM (with 0.25 micrometers design rule) and extend the life of the PBS production line for another 3 - 4 years.

There seem to be much activities, in the mask-making industry at present, in re-establishing a new process technique and condition for a conventional high-molecular polymer resist such as PBS rather than in seeking an alternative for advanced e-beam reticle fabrication. We had investigated, and reported at PMJ'96, how to utilize e-beam `thin resist' in order to obtain an improvement in CD linearity as well as an expansion in total process latitude. We found in particular that there seemed to be a threshold in exposure dose required for each individual resist to improve or even to maintain quality pattern edge and CD linearity, which looked to be independent of coating thickness. Accordingly we tried to determine an exact threshold for PBS and ZEP-7000 this time, and re-examined impacts of baking condition and coating thickness in details under a right exposure dose determined beforehand. This paper describes our findings and approaches to improve pattern edge quality, pattern linearity, hole pattern fidelity and so on by optimizing pre-baking condition, by reducing coating thickness, under a desired exposure dose which was determined by our detailed investigation, in regard to high-molecular polymer resist, for advanced e-beam reticle fabrication.

A chemically amplified negative electron beam resist process is developed for fabricating 0.25 micrometers devices 4X magnification reticles. In order to improve a critical dimension (CD) uniformity on reticles, a precise temperature controllability in post exposure baking is realized. Dry etching for Cr is also adopted for getting an appropriate pattern fidelity down to submicron on reticles to meet requirements for such an optical proximity effect correction. Under the optimized process conditions, a CD uniformity of 16 nm in 3 (sigma) is achieved on reticles. A CD linearity of down to 0.8 micrometers on the reticles is ensured as well. Actual 0.25 micrometers device reticle production results show that the CD mean variation is controlled within +/- 43 nm. It is confirmed that an advanced 4X reticle fabrication process for 0.25 micrometers device generation is realized.

The advanced requirements for resolution, critical dimension control, and linearity on photomasks require chromium etch processes, which are highly anisotropic and extremely uniform. They must also allow lithographers to expose the resist at the iso-focal point of the exposure system. A reactive ion etching (RIE) process of chromium films, previously reported, met some of these demands. However, it was reported that system loading effects significantly reduced the etch rate, etch rate uniformity, and CD uniformity. An alternative plasma etch system technology, inductively coupled plasma (ICP), has been investigated, which makes it possible to achieve much higher plasma densities, operate at lower pressures, and control ion energies independently. The combination of these features minimizes loading effects, makes it easier to control selectivity, accelerates etch rates, decreases undercutting,a nd improves etch uniformity. This allows the maskmaker to transfer the resist image into the chrome with minimal nominal CD shift and degradation of CD uniformity. Furthermore, with little or no undercutting, resolution and CD linearity are greatly improved. Results using the ICP configuration will be presented and compared with the RIE process. Preliminary studies of defects will be presented. Defect densities will be quantified using a plasma etch process and compared to a wet etch process. Sources of defects and possible solutions to defect reduction will be discussed.

Photomask defect criteria normally specify pinholes by size, regardless of other factors that may affect printability (e.g., transmission, reflectivity, phase, or proximity to features). This paper is an initial study of the effect of pinhole defect transmission, size, and location on inspection tool sensitivity and defect printability. Two masks with varying defect properties were built, inspected on a mask inspection tool, and exposed on a DUV wafer stepper and an aerial image simulation microscope. On the wafer, defect printability was measured as either the presence of an isolated defect or the variation in line size for a line with a defect in close proximity. A prototype inspection-based tool was used to estimate defect printability, and the results were correlated to on-wafer defect printability. The goal of this work is to find a quantifiable measurable that may be used at inspection review to reliably predict printability.

Photomask manufacturing for sub-0.5 micron ICs faces two principal challenges: efficient cleaning within cycle time requirements and minimization of water and chemical usage. A new technology using only deep ultraviolet photons and a flowing inert gas cleans state-of-the-art optical and x-ray photomasks. The technology uses no water or chemicals. Tools can handle single or multiple substrate sizes. There is no optical reflectivity change in the chrome. No damage is observed on chrome patterns from either backside or front side exposure whether the features are positioned horizontally or vertically. Our work to date indicates that this technology meets the cycle time requirements in the photomask manufacturing process.

Until recently, opaque defects on photomask have been removed by laser evaporation. However there are several disadvantages in this repair technique. First an operator must visually align the laser irradiation spot and pattern edge, so high repair accuracy can not be obtained. Also when the area to be repaired is large the laser sputters the surrounding surfaces and the evaporation tends to redeposit on pattern edges causing them to swell thus increasing the probability of detection by inspection equipment. Additionally the laser repair technique has proven to be very difficult if the pattern defect is any complex geometry. The repair using Focused Ion Beam (FIB) has been developed to solve some of these issues, however this method has not yet been applied to the production line because of degradation of transmission rate and the damage to the glass substrate. This paper reports the evaluation on the performance of the FIB repair using a newly developed gas assisted etching (GAE) technique and the status of using FIB with GAE in the production line. By using GAE, the degree of glass damage has been reduced by a factor of ten as compared with the FIB repairs without GAE. A transmission rate of 94% (i-line) could be obtained on a conventional mask and 92% (i-line) on the Halftone Phase Shift mask (HT-PSM). Furthermore, by applying the post-treatment after the repair, the transmission rate of the repaired area could be recovered to the same level as the normal glass area. The printing characteristics for i-line of the GAE repaired in both conventional mask and HT-PSM has been also good, showing that the GAE FIB repair is applicable in the normal photolithographic process window and that this method can achieve the similar printing result in comparison with non- repaired area.

Controlling line width error is one of the most challenging tasks in mask production. Current methods for monitoring line width are based on either CD measurements or edge defect detection. Neither method guarantees the detection of line width errors with good sensitivity over the whole mask area. For example, a combination of small edge defect together with CD non-uniformity within the allowed tolerance, can yield a significant line width error, which will not be detected using conventional methods. In this paper we are presenting a new approach for line width error detection. The method, a new feature of Orbot's RT-8000 die-to-database Reticle Inspection System, is a sensitive detector that operates during the inspection of all the reticle's area. The detection is based on a comparison of measured line widths on both the database and the scanned image ofthe reticle. In section 2, the motivation for this new detector is driven. An analysis of various defect types, which are difficult to detected using edge detection approach or CD measurements, is presented. In section 3, the basic concept of the new approach is introduced together with a description of the new detector and it's characteristics. In section 4, the calibration process that took place in order to evaluate the sensitivity of the detector is described. The experimental results of this evaluation are reported in section 5.

There have been several studies on the printability of subhalf-micron defects using reduction steppers. These studies typically involved 1X reticles with defect sizes greater than 0.3 micrometers . Because submicron 1X projection systems are being incorporated into numerous fabrication lines, there is a clear need to determine the impact of subhalf-micron defects using these systems. This paper examines defect detection and measurement capability on 1X reticles and the printability of those defects on production submicron 1X steppers. This analysis will enhance the understanding of the relationship between defect size and 1X projection optics and allows for determination of optimal defect specifications. A test reticle representative of a 64 Mb DRAM metal layer was manufactured with a programmed series of attached and isolated defects ranging from 0.15 to 0.5 micrometers . Both clear and opaque polarity defects were designed. The defects were identified and measured on two different reticle autoinspection systems. The performance of the two systems was compared to the reticle database to evaluate capture rates and efficiency. Actual reticle defect sizes were measured using low voltage SEM metrology. Defect printability was determined using a 1X i-line projection stepper with focus and exposure optimized for nominal critical dimensions (CD). The defects that printed on the wafer were measured and compared to the defects measured on the reticle. The effects of varying wafer exposure dose and focus within a 10 percent CD process window on defect printability were also evaluated. The results of the mask inspection comparison and the reticle versus wafer defect maps are compared.

The fourth annual mask industry assessment will again survey various industry companies for key performance measurements in the areas of safety, quality, delivery, and throughput time. The data complied includes safety incident rate, mask shipment volume, customer return rate, customer return reason, mask survival yield, delivery performance to schedule, and throughput time from 1988 through Q2, 1996. To enhance the 1996 assessment, the performance data will be collected by categories of defect size specifications. Participant identity remains protected by utilizing Arthur Anderson & Company to ensure confidentiality. The 1995 assessment contained data from all major U.S. merchant and captive suppliers along with two international shops. In 1996, all mask makers world wide will again be invited to participate to further enhance the value of this benchmarking exercise.

Attaching a pellicle and frame to a photomask induces stresses which change the position of patterns on the mask by up to 100 nm depending upon factors including mounting procedure, mask thickness, and overall mask size. The Nikon XY-5i is used to measure the pattern placement before and after pelliclization to establish changes in the array due to the pelliclization process. The importance of this type of analysis as it relates to the overall mask error budget will be explored. The XY-5i system capability and technology to repeatably measure the pattern placement through the pellicle will be explained. Data gathered from several photomask sizes and thickness will be presented. In addition, automatic and manual pelliclization processes will be compared.

In the integrated circuit (IC) photomask industry today, dense IC patterns, sub-micron critical dimensions (CD), and narrow tolerances for 64 M technologies and beyond are driving increased demands to minimize and characterize all components of photomask CD variation. This places strict requirements on photomask CD metrology in order to accurately characterize the mask CD error distribution. According to the gauge-maker's rule, measurement error must not exceed 30% of the tolerance on the product dimension measured or the gauge is not considered capable. The traditional single point repeatability tests are a poor measure of overall measurement system error in a dynamic, leading-edge technology environment. In such an environment, measurements may be taken at different points in the field- of-view due to stage in-accuracy, pattern recognition requirements, and throughput considerations. With this in mind, a set of experiments were designed to characterize thoroughly the metrology tool's repeatability and systematic error. Original experiments provided inconclusive results and had to be extended to obtain a full characterization of the system. Tests demonstrated a performance of better than 15 nm total CD error. Using this test as a tool for further development, the authors were able to determine the effects of various system components and measure the improvement with changes in optics, electronics, and software. Optimization of the optical path, electronics, and system software has yielded a new instrument with a total system error of better than 8 nm. Good collaboration between the photomask manufacturer and the equipment supplier has led to a realistic test of system performance and an improved CD measurement instrument.

Resolution enhancement techniques, such as optical proximity correction and phase shifting mask, have been extensively examined to extend the optical lithography capability to meet advanced lithography requirements. It also challenges the mask technology. As the minimum feature size decreases, the effect of proximity effect, resist slope, chrome slope, chrome edge roughness, and pattern corner rounding on defining the true critical dimension is significant. Increasing the sample size of mask CD measurement is inevitable to support the wafer fabs to characterize the wafer lithography process and product performance. The need to define the pattern fidelity, such as corner rounding of the optical proximity correction patterns, is emerging. Low voltage CD SEM, confocal microscope, and Atomic Force Microscope have been developed to improve the precision of sub-micron mask CD measurement. The AIMS system, applying microlithography simulation techniques, provides a promising method to predict the linewidth versus stepper dose, linewidth versus depth of focus, and correction between mask and wafer geometries. At Intel, we have developed a technique to magnify the image on the microscope to improve the precision and linearity of the CD metrology system. This newly developed technique which digitizes the measurement image and uses a gray level selective display improves the signal to noise ratio and enhances the image contrast and pattern edge detectability. This enables us to measure high resolution patterns, contacts, and curves of rounded features with high precision. The automatic measurement file creation software developed by Intel through downloading coordinates, automating measurements, and uploading data improves the measurement throughput in the production environment. In this paper, the current issues concerning CD measurements will be discussed. These issues include the effects of proximity effect, resist slope, chrome slope, chrome edge roughness, and pattern corner rounding on the true critical dimension. This paper will also present the user with a discussion on the current CD metrology tool capability. Most importantly, this paper will outline the requirements of future CD metrology tools, which will be vital to the accurate and descriptive measurement of photomask critical dimensions.

We have developed a new system to measure the phase shift caused by the phase shift mask (PSM) for a deep-UV stepper. The system uses two types of 248 nm wavelength deep-UV light sources. One is an Hg-Xe lamp, and the other is a KrF excimer laser. These are interchangeable to match the illumination light source of the stepper. The phase is measured with a polarizing shearing interferometer. It has two wollaston prisms, a condenser lens, and an objective. The objective has a long working distance of more than 7 mm. This is necessary in order to measure the PSM with a pellicle. A light beam is separated into two orthogonal polarized light beams and sheared on the PSM, and then recombined on a sensor. The transmission beams through a shifter and a substrate interfere each other. Then the phase difference from the shifter is obtained from the difference between the two beams. We evaluated phase measurement repeatability and the measurable minimum pattern width. In the measurement of a halftone PSM, the repeatability was +/- 1 degree over several months, and the minimum pattern width was 2 micrometers or less.

The performances of alternating PSM and its modified type (Hybrid PSM: mixed form of alternating and outrigger PSMs) applicable to 200 nm contact holes were studied via both simulations and experiments with DUV lithography. We focused on the 1.5 dimensionally arrayed contact holes (X-pitch: Y- pitch equals 2:1). The pitches of Y direction were 320 nm, 360 nm, 400 nm and the pitches of X direction were twice those of Y direction, respectively. To find out optimum NA and (sigma) , simulations were performed with various conditions (NA equals 0.4 approximately 0.8, (sigma) equals 0.3 approximately 0.8). Comparison of aerial images showed optimum conditions depends on the pitch and the type of PSM. Optimum NA ranged 0.5 approximately 0.6 with different pitches, and (sigma) was kept at 0.3. Experiments were performed with the condition of 0.5 NA and 0.3 (sigma) . We found that the alternating PSM provided benefits in printing highly packed contact holes, but that the pattern fidelities need to be improved for the 1.5 dimensionally arrayed patterns. The hybrid PSM showed its effectiveness for those patterns. In other words, the hybrid PSM has improved both the fidelity and the depth-of- focus. It enabled 180 nm hole patterns printing, of 320 nm pitch. Therefore, either the alternating or the hybrid PSM may extend the life time of the optical lithography in printing of contact holes.

We surveyed more than 150 different materials as candidates for optically tunable (at 248 nm and 193 nm), attenuating embedded phase-shift masks. Multicomponent materials with four distinct microstructures: (1) composites, (2) cermets, (3) multilayers, and (4) copolymers, where one component was optically clear at the application wavelength and the other component more optically absorbing, provided a systematic approach for adjusting the needed optical properties: specifically, optical transmission and (pi) -phase-shift. From evaluation of optical properties and other mask manufacturability issues, including chemical and radiation durability, etch selectivity, alignment and inspection properties, film stress and adhesion, we identified promising nitride and oxide materials based on MN_x-AIN (M equals Cr, Mo, W,...) and M'O_y-RuO₂ (M' equals Al, HF, Zr...) as well as promising polymers.

A process for high-resolution patterning of the membrane- type masks used in the SCALPEL (SCattering with Angular Limitation in Projection Electron-beam Lithography) lithography system is described. SCALPEL is a 4X projection electron beam lithography tool with the potential to extend commercial lithographic capability well into the deep sub-micron range: the recently-completed SCALPEL proof- of-concept (SPOC) system has printed 0.08 micrometers lines in thick resist on Si. The details of the patterning process we currently employ and metrology results from the first series of masks are presented here. The SPOC mask blank consists of a segmented W-coated SiN (Si-rich) membrane, fabricated on a 4' Si wafer. The blank is patterned with 45 different test chips using a vector-scanned e-beam lithography tool. Metrology is performed on completed masks, and results from measurements of line-edge roughness, CD linearity, and pattern uniformity are presented. We examine the need for proximity effect correction of the pattern data, and compare the effect of correction on pattern data file size for a variety of mask technologies.

The present paper reports on the achievement of a 3% linewidth control on experimental 4X reticles for 0.18 micrometers (0.72 micrometers on mask), for all linewidths from 0.15 micrometers (0.6 micrometers on mask) up to several microns at the same time, regardless of the local pattern density (e-beam proximity effect). Keeping in mind that a reduction of the projection demagnification factor of optical wafer steppers from 5X to 4X represents an additional generation of reticles (with 1 micrometers on mask level), it is clear that, compared to commercial mask making, this result consists of successful patterning of experimental reticles of 2 mask generations ahead, with tight linewidth control. A similar accuracy has become required for the fabrication of masks with optical proximity corrected (OPC) data, because this technique needs medium--to far submicron resolution on mask. Results of such experimental OPC masks are presented. The present work focuses on pattern-generation related topics. Inspection and repair are not covered, although even more aggressive improvements may be required there.

Experiments were initiated that examined parameters for optimum proximity effect correction using GHOST with 400 nm of PBS and the standard PBS process. Parameters examined included the size of the correction spot and the ratio of the correction dose to the primary dose. Critical dimension CD deviations from the mean in proximity tower patterns were used to determine optimum parameters. Results show that without data bias, the minimum feature size that can be achieved with GHOST within +/- 5% of the mean CD at 10 keV and 400 nm of resist is 300 nm. The correction scheme begins to degrade rapidly with smaller features. CD control at optimum GHOST parameters can be achieved by introducing data bias of 100 nm per side of a line to account for the wet etch undercut in the PBS process. This effectively increases the minimum feature size achievable from 300 nm to 500 nm. A resist that can be dry etched, thereby eliminating the wet etch undercut, will alleviate this problem.

When considering optical lithography, there is no true substitute for the resolution enhancements afforded by a reduction in actinic radiation. However, as we move below 365 nm i-line systems, the optics and attendant materials considerations become acute. Additionally, there is an obvious economic impetus to breath new life into existing exposure systems. Various optical enhancement schemes have been developed over the past few years for these and other reasons. While many of these are well suited for the patterns and economics of memories, there are a few that enable sub 0.5 k_l ASIC imaging. These include: large NA, optical proximity correction, and the deployment of sub- resolution assist features. We will demonstrate a blend of these that will support 280 nm ASIC pattern delineation with i-line (365 nm) systems and binary (non-phase-shift) masks.

Optical proximity correction (OPC) adds complex convolutions to mask pattern shapes which can stress mask-making capabilities. Because the `decorations'--jogs, serifs, assist features--created by proximity correction are approximations to `ideal' corrected shapes there are many variations of corrected mask shapes that provide the same result on the wafer. With an understanding of specific mask- making limitations it is possible to leverage this leeway to create more `mask friendly' patterns with OPC.

Lithography is one of the most important techniques in the IC fabrication and has been extensively used in processing. The high resolution and accuracy of electron beam lithography is most appropriate for making mask of optical and X-ray lithography as well as direct writing on wafer. Two types of resist, ZEP-520 positive resist and SAL-601 negative resist, were prepared for used in the electron beam lithography. Three different patterns, which include isolated line, contact hole and line and space patterns were exposed on the tungsten, oxide, and metal substrates, respectively. The 0.15 micrometers resolution of lithography patterns was achieved. For the etching of polysilicon and oxide, well defined profile of polysilicon gate with 0.1 micrometers width and well-defined tapered profiles of oxide contact hole have been obtained successfully.

Pattern placement errors are a problem in the manufacture of masks for proximity X-ray lithography. Many of these errors are attributable to long term drifts in beam position relative to external fiducials. To address this we have developed a technique based on through-the-membrane monitoring of the electron beam position. This uses a solid state detector with high bandwidth and gain mounted near the back surface of the target membrane. An accurately patterned overlayer on the detector provides the fiducial reference. The overlayer is designed to modulate the electron-hole pair current generated in the diode by absorbing the incident beam. Position information is obtained by analyzing the image created from recording the diode current during patterning. The phase in a Fourier transform of the data at the spatial frequency of the patterned absorber gives a measure of the position of the incident beam. Changes in the observed phase from one frame to the next can then be used to correct position errors of the beam in real time. We report results from tests of various components of this system. Early indications are that the system will be sufficiently fast and accurate for proximity X-ray mask pattern placement correction.

The performance of optical pattern generators as well as other lithographic tools was often limited by the performance of the data processing equipment and software used to control these tools. In fact, many performance attributes of modern pattern generators is due to the power and flexibility of the modern computers and software. For example, imperfections in the stage travel due to the mechanical components are often mapped out and corrected for in the control software. When key components are upgraded, these older tools regain useful life, at relatively low cost, for research and development and certain production applications. This paper describes modifications which have been made to a GCA 3600F pattern generator resulting in improved performance in key categories such as reliability and throughput. The modifications have been done to the operating system, data storage subsystem, job creation and monitoring system. Also, a generator has been constructed to simplify production of photomasks consisting primarily of regular arrayed patterns.

A new mask and wafer metrology tool is under final development at Leica to be launched in 1996. Recent performance data on the LMS IPRO system at Leica's development center shows precision and accuracy performance required for the 0.18 micrometers design rule device generation.

Two of the key performance parameters in the manufacture of photomasks and reticles are composite placement and composite overlay. Multipoint fitting of the placement and overlay errors is typically used in specifying the performance of the printing tool, whereas two-point fitting is more commonly used in the mask-production environment. In this study, Monte Carlo simulation techniques are used to compare the placement results that may be expected using the two methods. Mask yields are evaluated sing the maximum observed error as the primary metric, because this criterion is prevalent in the industry. The influence of several factors is examined, including number of sample points, size of test pattern, and the presence of systematic printing errors. The baseline test case consists of a 7 X 7 grid of points with an extent of 132-nm square with two additional alignment marks separated in the x direction by 144 mm, and random normally distributed errors with standard deviation (sigma) . For this case, the expected maximum error at 95% mask yield from a two-point alignment is approximately 4.5 (sigma) , and (tau) equals 1.3, where (tau) is defined as the ratio of the maximum error for two-point versus multipoint fitting. The expected maximum error depends logarithmically on the number of sample points, and (tau) decreases as the number of points increases. A sinusoidal systematic error in the printing system greatly degrades the yield using two-point alignment but has relatively little effect upon the yield using multipoint alignment. Similarly, an orthogonality mismatch between the printing and metrology tool causes (tau) to increase significantly, whereas a scale mismatch decreases (tau) . It is also demonstrated that the result from two-point alignment has more statistical uncertainty than that from the multipoint methodology, and thus multipoint alignment is more accurate in determining the expected performance of a given printing system from a limited set of sample masks.

Although the traditional method of O/L measurement (relative comparison between two levels) has proved to be a practical and cost effective way of measuring overlays, in the future this method will have to be supplemented by other means that require measurement of feature position in an absolute coordinate system and then comparing the output with the database rather than with some other level the accuracy of which remains to be established. The use of well calibrated Coordinate Measuring Instruments (CMI's) is one way to achieve the desired accuracy. But calibrating CMI's is a chicken-or-egg dilemma; you can't calibrate one without an accurately measured artifact, and you can't make the artifact without a well-calibrated instrument. Or so it seems. Positional self-calibration methods were invented to solve this problem and show great promise. But still there are many subtleties that must be resolved before such methods can be trusted. This paper explains the geometric basis for lattice methods of self-calibration and concludes with a theorem that demonstrates one of the striking difficulties that must be faced when relying on self- calibration algorithms.

In order to develop 1 G bit DRAM of 0.18 micrometers design rule, it is required to generate 0.2 micrometers contact hole patterns with local DOF over 1.0 micrometers . One of good candidates is DUV attenuated phase shift mask (PSM), which improves the lithographic process margin such as depth of focus (DOF), especially in contact hole patterns due to edge enhancement effect. In the case of DUV attenuated PSM, the optimum condition for contact hole patterns near 0.2 micrometers has been investigated by simulations and experiments using chromium- based attenuated PSM with the transmittance of 6% at 248 nm wavelength. We obtained local DOF of 1.2 micrometers for 0.2 micrometers contact hole of 1 G bit DRAM with printing bias of -0.046 micrometers using KrF laser system (0.31 (sigma) , 0.55 NA). We evaluated the characteristics of contact hole with various duty ratios and defect printability using programmed defects.

Excimer laser projection ablation is a dry patterning process in which an intense beam of ultraviolet light from an excimer laser is used to directly pattern a material. This technique has been used extensively in the microelectronics industry for patterning both organic and inorganic materials. Excimer laser projection ablation requires the use of a mask which is similar to a conventional 1X photomask. The laser ablation mask must withstand significantly higher energy densities than conventional photolithographic masks. A dielectric mask structure which consists of a quartz substrate coated with a stack of dielectric thin films has been developed for this process. Although the dielectric mask has been used successfully in a manufacturing environment, it suffers from the disadvantages of a complex fabrication process and high cost. Alternatives to the dielectric mask have been explored and a new mask has been developed which consists of an aluminum film on a quartz substrate. This mask meets the requirements for the laser ablation process and has the advantage of a low cost fabrication process which is similar to conventional chrome on quartz photomasks. The mask development, specifications, fabrication and results are discussed.

Embedded phase-shift mask (EPSM) has an advantage in comparison with several other phase-shifting mask approaches because of its simple structure and fabrication process. We tried to modify MoSi-based EPSM blanks by re-examining the material and by optimizing sputtering condition in order to produce more useful EPSM blanks for Deep UV lithography technology. New MoSi-based EPSM blanks for which Nitrogen gas is used as the reactive sputtering gas has been developed. And it has been confirmed that the New MoSi-based EPSM (MoSi-N) blanks are superior to HOYA previously developed one (MoSi-ON) in chemical durability, manufacturing stability and Dry Etching property.

It is out of question, that current state-of-the-art lithography--printing 350 nm structures with i-line tools or 250 nm structures with DUV tools--needs to correct for proximity effects (OPC). Otherwise, all the well-known effects like line-end shortening, linewidth variation as a function of adjacent patterns, linewidth non-linearity, etc. will produce a pattern, that is significantly different from the intended design. In this paper, we report first evaluation results of OPTISSIMO, a software package for automatic proximity correction. Besides the ability to handle full-chip designs by preserving as much as possible of the original data-hierarchy, there are significant options for the user. A large number of choices can be made to balance between the precision of the correction and the complexity of the corrected design. The main target of our evaluations was to check for full-chip OPC for the gate level of a state-of-the-art design. This corresponds to print either linewidths in the 350 nm to 400 nm range with i-line lithography or 250 nm/300 nm linewidth with DUV lithography. Taking 400 nm i-line lithography as an example, 3% precision OPC which has been demonstrated. By using hierarchical data handling, it was shown, that even the data complexity of a 256 M DRAM can be managed within reasonable time.

Edge rounding effect and line pattern length shrink (pattern edge pull-back) are serious phenomena in nanometer optical lithography. Serif and opaque dummy patterns are typically used for optical proximity effect correction in R&D stages. In mass-production of next generation DRAM, the mask yield should be decreased by using these associated patterns. It comes mainly from small pattern defects which violate the design rule. Unfortunately it is difficult to detect these defects with a recently developed inspection system. Also mask fabrication induced defects are linked to the resolving power of inspection systems. In this work, we present efficiency analysis of mask fabrication process for storage node layer of a 1 Gb DRAM developed by using several types of optical proximity effect correction methods which are used to reduce corner rounding effect and edge pull-back for enhancing storage area.

This paper reports advances to the Parametric Anchoring methodology in which a small number of representative line- space process measurements can be used to fit a process model to be used for Optical Proximity Correction (OPC). A correction rule table generation program uses the model to create the correction lookup tables to be used by a rules- based OPC correction program. The results of matching the process model to a variety of processes are shown.

In the e-beam world a simulator, comparable to well established optical simulators, has not been available so far. SELID (Simulation of E-Beam Lithography in 3 Dimensions) closes this gap by providing a comprehensive simulation tool covering most aspects of today's advanced e- beam lithography, such as process optimization and parameter determination for the e-beam proximity effect correction. SELID consists of 4 major parts: the simulation of the exposure step, the post-exposure bake and the resist development, and the analysis part. On output it displays many different views into the exposed image, 2D exposure images, as well as 2D resist profiles and resist structures in full 3D rendering. This paper presents first results using SELID. The application to direct write will be demonstrated. A commercially available positive e-beam resist was used for electron beam direct write lithography applications. Process optimization and the accuracy of the simulator will be demonstrated. Moreover, the agreement between experiment and simulation will be investigated.

For binary masks, focused ion beam (FIB) repair of opaque defects is of importance to improve mask yield and turn- around-time with the increasing packing density and continuing shrinkage of feature sizes. The effectiveness of the FIB mask repair is mainly limited by riverbedding, glass overetching, gallium staining, knock-on chrome, and image shift due to charging effect. These effects exert more influence on contact patterns than line and space patterns in the case of the same size of opaque defects under given design rule. In this paper, we investigate FIB repair of test masks which have background contact patterns with a variety of programmed opaque defects including unopened contact for feasible application of 64 M DRAM level. The geometry of the background contact pattern is designed for an approximate target critical dimension (CD), 0.41 micrometers whose feature size is a rough limit of the contact CD with reasonable depth-of-focus (DOF) and exposure time (ET) margin for binary masks when printed on a wafer with a 5X reduction i-line (365 nm) stepper. After repair with an optimum dose, a series of procedures for confirming its success is presented, that is, detectability, aerial image measurement software results, and wafer printability results. In addition, atomic force microscope results and 3D optical lithography simulation are presented for comparison. In our experiment and simulation results, (Delta) ET equals 100 msec and DOF equals 0.6 micrometers are obtained for the repaired contact.

In this paper, the deep submicron complementary metal-oxide- semiconductor (CMOS) devices were fabricated. A conducting poly(aniline-co-N-propanesulfonic acid aniline) (PAPSAH) as charge dissipation layer for e-beam lithography is used for direct writing the critical polysilicon gate level. Most of the specification of the designed CMOS parameters are met. The optimum conditions for submicron to 0.15 micrometers line/space of polysilicon gate using PAPSAH as charge dissipation layer for e-beam lithography are also established.

Owing to spin process used for mask manufacturing, the CD difference can be inevitably made in the radial direction by the centrifugal force. This difference can degrade CD uniformity of the given mask. So the ability of good maskmaking is closely related to the reduction of this difference as much as possible. In this paper, the CDCM (Critical Dimension Correction Method) is proposed. First, CD tendency of the given mask is mapped out. Then overall CD correction is done by making additional exposure for dose correction on the relatively underdeveloped CD region. The CDCM can also give some possibilities of improving wafer resolution by applying to the underdeveloped CD patterns on the wafer which are not corrected by layout design.

Critical dimension (CD) uniformity performance has improved significantly over the years by optimizing postbaking and development steps. As we continue to improve CD uniformity with PBS, our focus has shifted to other steps in the process. The crossover step, the point in the process where both develop and rinse are introduced simultaneously, has often been thought to have little or no effect on the develop time or to influence the uniformity results. However, careful examination of this step has determined that the crossover time, spin speed, and concentration of the develop/rinse mixture can have a significant effect on CD uniformity. Reduction of the typical radial effect was noted. Several design of experiment projects were run to examine these variables. A screening experiment was first run to determine the significant variables and their influence on uniformity. An optimization run was then made on crossover conditions, which included examination of plates from several vendors. Repeatability and sensitivity of the process to other variables were also determined. Results and recommendations are presented in this paper.

For bringing the halftone phase shift mask (HPSM) for 248 nm exposure into production, it is necessary to establish the mask-inspection method and mask-repair method. In this report, we will discuss the evaluation results of mask- inspection sensitivity test, comparison of mask repair methods and repaired-hole printability test. We fabricated a test mask that involves four kinds of programmed defect (clear defect; pinhole and protrusion, opaque defect; dot and intrusion), placed in 1.37 um hole array. Evaluation of defect sensitivity of KLA301 inspection system and defect printability using aerial image measurement system (AIMS) were done. Based on these results, we conclude that defect sensitivity for dot and edge defect (protrusion or intrusion) was slightly poor. We couldn't evaluate the sensitivity of pinhole. Because we could not fabricate pinholes smaller than 0.48 um and all programmed pinholes were detected. In respect of repair of clear defect, we could obtain a printability equal to holes without defects by deposition of carbon patch on the clear defect area, using a focused-ion beam (FIB). Three repairing methods of opaque defect (Laser-zapper, FIB, gas-assisted-FIB (GA-FIB)) were evaluated. Based on results of printability test using AIMS, we confirmed that GA-FIB is the best method for repairing of opaque defect. And we also report the ability of our HPSM using a chromium fluoride for ArF (193 nm) exposure.

Lucent Technologies (formerly AT&T) Microelectronics completed transition of their mask manufacturing to external vendors 1Q96. Maintaining a digital interface of all mask data and ordering information was critical to optimize the front end handshake with our vendors and to provide continuing support to our wafer fabrication lines. The goals that were set and achieved by using a totally digital interface are: reduced critical path vendor customer service and front end time; clarification and accuracy in meeting requirements; and ISO 9002 compliance, traceability and repeatability.

We have recently presented an operational principle based on the differential interference contrast (DIC) technique that we call `A-DIC method.' This method can suppress the circuit pattern images, and enhance the contrast between the images of the contamination and the defect-less circuit patterns. It uses both the transmission and reflection images. In this paper, we describe the other simpler method that we call `B- DIC' method.' This method requires only the reflection images and can be built in a confocal scanning microscope.

Micromachining can be considered a fabrication technology for building devices with feature sizes from a few microns to a few mils, often with moving parts. MicroeElectroMechemical Systems (MEMS) is a term often used interchangeably with micromachining. MEMS devices, however, are distinguished from micromachining by their mechanical nature, because moving parts are often a required component to enable them to sense or manipulate the external environment.

As defects are getting to be one of the most critical issues for next generation mask fabrication, Photomask Japan '96 Symposium held a Round Table Discussion on mask defect issues including inspection and repair targeting 0. 1 8 im device design rule and below. To clarify issues on defect specification, fabrication process feasibility with Cr wet etching Idry etching, and equipments of defect inspection and repair, eight panelists made short presentations and discussed those issues with audience. From a device manufacturer's stand point, mask making is required one or two generations ahead from the present targeting design rules, which enables chip size smaller resulting in profitable memory devices. New type defects are listed from a mask shop as transmission defects, resolution limit defects, semi-transparent defects and corner rounding, for which the Cr dry etching is advantageous over Cr the wet etching. Equipment makers also summarized critical issues for next generation. Defect inspection systems have several tasks such as improvement in sensitivity tunable with defect printability, feasibility to new type defects above, shorter wavelength optics, new system performance benchmark procedure, and preprocess software improvement to cope with optical proximity effect correction and phase shifting masks. Defect repair systems have several tasks such as accuracy improvement, substrate damage reduction, non-Cr material repair, three dimensional repair for alternating phase shifting masks and capability of phase and transmission control. The Round Table was held on April 21, 1996 at Kanagawa Science Park, Kawasaki, Japan. Keywords; mask, photomask, defect, defect inspection, wet etching, dry etching, defect repair, focused ion beam, optical proximity correction, mask data.

As field size is increased and design rule goes down to O.l8pm, more tight control of defects on the reticle is required than before. T his requirement is m ore critical on contact patterns than line and space(L/S) patterns in the case of the same size of defect. In this paper, test reticles which have background normal contact patterns and background OPC(optical Proximity Correction) contact patterns, respectively, have been investigated for 256M DRAM level. We included a wide variety of programmed defects in background patterns. The geometry of the background contact patterns on the reticle has been designed for 0.3 1 pm m ore or less when printed on a wafer with a 4X reduction stepper. The programmed defects are used for the sensitivity evalution of a defect inspection system, i.e, detectability. Also printability of these defects on the wafer is performed by using a 4X reduction DUV(248nm) stepper to deterrn me not only the reticle defect specification but also the detectability required for the defect inspection system. Test results are as follows: First, the defect specifications required for 256M DRAM can be determined. Second, OPC contact patterns have better wafer results than normal contact patterns in view of depth of focus(DOF) and defect printability. Third, inspection technique for OPC patterns remains the task to be solved. Keywords : detectability, printability, programmed defect, OPC, DOF, serif, simulation

It is now generally accepted, that optical lithography will be the mainstream approach to manufacture the I Gbit DRAM device generation with minimum features between I 80 nm (first working samples) and I 50 nm (projected fabrication). This development demands a considerable tightening of mask specifications. The printing of 180 nm I 150 nm features on a 193 nm-, 0.6 NA- tool is a highly non-linear pattern transfer process (ki = 0.56 I 0.47). Therefore, mask irregularities (defects, linewidth variations) will print easier than at larger dimensions. This paper presents results of full lithography simulations for the printability of mask linewidth variations and mask defects. Assuming, that the wafer linewidth error budget is shared between the lithography process and the mask as it is currently done (75% for the process, 25% for the mask), the linewidth uniformity on a mask has to be better than 12 nm for 180 nm designrules and better than 8 nm for 150 nm designrules. Masks have to have no defects larger than 100 nm resp. 70 nm (all those numbers are given at the 4x mask!). These numbers are almost a factor of 2 tighter than assumed in published mask specification roadmaps. Besides tightening the current specifications, the mask roadmap has to include other important parameters: We show, that the butting error of the mask writing tool, corner rounding, and the quality of the mask repair have substantial impact on the linewidth error on the wafer. Since mask defects print proportional to the squareroot of their area, the definition of mask repair quality has to be revised. Current mask repair processes cannot provide adequate repair quality. Since masks with these specifications are - in our point of view - not easy to make, we propose to reevaluate carefully the sharing of the wafer Iinewidth error budget. In other words, 193 nm lithography must dedicate a significantly larger portion of its (actually not so large) process window to the mask. Keywords: Optical Lithography, Mask Specifications, Mask Defect Printability, Mask Repair, Linewidth Error Budget

Reticle cleanliness is becoming an ever more serious issue in today's sub-micron wafer fabrication facilities. Line-widths on reticle are at or pushing below one micron. The defined size of what is a "killer' defect has shrunk accordingly. The small particulates added in shipping, reticle storage, and reticle handling are having a much greater impact on device yield. Current methods of reticle protection and storage are proving inadequate in preventing an accumulation of particles that would previously have been considered inconsequential. Device yield as a function of reticle cleanliness has become a major concern. Cost and turn time also factor into the need for improved reticle protection and handling methods. The cost for a super high-end reticle has sky-rocketed into the 5 figure range for a "traditional" binary, chrome-onquartz photomask. Holding up a wafer lot for what can be a two week replacement turn-time on critical spec reticles can result in catastrophic delays to production. Protection of these manufacturing critical resources is an absolute necessity. As a result of these issues, increased focus is being applied to developing truly secure methods for the protection, shipment, and handling of reticles. Specific problems include: . Inadequate protection provided by current stepper storage boxes . Shortcomings of shipping containers and sealing mechanisms . Over-reliance on direct manipulation in reticle handling practices

One of the main challenges for the industrialization of the alternating phase shift mask technique is the geometrical design of the phase shifters. The problem is complex and software development is required to achieve automation. This data preparation is necessary in order to apply the most powerful optical enhancement technique, the alternating PSM, to the gate lithography of logic designs. After the implementation of an algorithm, very promising on small flat geometrical designs of a few hundred transistors, it has been decided to test this software on complete circuits. This paper deals with the issues associated with alternating pattern generation on complex circuits. In particular, the use of layout hierarchy is essential to obtain an acceptable amount of manual intervention and computer processing time. There are two kinds of limitations to this global approach. First, the overlapping or the proximity of cells in the hierarchy may not be compatible with a cell-by-cell generation of phase shifters. Secondly, some local design configurations may be difficult to phase shift. The question is: `is the number of occurrences of such cases manageable?' This presentation is illustrated by our experience on a multiplier design of about 15000 transistors, created using a standard cell library in 0.25 micron technology. This work shows that the theoretically difficult cases of layout hierarchy or geometrical design are not critical in practice. In particular, the design method based on standard cell libraries creates a hierarchy which is very favorable to our approach. This article presents a method for the alternating pattern generation on complex circuits which is based on two main points: first, on a software able to generate alternating phase shifters on small flat cells, and secondly, on the advantage of using layout hierarchy.

A mathematical model of resist heating was developed based on the 3D classic heat conductivity equation in a multilayer object. A tabulation of special functions used for numerical calculation allows the analytical solution to be simplified to one only integral for both variably shaped beam and Gaussian beam. As a result, a dynamic distribution of temperature field in resist can be simulated for an exposure process in terms of thickness and characteristics of resist and multilayer substrate, as well as size, placement and consequence of flashes. Results of simulation are shown for both short and long time range exposures.

Can the i-line resist OCG895i, normally exposed with UV radiation from an i-line (365 nm) mercury discharge or an argonion laser (364 nm), be exposed with the visible 407 and 413 nm lines of the krypton-ion laser. If so, with what quality? Visible radiation is desirable since it simplifies the optical design, and makes the system more reliable and easily serviceable. Simulations with PROLITH/2 show the lithographic performance to be equal or even slightly better at the longer wavelengths. Optical theory gives that at equal resolution there is a marginal loss in depth of focus (12%) at 413 compared to 364 nm. However, this is compensated for by the better optical design of the optical system —in particular the better corrected field curvature — made possible by the much wider range of glasses available in the visible. In a practical context there results no resolution penalty from using 413 nm exposure. Experimental reticle exposures with 413 nm and NA = 0.42 using a large-area writer for micro-tip FED lX masks confirm the viability of the 895i in the visible. Furthermore, the good image decay properties at 364 nm are retained at 413 nm.