Automated inspection of OPC and PSM masks
07/01/1997
Automated inspection of OPC and PSM masks
Wolfgang Staud, Karen Huang, Patricia Beard, Photronics Inc., Milpitas, CaliforniaYair Eran, Applied Materials/Orbot Instruments, Santa Clara, California
Candidate projection exposure methods for pattern dimensions below 0.25 ?m include e-beam and x-ray, but these processes are costly and complex. Deep-UV (DUV) optical projection lithography, combined with optical enhancement techniques like PSMs and OPC, is expected to be the industry norm in the medium term [1]. Attendees at a recent SEMATECH-sponsored Litho/Design workshop concluded that the eventual adoption of PSM is inevitable, despite many outstanding problems of design, fabrication, and inspection.
Optical enhancement
While a photolithographic mask should produce an image that is identical to the reticle pattern, diffraction and scattering effects prevent achievement of this ideal with conventional "binary" masks. Altering the mask`s transmittance parameters can substantially improve the quality and definition of a given image. In PSM masks, judicious modification of the relative phases and amplitudes of the transmitted light allows fabrication of extremely narrow lines and small features. OPC, a complementary approach, "shapes," or selectively distorts, the features on either binary or PSM masks to improve the quality of the resulting image. For example, an OPC mask might apply serifs to rectangular corners of the reticle pattern [2].
This article discusses some of the practical manufacturing issues facing optical enhancement masks that exploit PSM and OPC techniques, with particular emphasis on mask inspection. We also present some interesting conclusions for design for manufacturability (DFM) from mask verification experiments.
 |
Figure 1. Mask OPC1, 4 subfields a) uncorrected, b) 1.0D, c) 1.5D, d) assist slots. The subfields are arranged in clockwise order on the mask, as Fig. 4 illustrates.
Phase-shifting masks. PSMs produce extremely narrow and well-defined lines and features in an optical image. Improved DOF and expanded process latitude make PSMs a viable but expensive enhancement technique for up to six of the total 18-26 layers in typical device mask sets.
A conventional transmission mask controls only the intensity of the light passing through it and, consequently, the pattern feature resolution is diffraction-limited. PSMs control the phase and amplitude of the transmitted light through destructive interference in order to minimize the undesirable image-spreading effects of diffraction. In the simplest PSM arrangement, similar amplitudes with opposite phase (180?) are exposed on opposing sides of reticle line edges.
A conventional PSM is a standard transmission photomask with an additional transparent layer patterned so that waves transmitted through adjacent apertures are 180? out of phase with each other. As a rule of thumb, any optical system will project a phase-shifting reticle with almost double the spatial resolution and greater contrast than the equivalent nonphase-shifted reticle [3]. Until recently, two main varieties of PSM were in use: frequency-doubling masks with alternating, periodic, phase-shifting apertures; and edge-enhancement masks with small, phase-shifting regions arranged around the main apertures. The halftone PSM [4], a more recent version of the edge-enhancement mask, consists of a transparent substrate and an attenuated phase-shifting halftone layer with transmittance of between 5% and 20%. Halftone PSMs perform better than conventional PSMs and, consequently, are increasingly common.
 |
Figure 2. OPC2 corrections and preprogrammed defects.
Optical proximity correction. OPC uses pattern compensation modifications of the photomask to reduce systematic print and etch biases resulting from optical proximity effects in the mask- and wafer-manufacturing processes. Though OPC is an accepted technique for sub-0.5-?m lithography, controllable manufacturing of OPC masks in production quantities may be challenging. OPC systems promise increased wafer yield, wider process latitude, and tighter control of line width uniformity.
Design for manufacturability
Manufacture of reticles from software-generated OPC or PSM mask patterns has posed considerable difficulties for maskmakers. While the semiconductor industry seeks eventually to automate conversion of IC designs into production-ready optical enhancement masks, semi-automated or manual layout methods are currently employed [5, 6]. Manufacturability constraints on data densities; feature shape and size; and, most importantly, the linearity, uniformity, and minimum line width of resulting patterns (over 130 x 130 mm mask blanks), prevent designers from applying optimum corrections [7].
To keep up with the demand for extremely rapid turnaround, most fabs can only inspect new masks for critical dimension and registration. The fabs expect maskmakers to guarantee the quality of the mask. More complex inspection of optical enhancement features by the maskmakers could drive reticles to three, five, or more times their normal cost. The interdependence of pattern writing, processing, and inspection presents new challenges.
Manufacturing matrix
We used standard laser-based and e-beam writing tools, together with their respective OCG895i and poly-butene-sulfone (PBS) processes, to fabricate masks for our verification study. The test patterns contained very aggressive corrections and assist slots below the specified resolution limit of the laser tools, so laser writing was abandoned early on. The e-beam masks used a standard PBS process on 1000-? antireflective chromium (ARCr) with 4000 ? of resist, using DP13 developer, 110?C bake, descum, and CR7S wet etch. A proprietary "puddle" process in conjunction with e-beam proximity techniques enhanced the resolution, linearity, and overall uniformity of plate "OPC2."
Inspection systems. The Orbot RT-810 and RT-8000 mask inspection systems were used for this study. These die-to-database inspection systems provide different illumination and optical resolution and, consequently, different defect sensitivities. Data extraction for the die-to-database inspection and scanner control were performed on separate workstations. A third workstation performed automatic classification of defects on a specific slice, while simultaneously scanning the next slice on the mask. This capability allows the operator to determine, in only a few slices, whether the inspection parameter settings are appropriate for the patterns in question.
A matrix of test plates, including both OPC and PSM, was used to investigate some of the more challenging inspection tasks.
 |
Figure 3. PSM test plate with AA, rim shifter, and CL subfields.
 |
Figure 4. Defect map of OPC1 inspection by Orbot RT-810.
OPC patterns. We examined two OPC patterns. OPC1 is a SRAM device with four individual fields containing different OPC corrections generated by a TVT (TransVector Technologies) correction package. The four fields (Fig. 1) were: a) uncorrected, b) 1.0D correction (extending or widening lines), c) 2.0D correction (serifs to compensate for corner rounding), and d) assist slots for use with off-axis illumination. The OPC1 pattern was also imaged onto a MoSiON-based, embedded attenuated (EA) PSM blank with 8% transmission.
OPC2, also generated using TVT conversion software, had a range of patterns with various degrees of correction (in some cases "overcorrecting" the pattern), including assist slots with minimum resolution requirements around 0.2 ?m. The test plate also included preprogrammed defects in a simple line pattern and an OPC pattern with assorted corrections (Fig. 2).
PSM patterns. The study evaluated three different PSM fabrication techniques. In the first technique, the alternating aperture (AA) plate contained a specially designed test pattern consisting of contacts ranging in size from 1.0-2.0 ?m in 0.2-?m increments. The complete plate contained 27 million contacts (Fig. 3).
In the second technique, a simple test pattern was converted into a chromeless (CL) shifter by dry etching the SiO2 after chrome etch, then stripping the chrome to create a CL test plate. Finally, for the third technique, the OPC1 pattern was imaged and dry etched onto a MoSi EA mask blank to create mask OPC3.
 |
Figure 5. OPC2 test pattern defect triggers.
OPC mask inspection
For the OPC1 plate, the inspection was set up as a single file for all four sub-fields. The defect map (Fig. 4) confirms that the inspection system reliably handled OPC patterns with increasing aggressiveness. The uncorrected field (upper left) shows a typical e-beam scan line error. A similar error can be more clearly observed in the 1.0D field (upper right), which also contains several more flags owing to geometry distortion. The 2.0D field (lower right) shows a significant increase in the number of defects, mainly because of rounding effects in the plate write and process.
The inspection tool apparently fails in the field containing the assist slots (lower left). However, a closer look at the correction reveals serious problems with the original OPC data conversion. The conversion produced 0.2-?m line width minimum assist slots, and the data file left gaps as low as 0.15 ?m between these slots. The e-beam process cannot resolve these features. This result has important implications for establishing software "deny rules" for OPC features below 0.4 ?m. The MoSi version of the OPC1 pattern (OPC3) yielded similar results.
OPC2 was the most challenging plate due to the variety of OPC test patterns. The test patterns range from simple to very aggressive corrections, with assist slots and completely overcorrected 2.0D serifs (Fig. 5). The bottom row of this pattern also has over 1000 "preprogrammed" defects in various locations and sizes. E-beam proximity correction in the write process gave superior resolution and linearity.
Figure 6 compares the "decorated" plate patterns against the "undecorated" database for one subfield of OPC2. The defect map generated from this sub-chip contained over 1300 defects concatenated into 710 total defects, of which 24 were identified by the system as "false." The total inspection time for this small pattern was 51 sec; the autoclassification took an additional 102 sec. The defect map clearly identifies the area of the preprogrammed defects in systematic rows, but it also shows some additional flags in the neighboring patterns.
One defect trigger stemmed from an overcorrection: hammerhead serifs on the line-endings included small assist features. Some of these features were clearly resolved by the inspection tool, but smaller ones were no longer correctly identified. The scanning electron micrograph (SEM) of the same pattern area shows a 0.2 x 0.6 ?m assist slot.
 |
Figure 6. OPC2: inspection of one subfield comparing "decorated" plate to "undecorated" database.
 |
Figure 7. a) Measurements made by the KMS300, a confocal line width measurement system, and b) 3-D reconstruction of AA contacts.
PSM inspection
The die-to-database inspections on the AA pattern showed no major problems, although an interesting phenomenon led to further investigation of the plate. In order to hide the shifter edges beneath the chrome, the contact patterns did not undergo the usual "post-etch" treatment after dry etch. Every second contact appeared smaller in the inspection and also exhibited a slightly lower intensity. The CD measurements system in both incident-confocal and transmitted light showed the same effect (Figs. 7a and 7b). In the SEM, however, top-down measurements showed almost equal size with good linearity and normal edge roughnesss for both etched and unetched features. Most actual defects stemmed from contamination during handling.
The chromeless shifter mask was investigated to check the possibility of creating a database for such inspections. A simple resolution pattern was imaged, wet and dry etched onto SiO2, and the chrome completely removed in a wet strip. Initial scanning of the remaining step patterns showed an approximate line width for the steps of 0.6-?m in transmitted light. Geometric manipulations created a database containing only the outlines of the figures. Comparing the prepared database and the scanned image allows normal die-to-database inspection.
Conclusion
OPC inspection. Corrections from 1.0D-2.0D did not present major problems in our pattern inspections. A "manufacturability" line can be plotted at 0.5 ?m for minimum OPC features (assist slots and serifs), with jogs no smaller than 0.1 ?m at a minimum of 2 address units for resolution. A jog width/length ratio of 1:4 will help the inspection tools clearly distinguish between edge defects and intentional OPC jogs. Additional geometry-dependent defect triggers will occur at smaller feature sizes.
Database file size and, consequently, write and inspection speed, greatly influence mask production cost. E-beam proximity correction and dry etching substantially improve manufacturability.
PSM inspection. We inspected both AA and CL phase shifters in a die-to-database mode. Effects from the dry-etch process were the major contributors to defect densities at feature sizes of 0.3 ?m or less. Not surprisingly, cleanliness is an important factor in minimizing these defects. The same database creation techniques can, in principle, be used for the inspection of exposed shifter edges in multiphase AA shifters and CL shifters. Separate review of 0? and 180? patterns is feasible through judicious "coding" of the database according to the exposure levels. So far, no satisfactory explanation has been determined for the phenomenon of "missing" shifters in PSMs.
MoSi (or thin Crxx) EA shifter masks can be regarded as standard inspections, owing to the inspection tool`s ability to expand the gray-level scale to accommodate the halftone character of the material. Defect mechanisms are the same as observed on all other plates, due to the same processing and dry-etch conditions. Repair of defective PSM and OPC masks remains an open issue for the mask-manufacturing industry.
Acknowledgment
This paper is based on original research published at SPIE in September 1995, Vol. 2621, pp. 597-611.
References
1. Z-T. Jiang et al., "Optical Property Simulation of Single-layer Halftone Phaseshifting Masks for DUV Microlithography," Semiconductor Science and Technology, Vol. 11, No. 10, October 1996.
2. C.A. Mack, "Trends in Optical Lithography," Optics & Photonics News, Vol. 7, No. 4, pp. 29-33, April 1996.
3. M.D. Levenson et al., "Improving Resolution in Photolithography with a Phase-shifting Mask," IEEE Trans. Electron Devices, Vol. 29, p.1828.
4. T. Terasawa et al., "Imaging Characteristics of Multi-phase-shifting and Halftone Phase-shifting Masks," Japan J. Appl. Phys., Vol. 30, p. 2991.
5. O. Otto, J.G Garofalo et al., "Automated Optical Proximity Correction - A Rules-based Approach," SPIE Optical/Laser Microlithography VII, 1994.
6. J. Stirniman, M. Rieger, "Fast Proximity Correction with Zone Sampling," SPIE Optical/Laser Microlithography VII, 1994.
7. C. Spence et al., "Manufacturing Issues for OPC Masks," OCG Conference, 1994.
WOLFGANG STAUD received his Photo Ingenieur degree in photo-engineering from the Technical University of Koeln, Germany, in 1984. He has worked in lithography for more than 12 years in various engineering and management capacities, at companies like Fairchild, LSI Logic, Nikon, and Photronics. Staud`s interest is in the connection between write/process/inspect. He is presently senior technical sales engineer at Ultratech Stepper.
KAREN HUANG received her BS in photoelectronics engineering from Tianjin University, Tianjin, China, and her MSEE degree from Northeastern University. She has more than four years of experience in lithography. Huang was research and development engineer at Photronics, and is presently process engineer in the Micron Display Technology Division of Micron Technology.
PATRICIA BEARD received her BS degree from the University of California, Santa Cruz. She has 15 years of photomask engineering experience at companies like Photronics and Hewlett-Packard. Beard is presently senior optical inspection analyst at Phase Metrics.
YAIR ERAN received his BsC degree in physics from Technion University, Haifa, Israel, in 1977, and his MsC and PhD in computer science and mathematics from the Wiezmann Institute of Science, Rehovot, Israel, in 1981 and 1987, respectively. He works for Orbot Instruments, Israel, where he has been algorithm group head of the reticle inspection product line since 1988 and manager of the RT-8000 since 1996. Applied Materials-Orbot, Industrial Zone, POB 601, Yavne 81106, Israel; ph 972/8-942-5695, fax 972/8-942-1644, e-mail [email protected].