The lithography crsis is now
06/01/1998
The lithography crisis is now!
Cell teams show results
M. David Levenson
Mary Swedberg
The 1997 SIA Roadmap for future semiconductor manufacturing poses some extraordinary near-term challenges for wafer lithographers and maskmakers. Put bluntly, the lithography
eticle guidelines for critical levels are inconsistent with 1998 optical realities. One reality is that 193-nm exposure technology is very late, too late to affect production before 2002. Another is that the k1 factor (k1 = CD ? NA/l), which defines the quality of an ideal aerial image, is about to take a dive below 0.5 for the 180-nm generation. At k1 = 0.5, certain features (such as dense line-space patterns) begin to disappear from conventional images. While recent advances in photoresist and optical proximity correction (OPC) have allowed production with k1 factors near 0.6, nothing - certainly not wishful thinking - will recover features that have disappeared!
Worse, when something is just barely present in a low-contrast image, tiny variations in the photomask dimensions (or anything else) can cause large changes in the resist CDs. This is the problem that Wilhelm Maurer of Siemens has been predicting for some time. According to the Roadmap, 1998 is the year it becomes a crisis. CD variations due to defocus and exposure errors add to those from the photomask. If the semiconductor industry wants a process window for the 180-nm generation - with 140-nm gates and k1 = 0.5 - the full range of variability of 4? reticle features should be even less than the ~20 nm specified in the SIA Roadmap. After the 180-nm generation, scheduled for first shipment next year, it only gets worse.
There is, however, an alternative: "resolution enhancement technology" or "wavefront engineering." In the simplest scheme, lithographers would switch their illumination to "annular" or "quadrupole" and recover about half of the lost process window. The catch is that all the masks will need assist-feature OPC to print narrow isolated gates and other key structures. On a 4? mask plate, the assist features will be ~280 nm wide, and the CD uniformity spec would still be around 20 nm. The exposure latitude will still be small because the image contrast will stay low. Stepper makers are hoping that the resist chemists will find a way to develop good patterns even if the minimum dose is 60% of the dose-to-clear!
Phase-shifting masks (PSMs) could increase contrast and make things easier. Hua-Yu Liu of Hewlett-Packard has shown that strong PSMs print isolated gates down to k1 = 0.25 with minimal image degradation. With proper mask design and stepper settings, the deep ultraviolet (DUV) aerial image would also contain line-space patterns, contacts, etc., with the CDs and contrast required for the next several generations. Embedded-shifter attenuated-PSMs combined with properly tuned, off-axis illumination show similar imaging performance, especially if a patterned opaque layer eliminates unwanted sidelobes. Because of some subtleties of optics, the CDs in the resist patterned by these types of mask can be relatively insensitive to the mask dimensions. In particular, an alternating-aperture PSM will print a 140 ?14-nm isolated gate whether the chrome-width at the phase-edge is 400 nm (at 4?) or zero! The simulated process windows easily exceed the SIA Roadmap requirements.
The problem is that to make this "wavefront engineering" strategy credible, the photomask industry will have to produce a large number of phase-shift masks - economically, free of printable defects, and with good turnaround times - this year. If everyone who was buying 300-nm-and-below masks in 1996 and 1997 buys 180-nm PSMs in 1998-1999, the demand will be roughly 25,000 advanced DUV masks. About 10,000 of these will be alternating-aperture-PSMs, built with conventional plates, but requiring two to four pattern-writing stages. The rest would be DUV attenuated-PSMs of various types. Supplying these masks will be a considerable challenge. For one thing, production-quality DUV embedded-attenuated PSM blanks are unavailable because of some unsolved contamination problem in the sole vendor`s deposition process. The DUV-PSM market is small because of concerns about quality, and thus does not generate the revenue to motivate production-level quality. That has been the situation for some time, but the alternatives to achieving a high level of quality production in 1998 are not attractive.
One option for producing the 140-nm gates required in 1999 would be conventional chrome-on-glass (COG) masks (with CD uniformity well below 20 nm) exposed on advanced steppers with miraculous resists and heroic process control. The exposure latitudes would be below 7% with depth of focus (DOF) <300 nm, less than half the Roadmap value. Using COG masks with annular illumination improves the DOF, but the area of the 180-nm window would still be less than half that of the 250-nm generation. Some manufacturers are hoping that 1? synchrotron radiation (x-ray) lithography will come on-line in time for the 180-nm generation. If x-ray became the leading-edge production paradigm, the photomask industry would have to ramp up defect-free x-ray mask production to >10,000/year. Or, the semiconductor industry could stall, with no higher-capability new products to drive demand for several years!
The lithography crisis, which arrives this year, is not going to be a short-term problem. According to the SIA Roadmap, the minimum k1 factor will be 0.4 (and below) from now until the end of optical lithography, sometime around the 100-nm generation in 2006. We have long known in principle how to make good low-k1 resist profiles. According to the SIA, we have to do it in production next year.