FSI Lithography Breakfast: CD control, throughput, cost
10/01/2002
The 9th annual Lithography Breakfast forum sponsored by FSI highlighted the three key issues in production lithography today: CD control, throughput and cost. Ardy Johnson, the host of the forum and VP of strategic marketing at FSI, pointed out that the need for few-nanometer CD variation — both wafer-to-wafer and within-wafer — required that the stepper and track each contribute less than ~3nm of variation.
That is especially difficult for a track which must accommodate the >125 wph output of the latest steppers. Such tracks may have 40 hot and cold plates and 25 possible critical paths through the tool, all of which have to be balanced to produce the same result. Subtle process improvements such as pre-wetting wafers with overlapping developer dispense also contribute to uniformity.
The keynote speaker, Tony Yen of International Sematech, looked both forward and backward, pointing out that past wavelength changes — though not painless — enlarged process windows and thus made production easier. In particular, shorter wavelength allowed higher k1 imaging, which made the image dimensions less sensitive to projection lens aberrations. Currently, aberrations are <0.02 waves at 193nm, which simulations indicate should contribute <3% CD variation, even at k1 = 0.35.
The problem is that the 65nm generation will have to work at very low k1 values (and small windows), even if 157nm tools become available in time. Yen predicted that those tools will be available in 2004 in spite of intrinsic birefringence and hard pellicle difficulties. Beyond that, Yen offered immersion lithography as a way to shorten the effective wavelength to 115nm, facilitating 55nm half-pitch lithography at k1 = 0.43 in 2007. If that comes to pass, the decade-old Sturtevant plot (which depicts the date when optical lithography is supposed to collapse vs. the date of the prediction) will extend to 2009!
Phil Ware of Canon speculated as to whether the routes on the ITRS Roadmap should have safety chains to prevent speeders from going over cliffs. The "silly-con cycle" as described by Ware begins with the prediction that some technology cannot be extended, motivating research into a new lithography system, which is delayed, requiring the original system to be extended, at which point the process starts again! Ware said that Canon, rather than waiting for 157nm technology to be ready, was bringing out a 157nm-compatible platform with high-NA 248 and 193nm optics. The FPA-6000-ES5 and AS4 will have lenses with <0.01l RMS wavefront error throughout the field and stage errors <2nm. The FPA-6000 single stage (with twin linear motors and reaction masses on each axis) produces a 140wph throughput (300mm wafers), according to Ware.
Jitze Stienstra of ASML gave an update on the dual-wafer stage of the Twinscan system and claimed a throughput of >100wph under realistic exposure conditions: Since the dual stage system allows more time for alignment and wafer height measurement, process control can be improved. Stienstra claimed that the ArF Twinscan produces 100nm isolated lines with full-wafer CD uniformity of 5.7nm, when the mask employs assist features. Without reticle-based enhancements (but with annular illumination), that tool has printed 80nm line space patterns with a 450nm focus window. Stienstra also claimed that ~50 Twinscan systems had been shipped. If all were ArF 300mm tools, that would make it the most popular tool of its generation.
Gene Fuller of Nikon described the performance of Nikon's latest ArF scanner — the 306C. With an exceptionally low-aberration, large NA lens, it prints 90nm line-space test patterns, limited by resist collapse. (Fuller allowed that the 193nm resists were still not up to the standard that exposure tool manufacturers would like them to be.) The flare of the 306C lens was <1% throughout the field and relatively uniform no matter how measured. Nikon continues to produce advanced 248nm scanners and announced an i-line scanner with improved resolution, overlay and throughput. At 157nm, Nikon believes that hard pellicles will force more complex reticle stages, capable of moving in the z axis. In the NGL area, Nikon has completed an electron projection lithography test stand capable of printing 60nm contact holes and gates (with negative resist) at throughputs above 20wph.
Doug Anberg, senior director of marketing at Ultratech Stepper, presented the most controversial talk. He noted that the reduction in CDs has allowed the die sizes to shrink below those predicted when the exposure field size was standardized at 26×35mm. More than 90% of chip designs would fit into an 11mm2 field, according to Anberg. Might that market be better served with a low-cost 6× stepper with a small field than with today's very expensive large-field scanners Reticle costs would be reduced by the larger feature sizespossible at 6× as well as by the smaller area being written. Anberg speculated that such a $10M "production-field" stepper would halve lithography costs for 100nm generation devices with runs less than 2000 wafers.