BACUS Panel Considers Model-based Data Prep
By Marc David Levenson
“What does the e-Beam Initiative want to be when it grows up?”
That was the plaintive question asked by Douglas Resnick, V.P. of template technology at Molecular Imprints, at the end of a BACUS panel discussion hosted by Jan Willis and Aki Fujimura of the e-Beam Initiative, a consortium of 41 companies focusing on electron beam lithography technology.
The e-Beam Initiative’s emphasis had been on making chip production and e-beam technology more compatible with one another, but that has broadened. Fujimura pointed out that – whatever the production technology – the first physical pattern was always written by some e-beam technology and thus electron technology would always be critical and thus now and (probably) forever deserving of more funding. Motivating that R&D investment was the goal of the e-Beam Initiative, according to Fujimura.
Resnick replied that the Initiative needed more focus: the larger industry just could not comprehend one R&D program affecting EUV and DUV mask-making, imprint, “complimentary” and maskless lithographies. It had to pick one, or perhaps two.
Still, the presentations and discussions at the e-Beam Initiative event illustrated Fujimura’s point that e-Beam technology was essential and highly challenging on all time scales. The beam blur of today’s mask writers is such that mask patterns cannot reflect the design data without simulation–based correction. Otherwise, contact holes on masks come out small, variable and round rather than square, reliable and accurate. The solution, according to Fujimura, was model-based mask data prep (MB-MDP), in which e-beam shots are overlapped and the doses specified in the data sent to the mask writer, rather than relying on the writer’s ability to correct for proximity effects during write. The number of shots and mask cost can be dramatically reduced while improving wafer image quality, he said.
As it happens, D2S, Inc., Fujimura’s home company, offers such MB-MDP software, and Gek Soon Chua of Global Foundries described how that software reduced variation in a 20nm SRAM contact level, laid out via MB-MDP based on “ideal” inverse lithography patterns. The DOF, CDU, process-variation (PV) band and MEEF are all dramatically improved even though the number of shots was reduced by 33 percent, according to Chua. The worst-case PV-band fell to the level of the “ideal” OPC for <3000 shots whereas Manhattanized ILT required 13,000 shots and a 5nm resolution to come within 10% of “ideal.”
Christian Bürgel of the Advanced Mask Technology Center in Dresden pointed out a future issue: EUVL mask substrates scatter differently from fused quartz, resulting in a multi-range proximity effect, with scales from 10mm to 20nm. Such complex proximity effects alter the CDs on 2xnm node EUV mask SRAFs and must be corrected by dose modulation, implemented in the mask data prep stage, not just by the mask writer, according to Bürgel. Related technology has been developed for e-beam direct write and just needed to be applied to EUVL masks, he concluded.
The subsequent panel discussion included representatives of the major mask makers. They pointed out the other challenges to EUVL mask making, but predicted first use in 2013. MB-MDP will be needed sooner than 20nm, but will have to overcome hardware inadequacies – both in the mask writers and fracturing computers – and its unfamiliarity. While MB-MDP is optimized for today’s vector-scan shaped-beam tools, dose modulation and other aspects can be implemented on future multi-beam raster-scanned systems with further collaborative R&D, Fujimura argued. Jan Willis predicted that the industry would find common ground and work together to implement MB-MDP once it realized that image quality (and thus yield) are improved while mask costs are reduced.
