The eBeam Initiative could slash mask costs at 22nm
Ron Wilson - February 23, 2010
The eBeam Initiative, which has been tightly focused on eliminating the need for photomasks at critical layers in today’s leading-edge processes, today announced an additional project: making photomasks feasible at the 22nm node. That seems like a sweeping enough claim to require some explanation.
The issue starts with announcements at the SPIE Advanced Lithography Conference in San Jose this week. To no one’s surprise outside Intel, the eUV stepper program is delayed once again. Extreme ultraviolet lithography will not be ready for the 22nm logic node (the 32nm half-pitch, if you are a lithography aficionado.) And when it is available, eUV will probably only be financially and logistically feasible for lines that run enormous numbers of identical wafers. It may be another five or six years, lithography experts suggest, before eUV is available for and applicable to the lower-volume runs of SoC manufacturing.
Fortunately for Moore’s Law, the wizards who actually understand arcana like computational lithography and advanced optical proximity correction (OPC) have anticipated this delay. They have figured out how to make masks so that enhanced versions of today’s 193nm immersion steppers can print even the most critical masks, such as the contact layer, onto 22nm wafers.
Researchers at Samsung, for instance, have found that by using curvilinear shapes and curved assist patterns they can achieve a significant improvement in depth of field and variation tolerance compared to masks using conventional rectangular shapes and straight assist bars. Samsung photomask team principle engineer Dr. Seong-Sue Kim described these results at an eBeam Initiative presentation this afternoon. These attributes, Kim suggested, translate fairly directly into smaller process variations and/or better wafer throughput. Not only do the curved figures print more accurate patterns onto the wafer, but they pack more densely on the mask, allowing, for example, slightly denser arrays of contacts.
But the impact of these curvilinear figures on mask-writing time is horrendous. Mask writing, remember, requires fracturing—breaking the figures to be placed onto the mask into combinations of lines and rectangles that e-beam systems can project. If the figures you are trying to put on the mask are curved, the conventional fracturing process approximates the curves with huge numbers of little rectangles (or, alternatively, with large numbers of overlapping rectangles, which has problems of its own.) Depending on the size and shape of a figure, it may fracture into dozens of rectangles.
But dozens of rectangles per figure, warned Toppan CTO Franklin Kalk, will drive mask writing time to 50 hours per mask. That represents $60K per mask in capital costs alone, Kalk warned. Worse, others point out, neither the e-beam systems nor the resist materials today are stable enough to support a two-day writing session. "We need to get down to a ten-hour write time," Kalk insisted.
Here is where the eBeam Initiative rides to the rescue. The organization has repurposed the concept that makes it possible for them to write patterns directly onto wafers with acceptable throughput. The technique is to create custom patterns to replace the standard lines and rectangles that normally form the repertoire—or, more correctly, the character set—of the e-beam system. In their direct-write technology, the eBeam folks create a character set optimized to a particular cell library, so that they can, for example, write all the patterns for a complete flipflop cell onto a wafer in one shot.
But for maskmaking, the eBeam scheme uses not elaborate custom shapes, but a set of circles. It turns out, Samsung’s Kim explained, that a set of circles substantially lowers the writing time for the new curvilinear figures. A contact may be as simple as a single circle with a larger circular assist ring around it: two shots. A complex curved shape with smooth sides may be just a handful of overlapped circles. The result of using the circles, Kim said, is dramatically fewer shots per mask, and a chance of reaching Kalk’s ten-hour limit.
The eBeam initiative is still quite a ways away from announcing a product or service, according to D2S CEO Aki Fujimura. But the organization expects to have its technology up and running when the first production masks have to be made for the first 22nm processes. Part of this optimism, Fujimura explained, is that many of today’s advanced e-beam mask writers are already capable of using a circle character set.
Further into the future, Fujimura said, the character set for mask writing could evolve beyond just circles to include other curvilinear shapes. At that point, it might be feasible to talk about creating curve-friendly cell layouts for use in 22nm designs, just as today D2S creates cell libraries that are optimized for the character set of the direct-write e-beam systems. But Fujimura emphasized that the necessity for applying intensive OPC to cell layouts for use with masks made the evolution of optimized cell libraries much more complex that it is for the direct-write system, which requires very little optical correction. "This is a logical step," Fujimura said. "But I think it will be a while before we are able to take it."
Kim made another very interesting point about the future of the mask-writing technology. While circular figures help a lot with 193nm immersion masks for 22nm, they may also be essential for future eUV masks. Initial publicity for eUV suggested that because the wavelength of extreme ultraviolet radiation is shorter than the minimum feature size even at 22nm, no OPC would be necessary for eUV masks. But Fujimura points out that since that early optimism, effects such as flare and backscattering have appeared, both of which require manipulation of the mask patterns to correct for them. These manipulations at nodes below 22nm, where eUV is likely to first be used, can become as complex as the intricacies of OPC. So the eBeam Initiative’s circles may still be useful when—or perhaps we should say if—eUV is ready to take its place at the high-volume end of lithography.
In an additional announcement, Fujimura said that six new members have joined the initiative, at least partly in response to the 22nm mask work. The new members include foundries GlobalFoundries and Samsung, e-beam systems vendors JEOL and Nuflare, mask inspection giant KLA Tencor, and lithography technology specialist Petersen Advanced Lithography.