Practical Chip Design

As we follow the process curve from 65 nm to 45 and 32, the challenge of lithography at critical layers is intruding deeper and deeper into the SoC design process. What had been the otherwise-invisible added expense of OPC has become an avalanche of design rules, often rendering perfectly reasonable structures—from an electrical point of view—off limits. And the handwriting is on the wall: below 65 nm we will see more model-based lithography-correction tools, more restrictions on cell designers, and in many cases, highly restrictive design rules. The latter are so unpalatable to many designers that alternatives are already starting to show up.

One small group of players in the industry wants to avoid the whole problem of nanometer optical lithography by doing away with the optical part. Nano-imprint technology typically uses e-beam systems to create a mechanical template, not unlike the tools used to press CDs, but on a much finer scale. These folks make a template that is an exact 1:1 image of a resist pattern. They coat the wafer with a polymer, press the template into it, harden the polymer with either heat or UV light, and withdraw the template. Presto: a pattern in the polymer that can be used to mask portions of the wafer. Stepping and repeating creates a patterned polymer layer across an entire wafer.

In principle, the technique has a great future. It requires only inexpensive equipment, needss no DfM tools, OPC, phase-shifting patterns, or other adjustments to the mask to compensate for optical effects—what you see is actually what you get—and in the lab researchers have demonstrated forming features as small as 3 nm. But there have been disabling problems as well.

At SPIE this week we spoke with Mark Melliar-Smith, CEO Molecular Imprints, of one of these companies. His message is that not only can the problems with nano-imprint technology be addressed, they to a great extent have been, and his company at least is on the verge of shipping production tools. Melliar-Smith is confident that his technology will be a major factor in CMOS memory production, manufacturing of patterned media for disk drives, and some other selected areas by 2010.

To understand why, we first have to look at the problems with nano-imprint technology. These can be roughly grouped into a few categories: particle-related defects, irregularities in the polymer, overlay accuracy, and throughput. Let’s look at each in turn.

The first, and most stubborn, problem has been defects. Obviously, if you are going to spread a fluid on a wafer, stamp it with a tool, and set it, you are sensitive to template defects, and to particles getting into the process, either through the fluid itself, outside contamination, material build-up on the tool, or damage to the tool. This continues to be an issue, Melliar-Smith says, but it is an issue asymptotically approaching a solution. “Under the right circumstances today we can get below one defect per square centimeter,” he says. “We need to do better, but we are getting better.”

The stubbornness of the defect issue has channeled nano-imprint technology into applications that are inherently more tolerant of defects. At one time, experts were saying that we would have to evolve an entire style of high-redundancy logic design for the process to be useful. But looking at the current trend, Melliar-Smith says that imprinting is quickly going to be viable for critical layers on high-density CMOS memory chips, which already have redundancy structures built in. He is less optimistic about using nano-imprint technology for logic chips, where one defect is still disabling.

Another area already tolerant of a few defects is the emerging need to manufacture patterned media for disk drives. Not only can nano-imprint easily handle the pattern densities required, because the overlay accuracy needs are modest you can stamp an entire platter—both surfaces—at once. This leads to very good throughput compared to optical techniques, at a tiny fraction of the cost. Melliar-Smith says Molecular has already shipped five tools into the hard-drive media market.

The second big issue is irregularity. This, Melliar-Smith claims, has been to a great extent dealt with by Molecular’s development program. Unlike most others, this company puts the polymer onto the wafer as a low-viscosity liquid, delivered by ink-jet heads as 1-micrometer-high, less-than-10-picoliter, droplets. The print heads are driven from the design’s GDS-II files, so the density of the liquid matches the density of the pattern that will be stamped into it—eliminating problems with squishing a viscous goo around, waiting for it to ooze to the right places, and having channels in the template to permit oozing. Hence, better speed and uniformity.

Another irregularity problem occurs when the polymer sticks to the template as the latter pulls away from the wafer, tearing and leaving material on the template. This, Melliar-Smith says, has been addressed by a lot of surface-chemistry engineering, and by simply strengthening the UV-cured polymer.

Overlay accuracy is a matter of mechanical design. There is no way around the fact that an imprint machine has to repeatedly stamp the template down into the liquid across the face of a wafer, with position accuracy that is a fraction of a critical dimension. It’s hard, but with time and engineering it gets better.

Finally, there is the matter of throughput. The step-spray-stamp-flash-repeat process is slow, even compared to the exposure times required for modern high-NA immersion steppers. So wafer throughput—except in the disk media application—lags behind that for optical lithography. To mitigate this, Melliar-Smith talks about cost of ownership rather than throughput. The imprint systems cost a fraction of even today’s stepper prices. The templates Molecular uses are in effect ordinary photomasks, so there is no major cost difference there (and the infrastructure is in place.) A template is good for thousands of imprints, and there is a way to use a single master template to make multiple copy templates, further reducing the cost. Further, imprinting does not require double-patterning for a single layer, so one imprint step can replace a dozen or more exposures, etches, and intermediate process steps on critical layers. All of these factors lead Melliar-Smith to his prediction that on a cost-of-ownership basis Molecular Imprints should be an alternative by 2010.

To illustrate the flexibility of the process, another application Molecular is investigating is the formation of lenses on optical sensors. Today the lenses are generally formed by depositing a metered drop of polymer on each pixel, waiting for the fluid to slump into an optically-useful meniscus-like surface profile, and then hardening it. Since Molecular can impress virtually any surface shape on its liquid, and since the polymerized liquid is quite transparent to optical wavelengths, imprinting can form a wide variety of lens shapes, including square-perimetered lenses to fit the outline of the pixel, aspherical surfaces—whatever the optical folks want. While there are applications waiting in semiconductor memories and disks, the biggest pay-off for the technology may come in such a non-obvious area—perhaps something no one has even thought of yet.