Touted as a revolutionary change in the manufacture of semiconductors, a silicon manufacturing process called laser-assisted direct imprint (LADI) could extend the lifetime of silicon microprocessors by producing feature sizes as small as 10 nm at considerably less cost than optical-lithography processes.
In conventional optical lithography, a laser illuminates a mask to project a pattern of light on a resist-coated silicon wafer. The resist protects various sections of the wafer during subsequent processing, allowing manufacturers to build complex solid-state structures by iterating the process.
LADI, developed by Stephen Chou of Princeton University (Princeton, NJ), can essentially be thought of as an embossing technique. In LADI, a quartz mold is placed on a section of silicon wafer, and the two are sandwiched between plates of quartz (see figure). Screws through the outer quartz plates hold the assembly together. A 20-ns, 1.6-Jcm-2 pulse from a 308-nm xenon-chloride excimer laser (Lambda Physik; Gottingen, Germany) is expanded to 6.25 mm2, then passed through the quartz plate and mold, converting approximately 300 nm of the silicon surface from solid semiconductor to molten metal. The molten silicon stays liquid for a few hundred nanoseconds, during which the silicon fills the etched voids of the quartz mold. Once the silicon returns to solid form, the mold is removed. The process can be iterated as necessary.
LADI starts with a quartz mask on a silicon wafer (a). Laser light passes through the mask (b) to melt the wafer surface (c), causing the mask to emboss a pattern (c). After the silicon solidifies (d), the two are separated (e) to yield clean ridges (bottom).
The quartz does not absorb the laser pulse because the band gap is larger than the photon energy. Light from a 633-nm helium-neon laser reflecting off the silicon wafer surface measured the lifecycle of the molten silicon; molten-silicon metal reflects roughly twice as much light as solid silicon.
Critics may point to problems in the 1970s with contact lithography, in which the mask was placed directly on the wafer and then illuminated. In that process, resist stuck to the mask and created defects on the wafer surface. Dust on the mask was also a problem, which was chalked up to early cleanroom designs.
"I would hope that resist technology has improved to the point that this wouldn't be much of an issue," says R. Fabian Pease, professor at Stanford University's Electrical Engineering Department (Stanford, CA). "I'm very excited about LADI and the other microcontact printing techniques under development around the world. I suspect that this process will find use in less complex structures where defects are less of an issue. If you're making hard-disk memory, you know that defects exist and there are ways to get around it. In other applications, its not so easy."
Another hurdle may lie in the making of the masks. Chou admits that producing them is time-consuming, while Pease adds that industry is concerned about going back to 1X masks compared to the 4X or 5X masks that optical lithography uses today. "The cushion from larger masks is a huge advantage in most people's minds, but I'm not so sure. It was a huge boon when we went from 1X to 4X, but those issues may not be as strong today." Pease points out that electron beam lithography is up to the task of making masks with 10-nm precision.
Although it is unknown how long quartz masks would survive, or how the imprinting process would work for multi-layer microchips, LADI holds great promise for quick nanostructure manufacture and cost savings. In addition, industry is not all that unfamiliar with the benefits of imprinting. The compact-disk industry uses a similar process to copy CDs at a cost of around $0.50 per 100 cm2 in less than a second, Pease says.