|
Arizona highways: ASM's ALD is more than just another three-letter acronym |
|
|
|
Feb 05, 2008 at 08:00 AM |
Even without Intel's bleeding-edge Fab 32 in Chandler making the world safe for hafnium and sub-45-nm chip processing, the greater Phoenix metro area would still be a hotbed of high tech, with the semiconductor and related micro-/nanoelectronics sectors leading the way and the photovoltaics industry catching up fast.
During my recent trip to the Valley of the Sun, not only did I get a serious intellectual download from the Flexible Electronics and Displays conference, I also caught up with several local infrastructure companies in the area.
Speaking of the Intel HKMG vanguard, one Phoenix-based outfit (or at least the North American unit of a Dutch company) has played a critical role in the leading chipmaker's high-k/metal-gate technological and production breakthrough: ASM America. Home of ASM International's transistor products group, the Arizona facility primarily focuses on the company's suite of epitaxial and atomic-layer deposition (ALD) tools and processes. Although Intel has never "officially" acknowledged ASM's role as its ALD supplier, the word leaked out awhile back. (Of course, Intel rarely acknowledges it even has suppliers, except when it gives out its annual awards.)
ASM's Bob Hollands smiled when the topic of Intel came up in our conversation. Although he was reticent to say the "I word" out loud, he did acknowledge the reports in the media and investment community and didn't exactly deny their veracity. But then, when a company has 80% market share of the installed base for HKMG ALD, as ASM claims to have, all one has to do is connect the dots, since Intel is far and away the leader in that front-end-of-line 45-nm process. I also mentioned that IEDM participants were pretty open about which company supplies Intel's ALD tools, which kept the smile on Bob's face.
The ASM product marketing director noted an approaching milestone for the company's Pulsar thermal ALCVD process module: 96 units are in the field, with that number soon expected to hit and surpass the century mark. Some 51 Polygon platforms have also been installed, most of which sport Pulsar chambers. He told me that most of the other chipmakers delving into HKMG own ASM's ALD equipment for both pilot and volume production, and that the equipment company boasts the largest ALD development organization, with facilities in Finland, Japan, and Korea, as well as at IMEC's 300-mm fab in Belgium and the Phoenix site.
You might get an argument from the likes of Aixtron, Aviza, or Applied Materials about whether ASM is the "real" ALD market leader, but no other company has the number of chambers and systems in the field or such an experienced and vertically integrated multidisciplinary team.
As Bob told me, although ASM has installed ALD equipment for fabrication of gate metal electrodes, DRAM capacitors, hard-disk/magnetic-media structures, FeRAM diffusion barriers, and other apps, the gate dielectric stack remains the most important focus area. ALD's critical role will only increase as chipmakers move to 32 nm and beyond half pitches and flash memory companies, the last adopters, start to bring up ALD-enabled processes to get those floating gate structures to cooperate.
A review of ASM's ALD approach revealed several key differentiating factors. The platform can be equipped with multiple chambers, which can either do various deposition processes in sequence or be dedicated to one process in multiple chambers. Since the process has a relatively---and inherently---low throughput, the flexibility to enhance productivity by devoting additional modules to a single process is helpful for certain users. As for the "speed" of deposition though, Bob said ASM boasts the fastest Al2O3 rate, at >230 angstroms per minute.
Because atomic-layer deposition puts down one conformal monolayer at a time (in other words, really thin films), the material composition can be tailored any way that a process geek wants, so the technology has quite a bit of scalability. In a sense, it provides "infinite process control" from an engineer's perspective. Bob believes hafnium-based gate stacks can be pushed down to 22-nm node, but beyond that, some new materials will come into play.
Another difference between ASM and other atomic depositionaries is that it is the only one with a process of record using a solid source among its multiple sources, which has correlated with excellent leakage performance for same or similar film thicknesses. ASM and ATMI are working to optimize the material delivery aspects for the HfCl4 precursor, one of several ongoing partnerships the equipment company has with other suppliers and consortia.
Since temperature control is critical for ALD, the company touts Pulsar's design, with cross-flow deposition in a hot-wall reactor, as a unique feature which keeps the heat dialed in low and facilitates the high deposition rate. The chamber's process flexibility, based on a thin, low-profile reactor design with the valves set close to the chamber, allows for quick switching of gases, good gas distribution, and the ability to change rapidly from pulse to purge mode.
Bob showed me several examples of compelling performance data for the Pulsar's hafnium-oxide process. Results reveal sturdy, production-worthy equipment and a stable process, including <0.5% wafer-to-wafer nonuniformity, <1.3% within-wafer nonuniformity, and <10 particle adders (>0.1 micron) in 3000-wafer-plus marathon fab tests. Tool-to-tool matching stats are also impressive, with different Pulsar modules showing statistical equivalency, even though they employed different reactor packs and source combinations, and were used in different facilities.
The impact of chamber changes on film thicknesses was also negligible, and pre- and post-PM monitor wafer tests indicated that defectivity remained in check after preventive maintenance. Still, Bob told me that the average time between chamber changes ranges from 4000 wafers to 20,000 wafers, depending on the application, which helps explain why each system is shipped with a spare chamber. The company continues to work hard on improving tool uptimes and general cost of ownership metrics, recognizing there is ample room for improvement.
What's next for ASM and its ALD portfolio? Bob shared some details on development work that finds a little zirconium goes a long way in enhancing the scaling potential of the hafnium-oxide cocktail, by increasing the k-value, benefiting reliability, and improving mobility. Lanthanum oxide is also under investigation as a capping layer for NMOS work function tuning, something ALD can control quite accurately. It has shown minimal electron mobility and equivalent oxide thickness impact, something likely to make device physicists' and process developers' hearts all a-flutter. The flash crowd is very interested in the possible use of lanthanum as a high-k dielectric material (along with a tantalum metal gate) for next-gen TANOS apps.
With HKMG just entering volume production in the logic houses, DRAM makers (at least those buying tools these days) clamoring for ALD-enabled capacitor processes, every major HDD manufacturer insisting on multiple ALD units in their facilities, and the first flash applications only a few years away from production, it's clear that atomic-layer deposition is no longer a promising process technology in need of a killer app, but a killer process with a growing list of apps in need of its unique molecular-level capabilities---and ASM America seems to be in a killer position to benefit from and exploit the growing ALD market.
|
|
