by James Montgomery, news editor
August 25, 2010 – Chip density is a growing problem: Packing, and electrically isolating, 400M transistors on an area the size of a pinhead. And this will eventually require stacking them vertically — e.g., DRAM memory 4F2buried wordline, NAND flash "skyscrapers," and logic FinFETs. For device architectures, that means taller and narrower structures, different types of surfaces and materials that need to be filled under and around (conventional CVD from the bottom/sides tends to pinch at the top), and within a required thermal budget and at a reasonable complexity and cost.
AMAT’s answer: a liquid flowable CVD, dubbed Eterna FCVD, that "fills anything" with up to 30:1 aspect ratios. It can fill (full and partial) 5nm geometries with "very little overburden," and also around and underneath things like 10nm overhangs, they claim. The other benefit of the new technology is elimination of carbon in film, which hampers transistor isolation and causes voltage shifts and leakage. Current spin-on dielectric technology (SOD) requires multiple removal steps (up to 20) to address this, and at anneal temperatures that become problematic when dealing with the abovementioned complex device structures. That translates into a 50% lower integrated cost vs. SOD, they said. McClintock later in the Q&A noted that cycle time in fab could improved by almost one day with this tool (mainly avoiding SOD-purification type steps), though that will vary by integration scheme and by fab/customer.
Listen to our podcast with Ajay Bhatnagar, global product marketing manager in dielectric gap fill, who explains in-depth how the technology differs from conventional CVD, and how it is being applied to new device architectures such as DRAM vertical transistors, NAND vertical bit stacks, and FinFETs.
AMAT is targeting a $400M market for this new business, but the market could easily double ($700M+) with the needs of new structures (e.g. 3D). And the company is banking that customers will find new applications for this FCVD technology: e.g., surface defect masking (where SOD films would be far too thick), planarizing with capping films (for small openings at the top of devices), possibly for some areas that require an ALD-like film (though probably not ALD itself, where films are too thin for this type of tool), or even gapfill for aluminum interconnects. The company claims to have five memory customers using the tool ("every major memory customer"), and three logic customers in the queue: one has the tool, one is awaiting shipment in 4Q, and another is in demos.
Other details about the tool, gleaned from execs during the presentation and the Q&A:
Chipmakers probably need a couple of these new tools per layer — a minimum of two layers at 2Xnm node, and as many as eight or more in forthcoming vertical structures, according to Bill McClintock, VP/GM of the company’s dielectric systems and CMP unit. Throughput is "moderate," comparable to the company’s HARP tools.
However, throughput is the flip side of the coin from cost, noted McClintock and Randhir Thakur, EVP/GM of AMAT’s silicon systems group, emphasizing SOD’s many more post-treatment steps. They reiterated the quality of the new film, which behaves almost as a thermal-like film. "A good-quality film really takes care of all the unnecessary integration related things that you have to cost-wise put in," Thakur said.
This tool is made with leading-edge nodes and needs in mind, but AMAT expects that the technology will find traction at trailing nodes as well — e.g. qualified at 28nm but then later at 4Xnm manufacturing, where it could also simplify gapfilling and could save several dollars per wafer.
AMAT execs expressed confidence in the exclusivity of this technology, which they say relies on a "unique" chamber design that allows the precursor to be introduced and deposited in a certain way. But they were less forthcoming when discussing the actual inner workings of the tool, except that much of the benefit is in the hardware which is unlike other CVD chambers. (Answering a question from the audience, they acknowledged that the precursor does in fact enter the chamber as a liquid, broken up in the top part of the chamber.)
They also noted that there is no separate licensing fee for the precursor (>30 precursors were screened to get the final one), which can be obtained separately, mostly sourced from one supplier. McClintock added that the precursor costs "more than 3× less" than what is used with SOD — reminding that cost should take into account how much chemical is used in multiple-step SOD, as well as what is not used (e.g. lost off the wafer edges and wasted).
Another question from the audience invoked similar flowable technologies, e.g. Trikon. Thakur noted earlier efforts ran into problems with hardware (how to bring the precursor into the system) and precursor quality (e.g. carbon-free).
And removing the carbon ends up improving electrical performance and yields. Thakur indicated isolation and defects that show up in second-order yield problems in chips, "some of those electrical issues are not there" with this flowable CVD approach.
Filling gaps with a liquid precursor which reacts during deposition to form the film isn’t necessarily a new concept, points out Gartner’s Dean Freeman — "Trikon did this a few years back with TEOS and hydrogen peroxide," he told SST, but "never really got any traction." AMAT execs were coy about discussing the technology’s secrets, but Freeman noted they mentioned plasma, which could indicate how it works: possibly pulsed during deposition to harden the layer and react the SiO2. Also interesting is the claim about removing carbon from the film — it’s unlikely that the precursor starts off as carbon-free since it goes down as a liquid, he said, which means the secret is getting the carbon out afterwards, e.g. during the reaction.
"If FCVD works as expected, it would likely build market share in the gap fill market," says Steve O’Rourke from Deutsche Bank, in a research note. The concept of flowable CVD is similar to Novellus’ Cu flow PVD technology, he points out. A $400M SAM won’t be significant near-term, but could position the company very well for the transition to 3D structures, he notes.
Update 8/27/2010: Ed Korczynski, former SST senior technical editor and current blogosphere denizen at Betasights, digs further into the Eterna FCVD details and discovers that the precursor is indeed carbon-free from the start, and that it "never sees the plasma." He also examines why previous efforts at flowable CVD didn’t catch on.