Is chip R&D facing a dead end?

Intermolecular Inc. and Semiconductor Research Corp. (SRC) are expected to roll new R&D initiatives at the Semicon West show, with the former reporting on a memory R&D pilot line and the latter describing emerging programs in analog, energy, medical and multicore. The announcements come amid continued debate on whether semiconductor R&D is an endangered species or is successfully evolving to adapt to changing realities.

The IC industry still pours billions of dollars annually into R&D. But in recent years, as costs have skyrocketed, chipmakers have been outsourcing more of their R&D requirements—with implications for intellectual-property control—to consortia, silicon foundries and third-party vendors. Some observers fear the trends portend a distant scenario in which the vast majority of chipmakers cease R&D, relegating the once-mighty industry to a service sector populated by branding houses.

Meanwhile, some chipmakers that have outsourced their R&D are reportedly dissatisfied with the payback. And some critics are questioning whether emerging R&D collaboration models are workable.

Collaboration model
SRC is among the believers in the collaborative model. Founded in 1982, the consortium is spearheading several R&D initiatives. One is the Nanoelectronics Research Initiative, which has a 15-year goal of demonstrating novel devices with critical dimensions below 10nm.

SRC is also expanding its efforts in "applications research." For example, it has quietly signed a memorandum of understanding with the National Science Foundation to investigate multicore processors.

Furthermore, SRC is setting up an analog design center, which will explore analog and mixed-signal technologies across "a broad spectrum of applications," said president and CEO Larry Sumney. And a program still on the drawing board, the Topical Research Collaboration, will initially address energy and medical research through two separate projects.

Intermolecular road map
Intermolecular, too, is expanding its charter. Last year, it made a splash by rolling out the High-Productivity Combinatorial (HPC) platform, which provides several process steps in various cluster tools. HPC is said to facilitate R&D of IC materials, processes and device structures. The company can sell the cluster tools to chipmakers for R&D purposes. It also has set up an R&D line within its own facilities.

Now, Intermolecular has set up an R&D pilot line for use in developing a range of emerging memory technologies, such as phase-change memory and resistive RAM. The company is also mulling plans to develop similar technologies in the 3D chip and solar markets, said CEO David Lazovsky.

Business is booming at Intermolecular, which logged $55 million in bookings in 2007. One of its customers conducts no R&D in-house, having designated Intermolecular as its R&D arm. Companies like Intermolecular have "changed the landscape," Lazovsky said.

Few would argue the point, though some would call the shift troubling.

'Not dead'
Chip R&D "is not dead; it's still going on," said Dean Freeman, an analyst with Gartner Inc. But "as costs continue to escalate and R&D becomes more difficult, Gartner expects a greater percentage of R&D to be outsourced."

Some aspects of the in-house R&D model may simply no longer be viable. Many chipmakers still use high-volume manufacturing tools and fabs to conduct R&D, Lazovsky noted, adding that the industry can no longer afford to "use yesterday's methodologies for today's challenges."

The shifting environment has been a bonanza for R&D outsource entities. Along with Intermolecular and SRC, groups such as Albany Nanotech, IBM's technology alliance, Europe's Interuniversity Microelectronics Center, Japanese consortium Semiconductor Leading Edge Technologies, International Sematech and SVTC Technologies Inc. all report booming activity.

The R&D cost for a single semiconductor company to develop a 45-nm process is $1.5 billion, Gartner estimates.

Right partners
Chipmakers are scrambling to find the right R&D partners, since not even the likes of IBM and Intel Corp. can afford to go it alone in the current environment. In total, semiconductor R&D costs are expected to double from $50 billion in 2007 to $100 billion by 2012, according to Intermolecular.

The R&D cost for a single semiconductor company to develop a 45nm process is $1.5 billion, according to Gartner. The firm expects R&D costs to rise 15 to 25 percent at the 32nm node.

On average, the IC industry still spends 12 to 15 percent of total sales on R&D. But that figure is somewhat misleading, as vendors continue to outsource more pieces of the puzzle.

The acceleration of R&D outsourcing activity has been particularly acute for technologies that could advance Moore's Law, Freeman said. They include carbon nanotubes, heterogeneous ICs, high-k/metal-gate technology and nonplanar devices such as FinFETs.

One high-profile case of R&D outsourcing was Texas Instruments Inc.'s decision last year to cease its digital R&D efforts and hand the task to foundry partner Taiwan Semiconductor Manufacturing Co. Ltd. TI now focuses on in-house analog R&D.

Advanced Micro Devices, Freescale Semiconductor, Intel, NXP Semiconductors, Toshiba, Samsung, STMicroelectronics and many others are also relying more on outside sources for R&D.

Some pundits predict a radical consolidation of the IC industry. Freeman sees the industry boiling down to a handful of R&D/manufacturing clusters: IBM's technology alliance, Intel and TSMC. There may be room for three or four memory makers, he said. Analog is a different story; several companies could survive the storm, he believes.

IP concern
As R&D responsibility is shifted to other parties, some question whether IP control will follow. Outsourcing critics have hit the panic button as more IP has been funneled to Asia, especially China, given that country's poor track record for IP rights protection and its high aspirations for gaining technology ground against Europe, Japan and the United States.

But it's a moot point whether IP "is going offshore or not," argued SRC's Sumney, because "the semiconductor industry [itself] is global."

Until the 1980s, semiconductor R&D was driven by giants such as Bell Laboratories, General Electric, IBM and Xerox Parc. The golden era of semiconductor R&D ended some 20 years ago, when the U.S. government forced the breakup of AT&T, which owned Bell Labs. The research organization "could no longer do research for research's sake," and never fully recovered, Freeman said.

Around the same time, chipmakers, which had historically handled the bulk of their R&D efforts in-house, began to find the investment burdensome. Consortia were launched to enable collaborative R&D.

Other R&D models followed. IBM formed a technology alliance with chipmakers to offset R&D costs. Later, silicon foundries jumped into the fray.

Now, specialty R&D service providers such as Intermolecular, SVTC and a slew of niche players are entering the mix. Many promise the ability to bring technologies "from the lab to the fab."

R&D models under scrutiny
Each R&D model seems to have its advantages and its share of detractors. For example, said Freeman, chipmaking consortium Sematech was formed "for the common good," but "it didn't always work that way."

Alliances have had their problems as well. Crolles2, the high-profile joint R&D venture of Freescale, NXP, ST and TSMC, last year saw Freescale bail out of the venture and join IBM's technology alliance, while NXP, a founding member when it was part of Philips, left the group and aligned itself with TSMC.

Meanwhile, according to sources, IBM technology alliance member AMD has been privately grousing that it has seen insufficient payback on the soaring IP fees. An AMD spokesman, however, said those repeating the rumors were "barking up the wrong tree" and that there was no merit to the reports.

Others believe that only an elite group of companies can sustain R&D in a consortium or foundry. "Doing development in a foundry is cost prohibitive," said David Bergeron, CEO of R&D foundry vendor SVTC, which recently opened up a solar development center.

SRC's Sumney defended the consortium model but acknowledged that the consortia's "challenge is to balance the research needs" of their membership. For example, he said, as companies have gone fabless or fab-lite, their R&D focus has shifted from advancing Moore's Law to conducting "applications research" in a particular discipline.

- Mark LaPedus
EE Times