November 24, 2008
Is the U.S. falling behind in chip R&D?
Much ink has been devoted to the decline of semiconductor research and development in the United States. Pessimists say Bell Labs has thrown in the towel and pioneers like TI are following the path blazed by offshore foundries like TSMC. Optimists counter that Intel is still the world leader in next-gen semiconductors, IBM remains king in semi patents, IM Flash Technologies is making strides, HP Labs' memristors could make semiconductor memory obsolete and U.S. universities and national labs are inventing game-changing chip technologies. They say research alliances between U.S. industry, labs and universities are filling the gaps.
Many EEs lament the loss of U.S. R&D leadership embodied by the announcement that Bell Labs will no longer perform semiconductor materials and devices research. Bell Labs led the world by pioneering not only transistors, but also many major breakthroughs in semiconductor materials and devices, including the MOSFET, the charge-coupled device, molecular beam epitaxy, electron beam lithography, photovoltaic cells, the carbon-dioxide laser, the quantum cascade laser, the optical router and the first single-chip 32-bit microprocessor.
 | | LASER SHOW: IBM's T.J. Watson Research Center recently demoed a characterization technology that could lead to commercially viable carbon nanotubes. |
Bell Labs, based in Murray Hill, N.J., where semiconductors were pioneered, is now the research arm of Paris-based Alcatel-Lucent. The Labs will now devote the time of its 1,000 researchers and its $2 billion budget to the development of near-term technologies for wireless, networking, optics and computer algorithms. A small group will continue doing long-term research into high-speed electronics and nanotechnology, but the groundbreaking semiconductor work—for which Bell Labs won six Nobel Prizes—has ended.
Bell Labs' abandonment of long-term semiconductor research could be viewed as a nail in the coffin of semiconductor R&D in the United States. U.S. semiconductor makers themselves, however, claim that changes in business models at Bell Labs are merely endemic to the semiconductor climate, where chip makers need to roll with the punches to survive in the global low-margin, high-volume markets.
"The demise of U.S.-based R&D and semiconductor business has been prematurely declared," said Gregg Bartlett, vice president of design technology at Freescale Semiconductor. "People need to understand that what matters now is not what mattered when Bell Labs was setting the direction for the world in technology. It doesn't mean that there is not just as much innovation happening. There are still centers in the United States—Intel, IBM and the technology partnerships that IBM has—where [R&D] is a very core focus and there's a tremendous amount of innovation still happening in those."
Just saying so, however, will not convince the naysayers, who trace the seeds of change back to the memory business, in which almost every major U.S. chip maker participated at one time. Today, the only remaining U.S. maker of DRAM is Micron Technology—and it just reported a $1.6 billion loss for 2008. U.S. semiconductor fabs were next to suffer the same fate, with manufacturing moving offshore to the foundries of Taiwan Semiconductor Manufacturing Co. (TSMC), Chartered Semiconductor Manufacturing, United Microelectronics Corp. (UMC) and lately to foundries in mainland China. Pessimists point to these trends and predict that semiconductor R&D will be the next to go. Others, however, say U.S. semiconductor R&D is down, but not out.
"Some have chosen [to] get out of semiconductor R&D and change their business model, but there is still tremendous semiconductor technology leadership here in the United States, both through the IBM Alliance and at Intel—obviously two key powerhouses for advanced semiconductor research and development," said Gary Patton, vice president for IBM's Semiconductor Research and Development Center.
IBM has an R&D budget of $6.2 billion, with more than 3,200 engineers and scientists supported by "more than 200,000 technical roles complementary to core research," according to Patton. Intel's R&D budget is similar, exceeding $6 billion, and its corporate technology group—Intel Research—includes more than 1,000 engineers and scientists, as well as thousands of technicians and support staff.
IBM and Intel both claim that U.S. semiconductor R&D is alive and well despite both the losses from commodity businesses, like DRAMs, and the United States' increasing dependence on offshore foundries. After all, Intel was the world's first chip maker to scale down to 45-nm high-k metal gate designs, proving that the U.S. system still works.
 | | FAMOUS FIRST: Intel developed its initial silicon laser chip with the University of California, Santa Barbara. |
"You can judge our health by the steady pace of the technology," said Justin Rattner, Intel chief technology officer. "We were able to make the transition to high-k metal gates literally without skipping a beat."
IBM, likewise, claims it has moved to 45 nm on schedule for its high-end server microprocessors, and it plans to make the transition at its other fabs in short order. However, to weather the high cost of deep submicron development, IBM has had to form an alliance, consisting of joint-development partners Advanced Micro Devices, Chartered, Freescale, Infineon, Samsung and STMicroelectronics.
What has changed is not the roadmap but the economic realities of chip manufacturing. As more semiconductor wafer processing moves offshore, more development responsibilities move along with it. Today, starting a new wafer fab can cost upward of $3 billion; to meet those types of investment realities, semiconductor makers like Texas Instruments (TI) have begun relying on their foundry partners to develop the processes that will allow scaling beyond 32 nm.
"Honestly, much of the process technology that has moved over to Taiwan has not been as a consequence of not making the investments, but more the economic realities of manufacturing—as goes the volume production, so goes where the technology should be developed and deployed," said Freescale's Bartlett.
In a sense, U.S. semiconductor R&D is a victim of its own success. Today, financial health depends on scaling down to fit more chips per wafer, thereby lowering cost per die. Fabs are reluctant to introduce new materials or devices when, with tweaking, existing processes can be made to work at the smaller scales. And because offshore foundries have increasingly performed wafer fabrication, that tweaking has fallen to them.
TI acknowledges it is no longer a vertically integrated company. "Over time we shifted our investments closer to the customers, so what we focus on is less the nuts and bolts of process technology that goes underneath the chip, and more toward defining what will solve the customers' problems of tomorrow, making sure it solves those problems in the most cost-effective and power-efficient manner," said Martin Izzard, a vice president in TI's R&D Center.
Instead of racing down the semiconductor roadmap with the 1,288 other International Technology Roadmap for Semiconductors members, TI plans to invest its $2.2 billion R&D budget into designing better analog, mixed-signal and application-specific signal-processing chips. It's unclear what impact this strategy will have in the long run. But one thing is clear: To develop new materials and device structures—the kind of innovations needed to enable scaling in the future—will no longer be TI's problem, but that of its foundries.
In contrast, companies like IBM, Intel and Micron—that are not willing to go fabless—are left holding the bag.
"We are entering a time where all companies are producing technologies that are closer to the fundamental physical limits of the photo processes and the physics processes, so the challenges are steeper than ever," said Brian Shirley, Micron vice president of memory operations. "Now, we are looking for some of those challenges to be solved out of our own budgets in a 3- to 5-year horizon. Those are phenomenally expensive developments and the industry pressure is greater than ever."
This gap between research and development will tax U.S. semiconductor makers more as designs scale toward physical limits. As the gap widens in the U.S., and industry R&D budgets are squeezed by lower margins and oversupply, U.S. universities and national laboratories will have to shoulder more responsibility for long-term research.
Indeed, even IBM sees the writing on the wall. Displaced research scientists have been seeding universities nationwide for several years now, prompting IBM, Intel and others to form strong partnerships with universities. One of the best examples of this is IBM's recent announcement of the world's smallest SRAM—a 22-nm device fabricated at its Albany Nanotech research center, a cooperative R&D effort among the IBM Alliance, the University at Albany-SUNY and New York State.
IBM runs several well-financed programs with universities in New York. It funds a partnership with Rensselaer Polytechnic Institute on advanced 3D and other packaging techniques. IBM also has two partnerships with the University at Albany—one includes a 300-mm fab capable of 15-nm lithography; the other, to which it has donated an IBM Blue Gene supercomputer, focuses on computation methods to harness grid computing.
According to Intel's Rattner, "We are big sponsors of academic research, that's where the bulk of the really long-range work is going on—spintronics, carbon nanotubes, graphene. As those innovations start to move into an 8- to 10-year horizon, then we start to bring that work in-house and develop the most promising [ones] to make them a part of our production."
Perhaps the best example of industry collaborating with academia to perform the kind of long-term research that will keep the United States at the forefront of semiconductor technology is Semiconductor Research Corp. (SRC), which sponsors more than 100 research programs aimed at coordinating university and industry lab efforts.
But private funding of university research is dwarfed by the government programs from the Air Force Office of Scientific Research, the Defense Advanced Research Projects Agency (Darpa), the National Institute of Standards and Technology, the Office of Naval Research, and the Department of Energy's 13 national laboratories and eight tech centers. The single largest source of non-defense government funding to academia is the National Science Foundation, which has a $6 billion budget and accounts for approximately 20 percent of all federally supported research conducted at U.S. universities.
As a result of increasing financial pressures, funding has been cut to the longer-term research programs that don't promise immediate returns. Basic "electronics" research funding declined from $254 million in 2005 to $236 million in 2007 and dropped in 2008 to $197 million, according to Darpa.
Private and government funding is still available, but whether it is enough to fill the shoes of Bell Labs and keep the United States ahead in semiconductor R&D is unknown.
"I don't think we need new programs—we just need to restore funding to the programs we already have in place," said Intel's Rattner. "We have the mechanisms, we know they work—we just need to put the people and the money behind them to continue to make them successful."
|
