The Tunnel at the End of the Light: The Future of the U.S. Semiconductor Industry

Today, as it was 25 years ago, U.S. leadership in the semiconductor industry appears to be in peril, with increasingly robust competition from companies in Europe and Asia that are often subsidized by national governments. Twenty-five years ago, the United States responded vigorously to a Japanese challenge to its leadership. U.S. industry convinced the government, largely for national security reasons, to make investments that helped preserve and sustain U.S. leadership. The main mechanism for this turnaround was an unprecedented industry/government consortium called SEMATECH, which today has attained a near-mythical status.

The world has changed in the past 25 years, however. Today, industry is not clamoring for government help. In a more globalized economy, companies appear to be more concerned with their overall international position, rather than the relative strength of the U.S.-based segment. More over, the United States continues to lead the world in semi conductor R&D. Companies can use breakthroughs derived from that research to develop and manufacture new products anywhere in the world.

Indeed, it appears increasingly likely that most semiconductor manufacturing will no longer be done in the United States. But if this is the case, what are the implications for the U.S. economy? Are the national security concerns that fueled SEMATECH’s creation no longer relevant? Unfortunately, today’s policymakers are not even focusing on these questions. They should be.

We believe that there could be significant ramifications to the end of cutting-edge semiconductor manufacturing in the United States and that government involvement that goes beyond R&D funding may be necessary. But the U.S. government has traditionally been averse to policies supporting commercialization, and the current ideological makeup of Congress dictates against anything smacking of industrial policy.

But assuming that more government help is needed, and that Congress is even willing to provide it, what form should it take? In considering this question, we decided to reexamine the SEMATECH experience. We concluded that SEMATECH met the objectives of the U.S. semiconductor companies that established it but was only a partial answer to sustaining U.S. leadership in this critical technology. Moreover, as a consortium that received half of its funds over a decade from the U.S. Department of Defense (DOD) under the rationale of supporting national security, the SEMATECH experience raises some unaddressed policy questions as well as questions about how government should approach vexing issues about future technology leadership.

The origins of SEMATECH

In the late 1970s, U.S. semiconductor firms concluded that collectively they had a competitiveness problem. Japanese companies were aggressively targeting the dynamic random access memory (DRAM) business. U.S. companies believed that the Japanese firms were competing unfairly, aided by various government programs and subsidies. They contended that these arrangements allowed Japanese firms to develop and manufacture DRAMs and then dump them on the market at prices below cost. Initially, U.S. industry responded by forming the Semiconductor Industry Association as a forum for addressing key competitive issues.

In 1987, a Defense Science Board (DSB) Task Force issued a report articulating growing concerns about the competitiveness of the U.S. integrated circuit (IC) industry. The DSB study depicted semiconductor manufacturing as a national security problem and argued that the government should address it. A key recommendation was the creation of the entity that became SEMATECH.

The Reagan administration initially opposed an industry/ government consortium, considering it inappropriate industrial policy. But Congress, concerned with what it considered to be the real prospect that the United States would cede the IC manufacturing industry to Japan, approved a bill creating SEMATECH, and President Reagan signed it into law.

From the outset, there were some concerns about SEMATECH. One was the nature of a consortium itself, which is essentially a club with members who pay to join. SEMATECH was made up of about 80% of the leading U.S. semiconductor chip manufacturing firms. But some companies, for various reasons, declined to join, and were critical of SEMATECH for two reasons. First, SEMATECH was criticized for focusing on mainstream technology and thus defining the next generation of technology based on a limited view of the world. SEMATECH decided to focus on silicon complementary metal-oxide–semiconductor (CMOS) ICs. This technology is the basis for memory and other high-volume devices that were targets of Japanese competitors. Cypress Semiconductor, a chief critic that made application-specific integrated circuits (ASICs), believed that SEMATECH supported incumbent mass market technologies, not those of more specialized producers. Second, companies that had declined to participate in the consortium argued that because SEMATECH received half of its funding from the federal government, the results of its efforts should be equally available to all. But SEMATECH adamantly maintained its view that only those who had paid their fair share should reap preferential benefits.

Another concern about the creation of SEMATECH was the limited role given to the DOD despite the national security rationale for SEMATECH and the fact that the DOD would provide 50% of the funding for the consortium. In the enabling legislation for SEMATECH, Congress ensured that the DOD would have little direct input in the project planning and activities of the organization. Congress, following the position of SEMATECH’s commercial participants, concluded simply that industry knew best what to do and how to do it.

But from the start, it was clear that the interests of the government, and especially the DOD, were not the same as those of the commercial IC industry. This is made clear in an Institute for Defense Analyses (IDA) study done for the DOD on technology areas that needed attention from a defense perspective. One highlighted area was ASIC technology, because of the DOD’s need for affordable, low-volume specialty ICs. Although ASIC technology had great commercial potential, it was unlikely that SEMATECH would pursue that technology because of its business model.

The IDA study also emphasized the need to invest in manufacturing tools, especially lithography technology, for future generations of ICs. Although industry participants in SEMATECH did not object to government investment in advanced lithography, they saw this type of effort as separate from SEMATECH. However, this raised concern about how longer-term DOD-sponsored lithography R&D would integrate with the near-term SEMATECH focus on developing technology to improve processing yields.

A third area emphasized was the need to develop non-silicon IC technology, especially that based on materials such as gallium arsenide. These technologies are especially important to defense applications that require high-speed signal processing and had, in the view of some, great commercial potential. Indeed, gallium arsenide IC technology became the critical enabling technology for cellular phones.

Thus, it was clear early on that SEMATECH, with its narrow agenda and focus on the survival of its member companies, could not, from the DOD’s perspective, sufficiently serve national security interests in IC development. The DOD needed to develop new types of devices for future capabilities, the processes needed to fabricate them, and a U.S. industrial base with a first-mover advantage, so that the DOD could reap strategic and tactical advantages resulting from the development of the new technologies. Some saw the manufacturing of these other devices as vital to national security, perhaps even more so than the standard commercial CMOS ICs emphasized by SEMATECH.

This longer-term perspective led the DOD to sponsor research in areas SEMATECH was not emphasizing, including the Very High Speed IC program, which focused on analog-to-digital converter ICs. These were not standard CMOS ICs and required different fabrication processes. The DOD also funded research on non–silicon-based ICs, particularly those using gallium arsenide, through the Monolithic Microwave IC program. In addition, under the Defense Advanced Research Projects Agency (DARPA), the Very Large Scale Integration (VLSI) program supported research on advanced IC architecture, design tools, and manufacturing tools, especially lithography.

Diverging interests

The divergent interests of the government and SEMATECH came to the forefront in the1990s over the issue of lithography technology. Lithography processing tools produce intricate circuit design patterns on semiconductor wafers. Their continued improvement has enabled IC manufacturers to shrink feature size and pack ever-increasing numbers of transistors and functionality into a single chip.

Lithography tools are extremely complex and increasingly expensive. The current leading-edge tools cost more than $50 million, and next-generation tools are projected to cost nearly $125 million. Manufacturers, however, can make only one or two leading-edge tools per month. The highest profit margins for IC products come immediately after an advance occurs in lithography technology. Once the improved lithography tools become more widely available, the ICs they produce become commodity items with thin margins. Thus, the order in which IC manufacturers get access to the most advanced tools is an important component of their profitability. Tool suppliers use this as leverage to reward their largest and most loyal customers.

The DOD through DARPA and industry through SEMATECH supported the development of advanced lithography tools by two U.S. suppliers: GCA and PerkinElmer (later Silicon Valley Group, or SVG). These companies once dominated the global lithography market but were displaced by the Japanese firms Nikon and Canon. With federal and other external funding, the U.S.-developed tools became competitive with the tools offered by Nikon and Canon. The DOD wanted U.S. IC firms to buy and use the U.S. tools, thereby supporting a U.S. semiconductor infrastructure. The leading U.S. IC firms, however, were reluctant and made it increasingly clear that they wanted the U.S.-developed technology to be available to their Japanese lithography tool suppliers. In essence, key U.S. IC firms were happy to have this technology developed but wanted to continue to use Nikon and Canon as their suppliers.

This crystallized the divergent interests of the DOD and some of the major U.S. IC firms. Some in the DOD saw it in the nation’s interest for commercial and defense purposes to have U.S.-based lithography technology capabilities. The industry leaders, on the other hand, emphasized business concerns about the ability of the U.S. lithography firms to scale production and deliver and support the tools. Further, because they had established special relationships with Nikon or Canon, they had good early access to key tools, providing them with a competitive advantage.

From the government standpoint, this raised the following question: Why did companies encourage the government to fund these U.S. tools if they were not going to buy them? SEMATECH members paid for some of the tool development, which gave them the right of first refusal but no obligation to buy. Yet U.S. lithography firms would not survive unless major U.S. companies bought their tools.

As the U.S. lithography toolmakers foundered, the U.S. government now faced the question of what, if anything, it could do with the remnants of the U.S. industry. With government acquiescence, SVG acquired PerkinElmer’s lithography business in 1990 and formed Silicon Valley Group Lithography (SVGL). SVGL proceeded to develop Perkin Elmer’s breakthrough step-and-scan technology but still struggled to attract a customer base. In 1993, it talked with Canon about sharing the underlying technology, but the U.S. government objected to any such transfer. In 2000, ASML, a Netherlands-based lithography company, a joint venture of Phillips and ASM, announced its intent to acquire SVGL for $1.6 billion. After an initial objection by the U.S. Business and Industry Council, ASML completed the acquisition of SVGL in 2001, albeit with some very specific strictures to satisfy U.S. security concerns.

ASML, with strong support from the European Union (EU), industrial collaborations through the Belgium-based Interuniversity Microelectronics Center (IMEC), and the technology that it acquired from SVGL, developed a strong customer base among IC manufacturers in Europe, as well as those emerging in Korea and Taiwan that could not gain early access to leading-edge tools from the dominant Japanese providers. By addressing this underserved market, ASML in 2002 became the global leader, with 45% market share. Through a series of technical innovations that solved major problems for the IC manufacturers, it grew to 70% market share in 2011. Canon, meanwhile, lost most of its business.

Thus, it is arguable that U.S. industrial policy in the 1980s and 1990s, coupled with that of Europe, was successful. U.S. government–funded technology helped create a highly capable lithography competitor who offered an alternative supplier to then-dominant Japanese leaders, reducing the prospects of unacceptable concentration of a critical production tool. However, the firm that successfully implemented this technology was a Dutch one, which also received strong support from European firms and technology policies and programs.

Subsequently, another issue arose in the late 1990s between the U.S. government and U.S. IC firms over lithography. For years, IC manufacturers were concerned that optical lithography would reach its technological limits and the industry would no longer be able to continue shrinking features on IC chips. One promising, albeit extremely challenging, solution lay in the development of lithography based on extreme ultraviolet light (EUV), a technology that originated at U.S. Department of Energy (DOE) laboratories.

To access EUV technology, Intel in 1997 formed the EUV LLC, which entered into a cooperative R&D agreement (CRADA) with DOE. As part of this agreement, Intel and its partners would pay $250 million over three years to cover the direct salary costs of government researchers at the national labs and acquire equipment and materials for the labs, as well as cover the costs of its own researchers dedicated to the project. In return, the consortium would have exclusive rights to the technology in the EUV lithography field of use. At the time, it was the largest CRADA ever undertaken.

Once the EUV LLC executed the CRADA, Intel announced that it intended to bring in Nikon and ASML to help develop the technology. This unprecedented access by foreign corporations to U.S. national defense laboratories became an issue for DOE and Congress. In reviewing the available options, DOE rejected the Intel proposal to partner with Nikon but allowed it to set up a partnership with ASML. Among conditions for this access, ASML had to commit to use SVGL’s U.S. facilities for manufacturing.

Originally slated for production in early 2000s, EUV tools are still not available. Today, ASML is the only company in the world with preproduction EUV tools undergoing evaluation by IC manufacturers. Analysts predict that if ASML successfully delivers commercially viable EUV lithography tools, it will expand its global market share to 80%.

U.S. policy clearly had an effect on the semiconductor industry, but drawing lessons from this experience is not a simple matter. Tensions between commercial and national security goals were never fully resolved. Although some U.S. companies benefited from federal efforts, several foreign companies also reaped benefits, and the overall gains for U.S. interests were not as broad or as long-lasting as hoped. Now the nation’s IC industry faces a new and somewhat different challenge in a different global economic environment. Developing an effective policy response is a challenge that can be informed but not guided by previous efforts.

The government’s role

The DOD’s investments in IC R&D typically aim far ahead of the trajectory of commercial R&D. In that sense, the DOD’s R&D can be considered disruptive because it often leads to technologies that are not embedded in current practice. However, the last thing that the mainstream semiconductor industry wants is anything disruptive. The existing commercial IC industry follows a highly controlled evolutionary roadmap to maintain the Moore’s Law pace of doubling the number of transistors on an IC every 18 to 24 months. Only when industry hits a wall and can no longer proceed along an evolutionary path will it consider radical change.

Today, the industry again faces such a wall. The cost per transistor on an IC, which has been declining at an exponential rate for more than four decades, has leveled off, and continuing progress is simply not economically possible with mainstream technology. Industry is counting on ASML’s EUV lithography tools to restore the pattern of reliable cost reduction, but it is not certain that these tools will be able to meet the technical and economic requirements. There are both huge engineering and basic physics challenges to the development of new technologies. The lithography tool will have to register the successive exposure layers with an accuracy within four nanometers or about 20 atoms, and at this scale even the photon intensity of the light becomes a concern. We are literally reaching the tunnel at the end of the light.

DARPA has continued to fund R&D on next-generation lithography, albeit with an emphasis on technologies that support cost-effective fabrication of the low volumes of the specialty ICs needed by the military. It has funded technologies called nano-imprint lithography and reflected electron beam lithography. DARPA’s program, however, is limited to proving technical feasibility and does not address the investment needed to take the technology to a production-worthy tool.

Taking a tool from the technology feasibility demonstration stage to a production-worthy product is a bet-the-company proposition, similar to the stakes in developing a new commercial airliner. The penalty for missing the market can be devastating. In the past, U.S. lithography toolmakers had the best technology in the world, but failed to commercialize it. Today, the United States does not even have a firm that makes lithography tools.

The U.S. government has repeatedly invested in technology development but has deliberately avoided funding commercialization efforts. The U.S. aversion to policies supporting commercialization is a stark contrast to those of other countries. The EU, for example, through IMEC, Germany’s Fraunhofer Institute, and other institutions, supports applied research and product development. This has helped companies such as ASML navigate the treacherous waters of commercialization. Japan, Korea, Taiwan, China, and others also have government-funded applied technology programs.

As the world of electronics moves into the realm of nanoscale electronics, or nanotronics, the United States has focused on proof-of-concept technology R&D programs at DARPA, including some novel but unproven nanoscale production approaches, such as nanoimprint, dip-pen, and electron beam lithography. Yet today there is no concerted, focused national R&D program addressing the nanotronic manufacturing infrastructure. Corporate consortia and regional centers, such as the Albany Nanotech Complex in New York and the associated IBM-led Semiconductor Research Alliance, have sprung up without federal funding. However, such efforts are corporation-dominated and international in focus. They do not aim at a national agenda of technology leadership.

Without a parallel effort focused on developing the high-volume manufacturing technology and infrastructure necessary to make actual products, the United States will probably not reap the rewards of its investments. U.S.-based companies that develop the technology will have to go elsewhere to manufacture. The President’s Council of Advisors on Science and Technology 2010 report on the National Nanotechnology Initiative emphasized the need to put greater emphasis on manufacturing and commercialization in a nanoelectronics research initiative, stating that “over the next five years, the federal government should double the funding devoted to nanomanufacturing.”

The United States has a long history of funding the underlying technology and manufacturing capability in areas where national security is the primary application. The situation becomes less clear in dual-use situations, in which the technology has both commercial and defense importance. IC technology is clearly dual-use, given its pervasiveness in commercial products and its criticality in most aspects of military strategic and tactical operations.

The question is, given the national security implications of IC technology, should there be a concerted U.S. policy to address nanotronics manufacturing? If the health of the U.S. semiconductor manufacturing industry was so important for national security 25 years ago that it was the rationale for creating SEMATECH, should the United States be doing something similar today? If not, what has changed? If it undertakes such an activity, should it structure it in a way that gives the federal government an active role and voice?

The DOD has a strong interest in sustaining U.S. leadership in emerging technologies that will provide needed military capabilities. But the United States also needs to foster these technologies to maintain the health of an industry that has become a key component of a modern economy.

Who would fund this effort? Is this a job for the DOD, as it was in the past? It is certainly in the DOD’s interest to see key technologies mature and for the United States to be a leader in manufacturing. On the other hand, does this warrant a broader research funding agenda to include the Departments of Commerce and Energy? We contend that this is a much broader and more profound issue for national security, encompassing economic, energy, and even environmental security. Without the robust development of nanotronic-based industries, the United States faces the prospect of losing its leading position in the broader information technology sector, with cascading effects on other industries that depend on this technology to continue boosting their productivity.

Our concern is that there is inadequate focus on and discussion of these issues. The United States still has some tremendous advantages, including companies that are leaders in the most advanced IC technologies and their production, some robust tool and equipment firms, and a strong government-funded R&D system. However, other countries see the opportunity to claim the field and are funding national-level R&D programs in manufacturing at the nanoscale. Furthermore, the economic, geopolitical, and security landscape has changed fundamentally since 1985, which adds complexities to assessing the situation and determining potential approaches to address it. Given these dynamics, we conclude that it is time for concerted discussion to determine whether nanotronics manufacturing is an urgent national and economic security issue, and if so, what should be done about it.