Darpa: Chip Industry’s Best Bet
Pentagon Agency Retools for Next Revolution in Semiconductors
After it was created for exclusively military purposes in the Sputnik era, a small research and technology arm of the Pentagon spawned a technology in the 1970s that helped bring about the computer-chip revolution of the 1980s. Three decades on, that technology is reaching the end of its commercial life, and the agency is launching a new program that could reduce costs and increase efficiency in the chip industry.
The organization that helped give birth to those ultraviolet lasers is now known as Darpa, the acronym for the Defense Advanced Research Projects Agency. And these days, Darpa has started a new program, using electron-beam lithography, that has the potential to reduce costs and increase the speed of fabrication by a hundredfold over existing tools, in part because it eliminates a costly and time-consuming step in the current process.
For the Defense Department, it could mean the ability to produce relatively small numbers of specialized chips for its weapons and communications systems quickly and at significantly reduced costs. For the American computer industry, the Darpa program has the potential to reshape the semiconductor industry.
“High-resolution lithography manufacturing is dominated by foreign states at present,” Darpa’s director, Regina E. Dugan, told a House Armed Services subcommittee last month. Citing the Netherlands and Japan, she said, “The development of this unique tool at Darpa could restore the U.S. position in the supply-chain manufacturing of critical systems components for national defense and commercial applications.”
If innovation is the turning of ideas into marketable products, then Darpa and its predecessor, the Advanced Research Projects Agency, may be wellsprings of the modern age. From the birth of the Internet and the devices that access it (cellphones, laptops and, more recently, household appliances) to its work to improve prosthetics, the agency has helped to make advances more quickly than the commercial sector might have done on its own.
“It is not the place of dreamlike musings or fantasies, not a place for self-indulging in wishes and hopes,” Dugan said. “Darpa is a place of doing. A place where vision is paired with execution. Where new things are imagined and then turned to reality. Where doing is as powerful a force as thinking. And where it is commonplace for something once deemed impossible to become improbable and then inevitable.”
The agency’s focus is on improving the nation’s defense by sponsoring revolutionary, high-payoff research that bridges the gap between fundamental discoveries and their military use. But it achieves these ends by working closely with industry and academia to develop ideas into viable commercial technologies.
Darpa does this at a cost to taxpayers of about $3 billion a year and a payroll of 120 program managers, a number unchanged in the past two decades.
Created in 1958 as a reaction to the Soviet Union’s launch of the space race, the agency’s mandate was to anticipate technological advances by (and develop such surprises for) America’s Cold War adversaries. After a few years focused on space, missile defense and nuclear detection, Darpa’s work was transferred to the newly created National Aeronautics and Space Administration. By the 1970s, it was focused on direct energy programs, information processing and tactical technologies. It was during this period that Arpanet, the precursor to the Internet, was born at Darpa headquarters.
Making a Few, Making Them Well
In 1973, Congress limited Darpa’s focus to purely military technologies — which is what the agency had in mind a few years later when it developed the laser technology, called Exciplex, that has been a building block for today’s most advanced microelectronic devices, both military and civilian.
Exciplex lasers have been used to make almost every semiconductor integrated circuit, or chip, in the past 20 years. Specifically, they are used in deep-ultraviolet photolithography, which focuses intense beams of light on a thin film, or mask. That mask is patterned with transparent and opaque areas, which correspond to the pattern that will be etched into silicon to create the transistors, resistors and capacitors needed for an integrated circuit.
An integrated circuit is made by fabricating super-thin wafers from silicon. A layer of silicon dioxide on the surface serves as an insulating base and prevents oxidation. An outer layer is coated with a chemical called photoresist, which is then covered with the mask. Ultraviolet light focused on the clear parts of the mask makes an imprint on the photoresist. A process of etching, often using chemicals, forms the pattern, layer after layer, creating the circuitry, including connective wires. At the time of their discovery and subsequent application, these lasers were a thousand times more efficient and a thousand times more powerful than their predecessors.
These lasers have enabled circuit dimension miniaturization down to the size of 22 nanometers — or 22 billionths of a meter. By comparison, a human hair is about 100,000 nanometers thick, a virus is about 30 nanometers and a strand of DNA is 2.5 nanometers.
But as miniaturization continues, the mask used in the process has increased enormously in cost and the wavelengths of the lasers themselves tend to exceed the size of the circuits. The mask typically is about 2 centimeters by 2 centimeters and can be used for 100 circuits on each of about 1,000 wafers, according to Joseph Mangano, manager for the so-called Maskless Nanowriter program. The Nanowriter is an electron-beam lithography tool that divides the beam into a million “beamlets.” Formed by a unique reflection electron-beam process, they write directly on the silicon wafers.
Mangano said the process is painstaking and increasingly costly: about $3 million per maskset, each of which is made up of 30 to 40 masks.
Mangano said that commercial producers are able to amortize the costs over millions of circuits produced, but at the Defense Department, only a few thousand specialized chips are required, making the process exceedingly expensive.
“Due to challenging patterning requirements and complex circuit designs, costs of lithography tools and masks have become unaffordable for low-volume manufacture, i.e., military electronics or application specific integrated circuits,” according to a Darpa budget document.
The need to address the growing cost prompted the agency to develop a new tool with California-based KLA-Tencor that eliminates the need for masksets. (KLA-Tencor did not respond to interview requests.)
Mangano said this new program would enable the Defense Department to produce relatively few chips at a cost equivalent to those produced on a much larger scale. And whereas the Exciplex lasers of the past can go to widths no smaller than 22 nanometers, the electron beam is expected to go to about 10 nanometers.
“This new tool permits tight control over wafer writing (nanolithography) to ensure greater security and low-cost integrated circuit customization, which are critical to most major U.S. defense systems,” Dugan said.
In short, the military would have its own tools to produce its own chips in small lots. “Carefully, we can upgrade them, bring them to state-of-the-art design rules, and retrofit them into legacy systems,” Mangano said.
Further, the tools would permit rapid prototyping. While in the past a single wafer would have 10 costly masksets, “this tool can write 10 circuits independently,” he said.
KLA-Tencor and Darpa are splitting the cost of the development efforts. The agency received $32 million in fiscal 2010 for its half of the program. It requested about $26 million in fiscal 2011 and about $16 million for fiscal 2012. Darpa budget documents said the program “will address both the DoD’s need for affordable, high performance, low volume Integrated Circuits (ICs) and the commercial market’s need for highly customized, application-specific ICs.” Tools will be made available for testing at the end of 2013, according to Darpa documents.
Mangano said that should the commercial industry adopt such a tool, the cost savings would be significant, potentially leading to less expensive and more capable devices, such as computers, smart phones and other common commercial tools.
But revolutionary ideas don’t just drop like manna from heaven. Mangano said that industry comes to Darpa with ideas, and if agency managers express interest, they pull together formal proposals that are competitively evaluated and funded “if they are worthy.”
At a hearing last month Dugan told the House Armed Services Subcommittee on Emerging Threats and Capabilities that Darpa finances approximately three times as many small-business proposals as are required by law. “Our engagement with them is focused not only on getting them resources, but also increasing our speed to get them under contract because speed is so important to small businesses and as well simplifying our approach,” she said. “Our contracting time is down by over 20 percent at this point with small businesses.”
Contracts have been whittled from 50 pages to 10. Darpa has conducted two business summits over the past year, involving 200 companies, 70 percent of which were small businesses representing 30 states.
“So our engagement with the small-business community from Darpa is very robust, both in terms of the ideas that they bring and, as well, their performance in our portfolio,” Dugan said.
The chairman of the panel, Texas Republican Mac Thornberry, asked the director whether her agency works to supplement research under way in the commercial sector. “Program managers at Darpa are experts in their field,” she responded. “They are often very closely coupled with their colleagues and experts in private industry and in academia.”