DARPA rattles up a half century
It s hard to imagine where we would be without the US Defense Advanced Research Projects Agency (DARPA). Decades of funding haven t just helped to create a sizable military market for III-V chips: they have revolutionized manufacturing efficiencies and improved the performance of many compound semiconductor devices for commercial applications. These include the famous GaAs Microwave and Millimeter-Wave Monolithic Integrated Circuits (MIMIC) program, which led to multibillion dollar sales for RF Micro Devices and Skyworks Solutions.
DARPA, formerly known as ARPA, didn t support the compounds when it was formed in 1958. However, it wasn t long before this agency recognized the superior electronic properties of GaAs over silicon, which couldn t produce a good solid-state device for microwave frequencies. In the early 1960s DARPA funded Carver Meads development of the first GaAs FET at the California Institute of Technology, and by the end of the decade it was constructing its own Center for Materials Research (CMR) at Stanford University. This facility focused on the liquid-phase epitaxial growth of GaAs and the fabrication of microwave devices.
One of CMR s goals was to educate PhD candidates who would subsequently be involved in materials-related technology development issues facing government, industry and university labs. Some success came when two members of CMR – Al Joseph and Richard Eden – moved to Rockwell Science Center, where they continued to develop GaAs microwave devices. Internal funding for this project was under threat in the early 1970s, so they turned to DARPA for support and netted three years worth of funding in 1973.
Eden believes that DARPA s funding throughout the 1970s helped to spur the development of the GaAs IC through long-term investments in materials and fabrication. "If you had told me [then] what would have been the real winner for GaAs IC technology – the cell-phone power amplifier – it would have been the last thing that I d have guessed."
It wasn t long before the efforts on microwave devices were ditched at Rockwell in favor of ICs. Silicon-based efforts within the company were making circuits that could operate in the microwave region and required components employing logic functions. GaAs equivalents promised faster speeds, but the existing MESFETs were unsuitable for logic. "The standard deviation of pinch-off voltage was enormous, and if you wanted to make ICs from the devices, you would need a large voltage swing," recollected Eden.
Employing a radically different MESFET technology slashed the variations in pinch-off voltage. Out went the mesa-epitaxial process that provided device isolation by selective removal of active material, in came a simpler, cheaper ion-implantation approach. "We were able to drop the standard deviation of the pinch-off voltages by a factor of 20 or more, down to where they were comparable to silicon."
The team went back to DARPA and requested funding to develop large-scale integrated circuits with 1000 gates or more. DARPA s director, George Heilmeier, was interested, but was only willing to stump up the cash if the team could demonstrate a gate with a power dissipation of less than 1 mW.
In a bizarre twist, Eden and his colleague Bryant Welsh were taken off the project at that point and the responsibility was handed to more senior members of the microwave group. But this did not deter these two, who devoted their nights and weekends to Heilmeier s goal. They produced devices that were so small that they could fit into the saw kerf space between the die areas of the official GaAs IC mask set. "These were the only things that were measured," said Eden, "because we were able to get decent ring oscillator speeds out of devices with power dissipations of 0.5–1.0 mW per gate."
With Heilmeier s target reached, a $5.47 million effort followed, centering on an 8 × 8 bit parallel processor. Success came in September 1980, five months after the three-year deadline had expired.
Computing GaAs
DARPA s funding didn t just provide the military with cutting-edge technology. It also spawned two companies that built high-speed circuits for computational purposes – Gigabit Logic, which was co-founded by Eden in 1981, and Vitesse Semiconductor, which was started by Joseph in 1984.
Gigabit Logic also received help from DARPA. "When we started the company we had a desperate need for funding and validation from customers," explained Eden. DARPA and Westinghouse fulfilled this need by paying for studies on analog-to-digital converters and ICs.
This convinced venture capitalists to invest, but Gigabit was never a great commercial success. GaAs computer chips were deployed in the Cray 3, but this computer was never launched and when Cray s founder, Seymour Cray, lost his life in an automobile accident in 1996 the company lost interest in III-Vs. Vitesse fared better and it opened the world s first 6 inch GaAs wafer fab in 1998, but today it focuses on products for optical networks.
DARPA continued to fund GaAs development throughout the 1980s and 1990s, and between 1988 and 1995 it pumped $0.5 billion into the incredibly thorough MIMIC program. "The pull behind the program was the need to develop more affordable weapon-system front ends," explained Eliot Cohen, who was the manager for the vast majority of this program. He says that there was a general recognition that hybrid circuits – the common type of circuit for microwave and millimeter-wave frequencies at that time – were not doing their job. The effort was also expected to deliver key technologies for radar, communication and electronic countermeasure systems that would benefit all three military services.
Cohen says that there was none of the production capability that exists today when the MIMIC program began. "It was going to take a substantial amount of money to get to a point where you really had the capability to produce these devices affordably, in reasonable quantities and in high yield." Very little attention had been paid to statistical process control and the identification of yield inhibitors, while substantial improvements to material quality, packaging and computer-aided design were needed.
MMICs had been demonstrated before the program. However, none of the companies that made them could foresee the number of lucrative commercial applications that would result, which meant that they were unwilling to make the investments to develop the technology.
In fact, commercial applications were a prerequisite for financially viable production lines. "It was a win-win situation," said Cohen. "You needed the ICs for military applications, but commercial applications served a very useful purpose."
By the early 1990s, MMICs were starting to appear in neighborhood phones – cordless handsets with a limited transmission range. "Very quickly after that the prices dropped as chips became more available," said Cohen. "Then they started being used in wireless, which of course is the principal application today."
The other application that was very exciting at the time was collision-avoidance radar. Hughes Aircraft developed "Forewarn", a simple radar system to warn drivers when children walked in front of or behind school buses. Other applications also came to light as the program continued, such as commercial satellite communications.
The commercial markets created by the MIMIC program have left a wonderful legacy, but they should not obscure the significant gains to the military. "It really revolutionized the way people produce the front end of systems," said Cohen. "I don t think you ll find many modern systems that use hybrid technology anymore."
The MIMIC program s success resulted from its thoroughness. All of the issues relating to GaAs circuit manufacture were addressed, including substrate and epilayer quality, computer-aided design, process control, and chip design and testing.
"I emphasized early on that people had to use manufacturing techniques that would result in an understanding of what were the yield inhibitors, the critical parameters to be monitored and so on," explained Cohen. According to him, the other major contribution was the improvement in computer-aided design: "They started realizing that you could produce the same circuits in a much smaller area of GaAs." Shrinking these circuits slashed production costs for military and commercial applications.
The GaAs MIMIC program also helped to establish MBE as a viable production technique for high-volume chip manufacture. Growers were producing 3 inch material with very good quality by the end of the first phase of the program.
Chip design also flourished. GaAs MESFET gate lengths were reduced from 0.5 to 0.25 µm in the first phase, while the second phase delivered improvements in HEMTs and HBT technologies. Processes for these transistors were not orientated to any specific application, and the development of more accurate transistor and matching circuit models, alongside greatly improved computer-aided design tools, made it possible to produce MMICs in foundries for various applications.
From GaAs to GaN
The vast improvements in chip manufacturing have not only positioned GaAs as an incumbent technology within the military. They have also provided the foundations for future DARPA efforts, such as the program entitled "Wide bandgap semiconductors – RF applications", which started in mid 2002.
"Our program can t compete in terms of its size to the MIMIC program," admitted wide-bandgap program manager Mark Rosker. "But on the other hand, we ve had the benefit of an awful lot of infrastructure that didn t exist for the MIMIC program. Because we get the piggyback, it doesn t require the same amount of resources."
The motivation for developing GaN-based RFICs stems from the material s very high levels of power density, in terms of W/mm. "Improvements in going from GaAs to GaN are not factors of 50% or even factors of two, but an order of magnitude," said Rosker. This means that a smaller device can be used for the same RF power, which either improves efficiency or simplifies matching to other components, thanks to a wider frequency range.
Rosker believes that these advantages over GaAs will lead to the widespread deployment of GaN in various classes of radar, communication and imaging. "I can t imagine why anyone would use GaAs if GaN was available, reliable and cost effective." Penetrating commercial markets may not be so easy, however. GaN is a very promising material for base stations, but it is not clear whether it can unlock the stranglehold that GaAs has on the handset market.
The first 24 months of the wide-bandgap program focused on improving material quality. "We could have gone directly to making devices or MMICs, but we would never have solved some of the fundamental problems," said Rosker. When the program began, SiC growth was plagued with defects, and it was difficult to control GaN-on-SiC growth.
Since early 2005 the program has been focusing on devices with contributions coming from three teams. Raytheon is leading an effort to build an 8–12 GHz transmit/receive module, TriQuint is heading a team with the goal of making a 2–20 GHz wideband amplifier and Northrop Grumman Space Technologies is in charge of a project to build an amplifier operating at more than 40 GHz.
Rosker is unable to provide details regarding recent successes, but he says that all of the partners have delivered remarkable levels of performance against difficult metrics. Most of this effort has focused on improving device reliability. "When we started you could literally watch a part on the bench degrade before your eyes," explained Rosker. Accelerated projections of state-of-the-art devices are now showing mean times to failure of more than 105 hours.
The results that Rosker is getting are validating his views on GaN: "I think it will be a long time before a material really displaces GaN. I think that 20–30 years from now we will look at GaAs as a transitional material that led us to better materials, both in terms of bandgap and other things." And that is the march of DARPA – continual development of novel materials for military use, which can spawn lucrative application in the commercial world.
View pdf of article
