An office that doesn't function like the rest of government - and it worksargue, for direction
It is tempting to call ARPA a crown jewel. Few organizations -- government or private -- can boast a budget of $2.8 billion and just 150 people. About half that number are program managers, usually Ph.D's, and the remainder work in administration.
The genius of ARPA's organization lies in its simplicity: take about 75 of the world's smartest technologists and give them money to spend on the technologies they consider critical to the national defense.
ARPA -- which expects 10 percent of its projects to fail -- is the quintessential virtual organization. It relies on outside contractors, communicates electronically, and works through fluid, ever-changing partnerships and alliances. And because ARPA has no huge, internal research and laboratory bureaucracy to feed, it can sometimes avoid bitter political battles over staff cuts or facilities. ARPA simply shuts down a program and lets its contractors adjust.
For instance, ARPA is now exiting the space business -- despite a proven and impressive record of success. Beginning in 1988, with annual budgets of $15 million to $30 million, ARPA funded development of the Pegasus and Taurus rockets, the key advance needed to spark the small satellite industry. That industry is now on the verge of exploding, creating communication services such as Motorola's Iridium project.
It's not the first time ARPA has played midwife. ARPA was created to break down bureaucratic barriers, particularly petty jealousies and turf wars between the services.
From the very start ARPA's mission has extended beyond defense -- what today would be called dual-use. NASA was originally part of ARPA until it was spun off in 1960. Through the 1960s ARPA focused increasingly on new sensor technology, energy R&D, radars and infrared sensing.
The advent of the microprocessor in 1969 diverted more of ARPA's attention toward information technology. The Internet, originally known as ARPAnet to help maintain computer communications in a nuclear holocaust, comes from ARPA research in the 1970s. And the agency funded most of the developments behind supercomputers and artificial intelligence.
Beginning in 1972, the agency had a "D" appended to its front to more explicitly define its role in defense. And it began developing prototypes of weapons systems, including the F-117A and many technologies later transferred to the Star Wars program. Among ARPA's finest achievements: funding Very Large Scale Integration technologies, which spawned a whole new generation of computer powerhouses, including Sun Microsystems and Silicon Graphics.
"ARPA was extremely farsighted in having a hands-off approach. That was the key to the success of a lot of projects," said Forest Baskett, chief technology officer at Silicon Graphics. Baskett was a principal investigator at Stanford University for ARPA's landmark VLSI program. The techniques used in that program helped automate the increasingly tedious process of circuit design.
The initiative is classic ARPA: The main manufacturers of semiconductors were loathe to embrace new design methods, as they already had millions of dollars invested in existing processes. So ARPA funded a program at Stanford to prove the viability of these methods. When the entrenched semiconductor companies declined to embrace the technologies, Stanford investigators struck out to form their own companies -- Sun Microsystems, Silicon Graphics, and MIPS (which later merged with SGI).
The project is classic ARPA in other ways, too. Baskett explains there are essentially two R&D funding models, the National Science Foundation model and ARPA. NSF states an area of research, solicits responses, and subjects responses to peer-review -- a conservative process by nature.
ARPA, on the other hand, tends to bypass peer review, relying on the judgment of its program managers, who have wide latitude in making awards. According to Baskett, this process encourages innovative development, though he admits both approaches are useful.
To these two models one might add the TRP and ATP model, which embrace the concept of cost-sharing. This approach is conservative in a way that is, perhaps, similar to the NSF model: scientific peer review tends to generate support for ideas embraced by the mainstream scientific community. Likewise, cost-sharing tends to encourage industry to participate in only those technologies viewed by the financial community as potential money-makers. Both approaches can be hostile to innovation.
And TRP also competes with ARPA's traditional approach to making awards. The agency has long been known as a nimble and innovative purchaser. When ARPA managers speak to industry of future spending plans, they rarely use the term "request-for-proposal," or RFP. Instead, ARPA issues BAA's, or Broad Agency Announcements. They allow agencies to solicit industry solutions to broadly defined problems; respondents propose their own solution and write a statement of work. The agency then has the latitude to fund one or more responses.
Though usually reserved for developing prototype applications of new technology, ARPA has received congressional approval to try this approach to developing full-fledged weapons systems. This authority to enter into "innovative agreements and transactions" frees ARPA from the constraints of the Federal Acquisition Regulation, a contracting bible that must be followed to the letter. "I'd get thrown into Leavenworth if I tried to buy things like ARPA," said one Air Force official.