DARPA funds new electronics revolution
- By George Leopold
- Sep 14, 2017
Building on a half century of unprecedented chip scaling and technology innovation, DOD is going back to basics in an attempt to reseed a U.S. electronics sector that has contributed mightily to the nation's economic and national security.
The Defense Advanced Research Projects Agency put more skin in the game with a batch of program announcements worth $75 million over the next year as part of its Electronics Resurgence Initiative (ERI) launched in June. After gathering input from chipmakers and DOD users, DARPA program officials announced six new efforts this week focused on three key areas: semiconductor materials, architectures and next-generation chip design.
According to Bill Chappell, director of DARPA's Microsystems Technology Office, the goal, is to make "progress beyond just transistor scaling," a reference to Moore's Law, the paradigm for achieving ever-smaller, faster chips. The problem now is the soaring cost and complexity of chip design and manufacturing. "We are a victim of our own success" in scaling chips, Chappell noted.
Given the soaring cost and complexity involved in building new electronic devices (the price tag for a new chip fab is currently about $10 billion to build and equip), agency officials decided on an integrated approach in order to achieve its overarching goal of "inventing the ideas of the 2025-to-2030 electronics world," Chappell said.
To achieve that goal, it is spreading initial funding across industry partnerships in materials, design and architectures rather then putting all its funding in a single area. If partnerships between commercial firms and DOD suppliers pan out, DARPA said it is prepared to invest a total of $300 million on ERI over the next four years.
As with transformative technologies like the Internet, DARPA is thinking a generation ahead in order to create new technology paradigms. To do that, it will again fund high-risk, high-reward fundamental research that commercial chipmakers can't pursue due to its speculative nature while the military services must focus on immediate requirements.
"We are investing in some of the real basics of electronics that are often overlooked" as Moore's Law and chip scaling runs out of steam, Chappell added.
Initial DARPA funding includes a materials focus that will among other things seek new ways to integrate unconventional chip materials with conventional silicon circuits. From there, research will center on boosting performance associated with chip scaling. The agency has experience in this area, including funding research efforts into radiation-resistant materials such as gallium arsenide.
The materials effort also will tackle 3-D systems-on-chip (SoC) architectures designed to increase the size of semiconductor substrates. A separate effort will investigate new chip architectures that "transcend the conventional separation of logic and memory functions," the agency said.
As chip feature sizes shrink to as little as 7 nanometers, another ESI project will seek to reduce the time and complexity required to design current SoCs. A pair of chip design initiatives will look for new ways to automate the layout process while making greater use of open source design and verification frameworks.
Meanwhile, a Software Defined Hardware program will investigate new ways to design and manufacture reconfigurable hardware and software, with particular focus on big data applications. The resulting architectures would be able to run data-intensive algorithms underpinning future machine learning and autonomous systems. One goal is to develop reconfigurable systems that rival the performance of current application specific ICs.
Among the big data issues raised by the torrent of sensor and other data collected by ISR systems is the possibility of designing configurable chips that can handle both dense and sparse data sets. One line of inquiry, Chappell explained, is developing "course-grained programmability" in which a processor equipped with a "just-in-time" compiler could determine on the fly precisely what type of data set it is processing. No such capability currently exists, he added.
Agency officials acknowledged the effort to move beyond Moore's Law to the next generation of electronics technology would again require close cooperation among commercial chipmakers, DoD suppliers, the military services, university researchers and the national laboratories.
ERI is "unique because we are trying to find that line between commercial [viability] and DoD impact," Chappell noted.
The strategic importance of electronics was underscored this week when the Trump administration moved to block the acquisition of U.S. chipmaker Lattice Semiconductor by Chinese investors. The White House cited national security concerns in preventing the deal.