The Pentagon's R&D shop is looking to solve the impending slowdown in global computing power by working with industry to develop the next generation of computer chip technologies.
Since the onset of the microcomputer revolution in the 1960s, global economic and technological progress has been largely fueled by regular, exponential leaps in computing power. With each passing year, computer chips have halved in size while doubling their processing power.
This principle, famously known as Moore’s Law and named after Intel cofounder Gordon Moore, may be coming to an end in the near future. In 2016, Moore’s old company Intel predicted that by 2020 or 2021, computer chip manufacturers will have mostly exhausted their ability to reduce the size and increase the power of modern computer chips, potentially rendering Moore’s Law obsolete and slowing future technological progress to a crawl.
The Defense Advanced Research Projects Agency has been keenly aware of this impending global computing slowdown. On Sept. 13, the agency released a trio of procurements that commit $300 million in new funding over the next four years to address the problem head-on and develop new technologies to further sustain and extend the microprocessing revolution. DARPA’s Microsystems Technology Office Director Bill Chappell said the agency also expects that funding to be augmented in the future through cost-sharing agreements with industry partners.
"It’s incredibly important for our country to handle this inflection point," Chappell said. "You could argue that the basis of the economic success we enjoy in the electronic space and in national defense is very much tied to Moore’s Law."
The three procurements, collectively known as the Electronics Resurgence Initiative, are each designed to tackle two of the six different technological limitations that are contributing to the slowdown. A materials and integration component looks to add a third dimension to microchips to allow for more space while also developing "materials, components and algorithms" to deal with memory bottlenecks, where too much time and energy is being spent transporting data between storage and processors. A design component aims to dramatically reduce the complexity of electronics design and create an open source verification framework. Finally, an architectures component looks to better leverage specialized machine learning and artificial intelligence-directed algorithms while developing multi-application systems on single programmable framework.
Chappell said DARPA split up the procurements to better manage the individual aspects of the project. While the work is designed to collectively solve the larger problem of an impending slowdown of computer processing power, success with even one or two components would still represent an important milestone, with broad implications for the commercial and public technology sectors.
"They’re all related and overlapping, but each [of the breakthroughs] would have tremendous value on their own," Chappell said.
The solicitations all have response dates in November and December. If the acquisition timeline goes according to plan, DARPA expects to begin to the implementation phases in April or May 2018.
The agency plans to hold vendor information sessions in Arlington, Va., (Sept. 18-19, Sept. 22) and Mountain View, Calif., (Sept. 22) for different components of the project. Chappell said there has been intense interest overall from the commercial sector since the project was announced, and the agency will be bringing in consultants from Bell Labs, Intel Labs and other industry subject matter experts to help manage the initiative.
"We’re sensing a hunger [in the tech industry] for this. It’s really basic electronics work," said Chappell. "Oftentimes, we get caught up in solving today’s problems. This is a really good role for federal dollars: to do the type of work that otherwise wouldn’t be sponsored because it is so speculative."