What is the future electronics industry that DARPA thinks?

The US Defense Advanced Research Projects Agency (DARPA) promotes the Electronic Resurgence Initiative (ERI) program, which invests funds in technology development projects for next-generation electronics and micro systems that will transform the world from 2017. The budget amount reached 1.5 billion dollars (about 170 billion yen) at the time of article creation, and it has a policy of proactively investing for the development and realization of new technology. Mr. Bill · Chapel, who is responsible for ERI at DARPA, responds to the IEEE Spectrum interview on the three ERI's initiatives.

DARPA Plans a Major Remake of US Electronics - IEEE Spectrum
https://spectrum.ieee.org/tech-talk/computing/hardware/darpas-planning-a-major-remake-of-us-electronics-pay-attention.html

IEEE Spectrum (Q):
What are the challenges of the American electronics industry that triggered large-scale projects like ERI?

Mr. Chapel:
In modern times, integrated circuit products such as SoCs are becoming increasingly complicated, and the cost of designing and manufacturing is getting high. This is because as the product to be manufactured becomes more complicated, a large-scale design team is required, so that a great deal of labor cost is also reflected in the price of the product as a result. As a result, it not only affects the sales of the entire commercial but also the defense expenses are expensive.

Q:
What is the problem that the electronics industry can not solve this "cost increase"?

Mr. Chapel:
The industry itself is outstanding in problem solving skills and it does not mean that the problem can not be solved. However, the industry can not necessarily keep the best policy. So, we are forming a community that tackles big problems so that individual companies can tackle individual problems.

Q:
Although DARPA is doing three efforts with ERI, the first one is "structure of chip". Why is this?

Mr. Chapel:
Regarding the structure of the chip, I would like to create more specialized chips. To put it simply, the application itself is put on the chip. In fact, this field is a hot research theme in the field of machine learning and deep learning . However, our ideal is not just to apply to AI, but we would like to apply it to a wider range of fields.

Q:
Is it "wide field"? Is there something like a concrete example?

Mr. Chapel:
Examples include software-defined hardware (SDH), such as "hardware changes behavior according to software instructions" and "domain-specific SoC" that can provide high performance in response to processing in various fields there is. The ultimate goal we are thinking is "reconfigurable architecture". This is because the hardware flexibly changes the characteristics of the chip itself depending on the application such as "image processing" and "pattern matching", which is very difficult to realize.

As a matter of fact, if you know the required functions beforehand, you can create a chip specialized for a specific function. However, it is very difficult to extract high performance as dedicated chip in any field processing by using it for various purposes.

Q:
What is planned for "design" which is the second initiative?

Mr. Chapel:
The program we did the first time is " CRAFT ". This is a small design team with 2 to 3 people level so that you can design chips equivalent to sophisticated chips produced by 100 large team design teams. This program is actually done by the Chisel team of the University of California, Berkeley, and it can be realized.

Q:
What are you planning for ERI's third initiative, "Materials and integration"?

Mr. Chapel:
The program we are doing at DARPA is called CHIPS and, in short, it is to break down a very large design. Indeed, there are things in which all the functions are intertwined complicatedly and exists as one huge chip, but if we say that we divide this into small parts for each function, is it easy to understand? Although it is disadvantageous to divide it into small parts, there are obvious benefits as well.

This is because if the design team is small, it is possible to design chips with minimal resources by designing only the necessary parts and reusing parts designed in the past in the other parts . Of course, it is undeniable that performance will be lower than when operating as a huge chip because each part is modularized, but we want to minimize delays due to modularization as much as possible It is.

Q:
ERI is likely to affect not only the electronics industry but also engineers. Is this the goal?

Mr. Chapel:
Our goal is to change the electronics industry and gain great benefits for the US as well as the Department of Defense. I hope that our activities will make the electronics industry vibrant and become a more dynamic industry.