According to the official website of Science magazine, on July 24th, DARPA announced a total of US$75 million plan to reinvigorate the chip industry by upgrading basic materials including new materials and new designs including carbon nanotubes. Over the next five years, DARPA's project will grow to $300 million a year, totaling $1.5 billion, providing funding for academics and industry professionals.
In this regard, Erica Foss, a computer science policy expert at Carnegie Mellon University in the United States, said with joy: 'It is a crucial moment to make this step.'
Silicon chips are approaching physical limits
In 1965, Intel co-founder Gordon Moore suggested that the number of transistors that can be accommodated on a chip doubles every 18 months—this is what we know about Moore's Law.
In the following 30 years, the chip development has followed Moore's Law by reducing the size of components on the chip. However, in the 21st century, the practice of simply reducing the size has come to an end.
If the chip shrinks to 2 nanometers, then a single transistor will be only 10 atoms in size. The reliability of such a small transistor is likely to be problematic. As the connection of transistors becomes closer, another problem is highlighted. Power consumption will be getting bigger and bigger.
Max Surak, an electrical engineer at the Massachusetts Institute of Technology (MIT), said that in addition, the speed of the chip has been stagnant, and the energy efficiency of the next-generation chip can only be increased by 30%.
Gregory Wright, a wireless communications specialist at Nokia Bell Labs, points out that manufacturers are approaching the physical limits of silicon. Electronics are confined to silicon wafers of only 100 atoms wide, forcing scientists to adopt complex designs to prevent electron leakage. And it leads to mistakes, 'We don't have much room for improvement now, we need to find another way.'
Valeria Beataco, a computer scientist at the University of Michigan, Ann Arbor, says that only a few companies can afford a multi-billion dollar chip manufacturing facility that would kill a small startup-led area. Innovation.
Flowserve said that some big companies began to design dedicated chips for specific tasks, which greatly reduced their motivation to pay for basic research that can be shared. A study by Flowserve and colleagues pointed out that in 1996, 80 companies joined. The semiconductor research community in North Carolina, by 2013, was reduced by less than half.
New materials, new architecture is sought after
DARPA is working to fill this gap and provide funding for researchers including Surak. Surak is using 3D chips made of carbon nanotubes to make 3D chips faster and more efficient than silicon transistors. switch.
At present, many companies use silicon chips to make 3D chips, so that the logic and storage functions can be more closely combined to speed up the processing. However, because the lines for transferring information between chip layers are too large and scattered, this leads to The speed of the chip is slower. Moreover, since the two-dimensional silicon core layer must be fabricated separately at a high temperature of more than 1000 degrees Celsius, it is impossible to construct a 3D chip on the basis of the existing integrated manufacturing plan without melting the third layer.
Surak explains that carbon nanotube transistors can be fabricated at room temperature, providing a better way to integrate dense 3D chips. Although their team's 3D chips will be 10 times larger than the most advanced silicon devices, the chips The speed and energy efficiency are expected to increase by 50 times. This is a boon for data centers with huge power consumption.
In addition, the DARPA project supports research on flexible chip architectures.
Daniel Bliss, a wireless communications specialist at Arizona State University, and his colleagues hope to improve the effectiveness of wireless communications with chips that can be instantly reconfigured to perform specific tasks. Bliss is working to develop software and not hardware to mix and The radio chip that filters the signal, this advancement will enable more devices to send and receive signals without interference. He said, this can improve mobile and satellite communications, and accelerate the growth of the Internet of Things that allows countless devices to communicate with each other.
Another grant from DARPA will be given to researchers at Stanford University to improve computer tools used in chip manufacturing. These tools validate novel chip designs through artificial intelligence called machine learning. They will help detect Design flaws in chips made up of billions of transistors, most of which were previously done manually, and new tools help speed up the task and increase the company's ability to test and fabricate new chip architectures.
Stanford University's electrical and computer engineer, 3D Carbon Nanotubes and Circuit Verification Project researcher Sebashhi Mitra said that even if only a small number of new projects are successful, DARPA's latest funding program will revolutionize our design of electronic products. The way, 'he said, this will also prompt engineers to go beyond silicon that has been in the chip field for decades. 'It seems obvious now that silicon will follow a known path, but we clearly know that the future is not this. Look like ' .