Researchers in the United States have received a $15 million prize from the National Science Foundation (NSF) to develop superconducting chips that are expected to be much faster and use much less power than the hardware the world is obsessed with. supports today for computing.
A team from the Viterbi School of Engineering at the University of Southern California is leading the effort, and it goes by the name DISCoVER, a rather fun acronym that stands for Design and Integration of Superconductive Computation for Ventures Beyond Exascale Realization.
As the name suggests, scientists are looking to use superconducting materials as an alternative to current semiconductors to develop new types of ultra-fast and highly energy-efficient integrated circuits that can enable sustainable, large-scale exascale computing.
Creating supercomputers capable of delivering more than one exaflop, or a quintillion floating-point operations per second, has been an important strategic goal for the United States and other countries, including China, because they can accelerate significantly critical research projects, ranging from drug development to climate change modeling.
However, there are concerns that current approaches to designing and manufacturing silicon processors are hitting a wall and no longer scaling enough to deliver post-exascale performance, hence the need for things like research into superconductors to hopefully overcome this technological barrier.
Margaret Martonosi, NSF’s deputy director for computer science and information science and engineering, said the $15 million award to DISCoVER, announced Friday, is intended to “support efforts that envision future materials for computer systems in a post-Moore’s Law era”.
DISCoVER’s work is to develop comprehensive hardware and software systems that “enable the design, optimization, and demonstration of novel superconducting devices, single-flow quantum logic circuits, and ultra-high performance, ultra-high efficiency superconducting systems. energy, approaching the theoretical limit of Energy Efficiency.”
Specifically, the team is focusing on chips using niobium-based Josephson junctions, devices made of two superconducting materials joined by a non-superconducting material. These devices can only operate at the very low temperature of 4.2 Kelvin (equal to -452 degrees Fahrenheit or -269 degrees Celsius), and they store logical values of zeros and ones” by creating or removing currents persistent in superconducting loops”.
The real cool thing about these superconducting loops is that they “exhibit zero resistance and therefore, don’t lose energy,” as DISCOVER puts it. This is one of the main reasons why superconducting chips have such staggering implications for energy efficiency in computing.
A long way to go
There is a tremendous amount of work to be done, so DISCOVER the sharing between the USC Viterbi team and researchers from several other universities, including Auburn University in Alabama, Cornell University in New York, Northeastern University in Boston, Northwestern University in Illinois, University of Rochester in New York and Yokohama National University in Japan.
The USC Viterbi team will focus on developing superconducting circuits and architectures for a wide range of applications, including general-purpose processors, neural network accelerators, and classical control systems for computers. quantum.
The other university researchers, meanwhile, will work in tandem with the USC Viterbi team to enable “the design and prototyping of a superconducting system of cryogenic computing cores,” which goes by the incredibly cool acronym SuperSoCC.
The SuperSoCC’s work will focus on new materials and devices, on-chip memory design, and interfaces that will allow the SuperSoCC to interact with room-temperature electronics.
DISCoVER believes SuperSOCC could be at least 100 times more power efficient than traditional chips made using silicon-based CMOS transistors while delivering the same level of performance as “state-of-the-art” multi-core semiconductor chips. of technology”. On the other hand, the team believes that SuperSoCC could provide performance at least 100 times faster at the same energy level as CMOS chips.
Notably, DISCOVER said these major gains are possible despite the energy cost of cryogenic cooling, which is required for SuperSOCC.
Ultimately, DISCoVER hopes to “empower a new generation of diverse engineers and entrepreneurs who will bring superconducting devices and circuits to the mainstream of high-performance computing.”
“With the fundamental limits of CMOS scaling in sight, now is the time for an expedition to explore emerging disruptive computing technologies,” said Massoud Pedram, a USC professor and green computing expert who leads DISCoVER. . ®