In their quest to create a neural network-based supercomputing platform, scientists are looking to the human brain for inspiration. A team from California's Lawrence Livermore National Laboratory (LLNL) is tasked with creating a supercomputer that can watch over the safety, security and reliability of the United States nuclear arsenal and, along with IBM, is hoping to create the first-ever brain-inspired supercomputing technology for deep learning.
The team's neural network will be based on IBM's neurosynaptic TrueNorth computing chips, consisting of 5.4 billion transistors in an array of one million digital neurons, which are specially designed to help computers perform complex cognitive tasks such as sensory processing, giving them a leg up on conventional computer technology.
The unique supercomputer with be built from 16 TrueNorth chips and will contain the equivalent of 16 million neurons and four billion synapses, although its design will ensure minimal energy consumption - just 2.5 watts of power - despite its amazing capabilities. Furthermore, its extremely low clock rate of just one kilohertz is achieved thanks to its exclusive focus: it can only host neural network software.
"The low-power consumption of these brain-inspired processors reflects the industry's desire and a creative approach to reducing power consumption in all components for future systems as we set our sights on exascale computing," said Michel McCoy, LLNL's program director for Weapon Simulation and Computing.
The LLNL team hopes to design a program that can achieve a higher degree of computing power for the National Nuclear Security Administration's (NNSA) missions that involve cybersecurity, nuclear weapons control and the management of agreements regarding nuclear missiles around the world.
The team is currently working towards their goal using NNSA's Advanced Simulation and Computing (ASC) program, which is evaluating machine learning and deep learning algorithms to determine their value to the program.
"Neuromorphic computing opens very exciting new possibilities and is consistent with what we see as the future of the high-performance computing and simulation at the heart of our national security missions," said Jim Brase, LLNL's deputy associate director for data science.
With a deal worth $1 million from IBM, LLNL is currently testing whether their neural networks are useful for the standard needs of the lab and is hoping to eventually shift certain projects and problems from their standard computers onto their new brain-inspired chips.
