Brain-like supercomputing platform to explore new frontiers
In the old days, it was common to hear a computer chip referred to as an "electronic brain." Modern chip designs are now making such labels even more apt. Lawrence Livermore National Laboratory (LLNL) is set to take receipt of a brain-inspired supercomputing platform developed by IBM Research. The first-of-a-kind system is based on a neurosynaptic computer chip known as IBM TrueNorth, and can process the equivalent of 16 million neurons and 4 billion synapses while consuming just 2.5 watts of power.
The TrueNorth system is billed as a fundamental departure from the way computers have been designed for over 70 years and uses digital neurons and synapses that processes information in a manner similar to that of the living brain – specifically, the right hemisphere of the human cerebral cortex. This isn't the first time that such a computer has been attempted, but according to IBM, TrueNorth is so advanced that it not only overcomes certain critical bottlenecks in conventional von-Neumann-architecture, but requires new ways of thinking to exploit the new hardware.
IBM says that the TrueNorth technology capable of creating computers operating at exascale speeds or a billion billion calculations per second. This is fifty times faster than current petaflop computers, yet is much smaller and uses much less power.
Originally developed by DARPA as part of its Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) program with the help of Cornell University, the computer being delivered to LLNL is made of 16 TrueNorth chips – each of which is made of 5.4 billion transistors that are put together to create one million digital neurons connected by 256 million electrical synapses. In contrast, there are 100 billion neurons in the human brain.
Despite carrying out 46 giga synaptic operations per second, IBM says that a TrueNorth processor uses only 70 milliwatts of power at 0.8 volts. The 16 chips working together are equivalent of 16 million neurons and four billion synapses while consuming little more electricity that a tablet computer.
Part of the reason for this performance is that by imitating the human brain, the TrueNorth neuromorphic processor overcomes some of the limitations of conventional von-Neumann-architecture. For example, program instructions and operation data can pass along the same route at the same time, which isn't possible in conventional processors. In addition, the TrueNorth processor doesn't need to be turned on all the time, but only when needed, which produces considerable power savings.
Another advantage, according to IBM, is that where standard computers focus on language and analytical thinking like the left-lobe of the human brain, TrueNorth is more like the right side with an emphasis on pattern recognition and integrated sensory processing as well as the ability to infer complex cognitive tasks.
The LLNL's TrueNorth system will part of the National Nuclear Security Administration's Advanced Simulation and Computing (ASC) program, where it will be used to study machine learning applications and deep learning algorithms and architectures, and to conduct general computing feasibility studies. The end game is to find ways to improve cyber security – especially in regard to protecting US nuclear weapons and ensuring their reliability without underground test explosions.
Under its contract with IBM Research, LLNL with receive the 16-chip TrueNorth system along with an "end-to-end ecosystem" to produce machines that can imitate the brain's capabilities for perception, action, and cognition. This will include a simulator; programming language; integrated programming environment; a library of algorithms, applications, and firmware; tools for making neural networks for deep learning; a teaching curriculum; and cloud enablement.
"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," says Jim Brase, LLNL deputy associate director for Data Science. "The potential capabilities neuromorphic computing represents and the machine intelligence that these will enable will change how we do science."
The video below outlines the LLNL TrueNorth project.