IBM provides the US government with a supercomputer inspired by the human brain

A technological breakthrough in the field of computer chip architecture paves the way for a new generation of cognitive systems

IBM will provide the Lawrence Livermore National Research Laboratory in California with a first-of-its-kind supercomputer platform that draws inspiration from the structure of the human brain and is designed for deep learning tasks. The system, developed by IBM's research laboratories, is based on a technological breakthrough in the form of a neuro-synaptic computer chip, named "IBM TrueNorth". The platform is scalable, and is able to handle a processing volume equivalent to that of 16 million neurons and four billion synaptic connections - while consuming electricity similar to that of a tablet computer: 2.5 watts of electricity per 16 TrueNorth chips. The structure of the chip, which draws inspiration from the structure of the human brain, allows complex cognitive tasks such as pattern recognition to be performed, and integrated processing of information of a sensory nature, such as vision or hearing, in a significantly more efficient manner than conventional computer chips.
The new system will be used to research new computing capabilities in the areas of activity of the Nuclear Security Agency (NNSA) in cyber security and deterrence to prevent the spread of nuclear weapons. As part of ASC, NNSA's Advanced Simulation Program, applications in the fields of machine learning, deep learning algorithms and appropriate architectures will be tested - as well as applications in more general computing fields. ASC forms the basis of the NNSA's central program for the security of the US nuclear systems and the readiness of the nuclear weapons array - without the use of underground atomic tests.
According to Jim Bryce, Deputy Director of the Lawrence Livermore Laboratories for Computer Science, "computing systems with a morphological structure similar to that of the nervous system present exciting new possibilities, which correspond to the way we see the future of high-performance computing and computer simulations. The potential capabilities presented by these systems, and the machine intelligence they will enable, will change the way scientific research is carried out."
The new technology constitutes a fundamental change in the way computers are designed, as far as they have been known to us in the last seventy years - and it may be a central component in the development of the next generation of supercomputers, which will be able to perform processing on an exascale scale: fifty times faster speed, or two orders of magnitude Size, compared to the fastest systems operating today, at petaflop rates: a quadrillion floating point operations per second. Like the human brain, neurosynaptic systems also require an extremely low current supply, and a significantly smaller volume.
According to Michael McCoy, director of the nuclear weapons simulation program at Lawrence Livermore Laboratories, "the low current consumption of the processors built inspired by the human brain reflects well the ambition in the computing industry to reduce power consumption in all components that will be integrated into future systems."
Dharmendra Modha, the chief scientist of the field of computing inspired by the human brain at IBM's research laboratories in Elmden, estimates that "the delivery of this advanced platform is a milestone, and marks the entry point into the new era of cognitive computing. We appreciate the relationship with the national laboratory, and in fact, even before the production of the new system, we conducted simulations of it on the Sequoia supercomputer, built by IBM, which is already operating in the laboratory. The collaboration will advance the field of computing systems in architecture inspired by the human brain, and will allow us to offer unprecedented capabilities in the future - while reducing the costs of purchase, operation and programming."
Each TrueNorth chip includes 5.4 million transistors, forming a network of one million digital neurons that communicate with each other through 256 million electronic synapses. Working at a voltage of 0.8 volts, the chip consumes 70 milliwatts to run real-time applications, providing 46 gigabytes of synaptic operations per second. This is a lower current consumption by several orders of magnitude compared to traditional computers running a similar neural network. TrueNorth chips were originally developed as part of the SyNAPSE program of the US Defense Advanced Research Projects Agency, DARPA, in collaboration with Cornell University.
As part of the agreement with IBM, Lawrence Livermore Laboratories will receive a computing system with 16 TrueNorth chips, equivalent to 16 million neurons and four billion synapses. The labs will also receive a comprehensive, end-to-end environment that assists in the construction and design of highly efficient machines that operate in a manner similar to the human brain's ability to perceive, understand and operate in a cognitive process. The ecological environment surrounding the new computer includes a simulator, a dedicated programming language, an integrated programming environment, a library of algorithms and applications, firmware, tools for building neural networks for deep learning applications, a training program and cloud work capabilities.