Has the next phase of computer programming begun? On Thursday, IBM announced that its scientists had developed a breakthrough software ecosystem to program silicon chips whose architecture is inspired by the brain.
The new programming model takes a path that differs from traditional computer architecture and structured programming. As Dharmendra S. Modha, principal investigator and senior manager at IBM Research, said in a statement, "a new architecture necessitates a new programming paradigm." His team successfully demonstrated a building block of a brain-inspired chip architecture using neurosynaptic cores in August 2011.
IBM has said that these new, neurosynaptic chips could essentially act as the right brain to Watson's left brain, referring to the IBM-built computer system that defeated champions in a live version of the TV quiz show Jeopardy. Watson covers language and analytical thinking, while the new cognitive chips and programming language handle senses and pattern recognition.
Right Brain, Left Brain
The applications for a programming language and chip architecture with cognitive capabilities include sensory recognition in small, low-power devices. The research team has envisioned, for instance, low-powered, light-weight eyeglasses that could help the visually impaired recognize objects, or a thermometer whose sensors could detect smells and recognize the presence of bacteria.
The IBM researchers have now developed breakthroughs for every facet of the programming cycle, including design, development, debugging and deployment.
They have created a multi-threaded, massively parallel and highly scalable functional software simulator using a network of neurosynaptic cores. A neuron model has been created as a fundamental information processing unit with brain-like computation, and the company said a network of such artificial neurons can "sense, remember, and act upon a variety of spatio-temporal, multi-modal environmental stimuli."
Programs for such an architecture are based, at a high level, on reusable building blocks the researchers call "corelets." A corelet is a network of neurosynaptic cores that performs a base-level function, such as containing all of the individual cores required to process sound. But each corelet is treated as essentially a black box with only its external inputs and outputs exposed to other programmers. Corelets are then combined into more complex functionality, creating apps without the need to program individual cores.
A library with designs and implementations of large-scale algorithms and applications has also been developed, containing more than 150 corelets developed in less than a year. The researchers have also created a teaching curriculum with an end-to-and software environment.
Michael Facemire, an analyst with industry research firm Forrester, said that it is reassuring IBM "is still innovating and pushing the envelope" toward new modes of computing architecture, since we're still a long way away from matching the sensory and other capabilities of the human brain.
While such brain inspired computing models could create totally new use cases and their solutions for computing devices, they might also be able to better accommodate such challenges as analyzing the huge streams of Big Data from users and natural events that businesses and research organizations are sifting daily for insights.
IBM said that its long-term goal is to build a chip system containing 10 billion neurons and 100 trillion synapses, which would take up less than two liters of volume and draw 1 kilowatt of power.