Dharmendra Modha has something to share with the world, and it’s making a splash. Modha spoke today at the SNS FiRe conference in Park City about the creation of IBM’s DARPA-funded TrueNorth chip, a pattern recognition processor capable of vastly exceeding all previous processing capabilities while using a fraction of the power.
Mohda described his humble beginnings, working his way from 100 neuron models all the way through rats and monkeys to human scale. He noted that by modeling a processor after the human brain, we can mimic a social network of 10 billion neurons. A real time simulator of that magnitude would require 12 gigawatts of power, or the entire generation capacity of the nation of Singapore.
This is instructive. As our processors have gotten faster and hotter, we have reached the upper limits of speed and heat in processing. The brain, on the other hand, is extremely slow and cool while processing massive quantities of information. A bee brain, for example, is a supercomputer the size of a postage stamp with the power load of a hearing aid battery.
The architecture of brains, Mohda explained, is parallel, distributed, slow, and modular. If you connect cortices from any direction they can easily communicate. The TrueNorth chip operates more efficiently, therefore, in almost every way. As well as the above characteristics, it is far cooler than its predecessors. The end result? We can achieve pattern recognition using large quantities of censors while using orders of magnitude less power.
According to Mohda, “We are literally creating synaptic supercomputers, very very quickly.” The underlying concept is a sea-change in our understanding of processing. Until now, Mohda notes, we have been trying to put a square peg in the proverbial round hole in the processing world. Rather than picture small shifts in power and capacity, Mohda says, “We need to envision an entirely new ecosystem. Chips come and go, ecosystems endure.”
What are the applications? Imagine glasses for the visually impaired that help them navigate complex environments without any need for internet connections. Imagine a tumbleweed robot that can roll into a burning house and search for missing children. Imagine censors that can measure ocean temperature, turbidity and more to track pollution.
This, Mohda states, will transform “not only business, but society itself […]We see a whole new generation of chips, systems, programming languages, software ecosystems, algorithms, and applications; a completely different way to make the world work.”
Where does this put us now? Mohda, with a sly grin, noted that we are at a point with synaptic supercomputing best described by Winston Churchill. “This is not the end, it is not even the beggining of the end. But it is, perhaps, the end of the beginning.”
In short, Mohda says, “The brain remains, and will remain, the frontier of discovery for at least a century.”