New 'neuromorphic' hardware brings robot brains closer to reality
Scientists may be one step closer to building a machine that many believe can trump any intelligent robot: a natural brain.
The BrainScaleS project, a collaboration between 15 research institutions, is creating hardware that emulates parts of a natural brain, tech site The Next Web reported.
“Our goal is to create a working system that will be located in Heidelberg, but accessible online to scientists all over the world,” said senior researcher Dr. Johannes Schemmel.
But The Next Web said one major drawback of neuromorphe hardware is that hardware is not as flexible as biology.
“Come new fundamental findings from biologists, we might just have to change our hardware from scratch,” Schemmel said.
Leading the collaboration is the Kirchhoff-Institute for Physics in Heidelberg, Germany.
The project aims to develop “neuromorphic hardware” or electronic systems that reconstruct the behavior of synapses, using electrical components such as transistors and microchips.
The Next Web said the team launched its first prototype: an eight-inch large wafer with 51 million artificial synapses.
Yet, the prototype represents only a tiny fraction of a working brain, although it can help examine the processing of natural nerve signals in time-lapse.
Scientists expect the hardware brain model will re-create neurological processes 10,000 times faster than they take place in a natural biological system, once the project ends in three years.
“That means that if we want to study a behavior (in the nervous system) that would take a few minutes in the real biology, it will only take us split seconds,” said Schemmel.
Earlier attempts at brain research had involved mostly computer simulation with a high-capacity computer system.
There have also been attempts to create a synthetic brain such as Henry Markram‘s Blue Brain Project.
After meeting an initial goal of simulating a rat neocortical column in 2006, Markram's next project is building a functional simulation of a human brain.
Others like physicist-turned-neuroscientist Sebastian Seung use connectomics, or the wiring of the human brain.
Seung’s team at the Massachussetts Institute of Technology is trying to identify and describe the connectome, or the connections between the brain’s some 100 million neurons.
But The Next Web said the problem with the software simulation is that it is hooked on the processing power.
There is no system available yet that could keep up with a biological nervous system.
“Modeling is essential for neuroscience. If we don’t have models to reconstruct the performance of neurones and synapses, we will never understand how the brain works. We can’t just stuff all this into equations, the behaviour of each individual cell is far too complex for that,” Schemmel said.
The Next Web also said hardware like this will allow the development of intelligent control systems, which will have a tremendous effect on robotics.
"Again, no Transformers here – but we’ll see systems that are increasingly powerful, adaptive and more resilient to errors," it said.
Brain's ability to learn
Schemmel said one of the most interesting aspects in neuroscience is the brain’s ability to learn.
Neurons grow through cell division and simultaneously exchange signals with their environment and adjust to their individual functions within the nervous system.
One of the principal tasks BrainScaleS need to follow is that it creates a platform for collaboration between neurologists, biologists, physicists and IT scholars to combine as much knowledge as possible and create that system that allows as much flexibility as possible.
“How well it all works out we’ll find out in the next few years. We’ll find out the gaps, learn from them and design the next generation (of neuromorphe hardware),” Schemmel said. — TJD, GMA News