IBM 'cognitive chip' TrueNorth can capture video at 2,000 fps

"Neuromorphic computing" can get massive results at fraction of power, tests indicate

IBM's TrueNorth "cognitive chip", which is built to operation in a similar fashion to the human brain, has shown an ability to capture video at a far higher frame rate than traditional techology allows.

Samsung took the 4,096 cores, one million neurons and 256 million synapses of the TrueNorth chip and fed it through an IniLabs Dynamic Vision Sensor camera.

The result was a video capture of 2,000 frames per second (fps), perhaps giving an idea of what a disembodied human brain could do if wired up to a vision sensor.

The technology uses what is effectively a form of intelligent native compression that works by reporting changes only if and when they're necessary, as opposed to capturing and processing everything that's seen.

This "neuromorphic computing" means that the technology is free from the restrictions of conventional instruction-led processing, and can capture 2,000 fps rather than the more usual 120 of traditional cameras.

Eric Ryu, vice president at the Samsung Advanced Institute of Technology, explained that the chip's performance is based on efficiency, not higher technological grunt, as reported by Cnet.

"There is a huge gap between biology and modern silicon technology," said Ryu, explaining that TrueNorth consumes only around 300 milliwatts of power, around a tenth that of a mobile phone processor.

The human brain, meanwhile, can handle some cognitive tasks with 100 million times less power than the average computer.

Possible uses for cognitive computing-led video capture could include overhead surveillance work to monitor roads or busy pedestrianised areas as the system can easily differentiate many different instances of movement in crowded areas.

"If we think of today's von Neumann computers as akin to left-brain-fast, symbolic, number-crunching calculators, TrueNorth can be likened to the right-brain-slow, sensory, pattern recognising machines," said IBM cognitive computing group lead researcher Dharmendra Modha in a blog post.

"We envision augmenting our neurosynaptic cores with synaptic plasticity to create a new generation of field-adaptable neurosynaptic computers capable of online learning."