Stefan IK

This piece of metal hides about 1/16th of a rodent’s brain computing capacity.

It’s name is TrueNorth and it’s a neuromorphic CMOS chip produced by IBM that consists of 4096 hardware cores, each one simulating 256 programmable silicon "neurons" for a total of just over a million neurons. That will allow researchers to run AI deep-learning algorithms with great efficiency on it, and the best part, due to the non-Von-Neumann architecture, it consumes ridiculously low amounts of power for the processing efficiency it provides - on the order of 1/10 000 that of a conventional processor, be it ARM or x86. That’s 0.07W compared to a typical ~120W - the chips are interlinkable, and with that density and power consumption, stacking them up is a matter of industrial availability more than anything else. Last time I checked, we were very good at making chips.

The Singularity might be closer than we thought. If that concept sounds alien to you, and you’re still thinking about how a brain works, this is a good place to start.

In the meantime, IBM is kicking off a boot camp to get researchers and programmers familiar with the architecture and start writing implementations.

In the words of Dharmendra Modha: 

“Humans use technology to transform society,” Modha says, pointing to the room of researchers. “These are the humans.”

This deserved a spot in my blog.

“The Clock” by Gislain Benoit