News

IBM’s TrueNorth Chip is Close to the Human Brain

August 20, 2015 by Jennifer A. Diffley

The TrueNorth chip contains one million programmable neurons. 5.4B transistors, and 4,096 parallel and distributed cores.

When you're one of the world's oldest and largest computer companies and suddenly find your name losing cache and your stocks plummeting, what's your next move? For IBM, it means building a computer that mimics the human brain.

The TrueNorth chip is a milestone in many areas, but most notably in size and complexity: it contains one million programmable neurons, 5.4B transistors, and 4,096 parallel and distributed cores. A full breakdown of the complexity of the chip and its difference from standard architectures is here. In essence, it's less a progression in computational power than an entire reimagining of chips altogether, which is notable as most companies are racing to simply get smaller and faster with small tweaks to the underlying architecture and fabrication. 

The equivalent of a rodent brain built with 48 TrueNorth chips.

IBM's goal behind the chip is to build a "neuro-synaptic chip system with ten billion neurons and one hundred trillion synapses, all while consuming only one kilowatt of power and occupying less than two liters of volume." In essence, IBM is making its own products obsolete.

Another difference between traditional chips and TrueNorth is that, instead of being commanded to perform functions, TrueNorth has the ability to recreate the cognitive performance of a small rodent, which means it solves problems without being told to solve them, also surpassing the logic of most human teenagers.

IBM imagines the chip inside smartphones and other mobile devices, which would allow processing power usually aided by the internet to be performed locally on the phone itself, but it's likely that companies like Google would migrate to neuromorphic chips for their ability to process massive data and deep-learning algorithms. 

Is it the human brain? No, but it's a step in the right direction. 

4 Comments