Thursday, August 7, 2014

Skynet Edges Closer

Originally Posted on  by hsauro

There is a fascinating article in science this week about the construction in hardware of a neural chip. This isn’t a new idea but scale and flexibility is novel. The chip is made by researchers from IBM and Cornell and can emulate 1,000,000 ‘neurons’. The developers claim that the approach is scalable, and efficient. Apparently chips can be cascaded to make even bigger networks.

I did a back of the envelope calculation to guestimate how many transistors it might take to emulate the brains from different organisms since we know how many transistors it took to make the chip.

This sentence was taken from Ars Technica: “The new processor, which the team is calling TrueNorth, takes a radically different approach. Its 5.4 billion transistors include over (1 million neurons) 4,000 individual cores” Assuming 100 billion neurons in a human brain that means it would take:

54 Trillion transistors to make a human brain, that is or roughly equivalent to 5,200 modern PCs, assuming 10 billion transistors per PC.

In other words not currently possible (In addition the fact that the chip only talks to 256 or neurons whereas the each brain neuron talks to about 7000). The IBM team would have to connect over 5000 chips to make a human brain, not that many it seems. Some other numbers:

Jelly Fish: ~ 4 Million transistors
Pond Snail Brain: ~ 60 Million transistors
Fruit Fly Brain: ~ 540 Million transistors
Honey Bee Brain: ~5.2 Billion transistors
Cockroach: ~5.3 Billion transistors 
Frog Brain: ~86 Billion transistors
Rat Brain: ~1 Trillion transistors
Human Brain: ~54 Trillion transistors

The single chip could just about simulate a Cockroach brain.

Brain data from … of_neurons1 and

No comments: