One Human Brain = 8 Global Internets (Total Data Rate)

It's a trope of science fiction movies. Somebody gets the content of their brain sucked out and transferred to some other brain over a wire. So how much data would that be?

http://arxiv.org/abs/1307.2196

This question occurred to me a while ago when I attended "Downloading the Brain",  a panel discussion hosted by the World Science Festival. Michel Maharbiz of UC Berkeley talked about the "Neural Dust"  his group is developing. (Apparently the paper with their first results is in the review process.) The "Dust" is a new way to monitor neurons, without needing a wire. A tiny CMOS silicon chip just a hundred microns or so is place on the neuron. On the chip is a piezoelectric resonator. An acoustic wave is used to power and interrogate the chip, and thus read out the electric fields in the neuron. It's very similar to the way that RFID chips are interrogated by radio waves to do things like locating books in a library or trucks on a highway.


So imagine you could use this neural dust, one mote per neuron, to "read out" a brain. How much data would this be?

Get out your envelope backs!

http://dx.doi.org/10.3389/neuro.11.002.2010

The cerebral cortex of a typical human brain has about 20 billion neurons.  Which is a lot of neurons, so forget about dusting them. But pretend you could. I think ten thousand bytes per second ought to be enough to capture the output of a neuron. A single neuron isn't that fast- it fires at most 200 times a second. But we don't really know how much information is in the precise timing of the pulses, or even if the magnitude of a pulse encodes additional information. (There are a HUGE number of things we don't know about how the brain works!)  So 200 trillion bytes per second, or 1.6 petabits per second should be roughly enough to transmit the complete state of a brain. (Not including the connection patterns. That's a whole 'nother envelope!)


How much is that?

It's about 1 optical fiber. The maximum conceivable bandwidth of a single-mode optical fiber is set by the frequency range where it's clear enough to transmit light without melting. That number is about 1 petabit (10^15)/s, depending on the transmission distance. (see this discussion on StackExchange). Bit rates of about 10% of that have been achieved in the laboratory using "Space Division Multiplexing" (see this news article or this keynote from a real expert), while the current generation of optical networking products use multiple channels of 100 Gigabit Ethernet to achieve as much as 10Tb/s on a fiber, about 1% of the theoretical limit. A petabit per second is a ways in the future, but so is our neuro-dust.) Even now, we could probably fit a brain dump on 16 of the laboratory fiber systems.

So we can imagine putting the bandwidth of a brain onto a cable we can hold in our hands.

But how much is THAT?

Cisco puts out a report every year estimating the total traffic on the internet.  This year, they're estimating that the total IP traffic in the world is 62,476 petabytes per month. That's about 190 terabits/second. So a brain readout would be about 8 times the internet's total data rate.

Right now, powering that much dust would be impractical. Currently a neural dust mote uses a half a milliwatt of power, which means 10 megawatts to read the whole brain. So it gets fried, but it was just watching Netflix.