UPLOAD: advocatus diaboli

Eugene Leitl (Eugene.Leitl@lrz.uni-muenchen.de)
Mon, 6 Jan 1997 15:29:09 +0100 (MET)


10^11 neurons in toto
10^ 4 synapses/neuron on the average
100 ops/synapse/ms (vanilla computer ops, not important)
(1000 synapseEvents/s, 1 ms simulation step resolution)
10^20 ops/s summa summarum, human brain operational equivalent

10^ 8 Words/s (100 MW/s nonlocal memory bandwidth, 10 ns SRAM, optimistic)
10^12 current sequential PCs/human, or, 32 kYrs:1 sec timebase ratio

50 bit/synapse (weight, delay, connected-to-ID, just the basics)
0.5 Mbit/neuron, let's say 1 MBit/neuron (10^ 6 bit/neuron)
10^17 bits in toto, human brain representation storage equivalent

current bleeding edge:
10^ 9 bits/die (1 Gbit/die, current lab DRAM density)
10^ 8 dies human brain representation storage equivalent

0.5 g/die = 50 ton chips, weight equivalent
1 W/die = 100 MW power dissipation, power equivalent

future photolitho, 2d:
1 nm structures (_very_ optimistic, currently 200 nm tops, Si-Si bond
length is 0.24 nm)
10 Tbit/die (10^14 switches/die)
1000 "Omega" brand dies, future human brain representation storage equivalent

Facit:
Jit-availability of nanotech (here comes Dr. D. to save the daaay) is
uncertain at best, as no physical feasibility proof exists.

Anybody just said "cryonics"?

ciao,
'gene