I want to point out that people should be wary of this recent wave of press surrounding "neural" computing. A lot of this involves trying to emulate biophysical properties of neurons we observe in the brain, but it is not clear how these properties affect function or if they are even necessary for the types of computations we are interested in emulating in silicon.
That being said, memristors are a fascinating piece of technology and we need to make progress on all fronts if neurally-inspired computing is to become a reality. This latest development (capturing the nonlinear properties of sodium and potassium channels in a microelectronic device) is quite interesting, as these properties are crucial for reproducing the spiking behavior of real neurons. All I'm saying is take these results with a grain of salt, there is still a lot of work to do!
At least this time they're talking about something that can be made to fire like sort of like a neuron instead of yet another boring old perceptron that can't even change its own behaviour.
Even if perfect replication of a neuron electronically was possible, there are 86 billion neurons in an average human brain which is well over ten times the number of transistors present on modern chips.
Of course, functional electronic neuron models and scaling well before that point is likely to be immensely useful anyway.
Even if it took a hundred thousand transistors (/transistor-size devices) to emulate a neuron, that wouldn't actually be a problem. Compared to electricity-on-copper, the connections between neurons are slow. When operating on electricity, there are no problems whatsoever in making your brain consist of a million chips that fill a warehouse.
Indeed. So it might actually be possible to emulate a human brain in hardware rather sooner than we might have thought. Memristors are also memory devices, with access times on the order of modern RAM, and a small multiple of 86 gigabytes is not a huge amount of storage these days. So imagine a 1 billion "neuristor" processor backed by a few hundred GB of memristor storage and operating in a sort of time-shared fashion (load a "brain component" into the neuristor configuration, interface with the memory for a while, then move on to the next component and so forth).
A search on Wikipedia turned up this 1976 paper [1] (unfortunately, behind a paywall), which derived a memristor circuit emulating Hodgkin-Huxley dynamics, a good biophysical model of the membrane potential of a neuron. In particular, it can show the generation of action potentials.
The relevant part is on p. 210: "In particular, the potassium channel of the Hodgkin-Huxley model should be identified as a first-order time-invariant voltage-controlled memristive one-port and the sodium channel should be identified as a second-order time-invariant voltage-controlled memristive one-port."
That being said, memristors are a fascinating piece of technology and we need to make progress on all fronts if neurally-inspired computing is to become a reality. This latest development (capturing the nonlinear properties of sodium and potassium channels in a microelectronic device) is quite interesting, as these properties are crucial for reproducing the spiking behavior of real neurons. All I'm saying is take these results with a grain of salt, there is still a lot of work to do!