Considering the answers of this question, emulating a human brain with the current computing capacity is currently impossible, but we aren't very far from it.

Note, 1 or 2 decades ago, similar calculations had similar results.

The clock frequency of the modern CPUs seem to be stopped, currently the miniaturization (-> mobile use), the RAM/cache improvement and the multi-core paralellization are the main lines of the development.

Ok, but what is the case with the analogous chips? In case of a NN, it is not a very big problem, if it is not very accurate, the NN would adapt to the minor manufacturing differences in its learning phase. And a single analogous wire can substitute a complex integer multiplication-division unit, while the whole surface of the analogous printed circuit could work parallel.

According to this post, "software rewirable" analogous circuits, essentially "analogous FPGAs" already exist. Although the capacity of the FPGAs is highly below the capacity of the ASICs with the same size, maybe analogous chips for neural networks could also exist.

I suspect, if it is correct, maybe even the real human brain model wouldn't be too far. It would still require a massively parallel system of costly analogous NN chips, but it seems to me not impossible.

Could this idea work? Maybe there is even active research/development into this direction?

  • 1
    $\begingroup$ I appreciate that you found my answer useful, but you might consider un-accepting it at least briefly... It may help encourage others to answer and add other useful info. $\endgroup$
    – mindcrime
    Commented Sep 8, 2016 at 2:47

5 Answers 5


I'm not sure about "emulating the brain" per-se, but in a more general sense there has been some thought given to using analog computing for AI/ML. It seems clear that analog computers do have certain advantages over digital computers. For one, they can (depending on the application) be faster, albeit at the cost of some loss of precision. But that's OK, because I don't think anybody believes the human brain is calculating floating point math using digital computing techniques either. The human brain appears, at least superficially, to be largely probabilistic and able to tolerate some "slop" numerically.

The downside to analog computers, as I understand it, is that they're not as flexible... you basically hardwire a circuit to do one specific "thing" and that's really all it can do. To change the "programming" you have to literally solder in a new component! Or, I suppose, adjust a potentiometer or adjustable capacitor, etc. Anyway, the point is that digital computers are supremely flexible, which is one big reason they came to dominate the world. But I can see where there could be room for going analog for discrete functions that make up some or all of an intelligent system.

As for research in the area, you might look into whatever DARPA was / is doing. There was an article in Wired a while back, talking about some DARPA initiatives related to analog computing.


I am currently reading Superintelligence: Paths, Dangers, Strategies by Nick Bostrom. When he discusses whole brain emulation, although computing power (storage, bandwidth, CPU, body simulation, & environment simulation) is one of the three general key things we are lacking toward its success, he also seems to agree that computing power is the most feasible and attainable of the three general issues we have for attaining it as of now. However he also goes on to to say

Just how much technology is required for whole brain emulation depends on the level of abstraction at which the brain is simulated.ref

Which is an interesting thought, but a whole different discussion.

Anyways, so I think you are correct in thinking that we aren't far from having the computing power and maybe you are on to something, but rather are biggest hurdles are the other two key prerequisites that we need to attain before we can even begin trying, which are scanning, and translation.

Of the three, it would seem translation is the one we need to advance in the most, as of now. A modest prediction of attaining whole brain emulation is at least 15 years or mid century. Theres much more information in this book of all of the different paths that can be taken to achieve super intelligence, and it is well researched, I highly recommend it if you haven't read it already.


I see two main issues with this suggestion.

One: digital circuits take up a lot less space, and they're easier to design, so you can put together a bigger system this way. (not to mention connecting separate chips within a system) This is mainly because in digital circuits your tolerances can be a lot loose.

The bigger one is: we still don't know how neurons work. Artificial neural networks somewhat resemble the natural one, but they behave differently. There are various ion channels, there are electric signals, and with these neurons stimulate each other, and if one's threshold is reached, it fires a spike. When it's reached again soon, you can see a burst in the signal. As far as I know, researchers don't yet know what function you need to implement to simulate it. The closest ANN is the spiking neural network, but it's not very useful in practice.


If the universe is discrete, then analog phenomena (fluidity, curvature) are built on primitively discrete phenomena (bits and pieces).

If the universe is continuous, then discrete phenomena (bits and pieces) are built on primitively continuous phenomena (fluidity, curvature).

If the universe is discrete, the speed of seemingly analog phenomena will be bounded by the number of discrete phenomena that can occur in time and space.

If the universe is continuous, then time, space or matter may be infinitely divisible, which may allow for the execution of some phenomena faster than those phenomena appear to execute in natural environments (like protein folding or electric circuits) - so called "super Turing" computation.

The continuous universe idea begs the question, though: From whence came all this discreteness? A discrete universe can allow for apparent continuous behavior via approximation and randomness (or pseudorandomness), whereas a universe that is infinitely divisible affords no obvious definition of where things should start and end. This is one of the reasons many thinkers eschew considering infinities - they may be illusory.

So, can analog "circuits" execute faster than digital? As of right now, we know of some seemingly analog phenomena that appear execute faster than some digital phenomena (like electron spin vs a silicon logic gate). Whether analog phenomena are intrinsically more efficient than digital depends on the actual nature of the universe, which we have not yet determined.


my hunch (and this is strictly a hunch) is that building a human brain on a chip is actually alot easier than you might think.

my pet theory is that biological neurons are horribly slow, clumsy, and error-prone devices (at least mines are :lol:), but that the human brain overcomes this limitation by increasing the degree of parallelism several orders of magnitude over the current chip technology; and to that end it requires ~1.0e+11 neurons.

but the chip removes these limitations, and when the neurons have instantaneous relays, then you dont need nearly so many of them. if thats correct, then a human brain on a chip could probably run in only a few million neurons as opposed to 1.0e+11 inside the skull.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .