I like the connection to biology. But it seems incredibly inefficient to store or transmit this representation. A normal 8 bit integer is blown up into 32 bytes. How would you implement anything that needs a cache?
Somebody correct me if I'm wrong, but the stream isn't an integer, it's a probability, or a number between 0 and 1. The point of it is that you shouldn't need to cache it, that would defeat the point of it being a stream.
Still, it seems like if you wanted reduced power, and could accept reduced precision, wouldn't you just do the same thing in binary by using only a few bits? Not even 8 bits, but like 3 or 4 bits?
We're not talking about replacing every aspect of a computer with streams. Just the ones that make sense. So, for example, once you're done processing the streams then you might store them as the calculated probabilities.
Let's say you have an image sensor where the 1s in each stream represent arrival of a photon at a particular location. Then do your image processing with the streams. Finally convert to a traditional image format for storage.
It would still be inefficient to transfer these streams from the sensor to the processor. You'd have to fab the processor on the same chip as the sensor, and even transferring the data on-chip from the pixels to the processing units might be inefficient.
