That is irrelevant because in all the neural nets that I'm aware of, the activation functions we are using are very simple non-linear functions. We only use fp

That is irrelevant because in all the neural nets that I'm aware of, the activation functions we are using are very simple non-linear functions. We only use fp multiplication because that' the way we add up the weights to activations, nature has analogue ways of doing that.