Why the single layer perceptron has a linear activation function while the Multi Layer Perceptron has a non-linear activation function ? What is the potential of the Multi Layer Perceptron respect of the Single Layer perceptron ?
Your premise is wrong. A single-layer neural network (perceptron) can use a (nonlinear) activation function. Nothing prevents you from doing that.
It's common that the last layer of a neural network doesn't use any activation function, and instead is input into a softmax layer. If there's only one layer, that means that no activation function is used. But this is just a standard convention. There's no ironclad law that says you have to do it this way. You could have an activation function after a single-layer neural network, if you wanted.