Activation function
Also: nonlinearity
A fixed elementwise function such as ReLU, sigmoid or tanh applied after each linear layer. Without it, any stack of linear layers collapses into a single linear map, so depth would add no expressive power.
Example
Two linear layers with no activation compute W₂(W₁x) = (W₂W₁)x, which is just one linear layer; inserting ReLU between them breaks that collapse.