Is there any way to learn a neural network based on a cost function that makes use of a root finding algorithm?

Question

I need to integrate a root finding algorithm in a neural network. For that, and in order to be able to perform backpropgation I need the algortihm to be differentiable. Is t