What is the derivative of multiplyLayer in convolutional networks?

I am defining a class "multiplyLayer" which can multiply (element-wise) a list of inputs. The forward function of this layer is

function Z = predict( this, X )
            Z = X{1};
            for i = 2:this.NumInputs
                Z = Z.*X{i};
            end
end

X is a matrix of inputs, and NumInputs is the number of inputs. My question is what is the derivative (or backward function) of this layer?