Why is ResNet needed?

Any idea, why in Neural Network, training error increases after some time as we go deeper and deeper?

Please be specific about which lecture / time you are referring to.