Beyond Rate Coding: Surrogate Gradients Enable Spike Timing Learning in Spiking Neural Networks | Synapse