Spiking neural networks (SNNs), as the third generation of artificial neural networks, have gained significant research attention due to their biomimetic properties and low power consumption. However, for commonly computing tasks on static data, such as image classification, conversion-based SNNs typically require multiple simulation steps to produce the final output, which limits SNNs’ ability to effectively handle these tasks. To this end, we present an SNN construction method based on the proposed Highway Connection (HiwayCon) module, which transmits spikes from the previous layer directly to the next layer, enabling neurons in the latter layer to respond more quickly to input spikes. We also introduce residual membrane potential (RMP) neurons, which maintain a “residual” membrane potential above the firing threshold, achieving instant firing. Our proposed method is applicable to common spiking networks such as fully connected networks and convolutional networks. Experimental results demonstrate that HiwayCon improves both classification accuracy and computational efficiency, reducing the simulation time required for convergence, and enhancing the real-time performance of SNNs.