Neuromorphic computing is gaining traction due to its inherently low power consumption and massive parallelism compared to traditional von Neumann architectures. Spiking neural networks (SNNs) exploit event-driven spikes and spatio-temporal information encoding, making them well suited for these platforms. However, efficiently mapping artificial neural networks (ANNs) onto tile-based SNN hardware remains challenging due to scalability constraints, limited bandwidth, and communication latency in underlying Network-on-Chip (NoC) architectures. In this work, we propose NeuroMap, a modular and reproducible toolchain that addresses these challenges by optimizing both the volume and distance of spike communication. NeuroMap leverages public libraries such as METIS for recursive 2-way graph partitioning and introduces a PageRank-guided refinement scheme to reduce hotspots and distribute communication more evenly across tiles. We further redesign tile microarchitectures with memristor-based crossbars to provide an energy-efficient and scalable hardware substrate for SNN execution. Experimental results show that NeuroMap consistently reduces hop count, global delay, and energy consumption across diverse SNN models and mesh configurations, outperforming prior approaches such as SNEAP while maintaining stable operation even on large and irregular networks.