We propose a SNN modeling framework that incorporates synaptic heterogeneity, an essential property largely overlooked in previous studies, and demonstrate its computational advantages and generalizability.