Structural Reformation of Large Language Model Neuron Encapsulation for Divergent Information Aggregation