This paper introduces Semantic Token Bus Reduction (STBR), demonstrating that the CPU-RAM-GPU bandwidth bottleneck is semantic rather than physical. STBR replaces raw data transmission with semantic token references resolved locally, reducing interconnect load from O(|D|) to O(1) under high coherence. Introduces Semantic Compression Theory, the Semantic Coherence Function, and the Token Instruction Set Architecture (Token ISA).
Elizeu Almeida Prudente (Mon,) studied this question.