Persistent State Machine: Breaking the von Neumann Memory Wall for LLM Attention
The Persistent State Machine addresses the memory wall in LLMs, cutting energy use and data traffic significantly.
The Persistent State Machine (PSM) aims to overcome the memory wall in LLM inference. This paper defines PSM as a mathematical 6-tuple and proves its equivalence to Deterministic Finite Automata (DFA). The Active State-machine Memory Architecture (ASMA) implements PSM, reducing system bus traffic by 99.47% and energy consumption by 99.44%.
This synthesis was produced from its source by AI; there is no human editor or manual review step. How we work