Persistent State Machines: LLM Attention with INT4 In-Memory Cells
Summary
The Zenodo record presents a formal framework for attention operators in large language models via Persistent State Machines (PSMs), using in-memory cells and INT4 quantization. It provides complete mathematical proofs, hardware synthesis results (Vivado) on low-power FPGA platforms and SoC integration, and discusses energy estimates, timing, and patent status, illustrating a hardware-software co-design approach for AI acceleration.