ELSEIF
Your brief EB
183 stories from 71 feeds 32 clusters Refreshed 9 minutes ago next pull 13:20

AI Signal 230

Persistent State Machines: LLM Attention with INT4 In-Memory Cells

WHY IT MATTERS

This proposes an alternative hardware architecture for transformer attention that claims extremely low dynamic power consumption, which could matter for edge deployment if the simulation-based results hold on physical hardware. However, the energy figures exclude external memory and come from tool estimates rather than board measurements, so the practical advantage remains unproven.

Written by elseif from the cluster below · every claim links back to a source

The three things worth knowing

01

The framework mathematically proves quantization error bounds and deterministic finite-automaton equivalence for attention under a bounded-logits assumption, placing the approach in DSPACE(O(n)).

02

A 1024-cell array on a Zynq-7000 xc7z020 showed estimated dynamic power below 1.0 mW with normalized dynamic energy of 3.81 × 10^-5 pJ/op, while a 256-cell SoC on UltraScale+ xcvu9p closed timing at 62.5 MHz using only 0.67% of logic slices and zero DSP blocks.

03

All energy figures are simulation-based tool estimates with no physical FPGA board measurements, and system-level external memory energy is explicitly excluded from the analysis.

THE CLUSTER

Same story, 1 feed.

ORDERED BY FIRST SEEN
Hacker News Persistent State Machines: LLM Attention with INT4 In-Memory Cells Open ↗