A 28nm 47.3TFLOPs/W 894mJ/Inference Visual Autoregressive Accelerator with Differential-Amplifier Speculation and Chain-Reaction-Like Parallel Generation

View at publisher
Authors: Zhiheng Yue, Xujiang Xiang, Jiamu Fu, Shaojun Wei, Yang Wang, Yang Hu, Shouyi Yin
Cited by: 0
References: 4

Abstract

To accelerate Visual Autoregressive (VAR) applications, this work implements a 28 nm VAR accelerator achieving 47.3TFLOPs/W and$5.76 \text{mm}^{2}$chip runs at 400 MHz, accelerating DeiT/ViT VAR by$37.6 \times$with$2.75 \text{TFLOPs} / \text{mm}^{2}$area efficiency.

Fields of Study & Subjects

Advanced Neural Network Applications Low-power high-performance VLSI design Parallel Computing and Optimization Techniques

Publication details

Journal / source
Digest of technical papers/Digest of technical papers - IEEE International Solid-State Circuits Conference
Pages
316–318
Publication date
15 February 2026
Language
EN