AIDA: Associative DNN Inference Accelerator

TL;DR

AIDA is an innovative associative in-memory DNN inference engine that accelerates fully-connected layers by processing data in-situ, leveraging sparsity and low precision for significant performance and efficiency gains.

Contribution

The paper introduces AIDA, a novel associative in-memory processor for DNN inference that outperforms existing accelerators like EIE in performance and energy efficiency.

Findings

AIDA achieves 14.5x peak performance over EIE.

AIDA is 2.5x more throughput-efficient.

AIDA benefits from sparsity and low-precision arithmetic.

Abstract

We propose AIDA, an inference engine for accelerating fully-connected (FC) layers of Deep Neural Network (DNN). AIDA is an associative in-memory processor, where the bulk of data never leaves the confines of the memory arrays, and processing is performed in-situ. AIDA area and energy efficiency strongly benefit from sparsity and lower arithmetic precision. We show that AIDA outperforms the state of art inference accelerator, EIE, by 14.5x (peak performance) and 2.5x (throughput).