SEADA: An efficient methodology for optimizing mixed-precision DNNs on multi-precision spatial architectures

2026-06-26

AI Summary

This work addresses the challenge of efficiently deploying mixed-precision deep neural networks (DNNs) on multi-precision spatial architectures, which requires joint optimization of precision assignment, quantization sensitivity, and system constraints. The authors propose SEADA, a novel framework that integrates a bit-entropy-driven per-layer precision selection strategy with a system-level analytical cost model. By incorporating a near-optimal mapping algorithm and a floating-point layer utility evaluation mechanism, SEADA enables efficient exploration of the multi-precision design space. The proposed approach significantly enhances deployment efficiency for mixed-precision DNNs, achieving notable improvements in latency, energy consumption, and memory footprint while supporting fast and robust design space search.

Abstract

Mixed-precision computation has been introduced in deep neural networks (DNNs) as an effective approach to reduce latency, energy consumption, and memory footprint. However, efficiently mapping mixed-precision networks onto multi-precision spatial architectures poses several challenges. These include determining the appropriate precision for each layer, balancing layer-wise accuracy sensitivity to quantization against architectural heterogeneity and system-level constraints, and accurately estimating the system-level cost of heterogeneous precision assignments. This work presents SEADA, an efficient methodology designed to address these challenges. SEADA comprises: (i) a configurable system-level analytical cost model of a multi-precision spatial accelerator architecture; (ii) a fast mapping tool that identifies near-optimal mappings of DNN workloads onto the target integer accelerator; (iii) analytical models for floating-point layers to estimate the overall benefits of mixed-precision execution; and (iv) a per-layer precision selection methodology based on bit-level entropy, enabling efficient assignment across multiple numerical precisions. SEADA's efficiency provides designers with a robust framework for the design-space exploration of multi-precision architectures.

Problem

• mixed-precision DNNs
• multi-precision spatial architectures
• precision mapping
• system-level cost estimation
• architectural heterogeneity

Innovation

• mixed-precision DNNs
• spatial architectures
• analytical cost model
• precision selection
• design-space exploration