LionHeart: A Layer-based Mapping Framework for Heterogeneous Systems with Analog In-Memory Computing Tiles

2024-01-17

arXiv.org

AI Summary

To address time-varying accuracy degradation in analog in-memory computing (AiMC) hardware—caused by device non-idealities such as noise, drift, and aging—this work proposes LionHeart, a hierarchical mapping framework. Its core innovation is the first layer-granularity, time-varying non-ideality-aware dynamic mapping mechanism, enabling accuracy-constrained analog/digital co-scheduling and overcoming the adaptability limitations of conventional static mapping. LionHeart integrates circuit-level non-ideality modeling, inter-layer accuracy sensitivity analysis, a multi-objective optimization solver, and a cycle-accurate simulation platform. Evaluated across multiple DNN models and datasets, it achieves inference accuracy approaching floating-point baselines while reducing system execution time by 6.1× and improving energy efficiency by 6.3×, all while strictly satisfying user-specified accuracy thresholds.

Technology Category

• Machine Learning: Hardware-aware ML
• Cognitive Modeling & Cognitive Systems: Analogy
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

When arranged in a crossbar configuration, resistive memory devices can be used to execute MVM, the most dominant operation of many ML algorithms, in constant time complexity. Nonetheless, when performing computations in the analog domain, novel challenges are introduced in terms of arithmetic precision and stochasticity, due to non-ideal circuit and device behaviour. Moreover, these non-idealities have a temporal dimension, resulting in a degrading application accuracy over time. Facing these challenges, we propose a novel framework, named LionHeart, to obtain hybrid analog-digital mappings to execute DL inference workloads using heterogeneous accelerators. The accuracy-constrained mappings derived by LionHeart showcase, across different DNNs and datasets, high accuracy and potential for speedup. The results of the full system simulations highlight run-time reductions and energy efficiency gains that exceed 6X, with a user-defined accuracy threshold with respect to a fully digital floating point implementation.

Problem

• Mapping for analog-digital systems
• Addressing analog computation challenges
• Enhancing accuracy and efficiency

Innovation

• Layer-based mapping framework
• Analog in-memory computing
• Hybrid analog-digital mappings