SAfEPaTh: A System-Level Approach for Efficient Power and Thermal Estimation of Convolutional Neural Network Accelerator

TL;DR

The paper addresses the challenge of accurately and efficiently estimating power and temperature in tile-based CNN accelerators to enable thermal-aware design space exploration. It introduces SAfEPaTh, a system-level methodology that integrates AERO for latency/energy/area, HotFloorplan for floorplanning, and Hotspot for thermal analysis, including transient effects due to pipeline bubbles and inter-layer dependencies, using realistic workloads on the TANIA architecture. The approach eliminates circuit-level simulations and on-chip measurements, delivering end-to-end estimates within roughly 500 seconds and enabling rapid exploration across CNN models and accelerator configurations. Experimental results on ResNet18 demonstrate the ability to capture both steady-state and transient thermal behavior, including thermal-aware mapping and transient upper-bound analyses, highlighting SAfEPaTh’s practical relevance for designing energy-efficient and reliable CNN accelerators. The work broadens the toolkit for hardware-software co-design in CNN accelerators and lays groundwork for scaling the methodology to multi-tile and multi-cluster systems as well as diverse architectures.

Abstract

The design of energy-efficient, high-performance, and reliable Convolutional Neural Network (CNN) accelerators involves significant challenges due to complex power and thermal management issues. This paper introduces SAfEPaTh, a novel system-level approach for accurately estimating power and temperature in tile-based CNN accelerators. By addressing both steady-state and transient-state scenarios, SAfEPaTh effectively captures the dynamic effects of pipeline bubbles in interlayer pipelines, utilizing real CNN workloads for comprehensive evaluation. Unlike traditional methods, it eliminates the need for circuit-level simulations or on-chip measurements. Our methodology leverages TANIA, a cutting-edge hybrid digital-analog tile-based accelerator featuring analog-in-memory computing cores alongside digital cores. Through rigorous simulation results using the ResNet18 model, we demonstrate SAfEPaTh's capability to accurately estimate power and temperature within 500 seconds, encompassing CNN model accelerator mapping exploration and detailed power and thermal estimations. This efficiency and accuracy make SAfEPaTh an invaluable tool for designers, enabling them to optimize performance while adhering to stringent power and thermal constraints. Furthermore, SAfEPaTh's adaptability extends its utility across various CNN models and accelerator architectures, underscoring its broad applicability in the field. This study contributes significantly to the advancement of energy-efficient and reliable CNN accelerator designs, addressing critical challenges in dynamic power and thermal management.

Figures (10)

• Figure 1: TANIA tile-level architecture, consisting of PEs (AiMCore with AiMC arrays cosemans2019towards, digital Vector Functional Units (VFUs)), distributed activation buffers (ActBufs), and fanout network for intra-tile communication yang2022aero.
• Figure 2: Proposed system-level power and thermal estimation methodology.
• Figure 3: (a) Latency, area, and energy estimation of 3 CNN-layers mapped onto 2 PEs by using AERO; (b) Pixel-level power traces for different PEs and ActBufs, layer-level and inference-level power traces for PE1 (Assumes data is read from ActBuf0 and is written to ActBuf1 for computing L0).
• Figure 4: Floorplan of one tile in TANIA: Coarse-grained and fine-grained.
• Figure 5: Power traces of TANIA runs Resnet18 under AERO mapping.