IoTCO2: Assessing the End-To-End Carbon Footprint of Internet-of-Things-Enabled Deep Learning

TL;DR

This paper introduces \textit, an end-to-end tool for precise carbon footprint estimation in IoT-enabled DL, with deviations as low as 5\% for operational and 3.23\% for embodied carbon footprints compared to actual measurements across various DL models.

Abstract

To improve privacy and ensure quality-of-service (QoS), deep learning (DL) models are increasingly deployed on Internet of Things (IoT) devices for data processing, significantly increasing the carbon footprint associated with DL on IoT, covering both operational and embodied aspects. Existing operational energy predictors often overlook quantized DL models and emerging neural processing units (NPUs), while embodied carbon footprint modeling tools neglect non-computing hardware components common in IoT devices, creating a gap in accurate carbon footprint modeling tools for IoT-enabled DL. This paper introduces \textit{\carb}, an end-to-end tool for precise carbon footprint estimation in IoT-enabled DL, with deviations as low as 5\% for operational and 3.23\% for embodied carbon footprints compared to actual measurements across various DL models. Additionally, practical applications of \carb~are showcased through multiple user case studies.

Figures (4)

• Figure 1: Total IoT count.
• Figure 2: IoT carbon footprint.
• Figure 3: The energy consumption of conv+bn+relu with various configurations on Snapdragon 8 G3.
• Figure 5: The IoT device carbon footprint breakdown (NPU: employing NPU; rcase: recycled casing; rPCB: recycled PCB; and 22nm: fabricated by the 22nm process technology).