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

Consequently, the global count of IoT devices is projected (DL) models are increasingly deployed on Internet of Things to grow annually by approximately 40% [17], accompanied by a (IoT) devices for data processing, significantly increasing the carbon significant increase in their carbon footprint attributable to both usage footprint associated with DL on IoT, covering both operational and manufacturing. It is anticipated that the carbon emissions and embodied aspects. Existing operational energy predictors often stemming from IoT devices may surpass those of global data centers overlook quantized DL models and emerging neural processing by 2028 [17]. Despite extensive prior investigations [4] delving into units (NPUs), while embodied carbon footprint modeling tools the carbon footprint of MLaaS in cloud environments, a notable gap neglect non-computing hardware components common in IoT devices, remains in the comprehensive assessment of the carbon footprint creating a gap in accurate carbon footprint modeling tools for associated with DL models executed on IoT devices.