Self-Powered IOT System for Edge Inference

Dileep Kurian1, Tanay Karnik2, Mukesh Bhartiya1, saransh Chhabra2, Saksham Soni2, Jaykant Timbadiya2, Suhwan Kim2, Krishnan Ravichandran2, Ankit Gupta2, Angela Nicoara3
1Intel technologies, 2Intel, 3HSLU


Abstract

We present a net-zero-energy IoT system mote for X86 class compute, image capture and visual recognition at the edge. The mote is self-powered by a single-inductor, 6-input, 9-output multi-modal energy harvesting power management IC (EHPMIC). The system connects wirelessly to a gateway that connects to the cloud. 14000 lines of software stack was implemented and the system was demonstrated to consume only 0.2mW idle and 24mW peak power. Measured minimum energy point was 6.2pJ/cycle at 0.5V/100MHz running Dhrystone benchmarks. EHPMIC with multi-modal harvesting capability from DC & AC sources delivers up to 120mW with a peak efficiency of 81.7% and only 120µW control power overhead.