Analog-Mixed-Signal Computing Microcontroller Unit for Artificial Intelligence of Things: Toward Real-Time Carbon Monitoring and Management

Abstract

Climate change has become a critical challenge in the last few decades. It has a massive effect on our livelihoods, health, and future. As a result, the world leaders began to set an agenda for reducing greenhouse gas emissions. The agenda consists of mainly three parts: (i) monitoring, (ii) abatement, and (iii) offsetting and artificial intelligence of things (AIoT) is considered integral to implementing all the three parts of the emission-reduction agenda. For example, with myriad databases and systems involved with different carbon-producing assets, the labor and cost required to simply organize the data are immense. AIoT can enable seamless processing of real-time activity, enabling various entities to efficiently structure, collect, and transform data into reports. Unfortunately, however, today’s embedded/mobile devices cannot support the vision of AIoT. Those devices’ primary choice of computing hardware is a microcontroller unit (MCU) such as ARM Cortex M Series. As of now, such units do not have the desirable computing speed and energy efficiency to tackle artificial intelligence (AI) and machine learning (ML) workloads such as convolutional neural networks (CNN). In light of this, we aim to create a new MCU integrating analog-mixed-signal (AMS) computing hardware. Multiple recent research indicated that AMS computing hardware could enable orders of magnitude improvement in energy efficiency and throughput for AI and ML workloads. We will prototype such a chip in advanced CMOS technology and will benchmark its performance and energy efficiency to verify the chip’s supportability for AIoT devices.