Convergence of Edge AI with Cloud Computing for AIoT (AI of Things) enabled by MEMS sensors
MEMS sensors with compute capability now enable intelligence at the edge, driving a paradigm shift in IoT data management. This shift is crucial in addressing the challenges posed by the traditional sensor-gateway-cloud architecture, which is burdened by the rapid proliferation of 18.8 billion and growing number of cloud-connected IoT devices. This massive scale of connectivity has led to the transmission of tens of zetabytes of data to the cloud, creating significant challenges in data management. While this data is crucial for the development of AI/ML models across IoT systems, the traditional architecture requires extensive coding and hardware expertise, posing major engineering hurdles.
ST AIoT Craft offers a revolutionary solution as a unique no-code/low-code sensor-gateway-cloud platform. This platform employs a distributed computing paradigm where intelligence is distributed. Machine learning algorithms run on sensors at the edge. Gateway logic manages sensor data transmission to the cloud and downloads new Edge AI algorithms to sensor nodes. AutoML algorithms generate the best Edge AI solutions using sensor data archived on the cloud. This platform allows users to manage intelligent IoT devices, gateways, and cloud infrastructure directly from a web browser. Users can program intelligent sensors to run machine learning models on-sensor and automatically update these models with new data over time. Some examples applications that can be built with this platform are: asset tracking devices enabled with MEMS sensor nodes installed in the asset, and, predictive maintenance application with classification logic running in the MLC of MEMS sensor. The platform provides precompiled examples and templates, an AutoML function to optimize machine learning classifiers on sensors, and data sufficiency logic to ensure only relevant data is sent to the cloud. This architecture conserves communication, compute, and energy resources. This solution achieves significant efficiency improvements in conserving communication bandwidth by sending only new data to the cloud for improving the classification or inferencing models. Since the classification logic executes in Machine Learning Core (MLC) in MEMS sensor, current consumption of sensor node is most efficient.
BIOGRAPHY
Mahesh Chowdhary, Ph.D. is a Company Fellow and Senior Director of MEMS software solutions at STMicroelectronics based in Santa Clara, CA, USA. He leads effort on development of solutions and reference designs for mobile phones, consumer electronic devices, automotive and industrial applications that utilize MEMS products, computing and connectivity products. His area of expertise includes AI/ML, MEMS sensors, IoT, digital transformation, and location technologies. He has been awarded over 50 patents. He has spoken extensively internationally about AI/ML, smart sensors, and IoT. Mahesh received Ph.D. in Applied Science (particle accelerators) from the College of William & Mary in Virginia.