The Internet of Things (IoT) marketplace continues to evolve. It has been estimated that there are currently ~ 23 billion connected devices operating worldwide. And yet, we are still awaiting the inevitable hockey stick in that growth curve. Why is that? What have been the limiting factors?
From household appliances, to health tracking devices, to smart manufacturing equipment, many use cases have not been able to find an adequate ROI due to the costs associated with traditional artificial intelligence (AI) enablement. The cost of extraneous hardware, data bandwidth, and cloud processing have caused many potential AI-enabled IoT adopters to delay their investments. There is also the lingering concern about the security of data that is collected and processed in the cloud.
The ability to embed and process AI directly at the endpoint of the IoT device will be the key factor in achieving exponential growth in the IoT market. This trend already has a name, “edge native applications.”
Endpoint AI means AI that is embedded and trained right at the device or machine. Machines are not computers, and they usually run on embedded systems composed of microcontrollers and components. According to IC Insights, there are about 250 billion microcontrollers in the world today, and 38.2 billion will be sold annually by 2023.
This increase in growth of microcontrollers sold is mostly due to the need to connect IoT devices to the Internet and start making sense of the data that those devices generate. With the data now available to process, running AI models and analytics on microcontrollers and sending only mission-critical data to the cloud seems to be the next logical step. It will decrease bandwidth cost and provide security and privacy to the base datasets, creating a very promising pathway to develop those new “Edge Native Applications.”
The Future in Motion
And this future is already in motion. Semiconductor companies are investing billions of dollars in developing more energy-efficient hardware and algorithms. Software companies like Google are also developing frameworks, such as Tensorflow lite, specifically for this purpose. Corporate and venture capital investment has now reached more than US$ 54.8 Billion, and start-ups like Kneron and ETA compute are making steady progress. The microcontroller companies themselves are also working on these technologies, and some of them are already using ARM Cortex M processers and can also leverage ARM’s Ethos-U55 AI. TinyML, the industry association promoting Tensorflow lite, believes that advanced machine learning should hit the market in two to three years, and killer apps seem to be 3 to 5 years away.
MicroAI™ Atom – A Paradigm Shift
Suppose you want to test the concept of edge native applications within your new or existing use cases? Suppose you are looking for a more asset-centric–cost-effective– approach to IoT asset management? There is now an existing solution available to download on GitHub.
On January 12th, 2021, ONE Tech announced launch of the MicroAI™ Atom SDK. Atom enables AI and machine learning algorithms to be embedded and trained directly onto the microcontroller unit (MCU) of the IoT asset. MicroAI™ Atom still requires 32-bit architecture microcontrollers, so it does not run on many smaller hardware units. The install base of 32-bit MCUs is big enough to enable many companies to develop and run their apps on existing devices and machines while waiting for the technology to catch up with the smaller devices.
Access the MicroAI™ SDK here