Did You Know TinyML is Big?

Posted June 28, 2021 in Technology

Micah Bevitz, iTelecom CEO

How does your phone respond when you say “Hey Siri” or “OK Google” if your phone is not even connected to the internet? How is your phone continuously listening?

Small neural networks, as small as 14 kilobytes, are on specialized microprocessor chips in your phone, and they are “trained” to detect these verbal commands. These chips use very little power, just a few milliwatts, instead of using the phone’s CPU, which consumes more power.

This is TinyML, and it stands for tiny machine learning. TinyML is an emerging technology to the public domain, but it has gained traction across many industries.

TinyML = Embedded System + Microcontroller + Machine Learning

TinyML is an embedded system that runs complex models on ultra-low-power microcontrollers. That’s a mouthful. Let’s break that down.

Embedded System

An embedded system is a computer system with a dedicated function within a larger device or electrical system. With one specific task, design engineers can optimize it to reduce the size and cost of the product and increase reliability and performance.


Microcontroller (MCU) is a low-cost, small computer on one single chip designed for embedded applications. They have the ability to retain functionality while waiting for an event, such as a button press or verbal command. These are also called ultra-low-power (ULP) microcontrollers and are in many IoT devices. Their power consumption is very low and well suited for long lasting battery applications.

Machine Learning (ML)

Machine learning is the study of computer algorithms that improve automatically through experience and data use. Machine learning algorithms build a model based on sample data, known as “training data,” to make predictions or decisions without being explicitly programmed to do so. Numerous applications use machine learning algorithms, such as email filtering, speech recognition, and image recognition.

Tiny is Mighty

Machine learning has been driving digital transformation. Previously, machine learning was computationally expensive, requiring huge data centers and powerful computing systems. Big tech companies were the only players that could develop and train successful machine learning models.

Now, TinyML microcontrollers can run deep learning algorithms. At a low cost and using open-source software, TinyML can democratize technological advancements. TinyML opens up opportunities for a broad set of developers, engineers, and businesses.

The power of TinyML enables devices to operate independently at the point of data collection. This on-device machine learning is fundamentally different from running machine learning in big data centers or the cloud. When you ask the device a question, it can answer right away using local functions and very little power.

IoT propelled rapid distribution of devices with over 250 billion microcontrollers in use today. That’s like every human on earth owning 32 Apple watches! Tiny? I think not.