Going Green with TinyML

As the Internet of Things (IoT) continues to grow and become increasingly integrated into our daily lives, concerns are rising about the environmental impact of these devices. From e-waste to carbon emissions, the rapid proliferation of IoT devices is having a significant impact on the environment.

E-waste is a major issue when it comes to IoT hardware. With an estimated 50 billion such devices connected to the internet by 2025, the amount of e-waste produced by these devices is expected to increase dramatically. Many IoT gadgets are designed to be disposable, and are often not easily recyclable due to the complexity of their design and materials used. This results in large quantities of e-waste, which can have a serious impact on the environment.

In addition to e-waste, IoT hardware also contributes to carbon emissions. Many of these devices are powered by batteries or otherwise require energy to operate, which can have a significant impact on the environment. In fact, a study by the International Energy Agency found that by 2025, the energy consumption of IoT devices could be responsible for up to 1.9% of global greenhouse gas emissions.

A group of researchers centered around Harvard University asked a question that at first might seem counterintuitive — can producing even more micro-sized hardware platforms help to alleviate these environmental concerns? The platforms in question are of a particular type, referred to as tinyML devices, in which machine learning algorithms run on low-cost, low-power microcontroller systems. They hypothesized that the intelligence baked into these algorithms may be capable of offsetting the carbon emission that they generate by reducing emissions in other sectors.

For those familiar with some of the typical applications of tinyML in consumer devices, keyword spotting, image classification, and anomaly detection are likely among the first to come to mind. However, the researchers point out that this technology is also very well-suited to improving the sustainability of global agriculture and assisting with wildlife conservation, for example. Existing systems have already proven themselves to be more accurate and efficient than humans in detecting plant diseases, which helps farmers to substantially improve both yields and profits. Other systems have shown themselves to be very valuable in preventing elephant poaching and collisions between boats and whales in busy waterways.

This evidence that tinyML may have a net positive impact on the environment is thus far anecdotal, so the researchers set out to quantify it. After all, while the environmental impact of an individual tinyML device may be small, when that impact is multiplied by the tens of billions of devices in operation, it becomes a major concern. So, to quantify the environmental impact of these systems, an open source TinyML Footprint Calculator was developed.

The algorithm behind this calculator takes into consideration many factors, including the entire life cycle — from raw material harvesting through end of life — of microcontrollers. Other factors like water demand, freshwater eutrophication, and photochemical oxidant formation are also factored into the algorithm. The result of the calculation is a metric estimating the total carbon footprint over the lifetime of a tinyML device.

An analysis was conducted in which a variety of tinyML devices were compared with commercial products, and it was found that they did in fact have a relatively small environmental impact. A low-cost tinyML device, like a keyword spotter, would be expected to have a 38 times smaller carbon footprint as compared to an Apple Watch Series 7, for example.

The research has shown that tinyML has a place in contributing to a sustainable future, however, the team showed that there are additional areas of research that can further contribute to that goal. Energy harvesting, for example, can reduce both carbon emissions and e-waste. Designing and building more efficient sensors, and focusing on recycling and upcycling are also areas that could pay dividends in the future if given more attention. It appears as though tiny devices could have a big impact on the future health of the planet.