It's nice to use the death time between active processes. It's nicer to use it with AI! However, it's done on purpose to make the microcontroller to sleep as long as possible, so the power consumption is reduced to the maximum. Many of these devices are battery-powered.

I can imagine a Cortex-M0 and -M3 cores running FFT's, hidden Markov filters, bayesian networks, Kalman filters, etc. However, they are going to suffer from the same illness that my mobile phone (and yours) suffers now: I must recharge the phone battery as frequently as I can. And nobody wants to change the batteries when the device has been located at the top of the highest tree in the neigborhood.

But, if the node is going to be powered from the wall ... well, thats another tale. Make it active 100% of the time!

Hey Xavier,

You are absolutely right on the MCU idle time and energy. Any computation comes with the price of energy. Nokia phones used to last weeks, whereas, the iPhone only last days. The way I see it, bring AI to MCUs enable us to do more with the expense of energy. But it opens doors to new applications.

It all depends on the use-scenarios. The nice thing about MCUs is their standby power compare to more powerful processors. This lower standby power could mean that devices can be sustained using energy-harvesting technologies.

However, if it's on a small battery, well, we need to be really smart on how we budget the energy.