This is part fifteen of a series. Read part fourteen here.
The amount of performance needed to function is always a difficult question to answer. I suggest the reason for this difficulty when considering the needs of a smart sensor is the fact that we have grown up in a performance world with an understanding that every 18 months or so we would have more performance. So, if we don’t have enough performance today, we can just wait a bit and we will have it. But when it applies to smart sensors, a new performance metric emerges. That new performance metric is energy efficiency.
It surprised me when I realized that as much as I was trying to convince people that power dissipation was important, I was being ignored. That is until it became obvious that we couldn’t increase the performance of our devices and systems because of the increased energy being dissipated as we drove the performance up.
We had reduced the voltage rails as low as we could take them while still not sacrificing performance. We had exhausted packaging techniques which could conduct the heat away from our processors. We had created clever power management systems to make sure we were not wasting any energy on non-critical activities. We turned off circuits when they weren’t being used (I call that method the “father’s solution to wasting power.” My father would say things like, “Turn off the lights when you leave your room,” or, “Were you born in a barn?”).
But we focused on the needs for cloud computing more so than for smart sensors. Now, in the IoT system we have a need for ultra-low powered smart sensors along with energy efficient high performance cloud computing devices.
So far, I have not answered the question of how much performance we need in a smart sensor and have cleverly diverted the discussion to power dissipation of high performance processing systems. I’ll now attempt to rein my discussion back in and give an answer.
The answer is smart sensors will need just enough performance to obtain the input from the sensors, process the signal and send the results to the communications system to be transmitted to the aggregator. But this doesn’t answer the question, does it? To help, let me complicate it even further with a few questions:
- Can we afford to have an operating system in a smart sensor?
- If we do need some sort of an ultra-low powered operating system? Does one exist?
- Can we afford to use a sophisticated language to program the processor?
- Will we need to revert to assembly code to get the power efficiency we need to maximize the real performance?
- Security would seem to be a luxury and need to be sacrificed, right?
- Is it time to get rid of the clock and go clock-less? Can non-uniform (e.g., random) sampling be OK?
- Is it time to go back to analog computing?
Now I feel better. Why? Because I’ve turned the tables on you. I’d like to hear your answers on the topic.