Software-Defined Sensors

Whenever we have tired of owning boxes, we have gone out and bought even more. Only, this time, the new boxes are software-defined. Our technological generation is going through virtual appliances, software-defined storage, networks, and radios. My question is - where are we headed next, especially with the advent of IoT?

At first, the motivation is to reduce costs, dependence on certain vendors, and simplify management. However, we quickly realize that software-defined boxes are most useful for speeding up innovation. If this were the right thesis, then what is the obvious next software-defined box?

One of the elephants in the room has been the humble sensor. One sensor package lives in our pockets - the ubiquitous smart phone carries more dozens of sensors, many of which are used only for the phone to work, but at least a dozen are about the environmental conditions around the phone. There are several thousand more sensor packages created by companies from Memsic to Analog Devices and from Yokogawa to Texas Instruments.

My postulation is that just like there is one dominant compute company - Intel - there will be one dominant sensor company. A corollary is that just like more and more functions ended up in the CPU, such as DMA, timer, processing, graphics, etc. we are going to have a mammoth sensing system built on a single "die".

Already MEMS and nano-materials are making possible mechanical and chemical measurements at dizzying precision and versatility. Right now, we see an unending list of sensor packages being built by a variety of startups to pursue a particular group of use cases. However, in the absence of easy to make software, different sensors have to be combined with bespoke software to make sense of what the sensors are producing. All that is very expensive, laborious, and risky.

What happens when a single software package can make sense of all the data coming from any number of correlated or uncorrelated sensors? The entire need for building custom sensor packages would disappear.

That is already happening - my company, Falkonry, with its AI is able to make sense out of any sensor data within minutes. That means a mega sensor package kind of like the modern day CPU will be easily programmable to sense practically any behavior invented on Earth. That day isn't far when Intel, Texas Instruments, or Qualcomm start to go after the IoT sensing business with vengeance.

Today companies, like Bosch, are packaging a heterogeneous set of sensors into a single form factor. Such general purpose sensor packages are going to steamroll specialized sensors just like packagers overran computer assemblers during the early PC days. Then you marry the mega sensor package with a super-intelligent AI software and high schoolers will create the next Dell or Compaq.

I invite you to violently disagree, chip in with your observations, and join us on this journey to an uber software-defined sensor.