By John Blyler, Content Officer
Challenges abound as designers deal with the analog nature of sensors, IP issues and the new algorithms required by the IoT.
Sensors represent both the great enabler and unique challenge for the evolving Internet of Things (IoT). Innovation in the market will come from surprising places. These are just a few of the observations shared by ARM’s Willard Tu, Director of Embedded and Diya Soubra, CPU Product Manager. “System Engineered Design” caught up with them during the recent ARM Tech Con. What follows is a portion of that conversation. – JB
Blyler: Everyone talks about the importance of sensors to enable the Internet of Things (IoT) but few seem to appreciate what that means. Would you elaborate?
Tu: Sensors are one of our key initiatives, especially from a microcontroller viewpoint (more on that shortly). But there is another aspect to sensors that both designers and even companies overlook, namely, the algorithms for processing the sensor data. These algorithms, from companies like Hillcrest, bring a unique value to the IoT market. And the algorithm software represents a real intellectual property (IP). I think people are missing out on the IP that is being created there.
Blyler: So you think that most people overlook the IP aspects and simply focus on the processing challenges needed to condition analog sensor signals into a digital output?
Tu: Processing power is critical, which is where distributed local and cloud computing comes in. But there are many other factors, such as energy harvesting to power sensors in areas you never thought of before. Both body and mess network communication challenges are another factor. Conversely, one enabler of sensors is their inexpensive cost. Ten years ago, an accelerometer was a really expensive piece of silicon for an automotive airbag system. Now, they are everywhere, even in cell phones which are very cost sensitive.
Blyler: Is this volume cost decrease due to innovation in MEMS design and manufacturing?
Tu: Yes, the MEMS market has evolved immensely (see Figure 1) and that’s the reason. And I think there is still a lot of evolution there. You see a lot of new comers with MEMS applications but I think you’ll see a lot of consolidation because only the strong will survive.
Soubra: Another factor is that few vendors use only one sensor, but rather a lot of sensors. A common example is multi-sensor accelerometers (see Figure 2): one sensor gives you a good pitch, the others give you yaw and roll. So you will always need three of four sensors, which means that you have to have software to handle all of them.
Blyler: Do you mean software to control and integrate data from the various sensors or software algorithms to deal with the resulting data?
Tu: Both – Software is needed to control and ensure the accuracy of the sensors. But developers are also doing more contextual awareness and predictive analysis. By contextual, I mean that a smart phone turns on when it’s being held next to my head. Predictive refers to what I’ll do next, i.e., having the software anticipate my next actions. Algorithms enable those capabilities
This is the next evolution in handling the data. You can use sensor fusion (sensors plus processors) to create contextual awareness. That’s what people are doing today. But how does that evolve into predictive algorithms? Anticipating what you want is even more complex than contextual awareness. It’s like using Apple’s Siri to anticipate when you are hungry and then order for you. Another example is monitoring a person’s glucose level to determine if they are hungry – because their glucose levels have dropped. It could be very intuitive or predictive down the road.
Blyler: These smart algorithms are another reason why processing power is a key enabler in the IoT evolution.
Tu: What you really need is scalable processing power. Sensors require a microcontroller, something with analog inputs. But there are still lots of designers who ask, “Why do you need to integrate the microcontroller with the sensor? It’s just an accelerometer.” They seem to forget that data acquisition is an analog process. The sensor data that is acquired must be conditioned and digitized to be useful in contextual or predictive applications. And that requires lots processing.
Another thing designers forget about is calibration (see Figure 3). Calibration is a big deal to get the accuracy necessary for all the contextual awareness applications. Calibration of the MEMS device is only part of the issue. The device must be recalibrated as part of the larger system once it is soldered and packaged to a board, to deal with temperature affects (of the solder) and flexing of the board. All of these things play a part of the system-level calibration.
You might think that, well, the sensors guys should do that. But the sensor guys are good at making a MEMS device. Some MEMS manufactures are vertically integrating to handle calibration issues, but others just want to make the device. This is another area where innovate IP can grow, i.e., around the calibration of the MEMS device to the system.
Blyler: Where will innovation come from as the IoT evolves?
Tu: I think the ecozystem is where innovation will emerge. Part of this will come from taking application developed in one area and applying them to another. Recently, I talked to several automotive developers. They admitted that they lack of expertise in developing certain types of algorithms – the same algorithms that companies like Hillcrest have already created for mobile consumer applications. I would like to introduce the automotive market to an algorithm company (like Hillcrest), a sensor platform provider (like Movea) ant a few other leaders in the mobile space.
I think you will see IP creation in that space. That is where innovation is coming, by taking that raw sensor data and making it do something useful.
Blyler: Consolidation is occurring throughout the world of semiconductor design and manufacturing, especially at the lower process nodes. Do you see similar consolidation happening in the sensor space.
Tu: Right now there is an explosion of sensor companies, but there will be a consolidation down the road. The question one should ask is if integration key to the sensor and IoT space. I don’t know. As a company, ARM would like to see a microcontroller (MCU) next to every sensor or sensor cluster – whether it is directly integrated to the sensor array or not. This is where scalability is important. Processing will need to be distributed; low power processing near the sensor with higher performance processing in the cloud. It is very difficult to put a high-powered fan based system in a sensor. It just won’t happen. You have to be very low power near the sensor.
Not only is the sensor node a very power constrained environment but it is also resource constrained, e.g., memory. That’s why embedded memory is critical – be it OTP or flash. In addition to low power, it is the cost of that memory is actually more influencing than the CPU.
Blyler: Thank you.