Making Sensors More Reliable

Experts at the Table: Semiconductor Engineering sat down to talk about the latest issues in sensors with Prakash Madhvapathy, director of product marketing, Tensilica audio/voice DSPs group at Cadence; Kevin Hughes, senior product manager for MEMS sensors at Infineon; and Matthew Hogan, product management director at Siemens EDA. What follows are excerpts of that conversation.

[L-R] Kevin Hughes, Infineon; Matthew Hogan, Siemens EDA; Prakash Madhvapathy, Cadence.

SE: What do you see as the overall trends in the sensor space?

Hughes: Infineon has focused on two macro trends: digitalization and decarbonization. One of the areas where these align for sensors is in corporate sustainability. Everyone is looking to be more energy efficient, not only for environmental reasons, but because it also has an impact on the bottom line. We’re working with a lot of companies to implement sensors as a way to drive energy efficiency and reduce waste. For example, I’m currently working on automation systems for buildings, using occupancy detection, air-quality monitoring, and other sensors to predict and understand how buildings are being used, so lighting and HVAC can be operated on demand. Instead of wasting energy, you’re only using the amount that you need when you need it. It allows for predictive maintenance, which also can help save energy. By using sensors on critical infrastructure and manufacturing equipment to monitor their health in real time, you’re able to send early maintenance warnings, so that you can make sure that all the machinery is operating at optimal efficiency and extending lifetimes. On a longer-term basis, you’re reducing the overall industrial waste of having to cycle through these machines.

SE: Does that mean your own office becomes a testbed?

Hughes: Yes, many of our offices have radar sensors and CO₂ sensors, collecting data all the time.

Hogan: There has been an explosion in different types of sensors, their applications, and the desire to use them in different areas. But from a manufacturing perspective and IC-integration perspective, we’re also seeing an emphasis on being more energy efficient, which means that the actual chips that are used in these environments are more energy efficient, as well. A lot of our customers are looking at using lower voltages, smaller devices, smaller components, greater integrations of those systems. What we’re seeing is that systems are becoming more sensitive to things like electrostatic discharge. From a reliability perspective, it’s making sure you’re covering ESD events so the sensors are going to have longevity in the field and can last for many years. They’re not just a short-term proposition that needs to be constantly replaced. During the design process, making sure that different aspects of the design are considered is where we see a lot of people working. They’re trying to understand how to make these sensors smaller, faster, better, but also more reliable than the previous generation because lots of them are being deployed. And as you start deploying more and more, your failure rates start to become more noticeable. How do we provide an order-of-magnitude-lower failure rate so that all these sensors being deployed can be seen in the environment?

Madhvapathy: The overall trend depends upon the industry that you’re looking at. Certainly, one trend is power efficiency, because energy is becoming very, very important. With the explosion in the number of sensors being deployed in industry, in smart cities, and in consumers’ homes, it’s important to keep the energy profile very low. Another trend is choices in user interfaces. We’re seeing a big expansion of applications for different ways to interact with our devices. Of course, there’s voice with Amazon Alexa and ‘Okay-Google’ types of devices, but that is now moving toward other devices that are not connected to the internet. Being able to talk to them in a more natural language, one can get to a better level of service from the devices, which includes things like washers, dryers, ear buds, robots — any IoT device in the field. If they can respond to voice commands or gestures, or any other methodology, that can greatly enrich and speed up the process of interaction with those kinds of devices, even including automobiles. For example, now they’re deploying things like ultrasound waves, so you can use gestures instead of physical buttons. It’s easier to interact with devices that way than to go through a plethora of menu items on the screen while you’re driving. There are also a lot of security use cases, where a device or building wants to figure out if a person should be granted access. In such cases, both voice biometrics and vision can be combined to create a personal profile, which is then used every time for authentication. For health use cases, there is continuous monitoring. Today when you have health issues, you have to go to the doctor and get monitored there, and you may have to go every few months to get re-checked. But if you have a device that you can wear or get implanted, you get continuous monitoring 24/7. When there’s an issue, the doctor can be alerted and timely action can be taken, rather than having to wait for a clinical appointment.

SE: That’s happening today with glucose monitoring, right?

Madhvapathy: Yes, and combined with that is the drug delivery aspect. You can have something embedded in the body that can create a control loop for timely action. For example, whenever a certain parameter goes beyond limits, it can release a drug into the body. Those are the kinds of trends that we’re seeing.

SE: All of this suggests that human factors engineering is going to come to the forefront.

Madhvapathy: Yes, if a wearable device is uncomfortable, a person is most likely not going to use it, even if they know it can be life-saving. For example, we have all these Fitbits and so on that can monitor your sleep, but most people don’t wear them because they are intrusive. When you have an implantable device, that’s a bigger deal, because that has to be FDA approved to make sure it doesn’t cause any of the problems just by inserting that in the body. And that device cannot be uncomfortable to the user, causing pain or discomfort, because it will be in the body for a long duration. And, of course, educating the patient that this is not going to cause long-term problems and it is actually better for them is another aspect that the medical community will have to take account of.

SE: What are the biggest challenges to reliability in sensors?

Hughes: It comes down to the industry that you’re in. One of the areas I’m working on is environmental sensors. You have to understand whether the part will be used in a sealed/controlled environment or an exposed environment. We spend 90% of our time indoors, which is a relatively narrow, controlled climate range, so this provides more predictable lifetime conditions for a sensor integrated into something like a personal device or indoor monitor. Whereas if you’re developing an outdoor agricultural sensor, soil sensor, or something similar, you have to take into account that you’re going to be experiencing very high climate extremes, high levels of moisture, dust, and potentially chemical exposure. It all comes back to the design process. You need to take the end use case into account when you’re doing your failure-mode analysis. You need to understand, for example, what the impact of moisture or chemical exposure is going to be on your sensor and make sure that you design for that because your device will be expected to operate reliably in the field, within datasheet specs, regardless of extreme temperatures or harsh environments.

SE: How is that affecting design choices and the future of certain technologies? Is it going to push innovations for memory or other components?

Hughes: Yes, absolutely. When you are exposed to temperature and humidity extremes, you have issues with outgassing and delamination and things like that, so there are lots of opportunities to develop more reliable materials. But a lot of it comes down to packaging, as well. My professor for my MEMS courses said that the effort for commercializing a sensor was 10% designing the sensor and 90% packaging it. You need to make sure that your system — not just the materials in the sensor, but the whole system that encompasses it — is designed to reject mechanical stress due to swelling or wire-bond delamination, and things like that.

Madhvapathy: To go back to the implantable sensor use case, the body has a lot of fluids. All these fluids are conductive and they’re very harsh. Even a tiny pinhole in the material can allow the liquid to seep in, and over time it can flood the electronics and cause short-circuiting. It’s extremely important to create the proper materials and appropriate layers so that you have some barrier against the biofluids. At the same time, you cannot have so much barrier that it affects your sensing, as well. You have to be able to sense whatever signal you’re trying to send, while protecting the internal circuits from the biofluids. It comes down to testing it before deploying it. For example, MEMS sensors are tested very thoroughly by dropping them from different heights or onto different materials, such as concrete or steel. There also comes a point that if you don’t test thoroughly or design very well, it can affect your yield, which can cut into your profit margins. So everybody tries for five-nines reliability. Otherwise, the profit margins get affected. You do a lot of thorough testing in different environments, including immersion in different liquids and so on to ensure the device is performing well after it’s taken out.

Hogan: On the automotive side of things, they’ve adopted some standards and refinements and enhanced those from a manufacturing side. For automotive, we have ISO 26262. It helps the automotive ecosystem understand whether or not what they’re designing is going to be compliant and reliable enough, and they refer to this concept called mission profile, which is understanding how the device is going to be used, what are its use conditions. More importantly, from an automotive and functional safety perspective (FuSA), it’s about being able to fail safely, making sure that when your device does fail, that it fails in a safe or appropriate mode that’s not going to impact the functional safety of that system — particularly if it is a safety-critical element. There’s also the idea of re-using the IP inside of your designs. We’ve had many occasions where we’ve been able to help customers out. They’ve used a piece of IP, one design, one mission profile, that they think is great and wonderful. They use it in the next design, and they hook it up differently. It’s used in different use scenarios and use cases. From a verification perspective, we’ve been able to provide an environment where they can make sure their IC is going to function correctly. Because one of the caveats of ISO 26262 is you can have an expert who says, ‘We’ve used this enough. We have enough experience with this. We’re going to accept it.’ But if you don’t understand those mission profiles, and if you’re not validating from an IC verification perspective appropriately, there may be reuse of IP that is problematic, because you’re actually changing the way it’s being used. You’re changing the environment it’s behaving under and you’re seeing some unexpected conditions or use models or environments now that weren’t originally planned or designed for.

Related Reading
Confusion Grows Over Sensor Fusion In Autos
Multiple approaches are being explored for multiple data types, but it’s still too early to say which is best — or whether any of them will shorten time to market for autonomous vehicles.
Machine Vision Plus AI/ML Adds Vast New Opportunities
But to fully realize its potential, MV must boost performance and keep pace with changing security and market needs.