behrooz abdi, tdk invensense

An interview with long-time semiconductor executive and Chief Executive Officer of TDK InvenSense

 


The following is a transcript of the audio available via the player above.  The audio file is the definitive source.

Ali Tabibian:                 Welcome, welcome, welcome everyone to this episode of Tech. Cars. Machines. I'm your host Ali Tabibian. We've crossed the thousand listener mark and we're glad you're all here. If you look at our website, www.gtkpartners.com or listen to our introductory episode of this podcast series, you'll know that we think Tech. Cars. Machines is about the intersection of three major areas [00:00:30] of technology and the world of course of cars and machines. The three areas of technology are sensors, connectivity, and software. 

                                    Today, we're going to have our first episode on sensors. As usual, we bring you the very, very best people to understand what's going on in the particular world we're about to talk about. Our guest today is Behrooz Abdi, currently the chief executive officer of TDK InvenSense. He's a very long tenured semiconductor executive, has been a chief [00:01:00] executive a couple times. Most recently and currently as chief executive of TDK's InvenSense unit.

                                    Behrooz was originally an investor and board member in InvenSense when it was venture backed. He became its chief executive when InvenSense was a public company and through a lengthy process that ended or came to fruition I should say in late 2016, he orchestrated the sale of InvenSense for $1.3 billion to [00:01:30] TDK Japanese multinational. At the time, InvenSense probably had half the world's market share in high end smartphones like iPhones and Samsung galaxies as far as some of the sensors in those devices were concerned and it had pretty good footprints elsewhere. Their new owner, TDK has a strong position in automotive sensors. That unit also now reports to Behrooz in the combined TDK InvenSense unit. 

                                    A quick note about something that Behrooz mentions in this interview. When describing [00:02:00] the bright future of sensing, he uses the word integration. Given his background at Qualcomm, there's more meaning to the word integration when it comes from Behrooz. You see, for years, the fortunes of Qualcomm, a successful and storied name in cell phones and smartphones especially in the communications technologies like CDMA, Qualcomm's success was driven by that one word, integration. Meaning this, every generation of cell phone required integrating greater numbers [00:02:30] of antennas and frequencies so the phones could work in a greater number of locations, in different countries and utilize multiple cell towers for maximum benefit. It seems like Behrooz sees the future of sensing to be at the same spot. 

                                    By the way, Behrooz has a very good sense of humor. It didn't really come out in the formal parts of the interview so we left in a few seconds of the informal parts of the taping so you can see what we mean. To go directly to the interview, go to minute [00:03:00] seven and a half of this episode. Otherwise, I'm going to spend the next few minutes explaining a few things about why sensors are a big deal. 

                                    A sensor is a device that measures some characteristic of the physical environment such as temperature or a relationship with that environment such as am I accelerating or coming to a stop. You're all familiar with the humble mercury thermometer. It was invented about 200 years ago by Mr. Daniel Fahrenheit in Amsterdam. To put a fancy phrase to [00:03:30] it, the thermometer is a temperature sensor. So there's an example of a sensor, a very simple sensor which has been around for centuries. Light sensors are pretty important too. Beginning in the 1840s, the most important light sensor of all was silver highlight film, the kind that made the Kodak company famous and made photography ubiquitous. 

                                    Another example, sensing motion is helpful in ships, aircrafts and rockets. Gyroscopes, which are devices originally with parts that spun like a top [00:04:00] are in this class of sensor. About 150 years ago is when their technology started becoming useful commercially. Early very precise versions that went into rockets and missiles were the size of really large balls, you know, the kind that people bounce around on in the gym to pretend that they're exercising. The big deal is that people have figured out how to make an increasing number of sensor types with semiconductor manufacturing techniques. These are the same techniques that deliver wondrous results in your smart phones and computers. [00:04:30] This means you are harnessing hundreds of billions of dollars of investment that the semi industry has made in miniaturization, precision, robustness, and critically in manufacturing chips that are easily integrated with electronic devices.

                                    Let's go back to our light sensor. Can you imagine trying to squeeze 35 millimeter Kodak film into your smartphone? The reason you don't have to do it is that a particular type of semiconductor chip called a charged couple device or CCD [00:05:00] was developed beginning in the late 1960s. A CCD is basically a matrix of tiny sensors each called a pixel, which detects light just like the Kodak film. Unlike film, obviously, the light is immediately converted to an electrical value for processing and it can be erased and reused repeatedly. When Apple says the iPhone X has a 12 megapixel camera, it means the CCD chip inside has 12 million tiny sensors in it. The same compact and cheap chips [00:05:30] enable your automobile to have multiple cameras installed including those that contribute to self-driving.

                                    Now film is a flat simple mechanism. Can we make a lot more complex devices in semi technology? Can we make a gyroscope and mimic the output of its spinning parts using semi technology? I don't know, Can we? 

Barack Obama:             Yes, we can. Yes, we can. Yes, we can.

Ali Tabibian:                 About 30 years ago, another big deal happened [00:06:00] in sensing, and that's called micro-electrical mechanical systems or MEMS. Basically, you're carving complex patterns on silicon to mimic physical devices. You're using the same techniques again that you use to carve transistors into these semiconductor devices. These MEMS units don't always work exactly like the mechanical device they're mimicking but achieve the same sensing result. Let me give you an example that was really striking to me and it's an example of a motion sensor. 

                                    I once saw a microscope [00:06:30] photograph of a shape that had been etched into a semiconductor chip. It looked like an empty swimming pool with a diving board over it. Now imagine someone kicking the device from below. The swimming pool would suddenly jerk up but the diving board would flex and actually its tip would get closer to the bottom of that swimming pool. That's pretty physically intuitive to you I'm sure. It turns out that by applying a voltage to that swimming pool and the diving board, you can measure the amount of flex, infer the type of motion that [00:07:00] caused that flex and do something useful like inflate an airbag to save your life.

                                    This type of motion sensor and others that measure magnetic field, the compass etc don't just make your smartphone smarter and your automobile better as we described, but also allowed the creation of new classes of machines such as the small consumer and commercial drones that are becoming fairly prevalent. Furthermore, you'll hear from Behrooz about how these modern sensors are starting to replace legacy versions in the industry.

                                    We hope you enjoyed this little explanation. [00:07:30] Let's get to it.

GTK:                             Tech. Cars. Machines. Subscribe here or at gtkpartners.com.

Ali Tabibian:                 Now that the timer has started we’re going to go live.  The only thing I’d say Behrooz is that the microphones are quite sensitive so if you tap the desk it picks it up really well.

Behrooz Abdi:               Let me sit back here a little bit that way I won’t get tempted to pound the table…

Ali Tabibian:                 …when you get excited!!

Ali Tabibian:                 We're here today with Mr. Behrooz Abdi, Chief Executive Officer of InvenSense, which is now a unit of the Japanese company TDK. Behrooz, thank you so much for taking the time, we really appreciate it.

Behrooz Abdi:               Well thank you very much for giving me the opportunity to talk about what's really a passion for me and really high degree of interest [00:08:30] for TDK. So great, thank you.

Ali Tabibian:                 Excellent. A lot of our listeners know that the podcast is an offshoot of our conferences which you were kind enough twice to present at. I think last year we had a scheduling conflict with your presence. One of the things you really did for conference attendees in those conferences is really lay out what the world of sensors are doing for the magic of IoT and industrial IoT to really happen [00:09:00] and happen in form factors and at price points that really make a lot of these devices possible to be consumed at volume.

                                    Your history in the semiconductor industry goes back for quite a long time. And essentially, you've been there since the beginning, you've seen how the influence of technology especially hardware technology and semiconductor technologies have sort of enabled communications and computing products, then started going into the world of personal communication products mainly phones and smartphones and are now spilling [00:09:30] out from there. Other devices and with TDK you also have their automotive business, sensing business reporting to you. Behrooz, tell us where you started and how you got here?

Behrooz Abdi:               Well, I've been in the semiconductor industry as you mentioned for a long time. Probably a little bit over 32 years. I started my career at Motorola semiconductor as an engineer analog and what later became mixed signal design engineer. I rose up [00:10:00] through the ranks during the early phases of growth in the disk drive and then later on with cellular phones. And as you know, Motorola was really big into cell phones and we actually designed some of the chips. I personally design a couple of the chips for the first what's called a Motorola StarTac, the first flip phone. And through the years I moved into operations marketing and general management. After 18 years with Motorola, [00:10:30] I was recruited into Qualcomm where I was the general manager for the chip business, what was called Qualcomm CDMA technologies. 

                                    When I was there, I'd learned about the power of platform and software and I got really exposed beyond analog and RF, which I had learned at Motorola, I got exposed into digital chips. So I got really into the Moore's law and how it's really transforming [00:11:00] communication. And after Qualcomm, I joined a small company, private company called RMI. At RMI, we were doing multi-core processors. I was the CEO there. These multi-core processors would go into some consumer devices but mostly really targeted towards data centers. So this is the early phase of multi-cores and multi-core processors were coming in to really compensate [00:11:30] for the fact that Moore's Law, if you had a single processor, the more you added processing power to it, the hotter it got and heat was getting out of hand and we had to deal with that.

                                    So, we sold RMI to a public company called the NetLogic. It was one of our technology partners, and we sold it for all stock deal. About two years later, we sold NetLogic to Broadcom for about [00:12:00] somewhere between $3.74 billion, that was at the time the largest exit in semiconductor in a decade, which I know these days, $4 billion is chump change for a lot of the companies. 

                                    After that, I took some time off to figure out what I wanted to do in my career, whether I wanted to stay in semiconductor or not. But I realized that semiconductor is actually very, very exciting. In fact, sensors are going to be where RF and analog was 20 years [00:12:30] ago. It's going to be a multi decade effort. Since I had led the strategic investment into InvenSense a decade before, when I was a Qualcomm, I was asked by the board to join the board and eventually became the CEO. 

                                    So that's my background and a little bit about the company. 

Ali Tabibian:                 Let me ask you, you said you could sense that, no pun intended, sensors are where RF was decades ago. [00:13:00] What do you mean by that? What was the intuition there? 

Behrooz Abdi:               Well, when I was dealing with RF in the 90s, I could see the complexity of multiple bands and multiple modes coming in. You started with a GSM phone and an analog phone and a CDMA phone, and pretty soon it was like well, analog and GSM together or GSM and CDMA together. Pretty soon, you saw technologies like Bluetooth come in or bands, you had to deal with different [00:13:30] bands because you couldn't get at one contiguous set of bands together in any company, in any country. So you had to deal with multiple bands. So the RF was getting complex and you couldn't have all these RF radios, all these radios in a phone in a form factor that was usable. So, RF miniaturization, multi-band, multi-mode technology had to evolve.

                                    That's what happened in 90s. I was very focused on what I call [00:14:00] the RF miniaturization. Make them small, make them cheap and they will come. Same thing now, when I look at a sensor, you look at these phones that have one or two or three sensors and I look at any phones or any device and I look at it and say it shouldn't be one or two sensors. It's like you can't have a human being with an eye or another human being with an ear. You got to have all these multiple sensors. So multisensory experience is extremely crucial [00:14:30] to any product, any device in the future. 

                                    Same thing I saw with processing. You would have either single core that's very powerful or you could start with a lot of very low power but fairly efficient, maybe less capability cores but you have a swarm of processing cores and you can have something that's much more power efficient, much more higher performance than a single core.

Ali Tabibian:                 What you're basically saying is in terms of aggregation [00:15:00] and integration and really driving the efficiency of how these sensors are built, you saw enormous runway, decades of runway. 

Behrooz Abdi:               And sensors that we hadn't thought of and really sensor fusion. How do you take multiple sensors together and form a decision? That's really what sensors are doing, it's helping you form a decision by sensing the ambient warmth, whether it's a mobile phone, in an industrial environment [00:15:30] or at home.

Ali Tabibian:                 Your work that brought you to InvenSense is about a specific type of sensors, ones based on MEMS technology.

Behrooz Abdi:               That is correct.

Ali Tabibian:                 What is that?

Behrooz Abdi:               In Layman's term, MEMS is a way to shrink mechanical devices through silicon. It's really miniaturizing mechanical devices. You could imagine any mechanical lever or a motor or anything that moves and shrinking it to the size of [00:16:00] a piece of silicon chip. So it's a processes using silicon technology, semiconductor technology to shrink mechanical devices. It started really in the 70s and 80s. One of the motivations for it back then was to utilize old fabs that were no longer keeping up with Moore's law. So people start to think about using them for shrinking mechanical devices. One of the first applications were accelerometers [00:16:30] for airbag deployment. It's a device that basically, a single axis of an accelerometer lever where if the car changes speed too rapidly, the mechanical device moves a little bit slower than the movement of the car.

Ali Tabibian:                 That's an excellent description. Now bookend the history of InvenSense for us as a way of showing us where the industry's come. What was InvenSense's first product and where is [00:17:00] it now?

Behrooz Abdi:               As we talked about MEMS being a mechanical device, shrinking it has challenges of robustness, quality and reliability in addition to performance. So InvenSense was founded in 2003 to solve those problems and really enable a fabulous model for MEMS. Meaning that we could put together MEMS together with CMOS which is a mechanical device plus the [00:17:30] intelligence in CMOS on one piece of wafer. And make it robust such that it could go into just about any product and make it small and really solve those type of issues. 

                                    The first product was something, product called the gyroscope and gyroscope measures rotational motion of the movement for the device, whatever that device is. 

Ali Tabibian:                 Like missiles have guidance systems based on gyroscopes.

Behrooz Abdi:               Exactly, based on gyroscopes. [00:18:00] The first application, the time, the biggest market at least on the consumer for gyroscopes were digital cameras or actually high-end cameras, SLR cameras where you would put some kind of a gyroscope into the lens. You compensate for the movement of the hand and you get a sharper image. At that time, digital cameras were starting to come in at a rapid pace. Digital cameras wanted to have the same stabilization technology. They adopted gyroscope and they adopted the InvenSense gyroscopes. 

                                    [00:18:30] But really, the first big market came from gaming. At the time, Nintendo, a very innovative company in Japan was looking to put motion sensing into their controllers. They were really looking to innovate through motion sensing and they approached InvenSense and our product got adopted into the motion controller. When I came into the company in 2012 as the CEO, and a year before that, [00:19:00] as a board member, we were starting to gain adoption into the mobile phone. Again, for the same reason, the first adoption of motion sensors in mobile phones were for sensing the rotational motion mostly for gaming and then also for digital image stabilization. 

                                    So image civilization and gaming were the two applications of the mobile phone. Because of my background in the mobile ecosystem, I was asked to come in and I joined the board and later on [00:19:30] I took over as the CEO and we started to really work the ecosystem and grow in that market. Really, a mobile and consumer is very good for any semiconductor product to grow volume, it's fast volume. The challenges with that is that you go up and you go down just as much as you go up and it's a very volatile market. So eventually, you need to diversify. 

                                    So for the last four or five years, we've done a couple of things. We really focused [00:20:00] on diversification on other sensors with the same technology platform. We focused on diversifying markets into automotive, home, IoT, and more recently into industrial. The third thing that we've done is really build a system around, a system solution around sensors. Coming from a background at Qualcomm, we looked at a modem chip having hardware and software [00:20:30] and that modem basically together with the RF chips sensed and transformed the RF technology and RF signal into something that was understood by the digital circuits and eventually process a voice and data for communication. 

                                    Same thing with the sensors, you are communicating with the ambient world. So being able to sense processing communicate is really important. So we really focus on the software that could fuse different sensor data and [00:20:30] not only sense but really create a much more accurate sensor data that eventually gets processed either in the cloud or in the edge for making decisions around the device where there's motion or after some time we picked up audio, so microphones is another sensor, MEMS space sensor. Processing and fusing motion data with audio data and more recently with other sensors such as pressure sensor and barometric sensors [00:21:30] for all kinds of applications. 

Ali Tabibian:                 So Behrooz, just very quickly, a smartphone that we have here sitting on the table today, how many sensors does it have? Just give us a sense, gyros, magnetometers. 

Behrooz Abdi:               It could easily have a north of 20 sensors, 20 to 30 sensors. The different variations of the definition of the sensor. Some people, for example, call GPS a sensor but it's an RF chip. But really, if you look at [00:22:00] the sensors, motion sensing, audio sensing, there's multiple microphones and some of the high end phones are as many as four, maybe even five microphones. Barometric sensing, temperature sensing, touch sensor, fingerprint sensor. So we can go on and on. There are many different sensors in these phones. 

Ali Tabibian:                 There's that growth and integration that you saw happening in the industry. You touched on something really interesting, which is trying to incorporate a lot more smarts [00:22:30] right where the sensor is and how sensors report something more interesting than just data. One of the big things in IoT and industrial IoT is this debate of where, actually in any disperse architecture, where's the intelligence, where should the intelligence be. That's been a big back and forth. What are your insights and findings if you all over the last year of where that debate is either happening or being settled?

Behrooz Abdi:               I think it's really going to be a heterogeneous architecture where it's [00:23:00] going to be distributed throughout the chain. What you really need the cloud for is infinite storage and compute power. But you don't need infinite power and storage for a lot of decisions that you need to make right away. The downside of communicating everything to the cloud and making decisions with the cloud just like has been that same issue for the last few decades is the latency, not having coverage all the time and power of transmission. 

                                    [00:23:30] So we think that there are some really key decisions that the sensor can make immediately with very low latency right there at the edge within the sensor in fact. And then there are decisions that need to be made in a bigger processor and most of these devices either have a microcontroller or have an application processor. That applications processor can process some of that and then send the rest into the cloud for infinite storage and compute so that learning [00:24:00] can happen.

                                    So it would be heterogeneous. So when we look at that, we look at what is good enough for sensor fusion. For example, a lot of decisions, if you have multiple axis in a motion sensor, like these days are six or seven axis in any kind of motion sensor, you can make decision as to the device activity. You can really collect contexts. You don't need to send raw data as to x y, z axis [00:24:30] where they're moving. What you can say is that the context that this device is that it's moving at this speed and rotating at that speed and doing this. You can batch that information and send it every few seconds or every few minutes or in industrial maybe once a day.

                                    So you can be very efficient but having some intelligence in the sensor is incredibly important, even more important now than it was before. That's why we built this company around [00:25:00] sensor, sensor fusion and more intelligence. In fact, we've had products where we've had microcontrollers embedded into the sensor. So we think that that's an evolution that will continue to happen with more Moore's Law and as people look to make more efficient decisions on the sensors.

Ali Tabibian:                 Yeah interesting. We were talking earlier and you mentioned that your new owner TDK, actually the software technology within InvenSense or something, both the culture and [00:25:30] the know how is an important rationale for that acquisition.

Behrooz Abdi:               Oh, it is. Actually very excited about the fact that we have software. With software, you can also fuse different sensors of the same kind. For example, you could have them have what I call multi-core or multi-sensor. You could have either one real powerful, power hungry and large motion sensor or you could have a swarm of motion sensors in an industrial device or in a car [00:26:00] and collect data from different parts of the car or from different parts of the robot or different parts of any industrial unit and fuse all that together, and with software and DSP or the microcontroller, you can make a decision and much more accurate and much more power efficient decision than you would have made with that one really expensive large sensor.

                                    So that I think is going to be, that's going to be more important. And it allows you to have, to use more [00:26:30] or less off the shelf consumer sensors but at a different level. 

Ali Tabibian:                 So, Behrooz, that's a really interesting point that's been bandied about in the industry which is, look, obviously the sensors that were developed for the smartphone apply themselves really well to many consumer devices. Drones actually being a quasi-consumer device being a great example of that. And there's always been the thought that the same technology could be applied to an industrial environment, maybe [00:27:00] even an automotive environment. To what extent has that all come true?

Behrooz Abdi:               That is coming true in a big way. In fact, we think that just like any other industry, I think that consumerization eventually comes to every industry. With the technology improvements down the cost curve and the power curve, what happens is that the capability of the sensors continues to go up and you can use [00:27:30] multitude of sensors. So to get a probability of a decision based on one really accurate sensor is no better and actually could be worse than using 10 sensors and different types of sensors to collect them and get the sensor fusion to make a much more accurate decision with lower latency and lower weight. Even in a car for ADAS or for an industrial robot. 

                                    So, [00:28:00] what consumerization, what the consumer market does, drive volume, drive cost curve, drive low power. As you know, it's brutally competitive. You have to innovate every year. That is very hard to do if you're only focused on the industrial market. We see that with companies, traditional companies that were only focused on the automotive and industrial market, what we see is that they're roadmaps running into a dead end. They can't innovate anymore. [00:28:30] Now they have to figure out how to shrink these sensors and make them lower power but their technology is so outdated. So we're coming in from behind and from the below. We're the low end disruptors that continue to improve every year and improve performance power and cost.

Ali Tabibian:                 That's very, very, very fascinating. Behrooz, I have two more questions. One of them which leads off directly what you just said. That convergence that you're describing basically to some extent is a story of the acquisition of TDK of [00:29:00] InvenSense. Take us into a little bit of the rationale for that transaction and how you now sit in the space versus other players in the space.

Behrooz Abdi:               Very good question. We felt that TDK would be a great partner for us from the very beginning. There might be, there's of course the financial part of the transaction but really for long term merger and integration and for this to be successful, you need to be with a company [00:29:00] that really truly values your technology and vice versa. What we valued from TDK is their expertise in material science. They've been a material company since the beginning for 80 years. Whether it's magnetic material, whether it's other technologies that they're innovating at the very nanoscopic level, we felt that for sensors to grow and continue to scale, you need access to more advanced materials. So from that angle, it was really good for us. 

                                    [00:30:00] They were already doing sensors dealing with some sensors. Then they had a great channel into the industrial and automotive market which we utilize. So that value came to us from that end. What we brought to them was really the system expertise, the softer expertise. It's really hard to describe the expertise at the system level until you have it and you see what it does for you. First of all, when you can put [00:30:30] a system together, you know where the weaknesses and strengths of the use case are. If your hardware is not perfect, you can fix it through some softer angle. You can always put some other component into the product and use software to improve your sensor fusion. You can really compensate for some of the deficiency in the hardware. 

                                    We see that and they see that now. And in fact, we've actually opened up new markets and customers for them where they had magnetic [00:31:00] sensors or some other sensors in the temperature and pressure and with software, we've been able to help them get access to new use cases and understand why customers drive certain specs. And that is very, very powerful. From that angle, it's been really great. On the industrial and automotive side, being able to visit customers with total system solution has been really powerful. With IoT is the same thing. They have both on the consumer IoT and on the industrial [00:31:30] IoT even more, they have really great connections. And that's what we're able to really pull together at the system level and take to their customer or to now our customers at a broader sense. 

Ali Tabibian:                 Where does TDK InvenSense sit in its space now? Who are the people who are adjacent to you?

Behrooz Abdi:               In terms of competition, there's the component guys. There's a typical Bosch, ST Micro, Murata and then Freescale, NXP [00:32:00] at different levels with different sensors, they're different competitors. When we look at our portfolio, we don't think that there's anybody who has the broad spectrum of the sensors that we have now. We started with motion sensors, then we got into microphones. We actually acquired the microphone business from analog devices several years ago and we've been able to advance it to the next level. More recently, we have gotten into barometric pressure sensor which allows us [00:32:30] to sense what floor you're on in an indoor navigation use case or for health and fitness. Calorie count, how many steps you're going up and down the stairs. 

                                    Then we have, more recently, we have gotten into ultrasonic sensors. There's so many applications for ultrasonic. One of the first products that we started with organically at InvenSense was for the fingerprint sensing using ultrasonic technology. And again, together with CMOS and MEMS, we were [00:33:00] able to do some really amazing things with the architecture. We very recently in the last couple months, we acquired a company in Berkeley called Chirp Microsystems. Their technology is used to detect gestures in several meters and objects and shapes. So very, very useful for future of object detection, object recognition, collision avoidance and robots and also in drones. So [00:33:00] many, many different applications and VR applications, which is one of the first applications for that ultrasonic technology. We can fuse motion and ultrasonic technology to very accurately locate the controller position in the VR mode and really solve some very innovative challenges and create a much more immersive experience with VR. 

                                    So sensor fusion, again, is something [00:34:00] that's very important for our company as now we've grown into a system solution company. 

Ali Tabibian:                 If you look past the time period of your typical development pipeline, if you sort of go past your development pipeline, what are your predictions or prognostications for the space. We're not going to hold you to it. I'm not going to come back in three or four years and say, hey, wait a minute, here's what you said.

Behrooz Abdi:               One of things is that, especially in the industrial and automotive, you will see a lot more [00:34:30] sensors and what I call a swarm of sensors that will be coming in together with the power of the cloud and the architecture. On the CMOS side you'll see that and with AI, you'll see a lot of innovation in that area. 

                                    One thing that I've always said around Moore's Law is don't bet against Moore's law. It's been continuing to enable smaller and more powerful devices. But the thing that I've been really [00:35:00] surprised about is that with AI is actually, the capability has been actually growing faster in the last five to 10 years. So you will have products with AI sensing and a lot more powerful processing in every device. What that means, especially as privacy becomes a bigger issue, you don't have to go to the cloud for a lot of the AI applications you can actually do AI on a chip, not necessarily [00:35:30] the sensor but at the edge. Together with a bunch of sensors, you can make really fast decisions and be private and not share that data with anybody else. You go to the cloud for learning. That's all. 

                                    So again, whether it's a personal device or an industrial device where security is even more important, you can do a lot more interesting things at that level. That's I think one of the things that we're going to see as sensors developed into lower costs and lower power.

Ali Tabibian:                 That's fabulous. [00:36:00] Thank you for that description. I hadn't really thought about what I think Apple calls differential privacy. Keeping things local as much as possible. About that being driven by a lot of the, been made possible by a lot of the developments and the learning not being lost in the process. 

Behrooz Abdi:               That's something that you can have the history, you can have the privacy and security have it all together. And when you think about IoT, the way I describe IoT is where are you, what are you doing, who are you [00:36:30] and how do you feel? And that's all about sensors. If you look at what IoT enabled applications like Uber or DoorDash, when you click on an app and you order food or order a ride, it knows that it is you and it knows where you are. That, now going into all kinds of devices, that capability and not sharing that data with their cloud, with their servers, I think it's going to be the next big [00:37:00] thing. I think sensors together with AI will enable a lot more of that.

Ali Tabibian:                 Great. Behrooz, thank you very much. I know we have to get you going for an important appointment. Is there anything else in the next 60 seconds or so that we should talk about?

Behrooz Abdi:               I think we covered most of it. I think it's really important to the audience to be energized about semiconductor and realize that there's a lot more innovation that's coming and we just seeing the start. [00:37:30] With sensors, you're going to see a lot of that innovation. Every device around you, whether you're at home, in the car industry, at the enterprise. One of the areas that long term I'm very excited about to see sensors come into is the medical field where it will really make people not just more productive but enable healthier lifestyles.

Ali Tabibian:                 Hey man, I certainly need to live more healthfully. Maybe if I could, do you have any sensors I could drink on my way out?

Behrooz Abdi:               [00:38:00] We'll pour some for you. It’s experimental, though. 

Ali Tabibian:                 Only the top shelf stuff please. Behrooz, thank you very much. 

Behrooz Abdi:               All right, well, thank you for your time.

Ali Tabibian:                 Thank you.

GTK:                             Tech. Cars. Machines. Subscribe here or at gtkpartners.com.

Ali Tabibian:                 Welcome, welcome, welcome everyone to this episode of Tech. Cars. Machines. I'm your host Ali Tabibian. We've crossed the thousand listener mark and we're glad you're all here. If you look at our website, www.gtkpartners.com or listen to our introductory episode of this podcast series, you'll know that we think Tech. Cars. Machines is about the intersection of three major areas [00:00:30] of technology and the world of course of cars and machines. The three areas of technology are sensors, connectivity, and software. 

                                    Today, we're going to have our first episode on sensors. As usual, we bring you the very, very best people to understand what's going on in the particular world we're about to talk about. Our guest today is Behrooz Abdi, currently the chief executive officer of TDK InvenSense. He's a very long tenured semiconductor executive, has been a chief [00:01:00] executive a couple times. Most recently and currently as chief executive of TDK's InvenSense unit.

                                    Behrooz was originally an investor and board member in InvenSense when it was venture backed. He became its chief executive when InvenSense was a public company and through a lengthy process that ended or came to fruition I should say in late 2016, he orchestrated the sale of InvenSense for $1.3 billion to [00:01:30] TDK Japanese multinational. At the time, InvenSense probably had half the world's market share in high end smartphones like iPhones and Samsung galaxies as far as some of the sensors in those devices were concerned and it had pretty good footprints elsewhere. Their new owner, TDK has a strong position in automotive sensors. That unit also now reports to Behrooz in the combined TDK InvenSense unit. 

                                    A quick note about something that Behrooz mentions in this interview. When describing [00:02:00] the bright future of sensing, he uses the word integration. Given his background at Qualcomm, there's more meaning to the word integration when it comes from Behrooz. You see, for years, the fortunes of Qualcomm, a successful and storied name in cell phones and smartphones especially in the communications technologies like CDMA, Qualcomm's success was driven by that one word, integration. Meaning this, every generation of cell phone required integrating greater numbers [00:02:30] of antennas and frequencies so the phones could work in a greater number of locations, in different countries and utilize multiple cell towers for maximum benefit. It seems like Behrooz sees the future of sensing to be at the same spot. 

                                    By the way, Behrooz has a very good sense of humor. It didn't really come out in the formal parts of the interview so we left in a few seconds of the informal parts of the taping so you can see what we mean. To go directly to the interview, go to minute [00:03:00] seven and a half of this episode. Otherwise, I'm going to spend the next few minutes explaining a few things about why sensors are a big deal. 

                                    A sensor is a device that measures some characteristic of the physical environment such as temperature or a relationship with that environment such as am I accelerating or coming to a stop. You're all familiar with the humble mercury thermometer. It was invented about 200 years ago by Mr. Daniel Fahrenheit in Amsterdam. To put a fancy phrase to [00:03:30] it, the thermometer is a temperature sensor. So there's an example of a sensor, a very simple sensor which has been around for centuries. Light sensors are pretty important too. Beginning in the 1840s, the most important light sensor of all was silver highlight film, the kind that made the Kodak company famous and made photography ubiquitous. 

                                    Another example, sensing motion is helpful in ships, aircrafts and rockets. Gyroscopes, which are devices originally with parts that spun like a top [00:04:00] are in this class of sensor. About 150 years ago is when their technology started becoming useful commercially. Early very precise versions that went into rockets and missiles were the size of really large balls, you know, the kind that people bounce around on in the gym to pretend that they're exercising. The big deal is that people have figured out how to make an increasing number of sensor types with semiconductor manufacturing techniques. These are the same techniques that deliver wondrous results in your smart phones and computers. [00:04:30] This means you are harnessing hundreds of billions of dollars of investment that the semi industry has made in miniaturization, precision, robustness, and critically in manufacturing chips that are easily integrated with electronic devices.

                                    Let's go back to our light sensor. Can you imagine trying to squeeze 35 millimeter Kodak film into your smartphone? The reason you don't have to do it is that a particular type of semiconductor chip called a charged couple device or CCD [00:05:00] was developed beginning in the late 1960s. A CCD is basically a matrix of tiny sensors each called a pixel, which detects light just like the Kodak film. Unlike film, obviously, the light is immediately converted to an electrical value for processing and it can be erased and reused repeatedly. When Apple says the iPhone X has a 12 megapixel camera, it means the CCD chip inside has 12 million tiny sensors in it. The same compact and cheap chips [00:05:30] enable your automobile to have multiple cameras installed including those that contribute to self-driving.

                                    Now film is a flat simple mechanism. Can we make a lot more complex devices in semi technology? Can we make a gyroscope and mimic the output of its spinning parts using semi technology? I don't know, Can we? 

Barack Obama:             Yes, we can. Yes, we can. Yes, we can.

Ali Tabibian:                 About 30 years ago, another big deal happened [00:06:00] in sensing, and that's called micro-electrical mechanical systems or MEMS. Basically, you're carving complex patterns on silicon to mimic physical devices. You're using the same techniques again that you use to carve transistors into these semiconductor devices. These MEMS units don't always work exactly like the mechanical device they're mimicking but achieve the same sensing result. Let me give you an example that was really striking to me and it's an example of a motion sensor. 

                                    I once saw a microscope [00:06:30] photograph of a shape that had been etched into a semiconductor chip. It looked like an empty swimming pool with a diving board over it. Now imagine someone kicking the device from below. The swimming pool would suddenly jerk up but the diving board would flex and actually its tip would get closer to the bottom of that swimming pool. That's pretty physically intuitive to you I'm sure. It turns out that by applying a voltage to that swimming pool and the diving board, you can measure the amount of flex, infer the type of motion that [00:07:00] caused that flex and do something useful like inflate an airbag to save your life.

                                    This type of motion sensor and others that measure magnetic field, the compass etc don't just make your smartphone smarter and your automobile better as we described, but also allowed the creation of new classes of machines such as the small consumer and commercial drones that are becoming fairly prevalent. Furthermore, you'll hear from Behrooz about how these modern sensors are starting to replace legacy versions in the industry.

                                    We hope you enjoyed this little explanation. [00:07:30] Let's get to it.

GTK:                             Tech. Cars. Machines. Subscribe here or at gtkpartners.com.

Ali Tabibian:                 Now that the timer has started we’re going to go live.  The only thing I’d say Behrooz is that the microphones are quite sensitive so if you tap the desk it picks it up really well.

Behrooz Abdi:               Let me sit back here a little bit that way I won’t get tempted to pound the table…

Ali Tabibian:                 …when you get excited!!

Ali Tabibian:                 We're here today with Mr. Behrooz Abdi, Chief Executive Officer of InvenSense, which is now a unit of the Japanese company TDK. Behrooz, thank you so much for taking the time, we really appreciate it.

Behrooz Abdi:               Well thank you very much for giving me the opportunity to talk about what's really a passion for me and really high degree of interest [00:08:30] for TDK. So great, thank you.

Ali Tabibian:                 Excellent. A lot of our listeners know that the podcast is an offshoot of our conferences which you were kind enough twice to present at. I think last year we had a scheduling conflict with your presence. One of the things you really did for conference attendees in those conferences is really lay out what the world of sensors are doing for the magic of IoT and industrial IoT to really happen [00:09:00] and happen in form factors and at price points that really make a lot of these devices possible to be consumed at volume.

                                    Your history in the semiconductor industry goes back for quite a long time. And essentially, you've been there since the beginning, you've seen how the influence of technology especially hardware technology and semiconductor technologies have sort of enabled communications and computing products, then started going into the world of personal communication products mainly phones and smartphones and are now spilling [00:09:30] out from there. Other devices and with TDK you also have their automotive business, sensing business reporting to you. Behrooz, tell us where you started and how you got here?

Behrooz Abdi:               Well, I've been in the semiconductor industry as you mentioned for a long time. Probably a little bit over 32 years. I started my career at Motorola semiconductor as an engineer analog and what later became mixed signal design engineer. I rose up [00:10:00] through the ranks during the early phases of growth in the disk drive and then later on with cellular phones. And as you know, Motorola was really big into cell phones and we actually designed some of the chips. I personally design a couple of the chips for the first what's called a Motorola StarTac, the first flip phone. And through the years I moved into operations marketing and general management. After 18 years with Motorola, [00:10:30] I was recruited into Qualcomm where I was the general manager for the chip business, what was called Qualcomm CDMA technologies. 

                                    When I was there, I'd learned about the power of platform and software and I got really exposed beyond analog and RF, which I had learned at Motorola, I got exposed into digital chips. So I got really into the Moore's law and how it's really transforming [00:11:00] communication. And after Qualcomm, I joined a small company, private company called RMI. At RMI, we were doing multi-core processors. I was the CEO there. These multi-core processors would go into some consumer devices but mostly really targeted towards data centers. So this is the early phase of multi-cores and multi-core processors were coming in to really compensate [00:11:30] for the fact that Moore's Law, if you had a single processor, the more you added processing power to it, the hotter it got and heat was getting out of hand and we had to deal with that.

                                    So, we sold RMI to a public company called the NetLogic. It was one of our technology partners, and we sold it for all stock deal. About two years later, we sold NetLogic to Broadcom for about [00:12:00] somewhere between $3.74 billion, that was at the time the largest exit in semiconductor in a decade, which I know these days, $4 billion is chump change for a lot of the companies. 

                                    After that, I took some time off to figure out what I wanted to do in my career, whether I wanted to stay in semiconductor or not. But I realized that semiconductor is actually very, very exciting. In fact, sensors are going to be where RF and analog was 20 years [00:12:30] ago. It's going to be a multi decade effort. Since I had led the strategic investment into InvenSense a decade before, when I was a Qualcomm, I was asked by the board to join the board and eventually became the CEO. 

                                    So that's my background and a little bit about the company. 

Ali Tabibian:                 Let me ask you, you said you could sense that, no pun intended, sensors are where RF was decades ago. [00:13:00] What do you mean by that? What was the intuition there? 

Behrooz Abdi:               Well, when I was dealing with RF in the 90s, I could see the complexity of multiple bands and multiple modes coming in. You started with a GSM phone and an analog phone and a CDMA phone, and pretty soon it was like well, analog and GSM together or GSM and CDMA together. Pretty soon, you saw technologies like Bluetooth come in or bands, you had to deal with different [00:13:30] bands because you couldn't get at one contiguous set of bands together in any company, in any country. So you had to deal with multiple bands. So the RF was getting complex and you couldn't have all these RF radios, all these radios in a phone in a form factor that was usable. So, RF miniaturization, multi-band, multi-mode technology had to evolve.

                                    That's what happened in 90s. I was very focused on what I call [00:14:00] the RF miniaturization. Make them small, make them cheap and they will come. Same thing now, when I look at a sensor, you look at these phones that have one or two or three sensors and I look at any phones or any device and I look at it and say it shouldn't be one or two sensors. It's like you can't have a human being with an eye or another human being with an ear. You got to have all these multiple sensors. So multisensory experience is extremely crucial [00:14:30] to any product, any device in the future. 

                                    Same thing I saw with processing. You would have either single core that's very powerful or you could start with a lot of very low power but fairly efficient, maybe less capability cores but you have a swarm of processing cores and you can have something that's much more power efficient, much more higher performance than a single core.

Ali Tabibian:                 What you're basically saying is in terms of aggregation [00:15:00] and integration and really driving the efficiency of how these sensors are built, you saw enormous runway, decades of runway. 

Behrooz Abdi:               And sensors that we hadn't thought of and really sensor fusion. How do you take multiple sensors together and form a decision? That's really what sensors are doing, it's helping you form a decision by sensing the ambient warmth, whether it's a mobile phone, in an industrial environment [00:15:30] or at home.

Ali Tabibian:                 Your work that brought you to InvenSense is about a specific type of sensors, ones based on MEMS technology.

Behrooz Abdi:               That is correct.

Ali Tabibian:                 What is that?

Behrooz Abdi:               In Layman's term, MEMS is a way to shrink mechanical devices through silicon. It's really miniaturizing mechanical devices. You could imagine any mechanical lever or a motor or anything that moves and shrinking it to the size of [00:16:00] a piece of silicon chip. So it's a processes using silicon technology, semiconductor technology to shrink mechanical devices. It started really in the 70s and 80s. One of the motivations for it back then was to utilize old fabs that were no longer keeping up with Moore's law. So people start to think about using them for shrinking mechanical devices. One of the first applications were accelerometers [00:16:30] for airbag deployment. It's a device that basically, a single axis of an accelerometer lever where if the car changes speed too rapidly, the mechanical device moves a little bit slower than the movement of the car.

Ali Tabibian:                 That's an excellent description. Now bookend the history of InvenSense for us as a way of showing us where the industry's come. What was InvenSense's first product and where is [00:17:00] it now?

Behrooz Abdi:               As we talked about MEMS being a mechanical device, shrinking it has challenges of robustness, quality and reliability in addition to performance. So InvenSense was founded in 2003 to solve those problems and really enable a fabulous model for MEMS. Meaning that we could put together MEMS together with CMOS which is a mechanical device plus the [00:17:30] intelligence in CMOS on one piece of wafer. And make it robust such that it could go into just about any product and make it small and really solve those type of issues. 

                                    The first product was something, product called the gyroscope and gyroscope measures rotational motion of the movement for the device, whatever that device is. 

Ali Tabibian:                 Like missiles have guidance systems based on gyroscopes.

Behrooz Abdi:               Exactly, based on gyroscopes. [00:18:00] The first application, the time, the biggest market at least on the consumer for gyroscopes were digital cameras or actually high-end cameras, SLR cameras where you would put some kind of a gyroscope into the lens. You compensate for the movement of the hand and you get a sharper image. At that time, digital cameras were starting to come in at a rapid pace. Digital cameras wanted to have the same stabilization technology. They adopted gyroscope and they adopted the InvenSense gyroscopes. 

                                    [00:18:30] But really, the first big market came from gaming. At the time, Nintendo, a very innovative company in Japan was looking to put motion sensing into their controllers. They were really looking to innovate through motion sensing and they approached InvenSense and our product got adopted into the motion controller. When I came into the company in 2012 as the CEO, and a year before that, [00:19:00] as a board member, we were starting to gain adoption into the mobile phone. Again, for the same reason, the first adoption of motion sensors in mobile phones were for sensing the rotational motion mostly for gaming and then also for digital image stabilization. 

                                    So image civilization and gaming were the two applications of the mobile phone. Because of my background in the mobile ecosystem, I was asked to come in and I joined the board and later on [00:19:30] I took over as the CEO and we started to really work the ecosystem and grow in that market. Really, a mobile and consumer is very good for any semiconductor product to grow volume, it's fast volume. The challenges with that is that you go up and you go down just as much as you go up and it's a very volatile market. So eventually, you need to diversify. 

                                    So for the last four or five years, we've done a couple of things. We really focused [00:20:00] on diversification on other sensors with the same technology platform. We focused on diversifying markets into automotive, home, IoT, and more recently into industrial. The third thing that we've done is really build a system around, a system solution around sensors. Coming from a background at Qualcomm, we looked at a modem chip having hardware and software [00:20:30] and that modem basically together with the RF chips sensed and transformed the RF technology and RF signal into something that was understood by the digital circuits and eventually process a voice and data for communication. 

                                    Same thing with the sensors, you are communicating with the ambient world. So being able to sense processing communicate is really important. So we really focus on the software that could fuse different sensor data and [00:20:30] not only sense but really create a much more accurate sensor data that eventually gets processed either in the cloud or in the edge for making decisions around the device where there's motion or after some time we picked up audio, so microphones is another sensor, MEMS space sensor. Processing and fusing motion data with audio data and more recently with other sensors such as pressure sensor and barometric sensors [00:21:30] for all kinds of applications. 

Ali Tabibian:                 So Behrooz, just very quickly, a smartphone that we have here sitting on the table today, how many sensors does it have? Just give us a sense, gyros, magnetometers. 

Behrooz Abdi:               It could easily have a north of 20 sensors, 20 to 30 sensors. The different variations of the definition of the sensor. Some people, for example, call GPS a sensor but it's an RF chip. But really, if you look at [00:22:00] the sensors, motion sensing, audio sensing, there's multiple microphones and some of the high end phones are as many as four, maybe even five microphones. Barometric sensing, temperature sensing, touch sensor, fingerprint sensor. So we can go on and on. There are many different sensors in these phones. 

Ali Tabibian:                 There's that growth and integration that you saw happening in the industry. You touched on something really interesting, which is trying to incorporate a lot more smarts [00:22:30] right where the sensor is and how sensors report something more interesting than just data. One of the big things in IoT and industrial IoT is this debate of where, actually in any disperse architecture, where's the intelligence, where should the intelligence be. That's been a big back and forth. What are your insights and findings if you all over the last year of where that debate is either happening or being settled?

Behrooz Abdi:               I think it's really going to be a heterogeneous architecture where it's [00:23:00] going to be distributed throughout the chain. What you really need the cloud for is infinite storage and compute power. But you don't need infinite power and storage for a lot of decisions that you need to make right away. The downside of communicating everything to the cloud and making decisions with the cloud just like has been that same issue for the last few decades is the latency, not having coverage all the time and power of transmission. 

                                    [00:23:30] So we think that there are some really key decisions that the sensor can make immediately with very low latency right there at the edge within the sensor in fact. And then there are decisions that need to be made in a bigger processor and most of these devices either have a microcontroller or have an application processor. That applications processor can process some of that and then send the rest into the cloud for infinite storage and compute so that learning [00:24:00] can happen.

                                    So it would be heterogeneous. So when we look at that, we look at what is good enough for sensor fusion. For example, a lot of decisions, if you have multiple axis in a motion sensor, like these days are six or seven axis in any kind of motion sensor, you can make decision as to the device activity. You can really collect contexts. You don't need to send raw data as to x y, z axis [00:24:30] where they're moving. What you can say is that the context that this device is that it's moving at this speed and rotating at that speed and doing this. You can batch that information and send it every few seconds or every few minutes or in industrial maybe once a day.

                                    So you can be very efficient but having some intelligence in the sensor is incredibly important, even more important now than it was before. That's why we built this company around [00:25:00] sensor, sensor fusion and more intelligence. In fact, we've had products where we've had microcontrollers embedded into the sensor. So we think that that's an evolution that will continue to happen with more Moore's Law and as people look to make more efficient decisions on the sensors.

Ali Tabibian:                 Yeah interesting. We were talking earlier and you mentioned that your new owner TDK, actually the software technology within InvenSense or something, both the culture and [00:25:30] the know how is an important rationale for that acquisition.

Behrooz Abdi:               Oh, it is. Actually very excited about the fact that we have software. With software, you can also fuse different sensors of the same kind. For example, you could have them have what I call multi-core or multi-sensor. You could have either one real powerful, power hungry and large motion sensor or you could have a swarm of motion sensors in an industrial device or in a car [00:26:00] and collect data from different parts of the car or from different parts of the robot or different parts of any industrial unit and fuse all that together, and with software and DSP or the microcontroller, you can make a decision and much more accurate and much more power efficient decision than you would have made with that one really expensive large sensor.

                                    So that I think is going to be, that's going to be more important. And it allows you to have, to use more [00:26:30] or less off the shelf consumer sensors but at a different level. 

Ali Tabibian:                 So, Behrooz, that's a really interesting point that's been bandied about in the industry which is, look, obviously the sensors that were developed for the smartphone apply themselves really well to many consumer devices. Drones actually being a quasi-consumer device being a great example of that. And there's always been the thought that the same technology could be applied to an industrial environment, maybe [00:27:00] even an automotive environment. To what extent has that all come true?

Behrooz Abdi:               That is coming true in a big way. In fact, we think that just like any other industry, I think that consumerization eventually comes to every industry. With the technology improvements down the cost curve and the power curve, what happens is that the capability of the sensors continues to go up and you can use [00:27:30] multitude of sensors. So to get a probability of a decision based on one really accurate sensor is no better and actually could be worse than using 10 sensors and different types of sensors to collect them and get the sensor fusion to make a much more accurate decision with lower latency and lower weight. Even in a car for ADAS or for an industrial robot. 

                                    So, [00:28:00] what consumerization, what the consumer market does, drive volume, drive cost curve, drive low power. As you know, it's brutally competitive. You have to innovate every year. That is very hard to do if you're only focused on the industrial market. We see that with companies, traditional companies that were only focused on the automotive and industrial market, what we see is that they're roadmaps running into a dead end. They can't innovate anymore. [00:28:30] Now they have to figure out how to shrink these sensors and make them lower power but their technology is so outdated. So we're coming in from behind and from the below. We're the low end disruptors that continue to improve every year and improve performance power and cost.

Ali Tabibian:                 That's very, very, very fascinating. Behrooz, I have two more questions. One of them which leads off directly what you just said. That convergence that you're describing basically to some extent is a story of the acquisition of TDK of [00:29:00] InvenSense. Take us into a little bit of the rationale for that transaction and how you now sit in the space versus other players in the space.

Behrooz Abdi:               Very good question. We felt that TDK would be a great partner for us from the very beginning. There might be, there's of course the financial part of the transaction but really for long term merger and integration and for this to be successful, you need to be with a company [00:29:00] that really truly values your technology and vice versa. What we valued from TDK is their expertise in material science. They've been a material company since the beginning for 80 years. Whether it's magnetic material, whether it's other technologies that they're innovating at the very nanoscopic level, we felt that for sensors to grow and continue to scale, you need access to more advanced materials. So from that angle, it was really good for us. 

                                    [00:30:00] They were already doing sensors dealing with some sensors. Then they had a great channel into the industrial and automotive market which we utilize. So that value came to us from that end. What we brought to them was really the system expertise, the softer expertise. It's really hard to describe the expertise at the system level until you have it and you see what it does for you. First of all, when you can put [00:30:30] a system together, you know where the weaknesses and strengths of the use case are. If your hardware is not perfect, you can fix it through some softer angle. You can always put some other component into the product and use software to improve your sensor fusion. You can really compensate for some of the deficiency in the hardware. 

                                    We see that and they see that now. And in fact, we've actually opened up new markets and customers for them where they had magnetic [00:31:00] sensors or some other sensors in the temperature and pressure and with software, we've been able to help them get access to new use cases and understand why customers drive certain specs. And that is very, very powerful. From that angle, it's been really great. On the industrial and automotive side, being able to visit customers with total system solution has been really powerful. With IoT is the same thing. They have both on the consumer IoT and on the industrial [00:31:30] IoT even more, they have really great connections. And that's what we're able to really pull together at the system level and take to their customer or to now our customers at a broader sense. 

Ali Tabibian:                 Where does TDK InvenSense sit in its space now? Who are the people who are adjacent to you?

Behrooz Abdi:               In terms of competition, there's the component guys. There's a typical Bosch, ST Micro, Murata and then Freescale, NXP [00:32:00] at different levels with different sensors, they're different competitors. When we look at our portfolio, we don't think that there's anybody who has the broad spectrum of the sensors that we have now. We started with motion sensors, then we got into microphones. We actually acquired the microphone business from analog devices several years ago and we've been able to advance it to the next level. More recently, we have gotten into barometric pressure sensor which allows us [00:32:30] to sense what floor you're on in an indoor navigation use case or for health and fitness. Calorie count, how many steps you're going up and down the stairs. 

                                    Then we have, more recently, we have gotten into ultrasonic sensors. There's so many applications for ultrasonic. One of the first products that we started with organically at InvenSense was for the fingerprint sensing using ultrasonic technology. And again, together with CMOS and MEMS, we were [00:33:00] able to do some really amazing things with the architecture. We very recently in the last couple months, we acquired a company in Berkeley called Chirp Microsystems. Their technology is used to detect gestures in several meters and objects and shapes. So very, very useful for future of object detection, object recognition, collision avoidance and robots and also in drones. So [00:33:00] many, many different applications and VR applications, which is one of the first applications for that ultrasonic technology. We can fuse motion and ultrasonic technology to very accurately locate the controller position in the VR mode and really solve some very innovative challenges and create a much more immersive experience with VR. 

                                    So sensor fusion, again, is something [00:34:00] that's very important for our company as now we've grown into a system solution company. 

Ali Tabibian:                 If you look past the time period of your typical development pipeline, if you sort of go past your development pipeline, what are your predictions or prognostications for the space. We're not going to hold you to it. I'm not going to come back in three or four years and say, hey, wait a minute, here's what you said.

Behrooz Abdi:               One of things is that, especially in the industrial and automotive, you will see a lot more [00:34:30] sensors and what I call a swarm of sensors that will be coming in together with the power of the cloud and the architecture. On the CMOS side you'll see that and with AI, you'll see a lot of innovation in that area. 

                                    One thing that I've always said around Moore's Law is don't bet against Moore's law. It's been continuing to enable smaller and more powerful devices. But the thing that I've been really [00:35:00] surprised about is that with AI is actually, the capability has been actually growing faster in the last five to 10 years. So you will have products with AI sensing and a lot more powerful processing in every device. What that means, especially as privacy becomes a bigger issue, you don't have to go to the cloud for a lot of the AI applications you can actually do AI on a chip, not necessarily [00:35:30] the sensor but at the edge. Together with a bunch of sensors, you can make really fast decisions and be private and not share that data with anybody else. You go to the cloud for learning. That's all. 

                                    So again, whether it's a personal device or an industrial device where security is even more important, you can do a lot more interesting things at that level. That's I think one of the things that we're going to see as sensors developed into lower costs and lower power.

Ali Tabibian:                 That's fabulous. [00:36:00] Thank you for that description. I hadn't really thought about what I think Apple calls differential privacy. Keeping things local as much as possible. About that being driven by a lot of the, been made possible by a lot of the developments and the learning not being lost in the process. 

Behrooz Abdi:               That's something that you can have the history, you can have the privacy and security have it all together. And when you think about IoT, the way I describe IoT is where are you, what are you doing, who are you [00:36:30] and how do you feel? And that's all about sensors. If you look at what IoT enabled applications like Uber or DoorDash, when you click on an app and you order food or order a ride, it knows that it is you and it knows where you are. That, now going into all kinds of devices, that capability and not sharing that data with their cloud, with their servers, I think it's going to be the next big [00:37:00] thing. I think sensors together with AI will enable a lot more of that.

Ali Tabibian:                 Great. Behrooz, thank you very much. I know we have to get you going for an important appointment. Is there anything else in the next 60 seconds or so that we should talk about?

Behrooz Abdi:               I think we covered most of it. I think it's really important to the audience to be energized about semiconductor and realize that there's a lot more innovation that's coming and we just seeing the start. [00:37:30] With sensors, you're going to see a lot of that innovation. Every device around you, whether you're at home, in the car industry, at the enterprise. One of the areas that long term I'm very excited about to see sensors come into is the medical field where it will really make people not just more productive but enable healthier lifestyles.

Ali Tabibian:                 Amen, I certainly need to live more healthfully. Maybe if I could, do you have any sensors I could drink on my way out?

Behrooz Abdi:               [00:38:00] We'll pour some for you. It’s experimental, though. 

Ali Tabibian:                 Only the top shelf stuff please. Behrooz, thank you very much. 

Behrooz Abdi:               All right, well, thank you for your time.

Ali Tabibian:                 Thank you.

GTK:                             Tech. Cars. Machines. Subscribe here or at gtkpartners.com.