On this episode of the Futurum Tech Webcast – Interview Series I am joined by Esam Elashmawi, Chief Strategy and Marketing Officer at Lattice Semiconductor, for an exciting conversation about the myths that exist around FPGAs and the things Lattice is doing to set the record straight.
Three Myths about FPGAs
In our conversation we discussed the following:
- The reality of FPGA offerings
- Whether or not you really need high performance FPGAs and GPUs for AI and machine learning
- A look into the investments that Lattice has made around FPGA design
FPGAs are an important piece of the compute architecture, but not everything that is said about them is true. If you’re interested in learning more about what Lattice is doing in the marketplace, be sure to check out their website or listen to the full episode below.
Watch my interview with Esam here:
Or listen to my interview with Esam on your favorite streaming platform here:
Don’t Miss An Episode – Subscribe Below:
Disclaimer: The Futurum Tech Webcast is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors and we do not ask that you treat us as such.
Daniel Newman: Hey, everybody. Welcome back to another episode of the Futurum Tech Webcast. I’m Daniel Newman, your host, principal analyst, founding partner at Futurum Research. Excited about this interview series. Today, I’ve got Lattice Semiconductor joining me, a guest that’s been here before, Esam Elashmawi. Esam is an awesome guest, good guy, likes fast cars like me so, I love having him on the show, because he likes to talk tech, and he likes to talk about cars. We aren’t going to talk about cars on this show. Really, maybe a little bit through the lens of chips, but definitely appreciate having him join the show. So, in a moment, I’m going to bring him on, but before I do so, as always, just quick disclaimer. This show is for information and entertainment purposes only. And while we will be talking about two publicly traded companies, please do not take anything I say, or is said on the show as investment advice. Excited for the conversation with Esam. So, without further ado, Esam, welcome back to the Futurum Tech podcast and webcast. How are you, sir?
Esam Elashmawi: Good to be here. How are you doing? I like for the entertainment purposes portion of the segment.
Daniel Newman: Doing all the on air on the various business broadcasts, writing on Market Watch, opining a lot about stocks and tech companies, I just figured it’s safe. What I don’t want to have people do is listen to you, get super excited, buy a bunch of Lattice. Maybe the market has a bad day, goes down and then I’m hearing about it like it’s something I did. I mean, people need to make their own decisions.
Esam Elashmawi: They do.
Daniel Newman: But of course, talking to folks like you, it’s always great to get that kind of inside scoop. You’re a top exec. You’re involved in strategy. I’ve been on the Investor Days. I’ve listened to you paint the strategy for Lattice. Always really enjoy hearing the direction of the company.
Know your CEO, Jim Anderson very well. Another impressive guy. By the way, he’ll be joining the Make It Markets pod in the spring. So you get a chance bounce over there. But Esam, I have a fun topic for you today. I thought this would be something that would be just amazing to talk about. We’re going to talk about some of the myths of FPGA since that’s a business that you’re in, but before I kind of hit you up, we’re going to do the three big myths, maybe like 3.5 big myths here on the show. But before I do that, just for everyone out there that maybe didn’t see you the first time you came on the show or haven’t run across you before, quick introduction about yourself and your role over at Lattice Semiconductor.
Esam Elashmawi: Yeah. Thanks Daniel. Esam Elashmawi, as you said, and you pronounced the last name correct by the way. I know you always worried about that. But I’ve been in the industry for going on 33 years and those 33 years had something to do with FPGAs. So I’ve been an FPGA person since I’ve started and entered the semiconductor industry. I’ve joined Lattice a little over three years ago. I run strategy and marketing for Lattice, all the fun stuff by the way. But I can tell you with Jim Anderson and the executive leadership, it’s been a fun three years. It’s been a blast, amazing employees, amazing talent within Lattice. Great company to be part of. And I’m excited by the way to be on this show and address some of these myths. I don’t know what those myths are. I have a feeling what those myths are, but we’ll address those myths.
Daniel Newman: Well, we’ve had a number of conversations and I think part of what stemmed me to come on, bring you on and have this conversation was that you are so passionate about what Lattice is doing. You’ve come to not only believe in it as an executive of the company, but in a role like yours, where you’re really depending on building a company to be rewarded for the success in a role you have to buy in, you have to believe. And part of the momentum that’s come behind the company is, I wrote a piece on Market Watch before the pandemic. I said semiconductors would eat the world. Everybody had been saying software would eat world, but what is software going to run on? And software has to run on something. And so we tend to think about semiconductors through the lens of general purpose.
And more recently as cloud has proliferated. We think a lot about the application specific, accelerators, but for a long time, FPGAs have been really an important piece of everything from development of next generation, general purpose and ASICs, all the way through becoming a very hot commodity for certain workloads that just need more programmability, more flexibility. Maybe if you don’t mind, because I do have three myths I’m going to have you respond to, but for the audience, because we do have varying levels of everyone that’s financial and investors, like I said, to highly technical people in the beltway that listen to this show, how do you describe besides obviously what the acronym means? Like just a quick, what is an FPGA for kind of that maybe little bit less experienced part of our audience?
Esam Elashmawi: Sure. Field-programmable gate array, that’s what an FPGA is. And by the way, you’re talking about semiconductors eating the world. I think with all the stuff happening in the news now around chips shortages, we can walk into now a Seven Eleven or supermarket and talk about chips and everybody knows what we’re talking about. So it’s really, it’s gained a lot more popularity.
Daniel Newman: Doritos.
Esam Elashmawi: People know what we’re talking about now. But what is an FPGA? FPGAs is a vital part of the semiconductor components that are out there. It’s the high end value portion of the semiconductor industry. And they are proliferated everywhere. I mean, whatever we do on a daily basis, you may not realize it, but there’s an FPGA involved. FPGAs are programmable devices, they’re programmable in the field. So a customer can actually buy these devices off the shelf. They’re standard parts. There’s nothing, no design pattern programmed into them and through our software tools, they can actually program the logic they want within these FPGAs.
They’re used everywhere. They’re not specific for a specific horizontal, vertical segment. They’re a horizontal solution that goes everywhere. So think about 4G, 5G communications. Think about client devices. Think about industrial factories, robotics, motor control, automotive applications, consumer, medical, defense. They can apply to all horizontal segments. So you make a phone call. You’re getting on the internet, you’re using a server, you’re in a factory using some equipment, most likely there’s an FPGA in that system. So they’re used everywhere and they give the customers flexibility because you can put your design in there, your secret sauce, and you can do that within hours or minutes, depending on how complex of design is, where it can get programmed within your device. So very flexible, used everywhere. Not everybody knows that what they’re using today has an FPGA in it, but that they’re all over the place.
Daniel Newman: Absolutely, Lattice coexist with many of the largest semiconductor company names in the world. Some of those that you probably hear more about, the Intels, AMDs and videos, you guys are in systems right alongside them in many cases. And by the way, the importance of the FPGA and the growth has led you guys to being able to expand a little bit up market, double your TAM, which is something you guys talked about at your most recent Investor Day. You’ve been making acquisitions on the software side that are going to just add more value to everything that you’re building. And of course, if you heard about AMD’s acquisition of Xilinx, that’s a big move that a AMD was making to… It just validates how important they see the FPGA playing, which I always say there’s nothing better company spending over $30B to validate the business that you’re in. That just doesn’t hurt your feelings probably very much.
So let’s start with the first myth. FPGAs are big, expensive, and use a lot of power. Talk about why that’s more myth than reality.
Esam Elashmawi: I mean, actually, if you even go back, people would say, FPGAs are only used for prototyping and that’s it. They don’t go into production and that’s not the case anymore. If you look at the FPGA market, FPGAs have been around now for almost 40 years and they’ll continue for the next 40 years and beyond. FPGAs, you can take the FPGA market, which is a close to $6B to $7B, and you can probably break it up into three categories. There are small FPGAs, there’s mid-range type F PGAs, and there’s these large FPGAs that are out there. And what we specialize in are the small power efficient FPGAs and we’re moving into the mid-range which we announced at our Investor Day. But, what’s in people’s minds sometimes, and this is where the myth started, is that FPGAs are very large, very power hungry.
And that stemmed from, if you go back to decades, there was a time when there was a race in the FPGA industry about who can build the biggest, baddest FPGA that’s out there. And it was about how do I build an FPGA with more gates, more logic. So we’re talking about million plus, 5 million, 10 million gates of logic. And in that time, there was a lot of advertisements specifically from Xilinx and at the time Altera, which has now been acquired by Intel, about who had the mantle on the largest FPGA out there and that sent a perception that to be successful in the FBGA market, you had to build the biggest FPGAs that are out there. But if you look at reality today, that’s not necessarily the case. If you look at how many FPGAs are actually used out there, there are more FPGAs in the small power efficient and in the mid-range and lot less that are being used that are very large, because the large ones are designed for specific applications. And that’s really around data center acceleration, or maybe a communications computing, which eventually by the way, because they’re very large, because they’re very high powered, get designed out by an ASIC.
But if you look at what Lattice does, we build these, and we specialize in these low power, small power efficient FPGAs, and now we’re introducing, in next year in 2022, our mid-range power efficient FPGAs. And that goes into a wide variety of applications. And so they’re very power efficient. In fact, when you take a look at what we can do with an FPGA compared to even an ASIC, you’re really not giving up much from an ASSP perspective or a performance perspective. In fact, you can even compare us to some as ASSPs, like MCUs, and we can actually do that function at a lower power.
That gap, or that myth, that FPGAs have to be much larger and higher power, it truly is a myth. And most of our customers will tell you, “Hey, I’d rather use an FPGA, I’ll get the power that I want, much lower power. I’ll get the performance that I want, because things happen in parallel.” You get that with the types of FPGAs that we focus on, where our competitors again, are focusing on the much larger FPGAs out there. And that’s good. We can do that, by the way.
Daniel Newman: I was going to say, it’s worth noting though, you guys have really accomplished a lot in terms of power efficiency and with all of the ambitions that the world, that enterprises, and of course, governance and regulators have for companies to be more power efficient, more carbon neutral, I mean, this is a footprint thing. I keep looking at everything from on Blockchain, the gas expense on using E and stuff.
A lot of people don’t realize this is semiconductor horsepower that’s costing all this money. And so I’m not saying you’ve got the answer yet to solve that problem, but I’m saying companies that are designing for power efficiency are solving a problem that’s much bigger than just lowering the electric bills and data center. You’re making an impact on the world. So, those ESG, all those ESG statements and stuff that’s out there. I know some of that sometimes looked at as soft or it’s looked at a little bit as posturing into the market, but there is real innovation that can be done that’s going to make a big difference to the world.
Esam Elashmawi: Absolutely. We’re doing sophisticated AI stuff at the milliwatts of power, which is actually pretty incredible. And when you compare that to the large FPGAs that do things in other large devices that do things in the hundreds of watts, there’s a huge difference in power time. We’re not talking about, “Hey, we’re 20% lower, 15% lower. We’re significant lower and power as the application.”
Daniel Newman: Numbers like 10 X, Esam. So you mentioned AI, let’s move on to that. The second myth I have is AIML is helping drive demand for semiconductors, but you can only really do that with high performance FPGAs or GPUs.
Esam Elashmawi: Yeah. That’s another myth. When you think about the amount of intelligence people are trying to put into systems, doesn’t matter if it’s in your car, your client device, industrial piece of equipment, robotics, we’re just putting more and more intelligence into equipment. And by the way, that’s great. It just makes things better for everyone, safer for everyone as well. You do that by adding artificial intelligence, machine learning. And the reality is there are wide range of applications for AI. If you’re in a data center and you’re trying to do heavy workloads around AI, yes, you need a large FPGA, GPU. That’s where the NVIDIAs, the Intel’s and others are trying to be successful. The that’s not the type of FPGA’s that we’re building.
We’re building these power efficient edge FPGA’s that can do AI very efficiently on the edge. I’ll give you some examples of that. We had a customer, for example, with doorbells. We worked with this customer a while back and in the application he said, “My doorbell, I need to take pictures today. I was doing it,” and this is a while back, “two to three frames per second.” That’s great. If somebody comes out to the door, you take his picture you’re fine. But the challenge he had was what if somebody’s not just coming up to the door and ringing the doorbell? What if they’re actually taking a package and running with it, two to three frames per second, isn’t it enough? I need at least 10 frames per second, eight frames per second, to be able to get there.
So we worked with this customer and he had a choice. He said, “Do I continue with an MCU that gives me two to three frames per second, or do I migrate to an FPGA, or do I migrate to some type of accelerator or CPU?” So it’s a good example where the customer was actually looking at three different options. So with the micro controller, the challenge was he couldn’t get to those frames per second. Micro controllers are serial in nature. You’re just not going to get that performance. With an FPGA because we do things in parallel, we can get to the frames per second that he needed and we demonstrated at a lower power than that MCU or a ASSP and so we gave him a better solution. But then conventional wisdom, or the myths said, “Well, why don’t I do a CPU or some type of GPU to do this?” Well, you could do that with a CPU and GPU. In fact, you’d probably get more frames per second than what you need. You’re at a much higher power consumption and the ASSP’s higher as well. So what’s the right fit?
That’s a good example where if you look at things across the broad spectrum of applications, not everything needs what’s going to be inside of a data center. On the edge, you need less performance than that. And that varies from application to application. The doorbell is one applicant and drones, it varies as well. We gave an example on client devices where we’re engaged with top OEMs on adding AI, intelligence to client devices. Again, you’re not going to put a thousand dollar solution in there, you need something that’s power efficient, small, ideal for those types of applications. And you can do that with an FPGA. So do you need to have a GPU? Do you need to have a large FBGA to do your AI applications? It goes back to the application that you’re trying to run, but on the edge, things around doorbells, drones, robotics, client devices, what you need are devices that are power efficient, but also have the performance to be able to do the application needs and our FPGAs are designed and suited for those types of applications.
Daniel Newman: Absolutely. You hit a number of things on the head. And I think one of the important things that’s the underpinnings of everything you said is, sometimes the question isn’t whether a certain type of technology can do it, because GPUs will be able to do a lot of the things that your FPGAs do. But going back to that first question, the power and the cost efficiencies when you don’t need, you’re not necessarily trying to train a vehicle to be able to self drive. You’re trying to do something very specific that has small amounts of maybe evolution and modification over time and the flexibility of the FPGA and the cost and power efficiency gives you those two layers where the investment isn’t going to be in vain. You can make those modifications and upgrade and evolve without having to turn over your whole hardware, your core hardware all the time, and then you can continue to program and upgrade and develop and of course save that money and that power in the process. It’s really a multi-pronged thing. Go ahead Esam.
Esam Elashmawi: Exactly. Actually, and you remind me of another myth that I hear sometimes from a customer, I had a customer that came up to us and said, “Okay, I can do this with your FPGA, but why don’t I do with an ASSP or an ASIC? Maybe I can, now that I know the function that I want, why don’t I harden it into an ASIC? And that’s another myth as well, because that same customer that we worked with, what we demonstrated to them was during the development cycle. So we had a customer we’re working on these AI algorithms. We adding smartness into their system, we’re adding intelligence to their system. And while we’re progressing through that, he actually brought in a company that was doing it on an as ASSP, or an accelerator, an ASIC. And he was doing a one-to-one comparison between us and that ASSP device. Through the progression of the evaluation, what we determined with the customer was, “Hey, we’ve got an enhanced algorithm, actually, that’s going to give you a better performance.”
So think about this as innovation in AI, it’s happening at a constant rate. Every quarter, every six months, every year, we’re getting more and more performance out of better algorithms, better ways of doing the AI itself, the neural networks. And so when we discovered this new way of doing this, that was more efficient. What he realized was the ASSP that he had would have to be re-spinned, the ASIC. And he actually coined a term with us. He goes, “Wow. These FPGAs are future proof,” meaning that as algorithms change your AI changes, or you want to add a new use case and another use case in, you don’t have to re-spin an F PGA. That’s the beauty of it. It’s programmable so they’re future proof. But you don’t get that out of a hardened ASIC or a hardened neural network that can do specific functions.
The other thing that you get out of an FPGA is you store your secret sauce. If you’re buying a specific device that everybody else has and has access to and does things the same way, where’s your secret sauce? And it’s not just around AI, but think about motor control and precision, or you’ve got other algorithms that you want to do on the way you’re doing trafficking of signals and managing things. Your secret sauce is in your FPGA. You’re not sharing it with anybody else. So, there are advantages of having programmable logic over hardened standard devices that are out there, ASSPs, because of just innovation in our industry, secret sauce, being able to modify things on the fly and not having to re-spin an ASIC. So, that’s another one.
Daniel Newman: Yeah. So we’re going to call it 3.5 because that sort of fits into a little bit of one and a little bit of two, but it’s notable on its own, but I think FPGA maybe needs to have a name change. If Facebook can change its name to Meta, maybe we can call it Future Proof Gate Array.
Esam Elashmawi: We’re going to take it under consideration.
Daniel Newman: The FP, you said FP and future proof. I wrote a book called Futureproof by the way, go ahead and everybody out there listening, go ahead and check that out on Amazon, buy that. All right, so let’s keep moving. We got one more. I’ve really appreciated the time, Esam. I always learned something when I speak to you. It’s funny because people always hit me as the expert in semiconductors, but I can always, always learn more about FPGAs, and by the way, I’m always fascinated. So here’s one that I hear all the time.
Esam Elashmawi: All right.
Daniel Newman: I’m constantly debunking is that the FPGAs are difficult to design and they require a specific set of expertise that’s very hard to find.
Esam Elashmawi: All right. So the first thing I’m going to, we have over 9,000 than customers. So it can’t be that difficult. But however, what most people will say is, “Hey, you have to know RTL to design an FPGA.” And there’s probably more software engineers out there than there are RTL engineers out there so how do you get to those engineers to be able to leverage the benefits of an FPGA? That’s a true statement and it’s something that the industry had addressed, the FPGA had addressed over the last several years, where we’ve developed and invested more in our software tools. And if you look at Lattice in particular, over the last several years, we’ve been investing a lot in what we call our solution stacks. And we’ve got award-winning solution stacks that are out there around Sensay AI that does artificial intelligence type applications, embedded solution, embedded vision solution stacks around computer vision. And we’ve got Century for security, Automate for robotics, and we’ve got a roadmap of more solution stacks that are going to be coming.
But the purpose of these solution stacks and why we invest in them is to help our customers get to market much faster. We’ve got over 9,000 customers. We can’t touch every customer. We have to have solutions around software that make it easy for customers to get to market much faster. And also by the way, makes it stickier because they’re using of our softwares in their product as well. They’re using a lot of what we provide them in their actual product and that makes it very, very sticky. But what we can do with these things, if you take, for example, Sensay AI, we take industry tools that are out there, open source tools like KIRIS and Cafe that allow people to do machine learning, an inferencing type learning for their AI application. And then we can take all of that and automatically translate that for them into FPGA language. So they don’t have to be an FGA expert.
And then we can take that with kind of what we provide them a lot of use case and models around object detection, facial recognition, key phrase, and they can take that from the libraries as well, put that in. So we actually make it a lot easier for them to implement their designs into an FPGA. Same thing from our security side, we have our Century solution stack and what we do using a soft risk five processors is we give the user an interface that looks like a processor so they can go in with their C language and configure how they want to configure the security function within their system and they don’t have to be these RTL experts. So if you go back a couple decades ago, maybe that myth was more real than it is today, but there’s a wide adoption of FPGAs even by non-FPGA experts because of the investments we’re doing in these solution stacks and our software tools to make it easier to adopt FPGAs into multitude of different types of applications.
Daniel Newman: Yeah. You hit it on the head and you’ve given the audience and of course me, I say take them to school Esam, but the long and short is the architectures are evolving. The ability to utilizing FPGA in a number of applications that maybe historically haven’t been thought about has clearly reached the surface. And this is also a good underpinning of the growth story that you’ve been sharing on your quarterly earnings, that you’ve been able to share at your Investor Days. It’s been part of why you’ve been able to come upstream in the market just a little bit. You’re certainly not trying to go after that ultra high end. You’re just saying that there is this gap in this white space and Lattice has shown with what it’s been able to accomplish, the 9,000 customers, that it’s clearly well positioned to do that.
So I want to thank you, Esam. We’re all out of time here, but really appreciate you coming on the Futurum Tech webcast. Appreciate Lattice Semiconductor letting you slip out here and spend a few minutes with me to talk about this. I think this should become a marque asset, a marque conversation for people that are really wanting to better understand what an FPGA is and what it isn’t, and what the opportunities are. So hopefully if you are out there and you listen to this, like I said, you went to school and now you know. But Esam-
Esam Elashmawi: Hey, I appreciate it.
Daniel Newman: Thanks for coming on. Hey, you’re going to be back soon.
Esam Elashmawi: Thanks for having me. And by the way, happy holidays and love to catch up and talk about cars. It’s what you started off with.
Daniel Newman: Yes, we always will and we always do. We won’t bore those of you that only want to talk chips right now, but just trust me. If you ever do want to hit either of us up. Esam, I’m going to send you off to the green room. I’m going to close up the show, but we will see you soon. You’ll be back.
Esam Elashmawi: Thank you. Happy holidays, everyone.
Daniel Newman: All right, everyone, that’s our show. Love having Lattice on the show, a great company doing a lot of interesting things. And of course, if you’re like me and you’re passionate about semiconductors, and you’re passionate about the innovation in the space, this is another company to rip and roar. I had them on my Market Watch Semiconductor Companies to Watch that maybe aren’t top of your radar and that’s because the company’s doing really interesting things and getting the chance to talk to folks like Esam and Jim Anderson, their CEO, has only painted a better picture of that for me, but that’s it for this episode for the myths. Thanks everybody for tuning in. See you later.