The Future of Wireless Networks, Academia Startups, & Intel: A Conversation With Dr. Andrea Goldsmith

Sriram Viswanathan: Welcome to the Tech Surge Deep Tech Podcast presented by Celesta Capital. Each episode, we spotlight issues and voices at the intersection of emerging technologies, company building, and venture investments. My name is Sriram Viswanathan, founding managing partner at Celesta Capital.

If you enjoyed Tech Surge Podcast, feel free to share it, subscribe, or leave us a review on your favorite podcast platform.

Today, we're delighted to welcome Goldsmith to this podcast. Andrea, Dr. Andrea Goldsmith. is immensely accomplished, is both as an academic leader and a entrepreneur. She currently serves at the Princeton University as the Dean of School of Engineering and Applied Science, as well as Professor of Electrical and Computer Engineering.

She's previously served as long and distinguished tenure at Stanford University, as well as Caltech. She's a pioneer in wireless communication and a scientist that holds dozens and dozens of patents, authoring several books. and winner of several notable awards, including the Marconi Prize, which is a highly, highly recognized and distinguished award for all her work.

She's co founded and served as the CTO of multiple innovative startups in the wireless space, including Quantenna. and Plume Wi Fi. She serves on a number of boards, including Intel, Medtronic, and is in the, is a member of IEEE and serves as an advisor to the President's Council of Advisors on Science and Technology.

Andrea, it's really a delight to welcome you to this podcast, and we're so lucky to have this conversation with you.

Andrea Goldsmith: Thank you so much for inviting me. Look forward to the discussion.

Sriram Viswanathan: So not often we have an opportunity to talk to someone as accomplished as you are, not just from a technical standpoint, but also someone who's gotten your hands dirty as an entrepreneur and also working with students and in the university construct and being on behalf Companies like Intel and others.

So we'll have lots to talk about. So if, if we may start first on your background, more from a technical standpoint, I know that you, you have a very rich sort of history. in the electrical engineering and communications area. And you were at Caltech before Stanford. So can we talk about your background leading up to Stanford?

And then we can, we can, we can get on to the next level of detail from there.

Andrea Goldsmith: Sure. I don't know how far back you want to go, but I was inspired to study engineering. My dad was a mechanical engineer. My mom was an artist and I saw engineering as a very creative. endeavor where you could really have a big impact on people.

And I became enamored with wireless towards the end of my undergraduate degree at Berkeley, and then working at a small defense wireless company in the late 1980s in Silicon Valley called Maxim Technologies. And I just found the technology of wireless magical. The fact that you could communicate with people.

anywhere in the world. And the technical challenges were met. So this was the late 80s. There was no Wi Fi cellular had just rolled out. And I was very intrigued about these emerging commercial technical systems. And that's what motivated me to go back to grad school to study wireless communication. I was very fortunate to work at AT& T Bell Labs when it was one entity as a graduate student in the early 1990s when the first digital cellular systems were being standardized.

And so when I finished my PhD in 1994, I had the option of, going to industry. I interviewed with Qualcomm and other companies. I interviewed with Bell Labs because I really enjoyed the research lab environment there. And they were just about to split. Lucent and AT& T were about to split. So it was a tumultuous time at Bell Labs.

And I interviewed in academia and had job options at. At, uh, MIT and Caltech decided to stay in California. My husband's startup was in the Bay Area, so that was part of that decision. So I started my academic career at Caltech and I was there for four years and, uh, it was a great place to start. I had wonderful colleagues, very well resourced university.

But then Stanford had an opening for a wireless communication professor, and I was very fortunate to get the job. So I came to Stanford in the beginning of 1999, kind of the bubble days.

Sriram Viswanathan: And this is with Professor Tom Kailath in Stanford? So Or in this

Andrea Goldsmith: I was just in the electrical engineering department, so I was recruited into STAR lab actually, which was more of the aerospace and satellite communication group, but I was part of the EE department and it was really the heyday of wireless communication at Stanford.

There were six or seven of us working in wireless and many, many students and classes and the startup and broader. commercial ecosystem for wireless was booming, Wi Fi was right around the corner. So it's a super exciting time to be at, at Stanford doing wireless communication. Yeah.

Sriram Viswanathan: So, you know, before we get into some of the specifics of your research and your particular work, if you just sort of step back, I, uh, had some experience running the wireless program office at Intel, and there was always this battle between the IEEE camp.

and the sort of the GSMA, you know, 3GPP, 3GPP2, the traditional cellular camp. How do you think about those two camps? Do you think that now that distinction is, uh, you know, difference without a differentiation? Is that where we are? Because most of the core technology is applicable, whether you're in the IEEE world or in the cellular world.

Andrea Goldsmith: It's a really interesting question, especially since That question has been asked for decades now, going back to when Wi Fi first burst on the scene as an 802. 11 standard. And the answer hasn't really changed that much. I think the way these two communities evolved is the telecommunication wireless standards, the cellular standards, the G's, if you will, came out of the telco industry.

Very conservative. grew out of, we need five nines, which means we need a reliability in our wired communication systems of a 0. 00001 percent chance that we drop a call. And so they were very conservative. They required that the spectrum was licensed so that they could fully control what was in the spectrum band.

And they evolved in a very conservative manner. Wi Fi was the Wild West. It came out of the IEEE standards, the Ethernet standards, which were best effort. And so it was okay in Wi Fi, especially the early days of Wi Fi and even today that it's best effort. If you, it's not licensed, there may be interference.

You're not guaranteeing any kind of service. And both camps have evolved according to those. Really core foundational principles where they started and it's why with cellular you can get coverage almost everywhere. It's not five nines. I mean, it's not that reliable in many parts of the Silicon Valley or the country or the world.

You'll drop calls, but there's still the philosophy of we have to have licensed spectrum. We have to have standardization. Everything that's operating in the spectrum is fully controlled. Versus the Wi Fi standard, which is not ubiquitous, right? It's only where you have Wi Fi access points. It's really an evolution of the Ethernet standard, the 802.

11, uh, standards. Um, and they coexist in a very synergistic way. I've been asked over the years, well, will Wi Fi go away or will cellular go away? Neither is going to go away because they serve different purposes. And those two kind of core. philosophies of those different types of systems serve those systems well, which is why I don't believe that they will ever go away.

They're complimentary starting from, you know, the 2000s when almost every, uh, portable phone started putting wifi in it because they needed the bandwidth. That's the other big difference is because the wifi community operates in the unlicensed band, there's a lot more spectrum. So you can get much higher data rates, but it's unreliable versus the cellular.

systems which are in licensed band, a lot less spectrum available, but it's more reliable because it's licensed. So I

Sriram Viswanathan: have a minor confession to make. I have a lot of battle scars in fighting that battle in the whole IEEE versus cellular landscape. But specifically your work now is looking at the next generation technology and next generation of next G, if you will, is the transition from 2G to 3G to 4G and beyond, is it largely The type of modulation techniques that are used is, is that the core reason why these transitions have occurred or is it enabling new, you know, use cases?

And I'm assuming I'm obviously that's a leading question because I'm assuming it's the latter. But if that is the case, if today, if I have 5G and if I have, you know, text, voice, video, data, IOT and all of that, what does one hope to achieve? In the next generation that you don't have in 5G, is it just faster, better, cheaper as you evolve from one generation to the next?

Andrea Goldsmith: It's an excellent question. And if we look at the history of how cellular has evolved, I mean, 1G to 2G was analog to digital. So obviously there was a compelling reason to move from 1G to 2G, better, cheaper, faster. You could do data, which you couldn't really do over analog. I mean, there were just many, many reasons to go to digital, but the digital cellular standards were split.

There were multiple different ways to do sharing of the spectrum for 2G users. Uh, when we went from 2G to 3G, we went to much bigger bandwidth. And so that was because 2G systems were mostly about voice, higher quality voice, and higher capacity systems for more voice calls and a little bit of data text.

Whereas 3G, that was the beginning of, okay, maybe people want to do more data than just text. And that was great until the middle of the 2000s when the iPhone came out and brought all of the 3G networks to their knees overnight because all of a sudden there was a huge demand for data. And that really spurred the development of 4G where the battles were at the physical layer.

Should we do co division multiple access CDMA or Orthogonal Frequency Division Multiple Access, OFDM, OFDM won out in, in 4G and OFDM, if I go back to my research as a PhD student, is the capacity achieving strategy for time varying wireless channels under certain circumstances. So it didn't surprise me at all that that's what ended up winning out, also won out in the Wi Fi standards for the same reason.

It's just a better, more flexible technology to do high speed data. So that's the foundation of 4G. Now, when we went from 4G to 5G, It was really about how do we sell more phones? It wasn't so much, what else can we enable? It's how do we get people to buy 5G phones? Well, we'll call it 5G, but it was very incremental in terms of the difference between 4G and 5G.

In fact, some 5G users want to go back to 4G because they got better performance. So the real challenge, I think, for researchers and, People in industry, entrepreneurs and big companies is what is it that we can enable with 5G that or 6G or beyond 4G that we can't do with 4G. So yeah, better, cheaper, faster, of course, better people going to be willing to pay for that is the expenditure to lay 5G base stations all over the country going to be worth the return that you get from people buying 5G phones.

And I think. As we look ahead to next G, I don't even call it 8G, but what really is going to change the paradigm for cellular, we have to enable things that we can't do now. AI may be a killer app because people want to do AI on their phones. A lot of the applications, whether it's automated driving or delivery or massive sensing, uh, is going to require a different type of communication, whether it's very short packets or very low energy or devices that run off a battery, or at some point we're going to move away from mobile phones and have different interfaces.

And so how do we build a cellular system that's going to support whatever killer apps emerge? That's what 3G and 4G did. It says, we don't know what people want to use these phones for. We're just going to enable whatever the paradigm is for the killer app that emerges. And we don't know what that's going to look like going forward.

What are people going to be willing to pay more for on their phones besides what they have today? But we need to build networks that are flexible enough to enable whatever those killer apps are.

Sriram Viswanathan: So, you know, is it fair to say. That in this transition that has occurred in the telecommunication wireless evolution, uh, there are step changes that have occurred, like the way you said from analog to digital, it was very clear.

You could now do text and, you know, getting on websites or whatever. And then moving from there to the next step change, where you had Multimedia potentially high speed video, high speed data and video is sold as a transition with 5G, a way of connecting machines and IOT and machine to machine and all of that.

Would you say that that has really not panned out to the extent? How people thought about it when they wrote the standard and if so, why

Andrea Goldsmith: yeah, absolutely it hasn't I think that's obvious If you look at the returns on for the telcos in terms of their cellular business The reason it hasn't panned out is it was not clear what killer app?

IOT would emerge that people would be willing to pay for. Of course, enabling every electronic device with an on off switch to be able to connect to the backbone network and to the cloud and to do data processing sounds like a compelling evolution of technology. But in order for it to be a business case, you have to be able to make money.

And I think that that's what hasn't emerged in 5G. And there were many things from the new radios being more efficient and having protocols that allowed for short packets and this kind of thing to slicing of the network, virtual network slicing to give higher quality of service, but none of those.

Things have emerged to support a killer application that people are willing to pay more money for than they were willing to pay in 4G for their high speed data and voice and video. And as a customer of a telco or a wireless customer, you expect that performance is going to get better and you're going to pay the same or less.

You're not going to be willing to pay more for the same performance and technology that you get. Uh, currently. And so you need to enable new things. And what hasn't emerged yet in the evolution of cellular is what is an application different from what we have today that people are going to be willing to pay for.

And that's why I say AI may be it. I mean. If you think about being able to do AI on your phones, it's going to require connectivity, the, in the cloud, it's going to require computation, both in the cloud and on the phone, and maybe at the edge as well, and people are probably going to be willing to pay more if they can get certain types of AI on their phone.

So that's one potential killer app. I don't know that it will pan out. If you think about automation or automated driving. The data rates that you need for that can already be supported by 4G and 5G. It, it is real time. And that's another area of cellular technology that we're not particularly good at, guaranteeing real time connectivity with very high reliability.

So that is a potential area of killer applications, but none have emerged that people are willing to pay for yet.

Sriram Viswanathan: So in the automotive example that you just alluded to, you know, we had a session with the CEO of Zoox in one of our podcasts. It sounds like the automotive autonomous driving problem. Is a compute problem and not so much a communication problem because most of the processing is happening in vehicle, you know, whether it's perception or fusion technology or whatever, if you apply a I.

to that, to the next level. Isn't that just extending AI at the edge, where it is really more of a compute problem? Where is the issue of that becoming a communication problem that justifies someone to pay huge infrastructure for 6G? Where I'm failing to understand what the connection would be for AI as a killer app that drives this evolution to the next generation.

Andrea Goldsmith: I agree that So, today, autonomous driving is mostly a compute problem in the car, so it's not a communication problem. But if you envision what autonomous driving might look like, suppose there were no people driving cars, right? It becomes a resource allocation problem, right? Where do you send the different cars?

It's a little bit like Waze operates today in figuring out for an individual where they should go based on traffic. But if every vehicle on the highway is automated, then you're going to need coordination across those vehicles to minimize traffic, to minimize accidents, to think about the impact of weather, how much do you space cars apart, so you can envision, um, just like networks today, you know, are packets of bits that are controlled in a centralized way, you could envision Automated cars, if every car is automated or a large percentage are, um, where they're going to need to communicate more for resource application safety, other types of things.

When I talk about AI as a killer app, it isn't so much using AI for a larger system, although that's part of it, but it's more just getting AI on your phone. If you want to be able to do AI tasks on your phone, you can't store the massive amounts of data. You can't even collect the massive amounts of data.

So where do you do the training? Where do you do the runtime operation and inference? I think that's an interesting communication problem that, as I said, is going to involve the cloud. It's going to involve the end device. And that means there's got to be communication across them.

Sriram Viswanathan: So in a way, would you say That the scaling and growth of hyperscalers and the types of services that they offer and more concentration at the core of the network happens, there's a need for greater communication.

But if all that intelligence and the inference happens at the edge, one could argue that you don't need that sophisticated, you know, next generation network. I mean, is that, is that the way to think about it?

Andrea Goldsmith: The devil's in the details in that, yes, if you're going to train all the models centrally and then you're just sending the trained models to the end device, you don't need a huge amount of communication, but there's a lot of reasons not to do all the training centrally.

Part of it is just security. You know, I don't want to it. give up all my data to some central entity to do the training. Part of it is you might want different models depending on the end device. And so you might want customized models that are going to be a lot easier to customize at the end device. So I think we're in the very early days of the AI revolution.

I'm not going to predict that that will be a killer app for communication and to drive demand and revenue in cellular. But when I look at the landscape of what might be a killer app, to me, that's the one that emerges as the most likely because automation, I think eventually will require a lot more communication.

It doesn't today because the way we're doing autonomous driving is very local. It's really just, okay, what's around me, you know, what's within range of the LIDAR, uh, that I can see. And, um, and it's not using the larger infrastructure, whether it's other cars or traffic lights or. road infrastructure sensing or weather or other things to do a better job of automation because we're still in the early days of automated driving as well.

Sriram Viswanathan: So in Stanford, you know, when you were faculty there and you were doing the research, your primary research was in extending the core performance of wireless networks, and you were also looking at new innovative algorithms for optimizing the wireless. Can you talk a little bit more about what exactly was the innovative idea, new idea that your, your research was focused on?

Andrea Goldsmith: Sure. When I Went back to grad school in 1989. I was incredibly fortunate to have the most amazing advisor, Praveen Varaiah, who was actually at the time just starting, uh, leadership of the PATH program, which was a program by Caltrans in California to automate one lane of every highway in California. And he was a control theorist, economist, didn't know much about wireless communication.

So he brought me in and, as a PhD student and said, well, how do we do wireless communication for these systems? And this is just when the second generation of cellular standards were being discussed. So there was this huge war about what the second generation should look like. And so I asked him, well, how do you figure out whether you should use time division or frequency division or code division?

And he said, well, you should look at the fundamental capacity limits of the channel. What's that? I have no idea. So that got me off on the beautiful field of information theory, which defines the fundamental limits of a channel and how you achieve those fundamental limits. So when I was doing my research at Stanford, wireless was booming.

And I said, as we look to the next generation and multi user systems and multi hop networks, what is the fundamental limit of these channels? And does that give me insight into the way to design the optimal algorithms and protocols and physical layer design to achieve those limits? So that's a lot of what I was doing at Stanford in my first

Sriram Viswanathan: decade.

Core information theory.

Andrea Goldsmith: Information, well, information theory, and then. And taking the insights that you get from that beautiful mathematical theory and applying them in practice. And so I've always, in my own work and my research with my students, try to look at the theoretical limits or theoretical aspects of a problem and then take the insights from that theory and try to develop practical schemes.

and see how well they do. And if they do well, then you've kind of closed the loop. If they don't do well, then you go back and say, well, what did I miss in the theory? And so I did that. That was my PhD research for wireless channels. When I got to Stanford, multiple antennas were just starting to come out.

The 802. 11n standard was just starting to be talked about, which was the first multi antenna standard. So then we asked the question, well, What's the fundamental limits of a multi antenna system? And that led to a lot of my work in beamforming and dynamic optimization of beamforming and multi user systems.

So it was all really inspired first by the theory. And then can we develop practical techniques that are achieving? that theoretical performance.

Sriram Viswanathan: It was interesting that you made the comment about information theory. So I was at the, you know, Celesta has a relationship with Bell Labs and we're actually working on commercializing a lot of the technologies coming out of Bell Labs as we speak.

And I had the opportunity to be at the famous you know, Shannon's little corner, Shannon's Law, the father of information theory. And I know that you spend time at Bell Labs as well, but you made a reference to multiple antenna that that's related to Dr. Paul Rogers work in MIMO. As you know, Dr. Paul Rogers is a senior advisor to Celesta.

So in most of your work, it is really identifying key core fundamental technology that actually extends the capacity and the capability of the wireless network, regardless of what generation that is in, because this sort of sustains generation to generation and beyond. What are you most excited about in this next five years of wireless evolution?

What is the hard problem that you think is going to be fascinating for the industry to have solved?

Andrea Goldsmith: That's a really good question. And going back to the 70s, there's been predictions that the physical layer is dead, which means that the way we communicate over point to point channels, we've already solved all those problems, so there's nothing more to do.

And that's been proven false for every generation. And I think that's true now as well. If we look at what should the next generation of cellular enable? One of the things is very high speed devices, devices that are moving at fast speeds, whether it's high speed train or an airplane or a drone. We don't have good techniques to communicate at high speeds when the channel is changing too fast to estimate it and adapt to it.

All, all the work I did starting from my PhD through the time at Stanford. assume that you could estimate the channel, feed that estimate back, and then adapt optimally at the transmitter to how the channel was changing. When you can't do that, you need a new physical layer. And there's interesting new ideas like orthogonal time frequency space modulation, which actually doesn't do that kind of adaptation.

Uh, so This is

Sriram Viswanathan: the follow on to OFDMA.

Andrea Goldsmith: Yeah, it's, it's, uh, well It's useful in some settings and OFDMA will continue to be the workhorse of future generations of wireless because it's so powerful when things aren't changing too fast, when you can estimate the channel and feed it back, you can prove that achieve the Shannon capacity of, of that wireless channel, but when you can't, then you have to ask, okay, well, if I don't know the channel, how do I communicate reliably over a channel like that?

And that's one of the interesting questions I think for the next generation, I think. The other questions are not so much about the point to point channel, but the network overall. So how do we ensure that users that aren't transmitting all the time can share the channel in an efficient way? If you think about energy consumption, if you have devices that you want them to operate on a battery for a year or five years or forever, we don't have good communication techniques that do that today.

If you think about some of these applications of automation, um, How do you ensure low latency and reliable communication? Because if you're creating a network of cars, obviously getting information between those cars quickly can be a safety issue and we don't yet have ways to guarantee latency. So I think what will drive research in wireless over the next decade or so is going to be application driven.

If we identify what are the interesting applications that. We can't support with today's technologies and then kind of bread and butter, faster, cheaper, uh, lower power, uh, where most of the work that we've done is really only been about faster, not about cheaper and not about lower power and not about higher reliability or security and privacy either.

Sriram Viswanathan: But this is fascinating. I think, you know, we can talk a lot about your technical depth of understanding and you're a pioneer in this, but you also sort of expanded from being an academic to being an entrepreneur. You actually started two companies, Quantana, which Celesta was involved in. And then of course, uh, the, the other company that you had, uh, was also, uh, was it Pulse, uh, uh, Plym WiFi.

Yeah. Those were really sort of putting into practice in commercial terms. lot of the research. Can you talk briefly about, you know, Quantenna and Plume and how they were different?

Andrea Goldsmith: Yeah. So I had been an academic for 20 years when I started Quantenna. So starting in 1994 when I began at Caltech. Um, but even when I was a professor at Caltech, I actually worked in a startup that was doing a non standardized five gig radio before 802.

11a came out. And so I'd always really been intrigued by practice. And I think it was to a large extent because all of my PhD research was inspired by working at the defense startup in the eighties. So after 20 years of being an academic and then that small foray into doing five gig wireless radio, I, wanted to see if all the research that I'd been doing in adaptive, uh, modulation, adaptive beamforming, multiple antenna communications mattered in practice.

And I was fortunate that I was approached by my co founder at the time, uh, Behrouz Razvani, who'd done a successful VDSL company. And he was intrigued about the emerging Mymo Multiple Antenna Wi Fi Standard Data 2. 11n, which was just starting to be written then. So he approached me and said, should we do a company together?

And I was naive enough to say sure. That's how Quantenna came about. We started thinking about doing multiple antenna Wi Fi for cell phones. And we were fortunate early on to have some key conversations with cell phone makers, including Tony Fidel of Apple and the Motorola Razor team, who said, we're never going to put a startup's Wi Fi chip in our phones.

It's too risky. So we pivoted to doing gateways and access points. And our timing was very good because by the time the technology was ready. That market was starting to take off and partly because we demonstrated that you could actually do video distribution through the home.

Sriram Viswanathan: Yeah. This is, I mean, this was a great IPO in 2016.

And you know, that, that turned out very well.

Andrea Goldsmith: And Plume, what happened to Plume? So Plume is still private, but they're very successful. All of the mesh networking that Comcast sells to its customers is based on Plume Wi Fi. And there's also standalone direct to consumer Plume mesh networking. The reason I did Plume, I wasn't actually thinking to do another startup.

I, Quantenna was quite a wild ride and very tiring. And I was happy to go back and I was actually elected to be the Stanford faculty Senate chair right after I came back and doing a lot of work in the IEEE was not thinking about doing another startup. But as you know, when a startup opportunity comes along, you can say yes or no, but not.

Give me another year or two to recover from my last startup. So Flume at the time we were going to do software in the cloud to manage small cells in cellular systems. And I worked in small cells going back to my Bell Labs internship days in the nineties. And it just seemed like such an interesting game changer for wireless technology to put software in the cloud.

This was 2010. to put software in the cloud to manage wireless devices. And I just said, this is too exciting to pass up. So that's why we started Plume. When small cells didn't materialize as we had hoped, we pivoted to doing software in the cloud to manage Wi Fi access points. And that's the premise of Plume.

Eventually, it became managing mesh networking Wi Fi access point. That's their business and, uh, we have an excellent technology in Plume as well. That's,

Sriram Viswanathan: that's amazing. Actually, uh, you know, I'm more fascinated by the straddling of these two worlds that you seem to have done exceptionally well. With two startups.

And, and of course, we'll talk about your, uh, journey in the academic realm, you know, going into leading one of the top universities in the country in Princeton, but, you know, we're actually investors in some other faculty members of your institution, you know, Chris Ray and Kunlei as part of Samba Nova, which is one of our companies.

And we've actually done lots of things. I'm always fascinated by how easily some faculty are able to. Switch between these two realms effortlessly. Is there some, is there a unique formula to that? Because, you know, not everybody not only has the opportunity to do that, but There seems to be an entrepreneurial genetic makeup, if you will, in certain faculty that are able to leverage their core ideas and research into commercially hugely successful companies.

I mean, John Hennessey is a classic example, who was the president of Stanford, who was the founder of MIPS. So not every faculty does that. So can you talk about what makes that distinction? of a faculty being able to do that?

Andrea Goldsmith: Yeah, absolutely. It's funny because most people view Stanford as the gold standard of faculty doing startups.

And in fact, in my department of electrical engineering and computer science at Stanford, only about 10 percent of the faculty founded companies. Many faculty worked in companies, worked in big companies, but there weren't as many founders as people imagine. Sometimes when I ask, well, what percentage of faculty in electrical engineering do you think start companies?

Oh, 50 percent or 80%. John Hennessy used to say that you'd That people think you need to start a company to get tenure at Stanford, and that's not true, but you need to start a company that's successful to buy a house in Silicon Valley, which may

Sriram Viswanathan: be true. But John was the founder of MIPS. Exactly. He got into Stanford after MIPS.

No, no, no, no. Oh, before, I'm sorry. No, he

Andrea Goldsmith: was. He was at Stanford starting from, he got his PhD at Stony Brook and came to Stanford and he did MIPS after he was tenured, I think it was in the eighties or early nineties. And then of course

Sriram Viswanathan: he's now chairman of

Andrea Goldsmith: Alphabet and he was an amazing president of Stanford, but coming back to your question.

So what I like to say is that most faculty should not do startups. There's a particular character for a faculty member who's willing to go out and do a startup because being a faculty member. at a great university like Stanford or Princeton is a really great gig. I mean, you're working with tremendous students.

You're working on whatever it is that you want to work on. You own your time. You're surrounded by brilliant colleagues and students. It's a great life. And so why would you want to give that up even temporarily to go and try your hand at a startup, which is most likely to fail because most startups fail.

So I think the character of a faculty member who I mean, there's some that are just naive and think it's going to be a cakewalk and, you know, I'll start this company and it'll be successful. And then I'll go back to my good gig, faculty job. And one thing about being a faculty member starting a company, and I felt this very acutely is that you have a backup plan.

Whereas nobody else does. So I knew with Quantenna that I could go back to Stanford. I was on a two year leave of absence, but my co founder didn't have a backup plan. My founding engineers didn't have a backup plan. And so that is something that's a little uncomfortable. I think as, as a faculty member, because you don't have as much at stake in the success of the company as everyone else, at least on paper, you don't.

I mean, for me. making Quantenna successful was tremendously important. And at one point, my co founder asked me if in order to make Quantenna successful, you have to leave Stanford completely, would you do it? And I said, well, I'll cross that bridge when I come to it, but I'm not going to say no now, because it is really important to me that this company be successful and all these people that have joined in part because of me.

I owe to them, the success of the company, everything I can do to make it successful. So coming back to faculty members, you have to be able to check your ego at the door. I mean, when I was fundraising for Quantenna and for Plume, I probably knocked on every door up and down Sand Hill Road. You get a lot of no's.

The fact that you're a faculty member with. a set of awards and papers and this kind of, it means absolutely nothing to, uh, I won't say it means nothing to investors, but it's, they're not going to invest in a faculty member unless they believe that person is going to be a successful entrepreneur and the skillset is very different.

So you have to check your ego at the door. You have to be willing to have people slam the door in your face when you're asking for money. You have to recognize that there's a lot of things you don't know, no matter how smart you are and no matter how well you know the research. That's not enough to get even from the research to an idea for a technology is a leap.

And then you've got an idea, you've got to turn that into a technology, then into a product and then into a company. And every one of those steps aren't things that most faculty members have experience with. So you have to recognize your limitations. You have to team up with other people, co founders, investors.

other founding team members and other employees that can get you through those different steps. And you have to be not afraid to fail. And like I said, there's not so many faculty members that have those qualities, but the ones that do are the ones that you see, the ones that, uh, they're successful. And if they're not successful the first time, they'll try again.

Sriram Viswanathan: That's right. I think you, as you rightly point out, I think Stanford is kind of like the Temple of all of this, right? It's, it's where all these great companies seem to be coming out of, you know, Rajiv Motwani's work on algorithms led to Google and a variety of others as well. But you seem to have now taken a pause at least in your entrepreneurial journey.

And now you've taken a bigger role at Princeton and talk to us about that. I mean, that's a. big job being the dean of one of the greatest universities in this country for engineering and for, and for the arts. Uh, so talk about your transition and what drove you to take that role as head of engineering at Princeton.

Andrea Goldsmith: I actually view it as a bit of an entrepreneurial journey, but in a different way than doing a startup. So. I was very happy with the two startups that I did. I'm very proud of what I accomplished with those two. I was ready for something different. I'd been a faculty member for 25 years. I actually retired from Stanford.

I was at Stanford long enough to merit a status. And I was thinking about how can I have a bigger impact beyond my own research or beyond a company that I could found. And that's why the Princeton job was appealing to me. So even though Princeton is an incredibly prestigious university, it's not so much prestigious in engineering.

And in fact, all of the Ivy League have engineering. as kind of second class citizens from most of the time since they were founded. If you look at the Ivy League schools from a decade ago or before that, engineering was basically on the physical and intellectual boundaries of the university. It was the, you know, the elite kids would go and study the liberal arts and the middle class kids would study engineering so they could run the machines in the industrial revolution.

Sriram Viswanathan: So you would be forgiving of people that say Princeton and. Other universities are Caltechs of the East.

Andrea Goldsmith: Well, I wouldn't, I wouldn't even say they're Caltechs of the East. I've been at four universities and all four are so different from each other in every dimension. So I did all my degrees at Berkeley, which is a big public university.

Caltech was a very small private tech university. Stanford. is a liberal arts university. Most people don't know that. Uh, but it's in the founding charter of Stanford that it's a liberal arts university, but it's of course, very well known in tech and in medicine. The med school is actually way bigger than any other part of Stanford in terms of the revenue it generates for Stanford as well as the faculty.

And then Princeton, which is an Ivy League liberal arts university with a relatively small engineering school. And so I went to Princeton because the new president in 2016, Chris Eisgruber, put in his strategic plan that he wanted engineering at Princeton to have bigger impact. It wasn't so much about rankings or particular metrics, but how can we be a great university in this era without having a really great engineering school.

And I thought, well, that's an interesting challenge to go to the East Coast that isn't known for tech at a time when New York is growing as a tech hub and a university which has the resources to grow engineering, but to do it in a different culture. than Stanford. So this is the Northeast. Princeton's a liberal arts university.

It's informal motto is to benefit the nation and all humanity, which is why I became an engineer was to build technology to benefit humanity. So to me, it was a entrepreneurial challenge to go to Princeton. And take its engineering school to new heights.

Sriram Viswanathan: That's great. We as Celesta believe in this model because as significant it is for core fundamental research to thrive in these academic institutions, which are fantastic places, the extra step that they can go to in making some of that research become commercially viable.

and benefit humanity is a big deal, which is one of the reasons why Celesta is actively working with a whole bunch of universities and, you know, the Waterloo and McMaster and other places to really commercialize. I want to ask you a very different question, which is with all this focus in this country, and for that matter, you know, lots of countries about the need for STEM education to really get to the next level.

Do you see Uh, positive signs that that's happening, or are you concerned that there is a general appetite in many of these great universities by the students to really look for, you know, the next shiny object to become an entrepreneur, to create the next Facebook. I'm not suggesting that, you know, Facebook doesn't use a lot of.

So it's not just about core technology, but it didn't come about as a way of commercializing fundamental research. It came about with an undergrad wanting to date, you know, young women in Harvard. So do you see there is a increase in focus in STEM or is there a cost for concern of the STEM focus in some of these universities?

Andrea Goldsmith: It surprises people when I say that 37 percent of Princeton students are engineering students, undergrads, and 11 percent are humanities students. That concerns me. And that's not unique to Princeton. It's just surprising at Princeton. If I told you that about Stanford, you wouldn't be surprised. Stanford actually has a little bit more percentage of undergrads studying engineering, more in the forties, and about the same in humanities.

So the students are voting with their feet, because they know where to go. the high paying jobs are. And it's more challenging if you get a degree in the humanities to have a career trajectory that takes you wherever you want to go. Now, I believe that undergraduate education is about a broad education and that you should study what you're passionate about and what you love and your career will take care of itself.

Maybe your first job won't be super high paying, but you will find a trajectory that's satisfying. Whereas if you just study engineering, even if you don't love it, I mean, I'm lucky I love engineering, but if you don't love engineering and particularly if you don't love computer science and you study computer science because you think you're going to get a great job, you're going to spend your life sitting in front of a computer.

You're probably not going to be very good at it. And interestingly, the computer science jobs are drying up. A lot of the big companies have hiring freezes right now. So Students that four years ago decided to study computer science because they thought there'd be a great job waiting for them, they didn't have a great college education because they were studying something they didn't love and they don't have the great job that they were hoping for at the end.

So I like to encourage young people to study what they love, to follow their passions. To get a very broad education because you never know when that class in philosophy or literature or history will influence what you do in your career, whatever that is. You look at Steve Jobs studying calligraphy in college as a college dropout.

That's why the Mac has the best fonts. And so. I think that there is a recognition amongst young people and their parents that college degrees are credentials that open the door for future jobs, and so there is more pressure on students from their peers and from their parents to study something valuable, to look at college as vocational.

as opposed to a broad education. And I think that's really unfortunate. And when I talk to freshmen at Princeton who have declared engineering, as I did last week, that was their first week of classes, I tell them, follow your passion and get a broad education because I am a better engineer and a better leader.

and a better citizen of the world for the philosophy and language and political science and history courses I took as an undergraduate. And there's many paths to professional success. Yeah. And so not to be so narrowly focused on what classes can I take to get that great first job.

Sriram Viswanathan: Yeah, yeah. This is truly remarkable because, uh, I mean, this is a hot topic.

As I'm sure you know, every VC, Peter Thiel onwards, are on a slightly different track where they are really encouraging students to not go to college and get that fundamental basic foundational technology and science and math understanding before they start. But if they don't have a passion for that, you know, as your example, you know, Steve Jobs, you know, became one of the greatest entrepreneurs of all time, but he had that deep passion for building unbelievably great products.

So that's,

Andrea Goldsmith: that's the. And he wasn't an engineer. I mean, and I think that that's the point. When I was at Stanford, I taught a freshman class on the art and science of engineering project design. And so these were freshmen, they didn't know very much. And because of the title of the class, it was the, one of the popular freshman classes at Stanford.

And we would get students. from engineering, from art, from science, from political science. And they saw when they came together to build a project that they needed expertise beyond technical expertise. And I think that's why I say you're a much better engineer. If you have a broader sense of the world, you understand what technology you need to develop to solve problems.

What technology will benefit humanity? What aspects of technology will not benefit humanity? And if you only study the math and science and engineering, you're not going to be as good an engineer.

Sriram Viswanathan: Yeah, that's absolutely, uh, you know, remarkable. I will be remiss if I don't talk about a favorite topic, which probably is not favorite for you, but it's one of your board positions.

You're on the board of Medtronic and you're on the board of Intel. Which has been in the news, you know, substantially in the last several months. Uh, and I'm not so sure how much, how much you can talk about it, but, uh, in general terms, I mean, you've observed this company, this industry going through just remarkable growth and innovation and all that, when you started a couple of years ago on the board, did you know what a mess it was?

Andrea Goldsmith: Um, well. I won't answer that question specifically because it's a loaded question, but what I will say is why I joined the board of Intel. I think that America, the high tech industry, and the world needs a alternative to TSMC as a foundry. And I do believe that Intel has the potential to be that alternative.

If Intel is not successful in that endeavor, the entire geopolitical landscape for tech changes. And so when I joined the board, I was really a believer in the need for Intel to come back from where it was at the time and be a strong foundry. Intel also has products and those products are also very compelling for the data center, for PCs, for wireless communication.

And I also think those are really important products. So I believe in the mission of Intel. It is going through a challenging time right now. That's one of the things I like about leadership is, is challenging times. And we have important decisions to make as a board, as the leadership team at Intel, I'm an optimist by nature, and I am optimistic that we will get through this period and emerge a stronger company.

Yeah.

Sriram Viswanathan: I mean, the famous story of Bob Noyce and Andy Grove and Gordon Moore. When Intel had a similar near death experience was when they pivoted from the DRAM business to compute and processor. So Intel has done it before. If there's one company that knows how to do it, you know, past experience would say they should be able to do this, but more generally, do you believe in the IDM model, the integrated device manufacturing model, which says semiconductor companies that inherently have a manufacturing capability bolted with the design.

versus a pure play fabless model, notwithstanding your comment about the constraint of having a sole source manufacturer with TSMC, do you believe in the segregation between fabless versus pure fab companies, or do you think the industry irrevocably is moving towards companies that don't do both? Is that a trend that you think is we're headed towards?

I think all the successful, you know, I mean, Lisa Su in AMD had a fab and one could argue, AMD shot up to what it is today, largely because of the divesting of the fab, uh, into Global Foundries. I mean, is that the model that you think is going to evolve over time?

Andrea Goldsmith: I mean, if you look at when AMD split from Global Foundries, it was a disaster for a while, right?

So it was not a given. That it was going to be the success that it is. I give Lisa a huge amount of credit for that. Obviously fabulous, uh, semiconductor companies have been wildly successful, including Quantenna. We were a fabulous company. So I think the, the question of, is this the right model for Intel?

Because that's the question on the table is the one that we're asking. I don't think it's clear cut that. It is the answer. It's not the answer and the devil's in the detail and in the timing. So again, if, if AMD had faced this decision a decade earlier, a decade later, it might've been a different answer.

And so I think that there's models, there's fewer, far fewer models for the joint model. And I think that's the challenge for Intel because it's one of. There's not so many successful models to look at and say, okay, that's the playbook we should follow. There's playbooks for companies that were one and split like AMD.

But like I said, if you look at what happened the first few years of that split, it was not at all obvious that that was the right thing to do in global foundries is not a wild success as a foundry. So I think that there's everything about the tech industry and questions like the one that Intel faces right now.

It's about timing, market. And the devil's in the details.

Sriram Viswanathan: And I guess there's no proxy for good exceptional execution, which one could argue prior to your time, you know, there may have been some missteps. Uh, before I close here, Andrea, are there more startups in you?

Andrea Goldsmith: I think as an entrepreneur, you never say never.

Uh, I, said after Quantenna that certainly I wasn't ready to do another startup right away, and I did anyway. Uh, when I look at where I am right now professionally, I really feel like higher education is at a very treacherous moment. where it's under threat by the government, where the value of higher education is unclear, where we're facing protests of students and they're very different from the protests in the Vietnam War because you have student against student, you have faculty against faculty.

This is such an important time for leadership in higher education that I feel that that is where my energies can have the most positive impact right now. So I won't say I will never do another startup because I loved both my startups. They were both a lot of fun and very rewarding and very challenging at heart.

But right now I think I'm answering my calling to where I can have the biggest impact right now. Yeah.

Sriram Viswanathan: Well, Andrea, it's been delightful talking to you, and I'm deeply honored that you took the time to, to talk to me in this podcast. And hopefully you will come back to talk to us again when there is a, you know, momentous decision on the Intel direction or in your new startup.

Yes.

Andrea Goldsmith: Okay. Very good. Thank you very much. Thank you very much. It was

Sriram Viswanathan: really fun. Thank you for being with me. Appreciate it.

Andrea Goldsmith: My pleasure.

Sriram Viswanathan: Thank you for tuning in to the Tech Surge podcast from Celesta Capital. If you enjoyed this episode, feel free to share it, subscribe, or leave us a review on your favorite podcast platform. We'll be back every two weeks with more insights and discussions of all things Deep Tech. Bye for now.

‍