I-Chip Workshop Inauguration by Mr. Lenin Patra (CTO of Marvel Semiconductors)

Up where you are today actually, so full credit to the older teachers who spent their lives teaching us. I think we can't say much. It's such a gratifying job when you know the work that they have done for us in the background, and more and more so when we grow up in the corporate ladder we realize how effective, how selfless the teachers have really been in bringing us up, not just in the technical aspect but also teaching us to be a successful human being, being very ethical, and at the end of the day a very good human being. I can't thank them enough. I clearly remember my interaction even with the questions quite a bit during my journey with Paul Jason. He's a legend, still affiliated with the institute. And one small correction: he did not recognize my face because I was actually part of IC, not EC. I did take a lot of classes from EC, that's how I'm very close to both. I think he'll remember me if he sees me. The others have done quite a bit while we were being taught. Those four years we spent in the United States are precious moments of our lives you can never forget: the night outs, hostel lives, interacting with the teachers. Very valuable, very precious.

Thank you, thank you. So a little bit of background about me: I have been with Marvell Semiconductor for the last 15 to 18 years, I don't remember exactly, I'm getting old. Almost 18 years I guess. Currently my role is Vice President, and now promoted to CTO of Marvell Technology for the Networking Fire Business Unit. But going back to where it started, let me make it very clear so I can structure my talk accordingly. Don't worry about the titles or the position. My main intention in joining this talk was to share my insight and knowledge because I have been in the semiconductor industry for the last 20 years. I felt that when Saurabh reached out to me, the least I can do for my young friends at United Calendar is to share my journey: what I have done, the failures I have seen, and of course the successes. The successes are for everybody to see from your resume and title, but what is more important is your journey and the things you should not do if you have to grow as a leader in the semiconductor industry. The second part is I'll share a little more about the electronics industry, more importantly the semiconductor industry, how it is growing, revolving, and how it will be going forward. Hopefully that will give a big picture to all my young friends of why EE or ECE is very important. Feel free to interrupt me, ask me questions. Don't be afraid, because otherwise the conversation will be one-way and I will not enjoy this. Anything, don't worry about the quality of the questions. I'm here to help.

So with that background, let me tell you a little about how the overall industry is setting up and why the semiconductor industry is really growing at a phenomenal scale and why there is a tremendous opportunity for all the young people like you, and also my MTech fellows. There is a lot of innovation that needs to happen to disrupt this industry. As you guys know, everybody today has a Facebook account, a LinkedIn account, everybody is using online shopping. You hear every now and then big news because of machine learning, artificial intelligence, big data. But you know what? All these are possible because at the end of the day, data is the biggest economy right now. But you go one layer below, peel the onion a little bit, you'll find that to generate that kind of data, to process the data, to do real-time computation and real-time analytics, hardware infrastructure is the most critical factor. Semiconductor, compute processors, movement of data from your cell phone and laptop all the way to the cloud is all done due to semiconductor innovation. Today because of the COVID effect, everybody is plugged into the digital era. Even teaching, classrooms are online. How many startups and new companies have come into the market to step up the pace? But everything, when you connect the dots, connects back to hardware infrastructure. You must be hearing that everything is in the cloud: Google Cloud, Amazon Web Services, Microsoft Azure, Facebook's cloud. What does that mean? These big e-commerce companies like Google, Microsoft, Amazon, Facebook, to scale up and provide more features, are heavily banking on cloud infrastructure. And if you go below what cloud infrastructure is, you'll find it has three major components: one, how to move data from node A to node B, mobile to the cloud, laptop to the cloud, online transactions to the cloud — that is data movement. Number two, you have to store the data, that is storage space, which boils down to memory. The third thing is data computation, to provide real-time analytics. For that you need a data compute processor, or DPU, or IPU. The three major pieces are data movement, data storage, and data processing. To make all these happen, everything boils down to the semiconductor. To move data from one place to another, that is where communication expertise comes into the picture. That is where you sense how fast you can move data from one place to another — we are talking about Ethernet, moving data in the 5G era from a handheld device to the base station. Storage part: you have hard drives, SSDs, which are memory. To scale memory, whether DDR or SRAM, everything is based on flip-flops. The fourth thing is processing. A lot of innovations are taking place because you are processing humongous amounts of data, and that is where acceleration of data processing comes in, to cross a lot of real-time analytics information. That is where machine learning acceleration or ML or AI can come into the picture. Are you guys with me? Saurabh, any questions?

Now let's talk about one layer below. The other portion is the humongous amount of data being generated in the automotive industry. I'm sure you guys are hearing about self-driving cars and vehicles. Tesla is one of the pioneers in Silicon Valley to reinvent self-driving automobiles. But imagine, to make a self-driving car is not a trivial job. Essentially you have to replicate a human with the visibility that a human can see, with the cognitive knowledge that a human has to drive a car, with a computing processor that can mimic human intelligence. That means you are generating tons of data from lots of sensors plugged into a car: cameras, radar, lidar. These sensors are in the front, side, and back of the car, generating tons of data. That data gets sent to a compute processor inside the car, not to the cloud, because latency is very important. That compute processor has to store the sensor data in local memory, do the computation, then send data to the sensor controlling the steering, telling it what direction to take. All this happens in a fraction of microseconds, not even milliseconds, to control a car. But you know what? EE, electronics, semiconductor is enabling this, making this happen. I may be one of the persons who is very passionate about the core subject I was taught, but this is exactly why a lot of people can talk about the online economy, big data, data being generated, but at the end of the day, your department, your engineering discipline, is the major force to create that kind of economy. Make no mistake, there are a lot of innovations still needed, a lot of disruption that needs to happen in that industry to reduce latency further, make processing power even more impactful, move data at much higher speed while combating challenges of electromagnetic interference and radiation. But I am really hopeful that you guys will be the ones to solve these challenging problems. Make sense? Yes, sir.

Good. Now let's go one layer below. All these things I'm talking about today are because I have done my part and I'm still doing my part, and I can see how the industry is going. But let's look back 20 years when I was just graduating from college, just like you guys, a beginner trying to get my foot into the semiconductor industry. I want to reflect on that and tell you what kind of challenges I faced. It may be different now because a lot of things are available today, but hopefully that will give you a perception of my journey, what I have done, and what you guys can do to be a leader in that industry. When I got into Honeywell, my first job was designing sensors for aerospace, a cabin pressure sensing system. When I look back, I had no clue what I was getting into. I had no idea that somebody was asking me to design a cabin pressure sensing system that would fly 10,000 feet above, carrying 300 people, with everybody's safety depending on that sensor. Obviously I got a lot of help from talented peers, advisors, and engineering managers. But if I look back, that was probably the most valuable lesson I learned: how to be fearless, how to ask for help, how to be a team player, and more importantly, how to upgrade your knowledge base very quickly. In the classroom, nobody tells you that you have a pressure sensor, you are designing a sensor to measure pressure with strain gauges, convert that to an electrical signal with amplifiers, feed that into an A2D converter to digitize the signal and give real-time feedback on how the pressure sensor is behaving. The reason I brought that example is because it reminds me of what you are being taught: the core definition of digital electronics, NAND gates, NOR gates, universal gates, how to build flip-flops and memory cells. But on this foundation, you actually build a product. You never get a sense of how impactful your four years at NIT Calicut were. I was thinking I had a good time with friends and teachers, but when the sensor came back and worked, I was extremely elated. There were a lot of challenges in the process, but at the end of the day you always go back to your basics, your fundamentals. Is that clear? Good. I'm saying this because as you move into the industry, you will see a lot of focus on tools, compilers, languages. But make no mistake, my friends: all those things you can learn on the job. What you cannot learn is the basics, the fundamentals of digital electronics, analog electronics, how a common mode rejection works for an op-amp, how a current mirror works. Those are very critical and important. I'll give another example. In our final year, there was a subject called Fuzzy Logic and Neural Networks. Trust me, I'm not lying. We were desperate to mock up the lessons as soon as possible, write the exam, get a passing mark, and leave the hell out of that subject. That's how difficult that subject was. But let me rephrase: it was not the difficulty of the subject, it was our unwillingness to learn it because we thought it was not useful. Being 21 years old, we thought we knew the whole world, we were the smartest kids on earth. You know what? When I look back, and I'm doing a lot of work in machine learning and artificial intelligence, oh boy, it's all about neural networks, fuzzy logics, building systems with a lot of neurons. Everything goes back to the basics, to the fundamentals. The other day I was joking with my wife: I'm reaching 40, but I'm still thinking about the subject I was taught 23 years back. But hey, that's the fun part of your journey and career. You'll always look back and find a place to your home, where you have been taught.

Now let's go to the reality of what is happening today and how you can gear up to the upcoming challenges. I'll stop for another 10 minutes, then we'll have questions. Today, if you look back, I was telling you about big data, data economy, data infrastructure, why data bandwidth, storage, and compute are important. The reason is simple: every single day there are tons of applications running on this infrastructure. You have Flipkart, Patanjali, Zomato, and a lot of startups becoming unicorns. But no matter what, some of this data needs to be processed to give you insights, to show how things are being analyzed at a much faster rate. Why am I bringing this up again and again? My young friends, there are tons of technical challenges waiting to be solved. I told you about latency, why it is critical. A simple example is a self-driving car. You are driving, you suddenly see a scooter coming in front of your car. As a person, you press the brake and stop. If it is a self-driving car, you expect that to happen much sooner, much faster, because you don't see a guy sitting behind the vehicle to drive it, and you are scared. But to make that happen, latency is important. Somebody needs to cut down the latency. To cut down latency, you need to process data much faster. Today, I'm not bragging, I'm working on developing a bandwidth of 200 gigabits per second on a single pair, the world's fastest 200-gig system. Somebody from NIT alone is the architect of that system. It will go everywhere in the industry. It doesn't exist today. One year from now, you can talk about it. The whole reason is to cut down latency. The second thing which is also very important is power consumption of a device. How do you reduce power? Every communication system is complex. One core mechanism in a communication system is to extract maximum SNR, maximum signal-to-noise ratio. That means you have to suppress noise as much as possible. Look back at semiconductor devices: you'll find various sources of noise: thermal noise, flicker noise, 1/f noise. These are major dominant noise sources in any electronic system or chip design. But it always comes with a penalty. To suppress noise, you have to pump more current. When you pump more current into the circuit, it burns more power. That's the challenge. When you are processing data at such a higher rate, you have to find innovative ways to reduce power. In the semiconductor world, we talk about picojoules per bit. That is the figure of merit of any system design: power, bandwidth, and the die area. The die area is directly related to the cost of a product. Performance is the bandwidth of a system, the SNR, how good the communication system is. And all those things are related directly or indirectly to power. When I talk about picojoules per bit, it's not about reducing 0.1 or 0.2 picojoules per bit. For a 100-gigabit system, the industry standard is about 5 picojoules per bit, meaning for a 100-gigabit device, it consumes about 500 milliwatts of power. But there is a catch: while it consumes 500 milliwatts, it is also driving something called a 3-meter or 4-meter direct attach cable, which in communication language we call insertion loss, about 40 dB of loss. People can say they can do it at 3 picojoules per bit or 300 milliwatts, but your immediate question should be: can you drive the same performance? So don't get too biased by the picojoules per bit. The figure of merit really boils down to power, performance, and area, and it is calibrated in the industry as picojoules per bit with respect to the loss of the system you want to drive. Does that help?

Again, very good question. It's all about basics. How many of you have taken signal processing classes? DFT, discrete Fourier transformation? It's there in the next semester for BTech. For any communication system, you have to understand the impairment of the system. What is the native application you are trying to achieve? For example, in simple layman terms, how long is the cable you want to drive? Let's say you are driving a cable or a channel. The channel is burdened with a lot of uncorrelated noise sources. If you send a square wave from A to B with a 2-inch trace, you will probably find that A is equivalent to B. But imagine sending a square wave with a 13-inch standard PCB trace. The 13-inch trace acts as a low-pass filter, and that square wave at the end will look like an impulse response after 30 inches of PCB trace. But your main goal is to replicate the waveform sent at A and see it at B. Now imagine a 100-gigahertz square wave at a 200-gigabit system, and you want to drive that with a 13-inch PCB trace. Whatever you are sending at A, at B you will see approximately nothing. You hook up an oscilloscope, you will only see noise. The challenge is how to recover the signal you sent from A to B, because the channel is low-pass filtered, lousy, inundated with ISI, inter-symbol interference, noise, jitter. But that is the beauty of it. That is why you need to study DFT, FFT. Once you have a good understanding of fast Fourier transformation, you'll find the equalization technique to recover the signal. But everything comes with a penalty. You want to solve it with the lowest power. Everything boils down to the figure of merit. So summarizing, I'm sure when you go to the final year, you'll see a lot of things. But I'm leaving you with enough to go and study. These are the challenges you will see in real communication systems.

Again, that's a very good question. There is a very old saying: you are influenced by the society you live in. When you see a lot of your immediate seniors joining software companies, there is nothing wrong with that, but you will be influenced. However, even in the software industry, if you are doing core work, it's not that easy. It's not about running a bunch of JavaScript or Python scripts. There are a lot of distributed system problems, computing system problems, new compiler design problems that need to be solved. So I'm not saying that's not the right path. But coming back to the core, if you want to stay in the semiconductor area, the first thing you have to ask yourself is your personal commitment and interest. Are you excited by putting an operational amplifier or a NAND gate on a breadboard, hooking up an input, and seeing an output? It is a difficult job, nobody wants to do it, but trust me, there is a lot of fun in making that work. It takes a lot of thought and hard work, but if you can do things right, you will be valued immensely. Everything has a sweet spot. Also, when you go into the software industry, there is an exponential number of people you will be competing with. In the core industry, you may not see that many people, but the people you work with are one hell of a ride. You'll interact with smart guys from all over the world, from MIT, you name it. Don't miss that out. Follow your heart.

It's very simple. Part of it goes back to the curriculum you are taught. Nobody teaches you how to design a compiler thoroughly, or how to design a memory completely. But it is all about your personal interest. One thing I did and I encourage you to do is read a lot of books and open courses. I'll send you some links from MIT open courses that will clear up your fundamentals. It's not that our teachers are not teaching, but there are real-life lab examples from top-tier universities that are heavily funded. You can practice those and clear up your fundamentals quickly. On top of that, there are a lot of open-source tools today for VHDL, timing analysis, signal processing. Pick up MATLAB or Python, there are signal processing libraries. You can play with them. If you don't understand, you can hit me up. Saurabh is here, he's in the second year. Go and spend time on the weekend, for example, to see what really catches your interest. It's nothing more than that. You need to do introspection and cultivate the enthusiasm to pick up the nuances.

Thank you for raising that question. Tell me a little about your PhD status and what exactly you are doing. I'll take an exercise and connect you to the India head of Marvell Semiconductor. There is a lot of 5G work we are doing there. I'll connect you to them personally. I'll see if there are internship opportunities. I'll make sure you at least have a chance to talk to them and start the interview process.

I'm not quite sure whether we can directly translate that work into what Marvell is doing. But I know for sure that there is a lot of 5G-related work, from designing baseband complex circuits to custom basic division. In fact, one of our senior digital managers is at Marvell Semiconductor in Bangalore. Let me see what I can do. I'll connect you, and also through my connections, there are a few of my peers at Intel, which does a lot of device work, especially in simulation. Let me see what can be done. I'll try my best.

Yes, probably. The first thing I'll do tomorrow is send me an email. Saurabh has my email address. He can circulate it. Feel free to ping me. As much as I can help, I'm always ready. I'm happy to meet an alumnus in the same semiconductor device area.

Again, a very good question. I am from NIT Jalandhar. What is wrong with it? I am managing a bunch of guys from the best institutions in the world, from Stanford, MIT. Don't get kind of like, be fearless. Of course, there is quality of people at IIT and top-tier universities, but if you guys put your heart out to the world that you want to do, nobody is going to stop you. And like I said, you can see me. It's not about bragging, it's about inspiration. You always find a source of information to follow. Pick your things wisely. A lot of top-tier guys, you know who is the CEO of Microsoft? Satya Nadella. He is from Manipal Institute of Technology, not from IIT. So don't get kind of like, university plays a role, it can give you a step to enter an industry, but after that it is all about how you are doing, how you are scaling up, how you are performing, how you are solving the problem. That's what matters. Nobody looks at you and says you are from IIT but you cannot solve this problem. It's all about doing things effectively. So don't get worried. Everything has a center of gravity. You can be the center of gravity.

Yes, absolutely. Every time you go in your career, knowledge is extremely important. There are two ways to learn: one on the job, but trust me, on the job there are a lot of other things you have to do because you are always being judged. Take a step back. I always encourage young guys to go do your MTech, go do your PhD, because that will teach you how passionate you are about the core work you want to do. So from that regard, doing MTech is extremely important. Going further, doing PhD is also extremely important. Another simple example: probably 90% of startups in the semiconductor sector are based on the PhD thesis that the founder worked on at a university. 90% of the time, it is when you spend a lot of energy and time to get to the very simple nuances, very simple details. So I encourage you guys to go ahead and do it. And as much as I want to help, for example, Silica, because I know she finished her PhD, go and bug her, get her advice. Did you enjoy it? What are the things you want to avoid? Have that discussion. Always reach out.

It's all about time management at that moment. How effectively you can juggle your corporate career and part-time courses. It is valuable. I'll be very honest: while I was doing my work at Marvell, I took a lot of classes from UC Berkeley. It is time-consuming, sometimes it drains you out. So I encourage you: the best you can do after your BTech is join a company and start learning. But if you can delay that for two more years, even your entry point becomes slightly better because you are judged among other MTech students. So it is about whether you want to delay getting into corporate life immediately or spend two more years in MTech, maybe find something interesting, who knows, you can come up with your own company or startup. So everything is subjective, but I always encourage you to have a good chat with a lot of seniors, do your homework, but go for it, my friends.

Well, I mean, like I said, there are three simple things to follow in life. I think anybody can practice that. Always go back to your basics. Make your fundamentals extremely strong, clear of misconceptions and misunderstandings. Most of the time, we are afraid to ask questions. It is as simple as that. We are afraid to ask because you feel the professor will get upset or say you don't know this. But I'll tell you, my friends, along with you, there are probably 40 other people in the class who did not get it. By asking questions, you are not just doing a favor for yourself, you are doing a favor for the rest of the students as well. So don't be afraid, always ask questions. Number two, it is also extremely important to enjoy the course of the journey. You'll be surprised that even after 10 years of your corporate career, when you want to have a meaningful conversation about anything, you'll always go back to your college, your home, your close friends. Number three, one of the things I realized while at NIT Jalandhar was that we did not have access to advisors, professors, or the kind of work we wanted to do. That is something I am going to help you with. I will personally arrange a few things so you can reach out to me. I'll get some time to connect you to other industry fellows, probably other startup companies that I have personally invested in. I'll tell the founders to come and talk to you, sync up, and see if any opportunities open up, even for internships. Being a relatively new NIT, we do not have a strong alumni system to get connected. That is something we, as alumni, have to bridge. The other thing is I always strongly encourage entrepreneurship. Trust me, there is no fun in doing work for somebody else. There is always fun in doing work for yourself. So keep that spirit alive. Think about it. I don't know if you guys are allowed to go to the bus stop anymore in the middle of the night, but if you are, have a cup of coffee or tea and think about these things. Extremely important. The last thing: don't get bothered by the barrier of the department. I don't know how many of you are interacting with the computer science department. Be open, reach out. Because in reality, 90% of EE students go and work on computer science stuff. So get connected to other teachers and professors in computer science. Take the help of placement, take the help of your faculties to introduce you. Extremely important.

Thank you so much. My pleasure. I think one time I'm still there, or I think it's not. That's okay. No problem. Thank you guys for giving me the chance to talk to you and interact with you. Quite fun, I really enjoyed it. I missed that there is still a phobia of asking questions, but take that out of your mind. This is just the beginning. I'll be in your journey to give you help as much as I can.