This Indian Startup is Reinventing Chip Design | Neel Gala, CTO/Co-Founder, InCore Semiconductors

1. 0:00 – 0:54

   Introduction

   1. NGNeel Gala0:00

      So imagine a chip which has a million transistors. You have to ensure that all million of them are working. Now, imagine the odds against you, right? One transistor goes wrong, it's a bug, and you spend $5 million getting the first prototype and it doesn't boot. Instead of "Hello, world," it says, "Bye, bye, world." That's it. [chuckles] Yeah, you're dead in the waters.

   2. AMAmrut0:19

      To make chip designers better, faster, but not necessarily out of a job.

   3. NGNeel Gala0:24

      Oh, I, I don't think that's possible. I, I think we are here to... We're here to rule. Uh, he signed the docs, and he said, "If you truly believe in what you're saying, you shouldn't be applying to these universities." And right then and there, uh, you know, that's where I shred those documents. If the defense does not have a processor, if we are depending on foreign technologies from China, from Israel, from XYZ, to procure these chips, deploy it in our, you know, artilleries, imagine what the world would look like.

   4. AMAmrut0:52

      Hi, this is Amrut. We are at IIT

2. 0:54 – 1:31

   Welcome to the Best Place To Build Podcast

   1. AMAmrut0:54

      Madras, my alma mater, and India's top university for people who like to build. We are here to meet some builders, ask them: What are you building? What does it take to build? And what makes IIT Madras the best place to build? [upbeat music] Hello, and welcome to the Best Place To Build Podcast. Today, I'm sitting with Neel Gala. He's the CTO of

3. 1:31 – 1:49

   Introducing Neel Gala | CTO, Co-Founder of InCore Semiconductors

   1. AMAmrut1:31

      InCore Semiconductors, a fabless based out of IIT Madras. They are a Peak XV invested DLI awardee company. Hello, Neel.

   2. NGNeel Gala1:39

      Hi.

   3. AMAmrut1:40

      Neel, um, a lot of our conversation is going to revolve around microprocessors. So can you broadly explain what a microprocessor is, and then we'll start from there?

4. 1:49 – 3:04

   Understanding Microprocessors

   1. NGNeel Gala1:49

      Sure. Um, I'll probably start with what I was told when, when I was in eighth, right? Uh, a processor is basically the brain of any electronic device, right? Uh, it's what controls that d- device, uh, how it communicates, how it interacts with the world in some sense, right? And in very layman terms, it would be, uh, sort of a, a universal translator, which can sort of convert human intent, which would typically be written in code, convert them to ones and zeros, and make your digital world dance, right? So at the end, every single device that you think is smart or you think is capable of working by itself has a processor in it, and it basically fundamentally figures out what needs to be done, how it needs to communicate to the world, and how it can serve human intentions at its best.

   2. AMAmrut2:43

      Hmm. Okay, so this, uh... Now, in the processor world, there are very few companies which actually build processors.

   3. NGNeel Gala2:49

      True.

   4. AMAmrut2:49

      They're all mega companies: NVIDIA, Intel, Arm. And in this pack of processor megawatts, um, India has developed this processor called the Shakti Processor, and you were there from day one-

   5. NGNeel Gala3:01

      Absolutely

   6. AMAmrut3:01

      ... in the team that built Shakti.

   7. NGNeel Gala3:02

      Yeah.

   8. AMAmrut3:02

      So can you tell us a little bit about what

5. 3:04 – 10:25

   What is the Shakti Processor?

   1. AMAmrut3:04

      the Shakti Processor is? How did that project start, and how did you get into it?

   2. NGNeel Gala3:08

      Very nice. Um, so I, I started my, uh, PhD, or my master's program, with, uh, Professor Kama back in 2012, 2010, 2012 timeframe. Uh, I always was, uh, excited about processors. I've, I've done some projects before that, and it's, it was, um, something very close to me, and that's something I wanted to pursue as, as my research. Uh, it- what happened to me at that point was, uh, we did not have, uh, enough resources, funding, or access to the cutting-edge stuff, right? These companies that you named would give the Tier 1 institute, like IIT Madras, access to at least two decades old of technology to do research on, and that was pointless. And at that point, you know, me and Kama-

   3. AMAmrut3:53

      Sorry, can I, can I just understand this better? As a master's student, you wanted to work on microprocessors. You needed some tech from, say, an Intel or IBM or whatever-

   4. NGNeel Gala4:03

      Yes

   5. AMAmrut4:03

      ... and you didn't have access to it because they wouldn't share their proprietary tech?

   6. NGNeel Gala4:07

      Yeah, definitely the geography played a bigger role. Uh, but also, if-

   7. AMAmrut4:10

      Or they might share it with Stanford, but not with IIT Madras.

   8. NGNeel Gala4:13

      True.

   9. AMAmrut4:13

      Okay.

   10. NGNeel Gala4:14

       And, uh, even if they wanted to, we needed resources, right?

   11. AMAmrut4:17

       Right.

   12. NGNeel Gala4:17

       We needed funding to do it, right?

   13. AMAmrut4:19

       Right.

   14. NGNeel Gala4:19

       So I, I always say this, right? Innovation warrants curiosity, but curiosity warrants funding, [chuckles] warrants resources, right? They have to come out of somewhere. And, uh, it just wasn't the right time at that point, and that's when me and Kama and, uh, a couple of other folks, we thought-

   15. AMAmrut4:36

       I, I know you, you're calling him Kama, but you're talking about Professor Kamakoti.

   16. NGNeel Gala4:39

       Yes. I will... Uh, I share a very close bond with him. Uh, I've spent more than a decade in his office with him on a lot of things. Uh, he's given me the liberty to address him. Uh, hopefully, I will- I won't get a call after the podcast. [laughing]

   17. AMAmrut4:55

       [chuckles]

   18. NGNeel Gala4:56

       But yeah, and, uh, that's when we decided if it doesn't exist, we should build it.

   19. AMAmrut5:00

       Okay.

   20. NGNeel Gala5:00

       Why not? Uh, let's venture into that. So we started this effort. We started doing research on, you know, what's open, what's not there. Uh, my master's and PhD was, now had to be on something different. I, I was looking at approximate computing. That was the gung ho at that point. But, uh, for this processor project, we started looking at PowerPC. They had something called OpenPOWER. Now, that's a, that's a funny thing because the name's OpenPOWER, but you've got to pay to have access to it, and it was more than 600 pages long, and there was almost a-

   21. AMAmrut5:32

       It's like also AI.

   22. NGNeel Gala5:33

       Yeah. I mean-

   23. AMAmrut5:33

       Open AI-

   24. NGNeel Gala5:34

       ... what would you do with it? It's, it's gonna take me multiple PhD avatars to finish that book, right? And just around the corner-

   25. AMAmrut5:40

       One, one sec, sorry.

   26. NGNeel Gala5:40

       Yeah.

   27. AMAmrut5:41

       Can I understand that a little better? Uh, OpenPOWER is a processor?

   28. NGNeel Gala5:46

       And... Yeah, so OpenPOWER is an instruction set.

   29. AMAmrut5:49

       Okay.

   30. NGNeel Gala5:49

       An instruction set is the fundamental layer between hardware and software.
6. 10:25 – 15:15

   The InCore Journey

   1. AMAmrut10:25

      It's amazing. Uh, I know that, uh, in the s- in the circles that I really know, they of course know you as somebody who maintained the code for a while also.

   2. NGNeel Gala10:32

      Yes.

   3. AMAmrut10:33

      Um, [clears throat] but eventually you sort of, uh, moved on and started this, uh, company-

   4. NGNeel Gala10:39

      Correct

   5. AMAmrut10:39

      ... called InCore, right?

   6. NGNeel Gala10:40

      Yeah, yeah.

   7. AMAmrut10:41

      So what does InCore do?

   8. NGNeel Gala10:42

      So InCore is in the business of providing RISC-V solutions.

   9. AMAmrut10:46

      Okay.

   10. NGNeel Gala10:47

       Uh, and by solutions, I wanna emphasize that we're not just a core company. We don't build just the core, but we build subsystems, we build SoCs, and in fact, everything else that goes around to making a product or enabling a product based on this file.

   11. AMAmrut11:03

       So, sorry, sorry, sorry. I have a question here. Um, I want to unpack these words. So there'll be... There's a microprocessor, there's a core. I want to understand what was the difference between a microprocessor and a core?

   12. NGNeel Gala11:13

       Okay.

   13. AMAmrut11:13

       Then you also mentioned, uh, a SoC, and then we said, uh, microchip, so-

   14. NGNeel Gala11:18

       System, yeah

   15. AMAmrut11:18

       ... a subsystem, right?

   16. NGNeel Gala11:18

       Yeah.

   17. AMAmrut11:18

       So can you just build it up? Like, if I look-

   18. NGNeel Gala11:20

       Sure

   19. AMAmrut11:20

       ... at a final product like this, this-

   20. NGNeel Gala11:23

       Correct

   21. AMAmrut11:23

       ... this, uh, microphone-

   22. NGNeel Gala11:25

       Yeah

   23. AMAmrut11:25

       ... what is inside that?

   24. NGNeel Gala11:27

       So it all starts with a chip-

   25. AMAmrut11:29

       Mm

   26. NGNeel Gala11:29

       ... right? And that's what you're gonna get as a, as a device from the fab, our packaged, uh, device, and you're gonna solder it and build this chassis. Now, inside the chip, there are gonna be multiple components-

   27. AMAmrut11:39

       Mm

   28. NGNeel Gala11:39

       ... right? One of the most important components is, is the core. The core is sort of the computational engine, the primary engine.

   29. AMAmrut11:46

       Okay.

   30. NGNeel Gala11:47

       Right? And everything around it is, is a subsystem or a SoC, right? Uh, let me give you an analogy, right? Um, since we are in campus, let's imagine, uh, the admin building at IIT Madras. That's the core.
7. 15:15 – 19:44

   What is an IP?

   1. AMAmrut15:15

      you're building around all these things. Quick question: What is IP in this business, and how do you sell IP, and how does someone buy IP? How is it protected?

   2. NGNeel Gala15:25

      Okay. Um, IP is, you know, as it stands, intellect property, intellectual property, and, uh, uh... But the default definition would mostly just be, you know, patents and copyrights-

   3. AMAmrut15:36

      Right

   4. NGNeel Gala15:36

      ... that you have on something-

   5. AMAmrut15:36

      Yeah

   6. NGNeel Gala15:36

      ... that you've created, you've created.

   7. AMAmrut15:38

      Yeah.

   8. NGNeel Gala15:38

      And, uh, a lot of the industry depends on legal, uh, records to protect themselves-

   9. AMAmrut15:44

      Yeah

   10. NGNeel Gala15:44

       ... in some sense. Uh, but for me, personally, I believe, uh, the IP at InCore, it's, it's the intellectual... It's the, it's the productized intellect that exists at InCore, right? Can we think in terms of what we can sell, right? Not think in just pushing boundaries of research, not just building the next big thing, but, uh, it's unsellable of sorts.

   11. AMAmrut16:09

       Mm.

   12. NGNeel Gala16:10

       So yeah, the industry definitely, uh, piggybacks on IP as a way to protect itself. Uh, we encrypt them, we obfuscate them.

   13. AMAmrut16:19

       But how do you, how do you... Oh, okay. Okay.

   14. NGNeel Gala16:21

       Yeah.

   15. AMAmrut16:22

       So I, I'll ask you, like, from the way I understand. Suppose I run a chip design company, say, and I want to buy your IP. What is it that you actually give me?

   16. NGNeel Gala16:32

       Great question. So we don't give you a physical chip. Let me make that sure, right? An IP company gives you softwares.

   17. AMAmrut16:38

       Okay.

   18. NGNeel Gala16:39

       So in order to build this IP, we use a lot of tools and languages, things like Verilog, VHDL. We use languages like Bluespec system, Verilog, Chisel sort of things. Uh, we write code in that.

   19. AMAmrut16:52

       So the, the code that you are developing in all these languages will go into my code?

   20. NGNeel Gala16:55

       Yes.

   21. AMAmrut16:56

       Okay.

   22. NGNeel Gala16:57

       So what we will give you is, we will optimize, we will write this code super efficiently, so that when you take this code to the fab, you get an amazing, you know, low-power chip or high-performance chip and whatnot. That's the idea.

   23. AMAmrut17:10

       So related question on that, when you give me your IP as code, what stops me from just duplicating it?

   24. NGNeel Gala17:16

       So when we give you some IP, we definitely are aware of, uh, leak or security problems when it comes to, uh, you know, giving it to our customers. We do some of eng- some part of engineering there, where we obfuscate the code so that it's not easily readable. Uh, we encrypt it so that you can't decode it back, you can't figure out the intrinsics of it. But these are all things to protect ourselves, but they make the life of our customers very difficult-

   25. AMAmrut17:42

       Mm

   26. NGNeel Gala17:42

       ... right? Because they've put this IP into a bigger system, and once you don't have visibility inside it, uh, they're not sure if the system is broken because of this or something else, right? So more often than not, uh, you have to share source, right? Now, the good part about InCore is, uh, we give them something called a Verilog IP. It's written in Verilog. Verilog is an industry-standard language. As I said, we develop all of our IP in Bluespec. So we develop all of our IP in Bluespec system Verilog. Now, that's a high-level language. That's what we keep closed. We convert this into Verilog, and we share this with our customers, right?... So this way, we've protected ourselves from the high-level stuff. We don't expose this out. That is truly our IP, that is truly something of our own secret sauce. We've built tooling or we've adopted tooling, uh, which helps convert BlueSpec to low-level languages, and that's what we make visible to our customers. So they have access, but it's, it's gonna be very difficult for them to reverse it.

   27. AMAmrut18:44

       And this is common in the industry, right? Like-

   28. NGNeel Gala18:45

       Absolutely.

   29. AMAmrut18:45

       I mean, they-

   30. NGNeel Gala18:46

       They make it difficult for people to-
8. 19:44 – 20:55

   How is InCore Different From Companies like NVIDIA?

   1. AMAmrut19:44

      NVIDIA and ARM are IP companies, right?

   2. NGNeel Gala19:47

      Yeah.

   3. AMAmrut19:47

      Uh-

   4. NGNeel Gala19:47

      They're fabulous chip companies, yeah.

   5. AMAmrut19:49

      Yeah, and I think they, they scale because the, the chips they were making at some point scaled, uh, because of a particular use case, NVIDIA scaling-

   6. NGNeel Gala19:57

      Yeah

   7. AMAmrut19:57

      ... because of, uh, the crypto boom, and AI, uh, ARM scaling because of the GSM phone boom.

   8. NGNeel Gala20:03

      Yeah. Yeah, yeah. So, um, yeah, I was just gonna add, uh, each of those are, uh, they're slightly different than InCore. We are, we are still ahead in our- in the supply chain. Uh, NVIDIA owns the chips that it gets manufactured outside of itself.

   9. AMAmrut20:18

      Mm.

   10. NGNeel Gala20:19

       It may not own a fab, but it definitely goes... It builds the IP, builds the GDS, goes to the fab, gets the chip, and then sells the chip.

   11. AMAmrut20:27

       Okay.

   12. NGNeel Gala20:28

       Uh, we basically provide an IP which allows customers to go build the chips themselves-

   13. AMAmrut20:34

       Okay

   14. NGNeel Gala20:34

       ... through the fab or not. And it takes massive cycles, uh, of iterations, of optimizations, and, uh, a lot of effort to realize that IP into a value add in the final chip.

   15. AMAmrut20:46

       Can I just come back to something you spoke about earlier? You said that the thing that you're focusing on is reducing the design life cycle, and you said four months, and I said maybe-

   16. NGNeel Gala20:54

       Yeah

9. 20:55 – 25:39

   What Does a Microprocessor Design Life Cycle Look Like?

   1. AMAmrut20:55

      ... maybe it feels like it should be more.

   2. NGNeel Gala20:56

      Yeah.

   3. AMAmrut20:57

      Um, to few weeks... No, no, few days.

   4. NGNeel Gala21:00

      Right.

   5. AMAmrut21:00

      So w- what is it that... What, what does that mean? Uh, how long is a design life cycle? How are you reducing it?

   6. NGNeel Gala21:07

      Amazing question. So, um, we are in the phase of spec to RTL freeze, right? Design freeze. Okay, now, when you wanna build a chip, there are multiple phases. You start with a spec, uh, you get the first design, then you start physical design, then you go to the fab, the fab goes to packaging, then you build chips, you build boards, and then you do a demo, right? This entire life cycle is, is at least 18 months. You could cut corners and, you know, skip a few steps and maybe do it under 12 months, but it's at least that much. Now, within these 12 to 18 months, it takes you at least the first four months just go and figure out what you wanna design, right? You're gonna start with, "Okay, I wanna build a smart bulb, and I wanna build a chip for that. I wanna have some intelligence into it. I'm gonna need some USB, I'm gonna need Wi-Fi, I'm gonna need Bluetooth, I'm gonna need, uh, some other memory devices and stuff." You build this nice little diagram. That diagram now needs to get converted to actual design.

   7. AMAmrut22:05

      Mm.

   8. NGNeel Gala22:05

      Someone has to go procure these IPs because you're not a design house. You have-

   9. AMAmrut22:09

      Sure.

   10. NGNeel Gala22:09

       You don't build everything in the world. You go license it. Whether it works or doesn't work, someone else has a competitive part. You go buy that, you evaluate them-

   11. AMAmrut22:17

       Yeah

   12. NGNeel Gala22:17

       ... you see whether they get-

   13. AMAmrut22:18

       In fact, the things you spoke about, Bluetooth is a-

   14. NGNeel Gala22:20

       Exactly

   15. AMAmrut22:20

       ... protected IP.

   16. NGNeel Gala22:21

       Yeah.

   17. AMAmrut22:21

       You can't... Even if you knew how to-

   18. NGNeel Gala22:23

       Exactly

   19. AMAmrut22:23

       ... you can't design it.

   20. NGNeel Gala22:24

       You can't design it in day one. So then you've got the pieces together, you've, you've basically now have a recipe, and then you start executing the recipe. You put it together.

   21. AMAmrut22:33

       Yeah.

   22. NGNeel Gala22:33

       Now, you gotta check whether it works-

   23. AMAmrut22:34

       Mm

   24. NGNeel Gala22:35

       ... before you actually go to silicon. So you're gonna have a software team which is deployed, and they're gonna build a complete stack of the application, the firmware, the kernels, the... all of that stuff. Then you're gonna have people working on emulation. So you're gonna prototype it on FPGA, so that you can see what the silicon would look like. How is it gonna feel? Does it need timing? Does it need other parameters? And then there's a bigger beast of verification, right? Am I sure that everything works? Am I bug-free? And, and that's the intent of these teams. Uh, each of these things can easily take four to six months, right? And the problem with the conventional way is, they all start one after the other. They're se- they're serialized. You first have to get the design, so the verif starts. Only when the verif starts, you can then put it on FPGA. It's only on the FPGA you can do software-

   25. AMAmrut23:24

       Mm

   26. NGNeel Gala23:24

       ... right? So by the time your software team is kicked in, you're already two months in or three months in. What we're trying to do at InCore is, we wanna reduce this time period to weeks.

   27. AMAmrut23:33

       Okay.

   28. NGNeel Gala23:34

       Right? And within this week, what happens is, you can iterate, you can keep changing stuff. We will give you collaterals for software, for hardware, for emulation on day one. So your teams are not just on the bench, they can actually start hacking as soon as possible, right? The biggest add over here is, you still have four months, which means you can now focus more on your innovation, right? You wanted to basically put in your AI accelerator, but you needed a Bluetooth, you needed a Wi-Fi, you needed stuff so that, you know, the rest of the world works. Your value add, your innovation was in this accelerator. You got derailed by doing all these other things that the world does, and you spent four months there. I reduced that to four weeks or four days. I've given you back three and a half months to focus back on your accelerator.

   29. AMAmrut24:22

       So now-

   30. NGNeel Gala24:23

       So now you've got more liberty to innovate.... and that's where InCore is going, right? We-
10. 25:39 – 28:00

    The Truth Behind Building Chips

    1. NGNeel Gala25:39

       it's always a massive learning curve, right? So which is why I always say that today, building a chip is a marathon, and it's a marathon with a condition that you can't quit. You just can't step out because you've invested millions of dollars to see that first damn prototype.

    2. AMAmrut25:53

       Right.

    3. NGNeel Gala25:54

       Right?

    4. AMAmrut25:54

       And that's not even the opportunity cost. You're talking about actual money as well.

    5. NGNeel Gala25:57

       Actual investment. That is the patient capital that this industry demands. I need 18 months of investment, no questions asked, and we are on a mission to make that marathon a sprint.

    6. AMAmrut26:09

       Mm.

    7. NGNeel Gala26:09

       I want people, I want engineers in the future to truly commoditize the thinking of silicon, right? Today, people innovate on software, they should be able to innovate on silicon the same way. They shouldn't just fear it and survive it.

    8. AMAmrut26:23

       That's a very interesting-

    9. NGNeel Gala26:24

       Interesting point

    10. AMAmrut26:25

        ... that I want to just focus on a bit.

    11. NGNeel Gala26:27

        Mm-hmm.

    12. AMAmrut26:27

        In software, the innovation cycle is so quick, and the resource requirement is so less, that if you come up with a prototype that doesn't work, you just junk it and move on. In silicon, junking a project is very tough because, you know, you're, you're basically-

    13. NGNeel Gala26:42

        Bankruptcy. [chuckles]

    14. AMAmrut26:43

        Yeah.

    15. NGNeel Gala26:43

        It's close to a bankruptcy problem.

    16. AMAmrut26:44

        Yeah. And if you're in a bigger company, that could simply mean that your team is going to be quite broke.

    17. NGNeel Gala26:48

        Oh, definitely. Yeah, this, this industry is extremely risk-averse, right? There have been multiple instances in the past, uh, where large MNCs, right, who've been chip vendors and OEMs, had to recall millions of their chips, and that's, that's the biggest problem, right? A single bug... So imagine a chip which has a million transistors. You have to ensure that all million of them are working.

    18. AMAmrut27:14

        Mm.

    19. NGNeel Gala27:14

        Now, imagine the odds against you, right? One transistor goes wrong, or one element, or one gate goes wrong, it's a bug, and you spent $5 million getting the first prototype, and it doesn't boot. Instead of "Hello, World," it says, "Bye bye, World." That's it. [chuckles] Yeah, you're dead in the waters. You've got to restart. That entire cycle starts again, and-

    20. AMAmrut27:34

        Interesting

    21. NGNeel Gala27:34

        ... the aversion makes you a lot more paranoid. I wanna make sure that there is no bug. I spent eight months of verification cycles with as much confidence to say that, "Yes, I have hopefully gone through anything that a customer would go through, any corner cases, and my chip would still work."

    22. AMAmrut27:52

        Mm, very nice. Uh, I'm looking at, uh, Deepak's notes again. Thank you, Deepak. [chuckles] Uh, it talks about something

11. 28:00 – 32:00

    Exploring the Concept of SoC Generator Platforms

    1. AMAmrut28:00

       called an SoC generator platform. So can you tell us about what this term means? And-

    2. NGNeel Gala28:04

       Yeah

    3. AMAmrut28:04

       ... I know you're explaining the same thing, but can you sort of give it, give it the... How, how do I go from that term? Like, if I'm in the chip industry, and I read SoC generator, what do I understand from it?

    4. NGNeel Gala28:16

       Oh, wonderful. Uh, as I said, you know, the subsystem is, is much bigger than the core, right?

    5. AMAmrut28:22

       Mm.

    6. NGNeel Gala28:22

       The core becomes the fundamental building block. Uh, but a subsystem or a system on chip, that's what an SoC stands for, uh, is much closer to the end goal, right? You've brought in all the IPs, you've brought in together. Now, the problem is, every time you start a project, you're gonna be using the same IPs, but differently, right? It's like a recipe. Whether you add the salt first or the sugar later, and jaggery, right? It gives you a different flavor and different trade-offs. The generator that we have built allows you to do those trade-offs at rapid speed.

    7. AMAmrut28:53

       Yeah.

    8. NGNeel Gala28:53

       Right? I could go around, play around, change, and I could iterate and say, "Okay, if I mix these components in this order, do I get good performance? If I change the order or if I bring in something else, do I get better area? Do I get better power?" And that's where the USP is, right? So we've un- we've dealt with customers who have gone through 20 iterations in a matter of twen- two weeks, right? They started off saying, "Okay, I want to match this part from the industry." We give them a solution, and they said, "Okay, now can I optimize on this?" "Okay, two days, let's iterate that." We give them a new draft. "Okay, that's good, but I've gone up in power. Can we go and restart?" "Okay, let's do something else."

    9. AMAmrut29:32

       Mm.

    10. NGNeel Gala29:32

        This would typically take weeks of iteration.

    11. AMAmrut29:34

        You're, you're saying it, uh, but I think PPA is a-

    12. NGNeel Gala29:38

        Power, performance

    13. AMAmrut29:38

        ... term in the industry, right?

    14. NGNeel Gala29:39

        Yeah.

    15. AMAmrut29:39

        So what does PPA mean?

    16. NGNeel Gala29:40

        So PPA stands for power, performance, and area. Uh, you have to be good on all of them. You need to be low power, uh, you need to be high performance, and low area.

    17. AMAmrut29:50

        Uh, low power because you want to use lesser battery-

    18. NGNeel Gala29:53

        Yeah

    19. AMAmrut29:53

        ... which is obvious, like on a smartwatch-

    20. NGNeel Gala29:55

        Yeah

    21. AMAmrut29:55

        ... it would mean a lot more time.

    22. NGNeel Gala29:57

        Absolutely.

    23. AMAmrut29:57

        And that's a feature that customers would... In fact, in a remote operation, you may want your chip to work for five years-

    24. NGNeel Gala30:03

        Yeah

    25. AMAmrut30:03

        ... 10 years.

    26. NGNeel Gala30:03

        Some battery-operated systems.

    27. AMAmrut30:05

        Uh, performance means?

    28. NGNeel Gala30:08

        Uh, mostly the frequency. When people tell you, "I'm running at 3 gigahertz and 4 gigahertz of processing power," that's basically how fast you can consume data, compute it, and, you know, infer it.

    29. AMAmrut30:18

        Mm.

    30. NGNeel Gala30:18

        So-
12. 32:00 – 36:00

    Why Can’t AI Take Over the Semiconductor Industry

    1. AMAmrut32:00

       question I have on that is that why can't AI do it overnight?

    2. NGNeel Gala32:03

       Amazing question, right? Uh, if you, if you look at where AI is being deployed in this industry, right, it is still a co-pilot. It is... It's not something that, uh, we can completely depend on to produce an output which can be consumed without human intervention.

    3. AMAmrut32:20

       Mm.

    4. NGNeel Gala32:20

       Right. So it helps a lot with verification, so it, it helps identify corner cases, but because of the risk-averse nature of the industry, we still have to have humans in the loop. We still have to go through those iterations, and regressions, and confirmations whether we are actually doing what we set out to do, right? And this dilemma exists because of hallucinations in AI, right? That's, that's a big problem. I don't know whether it's going to give me the optimal or not, right? So if we start deploying Gen AI or generative AI to these problems of building chip design, it's probably, uh, not going to take off too soon. But having said that, uh, there have been, you know, learning algorithms which have been deployed in the industry over the decades-

    5. AMAmrut33:08

       Mm

    6. NGNeel Gala33:08

       ... but these have been limited to the big companies like Intel, NVIDIA. They definitely have much more smarter algorithms, smarter tools, which look at their data sets and help their engineers produce better based on its own internal learnings, right?

    7. AMAmrut33:24

       Okay.

    8. NGNeel Gala33:24

       But does a startup like ours have access to that AI, and does... Do we have access to that smart tool? Definitely not. That's proprietary.

    9. AMAmrut33:30

       So you're saying that Intel has 40, 50 years of designs, it can train an internal tool on those designs to generate-

    10. NGNeel Gala33:38

        Yeah

    11. AMAmrut33:39

        ... future designs?

    12. NGNeel Gala33:40

        Correct.

    13. AMAmrut33:40

        Which may or may not be accurate or reproducible.

    14. NGNeel Gala33:42

        Yeah.

    15. AMAmrut33:42

        But in any case, the rest of the world doesn't have access to that, uh, that AI, which-

    16. NGNeel Gala33:48

        You can't just take GPT or, you know, the O3 model or the Claude models, because they, they have not been trained enough. There is just lack of training data sets in the open source world, which tells you how to do the best designs out there. A lot of it is academic effort, so you have to know that if the AI was trained on academia, it's going to give you an academic output.

    17. AMAmrut34:07

        Right.

    18. NGNeel Gala34:08

        To productize it, you still need the people there. So it's going to be a long way off, and hallucinations don't make the cause better. Uh, reproducibility makes it even more difficult to rely on AI. But as you said, uh, AI can always be used as an interface to the SoC Gen tool. That's something that, uh, we hope to work on in the next, uh, couple of months, is, "Can you give me a prompt which converts it to an input-

    19. AMAmrut34:33

        Mm

    20. NGNeel Gala34:33

        ... to this tool?" But the tool does the generation, the tool is what creates the SoC. I would use the AI to analyze the inputs and outputs to the tool, but never deploy it to generate the, uh, SoC itself.

    21. AMAmrut34:46

        Sure. So it becomes like a co-pilot, which is what you started with.

    22. NGNeel Gala34:49

        Which is what I said.

    23. AMAmrut34:50

        Mm. To make chip designers better, faster, but not necessarily out of a job.

    24. NGNeel Gala34:56

        Oh, I, I don't think that's possible. I, I think we are here to... We're here to rule quite some decades before AI takes over our jobs. [chuckles]

    25. AMAmrut35:04

        We being chip designers?

    26. NGNeel Gala35:06

        Yes.

    27. AMAmrut35:07

        Right.

    28. NGNeel Gala35:07

        That's, that's my motivation to all freshers to, you know, join this industry.

    29. AMAmrut35:11

        Nice. Um, yeah, I think, uh, sorry [chuckles] just to pick on that, the semiconductor industry or the talent pool in India is actually growing very, very fast.

    30. NGNeel Gala35:21

        Yes.
13. 36:00 – 40:50

    The Choice to Pursue Higher Studies at IITM Instead of Abroad

    1. NGNeel Gala36:00

       Mm-hmm.

    2. AMAmrut36:00

       India has always been rich on electronics and semiconductor engineers, right? Um, but more often they just migrate out of India. And, uh, you told me of a very interesting and exciting story of you and a paper shredder.

    3. NGNeel Gala36:14

       Yes. [laughing]

    4. AMAmrut36:14

       Can you tell us that story?

    5. NGNeel Gala36:17

       Absolutely. Um, so this was, uh, this is while I was working with Professor Gama as a project staff, and, uh-

    6. AMAmrut36:25

       That would be what, 2000?

    7. NGNeel Gala36:26

       This was around, uh, 2010, uh, timeframe. I had, I'd graduated in 2010. 2011 is when I was, uh, like any other peer of mine, I was, was behind the American dream, you know, "Let's go do master's there." I started applying. I had this list of, uh, six, seven universities, ambitious, moderate, backup universities. At that time, we had to, you know, write recommendation letters by hand.... and the prof had to sign it, and we had to courier it.

    8. AMAmrut36:53

       Yeah, and Professor Kamakoti's recommendation letter carries a lot of-

    9. NGNeel Gala36:56

       Oh, definitely. Yeah, it was, uh... And I was- you know, I would do anything to get that letter. So I, I- one fine day, I had two more universities, which were my backup universities, and, um, I was waiting for him, uh, in his office. He comes out of his office at, uh, close to 8:00 PM, right? And he tells me, "Neel, what's up?" And I tell him, "You know, I have these two universities. You've sent six. Why don't you send these two? These are my backup." "Then why are you going there? I mean, these are backup. These are even worse than doing a master's in IIT Madras. Why do you wanna go there? Uh, why, do you want to abandon your parents?" I was like, "No, no, I will come back for sure." Uh, he's like, "Nah, if you wanna come back, then why go at all? I mean, what can I not give you that these universities are giving you?" And that made me realize, uh, this is, this is a prof I spent 18 months with. Uh, he sees a potential in me. He believes that I can do much better here, and yet I'm on this, uh, wagon to go try my luck out in a foreign land with foreign people, with, uh, zero idea of where my future is. And, and that very instant, uh... He always keeps a shredder because he does a lot of, uh, confidential stuff there. Uh, he signed the docs, and he said, "If you truly believe in what you're saying, you shouldn't be applying to these universities." And right then and there, uh, you know, that's where I shred those documents, and I stopped applying. Uh, fortunately, his recommendations gave me, uh, access, or got me admissions to an Ivy League school as well. Uh, but Professor Kama is a force of nature, right? It's, it's impossible, you walk out of his room with less patriotism, right?

    10. AMAmrut38:31

        Yeah, he's very inspiring.

    11. NGNeel Gala38:32

        You have- you're just filled with heart of patriotism, and it would be very tough to, you know, wean that off. And, uh, that day forward, I, I think we were aligned on, you know, the belief that India has to create and not just consume-

    12. AMAmrut38:47

        Mm

    13. NGNeel Gala38:47

        ... especially on some of the most fundamental technologies of the digital era.

    14. AMAmrut38:50

        And this is, of course, before the Shakti project started-

    15. NGNeel Gala38:53

        Yeah, yeah

    16. AMAmrut38:53

        ... so after that, you, you guys collaborated further on that.

    17. NGNeel Gala38:54

        Yeah, then it became sort of, uh... I mean, we doubled down on each other, right? Uh, he was the access, he was the guy with the vision, and he would convince the government strategic sectors to, you know, talk to us, to work with us. And he was able to make a, a call. Before Atmanirbharta was- became a norm, you know, he was literally driving it for us, saying that, "If the defense does not have a processor, if we are depending on foreign technologies from China, from Israel, from XYZ to procure these chips, deploy it in our, you know, artilleries, imagine what the world would look like."

    18. AMAmrut39:30

        Yeah.

    19. NGNeel Gala39:31

        And that's, that's enough to set paranoia [chuckles] in some of these sectors. And that's how Shakti truly took off. It, it started off as a mission to satisfy indigenous requirements first.

    20. AMAmrut39:42

        Yeah.

    21. NGNeel Gala39:43

        Yeah.

    22. AMAmrut39:43

        I mean, if you say it like that, it's a very scary thought that our power plants, our grid systems, our CCTV cameras-

    23. NGNeel Gala39:51

        Your Aadhar card, your bank [chuckles]

    24. AMAmrut39:53

        ... Aadhar database, your banking database runs on infrastructure built on chips. Maybe these are chips that we trust the vendors from, but sometimes maybe not.

    25. NGNeel Gala40:02

        It's mostly not, because, because of how technology is. You can't reverse engineer a chip.

    26. AMAmrut40:08

        Mm.

    27. NGNeel Gala40:08

        If I give you this... Okay, this-

    28. AMAmrut40:11

        Correct

    29. NGNeel Gala40:11

        ... this connects to WFI and Bluetooth, I tell you that.

    30. AMAmrut40:13

        Fair. Nice.
14. 40:50 – 45:14

    From the Lab to Market: A Journey

    1. AMAmrut40:50

       Okay, so in your journey, you have worked a lot on the Shakti ecosystem. Um, I mean, congratulations on reaching so far. The Shakti processor ecosystem has also inspired a bunch of startups, yours included.

    2. NGNeel Gala41:00

       Yeah.

    3. AMAmrut41:00

       Uh, you've moved on, uh, from just being the Shakti, um, the, the person who was heading the Shakti Technical Group, to now being the CPO of InCore. InCore itself has moved on to being a IP provider company across a bunch of things, and now you are betting on this, uh, SoC generator idea-

    4. NGNeel Gala41:18

       Yeah

    5. AMAmrut41:18

       ... which is shortening the design life cycle for a bunch of chip companies, right?

    6. NGNeel Gala41:22

       Yes.

    7. AMAmrut41:24

       Okay, that's a, that's... [laughing]

    8. NGNeel Gala41:26

       It's a short brief. [laughing]

    9. AMAmrut41:29

       Lot. Uh, I wanna ask you, in this process, you've gone from a researcher, uh, to a sort of a researcher working on a product, to a technical co-founder, and that's a journey in itself, right?

    10. NGNeel Gala41:42

        Absolutely.

    11. AMAmrut41:42

        Because you're now running a business.

    12. NGNeel Gala41:44

        Yes.

    13. AMAmrut41:45

        Uh, it's a difficult business, but it's still a business.

    14. NGNeel Gala41:47

        Absolutely, yeah.

    15. AMAmrut41:48

        So how has that journey been?

    16. NGNeel Gala41:51

        It's, um, it's been an uphill for sure, right? Um, once you've been in academia for so long... I was doing a PhD. I mean, I, I spent more than a decade in RISELab, and it's a comfortable space. It's, it's very tough to get out of that space because Kama spoils you. He will give you all the luxuries in the world, all the access to everything, and it's... You, you simply start believing that tech is the answer to everything.

    17. AMAmrut42:14

        Mm.

    18. NGNeel Gala42:14

        If you build the tech, that should solve all problems in the world, and that's where we get it wrong.

    19. AMAmrut42:20

        Mm.

    20. NGNeel Gala42:21

        Right. The transition... So what academia teaches you to think is what's possible. What entrepreneurship demands you to think is what's needed.

    21. AMAmrut42:32

        Mm.

    22. NGNeel Gala42:33

        Now, I was lucky that I was at the inflection point where the answer to both of them was, was RISC-V or Shakti, and, you know, I could connect the dots. Uh, but when I switched on, or when I, when I put on the entrepreneurship hat, uh, when I became a founder and a business owner, uh, listening to the customers, trying to understand their pain points, because they can't be as articulate as you would like to be, right?... you truly have to probe them enough. You have to sit them in, you know, sit with them in the same room, and then realize that this is a problem which requires a much different solution than what I'm doing. Now, there are two ways. You could always say that, "Okay, I'm not gonna do this because I've built tech for something else." But if you could take two steps back and said, "If I make some changes and if I can address this, then this market opens up."

    23. AMAmrut43:19

        Sure.

    24. NGNeel Gala43:19

        So the ability to listen, right? You need to sell before you build. That's a very difficult thought process to an engineer.

    25. AMAmrut43:30

        Mm.

    26. NGNeel Gala43:30

        For a good part of my journey, I was scared to go sample with customers until I got it perfectly done, right? Because I would come up with my corner case scenarios, I would cook up things that could go wrong, and I would start fixing them. And I would keep telling Deepak, "No, no, one more month. No, no, one more week. I just need to do this, I just need to do that." Uh, but at some point we put our foot down and we said, "You know, wherever it is, let's just, um, throw it over the wall. Let's see what happens, what breaks," right?

    27. AMAmrut43:54

        Mm.

    28. NGNeel Gala43:55

        Amazingly, it was a very different response than I expected. People were warm to it. People wanted to see how we could do things differently because they saw value-add even otherwise, right? So for anyone who is making this transition, I would suggest, you know, stop, stop thinking that if you get the tech right, the, the job is done. The job- the job only starts when you know what needs to be deployed and what is sellable.

    29. AMAmrut44:21

        Very nice. I'm reminded of this story, um, a really old story of how Bill Gates sold, uh, DOS to IBM before they built DOS.

    30. NGNeel Gala44:29

        Yeah!
15. 45:14 – 49:21

    The Road to Being Silicon-Proof

    1. NGNeel Gala45:14

       I think, uh, a customer doesn't just bring, uh, money to the table. I think they bring a lot more value add, because in this industry, people will bet on you if someone else has bet on you. So it's very difficult to get the first customer because, uh, you are a startup who has just got the first tape out done, first silicon of your own, but no one else has actually deployed it-

    2. AMAmrut45:39

       Right

    3. NGNeel Gala45:39

       ... no one has placed a bet on-

    4. AMAmrut45:40

       I have heard a word called, uh, market-proof?

    5. NGNeel Gala45:44

       Silicon-proof.

    6. AMAmrut45:45

       Silicon-proof?

    7. NGNeel Gala45:45

       Yeah. Market-proof, silicon-proof. So someone should have placed a bet on you, uh, so that others can. And finding that first guinea pig, finding that first customer who truly wants [chuckles] wants... Who believes in your story-

    8. AMAmrut45:58

       And is risking a recall-

    9. NGNeel Gala46:00

       Yeah

    10. AMAmrut46:00

        ... basically.

    11. NGNeel Gala46:00

        And at that point, you are selling your soul. You're doing everything possible, right, for that one customer. He tells you to build a quantum chip, you as well start learning quantum the next day. Because that's the one thing that will make your third, fourth, fifth product sell at rapid pace-

    12. AMAmrut46:16

        Mm

    13. NGNeel Gala46:17

        ... right? So that's a very different mindset, and that's a very different way, and you have to be fluid to adjust to market trends.

    14. AMAmrut46:25

        Nice.

    15. NGNeel Gala46:25

        Right? If you start off saying, "Oh, I'm gonna build this big-ass processor or this tiniest core," that's a bad thing to do, right?

    16. AMAmrut46:32

        Neel, also in this journey, you have raised funding.

    17. NGNeel Gala46:35

        Yeah.

    18. AMAmrut46:36

        Uh, you have investors backing you up. You've, uh, you, you're a DLA awardee.

    19. NGNeel Gala46:40

        Yes.

    20. AMAmrut46:41

        Um, and, uh, that means you have to not only listen to your customers, but also to your investors and, uh, the government, which-

    21. NGNeel Gala46:49

        Uh, yeah

    22. AMAmrut46:50

        ... for good reasons-

    23. NGNeel Gala46:51

        Yeah

    24. AMAmrut46:52

        ... may want to take a direction, may want you to take a direction that's not aligned to your, uh, customers, and so on. So is that also a journey that you have had, too?

    25. NGNeel Gala47:00

        Oh, absolutely, absolutely. I think, um... So once we got invested by Peak XV, they have these surge, uh, sessions where they, they take you around meeting with other, uh, eminent personalities in the field and as founders. You know, one of the investors mentioned that as a founder, you're constantly selling. You know, whether- your every waking minute, you are selling to your employees that you're a good company to be in, you're selling to your investors that they've invested in the right place, you're selling to your customers that you're the right piece of product to buy. And managing this is, is chaos, right? Because as you said, I would have spent a couple of years building a product and the customer wants something different, and they're willing to pay upfront. Now, do I make the decision to give up this? Because I do see potential, but that is the first customer which requires us to go in a different direction. Um, and you're gonna make a call. That's where, you know, people like Deepak and, uh, Madhu would come in and they would say, "You know, what does the market look like? What are the trends?" And that's where you stop being technical and you say, "Stop thinking of engineering hours. Think about the company first," right? "Stop putting yourself as, uh, the guy who needs to implement it, rather than the guy who needs to sell it." And that makes life a lot more-

    26. AMAmrut48:16

        Mm

    27. NGNeel Gala48:16

        ... clearer, right? So managing expectations is definitely difficult. Uh, people have, uh, pushed us or pulled us in different directions. Uh, I can tell you, uh, one of our initial, you know, customers that we were talking to, they spent a lot of time with us, but didn't materialize, right? And we thought we could win them over with our tech. [clears throat] And what we didn't do was, uh, uh, we didn't get enough guarantees from them, saying that, you know, they will actually go to market, they will actually deploy us. Uh, and these are learnings that help us. I don't think there is a shortcut or there's a book which tells you, "Don't do this, and do this." I think every journey is of their own.

    28. AMAmrut48:59

        ... and, uh, actually, even if you knew that, even if you knew that, you know, you're going down- you're being led by a larger player down this path, at the moment, you're very small, you don't have as much-

    29. NGNeel Gala49:10

        Right

    30. AMAmrut49:10

        ... bargaining power. Like, you can't go and tell a large company that I want this in writing.
16. 49:21 – 50:20

    Tackling the Fear of Attempting

    1. NGNeel Gala49:21

       and they don't have a rational behind their answer-

    2. AMAmrut49:25

       Then you-

    3. NGNeel Gala49:25

       They should throw up red flags, right?

    4. AMAmrut49:27

       Yeah.

    5. NGNeel Gala49:27

       So I think it was the fear of not attempting. I think the fear of failure is so high with us-

    6. AMAmrut49:32

       Mm

    7. NGNeel Gala49:33

       ... we don't even attempt to ask those questions.

    8. AMAmrut49:35

       Mm.

    9. NGNeel Gala49:35

       It's okay to be thrown down. It's okay to be told you're small. You are small, right? No, no harm in accepting that, but you don't wanna hear it, and those are the mindset changes that you need to make. You should be okay in people telling you your reality. You being in cognizant of where you are and where you need to be, that's very different than writing code.

    10. AMAmrut49:58

        The advantage that you have, though, is that the, the product you're building or the IP you're building is globally scalable.

    11. NGNeel Gala50:04

        Yes.

    12. AMAmrut50:05

        Uh, right? And it's not a, it's not a location-specific thing, so once it starts picking up and once customers st- once your first big clients come in, uh... a- and I, I, I don't know how close you are-

    13. NGNeel Gala50:16

        Mm

    14. AMAmrut50:16

        ... to those kind of milestones.

    15. NGNeel Gala50:18

        Right.

    16. AMAmrut50:18

        But then it, in semiconductor

17. 50:20 – 51:52

    Scaling Up in the Semiconductor Industry

    1. AMAmrut50:20

       in general, the scale happens very quickly. It happens-

    2. NGNeel Gala50:23

       Absolutely.

    3. AMAmrut50:23

       It takes a long time, but once it happens-

    4. NGNeel Gala50:26

       Yeah

    5. AMAmrut50:26

       ... it's like-

    6. NGNeel Gala50:26

       Yeah

    7. AMAmrut50:27

       ... very fast, right?

    8. NGNeel Gala50:28

       Absolutely. It's... The first customer is the, is the toughest one, because he's the guy who's taken the risk, and he becomes your well-wisher if he's happy, right? And that well-wisher is gonna close deals at, at half the speed than you can imagine, right? If it typically takes six-month sales cycles today, as InCore, uh, post our first customer wins, couple of wins that we've had, now the sales cycles are, like, two months, a month. Because they know someone has taken your IP, gone to product, they've made product, they've made money out of it, and they've not complained. These guys want the same stuff or they want something more modified. Now they have confidence that InCore is really a semiconductor company to scale. Now, when it comes to being global, uh, the space that we are in, InCore is in, uh, we are more of, uh... I don't think India is the right market today. I think we're, we are a little ahead of time. [clears throat] Right now, the wave of, uh, uh, in the Indian semicon, uh, sector is mostly on chip designers. You're gonna find a lot of companies and products who will actually sell chips. Uh, this is the first wave, and this is by obvious by design because of, you know, the kind of, uh, incentive schemes that the government has come up with for fabs, for OSATs, and stuff like that. So people have been pushed to build products first, right? No matter what it is. I believe the second wave is where they

18. 51:52 – 54:18

    The Shift Towards Innovation on Silicon

    1. NGNeel Gala51:52

       would want to build designs smarter or better.

    2. AMAmrut51:55

       Mm, mm.

    3. NGNeel Gala51:55

       That's where we will be, you know, uh, playing a major role. This always happens globally.

    4. AMAmrut52:01

       That, that makes sense because in my mind, uh, India is importing a lot of electronics-

    5. NGNeel Gala52:05

       Yes

    6. AMAmrut52:05

       ... so the government is thinking first let's-

    7. NGNeel Gala52:07

       Let's just-

    8. AMAmrut52:07

       ... slow down on the import.

    9. NGNeel Gala52:08

       Yeah.

    10. AMAmrut52:08

        Yeah. Uh, instead of importing, like, 2% of our GP-

    11. NGNeel Gala52:11

        Correct

    12. AMAmrut52:12

        ... of our imports as just m- mobile phones, let's make them here.

    13. NGNeel Gala52:15

        Yeah.

    14. AMAmrut52:15

        Uh, and that would mean that we need to make those-

    15. NGNeel Gala52:18

        Yep

    16. AMAmrut52:18

        ... chips first.

    17. NGNeel Gala52:19

        Yeah.

    18. AMAmrut52:20

        And then we would go down that-

    19. NGNeel Gala52:21

        And then you would innovate.

    20. AMAmrut52:23

        Yeah. Then you would see we are making so many chips, but the, but the underlying IP is coming from outside-

    21. NGNeel Gala52:29

        Correct

    22. AMAmrut52:29

        ... so then let's focus on the IP.

    23. NGNeel Gala52:30

        Yeah, yeah.

    24. AMAmrut52:30

        But you're saying that globally, IP is a big play?

    25. NGNeel Gala52:34

        Oh, definitely. Globally is an IP... And this happens especially in the last, uh, decade, where this concept of, uh, custom silicon or domain-specific architectures have become a default, right? Uh, the idea is very simple: I, I can't deploy a, an i7 processor for a, you know, smart bulb. That's a different class of process.

    26. AMAmrut52:54

        Sure.

    27. NGNeel Gala52:55

        Right? So everything are now... People are gonna go lower in their stack. They're gonna start by innovating of the software or the application at the kernel, but now they have to innovate at the silicon to stand out.

    28. AMAmrut53:08

        Mm.

    29. NGNeel Gala53:08

        Right? So people will want st- a different AI accelerator. They will want different security protocols for, you know, handle, you know, the POS machines, right? Point-of-sale machines that you have. You want higher amounts of security there. You want to enable Wi-Fi there. You want to enable better experiences there.

    30. AMAmrut53:24

        Yeah.
19. 54:18 – 59:10

    What Does RISC-V Mean in the Global Context

    1. AMAmrut54:19

       mean in its global context? Uh, and, uh, why is it so important? Why is, why is, why is India focusing so much on RISC-V?

    2. NGNeel Gala54:27

       Amazing question. Um, RISC-V, beyond an open standard, is an open opportunity, right?

    3. AMAmrut54:35

       Open opportunity, okay.

    4. NGNeel Gala54:36

       It's an opportunity to play. So let me go back, right? Instruction sets were released, uh, much earlier, right? They, they came back with IBM 360s, where the idea was, "I wanna have iterations on hardware and software independent."

    5. AMAmrut54:55

       Okay.

    6. NGNeel Gala54:56

       "So I can keep building more better software, whether it works on Gen 1, Gen 2, Gen 3 of the device." So you can have two teams-... but they always communicate with the same PDF, and that PDF tells you A for apple, B for ball, sort of a thing. It tells you the instructions that you can use to build software and instructions that the hardware should understand. Now, once that was introduced, everyone realized that's, that's the Holy Grail, right? If this is kept proprietary, you can control market. If you change this, you can improve performance. If you enhance this, you can get better software, and this became what people started keeping closed source, keeping as a secret source.

    7. AMAmrut55:38

       So basically, all the tools that chip designers use will use this ISA, right?

    8. NGNeel Gala55:43

       Yeah.

    9. AMAmrut55:44

       Okay.

    10. NGNeel Gala55:45

        So for example-

    11. AMAmrut55:46

        So I may not be actually writing an ISA, but anything that I use will-

    12. NGNeel Gala55:49

        Yes

    13. AMAmrut55:49

        ... somehow-

    14. NGNeel Gala55:50

        Every time you write any code-

    15. AMAmrut55:51

        Uh

    16. NGNeel Gala55:52

        ... it has to get converted to something we call machine code.

    17. AMAmrut55:54

        Which is w-

    18. NGNeel Gala55:55

        Which is-

    19. AMAmrut55:55

        As per the ISA.

    20. NGNeel Gala55:56

        Yes.

    21. AMAmrut55:57

        Got it.

    22. NGNeel Gala55:57

        So that machine code tells you that if software has written add instruction, convert it to one, zero, one, one, one, whatever the sequence is.

    23. AMAmrut56:05

        Okay.

    24. NGNeel Gala56:05

        That's the idea, right?

    25. AMAmrut56:06

        And so if a company-

    26. NGNeel Gala56:06

        It tells you what that behavior is.

    27. AMAmrut56:08

        So if a company is controlling the ISA, it's indirectly controlling all the tools, all the ecosystem, all the libraries-

    28. NGNeel Gala56:14

        Absolutely

    29. AMAmrut56:14

        ... uh, everything. So basically it's like total control.

    30. NGNeel Gala56:18

        It's a monopoly.
20. 59:10 – 1:02:30

    The Genesis of RISC-V

    1. AMAmrut59:10

       you were part of the international RISC-V community as a operator.

    2. NGNeel Gala59:14

       Yes, yes. I was, uh... [clears throat] So IIT Madras was one of the founding members of RISC-V. So because we adopted before RISC-V was there, we were amongst the first 10 or 12 entities across the world who became the founding members of RISC-V Foundation. Uh, personally, at InCore, uh, I was also contributing as the vice chair for the compliance of RISC-V. Basically, if you have to use the trademark for RISC-V, claiming that your chip or your product is RISC-V compatible, you have to pass a bunch of s- tests, and that entire test suite, infrastructure, automation, was all built by InCore today.

    3. AMAmrut59:52

       So in your period that you've been associated with RISC-V, is it something that's grown? Are there actual chips out?

    4. NGNeel Gala59:56

       Oh, definitely.

    5. AMAmrut59:57

       Okay.

    6. NGNeel Gala59:57

       Yeah. The... For example, the [clears throat] there's just an article yesterday. NVIDIA has shipped more than a billion RISC-V parts just in 2024.

    7. AMAmrut1:00:08

       Nice.

    8. NGNeel Gala1:00:09

       So it's-

    9. AMAmrut1:00:10

       And what does it-

    10. NGNeel Gala1:00:10

        ... yes

    11. AMAmrut1:00:10

        What does it mean for a count- for a country? Like, why is RISC-V India's ISA?

    12. NGNeel Gala1:00:15

        Oh, yeah, I was just getting to that. So as I was saying, RISC-V being open gives you the, uh, the permission to play, right, in a space that you've been locked out for decades, for geographical reasons, for talent reasons, for brain, brain drain and all of that. Uh, all the smart people have been going abroad or we've been a services sector. RISC-V has opened the opportunity to create, to innovate, and that now becomes a way to segue, uh, you know, segue into what I was- what we were talking earlier, is the fear and the paranoia that the strategic sectors have. They can now truly control, they can ensure that a chip does not have a backdoor, right? You are now completely in control of the entire loop from spec to production, right? And that becomes an important factor because you can now build something that is more India-specific. See, I think the problem that we've been facing is we keep importing foreign solutions for Indian problems, right? And-... Professor once told me this amazing question, uh, amazing, uh, story, right? He said, "If you go to the US, right, you- the tap water, it's the same water that goes into your sink, and it's the same water that goes into your washrooms and what you drink and everything. It's, it's all from the same source." Imagine if the campus did that in India. It wouldn't fly, right? You would have Brahmins, or you would have these religious folks who would come and say, "Oh, you know, how can I use the bathroom water to, you know, do my religious activities?" Or, "How can I drink that water?" You need separate sources. In that context, I don't think applying a foreign solution to Indian problems is just gonna fly out of the box. At some point, you have to innovate for Indian problems. You can only do that if you control the ecosystem.

    13. AMAmrut1:02:00

        You need to build ground up.

    14. NGNeel Gala1:02:01

        Exactly. So, for example, what, uh, Bhavish Aggarwal did with Krutrrim. He said, "You know, I, I own the e-bikes. I can build stacks," but he now wants to control silicon, right? He wants to control data. Why should it go to the Europe borders, and why should the data centers be there? Why can't it be within us? Why should I share my data to anyone else? Because this is Indian stuff for Indian problems, for Indian solutions, and we need a lot more resources for us to innovate in that space.

    15. AMAmrut1:02:28

        Fair enough.

21. 1:02:30 – 1:11:47

    Closing Thoughts & Reflections

    1. AMAmrut1:02:30

       Nice! This has been a great conversation. Um, and, of course, Professor Kamakoti was here, uh, uh, and, uh, we spoke to him also about RISC-V and about how India needs to be self-sufficient. It's not enough for us to say that we have built a Shakti processor, so we know how to do it. It also needs to go into our power plants-

    2. NGNeel Gala1:02:49

       Yeah

    3. AMAmrut1:02:49

       ... and, uh, the commercialization has to happen, and i- it's not enough to say that we can build CCTV cameras on our own. We have to actually go and do it.

    4. NGNeel Gala1:02:58

       Yeah.

    5. AMAmrut1:02:59

       Uh, and so that's amazing that you're also, uh, pa- you're a part of the part... part of the journey where, uh, at a university level, you are thinking it through, but also part of the journey where it's actually going to industry and-

    6. NGNeel Gala1:03:09

       Yeah, yeah

    7. AMAmrut1:03:10

       ... the commercialization is happening, and, uh, long journey.

    8. NGNeel Gala1:03:13

       Oh, definitely. It's, it's, uh, it's a fruitful one. It's, it's one bearing, uh, a lot of, uh, good memories, a lot of good learnings, I would say.

    9. AMAmrut1:03:24

       Would you advise young people to get started on the same journey now?

    10. NGNeel Gala1:03:28

        Oh, most definitely. I think, uh, the world is, is much more accepting, is much more accepting than it was a decade back. I think the challenges we faced, uh, have reduced today. I think, uh, the government is much more aware, the investors are much more aware of semicon, of [clears throat] of startups coming out of India, the-

    11. AMAmrut1:03:48

        You mean the problems that a startup founder would have?

    12. NGNeel Gala1:03:52

        I think we spend a lot of time educating [clears throat] investors, educating the government as to why what we're doing is of strategic value, why this is of commercial value, why this has to be the next big thing, right? The awareness has grown. The access to information has grown. People are participative. People are forthcoming. Uh, the DLI, for example, is, is an amazing initiative, right? I mean, we've saved millions of dollars of tools spend, so... Otherwise, with these- with the 3 million seed round, it's impossible for us to have such on our own. We are definitely much higher on, uh, in terms of output when it comes to the dollar invested, so-

    13. AMAmrut1:04:34

        Per dollar-

    14. NGNeel Gala1:04:36

        Returns are much-

    15. AMAmrut1:04:37

        Much higher.

    16. NGNeel Gala1:04:37

        So even-

    17. AMAmrut1:04:37

        So even investors-

    18. NGNeel Gala1:04:39

        Yeah

    19. AMAmrut1:04:39

        ... who are, uh, in Europe and US.

    20. NGNeel Gala1:04:41

        So all to the young people, I think the, the biggest, uh, comment that I would give or suggestion I would give is, uh, stop fearing failure. Uh, you would be surprised the number of people who are watching for your success, right? It is amazing that the patriotism, the Atmanirbharta, is, is actually there. It's, people are forthcoming, people are willing to be your well-wishers, even if they've not seen you. As long as it's in India, as long as, as you're willing to put in those hours, those efforts, and, and don't be afraid to ask the silliest and the stupidest questions. You're not an AI, and AI can learn the entire BTech degree in 10 minutes.

    21. AMAmrut1:05:24

        Mm.

    22. NGNeel Gala1:05:24

        Humans have to go through it, right? It's impossible to do that, so don't fear the effort. That's what I would say.

    23. AMAmrut1:05:29

        To that student in university now, who wants to go abroad, uh, who wants to, um, who's going to his professor now and taking that rec letter like you did, uh, 15 years back, um, I know you're saying it, but do you think that... H- how do they build the courage to sort of stay here and build, knowing that maybe a US life is easier?

    24. NGNeel Gala1:05:54

        That's, uh-

    25. AMAmrut1:05:55

        I, I mean, and-

    26. NGNeel Gala1:05:55

        Yeah

    27. AMAmrut1:05:56

        ... especially in semiconductors, right? Because in semiconductors, the, the bulk of the industry is in the Valley and, whatever, in the US.

    28. NGNeel Gala1:06:04

        No, definitely, I, uh... This is, this is a conversation I have more often, uh, even today because, uh, we hire freshers a lot. You know, they, they have these dilemmas: "Okay, maybe I'll work for two years, and then I'll go to the US." So I still have the same conversations, and, uh, to be honest, I, I don't think there is one answer that I can give, but, uh, society pressures definitely play a, a major game, right? Do you have commitments at home? Do you have pressures which warrant you to earn? Because doing a startup here, you have to have the courage to go through a lot of hurdles financially, society-wise. You're gonna have constant, you know, uh, chachas in the next door saying, "Oh, you've not done this," or, "You've not hit this milestone. You don't have a kid, you don't have a car, you don't have a house." Those things still exist.

    29. AMAmrut1:06:57

        Right.

    30. NGNeel Gala1:06:58

        Uh, it's, it's, it's gonna be a journey which is gonna, gonna be, uh, rewarding at least five, six years later. It's not instantly rewarding. So to the guy who is- who really wants to go to the US, I would say-... if you truly believe that an IIT or an BITS or, you know, any top tier one institute can't give you that here, uh, maybe, yeah, because it, it takes a very different mindset to be here, right?

Episode duration: 1:11:52