Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
A deep dive into the Indian semiconductor dream with Shashwath
200 Plays1 year ago
Discover why microchips are essential for the AI economy and how they've propelled Nvidia to become the world's third most valuable company. Shashwath is the founder of one of India’s top microchip companies and in this episode, he goes deep into the semiconductor business. Gain insights into the importance of microchips for AI and their impact in shaping India's technological self-reliance.
Get notified about the latest releases and bonus content by subscribing to our newsletter at www.founderthesis.com
Read more about Mindgrove:-
1.Designed-in-India chips for the World
2. Mindgrove Technologies’ Shashwath TR on Making India’s Most Secure Chip
3.Redefining Chip Design: Mindgrove’s Tailored Approach to Innovation
Recommended
Transcript
Introduction to the AI Economy and Semiconductors
00:00:00
Speaker
Hi, I'm Shashvit, CEO and co-founder of Mindgroup Technologies.
00:00:17
Speaker
I'm sure you've heard a lot about how data is the new oil. I would take this a little further and say that AI is the engine that is fueled by data. And the nuts and bolts that make this whole thing move forward is microchips. Microchips or silicon or semiconductors are the backbone of the new data powered AI economy. And that is the reason why the chip maker NVIDIA is the third most valuable company in the world.
00:00:43
Speaker
To better understand this fascinating semiconductor ecosystem, your host Akshay that speaks to TR Sashwat, the founder of Mindgrove Technologies, which is among a handful of Indian microchip companies. Stay tuned for a masterclass on all things silicon on this episode of the Founder Thesis podcast.
Mindgroup's Focus and Foundation
00:01:09
Speaker
Give me an elevator pitch of Mine Grove. So Mine Grove is a fabulous semiconductor company, which is among the first to make advanced microcontrollers from within India. And what is unique about us is that we are actually doing all of the design work over here. And we actually own the design of the microcontroller completely, the system on chip completely.
00:01:40
Speaker
we are targeting our microcontrollers towards computer vision and signal processing at the edge.
Understanding Semiconductors and Their Components
00:01:49
Speaker
Okay, so for listeners who are not familiar with a lot of language that you used, what's fabulous?
00:02:00
Speaker
First, could we start with a semiconductor 101? What's a semiconductor company all about? What types of companies exist in the world? What's the evolution? Should I start with the technology? Yeah. Okay. So semiconductor, I mean, we use it all the time. Everything from your LEDs to your iPhone is a semiconductor.
00:02:29
Speaker
And basically, what a semiconductor is, at its heart, is a very special kind of material that could be a, like for example, the most commonly used one would be silicon, where it is not a great conductor of electricity. It's a decent conductor. It's not like a metal. But what it does is it is selectively conductive.
00:02:58
Speaker
So for example, it may conduct in one direction, but not in another. So if you have two terminals, it may conduct in this direction, but not in this direction. And what that does is it allows us with a clever arrangement of different types of semiconductors. It allows us to create things like transistors, diodes, switches, and mostly
00:03:26
Speaker
transistors because that then allows us to chain those transistors into actually useful things like logic gates. That forms the core of everything that we've come to know as computing industry and our entire electronics industry. Everything that revolves around this particular tiny little thing called a semiconductor. Now, yeah. One quick question.
00:03:55
Speaker
A layperson's understanding of transistor is like a radio, which is obviously not what you're referring to. What do you mean by a transistor? A transistor is a very specific type of a semiconductor, which has three layers. There are two types of semiconductors, typically semiconductor materials. One is called a positive and the other is called a negative.
00:04:18
Speaker
So the most simple kind of transistor is called a PNP or an NPN, where you have a positive semiconductor, which is sandwiched between two negatives or two positive sandwiching one negative. And from there, we have a more complex one called a metal oxide semiconductor field effect transistor or MOSFET, which is actually what we get to use in most circuits today.
00:04:44
Speaker
the older type which is called a bipolar junction transistor is kind of antiquated. So what this does is this device, so it has three terminals, right? Each of these elements has one terminal. And if you use one terminal as an input, you can use it to control the flow of power or current between the other two terminals. Think of a valve,
00:05:16
Speaker
in a piping system. And imagine if that valve was also controlled by the pressure of water coming in on itself. And the more pressure that comes, the more force it allows through the other side. So that's really what it is. And
00:05:35
Speaker
It finds all kinds of crazy uses. The obvious reason why we use it in radios, for example, is because it's an amplifier. So you can have a switch or you can have a valve which is very, very sensitive to small currents coming in when you're taking it from the air.
Advancements in Integrated Circuits and VLSI
00:05:54
Speaker
and use that to regulate the flow of a much larger flow of current from a battery or mains power into your audio system. So this valve, so if the audio moves very lightly on this side,
00:06:12
Speaker
on this input side, it'll create a very large variation on the output side. So that's where everybody, the most well-known thing about a transistor being a radio comes from, right? Other ways of using a transistor are as a switch. So when we use it as a switch, what we do is, if the pressure, in this case, pressure would be voltage or current.
00:06:41
Speaker
So if the voltage coming in is less than a particular amount, you don't conduct at all on the output. If it breaches a particular threshold, then you start conducting fully. So that's called a saturation mode. You're either on or you're off. And arranging transistors operating as switches together in clever arrangements, you'll get gates.
00:07:07
Speaker
which are logic gates, which are logic elements such as you could have an AND gate which says you have two terminals as input and only if both of them are simultaneously on will the AND gate produce an output. The opposite would be an OR gate where if one terminal is on and the other is off, any one or both have to be on and then it will produce a true output or a non-output.
00:07:37
Speaker
If both of them are off, it will produce an offer. And then you chain all of these together to form logical elements. So when you're talking about binary arithmetic, you can create a circuit which adds two numbers using a bunch of OR gates and what is called an exclusive OR gate. So all of these are created using the transistor at the lower level, right?
00:08:07
Speaker
Okay, so essentially computers understand binary math, which is zero one. Yeah. And that zero one is represented as on state or off state. Yes. And the reason that they understand the binary math, they don't actually understand anything. It's just that the circuit has been designed to produce binary arithmetic to be able to operate in the binary domain.
00:08:35
Speaker
So actually the origin, the very first couple of computers that were built will operate, did operate in decimal system, but then it was found to be far more efficient and far easier to program and we did it in binary. Okay. Understood. Okay. Right. So now it's these transistors and putting them together that actually creates a computer. And, um,
00:09:03
Speaker
Earlier, when we're in college and we're doing BTEC, we have electronics lab. We get these transistors which are this big. Those were one state of the art, but now they're basically used to train young engineers. Anything which is being used in serious work today would be something which is called a very large integrated circuit. What an integrated circuit means is to create a not gate, where
00:09:33
Speaker
If the input is zero, the output is one. Or if the input is one, the output is zero. So to create a NOT gate, you would have two transistors in, acting together. So that is, in this case, these are no longer the BJTs, but they're what we call MOSFET transistors, M-O-S-F-E-T.
00:09:59
Speaker
And there are two types of MOSFET transistors. There's a positive MOS and a negative MOS. The current technology uses both together. And so that is called a CMOS, complementary metal oxide semiconductor.
00:10:15
Speaker
And so when you actually create, like if you're creating an adder, then you would have several of these. If you're trying to add two eight digit decimal numbers together, eight bit numbers together, you would have something like 30-ish gates total. And that would be about 60 to maybe 120 transistors. So you,
00:10:43
Speaker
When you create a modern semiconductor, when you create a VLSI semiconductor, what you do is you integrate all of these onto one piece of silicon, which is the substrate. And then you actually create the circuit, which you draw the transistors on top of that substrate, right? So what happens when you do that is that the circuit becomes small.
00:11:13
Speaker
Because it's smaller, it uses less power. Because it's using less power, it is also more efficient. And because it's smaller, the distances are smaller, it is much faster for current to reach from one end to the other, so it is much faster. So the game in all semiconductors is to reduce the size of the transistor as much as possible.
Chip Manufacturing and Photolithography
00:11:34
Speaker
And that gives you speed, it gives you power, it gives you all the things that we're used to when we buy an iPhone this year, which was twice as fast as the iPhone last year.
00:11:45
Speaker
Okay, amazing. And so this VLSI is what we call a chip. This VLSI is the technology that creates a chip. And so what happens is this VLSI, it's a specific technology, right? And it's a specific manufacturing process now that gets used.
00:12:12
Speaker
And basically, these circuits are etched on two silicon wafers. The current state of the art is a 300 mm wafer. So that means it's about one foot in diameter, think of a dinner plate. And on that dinner plate, a modern processor, let's say a simple microcontroller like what we are building right now, that would have like a million transistors.
00:12:42
Speaker
the Apple M1 chip or your typical Intel chips would have upwards of a billion, maybe close to 10 billion kind of transistors. It's enormous. And that fits into what size wafer? Typically, so the wafer is 300 mm in diameter. And typically you have chips, chips could range anywhere between two square mm
00:13:11
Speaker
one square mm, I've heard of even smaller than one square mm, all the way up to behemets like the Intel Xeon, which is rumored to be something like 800 square mm. So now it's actually the manufacturing process that comes next, right? So what they do to manufacture this is they create a, this is called photolithography.
00:13:40
Speaker
So you have a base silicon and then you expose you draw the circuit as a mask that is like a die and then you use ultraviolet light to expose the silicon. Wherever it is exposed it turns from silicon to silicon dioxide and that can be etched away using acid.
00:14:00
Speaker
and then on top of that you grow a little bit more material and then you etch again and then you add more circuit and then you etch it off and then you add more circuit to etch it off, you grow it in layers like that. As you grow it, the circuit builds itself.
00:14:16
Speaker
And like I said, let's say there is a 10 by 10 chip, right? That's 100 square mm. So you'll have a 10 by 10 mask. And the 10 by 10 mask will be put into the UV photolithography machine with the silicon wafer coming under it. It'll keep etching each. So each layer will have its own mask. So it'll etch each mask. Could be anything between 50 and 100 layers, depending on the chip and the complexity and so on.
00:14:46
Speaker
And it's like 3D printing. It's more like CNC than 3D printing. But the printing industry is a very good analogy for the business of semiconductors. And I'll get to that. So once this is done, you have these 100 mm 10 by 10 squares on basically edged onto a dinner plate.
00:15:13
Speaker
And then we cut it kajukatli style. We cut it like that. It's called dicing. You cut it using lasers. And then you get these 10 by 10, 10 by 10, 10 by 10 little wafers. They're called dyes. And these dyes are the actual chip. After you get the dyes, you have to package them typically in either metal, plastic, or ceramic.
00:15:42
Speaker
And then you have the little metal leads that come out of it or the metal balls that go on the bottom. And that is a chip that everybody can recognize whenever you know your electronics breaks and you see the chips inside, that's actually what it is.
Roles in the Semiconductor Industry
00:15:58
Speaker
Usually the packaging is much bigger than the chip, right? So, and depending on the packaging could be much bigger or several times bigger.
00:16:11
Speaker
and there are various packaging technologies. So once you package it, then it is taken out and then you put it on a PCB and you solder it. And then you actually, with all the other components that go onto that particular board, and then you get what you would actually think of as, let's say a computer or a phone or whatever.
00:16:35
Speaker
Now, this entire process, which I have described to you happens through multiple entities, multiple companies, multiple countries, multiple agencies, multiple disciplines, because it has everything. All of this etching and growing and silicon, all of that is hardcore material science. It's hardcore chemical industry process engineering.
00:17:02
Speaker
which I have only a very peripheral knowledge about because I'm an electronics engineer and my skill is in design. So taking the die, the door silicon die, and then packaging it in that plastic package is a specialty all by itself. And so there are agencies which do this, agencies which do this, and then there are agencies which do the design. So what we have now
00:17:30
Speaker
in the landscape is you have the design company on one end who's creating the design and then they send it to a semiconductor foundry, which is where the wafer is taken and etched. And then the foundry gives you bad eyes, typically, unpackaged eyes. Then it is taken to an assembly testing, marking and packaging
00:17:59
Speaker
company who does assembly, which is putting it into the package, testing. So they take that package, they put it into a testing machine and check whether the circuit is actually working per spec. Then they do the markings on it. And then they sort it by functionality, whether it's, that's why you get your different speeds of computers run, right? They do the marking and then they package it.
00:18:30
Speaker
I'm making markings on the chip. Is it like writing a label essentially? Yes, but it's enough of a specialization that it gets its own name. And then packaging. Packaging is where they put it into a tray or a reel of like 100, 1000, 10,000, 100,000 chips. And then they shut it off.
00:18:59
Speaker
So each chip is individually tested. Yes, it should be. But the tests at this stage are very rudimentary. So now a company like Intel has all of this, actually only Intel to date among the major companies, has all of this in-house.
00:19:28
Speaker
design, manufacturing, packaging, everything is in house, right? So there were other such companies in the past, but either they have all shut down or they have split up. So this is called an integrated device manufacturer, IDM.
00:19:49
Speaker
I think Intel is probably the last man standing of the lot. There are a couple of others like Texas Instruments, which have their own foundry for some lines, but not for everything. Everybody else, right? AMD, Nvidia, Qualcomm, Broadcom, MediaTek, us. Everybody else is called a fabulous semiconductor entity.
00:20:17
Speaker
So the fab here is the fabrication unit. We do not own the fabrication unit. We only own the design. And then we send it out to the fabrication unit of our choice. You have to work with them very carefully to adapt your design to their process and get it out from there. And then you give it for packaging. Packaging is another company. So it's become a very,
00:20:43
Speaker
diversified a very open kind of an environment where different people can play at different levels.
The Economics of Specialization in Semiconductors
00:20:50
Speaker
So the fabricator companies, I believe the biggest one is the Taiwanese company called TSMC. Yes, by like a crazy margin.
00:21:04
Speaker
Like the Apple M1, M2, M3 chips, the Google chips. So all of Apple chips, most of Qualcomm's chips, most of Broadcom's chips, most of MediaTek's chips. So that's all of your cell phone, AMD, and Nvidia are all their biggest customers. And they are the ones who keep pushing the boundaries. So we measure the size of a transistor in nanometers.
00:21:34
Speaker
and they're the ones who keep pushing the boundaries, five nanometers, three nanometers, 0.1 nanometer, 0.5, 0.18. So they are, they're very, so one of the things about DSMC is they give you a guarantee that they will never do a design. So you can be sure that your manufacturer will not be competing with you.
00:22:04
Speaker
So why was there a need for this to become a specialized unit, like a specialized fabricator and a specialized design house? Is it that there is too much complexity and it's best to specialize? Yes, pretty much in a nutshell. So what happened is as the technology became bigger and became more and more complex, two things. One is it's become so complex that you need to specialize and you need to create something which is very
00:22:35
Speaker
focused on either this part of the problem or that part of the problem. If you try to mix the two, it becomes a mess. Intel found that out because they failed in their post seven nanometer process. They did not get the yields. I will talk about yields later, but they were basically not able to make it economically work. What happened is
00:23:06
Speaker
for a fab to actually be making money, it needs to run 24 seven and a small foundry, a really small foundry would be making about 40,000 wafers per month, right? Well, taking 40,000 wafers into production per month, it may take about three months to actually manufacture. A large foundry like DSMC has a volume measured in several million wafers per month.
00:23:36
Speaker
So if, but a particular, a single company, let's say MindGrow would not have that kind of volume definitely to start with and probably never in their lives would they be making like 40, 50,000 wafers a month. So it's, it's really hard for us to own our own foundries.
00:24:00
Speaker
And therefore, this particular model of an outsourced foundry and an outsourced assembly house, all of that came up. But as you can see, all the names that I mentioned, NVDR or AMD, these are the big names in the semiconductor industry, which are household names. And they're huge market cap companies. So a lot of the value is actually aggregating to the design.
00:24:30
Speaker
and not to the manufacturing. The foundry margins can be relatively low, whereas design margins can be relatively high. So that's basically what it comes down to. It's very hard to do foundry unless you're aggregating several designs. And it's very hard for, I mean, the process control is nuts enough that, I mean, you're talking about things happening at the scale of a nanometer.
00:25:00
Speaker
So you're talking about something which is so hard to control. So naturally it aggregates to a few strong players who can actually get it done. And the biggest of that, as you mentioned, is TSMC.
00:25:25
Speaker
You had mentioned that chip making and 3D printing analogy you wanted to say? Not 3D printing, I would say more like the publishing industry. So imagine like there's an author who writes a book and then he takes it to a publisher.
00:25:46
Speaker
Now what the publisher does is, the publisher does the copyright or artwork, puts a cover on it, editing, copy editing, all of that, and then does not necessarily own a printing press. So they give the print, also is the printing to somebody. Probably the binding happens elsewhere. It's packaged, it's distributed through various different channels.
00:26:15
Speaker
the actual author and the publishing house would be in the analogy, something like a fabulous semiconductor entity. Okay. Got it. Okay.
Mindgroup's Founding Story and Indian Workforce
00:26:27
Speaker
Okay. So what is the, the gap in the market, which you saw and you know, like what made you start Mindrow?
00:26:40
Speaker
So what made me start mangrove is not that I saw a gap in the market. What made me start mangrove is that I wanted to run a company. So the process is slightly different, right? And we looked around for what it is that we could do. Well, who's here? Just give some context here. So my co-founder is Sharan and we were colleagues in the previous job.
00:27:08
Speaker
And what is that? Well, uh, well, actually I was a senior to begin with, and then we were colleagues. So we were, uh, we were working as software developers for what is called non-destructive testing. And that is, uh, where you use stuff like ultrasound, x-ray, um, for industrial purposes, like for testing a plane.
00:27:33
Speaker
And what happened there is, I'm a hardware engineer and I wanted to always make my own hardware. I did not want to be a software developer for somebody else's hardware. I wanted to make my own hardware, probably my own software also. So I looked at doing that. And one of the gaps that I found over there is that I could not find a processor, which is,
00:28:01
Speaker
conducive to the kind of things that I want to do. Either I have to settle for something which is expensive, which is a poor fit. It's expensive because then that's the only thing that gives the performance I need. But in terms of other features that I would need for these kinds of domains, it's a poor fit. Or I would have to chain together a bunch of half integrated
00:28:28
Speaker
little cores and then write a million different programs to actually make anything work. So, what is this use case that you're talking about for which you did not find? Well, in that case, it was ultrasound. Okay. So, a chip to control the ultrasound machine? Yeah. Okay. Which is a processor, basically. It's a processor with special interfaces. That's all. Okay.
00:28:55
Speaker
And well, the other thing is in the place where we were working, it's we never got an opening to do this kind of research and development and actually create the product. So during the pandemic, we just kind of decided, why don't we try it on our own?
00:29:19
Speaker
So we came out, ultrasound the next day, especially industrial is a tiny market. So we tried to think of doing radar because there's a lot of different applications you have for it. And so by that time, Sharon had already started his part-time PhD at IIT Madras. His guide was Professor Kamakoti, who is also an old family friend of mine. And we would keep meeting him and we mentioned to him that we wanted to start a company.
00:29:49
Speaker
We wanted to do this. He said, take and use Shakti. We had known about Shakti since about 2017. What is Shakti? I'll come to that. So the interesting thing is, I'll just take a digression and come back to Shakti. Just talk about the landscape as it existed.
00:30:16
Speaker
So the interesting thing is there is about 20% of the semiconductor workforce, especially the design workforce, is in India. They're Indians working in India, either out of Chennai, Bangalore, Hyderabad, Pune, little bit of Noida, I guess. And a large portion of the demand
00:30:42
Speaker
Though it's not visible because a lot of the chips get ordered outside the country and assembled outside the country and you get finished products chipped into the country. A large portion of the demand is also here. I would say about 15 to 20% as it stands. Huge market for it. 15, 20% of global demand of chips. Global demand of chips would be in India, yes. Okay, why is that? Because of the mobile revolution.
00:31:08
Speaker
Well, mobile revolution is one thing, but it's not just that. You have IoT, you have, well, you know, your LEDs are chips. Any smart device, basically. Any device, any electronic device today is a chip.
Processor Instruction Sets and Open-Source Alternatives
00:31:23
Speaker
If you have, let's say you have an LED with a dimmer, right? In the old days, when you had a light with a dimmer, the dimmer was a coil which you turned.
00:31:34
Speaker
Now it's a coil which you turn that is read by a very tiny processor, which then generates a signal for that LED to reduce its. So it's a lot more complex, but it's just easier to do it rather than building a bunch of analog electronics. So that's why we do it that way, right? So
00:32:00
Speaker
Shakti was born out of this. We don't own any of the IP in India. All of this 20% of the workforce is working for MNCs. Their IP is owned outside. And Professor Kamakoti and his students wanted to do some security research on a processor. They approached Arm. Arm is one of the biggest suppliers of processor IP in the world.
00:32:30
Speaker
So, ARM is an interesting stakeholder in this ecosystem that we did. Oh, yes. What is ARM? Where does ARM fit into this ecosystem? So, now you're dragging me into a rabbit hole called IP. Now, design itself is such a vast ocean that you can't, I mean, nobody designs the entire
00:32:56
Speaker
in their own little fabulous or even very huge fabulous company, not even Apple does it. So different parts of the chip, you might assemble it using IP from different people. One of the biggest suppliers is ARM and what ARM supplies is different kinds of processors. Specifically, what they supply is different kinds of mobile processors
00:33:25
Speaker
and they have something like a 99-ish% stranglehold on the smartphone, especially the smartphone industry, and maybe about 98.5% of the total cell phone industry. So if you have an Apple phone, all Apple iPhones, iPads, MacBooks,
00:33:52
Speaker
even Apple watches are using ARM processors inside. So, ARM processors or the designs from? Designs made by ARM, which is the processor IP, which is integrated into the rest of the Appleship. In a way, it's like how if today you want to build a website, you will just use WordPress. You will not build it from scratch. Well, that's a good analogy. And if I can go back to the publishing analogy again,
00:34:23
Speaker
Imagine that, you know, it's a multi author book. And each author is writing a chapter. So the chapter called processors would be written by arm. Okay, it's not that simple, especially if your name is Apple. So what Apple does is the license to the processor and then they may they tweak it. Okay, okay. Um, so
00:34:48
Speaker
companies like Qualcomm or Mediatek, all of them also use the same ARM processor. And a lot of ARM also finds its way into these small IoT kind of devices. If you used probably all your modems, Wi-Fi routers, all of them, they might have an ARM processor in them.
00:35:18
Speaker
The classic version of the processor that you see in cell phones is called the A series. And the one that you see in small microcontroller applications is called the M series. So you get A57, A72, whatever, A76. And then you get the M0, M4, M33, M55. So they have their different lines.
00:35:50
Speaker
Okay, understood. So coming back to Shakti, which is where the rabbit hole of ARM started. Yes. So we come back to Shakti. Picture yourself in 2012, ARM is one of the biggest suppliers already. Not as much of a bully monopoly as it became later, but then already pretty much the biggest skin on the block when it came to mobile. So they wanted to do some research on the processor.
00:36:19
Speaker
And to do that, you actually need to take the processor and start making tweaks to it. Like what Apple would do. Aren't refused. So the IITM team said, to help with it, we'll make our own processor. And then they looked at a particular... Now I have to go into another rabbit hole called the instruction set architecture, right?
00:36:46
Speaker
I'll try to make this as small as possible because I don't want to drag this. So an instruction set architecture is like a specification for a processor. It is how the processor itself interacts with the program. Or how you write a program for that particular processor is called the instruction set architecture. And arms is proprietary. They developed it themselves starting in the 1980s.
00:37:16
Speaker
They had a major revision in the mid 2000s and the late 2000s. So we are on version eight, version nine of the ARM ISA. Similarly, Intel and AMD shared ISA called the x86. That comes from 8086, 80386. If you're an old timer, you will remember these numbers.
00:37:44
Speaker
Yeah, my first computer was a 486. Mine was a 286, so yes. We all remember those, but that's where that comes from. The X is the 80X86. AMD was licensed by Intel to use the same instruction set architecture. There were other licenses that
00:38:09
Speaker
at different periods of time, I don't think there are too many more anymore. Anyway, so x86 is another well-known one. Another major one is called the IBM Power. And if you remember, Apple used the IBM Power before they switched to the Intel process in like the late 2000s. The IBM used to be a chip maker. IBM used to be everything.
00:38:35
Speaker
Okay. IBM used to be IBM research is still cutting edge in terms of chip making, but then they only do the research. Okay. Okay. Okay. So IBM, Apple, and Motorola together came up with this thing called PowerPC. And that was there.
00:38:58
Speaker
when IBM kind of started exiting, making chips by themselves. So, power is still being used in IBM servers. They make their own chips for their servers sometimes. Though, of course, now hilariously, IBM is also a fabulous semiconductor entity. We can come to that if you want. So, IBM
00:39:20
Speaker
open sourced or open out in a kind of a way, it's not exactly open source in that sense, they're power architecture. And that was very well known. So the first idea was to make a processor based on the power architecture. They took one look at power, and it is exactly like something that you would expect out of a behemoth like IBM, it's too complex to implement. Right around that time,
00:39:50
Speaker
in Berkeley in the US. The guru of microprocessor design is a professor called David Patterson. So every computer science engineer who has ever read a computer architecture will be familiar with Hennessy and Patterson, which is the textbook that we all use for that. Hennessy is Jack Hennessy who is now chairman of Google.
00:40:20
Speaker
And so Hennessy and Patterson working at Berkeley, they came up with something called a reduced instruction set computing architecture. This is in the 80s. ARM follows a reduced instruction set architecture. Intel on the other hand, x86 on the other hand is what is called a complex instruction set computer, CISC.
00:40:47
Speaker
So a major debate that raised in the 80s, 90s is, excuse me, CISC versus risk. Ultimately the answer was- Who won that debate? No, stalemate. Okay. So the biggest CISC that you know of is x86 and the biggest risk that you know of is ARM. And each one is dominating a market, so there is no winner. There's no clear winner.
00:41:17
Speaker
What we figured out is that Cisco and RISC don't matter after the point. The underlying micro architecture matters a lot more. But anyway, so there were several iterations of the RISC philosophy which were made as research processes inside Berkeley.
00:41:45
Speaker
Now Dave Patterson's student, Krista Astanovich, and his team was starting to build what they called RISC-V. And what they did is they, from the start, they open sourced it. So anybody can download the spec and they'll provide you references and all kinds of other things. And then you can royalty-free, you can just use it and build your own processor. That's what the Shakti team did.
00:42:15
Speaker
So turns out that after Berkeley and MIT, Shakti was the third team to actually implement it. And that is purely Indian owned IP. It is IP coming out of IIT Madras.
00:42:37
Speaker
and it is IP coming, open source, so anybody can take it and use it. And especially a company which happens to be, well, its mentor happens to be the creator of Shakti, so we got privileged access in terms of talking to the people who originally built it. Sharon is a member of the lab, so we know it very intimately, so we started using it.
00:43:08
Speaker
Jumping back out of all the rabbit holes, we are sitting in Professor Kamkoti's office and he says, use Shakti.
Mindgroup's Chip Design Process
00:43:17
Speaker
And I'm compressing a lot of conversations that happened over like three, four months over here, but basically the conversation went something like this. Sir, we can use Shakti, but is the chip going to come out? Yes, of course the chip will come out when you make it.
00:43:34
Speaker
So you do realize that I don't have $10 million to sink into a semiconductor startup? Yes, I know that. IIT Madras will support you. Get yourself incubated. So just at that time, they started something called the IIT Madras Prabhata Technologies Foundation, which is our incubator. And what Prabhata did for us is
00:44:03
Speaker
They are a technology innovation hub. So one of the things they do is they incubate startups. So what Pravata did for us is that Pravata gave us two very important things. They gave us a discount on third party IP that we need to build the chip. It should be insanely hard for us to build internally. And even if we built it, it would be equally hard for us to get it certified as working.
00:44:32
Speaker
and you're not gonna be able to sell a chip which is not certified that way. And the second thing they did is even more importantly, they gave us tools. The software tools for semiconductor design can run into, well, three, four times the typical seed round that a startup might raise. So the only way that semiconductor startups are possible
00:44:57
Speaker
is by something, some kind of an aggregation mechanism for these tools to be made available to them. Pratak was among the first in the country to do it. We were the first beneficiary of that. There are a few others like FabCI in Hyderabad and Safil in Bangalore, which do the same thing. The cost has not gone anywhere. We have to pay them back once we actually start selling chips.
00:45:28
Speaker
but at least it's not front loaded on top of the company and you can't actually build anything. So that is basically how we got started. The third thing that IIT and Bravattak did for us is there will always be a lot of, a lot of people know about Shakti, right? And there's always a lot of questions coming to the lab as, when will Shakti be available? Can I use it to build a mobile phone? Can I build it to use it to build a,
00:45:58
Speaker
electricity meter, can I use it to build a toy? And the answer was always blink, because I mean, IIT Madras is a research organization, it's an educational institution, its job is not to commercialize IP. And all of those queries got directed to us.
00:46:22
Speaker
We looked at all the different things that people wanted to build. We looked at the markets. We looked at the volumes. We looked at the complexity of addressing that particular market. And we settled on chip number one and chip number two, which we have designed slash are designing. And
00:46:46
Speaker
There's a lot of reasons why we settled on these things. I could build anything. I could build a server grade processor. I could build something for the mobile phone industry. I could build something for desktops and PCs. But a couple of things over here. One is you need to build a chip which you are actually capable of building, which the team is actually capable of building within a reasonable amount of time. I could build a server grade chip in about six or seven years. I wanted something that we could build in a year or so.
00:47:17
Speaker
It needs to have a decent market. And it needs to be something, it needs to be a measurable success. So we scouted around for different markets. And one additional constraint that we placed on ourselves is stick to what you know best. I know signal processing, I know machine vision, and I know what all these things involve
00:47:49
Speaker
I did that for 15 years, so let's build a chip that I would find really great to work with. So that's how chip number two was actually born. But then a lot, as we spoke to people, they said, you know, that's too complex a spec. You're talking about this and that, and that IP and this IP, I don't want to pay for all of that.
00:48:16
Speaker
I don't want those kinds of peripherals. I have, I want something that I can put into a meter. I want something that I can put into a small, you know, embedded device. It still has a lot of crossing that it needs to do. It needs to, um, you know, in some cases like battery management and stuff, it needs to actually run the small neural network, but I don't need anything more than, I mean, I need basic stuff on the outside and I need some decent performance.
00:48:43
Speaker
what I have on the market are things which don't have that kind of performance that you're talking about. So I'm very interested in that the performance that you're giving me, but at not with the kind of spec that you would use for machine vision. So we asked what's the volume? The volume was very interesting. So we said, okay, that we can do.
00:49:09
Speaker
It's a fork on the road to the second ship anyway. So we'll build it. And that fork has taken us about a year to build. We literally got started in, so we got started with the concept of the design in about November 22. And we actually started work
00:49:38
Speaker
on the design itself, like April 4th, April 5th, 23. We taped out on November 15th. Tape out is the technical term for when the industry jargon for when the design goes to the foundry and they start manufacturing. So we taped out the prototype run on November 15th.
00:50:03
Speaker
Oh, it's just like a week back. It's just like a week back. This is only a prototype run. So we have to get it back. There's still like months of wait time before we actually get it back and we look at it and see if it's actually working. We may have to do another prototype run to actually fix bugs and so on. So we have a few months wait before we can actually launch the product.
Integrating IPs and Chip Design Techniques
00:50:30
Speaker
I don't want to, I mean, I want to emphasize that it's still a prototype run and we're not out of the woods yet to that product. But it's still nice. It's a nice closure to a lot of design activities that happened. And so we did that and then we've gone back to building the second chip. Okay. I have a couple of questions.
00:50:58
Speaker
Why did you need to design a chip? Shakti was a chip design, right? Like a open source chip design. No, no, no. Shakti is a bunch of IP. Okay. More to the point, Shakti is a core. So it can't do anything by itself. It can't be a chip by itself.
00:51:26
Speaker
You need to do two things to have a chip, right? You need to have peripherals, things to connect to the outside world. The Shakti code does not have that. There are some peripheral devices which were developed at IIT Madras, which you also used. And the second, the most important thing that you need to do is you need to pick a foundry and you need to adapt the design to the foundry. You need to characterize it, qualify it, make sure that the design is working correctly. You have to simulate it and all of that.
00:51:56
Speaker
So if you think about it, one of the biggest semiconductor companies in the world, Qualcomm, a very disparaging way of dismissing what they do with processors is they buy IP from different people, from ARM, from this guy, from that guy, from that guy, and they just put it together. Just put it together is massive amounts of work.
00:52:25
Speaker
Got it. Got it. Got it. Okay. Okay. So, what all did you put together? Where did you get the instruction set from? So, it's Shakti. So, the instruction set is RISC-V. Okay. Okay. So, Shakti is the instruction set. And where did you get the design IP from?
00:52:46
Speaker
So the rest of the IP, some of it was from IIT Madras's own IP, which they did for testing and qualification. By the way, Shakti has been fabbed up before at least four times by IIT Madras itself. One of those was at a very advanced process node called the Intel 22 nanometer process node. Intel sponsored that chip out.
00:53:10
Speaker
And so there's a lot of IP that was built for those tapeouts and there's a lot that was built for simulation and stuff like that. So many of the peripherals came from there. We made lots of modifications to them. We fixed some bugs, we adapted it to things that we wanted to do and so on. So that is Shakti itself and some of Shakti's
00:53:37
Speaker
And then we bought some other IP, most critically, the clock, which goes into the chip. All of these chips are clocked, right? That's where you get your gigahertz number from. And by the way, our first chip is clocked at 700 megahertz, which already makes it something like nine times faster than the 486 that you used.
00:54:05
Speaker
Wow, okay. So this is a 700 megals chip. So we need a clock for that. You can use an external clock, but then you can't usually breach about 200 or 300 megals if you go. The RAM that goes onto the chip, and the RAM is not only what is accessible to the program. There are other bits of RAM scattered around the chip for doing various things.
00:54:32
Speaker
That is provided by the foundry. So we had to create that and we had to get that, create, adapt it for the design, then put it into the design. Then the clock we bought from outside, we bought an analog to the computer from outside. Yeah, you could say that. It's a circuit design. Okay, okay, okay, okay, got it.
00:55:03
Speaker
So the way we design these circuits, we write code, and then there is a compiler that converts it to design. It's too complex to actually do these kind of circuits by actually drawing the circuit diagram. It won't work. So we write the code and then the code is in special languages called a hardware description language.
00:55:27
Speaker
And then that gets transformed into circuit. And that circuit, so it's under supervision, it's not completely automated. That circuit has to be tested, verified, qualified. And then that circuit has to be laid out in silicon. So we create the manufacturing design for that.
00:55:52
Speaker
Like the various, like it is multiple mask each layer. Each layer gets its own mask. Okay. So you design each of the mask. Okay. No, we don't design each of the mask. We design the layers. The actual design of the mask, layer to mask is something the foundry does. Okay. And the reason that the foundry does it is because it's very specific to their process.
ASML's Monopoly in Photolithography
00:56:20
Speaker
and they will actually use an electron beam to etch that mask. So the processes here, they all sound like science fiction, right? Just one more aside over here, because I said science fiction, let me tell you this one aside on the technology that goes into really advanced chips like the M1, M3, not even M1 like M3 and all of that.
00:56:46
Speaker
the latest series of AMD Intel processors, the latest Nvidia GPUs. There is a company in Netherlands called ASML. ASML is a vendor for TSMC, right? Yes. They provide the machine. They provide the photolithography machine.
00:57:06
Speaker
Each machine is like $600 million or $700 million or something like that. It requires like 10 747s to transport it from the Netherlands to Taiwan. It takes a bunch of time to assemble it. And basically what this machine does is it produces light of a very specific frequency.
00:57:33
Speaker
And the way it does it is, there is a chamber in which there is a drop, tungsten is melted and made into drops. As it's dropping, a laser hits it once to take a circular or teardrop shape and make it into a kind of a concave shape. And then it hits it again to convert it into a plasma. That plasma gives out light in several frequencies
00:58:02
Speaker
Out of those several frequencies, one particular frequency is filtered. That filtered frequency is too small to be focused using lenses. So they have these special mirrors, which are the only thing that it actually interacts with. To bounce it like a concave mirror is like a funhouse mirror. Bounce, bounce, bounce, bounce. Each time it bounces, it focuses a bit, bit, bit, bit more. Then finally it ends up
00:58:30
Speaker
the size of the mask and then goes and illuminates the chip. So, I mean, that's the kind of technology that these people are talking about. Yeah, incredible. And I read about this company because I believe US kind of told them that they can't sell their machines to China. Yeah. Right.
00:58:56
Speaker
So ASML is currently a monopoly in the field of extreme ultraviolet. And like I just explained, it's extreme. ASML is a monopoly not because of any other structural reason. It's simply because everybody else gave up. Amazing. So there are other companies which can build older process nodes.
00:59:25
Speaker
Nikon, Canon, Applied Materials, Lamb Research. There are all kinds of companies which can do other processes. These extreme ultraviolet processes, currently ASML is the one vendor in the world. Okay, interesting. And so the thought is that if they prevent ASML from selling, and ASML, even though it's a Dutch company, it uses a lot of American IP.
00:59:54
Speaker
So the American government has a certain strong hold over what they can and cannot. Okay, got it. So which fab are you using for your prototype run? I will tell you that it's in Taiwan and NDL prevents me from saying any further right now.
01:00:20
Speaker
Why is there an idea? Why would a vendor ask their customer to sign the name? Because they don't know if the chip will work yet. Okay. So they don't want to get out ahead of it and say that they always prefer, most of the foundries always prefer to sit in the background. I mean, unless you're a industry insider, well, now a lot more people know because of all the
01:00:48
Speaker
noise being made about it. Five years ago, had anybody heard of the name TSMC? No, it was an unknown company. It was an unknown company which even then made something like 60% of the semiconductors by value. By shipments, it's probably a little less.
01:01:13
Speaker
Okay, interesting. What kind of customers are you making this for? We're making this for people who are doing electronic designs, embedded IoT kind of electronic designs. We're talking to people who are doing biometrics, the biometrics that let you into your office.
01:01:41
Speaker
We're doing it for people who build smart meters, energy meters, water meters, those kinds of things. That's the first chip. All your connected devices in your house. So one of the things we do is we built in security into this chip at a level which is not typically available on microcontrollers. And the reason is that these are all things that you would never have thought of connecting to the internet
01:02:10
Speaker
even a couple of years back. Now you're connecting them to the internet and they are a security hole. So if you can provide encryption, if you can provide those kinds of security features over there, makes it easier for total system security.
01:02:33
Speaker
So your pitch to the manufacturers or the device manufacturers who will use your chip is that it's a made in India chip A and B, it has more security.
01:02:48
Speaker
Actually, for a lot of things, those pictures don't work, right?
Mindgroup's Market Strategy for IoT Devices
01:02:55
Speaker
But they do exist, of course. It's always there in the background that's made in Diatchib. We aim to be world class in quality. And we are always measuring our quality against the top in the world in this class, which is Texas Instruments, NXP semiconductors, which used to be a division of Philips.
01:03:17
Speaker
and renaissance electronics in Japan. The fourth is Infineon, but then they're a bit more known only for automotive. So it's not so easy to measure against them. So we aim to measure ourselves against that. We're not trying to just, I mean, we're trying to cost optimize, but we're not trying to cost cut.
01:03:42
Speaker
So there are security reasons why you may want a made in their chip, but you may also want, like for logistical reasons, you would want an Indian supplier, right? So that works a lot. That is one pitch. The second is, yeah, like I said, one pitch is that you get the highest quality at slight decent discount.
01:04:10
Speaker
and only of the features that you need. For most of these kinds of applications, you don't need all the features that the top end would give you. So I'll cut some of the features and give you a discount. The second is, the next one is security.
01:04:29
Speaker
And the promise here is that we give you security of the fly without actually charging extra for it, and in a way that you can actually use it without making it too difficult on yourself. The next one is support, which is something that we give from within India. So that's important for a lot of people. Support for the developers to write the code that will attract them. Both software developers and hardware people.
01:04:59
Speaker
We give you reference designs, all of that. Okay. Okay. And what is the, what's the way in which, what's the go-to market for this? Like in a sense, we've already gone to the market. We've been talking to a lot of people and the way we do it. So this comes to our philosophy, right? We want to be a product company. We don't want to take projects.
01:05:28
Speaker
but then it has to be something, a product that will actually sell in the market. So what we do is we study the market and we propose, we create a proposed solution. We call it a 60% spec. Then we shop it around to people, ask them what they think of it and show them how their product would look with our chip in it. And then we ask them questions like, does this look
01:05:58
Speaker
doable to you, does this look worth it to you? What other kind of acceptance criteria do you have? Is there anything in here that you absolutely need, which we haven't put in? Or is there anything over here which you think is redundant or irrelevant and you want me to take it out? And what we do is we, so it doesn't make sense for us to develop one chip per vertical.
01:06:25
Speaker
you won't get volumes. So if you want to get volumes, what we do is we design the chip which is flexible and adaptable to situation. And we also have a flexible model of working with people where we can either just give you the chip if you are a company which has that kind of a design job, hardware design jobs, or we will give you a reference design and you can write your own software for it.
01:06:56
Speaker
We own our own spec. And if something is not possible in hardware, we do it in software. Or rather, it's software first. It's software driven. If something is possible in software, you do it in software. And you do stuff in hardware to make the life of the software developer easy.
01:07:26
Speaker
and not just because you can do something in hardware. So there are all kinds of little things over there where we talk to customers and this chip has been designed with a whole bunch of inputs from a whole bunch of people. And all of them are interested as soon as they get samples, many of them are ready to place orders on us.
01:07:54
Speaker
Give me an example of something which can be achieved through both hardware and software and what would be the way which you achieve it through software instead of hardware. Okay. So very, very simple example, right? Let's say that you have a, no, we are talking through mics and speakers here.
01:08:20
Speaker
And let's say that you know that there is a particular harm at a particular frequency and you want to cut it off. You could do that in hardware. You could actually do that at the mic level by adding stuff to the mic. Or you could do it in hardware by building an analog circuit with a bunch of transistors to do it.
01:08:41
Speaker
But for a DSP engineer, it's always easier to just take the entire thing into the system, into a DSP chip, convert it to do a Fourier transform on it, apply a filter and then Fourier transform it back, the hum is magically gone. And the advantage of doing it in software is that tomorrow you want to change the frequency, you can change the frequency easily.
01:09:09
Speaker
More importantly, tomorrow if you want to change the exact algorithm, I don't want Fourier transforms anymore. I want to do something more complex. I don't do machine learning on it. And you can do machine learning on it. There's nothing stopping you. So what I would do there to accelerate it in hardware is if I'm doing a Fourier transform, there is a particular element called a multiply accumulate, which is the heart of a Fourier transform, to fast Fourier transform today.
01:09:37
Speaker
you build a bunch of multiply accumulate units. And whenever the programmer encounters that point, instead of writing the code for it, he'll just call the multiply accumulate unit. Okay, so very, very trivial example. I mean, that's very old. But then what we do in terms of software driven code design is a lot more is along those lines, and it's a lot more complex.
01:10:08
Speaker
Okay. So your value add to a customer is that you are not only an India supply chain vendor, but you will also help them to design. Yes. That is something which a foreign chip supplier may not necessarily help them to design their hardware and software. So a lot of these people, right? They don't use the top end of the market. They use
01:10:35
Speaker
other chips, just put it that way. And they do that typically for cost or other supply chain issues. And you often find in that kind of a market that there's no documentation or it is made needlessly complex for the user because it would have made life marginally easier for the designer
01:11:03
Speaker
The documentation is in languages that you don't follow. There is no support. The support forums are empty or just filled with questions. What we can guarantee is that we are there. And we'll help you build it. Who certifies that a chip is ready for sale? Like you said that there is a cost of getting a chip certified.
01:11:34
Speaker
Not exactly certified to the chip is ready for sale. So if you're claiming anything security, then there are certifications on security. Do you have the stuff that you came?
Business Model and Market Challenges
01:11:48
Speaker
Then there is a certification for temperature sensitivity.
01:11:57
Speaker
So like if you're, for example, if your chip is going to go into the engine compartment of an internal combustion engine vehicle, needs to be able to handle like 125 degrees Celsius. Then there are certifications for safety. As in, can it recover from a fault? And you have to have documentation to prove that. That is called functional safety.
01:12:24
Speaker
That is very important in aerospace industry, in the automotive industry. Wherever life is involved, you have functional safety. So there are these kind of certifications that you would typically have to do. Electromagnetic interference is a major certification that you would have to do, depending on where you're deploying it. So all of those things. And there are separate agencies for each of these type of certificates.
01:12:54
Speaker
Yeah. And annoyingly, not enough of them in India. Yeah. Okay. Okay. So you have to send it out for some. Yeah, we can have to send it out. Okay. So the reason why it takes 10 million to start a chip, a fabulous chip company is the certification cost, the cost of the software, which you need to use to design the chip, the cost of people and so on. How much have you raised till now?
01:13:24
Speaker
which plays 2.33 million. And that is enough to make one chip. Okay. And go to the market and see whether the chip will stick. Okay. Is there a working capital requirement here? You'll have to pay your fab in advance, like if you order, let's say. Yeah, it's not that much for a production, for a prototyping run.
01:13:53
Speaker
When it comes to production plans, you are going to a very, very significant chunk, and we will have to find out a way to raise that. Right. So before you actually are able to sell, you'll need to raise more money, right? Because you need the working capital. All right. Depends. So if you have enough orders, we can raise that easily.
01:14:16
Speaker
through debt or something like that. Yeah, through debt. No, just raise another round. I mean, if we raise a substantial round, we can use it to finance, go to market for the first trip, we can use it to finance part of the second trip. There's a lot of other things that we can do. But then what we're waiting for is to get the first prototype out, get a little bit of market traction, like tentative orders or something. Then that's enough for us to actually raise a decent amount of capital.
01:14:46
Speaker
What would be the next phase, like 10-15 million types? Possibly. I mean, I'm not going to give you any direct numbers because we're still working on it. Frankly, the answer there is I don't know. Because a lot of it depends on two, three things, right? One is how many orders we get.
01:15:12
Speaker
But how soon we have to ramp that production up. Another is how much of the second ship we want to finance in this round? Or do we want to raise another round for that? It depends on how many extra tools or IP that we will need to buy.
01:15:36
Speaker
because we get a decent set now, we get almost everything right now, but then there are certain things that we don't, we can't do with our current tool set. We get a limited number of licenses. So there are a lot of other considerations that we have to go through to understand what is this we wanna do. So one of the main things is right now, since we have products backing, a lot of that cost,
01:16:04
Speaker
which we would have to raise could go into hiring great engineers rather than circling uselessly around IP and tools. And a couple of really key engineers can make all the difference in chip design.
01:16:28
Speaker
So we have this thing called a front-end or back-end design team. So the front-end team is what actually designs the logic. And often these guys can be rock stars. The back-end team is the one that transforms the logic into a printable circuit. And they're rock stars of a different kind. They're rocks.
01:16:53
Speaker
They're very patient. You need to be very patient with the tool. The tool will give you all kinds of stupid errors. You have to be very patient. You have to be very methodical. We are fighting against companies like Nvidia and AMD when we're hiring these engineers. Yeah, they cost a lot. And Apple.
01:17:16
Speaker
But the good thing is that because India has a pool, so it's not like this kind of talent is inaccessible. It is not inaccessible. And frankly, there's a lot of other things that we've learned. A large portion of these skills, you don't need to hire the top talent. You can hire mid-tier and train them up.
01:17:41
Speaker
And there are things that I would like to call force multipliers like for example the specific tools that we use. We inherited a tool from the Shakti team called the BlueSpec language BlueSpec system Verilog, which is a hardware description language, which is very high level advanced on one side. In the sense that you can just specify something and then
01:18:04
Speaker
not have to worry about each individual buyer. But then it still feels a lot like actually designing hardware and not writing. So a lot of these other high level languages are software languages which have been adapted. So glow spec feels a lot more like you're writing real hardware. And so it's easier for a hardware engineer to think that way. And that's a huge force multiplier.
01:18:29
Speaker
We use a lot of automation inside the team. We use continuous integration. These are all things which are learned from the software industry, not necessarily always deployed in the hardware industry. We use version control, which endlessly fascinates some of the old generation hardware engineers. Because we can keep track of, very precise track of what's going on with our code.
01:18:57
Speaker
I can let people experiment a lot more safely. So there are force multipliers. So you don't need as many engineers as you think you would to build a chip. Okay. Interesting. We did it with a team of like six. That, that is a really amazing. If things go as per plan, how many chips would you sell next year?
01:19:27
Speaker
As per plan, the plan is still being worked out, but I would really, the target that I would like to reach is lot of two million chips. Wow, okay. That's a small order. Okay. Yeah, for a fab it would be a small order, right? Yeah, exactly. That's a problem, right? We need to have, it's huge for us, but then for the fab it's just,
01:19:50
Speaker
I don't know. It doesn't even cross decimal rounding error in the bottom of their excel sheet. And how much revenue would that net you? 2 million chips, selling 2 million chips would give you how much revenue? Quite a lot.
Future Aspirations and India's Semiconductor Potential
01:20:11
Speaker
We haven't worked out the price exactly yet. We need to get quotes from the Foundry info, quotes from ATMP to actually narrow it down.
01:20:22
Speaker
assembly testing marking packaging. Okay. Okay. So we need to get those quotes and a lot of that can only be fixed after the prototyping has been one successful done has been complete. I mean, so those are the costs, but you have been in the market. You would know that I can sell this chip for a dollar each or 10 dollars each. So yeah, just from that estimate, what kind of revenue would that give you?
01:20:50
Speaker
I really don't want to quote a figure right now because, so the other side of the market is also important, right? What is it the market is willing to take? Now, if I can happily quote a TI price for this. TI is Texas Instruments. Texas Instruments. And if I quote a Texas Instruments price, nobody will buy it, but then the revenue number will look really good because TI sells some of these chips for like $8, $10, and so forth.
01:21:20
Speaker
You're looking at 20 million in revenue. That's huge for a three-year-old, four-year-old company. But I mean, I can't quote a TI price, right? And there is a lot of value add. What exactly are the value adds that you're going to give with this chip, and how are you going to price those things? A lot of the pricing models are still up in the air. So it's not a flat x-dollar per chip, but there will be add-on. And the bigger volume you'll be.
01:21:51
Speaker
that a customer gives us, the bigger the order I can place at the foundry, the bigger the order I place at the foundry, the foundry starts giving me discounts and I pass those discounts on to the customer. Right. Yeah. Yeah. Okay. So, and the foundry, negotiating the foundry is exactly like negotiating with an auto at the railway station.
01:22:15
Speaker
Yeah, you know, you have no choice. You have no choice. You can address him higher, please. Interesting. So that's a very... I mean, it's that way for a reason. And the reason is that for the foundry, it's all about keeping their lines running 24-7.
01:22:43
Speaker
So if you give them a slightly larger order and say that you can keep it active for another five days, and purely on the basis of my order, then they're very happy to have that conversation. You're at scale already. I mean, the fact that you're talking to these big foundries is you're already at that scale.
01:23:08
Speaker
yeah okay what would be the price differential between chip one and chip two just to understand how much is the value add that is happening in chip two like if chip one you were to sell let's say five dollars each what would you be selling chip to it
01:23:25
Speaker
So I'll just take your $5 number. It's somewhere in the middle of the range where I would like to price it for chip one. Let's say then then chip two would be in the $15, $20 range. Okay. And who are the customers for chip two? What is the use case there? Security cameras is a big one. Why is there any specialized chip for security cameras?
01:23:54
Speaker
Well, not just security cameras, IP cameras. Let me be a little bit more broad on that. You can have industrial cameras, you can have baby monitors, all of them can use this engine. These devices need to take input from the camera, compress it really fast and set it on.
01:24:16
Speaker
So they need speed forward and often today you have these event driven kind of things where you don't record everything, you only pass on what is an event.
01:24:27
Speaker
So you need to have processing to read the camera, which is already quite fast. Let's say even if it's at HD, 30 frames per second, that's already, one HD frame is two MB into 30 frames per second. That's 60 MB per second, MB per second on a tiny chip. And then you need to compress that
01:24:54
Speaker
like whatever MP4 or whatever. And then you need to stream it to, if you're a security camera and you're a battery operated security camera in some godful second corner of a campus, you need to do that on a battery budget.
01:25:17
Speaker
And then you're streaming it, and then you're doing inference to see whether someone, I mean, like, I don't know, face detection, voice detection, something like that. You may have audio, and you have to process the audio also. So, yeah. It means it's processing power. It's not. So you need a, which is purpose-designed to achieve speed? It's not purpose-designed to achieve for the security camera.
01:25:44
Speaker
It's purpose-designed for computer vision load workloads. And the thing about computer vision workloads is that if I design for CV workloads, then it's automatically equally well-designed for certain other workloads, like I'm talking about putting it in a security camera. It can be used on the other side. It can be used on a TV. You're encoding here, you're decoding here. It's very similar kinds of workloads.
01:26:21
Speaker
Okay, got it, got it. Who did you raise this 2.3 billion dollars from? We raised it. It was led by what was then Sequoia India, Sequoia capital India, now called peak 15 ventures. We had two other VCs on the ground and a couple of other engines, so special invest and whiteboard capital. And yeah,
01:26:53
Speaker
Was there some amount of struggle story behind raising the fundraise or was it relatively easy to raise? Well, I don't have a first-hand account, I mean experience of how difficult it has been for others.
01:27:13
Speaker
But then I'll just give you an example. I'll give you a timeline and then you'll understand, right? We started talking to Sequoia in February of 2022. We signed the term sheet on August 15th, signed the SSHA in November, late November, and got the money in January.
01:27:41
Speaker
One year. And a lot of that, the initial phase convincing people that this is something that you can do, this is something that you can believe in it. I'd say it's probably a little easier now, but the previous cohort of semiconductor and hardware startups from India had it really rough.
01:28:12
Speaker
because nobody could believe that you could do hardware from India. And you could do products from India. This is a very, very important thing. I mean, a lot of services people doing, like, for all the foundries, their process design kits are designed by our IT majors, you know, head CLTCS in fee. A lot of Intel's, the Intel's and AMD's of the world outsource to India.
01:28:43
Speaker
But I can own a product and I can actually build it from India's story that not everybody was willing to believe. It's gotten a lot better over the last couple of years. I'd say since about 2020. India doesn't have a track record of hardware success as such. The track record is mostly software success.
01:29:07
Speaker
And I would say that that's unfair because India has had some really brilliant early hardware successes. And it's just that we didn't follow through on that. Which ones? Well, for example, in the early days, HCL had their own hardware division. Yes, they did. They had a PC. Yes. And it's not just PC. They were doing servers. They were doing all kinds of stuff.
01:29:35
Speaker
they could have been a competition to well in those days, the car packs and delts of those days. Yeah, I believe HCL started as a PC company, right? Yeah. Something they pivoted to. But they pivoted to software later. So HCL is one early HCL is a nice model to think of. We've had other hardware. For example, I mean, think of Tata Motors.
01:30:04
Speaker
It's a hardware company for all. A modern-day car is basically a computer on wheels. True, true, yeah. Tata Mahindra, they're basically computer companies. They may not do semiconductors, but they do hardware. There are lots of other companies under the radar which are actually supplying the big guys. You would never know that these companies are Indian.
01:30:37
Speaker
And what happens a lot in the industry is that, you know, there are Indian startups at start, and then they very, very, very early get acquired by somebody else. And so again, that doesn't look like an Indian company anymore, but for all intents and purposes, it is. What is going on? What's your role at Mindgrove? My role at Mindgrove? To begin with, my role at Mindgrove was
01:31:07
Speaker
50% of the work because there were just two of us. But one of the things that I'm good at, I'm definitely good at and I know I'm good at is delicating and letting people do their work. So my role at MindGrow is to be the place where the buck stops.
01:31:35
Speaker
My role would be to... The other thing that I'm very good at is going in, taking a problem, looking at it, and immediately finding a solution, even if I haven't seen the problem before. So my role is on the engineering side. It is to provide guidance, to troubleshoot, to firefight, to solve problems. We are still doing about 80, 85% of sales and marketing
01:32:04
Speaker
as founders and that's what we've been encouraged to do by everybody. You need to do it at least a few times before you can guide somebody else to do it.
01:32:17
Speaker
So, I mean, we don't have very strictly defined roles. I call myself CEO, he calls himself CTO, but Sharon and I don't strictly define roles between the two of us. If there's an admin task and I'm not available, he'll take it up and do it. If there's a technology task and he's not available, even if he's available, even if I'm available, we might just take it up and do it because it's just easier to do with that. It's something that I know and he doesn't know or he knows and I don't know, whatever.
01:32:51
Speaker
So yeah, that's, it's a very fluid kind of a role. And I like to keep it that way. So, you know, like, say, in a 10 year timeframe, where do you see minegrove being? Where we would definitely like to be is, you know, four, five, six established chips in the market.
01:33:20
Speaker
And I told one of my investors this, I know I made it when I open a product and I find a mine grove chip in there and something that I bought off the shelf somewhere. I find a mine grove chip in there and I did not know beforehand that there was going to be a mine grove chip in there. Amazing, amazing. More power to you in making that dream come true.
01:33:48
Speaker
So that is my measure of success. All other things, I mean, but it's very hard to predict exactly what kind of products we will build even three years down the line because the industry is changing very, very rapidly. So the technology is changing very rapidly. The industry is changing very rapidly. The way that we think about building certain things is no longer the same as it used to be.
01:34:17
Speaker
And we are looking to sidestep some of these technology ladders and just go straight to them. You know, not have to go through the route that some of our seniors in the industry had to go through. Just take the learnings and go straight to where we need to reach.