Hardware Architect Answers Microchip Questions | Tech Support | WIRED

[00:00] chief technology officer for systems design at IBM. This is Microchip Support. [upbeat music]

[00:12] "How are microchips programmed to know why zeros and ones are so important in computing. as a long series of zeros and ones.

[00:27] It enumerates all the characters, the letters, of zeros and ones to represent each of these letters.

[00:40] and just string all the letters, and characters, to create a long sequence of zeros and ones. a zero is usually represented by no voltage,

[00:54] like one volt or 1.5 volts. Transistors in a computer chip can then modify the signals and performing certain computations.

[01:07] like adders or multipliers, and perform ever more complex operations that you're sending into the computer chips.

[01:22] "Just how on earth does a transistor physically work?" You have an input and an output we call the input and output source and drain,

[01:36] An electrical signal connected to the gate or it keeps it closed so that no electricity can flow. there are billions of transistors,

[01:50] but they can switch billions of times per second. Well, the modern chips are designed

[02:02] like we're down to five, four, two nanometer design points, and the manufacturing is extremely complicated. is extremely costly.

[02:16] to get to the next technology node is an extremely costly undertaking as well. and we only have a few companies

[02:31] and build those fabs. are TSMC in Taiwan, Samsung in Korea, All these companies are also building fabs in the US

[02:45] Ventynine asks, "Why do computers get slow with time?" When you own a computer over a period of time, you are getting firmware and software updates,

[03:00] So it's not that the hardware gets slower, a lot of junk on the device. It's just you're asking more of it.

[03:15] why do they need so many new data centers anyway?" With what's going on in AI over the last few years, and so let's just step back.

[03:30] over the last three, four, five years, and it's really driving worldwide productivity Now, we spent trillions and trillions of dollars in wages

[03:45] and if we can make knowledge workers more productive by only a few percent, that is a massive market, And so you see a lot of companies building data centers

[04:00] Now, these data centers are really complex. because they get filled with computers and then of course a lot of GPUs for all the AI processing.

[04:18] are really huge infrastructure projects, and power supplies and cooling, ultimately there are millions of chips.

[04:30] general purpose processors, and of course lots of GPUs for the AI processing. of transistors in my CPU actually doing?"

[04:43] A modern day transistor has only a few nanometers of size. So when we are talking about five nanometers,

[04:55] of the width of a human hair. each one of them is a tiny switch, two or three or four signals

[05:08] or the or of all these signals, right? or is one of them not a one, and form more and more complex circuits.

[05:23] we can perform those computations in a loop. so that you can actually program the chips And so because we're putting all of these circuits

[05:37] it adds up to billions of transistors. Minoshi asks, "How can chips have billions of transistors It really matters how much data you need

[05:51] versus how much computation you perform on the chip These are all the connections on the backside. as well as the input and output signals,

[06:04] and a lot of memory on this chip DickheadNL is asking, "Why do computer chips warm up?" and every time they do a switch,

[06:18] and the metal stack, and when that current flows, and they push against the atomic structure of the metal, almost as if your hands are rubbing together.

[06:31] That friction is causing the heat in the chips. "If transistors are so small like a few atoms, We start in the manufacturing process with a blank wafer,

[06:46] We coat the whole wafer with that. of the design is used to shine a light and then we etch out the areas

[07:01] and we can deposit metals, or we can dope the silicon And then this happens in many, many, many layers. using repeated steps of photo resist,

[07:18] and then after the transistors we then put the metal stack on top Nowadays with the fine structures that we have,

[07:32] we are using for that imaging. because the wavelength of the light itself to show the fine structures that we need on these chips.

[07:50] The machines that do all that work are massive, because they need to super precisely position the wafer. and all of that needs to be like really in lockstep

[08:06] to be able to create these super fine structures Then there are super fine machines and we call that dicing.

[08:18] and then these individual chips That's the little green board with two chips, that interconnects the two chips on this module,

[08:32] where we have the pins A Reddit user asks, with no computers to create it?"

[08:44] were designed by hand on a piece of paper. and then people would connect the different components I myself, when I was at Saarland University in Germany,

[09:00] what were called wire rep boards. and you would connect little wires on the backside And we build a small calculator

[09:13] But in the 70s, whole computers were built Nowadays, of course, we have very powerful computers, to build ever more powerful computers.

[09:25] to validate the functional correctness of chips, InternalGoal955 is asking, "AI conquered software coding,

[09:39] How do we prepare for inevitable displacement?" is really strong here, too strong. that make us engineers more productive.

[09:53] That's also true in chip development But it's another set of powerful tools and allows us to build better chips going forward.

[10:06] and enable us to build better chips. in the manufacturing of computer chips? So modern manufacturing processes for semiconductors,

[10:24] cell phone chips, et cetera, are based on silicon. very far in terms of how many transistors how we can manufacture them in a very reliable way.

[10:41] that can be used as semiconductors. But for the most powerful computer chips, A semiconductor is a material

[10:55] but that can be configured to sometimes conduct So you can build a transistor with the gate, the semiconductor is either conducting or not conducting.

[11:12] That is the fundamental building block for modern chips. that allow us to make faster processors?" The silicon node that's at the base,

[11:27] a four, three, two nanometer chip? Then micro architects like myself, and build faster processors

[11:40] how to make storage faster, how to make network faster. computers are getting faster, faster and faster. in the broad field of computer engineering.

[11:54] who basically lays out the big picture architecture into the different components and subunits Dudewiththebling asks, "Theoretically,

[12:10] If you go back to computers from the 1930s and 40s, they were built using magnetic relays or vacuum tubes. with the transistors on silicon chips, for example.

[12:27] In the span of my career over the last 25 or so years, to five and two nanometer transistors nowadays. on how far we can continue to drive this,

[12:41] Nobody knows exactly how we'll build these chips because there's gonna be some scientific breakthroughs. how we would manufacture the chips

[12:54] That was an unknown. So I believe we'll see the innovation continue and therefore add more and more transistors

[13:08] and we're entering really the research We're calling that the Angstrom age, of the size of just a few atoms.

[13:23] R2002 asks, "Semiconductor super cycle, Crash coming?" we're building massive new data centers,

[13:35] and it's really hard to build additional supply are so enormously complex and expensive. from the new data centers,

[13:49] because it's hard to build more manufacturing fabs. Microchips have always gone in cycles. has always gone up for a few years,

[14:04] Right now, we're in what we call a super cycle. there's so much demand for microchips, and it's really hard to scale up

[14:17] because these fabs are so incredibly expensive, driving the current cost of the microchips up. That's really anybody's guess.

[14:32] I personally believe AI is such a transformative technology DoomCrystal asks, because the transistors are physically too small,

[14:45] There's physical limits to how big we can make chips, the more expensive it is. When manufacturing chips, we're using masks

[15:00] and these masks can only be produced in a certain size, and so building chips above 750 or 780 square millimeters,

[15:12] and therefore expensive. between a GPU and CPU?" There's memory chips, there's chips in a camera

[15:30] and turned the light into electrical signals, et cetera. A CPU is a historically very versatile type of microchip that is programmable and can execute all kinds of software.

[15:43] or the heart of a traditional server computer. They came about maybe 20 so years ago,

[15:55] used, for example, in either gaming It turns out that the capabilities like real, strong high performance

[16:10] are also very relevant to AI processing. have actually been kind of built around the GPUs is similar to the kinds of math

[16:24] Prgmmr7 asks, "Could someone explain and the differences?" It starts with the people who develop the silicon process,

[16:40] and then we have the engineers who design the chips. who sort of lays out the big picture Then logic design engineers

[16:54] the floating point units and the caches, for example. that the logic design is functionally correct when it computes on the data.

[17:07] and turn it into what we call a layout. Which function goes where? And then as the chip gets manufactured,

[17:20] to actually put a system around the chip. Somebody designs the card. Somebody puts it all together and validates it,

[17:34] and the card works from manufacturing. and make sure that we have Pyros_it asks, "What were the tech leaps

[17:50] so much faster than the ones in the 1990s?" and everything gets better all the time. So it's faster transistors, smaller silicon nodes.

[18:02] It's faster memory, faster network, faster storage. If you kept one thing the same as it was in the 90s, So it really takes all of it

[18:16] to create these breakthroughs. "Why does Moore's Law keep ending every decade Moore's law was postulated not really as a law,

[18:34] we can double the number of transistors That law is still around, despite it has slowed down a little bit, right?

[18:47] but we can continue to grow the numbers What really has broken down is Dennard scaling. smaller and smaller, put more of them on the chip,

[19:02] they end up consuming the same amount of power That scaling has really ended, it's really hard to stay in the power budget

[19:18] processors consuming more power now So with chip design now, is how do we manage the power consumption of the chip?

[19:31] as we put more and more transistors into a chip, and so there's a few key challenges here. and then that power creates heat,

[19:44] and that's why you see fans in your computers. you see massive power lines go into the data centers, They use a lot of water to cool the air in the data center,

[20:01] to cool the chips with water. with no imperfections?" with billions of transistors,

[20:14] And we're designing to deal So for example, when you're designing a memory element, you're designing maybe 10% more.

[20:30] where you can block out a bad memory cell Or think of some strange numbers of cores on a chip, like you could have a chip with 28 cores, for example.

[20:46] is there's actually 30 cores on the chip, are actually working, we can sell that as a 28 core chip.

[20:58] So we just need to prepare for that, so that we can also sell partial good chips.

[21:10] "Putting chips in people's brains would be great." versus what might happen in the future We've put chips into the human body for decades already.

[21:26] It measures the electric signals in your heart and it can send a pulse Modern pacemakers also contain memory, and take traces,

[21:42] that can be read out at a doctor's office. There's already research happening, for example, to have artificial eyesight where a camera is connected,

[21:55] into the visual cortex of the brain. where I'll just say loosely, we can mitigate disabilities, a patient has a stroke and a chip could be used to repair

[22:12] That already is happening, Where it gets a bit more complex and controversial the capabilities of the brain.

[22:27] that has emotion, and intuition, and it makes us think, it makes us be innovative putting an additional chip

[22:41] that's out on the internet would actually help or hurt. besides all the ethical concerns it would create. "Why does making chips require clean facility?"

[22:55] A dust speck is thousand times that. that you have a dust speck settle on the wafer

[23:08] Well, then the chip won't be able to work. are super, super clean room onto the chips that you're producing.

[23:22] what was your career path like?" Mine started as a computer science student and then I joined the IBM Development Lab

[23:37] and I kind of learned chip design as part of my job. and develop next generation mainframe chips, and went through different aspects of different chips.

[23:52] I designed IO circuits. And then as my responsibility, I ended up in my current role as CTO.

[24:07] with an electrical engineering background. in terms of how programming works, as part of doing my job.

[24:21] Thanks for watching. [upbeat music]