WEBVTT - The Intel Story Part Two 0:00:04.519 --> 0:00:12.840 Technology with tech Stuff from stuff works dot com. Hey there, 0:00:12.880 --> 0:00:17.000 and welcome to text Stuff. I am your host, Jonathan Strickland. 0:00:17.200 --> 0:00:21.000 I'm a senior writer here with how stuff works dot com. 0:00:21.040 --> 0:00:24.760 And today we're going to continue our story about Intel, 0:00:25.440 --> 0:00:29.880 a very important company in the history of Silicon Valley 0:00:29.920 --> 0:00:33.760 and electronics and computing in general. Now, in our last episode, 0:00:34.080 --> 0:00:37.600 we really focused on the origins of Intel and traced 0:00:37.640 --> 0:00:40.080 it all the way back to the birth of the 0:00:40.159 --> 0:00:45.720 semiconductor industry itself. Two of the co founders of Intel, 0:00:45.920 --> 0:00:50.280 Gordon Moore and uh and Rob Noyce. They had come 0:00:50.360 --> 0:00:56.480 from a a an organization that didn't treat them very well, 0:00:56.640 --> 0:00:59.840 and so they, along with six other people, left that organization. 0:00:59.880 --> 0:01:03.640 They became known as the Traitorous Eight and founded the 0:01:03.840 --> 0:01:07.959 fair Child Semiconductor business under the umbrella of a larger 0:01:08.040 --> 0:01:13.240 company that did camera and instrumentation work. Worked there for 0:01:13.240 --> 0:01:17.160 about eleven years before they left that company to co 0:01:17.360 --> 0:01:21.399 found Intel. And in our last episode, I ended right 0:01:21.440 --> 0:01:25.520 around the time that they had introduced their first micro processor, 0:01:25.680 --> 0:01:28.960 the four zero zero four. Now we're gonna pick up 0:01:29.000 --> 0:01:31.039 from that point forward. We're also going to backtrack a 0:01:31.040 --> 0:01:34.720 little bit just to explain some other details, and we're 0:01:34.720 --> 0:01:38.320 gonna explore what Intel was all about since its founding 0:01:38.440 --> 0:01:43.200 up until present day. Now, the four zero zero four 0:01:43.280 --> 0:01:46.760 started off as one of a set of four chips. 0:01:47.200 --> 0:01:50.400 It's like a package of four chips that were designed 0:01:50.440 --> 0:01:53.880 specifically for another company. That company was the Nipon Calculating 0:01:53.920 --> 0:01:58.320 Machine Corporation. Now, the original designation for those four chips 0:01:58.480 --> 0:02:04.880 was the MCS DASH four. Along with that four oh 0:02:04.960 --> 0:02:08.400 four or four zero zero four, I should say chip, 0:02:08.840 --> 0:02:12.520 were three others. Right. You had a chip that was 0:02:12.600 --> 0:02:17.480 a read only memory chip or ROM chip that was 0:02:17.639 --> 0:02:21.800 responsible for storing the custom applications for calculating machines. So 0:02:21.919 --> 0:02:24.720 essentially the programs of these calculating machines were stored and 0:02:24.760 --> 0:02:26.919 read only memory. That meant that you could not write 0:02:27.000 --> 0:02:30.359 over that information. It was always going to be there. 0:02:30.360 --> 0:02:33.480 It was hard coded onto the chip itself, and that 0:02:33.600 --> 0:02:36.000 way it made it efficient and reliable and you didn't 0:02:36.000 --> 0:02:39.640 have to worry about accidentally erasing it. There also was 0:02:39.720 --> 0:02:44.000 a random access memory chip or RAM chip. Random access 0:02:44.040 --> 0:02:48.760 memory allows you to store data in a temporary storage space, 0:02:49.040 --> 0:02:52.280 so that way the processor can call upon that data 0:02:52.360 --> 0:02:57.080 quickly without having to search a deeper storage UH solution. 0:02:57.880 --> 0:03:01.560 The fourth chip was a registered chip, and that was 0:03:01.600 --> 0:03:06.080 handling the input output port, so taking in the information 0:03:06.160 --> 0:03:09.639 from the input devices and being able to translate that 0:03:09.720 --> 0:03:13.240 into whatever the commands were for the CPU to process. Now, 0:03:13.240 --> 0:03:16.960 it was the CPU that really ended up changing everything. 0:03:17.760 --> 0:03:21.480 So Intel designed the CPU as sort of a general 0:03:21.520 --> 0:03:26.720 purpose programmable processor. Now it wasn't a true general purpose 0:03:26.760 --> 0:03:29.160 processor yet. That wouldn't come until a little bit later, 0:03:29.560 --> 0:03:34.920 but it showed the promise of this strategy. It wasn't 0:03:35.000 --> 0:03:39.320 locked into a single practical application. So up until the 0:03:39.360 --> 0:03:45.080 early nineteen seventies, hardware for computing machines was heavily customized 0:03:45.160 --> 0:03:48.800 for each specific machine. In other words, the chips you 0:03:48.800 --> 0:03:52.880 would find in one type of computer, we're completely incompatible 0:03:53.200 --> 0:03:56.200 with every other type of computer. To design the processor 0:03:56.320 --> 0:03:59.240 for these machines, you essentially had to go back to 0:03:59.600 --> 0:04:03.160 the big inning and rebuild the wheel every single time 0:04:03.200 --> 0:04:05.960 you wanted to do it. Uh. There wasn't really a 0:04:06.000 --> 0:04:09.480 concept of a general purpose processor the way the four 0:04:09.600 --> 0:04:12.840 zero zero four was, and the concept of plug in 0:04:12.920 --> 0:04:17.160 play would be decades further into the future. Intel's move 0:04:17.240 --> 0:04:21.159 to create this programmable processor allowed for an explosion in 0:04:21.279 --> 0:04:25.799 various uses. So the four zero zero four would provide 0:04:25.800 --> 0:04:28.960 the foundation for Intel's future chips, which then would become 0:04:30.000 --> 0:04:34.280 an instrumental component in countless types of technology, not just 0:04:34.400 --> 0:04:40.640 computers or calculators. It virtually guaranteed Intel's success in the market. 0:04:41.560 --> 0:04:44.400 Beyond that, the processor was able to take advantage of 0:04:44.400 --> 0:04:48.040 the advances and manturization that had followed the invention of 0:04:48.080 --> 0:04:51.360 the transistor. Now you'll remember from part one of this 0:04:51.520 --> 0:04:55.280 series that Gordon Moore the one of the co founders 0:04:55.320 --> 0:04:58.360 of Intel. He was the guy who came up with 0:04:58.440 --> 0:05:01.559 what we now call Moore's law. All he had made 0:05:01.720 --> 0:05:05.159 an observation that observation would become Moore's laud But the 0:05:05.200 --> 0:05:10.719 observation said that the balance of technology, economics and manufacturing 0:05:10.760 --> 0:05:15.679 processes would mean that for the foreseeable future. And remember 0:05:15.680 --> 0:05:18.479 he made this production back in nineteen sixty five, the 0:05:18.560 --> 0:05:23.919 number of discrete elements on microprocessors, on semiconductor chips would 0:05:23.960 --> 0:05:27.440 double every two years or so. So if you were 0:05:27.480 --> 0:05:30.520 to get a semiconductor chip in nineteen sixty seven, it 0:05:30.560 --> 0:05:33.000 would have twice the number of transistors that you would 0:05:33.040 --> 0:05:36.039 have found on a on a semiconductor in nineteen sixty five. 0:05:36.440 --> 0:05:39.000 The ones in nineteen sixty nine would have twice as 0:05:39.040 --> 0:05:43.360 many as the ones from nineteen sixty seven, and so on. Now, 0:05:43.400 --> 0:05:45.719 by the time Intel was ready to produce the four 0:05:45.839 --> 0:05:49.000 zero zero four chip, they could make a single CPU 0:05:49.120 --> 0:05:55.120 microprocessor that was as powerful as the famous Eniac computer. Now, 0:05:55.120 --> 0:05:58.280 the Eniac computer was one of the first electronic computers 0:05:58.640 --> 0:06:03.240 ever in the history of mankind. Uh. It was constructed 0:06:03.520 --> 0:06:08.760 in the forties, came online more or less in and 0:06:09.640 --> 0:06:12.679 at the time it was one of it was really 0:06:12.720 --> 0:06:16.120 the most powerful electronic computer in existence. It was one 0:06:16.160 --> 0:06:18.800 of the first ones, so obviously it didn't have a 0:06:18.800 --> 0:06:23.400 whole lot of competition, but it took up an entire room. 0:06:23.440 --> 0:06:26.160 It had a lot of very large components that generated 0:06:26.160 --> 0:06:29.200 a ton of heat. It used vacuum tubes instead of transistors. 0:06:29.839 --> 0:06:34.680 But for this supercomputer of the early days, it would 0:06:34.720 --> 0:06:38.680 take up an entire room. Well until's first microprocessor chip 0:06:38.760 --> 0:06:43.880 had the equivalent amount of power stored on a chip 0:06:43.920 --> 0:06:46.880 of semiconductor material that was the size of your fingernail. 0:06:47.520 --> 0:06:50.000 So you went from a device that took up the 0:06:50.200 --> 0:06:52.720 entire room in a building, and it was a big room. 0:06:52.760 --> 0:06:55.599 It wasn't like a little office to something that could 0:06:55.640 --> 0:06:59.839 fit on your fingernail with the equivalent amount of processing power. 0:07:00.080 --> 0:07:06.839 This was a transformational moment in computer history, and one 0:07:06.839 --> 0:07:09.039 of these days I'm doing I'll do a full episode 0:07:09.440 --> 0:07:12.520 about the Eniac computer and also how it came to 0:07:12.600 --> 0:07:15.000 be and the people who worked on it. It was 0:07:15.040 --> 0:07:18.520 a fascinating device all on its own. In the decade 0:07:18.800 --> 0:07:22.560 that it was in operation from around nineteen to nineteen 0:07:22.600 --> 0:07:27.920 fifty five, researchers estimate that it ran more calculations during 0:07:27.960 --> 0:07:31.680 that ten year period than all the calculations that had 0:07:31.720 --> 0:07:36.760 been performed by humans leading up to that moment. So 0:07:36.880 --> 0:07:40.960 in ten years, it did more calculations than all of 0:07:41.080 --> 0:07:45.600 humans had done throughout history leading up to the creation 0:07:45.640 --> 0:07:52.840 of the Eniac. That is incredible. But then it ended 0:07:52.920 --> 0:07:57.280 up dying after it was struck by lightning. So it 0:07:57.320 --> 0:07:59.480 has a tragic end to that story, which also means 0:07:59.480 --> 0:08:02.120 it would make a great subject for a podcast in 0:08:02.160 --> 0:08:06.239 the future. I think that maybe thour got a little 0:08:06.280 --> 0:08:09.400 miffed that we humans that we're getting really confident, and 0:08:09.440 --> 0:08:11.200 so he can kind of nipped in the bud. But 0:08:11.440 --> 0:08:15.360 that's another story. Let's get back to Intel Now, if 0:08:15.400 --> 0:08:18.680 you want to know what the clock speed was of 0:08:18.720 --> 0:08:21.960 the four zero zero four, it's initial speed was one 0:08:22.200 --> 0:08:26.640 d eight killer hurts. Now, clock speeds, in case you 0:08:26.640 --> 0:08:30.239 aren't aware, they're measured in hurts, and it's a measure 0:08:30.280 --> 0:08:33.520 of how many clock cycles a CPU can perform in 0:08:33.559 --> 0:08:37.360 a second. So a hundred eight killer hurts clock speed 0:08:37.400 --> 0:08:42.160 means that the processor can perform one eight thousand clock 0:08:42.280 --> 0:08:48.040 cycles every second. In an ideal world, every single individual 0:08:48.160 --> 0:08:52.240 instruction sent to a processor takes up one clock cycle. 0:08:52.760 --> 0:08:56.440 So you could translate this as saying this processor was 0:08:56.480 --> 0:09:01.600 capable of completing one hundred eight thousand inst ductions every second. 0:09:02.400 --> 0:09:05.920 That's not exactly true because not all instructions take a 0:09:06.000 --> 0:09:10.720 single clock cycle, and efficiency makes a big difference, but 0:09:11.120 --> 0:09:14.240 it gives you a general idea of the capabilities of 0:09:14.240 --> 0:09:19.520 this microprocessor. Now, really, when you get down to it, 0:09:19.600 --> 0:09:23.040 everything that a processor does takes up a certain number 0:09:23.040 --> 0:09:26.439 of clock cycles, and it depends on what the processor 0:09:26.559 --> 0:09:29.360 needs to do, but it tells you there's a physical 0:09:29.520 --> 0:09:33.760 limit to the number of tasks a processor can complete 0:09:34.080 --> 0:09:37.200 within a given amount of time. So a second in 0:09:37.200 --> 0:09:42.079 this case, if you are throwing stuff at the processor, 0:09:42.160 --> 0:09:45.720 that takes fewer clock cycles than what it is capable 0:09:45.760 --> 0:09:48.880 of doing. Things should run pretty smoothly, right. If the 0:09:48.960 --> 0:09:51.880 number of instructions you're sending to the processor is less 0:09:51.880 --> 0:09:56.400 than the processors clock speed, it should be a pretty 0:09:56.440 --> 0:10:00.720 smooth experience for you on the user side of the computer. 0:10:01.160 --> 0:10:03.960 But as you start to throw more processor hungry tasks 0:10:04.080 --> 0:10:06.800 at the chip, you can start to slow down because 0:10:06.920 --> 0:10:11.120 you might be throwing more directions per second than it 0:10:11.200 --> 0:10:13.840 can handle, and then you start getting lag and things 0:10:14.160 --> 0:10:17.200 get jittery and slow. There are a lot of other 0:10:17.200 --> 0:10:19.760 factors that affect us as well, including how efficient that 0:10:19.840 --> 0:10:24.280 processor is. Some processors are more efficient and can do 0:10:24.840 --> 0:10:30.280 the same task with fewer clock cycles than a similar processor. 0:10:30.640 --> 0:10:34.240 So if you get two different processors and they have 0:10:34.320 --> 0:10:37.640 similar clock speeds, but one of them is designed and 0:10:37.760 --> 0:10:42.000 optimized so that it's much more efficient, you can do 0:10:42.240 --> 0:10:45.040 more with that processors than you can with the other one, 0:10:45.120 --> 0:10:48.280 even if they're both rated at the same clock speed, 0:10:48.600 --> 0:10:52.920 just because the other one handles tasks more efficiently than 0:10:53.040 --> 0:10:56.440 it's uh than it's peer. This is true even if 0:10:56.480 --> 0:11:00.920 you have two processors that have different clock speeds. If 0:11:00.920 --> 0:11:04.959 you've got one processor that has a measurably faster clock 0:11:05.040 --> 0:11:08.640 speed than your first one, but the first one is 0:11:08.679 --> 0:11:12.960 still more efficient. You can still end up having a 0:11:12.960 --> 0:11:19.600 better experience using the quote unquote slower microprocessor because it 0:11:19.720 --> 0:11:23.000 is more efficient in its design. So that's something to 0:11:23.080 --> 0:11:27.200 keep in mind if you're ever shopping for microprocessors. Just 0:11:27.360 --> 0:11:30.679 looking for that clock speed is not really an indicator 0:11:30.760 --> 0:11:35.199 of how good that microprocessor is. It is an indicator, 0:11:35.720 --> 0:11:37.640 but it is not the only one you should pay 0:11:37.679 --> 0:11:40.199 attention to. It's kind of like when you go shopping 0:11:40.240 --> 0:11:43.480 for digital cameras and you start looking at megapixel numbers. 0:11:44.200 --> 0:11:49.440 Larger megapixel numbers doesn't necessarily equate with better pictures. There 0:11:49.480 --> 0:11:52.880 are a lot of other factors that are very important 0:11:52.920 --> 0:11:57.200 when it comes to color, representation, contrast, all of these 0:11:57.240 --> 0:12:01.640 other elements that go into creating digital images. So I 0:12:01.679 --> 0:12:03.559 just want to bring that out here in this part 0:12:03.600 --> 0:12:06.040 of the podcast, just in case you are looking at 0:12:06.440 --> 0:12:08.480 building a computer and you want to find a really 0:12:08.480 --> 0:12:12.040 good micro processor. Just going for that high number is 0:12:12.040 --> 0:12:14.280 not a guarantee that you're getting the absolute best for 0:12:14.320 --> 0:12:17.080 your money. You have to take these other elements into consideration. 0:12:18.000 --> 0:12:20.880 Uh So, I know it's a bit of a tangent. 0:12:21.400 --> 0:12:26.080 But it's important to remember just because otherwise we oversimplify 0:12:26.200 --> 0:12:31.239 and say a higher clock speed equals a faster, better processor. 0:12:31.400 --> 0:12:38.040 That's not always the case. Uh. Anyway, back in Killer 0:12:38.120 --> 0:12:43.520 Hurts was wicked fast. That was a really fast processing time. Today, 0:12:43.640 --> 0:12:47.520 something like Intel's Core I seven KB Lake processor can 0:12:47.600 --> 0:12:51.160 run at four and a half giga Hurts. That means 0:12:51.240 --> 0:12:55.120 it can run four and a half billion clock cycles 0:12:55.160 --> 0:12:57.600 per second. It can handle four and a half billion 0:12:57.640 --> 0:13:00.840 instructions per second. Compare that to the four zero zero 0:13:00.880 --> 0:13:03.520 four that was a hundred eight thousand, four and a 0:13:03.520 --> 0:13:07.559 half billion. You start to see the power of Moore's 0:13:07.640 --> 0:13:11.400 law how that plays out over time, and being able 0:13:11.440 --> 0:13:16.760 to go into the billions of instructions per second would 0:13:16.760 --> 0:13:19.840 probably have amazed even Gordon Moore back in the day, 0:13:19.920 --> 0:13:23.360 because when he was making this observation he didn't necessarily 0:13:23.440 --> 0:13:28.239 think this would be sustainable indefinitely. He thought that eventually 0:13:28.240 --> 0:13:32.080 we would run into some sort of fundamental limit as 0:13:32.120 --> 0:13:34.840 to what we are able to accomplish, and at that 0:13:34.920 --> 0:13:38.680 point the actual progression would break down and you would 0:13:38.720 --> 0:13:43.280 no longer be able to have a processor that's effectively 0:13:43.320 --> 0:13:47.640 twice as powerful two years into the future because we 0:13:47.679 --> 0:13:51.360 would have bumped up against some sort of fundamental limit 0:13:51.520 --> 0:13:55.120 that would prevent us from doing that. Now, for the 0:13:55.160 --> 0:13:58.560 four zero zero four, Intel was using two inch wafers, 0:13:59.120 --> 0:14:04.079 which is interesting because typically semiconductor companies were using twelve 0:14:04.120 --> 0:14:08.559 inch wafers to design microchips, but Intel was using two 0:14:08.559 --> 0:14:11.439 inch wafers for this one. And that might sound delicious. 0:14:11.480 --> 0:14:15.240 You're hearing wafer, you might be thinking cookies. It's not. 0:14:15.600 --> 0:14:18.360 The wafer is called that because it's a big circular disk, 0:14:18.760 --> 0:14:21.040 or in the case of the four zero zero four, 0:14:21.080 --> 0:14:25.400 they're using small circular discs. But a wafer is a substrate. 0:14:25.960 --> 0:14:30.760 It's a foundation. It is the ground upon which you 0:14:30.800 --> 0:14:34.920 build a processor or you build a circuit. It's made 0:14:34.920 --> 0:14:39.600 of semiconductor material It's frequently silicon these days, it's more 0:14:39.640 --> 0:14:42.280 often silicon than anything else, but there are other types 0:14:42.320 --> 0:14:45.560 of semiconductor materials, so I don't want you thinking silicon 0:14:45.760 --> 0:14:50.160 is the one and only type. There are others. Silicon 0:14:50.640 --> 0:14:55.760 wafers are usually twelve inches in diameter, not two inches 0:14:55.800 --> 0:14:58.479 in diameter, so this was a bit of a departure. 0:14:58.720 --> 0:15:01.640 But you have to imagine in a really shiny disk 0:15:01.800 --> 0:15:04.440 that has like a grid like pattern across it that's 0:15:04.440 --> 0:15:08.840 repeated over and over and over again. Uh, the patterns 0:15:08.880 --> 0:15:11.800 that you see, that's actually the patterns of where circuits 0:15:12.040 --> 0:15:15.120 will be laid. So it's not just a design, that's 0:15:15.160 --> 0:15:19.080 actually the physical architecture of where the circuits are going 0:15:19.120 --> 0:15:22.200 to be. They don't just come out that way. They 0:15:22.240 --> 0:15:25.080 have to be etched that way. When you first create 0:15:25.320 --> 0:15:28.320 these silicon wafers and you and you get them polished, 0:15:28.600 --> 0:15:32.520 they are just reflective. They don't have any grid like 0:15:32.600 --> 0:15:36.240 patterns on them. Now, for the geometrically minded out there, 0:15:36.280 --> 0:15:38.840 you might wonder, why the heck would you produce round 0:15:39.040 --> 0:15:44.640 silicon wafers when microprocessors are rectangular. Now, obviously you can 0:15:44.800 --> 0:15:48.080 build a bunch of microprocessors on top of a wafer 0:15:48.080 --> 0:15:51.040 and then you cut them out. So you use your 0:15:51.080 --> 0:15:55.720 wafer to be the the foundation for all of your microprocessors. 0:15:55.760 --> 0:15:58.800 You you etch it out using a repeated pattern over 0:15:58.840 --> 0:16:02.640 and over again. You build out the microprocessors, then you 0:16:02.680 --> 0:16:04.480 cut them all out, and you get your little dies 0:16:04.840 --> 0:16:10.200 of of CPU chips or other microchips, doesn't have to 0:16:10.240 --> 0:16:13.320 be just ae CPU. Other microchips also followed this pattern, 0:16:14.040 --> 0:16:16.680 but you're thinking, well, if it's round and ultimately the 0:16:16.760 --> 0:16:19.360 chips are square, that means there's a lot of waste, right, Like, 0:16:19.440 --> 0:16:23.960 you start to get where the edges of the the 0:16:24.040 --> 0:16:28.400 individual microprocessors are coming up against the edge of the disk. Well, 0:16:28.440 --> 0:16:30.960 square and circle, they're not going to match up perfectly, 0:16:30.960 --> 0:16:33.280 so you're gonna have some wasted material. Why would you 0:16:33.320 --> 0:16:37.360 go with a disc in the first place. Well, in 0:16:37.440 --> 0:16:39.160 order to answer this question, I'm gonna have to talk 0:16:39.200 --> 0:16:42.000 about how these chips are made today for the most part. 0:16:42.080 --> 0:16:45.200 But remember that the process back when Intel made it's 0:16:45.240 --> 0:16:48.560 four zero zero four chip was similar but not nearly 0:16:48.600 --> 0:16:53.160 as sophisticated as the way we do it today. Um, 0:16:53.200 --> 0:16:58.600 the microprocessor manufacturers cut wafers up to make several microprocessors 0:16:58.600 --> 0:17:03.640 per wafer, but that does create some way. So here's 0:17:03.720 --> 0:17:09.560 why we have round wafers instead of say, square wafers. 0:17:09.600 --> 0:17:15.480 It's because we grow the wafers and it's done in 0:17:15.480 --> 0:17:18.240 a very interesting way. So imagine that your goal is 0:17:18.280 --> 0:17:22.800 to create a sheet of silicon that's perfect from a microprocessor. 0:17:23.040 --> 0:17:25.720 That means you have to have an incredible amount of 0:17:26.200 --> 0:17:29.000 pure silicon, or as close to pure silicon as you 0:17:29.040 --> 0:17:34.680 can possibly manage. Any sort of impurity will introduce electrical 0:17:34.800 --> 0:17:39.080 elements into your substrate that you don't want because that's 0:17:39.119 --> 0:17:41.200 going to create errors. So you need it to be 0:17:41.240 --> 0:17:44.040 as pure as you possibly can make it. You then 0:17:44.080 --> 0:17:47.040 treat it to allow transistors and other components to transmit 0:17:47.080 --> 0:17:51.800 electricity across the circuit without having it bleed through or 0:17:51.880 --> 0:17:55.960 leach away the substrate. Because silicon is a semiconductor and 0:17:55.960 --> 0:17:58.760 can either conduct or insulate based on properties, it is 0:17:58.840 --> 0:18:03.639 ideal for this. And we start with sand. Sand has 0:18:03.680 --> 0:18:07.760 a very high percentage of silicon in it, and again 0:18:07.840 --> 0:18:11.720 silicon is our semiconductor material of choice. The sand gets 0:18:11.760 --> 0:18:14.800 melted down and then you eventually end up with a 0:18:14.840 --> 0:18:19.360 material called polly or poly silicon, and you use this 0:18:19.480 --> 0:18:22.960 to um you separate out as much of the impurities 0:18:23.000 --> 0:18:27.800 as you possibly can to you get to pure silicon. 0:18:28.480 --> 0:18:32.240 You melt down ingots of this stuff into a purged 0:18:32.480 --> 0:18:35.760 furnace at a temperature of more than two thousand, five 0:18:35.840 --> 0:18:39.640 hundred degrees fahrenheit, which is more than one thousand, three 0:18:39.880 --> 0:18:45.760 d seventy degrees c intigrade. The furnace must be purged 0:18:45.880 --> 0:18:50.480 with oargon gas to make certain there's no unintended impurities present. 0:18:50.600 --> 0:18:52.720 You want to make sure that you've gotten rid of 0:18:52.760 --> 0:18:56.919 anything that could end up affecting the silicon as you 0:18:56.960 --> 0:18:59.920 try to create a wafer. Now, at this point there's 0:19:00.000 --> 0:19:07.159 about one non silicon atom per billion silicon atoms, so 0:19:07.240 --> 0:19:10.480 that's incredibly pure. You get a billion silicon atoms and 0:19:10.480 --> 0:19:15.240 then one thing that is not silicon. That's pretty pure. 0:19:17.119 --> 0:19:21.920 You then have all of this molten silicon inside a crucible, 0:19:22.000 --> 0:19:27.399 which is like a giant, uh cylindrical column, and it 0:19:27.520 --> 0:19:31.919 keeps that molten silicon nice and hot. The crucible starts 0:19:31.960 --> 0:19:36.359 to spin in a given direction. For the purposes of 0:19:36.400 --> 0:19:41.080 this discussion, will say it's spinning witter shans, also known 0:19:41.119 --> 0:19:45.800 as counter or anti clockwise, so it's spinning anti clockwise. 0:19:46.760 --> 0:19:52.199 You then insert a seed crystal of silicon. It's about 0:19:52.280 --> 0:19:55.160 the size and shape of a pencil, and you put 0:19:55.200 --> 0:19:59.880 this it's lowered down by machine, into the molten sil 0:20:00.080 --> 0:20:03.879 con and it acts as sort of the nucleaic spot 0:20:04.080 --> 0:20:06.679 for other silicon crystals to form. And this way you 0:20:06.720 --> 0:20:14.240 get a a very regular crystalline structure, a mono crystalline 0:20:14.280 --> 0:20:19.400 structure of silicon, and this seed will turn in the 0:20:19.400 --> 0:20:22.480 opposite direction of the crucible, So in our example, it 0:20:22.520 --> 0:20:28.400 would turn clockwise while the crucible turns counterclockwise. Now, as 0:20:28.440 --> 0:20:35.320 this happens, you end up creating this column, this cylindrical 0:20:36.040 --> 0:20:42.520 column of pure silicon or mostly pure silicon. Uh, and 0:20:42.560 --> 0:20:45.679 you get that monocrystalline structure all the way throughout the 0:20:45.880 --> 0:20:51.960 entire thing. When you withdraw the cooled silicon from the crucible, 0:20:52.440 --> 0:20:56.120 it looks like a giant silicon log that tapers at 0:20:56.119 --> 0:20:59.840 an end because that's where you've been pulling the sea 0:21:00.160 --> 0:21:03.679 crystal out, where you've been very slowly drawing it upwards. 0:21:03.720 --> 0:21:08.320 You you draw it incredibly slowly, like a millimeter and 0:21:08.320 --> 0:21:12.399 a half per minute, so it's a very very slow process. 0:21:12.480 --> 0:21:16.200 You're not just whipping the crystal out of the molten silicon. 0:21:17.320 --> 0:21:20.400 Doing this ends up creating, like I said, a tapered column. 0:21:20.960 --> 0:21:24.399 And the silicon has great tent sile strength, meaning you 0:21:24.440 --> 0:21:29.600 can suspend it from a thread like amount of silicon, 0:21:29.760 --> 0:21:35.119 and even though it weighs between two and four pounds 0:21:35.240 --> 0:21:40.080 or between a hundred and two hundreds, it'll hold it. However, 0:21:40.280 --> 0:21:43.080 is very brittle, so while it has great tent sile strength, 0:21:43.160 --> 0:21:44.919 if you were to just try and cut it. It 0:21:44.960 --> 0:21:50.120 would cut very easily. The column or pole of silicon 0:21:50.560 --> 0:21:54.520 would be about twelve inches or so in diameter, and 0:21:54.560 --> 0:21:57.040 you would then use that to slice it into wafers. 0:21:57.080 --> 0:22:01.320 So imagine this log of silicon and you put it 0:22:01.359 --> 0:22:06.280 through a device that uses very thin wire cutters or 0:22:06.440 --> 0:22:09.480 wire blades. I guess I should say they're not cutters, 0:22:09.480 --> 0:22:12.639 they're they're blades made a very very thin wire that 0:22:12.800 --> 0:22:17.760 then just zoom through this column and chop it up 0:22:17.800 --> 0:22:20.800 into these very thin wafers. They're about a millimeter thick, 0:22:22.200 --> 0:22:25.840 and you can get quite a few out of one 0:22:25.880 --> 0:22:30.400 column of this silicon material. At that point, you send 0:22:30.440 --> 0:22:33.199 them to another device to polish them out, because the 0:22:33.359 --> 0:22:38.919 cutting creates uneven set sections on the surface of the wafers, 0:22:38.960 --> 0:22:41.120 so you have to polish it to remove as much 0:22:41.160 --> 0:22:43.320 of that unevenness as you can. But even that's not 0:22:43.400 --> 0:22:48.520 good enough, because when you're building microprocessors, you're working on 0:22:48.600 --> 0:22:52.160 the microns scale or smaller. These days, you're working on 0:22:52.560 --> 0:22:58.080 the nano scale. At that scale, even the tiniest of 0:22:58.480 --> 0:23:02.040 bumps or grooves is going to look like an enormous 0:23:02.040 --> 0:23:05.240 mountain range or incredible valley. You have to buff it 0:23:05.280 --> 0:23:08.000 all out and get it as level as you possibly can. 0:23:08.400 --> 0:23:11.560 So after putting it through the polishing machine, you typically 0:23:11.560 --> 0:23:15.880 would need to chemically treat the wafers to get them 0:23:15.920 --> 0:23:20.400 as smooth as they possibly can be. Now, the next 0:23:20.440 --> 0:23:24.040 step would involve etching the wafers to create the pattern 0:23:24.480 --> 0:23:29.320 for your circuits. So this is sort of like building 0:23:29.359 --> 0:23:34.840 the blueprint for the circuits directly onto the substrate itself. 0:23:35.640 --> 0:23:39.040 Uh So, obviously, any sort of flaw in either the 0:23:39.119 --> 0:23:45.640 silicon material or it's physical properties, whether it's the unevenness 0:23:45.880 --> 0:23:49.399 or maybe there's an impurity that has fallen down and 0:23:49.480 --> 0:23:52.760 touched the wafer, it's enormous when you get down to 0:23:52.800 --> 0:23:57.200 that scale, so the tiniest of imperfections can completely ruin 0:23:57.640 --> 0:24:06.960 a chip. For etching, Intel uses a light sensitive layer 0:24:07.400 --> 0:24:11.280 called photo resist and coats the surface of the wafer. 0:24:12.240 --> 0:24:17.520 That's essentially an etch resistant material. Anything that UH encounters 0:24:17.520 --> 0:24:21.200 it is going to resist being etched. And they let 0:24:21.280 --> 0:24:25.040 this layer harden and then they use other little stencil 0:24:25.240 --> 0:24:29.840 like devices called masks. The masks cover parts of the 0:24:29.920 --> 0:24:32.920 chip while allowing other parts of the chip, or rather 0:24:33.000 --> 0:24:35.760 I should say cover parts of the wafer and allow 0:24:35.840 --> 0:24:40.000 other parts of the wafer uh to be exposed. You 0:24:40.080 --> 0:24:44.960 then use ultra violet light, which turns the photo resist 0:24:45.040 --> 0:24:50.720 material it encounters soluble, and through the process of building 0:24:50.720 --> 0:24:52.760 a circuit, you have to use lots and lots of 0:24:52.760 --> 0:24:56.320 different masks, essentially, lots of different stencils, and lots of 0:24:56.359 --> 0:24:59.880 applications of this ultra violet light. Because circuits are really 0:25:00.040 --> 0:25:03.760 three dimensional creations. They're not just two dimensional. They're not 0:25:03.840 --> 0:25:08.000 just width and and height. There's also depth to them. 0:25:08.080 --> 0:25:11.080 So you use a sequence of these masks and you 0:25:11.720 --> 0:25:15.240 expose the wafer to ultraviolet light. Each time the ultra 0:25:15.320 --> 0:25:19.840 violet light hits through the mask and contacts the photo 0:25:19.960 --> 0:25:23.119 resistive layer below, it makes it soluble. You then treat 0:25:23.520 --> 0:25:27.480 the wafer with a chemical that removes all the soluble material. 0:25:28.000 --> 0:25:32.080 So that's you're left with everything else. You've etched away, 0:25:32.119 --> 0:25:34.520 all the stuff you do not want. All the stuff 0:25:34.560 --> 0:25:38.199 you do want remains on the wafer. That is the 0:25:38.200 --> 0:25:43.960 blueprint for your circuit. At that point, then you would 0:25:44.520 --> 0:25:47.639 implant some ions, which are charged particles. They can have 0:25:47.680 --> 0:25:50.000 either a positive or a negative charge, depending upon the 0:25:50.040 --> 0:25:52.320 application you need them to be. You would either put 0:25:52.320 --> 0:25:55.240 in positive ions or negative ions, but you use that 0:25:55.640 --> 0:25:58.359 in the silicon itself. This is called doping. This is 0:25:58.480 --> 0:26:02.600 used in semiconductors all the time to specifically dictate how 0:26:02.640 --> 0:26:08.640 the semiconductor performs under specific circumstances. While you're etching, you're 0:26:08.640 --> 0:26:13.400 creating channels for what will become transistors. The transistors themselves 0:26:13.520 --> 0:26:17.080 must be deposited into the channels, and today Intel uses 0:26:17.160 --> 0:26:21.879 a method called atomic layer deposition to apply materials to 0:26:21.920 --> 0:26:24.600 the wafer surface at a level of precision necessary for 0:26:24.640 --> 0:26:28.800 components on the nanoscale, because at that scale you're talking 0:26:28.880 --> 0:26:31.199 about something so small you can't even see it with 0:26:31.280 --> 0:26:34.959 an optical microscope. You would need a scanning electron microscope 0:26:35.000 --> 0:26:36.880 or something similar in order to even get a look 0:26:36.920 --> 0:26:40.000 at it, so obviously you have to have atomic precision 0:26:40.080 --> 0:26:45.520 with this. Intel also uses electroplating to deposit copper ions 0:26:45.640 --> 0:26:48.320 onto the transistor, and if you listen to my History 0:26:48.359 --> 0:26:52.040