WEBVTT - China Made a Chip Breakthrough That Shocked the World 0:00:10.680 --> 0:00:14.560 Hello, and welcome to another episode of the Odd Lots podcast. 0:00:14.640 --> 0:00:16.919 I'm Joe Wisenthal and I'm Tracy Alloway. 0:00:17.040 --> 0:00:20.280 Tracy, it's been too long since we've done a semiconductor episode. 0:00:20.400 --> 0:00:23.480 I know, I actually feel really behind on what's been 0:00:23.520 --> 0:00:26.520 happening in the industry. I've seen some headlines. I mean, 0:00:26.560 --> 0:00:29.280 obviously we talked about in Video earlier in the year 0:00:29.360 --> 0:00:33.440 in that stock just exploded, and recently they released their earnings. 0:00:33.479 --> 0:00:38.760 I've seen some news about additional export restrictions and things 0:00:38.920 --> 0:00:41.080 like that. But I haven't been paying close enough attention, 0:00:41.120 --> 0:00:42.400 and I feel really bad about that. 0:00:42.840 --> 0:00:47.000 Do you ever read, Tracy, or have you ever ever read, Yes, Joe, 0:00:47.159 --> 0:00:50.120 you read. Have you ever tried to read those like 0:00:50.280 --> 0:00:55.640 really technical semiconductor publications that talk about, oh, the new 0:00:56.120 --> 0:01:00.120 in video chip is four different cores and a thing here, 0:01:00.160 --> 0:01:01.960 And have you ever seen those sites? 0:01:02.080 --> 0:01:02.320 I have? 0:01:02.520 --> 0:01:03.560 Sometimes people drop. 0:01:03.440 --> 0:01:05.720 The links at our discord and I try to catch 0:01:05.800 --> 0:01:07.880 up to speed on what the hot new chip is, 0:01:07.920 --> 0:01:10.039 and I just I always give up one third of 0:01:10.040 --> 0:01:10.840 the way through the post. 0:01:11.000 --> 0:01:13.280 It's really difficult. I mean, even the ones where they 0:01:13.319 --> 0:01:16.039 talk about like wafer thinness and stuff like like, I 0:01:16.040 --> 0:01:19.640 can kind of understand it, but there's so much within 0:01:19.760 --> 0:01:22.959 this specific subject. There's sort of the big picture things 0:01:23.000 --> 0:01:26.880 like geopolitical tensions and trade tensions and things like that, 0:01:26.920 --> 0:01:29.679 and then you can really cut into I don't want 0:01:29.680 --> 0:01:32.360 to say the nuts and bolts of the specific technology, 0:01:32.480 --> 0:01:35.120 but you know, I don't know the wafers and pins 0:01:35.160 --> 0:01:36.280 of specific technology. 0:01:36.600 --> 0:01:38.600 So the one thing is it seems like the lower 0:01:38.680 --> 0:01:42.080 nanometer number the better, Yes, right, But even there, and 0:01:42.120 --> 0:01:44.560 I know we've done some episodes in the past with 0:01:44.680 --> 0:01:47.680 us Stacy Rasken, etcetera. Even there, that only tells you 0:01:47.760 --> 0:01:49.720 so much about a chip's performance, and there are all 0:01:49.720 --> 0:01:54.840 different kinds of architectures and yields. We've talked about that too. 0:01:54.920 --> 0:01:57.760 Or in theory, you could have a really powerful chip, 0:01:58.000 --> 0:02:00.680 but maybe it's not economical because you lose so many 0:02:00.760 --> 0:02:03.680 chips in the process, et cetera. So wrapping one's head 0:02:03.720 --> 0:02:05.320 around chips is tough. 0:02:05.520 --> 0:02:08.360 Yes, agreed, But what will we be doing on this 0:02:08.440 --> 0:02:10.480 episode China wrap our heads around chips. 0:02:10.800 --> 0:02:11.080 Yeah. 0:02:11.120 --> 0:02:14.840 So, you know, there's been some news particularly related to China, 0:02:14.919 --> 0:02:18.120 and I think Huawei came out with a phone fairly 0:02:18.200 --> 0:02:20.720 recently in the last couple of months, and what claued 0:02:20.760 --> 0:02:25.000 people's attention was it seemed to have performance that people 0:02:25.120 --> 0:02:27.840 didn't assume it could have given what was known about 0:02:27.840 --> 0:02:31.240 the state of domestic Chinese semiconductor compress seven. 0:02:31.080 --> 0:02:34.600 Nanometers, Yeah, seven nanometers. I shouldn't whisper on a podcast. 0:02:35.160 --> 0:02:36.880 I don't know why I did that. It's not a secret. 0:02:37.240 --> 0:02:40.440 So supposedly it had a chip made by SMI C, 0:02:41.200 --> 0:02:45.600 a Chinese chip maker, that was seven nanometers and something 0:02:45.800 --> 0:02:50.200 that people thought China wasn't able to produce just yet. 0:02:50.240 --> 0:02:53.000 And yet here we are talking about the seven and 0:02:53.280 --> 0:02:54.239 M in these phones. 0:02:54.440 --> 0:02:56.560 And I guess the question is, is this represent a 0:02:56.560 --> 0:03:01.320 major domestic breakthrough for China's semiconductor into Does it mean 0:03:01.360 --> 0:03:03.840 that some technology which wasn't supposed to get into the 0:03:03.840 --> 0:03:06.639 country somehow got in? So there's some sort of US 0:03:06.800 --> 0:03:08.280 national security implications. 0:03:08.639 --> 0:03:11.000 But it's a good time. You know, we've talked about this. 0:03:11.000 --> 0:03:13.360 For years, you know, with people like Dan Wong, like 0:03:13.600 --> 0:03:16.799 what is the state of Chinese semiconductor? Are they catching up? 0:03:16.960 --> 0:03:20.040 So I think it's a good time to take stock 0:03:20.200 --> 0:03:20.919 of the situation. 0:03:21.120 --> 0:03:21.640 I agree. 0:03:21.680 --> 0:03:24.800 And also, you didn't mention the biggest thing that people 0:03:24.800 --> 0:03:27.160 are talking about now, which is the idea that have 0:03:27.400 --> 0:03:34.200 the restrictions basically created the exact opposite result intended and 0:03:34.320 --> 0:03:37.720 maybe accelerated China's semiconductor technology. 0:03:37.280 --> 0:03:40.560 Which is something people have warned about that ultimately, sure, 0:03:40.720 --> 0:03:42.640 maybe you set the country back a few years in 0:03:42.680 --> 0:03:46.080 its development, but if you restrict its capacity to get 0:03:46.120 --> 0:03:50.080 international technology, then it just builds faster its own homegrown 0:03:50.240 --> 0:03:52.840 and all kinds of questions related to chips right now 0:03:52.880 --> 0:03:54.880 that we need to do, we need to catch up on. 0:03:55.120 --> 0:03:57.680 Yes, And also, can I just say that I blame 0:03:57.920 --> 0:04:01.920 the semiconductor restriction on me having to read the Three 0:04:02.000 --> 0:04:02.840 Body Problem? 0:04:03.240 --> 0:04:03.960 Did you ever read that? 0:04:04.360 --> 0:04:06.800 And now these visions. I don't know how many people 0:04:06.840 --> 0:04:08.760 have read it, but it's a lot of people used 0:04:08.760 --> 0:04:12.560 it as an analogy for China's technological development. But now 0:04:12.600 --> 0:04:16.240 I have nightmares about like little shriveled up dead people. 0:04:16.920 --> 0:04:19.320 And if you've read the book, if you've read the book, 0:04:19.360 --> 0:04:20.000 this makes sense. 0:04:20.160 --> 0:04:22.000 Okay, Well, let's talk chips. 0:04:22.040 --> 0:04:25.120 And we really do have two perfect guests, one of 0:04:25.160 --> 0:04:27.840 whom we've had on the show before, Doug O'Laughlin. He 0:04:27.960 --> 0:04:31.880 is the chief analyst and Fabricated Knowledge, a semiconductor research 0:04:31.960 --> 0:04:35.919 service and Dylan Fitzel, chief analyst. It's Semi Analysis, a 0:04:35.960 --> 0:04:39.000 boutique semiconductor and AI research firms. So we are going 0:04:39.040 --> 0:04:42.640 to pick their brands about the state of Chinese chips. So, 0:04:42.920 --> 0:04:45.760 Doug and Dylan, thank you so much for coming on, 0:04:45.800 --> 0:04:46.600 odd Laws. 0:04:46.560 --> 0:04:48.679 Thank you for having us, Thanks for having me again. 0:04:49.320 --> 0:04:51.400 Let's just start. What was it? 0:04:51.440 --> 0:04:54.000 What's the deal with this Huawei phone that caught everyone 0:04:54.040 --> 0:04:54.599 by surprise? 0:04:55.200 --> 0:04:59.200 So Huawei has historically been a leader in technology. That 0:04:59.279 --> 0:05:02.240 wasn't a surprise anyone. And a few years ago, of course, 0:05:02.520 --> 0:05:05.520 they got banned from, you know, many aspects of the 0:05:05.640 --> 0:05:08.240 US and Western semi conductor supply chain. They got banned 0:05:08.279 --> 0:05:12.440 from utilizing TSMC, the world's largest chip maker, and all 0:05:12.440 --> 0:05:14.480 of a sudden they come out with their own phone, right, 0:05:14.520 --> 0:05:17.279 their new phone that has their own chip made in China. 0:05:17.600 --> 0:05:20.359 The expectations, of course when they first announced it were whatever. 0:05:20.880 --> 0:05:23.080 But then once people got their hands on the phone, 0:05:23.080 --> 0:05:25.920 it was like, oh my god, this is actually very good. 0:05:26.160 --> 0:05:29.039 You know, when you compare to foreign phones right with 0:05:29.200 --> 0:05:32.039 wou'd say, Qualcom chips from like say, Samsung, it is 0:05:32.080 --> 0:05:34.559 only you know, a year and a half behind, right, 0:05:34.640 --> 0:05:37.279 And in some respects. It's actually just as good in 0:05:37.320 --> 0:05:39.760 certain specifications. It depends on how how nitty gritty you 0:05:39.800 --> 0:05:41.760 want to get into it. But at worst one point 0:05:41.760 --> 0:05:45.000 five years behind at best on part right. And so 0:05:45.080 --> 0:05:47.760 that was a big, big shocker for everyone, just. 0:05:47.760 --> 0:05:51.520 Real quickly on those specifications. Like when you say the 0:05:51.600 --> 0:05:53.880 phone is really good, because I don't know, all phones 0:05:53.920 --> 0:05:56.040 sort of seem pretty good and I don't really notice 0:05:56.120 --> 0:05:58.440 advances and phone. So when you say that the phone 0:05:58.440 --> 0:06:02.440 surprised people by capacity, could you just be specific about 0:06:02.440 --> 0:06:03.120 what shocked people? 0:06:03.160 --> 0:06:04.240 And they picked up this device? 0:06:04.720 --> 0:06:08.040 Sure? So stage one is sort of the network performance, right, 0:06:08.080 --> 0:06:10.760 so it can download and upload data just as fast 0:06:10.760 --> 0:06:12.680 as any foreign phone, okay, which is the sort of 0:06:12.720 --> 0:06:14.520 on par thing right. In fact, it was better than 0:06:14.560 --> 0:06:17.440 the current iPhone. It was on par with the current 0:06:17.520 --> 0:06:20.640 Qualcom based phones. And so then that's like one specification. 0:06:20.720 --> 0:06:24.080 Other specification is your CPUs and GPUs and AI aspects 0:06:24.120 --> 0:06:27.159 of the phone. Right. And there they were not only 0:06:27.279 --> 0:06:31.360 using you know, they're again using domestic manufacturing capabilities, they 0:06:31.360 --> 0:06:34.839 were also using domestic design capabilities and were able to 0:06:34.920 --> 0:06:37.320 match again what folks had done a year and a 0:06:37.320 --> 0:06:40.479 half ago or in some cases even sooner more or 0:06:40.480 --> 0:06:43.000 more recently. Right, So this is sort of on on 0:06:43.080 --> 0:06:45.520 a performance basis, whether it's in gaming, whether it's in 0:06:45.839 --> 0:06:48.520 you know, uploading and downloading videos, whether it's in camera. 0:06:48.960 --> 0:06:51.360 Every every aspect of this phone was on par, if 0:06:51.360 --> 0:06:51.680 you will. 0:06:52.080 --> 0:06:55.560 I think that people just probably don't appreciate how impressive 0:06:55.560 --> 0:06:57.920 that is, given the fact that how little technology they 0:06:58.040 --> 0:07:00.320 have access to. So this they did this without EV 0:07:00.480 --> 0:07:03.680 which was the big October restrictions in twenty twenty two. 0:07:04.200 --> 0:07:06.760 And I think the thing that's most impressive about this is, 0:07:06.800 --> 0:07:09.400 like it's a really good chip with both hands tied 0:07:09.400 --> 0:07:11.960 behind their back. And I think that like, if the 0:07:12.080 --> 0:07:16.200 restrictions weren't there, the implication is that SMC could probably 0:07:16.480 --> 0:07:19.440 ship a leading edge phone as good as TSMC and 0:07:19.520 --> 0:07:22.240 maybe even better than Intel or domestic production. And I 0:07:22.280 --> 0:07:24.200 think that that's probably the biggest takeaway that I think 0:07:24.200 --> 0:07:27.400 people need to understand, you know, and there's definitely some 0:07:27.480 --> 0:07:29.640 DV tools that snuck in, and we can talk about 0:07:29.640 --> 0:07:32.880 the you know, all the mechanics of the cross. You know, 0:07:33.000 --> 0:07:35.880 how slippery the restrictions have been and how maybe poorly 0:07:35.960 --> 0:07:39.040 enforced it's been. But like they did an amazing phone 0:07:39.080 --> 0:07:40.920 with both hands tied behind their back. And I think 0:07:40.920 --> 0:07:43.520 that in the conversation that we've been having about semicuctors 0:07:43.680 --> 0:07:46.760 in China for a long time, it's always been like, well, 0:07:46.760 --> 0:07:49.040 they're really far behind, they'll never catch up. And I 0:07:49.040 --> 0:07:51.320 think that this is the first time where you can say, like, 0:07:51.440 --> 0:07:54.000 if they had what we had, they've caught up. Like 0:07:54.240 --> 0:07:56.360 I think that's Dylan. Do you agree or disagree with that? 0:07:56.680 --> 0:07:58.400 I would say, you know, if if you compare just 0:07:58.440 --> 0:08:01.239 what's shipping in the market, Yeah, of course TSMC shipping better, 0:08:01.440 --> 0:08:04.240 Samsung is shipped better. But and everyone talks about this 0:08:04.320 --> 0:08:07.720 Intel turnaround that may or may not happen. Intel's best 0:08:07.800 --> 0:08:11.200 chips that they ship today are seven animeter, the same 0:08:11.240 --> 0:08:13.640 as what China has today, right, so it's it's on 0:08:13.840 --> 0:08:16.920 par now. Of course, Intel's close to releasing their four 0:08:16.960 --> 0:08:19.320 N animeter, and you could you could caveat that in 0:08:19.320 --> 0:08:20.880 a hundred ways. But if you look at what's in 0:08:20.880 --> 0:08:24.080 the market today, the dentist right again, as you mentioned earlier, 0:08:24.080 --> 0:08:26.440 lower nanimeter ship that you can buy today from an 0:08:26.440 --> 0:08:29.720 American manufacturer is from Intel, and the dentist is from 0:08:29.760 --> 0:08:33.079 Smick and Huawei, and it's the same, right, It's it's 0:08:33.080 --> 0:08:34.200 a similar capability. 0:08:34.720 --> 0:08:38.800 Okay, so not since the invention of pringles have people 0:08:38.880 --> 0:08:43.160 been so excited about a single chip. But you sort 0:08:43.160 --> 0:08:45.400 of alluded to this, Doug, But can you maybe walk 0:08:45.480 --> 0:08:49.400 us through what is needed to produce a seven and 0:08:49.800 --> 0:08:53.040 M and what to your point about China basically doing 0:08:53.120 --> 0:08:55.880 this with a hand or two tied behind its back, 0:08:56.280 --> 0:08:59.120 what was actually available to them? 0:08:59.200 --> 0:09:01.720 So this is this is a really hard question answer. 0:09:01.720 --> 0:09:03.760 I'm gonna have to ask Dylan a lot about this 0:09:03.840 --> 0:09:06.720 as well. But the thing that I think probably differentiates 0:09:06.720 --> 0:09:09.640 it from let's say TSMC's process is that they did 0:09:09.640 --> 0:09:12.800 not have access to EV. That's clearly the big delineation. 0:09:13.160 --> 0:09:16.000 But if you remember the original the extreme yes EV 0:09:16.280 --> 0:09:20.200 Extreme Ultraviolet that's the latest and greatest from ASML, and 0:09:20.240 --> 0:09:22.079 they cost like, you know, three hundred million plus a 0:09:22.120 --> 0:09:25.960 pop their extremely advanced technology to make tiny, tiny wavelengths 0:09:26.000 --> 0:09:28.600 of light. But they managed to get around this with 0:09:28.840 --> 0:09:31.720 something called quad patterning self a line quad patterning, which 0:09:31.760 --> 0:09:33.320 is like we're not going to go into the details 0:09:33.320 --> 0:09:36.120 of that, extremely technical but an extremely hard thing that 0:09:36.160 --> 0:09:38.720 Intel got caught up on. So they managed to ship 0:09:38.720 --> 0:09:41.679 the seven animeter Chip much quicker than Intel managed to 0:09:41.720 --> 0:09:44.240 get through all of their problems doing the same quad 0:09:44.240 --> 0:09:47.360 patterning DV process. And I think that that's a big deal, 0:09:47.440 --> 0:09:50.160 Like it shows that there's a lot of technological umph 0:09:50.280 --> 0:09:53.120 underneath the restrictions alone. And then on top of that, 0:09:53.120 --> 0:09:56.160 the restrictions have been extremely poorly enforced. That's why they 0:09:56.160 --> 0:10:00.360 had some, you know, some restatements this year about specificities. 0:10:00.400 --> 0:10:02.959 But what happens is, like a good example is SMIC 0:10:03.200 --> 0:10:06.240 has a leading edge and a lagging edge factory, right, 0:10:06.280 --> 0:10:09.560 and they're both related entities. But my understanding is the 0:10:09.640 --> 0:10:13.760 lagging edge entity can go by extremely advanced deposition and 0:10:13.840 --> 0:10:16.160 etched tools, ones that are on the restrictions for the 0:10:16.200 --> 0:10:19.040 October twenty twenty two, and then they can just kind 0:10:19.040 --> 0:10:21.480 of shuttle those tools into the fab of the leading 0:10:21.559 --> 0:10:23.160 edge and then effectively be able to use it. 0:10:23.240 --> 0:10:25.840 That doesn't seem like it's doing you know, the restriction 0:10:25.960 --> 0:10:27.640 is doing what it was intended to do in. 0:10:27.559 --> 0:10:30.959 That case, Yeah, not at all so far. Pretty much 0:10:30.960 --> 0:10:33.200 what's happened, and this has happened every single time we've 0:10:33.200 --> 0:10:37.600 had restrictions on American semicap companies, is that the restrictions 0:10:37.640 --> 0:10:39.640 come out all the companies say, oh, this is going 0:10:39.679 --> 0:10:42.240 to impact us, and then they slowly find ways for 0:10:42.320 --> 0:10:45.240 loopholes to be pushed through. An example is Applied material 0:10:45.320 --> 0:10:48.480 how to South Korean factory that probably had I think 0:10:48.480 --> 0:10:51.560 they are right now under investigation, and I think the 0:10:51.640 --> 0:10:54.200 thing is it's pretty clear that some of the leading 0:10:54.320 --> 0:10:57.959 edge tools X THEEV stuff is getting into China and 0:10:58.000 --> 0:11:01.160 they're able to use clever engine hearing to make a 0:11:01.160 --> 0:11:03.839 better chip than we thought was possible. Yeah. 0:11:03.880 --> 0:11:06.680 I would say that while in spirit the regulations pretty 0:11:06.760 --> 0:11:09.160 much said hey, you can't have less than fourteen nanimeters, 0:11:09.559 --> 0:11:12.680 the specifics of what was actually banned were quite a 0:11:12.679 --> 0:11:14.880 bit more varied. Right. The tool that you can use 0:11:14.880 --> 0:11:17.600 for twenty eight and nanimeter, well you just use many 0:11:17.600 --> 0:11:19.640 more of them for seven, right. I mean obviously there's 0:11:19.640 --> 0:11:23.520 a lot more complications there, and so the government sort 0:11:23.559 --> 0:11:28.000 of handed out export licenses like candy to companies like Smick, 0:11:28.080 --> 0:11:30.600 saying hey, yeah, you can expand your twenty eight nanimeter 0:11:30.640 --> 0:11:32.720 all you want. Right, that's not the spirit of the regulation. 0:11:33.000 --> 0:11:35.520 Applied Materials, go ahead and ship whatever tools you want 0:11:35.800 --> 0:11:39.800 to the twenty eight nanimeter fab in Beijing for Smick, 0:11:39.840 --> 0:11:41.959 and then Smick then gets those tools ship to Shading 0:11:42.040 --> 0:11:44.400 High right where they are making fourteen nanaimeter, where they're 0:11:44.400 --> 0:11:47.640 making seven nanimeter. And this applies not only to you know, 0:11:47.720 --> 0:11:50.520 applied Materials, but also applies to ASML and every other 0:11:50.520 --> 0:11:53.360 equipment company. Right, So when you say, like, hey, yes, 0:11:53.440 --> 0:11:56.199 they're banned from having less than fourteen ananimeter on one hand. 0:11:56.280 --> 0:11:59.439 On the other hand, every single tool that they used 0:11:59.440 --> 0:12:03.400 for their seven nanimeter was the equivalent of what TSMC 0:12:03.480 --> 0:12:05.920 had when they made their seven animeter, or an upgraded 0:12:06.000 --> 0:12:08.120 version of it, right, So it's not like any specific 0:12:08.200 --> 0:12:12.000 tools were banned that were required for seven nanimeter. And 0:12:12.080 --> 0:12:14.400 so it's kind of like, you know, the regulation and 0:12:14.440 --> 0:12:17.480 the implementation were so far away from each other, and 0:12:17.520 --> 0:12:20.440 that's sort of what these recent regulations hopefully are going 0:12:20.440 --> 0:12:23.320 to try and help. But there's still maybe some holes there. 0:12:38.160 --> 0:12:41.600 So actually explain this in an abstract sense, why is 0:12:41.640 --> 0:12:45.640 it hard to align the implementation of the law with 0:12:45.679 --> 0:12:47.560 the letter of the law or the spirit of the law. 0:12:48.040 --> 0:12:49.440 I think it has a lot to do with the 0:12:49.480 --> 0:12:52.800 fact that the US doesn't want to wholesale band chip 0:12:52.800 --> 0:12:56.200 production in China. Pyriice right, while China's automotive you know 0:12:56.240 --> 0:13:00.480 sort of chip manufacturing is exploding bids of vertically integrated 0:13:00.520 --> 0:13:03.040 monster that is like really taking over the world with 0:13:03.160 --> 0:13:06.319 amazing vehicles and chips for those vehicles. I don't think 0:13:06.320 --> 0:13:09.240 the US has really an intention to block those, but 0:13:09.280 --> 0:13:11.640 the tools that are required there. It turns out all 0:13:11.679 --> 0:13:14.720 the advances and tools that have happened over the last decade, 0:13:14.840 --> 0:13:16.920 many of them would still also be applied to that say, 0:13:16.920 --> 0:13:18.880 twenty eight animeter chip, just the same as it would 0:13:18.880 --> 0:13:21.520 be applied to that seven animeter chip. It's just the differences, 0:13:21.559 --> 0:13:24.200 you know, throughput versus you know, accuracy, if you will, right, 0:13:24.240 --> 0:13:26.719 in a simplified sense. And then the other problem is 0:13:26.800 --> 0:13:30.240 this is so incredibly technical, right like Doug basically told 0:13:30.240 --> 0:13:32.520 me ten times, don't say, you know, like a list 0:13:32.520 --> 0:13:35.640 of words that are too complicated for audience because it's 0:13:35.720 --> 0:13:38.480 It's like, the problem is the government is mostly talking 0:13:38.559 --> 0:13:40.520 to hey, like, who's a lithography expert? Well, they all 0:13:40.520 --> 0:13:43.600 work at ASML, and what is their incentive, right, it 0:13:43.640 --> 0:13:46.480 is to ship as many tools period as possible, and 0:13:46.559 --> 0:13:49.200 including to China, Right, Like what is ASML care that 0:13:49.760 --> 0:13:51.679 you know, XYZ is happening. They want to be the 0:13:51.720 --> 0:13:53.880 monopoly and lithography and continue to ship, and if they 0:13:53.880 --> 0:13:55.400 don't ship, then you know, there will be a Chinese 0:13:55.400 --> 0:13:57.600 company eventually one day. Right, So their incentive is for 0:13:57.840 --> 0:14:00.280 the government to have as week of a control around 0:14:00.320 --> 0:14:02.880 lithography as possible, and they want to be like, well, hey, 0:14:02.920 --> 0:14:05.079 like this tool's used heavily here, you can't just blanket 0:14:05.080 --> 0:14:06.880 ban it. No no, no, no, just ban it for 0:14:07.160 --> 0:14:09.319 that fab and we promise we'll make sure that we 0:14:09.360 --> 0:14:11.439 won't ship it to that fab. But if the customer 0:14:11.480 --> 0:14:13.040 decides to move it from Fab A to Fab B, 0:14:13.200 --> 0:14:14.160 then you know, oh no. 0:14:14.520 --> 0:14:17.280 And also I think there's like an implication that China's 0:14:17.280 --> 0:14:20.320 going to play extremely fair with the regulations, and that's 0:14:20.680 --> 0:14:24.000 just clearly like I feel like we keep making these 0:14:24.040 --> 0:14:26.560 regulations and they're really cute and it's like, well the 0:14:26.600 --> 0:14:28.960 spirit of the law and stuff. And then like, meanwhile, 0:14:29.000 --> 0:14:30.960 you look at what China is doing for their domestic 0:14:31.000 --> 0:14:33.840 semicucter production and they're like they don't care, Like they 0:14:33.880 --> 0:14:36.240 do not care. And what they're doing is probably one 0:14:36.280 --> 0:14:39.480 of the most aggressive industrial policies ever to ramp their 0:14:39.560 --> 0:14:41.160 leading edge semicuacter production. 0:14:41.240 --> 0:14:42.720 Like I think it looks like the US in the 0:14:42.800 --> 0:14:43.600 nineteen thirties. 0:14:43.840 --> 0:14:46.360 Yeah, it's totally different. Like, you know, the Chips Act. 0:14:46.440 --> 0:14:49.200 I'm sure everyone is like heard of and understand the 0:14:49.240 --> 0:14:51.440 Chips Act. Right, it's like fifty two billion dollars and 0:14:51.840 --> 0:14:55.479 then plus like another twenty something in tax credits. That's peanuts. 0:14:55.520 --> 0:14:58.120 Like we are talking a completely different game, and how 0:14:58.240 --> 0:15:01.320 meaningful the incentives are there. We could go on about 0:15:01.320 --> 0:15:03.680 this forever, but like essentially every step along the way 0:15:03.840 --> 0:15:08.160 has massive cost and taxes and R and D credits 0:15:08.160 --> 0:15:10.800 and rent reductions and you know, the big fun one, 0:15:10.880 --> 0:15:13.840 two and subsequently three are about to launch and all 0:15:13.840 --> 0:15:16.800 of these things together is probably putting hundreds of billions 0:15:16.840 --> 0:15:20.520 of dollars in subsidies to encourage the semiconductor industry to 0:15:20.920 --> 0:15:23.000 figure it out. And I think that, you know, with 0:15:23.160 --> 0:15:26.160 that much push from the top, you know who cares 0:15:26.240 --> 0:15:27.920 if a tool is not being used in the spirit 0:15:27.920 --> 0:15:30.000 of the wall, right, Like, like China's government isn't going 0:15:30.080 --> 0:15:33.080 to be like, well, darn you for using this for 0:15:33.120 --> 0:15:35.920 this using this deposition tool for not its intended purpose. 0:15:36.120 --> 0:15:38.960 It's pretty clear that they're weaponizing the split in the 0:15:39.000 --> 0:15:42.160 East and West semicopter supply chains, and they're trying to 0:15:42.200 --> 0:15:44.200 do it as fast as possible. And I think that 0:15:44.200 --> 0:15:48.160 that's something that the West just continues to underestimate. How 0:15:48.280 --> 0:15:49.720 much gump should they have toward that? 0:15:50.240 --> 0:15:54.080 Yeah, I remember Dan Wong, another one of our favorite 0:15:54.120 --> 0:15:58.120 Odd Lots guests. He called it China's Sputnik moment when 0:15:58.160 --> 0:16:01.480 the tech restrictions started coming into play, because like, basically 0:16:01.840 --> 0:16:04.080 it was a huge wake up call for China that 0:16:04.120 --> 0:16:06.920 it could no longer count on the US to supply 0:16:07.120 --> 0:16:10.440 its technology and that it would have to basically encourage 0:16:10.440 --> 0:16:14.400 its own domestic alternatives. So on that note, I take 0:16:14.440 --> 0:16:17.360 the point about industrial policy and the scale of the 0:16:17.360 --> 0:16:20.400 way China is doing it here, and certainly China has 0:16:20.480 --> 0:16:24.320 a lot of experience in both industrial policy and just 0:16:24.400 --> 0:16:28.120 generally a centrally planned economy. But what sense do we 0:16:28.280 --> 0:16:34.840 have of how efficient their semiconductor manufacturing process is so far. 0:16:34.880 --> 0:16:37.400 I've seen bits and pieces about this, you know, I've 0:16:37.440 --> 0:16:40.480 seen people talk about there is a huge profit drop 0:16:40.600 --> 0:16:43.800 in SMIC earnings for the third quarter, I think something 0:16:43.840 --> 0:16:46.440 like eighty percent. So a lot of people are going like, yeah, Okay, 0:16:46.480 --> 0:16:51.280 they're producing these new chips. The seven nanometers, but maybe 0:16:51.320 --> 0:16:54.480 they're spending an insane amount of money to do it. 0:16:54.880 --> 0:16:57.240 And also I think there was an inventory shortage of 0:16:57.280 --> 0:16:59.800 the phones, so maybe there's a sense that they're not 0:16:59.800 --> 0:17:02.320 able to produce these at scales just yet. 0:17:02.720 --> 0:17:05.680 So the thing about the manufacturing here is that, you know, yeah, 0:17:05.720 --> 0:17:08.160 their volumes are limited, right, only about seven million phones 0:17:08.160 --> 0:17:10.360 will ship this year, looking like they're going to ship 0:17:10.359 --> 0:17:12.920 maybe forty million next year. Versus the you know one 0:17:12.960 --> 0:17:15.480 point four billion or now one point two billion phones 0:17:15.520 --> 0:17:17.399 that ship a year. This is, you know, kind of 0:17:17.400 --> 0:17:19.800 like an initial drop in the bucket, but the scale 0:17:19.840 --> 0:17:22.119 of the ramp is huge. And what's more important, I 0:17:22.119 --> 0:17:24.199 think that's hard to recognize for most people is that 0:17:24.440 --> 0:17:28.800 semi conductor manufacturing is literally the most complex manufacturing supply 0:17:28.880 --> 0:17:31.520 chain in the world, bar none. There's more processed steps, 0:17:31.760 --> 0:17:34.280 there's more complex there's more R and D in this 0:17:34.400 --> 0:17:36.960 field than any other field. It is the most complex 0:17:36.960 --> 0:17:39.480 supply chain, period and so when you're talking about thousands 0:17:39.520 --> 0:17:42.280 and thousands of processed steps, each step has one hundred 0:17:42.280 --> 0:17:44.919 different knobs that you could turn on each tool getting 0:17:44.960 --> 0:17:47.760 good yield, right, getting the number of chips you try 0:17:47.800 --> 0:17:49.640 to make for how many actually work in the end 0:17:49.960 --> 0:17:53.040 is very very difficult. But the way you learn is 0:17:53.119 --> 0:17:57.360 by ramping right by by producing and then hey, if 0:17:57.359 --> 0:17:59.720 I'm running a thousand experiments in flight, and each one 0:17:59.760 --> 0:18:02.359 has one knob slightly turned differently, and then I see, oh, 0:18:02.400 --> 0:18:02.920 which one work? 0:18:03.119 --> 0:18:03.439 Awesome? 0:18:03.480 --> 0:18:05.119 Now I know that I leave that knob turned, and 0:18:05.160 --> 0:18:06.720 now I move on to another piece. Right, sort of, 0:18:07.000 --> 0:18:10.080 there's this like complex like you know, constant tweaking of 0:18:10.080 --> 0:18:13.199 the process, right, this is what's made TSMC, you know, 0:18:13.280 --> 0:18:16.440 so amazing, versus say a company like Intel. Is that TSMC, 0:18:16.840 --> 0:18:19.479 their engineers are constantly doing this even on fifteen year 0:18:19.520 --> 0:18:22.560 old process technology, still getting they old better, still getting 0:18:22.560 --> 0:18:25.480 the performance better on say ninety to animeter, whereas Intel 0:18:25.520 --> 0:18:28.320 would move on to work on the newest technology every 0:18:28.359 --> 0:18:31.399 single generation. And so Smick by you know, yes, they 0:18:31.440 --> 0:18:34.600 are not profitable, especially if you strip away the subsidies, right, 0:18:34.600 --> 0:18:38.240 the Smick Shanghai joint venture, the subsidies are getting from Beijing. 0:18:38.640 --> 0:18:40.880 These are massive. If you were stripped those out, they'd 0:18:40.880 --> 0:18:43.840 be losing billions a year, right, But you know that's 0:18:43.960 --> 0:18:46.600 that's the thing is like, hey, let's skip forward now, 0:18:46.720 --> 0:18:49.440 is there twenty eight animeter profitable probably, Yeah, it's their 0:18:49.440 --> 0:18:51.760 seven animeter that's not okay. But we skip forward a 0:18:51.800 --> 0:18:54.080 couple of years and it's like seven animeter will be 0:18:54.160 --> 0:18:56.879 very profitable. And guess what, over eighty five percent of 0:18:56.880 --> 0:19:01.000 the world's chip value is not under seven animeter today, right, 0:19:01.760 --> 0:19:04.359 And so like there is a long tail of like, hey, 0:19:04.359 --> 0:19:06.719 my car has zero seven animeter chips, right, But there 0:19:06.720 --> 0:19:08.440 are a ton of chips that are made by Texas 0:19:08.440 --> 0:19:11.000 Instruments and analog devices and microchip and you go down 0:19:11.040 --> 0:19:12.720 the list and it's like, well, China's going to compete 0:19:12.760 --> 0:19:14.239 with these companies and they're going to compete with them 0:19:14.320 --> 0:19:14.880 very strongly. 0:19:15.080 --> 0:19:15.280