BRIAN SANTO: I’m Brian Santo, EE Times Editor in Chief, and you're listening to EE Times on Air. This is your Briefing for the week ending July 26th.
BRIAN SANTO：我是EE Times总编Brian Santo，你正在收听的是EE Times全球联播。这是截至7月26日的一周新闻概要。
…we’re mixing it up a bit. In this episode, we’re going to focus on a single topic. It’s how the world’s biggest companies are doing business in ways no company has before – and what that might mean for everybody – not just the technology industry.
There are fewer than a dozen of these giants. On Wall Street, you might still hear them referred to as “the FANGs” – “FANG” being an acronym for Facebook, Apple, Netflix, Google. These are companies that have grown so enormous, we’ve invented new words for them: “hyperscale” and “hyper-vertical.” Other hyperscalers include Alibaba, Amazon, Baidu, Microsoft.
These companies have exerted an unprecedented amount of influence in our society and culture. This is the Big-Picture stuff we all know – that Facebook has become a vital venue for socializing, as well as an unexpectedly powerful political forum. Amazon helped change the way we shop. Google has upended multiple industries, including advertising and publishing. These hyperscale technology companies have changed the way we live.
What most people have yet to comprehend is – they’re not done. They are not done. What they are doing now is remaking the technology industry itself. And since most of us – even those of us who work in the business – haven’t quite understood the extent to which this is happening, almost nobody has really thought about the implications — about the Big Picture.
Wally Rhines, the almost legendary chairman emeritus of Mentor Graphics, has thought about the implications, and they are more consequential than anything hyperscalers have been up to, to date. But to make sense of what Rhines is going to tell us, we need to run through what the hyperscalers are doing now.
富有传奇的Mentor Graphics名誉主席Wally Rhines考虑到了这个影响，它要比迄今为止这些超大规模公司已经做的更严重。但要领会Rhines将要告诉我们的内容，我们需要了解这些超大规模企业现在正在做的事情。
Every hyperscaler’s most valuable asset is information. They make money from getting it, processing it, analyzing it, and acting on it. That’s why every hyperscaler ranks among the biggest data center operators in the world. So, what’s important to a hyperscaler? Brad Booth manages the technology at Microsoft’s data centers, in Microsoft’s Azure unit. Here’s what he has to say about it.
BRAD BOOTH: What semiconductor vendors need to understand about hyperscalars is that energy is our cost. Power is our cost for running our data centers. And while we are all in different phases of deployment and have, maybe have different applications or different customers or different areas of focus, power is the core value of data center operations. And because of that, it's interesting to see how the industry and semiconductor vendors come in and they will look at their solution and quite often say, Oh, well, you know, the power for this CERTIS is, you know, only 50% more than it is for that CERTIS. Or, you know, the power here is, you know, only, you'll use a re-timer, and that's only a few extra hundred milliwatts. And on the small scale, that doesn't seem like much. But when you operate on a scale where you're deploying millions of these potentially, or hundreds of thousands or tens of thousands even, that number starts to add up. And that number starts to directly impact how many servers you can deploy. So, the more power inefficient you are being, the more re-timers or extra power being burned for no added value, the more you put that into your system, the less power you can put to the applications that your customers want to run. And I think that is an area where the industry will have to look at it more seriously. That's why you're seeing efforts by a lot of the hyperscalers to focus on things like moving optics inside, closer to the switch silicon to, you know, reduce that power. And I think overall that's going to be a trend of our industry, is how do we become more power-efficient.
BRAD BOOTH: 关于超大规模公司，半导体供应商需要了解的是，能源也是我们的成本。能耗是我们运行数据中心的成本。虽然我们都处于不同的部署阶段，可能有不同的应用、不同的客户或关注领域，但能耗是数据中心运营的核心所在。正因如此，有趣的是看到行业和半导体供应商如何带着他们的解决方案出现，经常会说——你知道的，这个提供给CERTIS的功耗，只有提供给另个CERTIS的50％多的功耗。或者，你知道，这里说的功耗，只是指你使用re-timer，而这只需几百毫瓦。从小范围来看，这似乎并不多。但是当你进行大规模操作时，你可能会部署几十上百万，甚至数百万台，那这个数字则开始叠加。这个数字将直接影响您可以部署的服务器数量。因此，您的能效越低，没有附加价值的re-timer或额外功耗越高，您将其放入系统的次数越多，您可以为客户想要运行的应用提供的功率就越少。而且我认为这是行业必须更加认真对待的一个领域。这就是为什么你会看到很多超大规模企业的人都把注意力集中在像将光学器件移到内部这样的事情上，集中在开关芯片上，是为了降低功耗。我认为整体而言，这将是我们行业的趋势——如何变得更加节能。
BRIAN SANTO: The point of quoting Brad Booth here isn’t that hyperscale companies like Microsoft want to push the semiconductor industry as hard and as fast as it can go. Of course they do. The point is the stakes. A tiny change in a chip here can lead to an immense saving in a data center there.
BRIAN SANTO: 在这里引用Brad Booth的观点，并不是像微软这样的超大规模公司想要尽可能快地推动半导体行业的发展。 当然，他们肯定会去推动发展，这事关重大。一颗芯片的微小变化都可以为数据中心大幅节省开支。
Let me tell you a quick story. Three years ago, Google and Facebook got together to formulate a standard for server racks. This is a subject so boring it brings tears to the eyes of other engineers. But it was of vital interest to Google, Facebook and anybody else with a data center. The standard would make racks more energy efficient by 30 percent. Thirty percent! Spread that across thousands of racks? Google and Facebook said they were saving millions of dollars a year in energy costs with the new racks.
让我给你讲个小故事。 三年前，谷歌和Facebook共同制定了服务器机架标准。 这是一个如此无聊的主题，让工程师都欲哭无泪。 但谷歌、Facebook和其他拥有数据中心的公司对此都非常感兴趣。 该标准将使机架可以节能30％。 百分之三十！ 在数千个机架上部署？ 谷歌和Facebook表示，他们每年通过新机架能节省数百万美元的能源成本。
What seem like picayune details to other companies are incredibly important to hyperscale companies. Real money is on the line. The second some better technology is available at the right price, in it goes. Data centers improve at the speed of improvement.
这些对其他公司来说，看起来微不足道的细节，对超大规模公司来说却非常重要。 这都是真金白银。第二重要的则是，更好的技术也要有合适的价格。 数据中心正以高速在前进。
The semiconductor industry… does not. Chip companies are on schedules. That’s the inherent property of Moore’s Law – progress on a schedule – and each new step is a big one. Data centers will take any incremental improvement as soon as they can get it. The way data centers consume new technology literally does not synchronize with the way chip vendors generate new technology. And the hyperscalers have stopped waiting around. They’re pushing the technology faster, and if they can’t push it fast enough, well, then, they’ll design their own chips. For years, Apple has been designing its own chips – what it calls systems in a package. Google designed its own artificial intelligence chip; Facebook is doing likewise.
但半导体行业却不然，芯片公司正有条不紊地推进。 这是摩尔定律的固有属性——按计划进展 ——每一个新步骤都很重要。 数据中心只要有机会，就会立即采取所能的渐进式措施进行改进。 数据中心采用新技术的方式与芯片供应商生成新技术的方式不同步。 所以那些超大规模公司们不再坐以待毙。 他们正在更快地推动技术进步，如果认为技术进步达不到期望值，那么他们选择自己设计芯片。 多年来，Apple一直在设计自己的芯片——称之为封装的系统。Google设计了自己的人工智能芯片；Facebook也设计自己的芯片。
EE Times just published a series of articles on how the semiconductor industry is changing in response to hyperscaler influence. The story that EE Times Editor Rick Merritt contributed to the special project is about the repercussions of the hyperscalers getting into the chip business. Junko Yoshida talks to him about it.
EE Tims刚刚发表了一系列关于半导体行业如何应对超大规模企业带来的影响、变化的文章。 EE Times编辑Rick Merritt为此贡献了一篇特别项目文章——关于超大规模公司涉猎芯片业务带来的影响。 Junko Yoshida将与他就这个问题进行探讨。
JUNKO YOSHIDA: You were the one actually in the beginning when we were setting up this special report, you were the one who made a very astute comment, saying that over the last few decades, every time when a new market segment pops up — like workstation or personal computer or game console or smartphone — chip companies were always put in a position to learn quickly how best to design and manufacture chips that fit the new customers' demands. So, my question to you then is, what's different this time around about hyperscalers?
JUNKO YOSHIDA: 当我们决定要做这份特别报告时，你就是我们想到的不二人选，你做出的评论非常精彩。你说在过去的几十年里，每当突然出现新的细分市场时 ——比如工作站、个人计算机、游戏机或智能手机——芯片公司总是能够快速学习如何最好地设计和制造符合新客户需求的芯片。 那么，我想问的是，这次超大规模企业的不同之处是什么？
RICK MERRITT: Yeah, this time around, instead of a broad market like consumer electronics, um, where there's many market segments and you have to figure out which ones you want to go after and then create a standard product to hit as many of them as possible, this time you have seven gigantic customers. And to some extent they all want the same thing, and to some extent each one wants something a little bit different. Kind of weird!
RICK MERRITT: 是的，这次不像消费电子这样的广阔市场那样有很多细分市场，你必须弄清楚你想要追求的是什么，然后再创造一个标准产品，以尽可能多地满足七个“巨无霸”客户的需求。 在某种程度上，他们都想要同样的东西，但在另外某种程度上，每个人又都想要些不同的东西。 反正挺怪的！
JUNKO YOSHIDA: Yeah. You said "gigantic," but certainly the biggest challenge for chip companies are these hyperscalers are flush with money. The money that a lot of chip companies don't have. Do you agree?
JUNKO YOSHIDA: 是啊。 你说“巨无霸”，但对于芯片公司来说，最大的挑战当然是这些超大规模公司都财大气粗。 许多芯片公司却没有这么“壕”。 你同意吗？
RICK MERRITT: Yes. And a lot of money that the individual customer for a semiconductor company doesn't usually have. I mean, Amazon alone is like a $250 billion a year company. There's not too many semiconductor customers that have that much cash.
RICK MERRITT: 是的。 并且半导体公司的单个客户通常不会拥有大量资金。 我的意思是，仅Amazon一家公司年收入就有2500亿美元。 拥有如此多现金的半导体客户并不多见。
JUNKO YOSHIDA: And also, I think in the past a lot of system companies — whether it's automotive companies or the, uh, PC companies or CE companies — you know, in the early days they may have had their own ASIC team, but system vendors typically never had their own design team because they didn't see that's their business. So how are things different this time around?
JUNKO YOSHIDA: 而且，我认为过去许多系统公司 - 无论是汽车公司还是PC公司或CE公司——你知道的，在早期他们可能拥有自己的ASIC团队，但系统供应商往往从未有过自己的设计团队，因为他们不把这部分视作他们的业务。 那么这次的情况又如何呢？
RICK MERRITT: Well we've got to blame Apple for this one, blame Steve Jobs. He created this model that said, I'm going to be vertically integrated. And I'm going to make the chip and the software and wrap it in metal and sell it to somebody with a service on top of it. And everybody said, “Wow, that's the model because they're one of the world's biggest, most successful companies.” So, the hyperscalers — because they're so big, and because they do control all their own software internally — it's their workloads and their software — they said, “Let's start our own semiconductor team, and at least we'll specify exactly what we want. Because sometimes nobody's making it.”
RICK MERRITT: 好吧，这得归咎于Apple，归咎于史蒂夫·乔布斯。 他创造了这种模式，并且说过，“Apple将垂直整合，我们将制作芯片和软件并用金属包装起来，提供服务，出售给客户。” 每个人都说，“哇，这就是榜样，因为他们是世界上最大、最成功的公司之一。”所以，超大规模企业 —— 它们太大了，而且确实在控制着自己所有的内部软件 ——他们的工作载量和软件。他们说，“让我们创立自己的半导体团队，至少可以明确我们想要的东西。 因为有时没人会去做这件事。”
JUNKO YOSHIDA: Exactly. Do you see that these giants are emerging as real competitors to chip companies then?
JUNKO YOSHIDA: 确实如此。 您是否认为这些巨头们正逐步成为芯片公司的真正竞争者？
RICK MERRITT: You know, I talked to a lot of people about this, and pretty much the consensus is, these guys have big problems to solve, enough problems to solve, that they don't want to have to make chips if they don't have to. But sometimes there's big holes. Like nobody really knows their workloads and what AI accelerators they need, so maybe they'll have to design one like Google did. Uh, the guy who heads the semiconductor group at Facebook used to run the data center group at AMD, and he said when he got there nobody had a video encoder suitable for the data center. So his team defined one, and they took it out and shopped it around, and now Broadcomm and Verisilicon are going to make it. That works for them.
RICK MERRITT: 我和很多人谈论过这个问题，而且他们的共识是，这些巨头们有很大的、足够多的问题需要解决，如果条件允许，他们也不想制造芯片。 但有时会有很大的漏洞需要弥补。 就像没有人真正了解他们的工作负荷以及他们需要的AI加速器一样，也许他们必须像Google那样自己设计一个。领导Facebook半导体团队的人，曾在AMD数据中心负责运营，他说当他刚到Facebook时，没人有适合数据中心的视频编码器。 所以他的团队定义了一个，然后把它拿出来售卖，现在Broadcomm和Verisilicon将会购入。 这招对他们挺适用。
JUNKO YOSHIDA: Right. But also, the way I see it is that those so-called "internet platform giants," they are in a lot of businesses, aren't they? I mean, they're not just making data centers, but they're talking about, their ambitions really range to things like autonomous vehicles, IoT devices, health care devices. I mean, their ambitions are so big and encompass such a huge market, is it not?
JUNKO YOSHIDA：对的。 但是我认为那些所谓的“互联网平台巨头”，他们涉及的行业颇广，不是吗？ 我的意思是，不只是运营数据中心，如他们所说，他们的野心实际上涉及自动驾驶汽车、物联网设备、医疗保健设备等。 我的意思是，他们的野心是如此之大，涵盖了如此多巨大的市场，不是吗？
RICK MERRITT: That's true. But they're also all a little different. Facebook tries to keep it simple by saying they have four big applications: Facebook, Instagram, Messenger, I forget number four. Amazon, you know, has its bookstore, but they also sell everybody cloud computing services with AWS. So, in some ways they have similar big applications. Some are more fragmented than others.
RICK MERRITT: 确实如此，但他们也都有些不同之处。 Facebook试图说服大众它们坚持较纯粹的业务，拥有四大应用平台：Facebook、Instagram、Messenger，第四个我忘了。Amazon有自己的书店，但它们也通过AWS面向所有人销售云计算服务。 因此，某种程度上，他们有类似的大型应用。有些则比其他公司业务更加分散。
JUNKO YOSHIDA: All right. So, give me a few examples of specific technologies that those hyperscalers want chip companies to develop or step up — you know, up the ante. Give me some examples here.
JUNKO YOSHIDA:好的。 那么请给我讲一些关于特定技术的例子，那些超大规模公司希望芯片公司能够开发或加速哪些项目。 给我举些例子吧。
RICK MERRITT: Well the biggest one is AI accelerators.
RICK MERRITT: 这些超大规模公司需求最大的是AI加速器。
JUNKO YOSHIDA: Right.
JUNKO YOSHIDA: 好的。
RICK MERRITT: They each see their ability to do deep learning to personalize services as their value-add. And that's how Amazon's going to compete with Google, who's going to compete with Facebook. So they are each trying to run faster than the other in doing better deep learning. So they need accelerators really desperately. But each one has a different workload, a different software that they understand really well. And they just want an accelerator that runs that fast, not Amazon's.
RICK MERRITT: 他们均把深入学习来制定个性化服务的能力视作附加值。 这就是Amazon如何与Google竞争，后者将与Facebook竞争。 所以他们都在努力比其他公司更快地进行深度学习。 因此他们真的非常需要加速器。 但每个公司工作载量不同，他们各自都拥有熟悉且用得顺手的不同软件。 他们只是想要一个能让现有软件运行更快的加速器，而不是Amazon自有的加速器。
BRIAN SANTO: AI is vitally important for the hyperscalers. And if there’s a glamorous part of the semiconductor industry, it’s the processor segment. When it comes to AI, the glamor is all in co-processing AI accelerators. Less emphasized but equally important, AI systems rely heavily on memory devices. Gary Hilson covers semiconductor memory technology for EE Times. His contribution to our special project was an article chronicling the extent to which hyperscalers are pushing memory technology so much harder and faster than ever before. So much so that they’re engaging in parts of the IC manufacturing process that chip customers have almost never been involved in before.
BRIAN SANTO: AI对于超大规模企业来讲至关重要。 如果半导体行业有一个光鲜亮丽的部分，那必然就是处理器领域。 在AI方面，共同处理AI加速器集焦点于一身。 不那么突出但同样重要的问题，是AI系统严重依赖于存储设备。 Gary Hilson负责 EE Times的半导体存储技术方面的工作。 他为我们这个特殊项目贡献了一篇文章，记录了超大规模企业比以往更加努力和快速地推动内存技术发展。这些企业的 参与程度极高，甚至参与了芯片客户以前都从未参与过的IC制造过程中的某些部分。
GARY HILSON: The hyperscalers are just so aggressive relative to other companies that they’re just pushing the momentum of things that will probably happen, would happen anyway over time but they’re not content, uh, for that to play out. So, they have a vision for that the things they want to do and they’re, they’re sharing that visions, uh, with the semiconductor industry and in particular the capabilities and requirements that they need from memory.
GARY HILSON: 相对于其他公司来说，超大规模公司是如此具有侵略性以至于他们只要推动可能发生的势头，这些事随着时间的推移就必然都会发生。但他们并不满足于此。 因此，他们对想要做的事情和他们的愿景是非常有远见的，他们正在与半导体行业分享这些愿景，特别是他们对存储器的容量和要求。
BRIAN SANTO: So, memory is interesting. So, we’ve been talking about AI inference engines and AI accelerators but tell us a little bit about some of the requirements they have uh, for new types of memory or new variations of existing memories.
BRIAN SANTO: 所以，存储器也是很让人感兴趣的。我们一直在讨论AI推理引擎和AI加速器，那请告诉我们一些有关它们的新要求，包括新的存储器类型或现有存储器的新变化。
GARY HILSON: Well I think for the hyperscale companies, they’re not necessarily interested in being the guinea pigs for new emerging memories and we talk about emerging memories, those are ones that have been around for a while, the MRAMS, the ReRAMS, the PCM, you know and that comes in the form of, you know intel octane. They’re definitely look at those. Um, but ultimately, they want to get, their trying to move data back and forth as quickly as possible for some of these applications to reduce the latency. Just as importantly, is to reduce the power, uh, because of the scale that they’re doing these things, it’s a huge scale so, uh, and, and in general we talk about how much power data centers, you know eat up and so in general they’re looking to keep the power down. So, there’s a performance requirement that they want to meet but it’s also a power consumption because that’s a big part of their cost is operating, you know, these devices with these memories. So, if an emerging memory can be made, and, and hit those you know key requirements, the performance, you know, the scaling, and the low power consumption, I think they’d go for it. But at the same time, if something say solution that combines DRAM and flash is 5 dollars cheaper per device, they’re going to go with that with works, with what works because that extra 5 dollars on the scale that they’re doing things is going to add up in terms of cost. So, I think they’d need a really strong value. through opposition on one of these emerging memories to say I’m going to, we’re going to pay this premium. You’d have to do something really unique. Otherwise, they’re going to look at DRAM and flash and similarly the GGDR Memory and the high bandwidth memory although that’s one of the more expensive ones so I don’t think. I think they would, might use it in select areas, it’s worth the premium. But they’re not necessarily just because they have all this money doesn’t mean they’re going to be throwing it around, per say.
GARY HILSON: 好吧，我认为对于超大规模公司来说，他们并不一定对成为新型存储器的实验小老鼠感兴趣。我们谈到新型存储器，就是那些市面上已经存在了一段时间的存储器，如MRAMS，ReRAMS，PCM，还有intel octane。他们肯定会注意到这些品类。但最终他们希望，且试图尽可能快地在这些应用之间来回移动数据，以减少延迟。同样重要的是，降低功耗，因为他们正在从事的业务规模非常巨大，且一般来说，我们谈论数据中心的能耗是很高的，所以通常他们都希望能把功耗降低。因此，他们希望性能达到要求，但这也带来功耗，运营开支是他们成本支出中的一个重要部分，这些设备、存储器之类的。所以如果能制造出新型存储器，并且能满足那些关键要求，如性能、规格和低功耗的产品，那我认为他们会选择这个产品。但与此同时，如果有解决方案，能结合DRAM和闪存，且每台设备便宜5美元，那么他们肯定会选择，因为对他们正在做的业务来说，额外5美元合算在成本上，将是很大笔的金额了。我认为它们需要具备非常强大的价值。从新型存储器侧面来看，我们将愿意支付这笔溢价。但这些设备商必须做些独一无二的事情。否则，超大规模企业会对DRAM和闪存，以及类似的GGDR内存和高带宽内存感兴趣，尽管这些都是更昂贵的存储器。我认为这些超大规模企业可能会在特定的，值得花钱的领域使用这些更昂贵的存储器。他们这些公司有足够财力，并不意味着他们会随便乱花。
BRIAN SANTO: But yet, they’re, they’ve got the money, they’re willing to spend it for what they want. And it’s kind of interesting what we’re hearing now is that um, when they decide what they want if they don’t see it on the open market, um, they’re increasingly willing to design it themselves with the example, the [UNKNOWN] that you mention before and now we’re actually seeing them hanging out at Semicon West with the semiconductor production equipment manufacturers and actually making suggestions about the process technology itself uh, required to, to get to the performance and uh, advances uh that they’re hoping to see. Is – that’s unusual, right? To get that involved?
BRIAN SANTO:这些超大规模公司确实是足够有钱，也愿意为他们想要的东西花钱。 我们现在听到的有趣消息是，当这些公司决定想要什么时，如果在市面上找不到，他们越来越愿意参照例子来自己设计，你之前提到的[ UNKNOWN]，实际上现在我们看到他们在半导体生产设备制造商的Semicon West参观，并且就制造过程技术本身提出建议、要求，以达到他们的期望性能，推进他们希望看到的结果。这很不寻常对吗？ 参与到这种程度？
GARY HILSON: It’s, it’s definitely a change uh, from how, how, you now previous big influencers have, have worked with the memory business. I mean, the, one of the, the big players that used to be able to and you know still does to a large degree is Apple saying hey, we need this. Uh, but they were more about can you add a pin to a certain kind of memory, um, but otherwise it wasn’t about massive innovation and of course, uh, the, the memory makers would respond to that because of the market size. I think the hyperscalers are definitely interested in working directly with, you know fabrication companies and coming up with a process, if they can, you know, pull together the materials and the memory more quickly that meet all their requirements. So they’re definitely collaborating more – it’s I think it’s early stages but certainly the, the fab, you know the manufacturing companies uh, you know that I’ve spoken to are certainly looking at that, their, their getting approach, they’re explore those options to come up with these really specific solutions that these hyperscalers uh, want to hit those performance targets and to hit those power consumption goals.
GARY HILSON: 就是，这绝对是一个改变，你也知道以前的大趋势是如何影响存储器业务的。我的意思是，过去能够达到这样影响力的大 “玩家”之一，也就是Apple，现在很大程度上仍然影响市场，他们会直接说：“嗨，我们需要这个”。但更多的，是关注你能否为某种特定存储器添加bin，而不是它是否有关于大规模的创新，当然，由于市场规模，存储器制造商会对此做出回应。如果能够更快地将材料和内存整合在一起，我认为超大规模公司肯定有兴趣直接与制造公司合作，并提出流程，以满足他们的所有要求。因此，他们肯定会更多地合作—— 我认为这是早期阶段，但肯定制造公司，会如我所说过那样，关注制造流程，他们重新探索这些选项，以为这些超大规模企业提出解决方案，满足他们希望达到的性能目标，并实现这些功耗目标。
BRIAN SANTO: So as Gary Hilson just observed, hyperscalers are even reaching down into process technology. EE Times editor Dylan McGrath was at Semicon West, the annual conference for semiconductor production equipment manufacturers. Most companies that buy computer chips typically find fab equipment to be a little too far downstream to get interested in. But with the hyperscalers, that’s changed. We asked Dylan about that.
BRIAN SANTO: 正如Gary Hilson刚才所说，超大规模企业甚至正在深入研究工艺技术。 EE Times编辑Dylan McGrath参加了半导体生产设备制造商年度会议Semicon West。 大多数购买计算机芯片的公司通常会发现制造设备有点过于下游而无法引起人们的兴趣。但随着超大规模企业的发展，这种情况发生了变化。 我们接下来与Dylan聊聊这件事。
BRIAN SANTO: So Dylan, you were at Semicon West earlier this month, uh, what did you see there?
BRIAN SANTO: 那么Dylan，你本月早些时候参加了Semicon West，你在展会上有什么新发现？
DYLAN MCGRATH: One of the most, uh, clear trends was just the presence of people from Facebook, Google, um, engineers from some of these big uh, web companies that are actually there are Semicon West now in increasing numbers over the past few years and really kind of getting as involved as they can in, in driving the process technology.
DYLAN MCGRATH: 来自Facebook，Google，以及来自一些大型网络公司的工程师的出现显示了会上一个最明确的趋势，实际上在过去几年中，Semicon West上出现的网络公司数量越来越多，在推动工艺技术方面，这些公司也想要尽可能地参与其中。
BRIAN SANTO: So put that in perspective — there have been enormous companies in the past. Uh, enormous organizations, big customers of Semiconductors, it might be GM or Ford or, or uh, a space organization like NASA, or a company like Cray Research. These are all companies that have had leading edge applications and interest in uh, having some very specific things delivered in the Semiconductors they’ve wanted. Have they every gotten this, uh, involved in the process before? Not only make, you know, designing their chips but getting involved in the process technology?
BRIAN SANTO: 因此我的观点是 —— 过去有很多公司， 大型组织，半导体大客户，可能是通用汽车或福特，或是像NASA这样的太空组织，或像Cray Research这样的公司。 这些具有前沿应用和兴趣的公司，在他们期望得到的半导体器件中提出了一些非常具体的要求。 他们每个之前都参与过制造过程吗？ 不仅要设计他们想要的芯片，还要亲自参与工艺技术吗？
DYLAN MCGRATH: So, I think the short answer is no, they have not. Um, and I’ll qualify that a little bit. I, I, you know, I think the space agencies have always had some role. They always have some type of presence but uh, there are a few things that are different now. I mean for one, the international technology road map for Semiconductors is no longer enforced. I think all the other companies you mentioned were, were buying the chips that were available based on the roadmap and for the most part, they didn’t have the need to really, they didn’t want anything special. They wanted the most advanced chips they could get or in some cases not even the most advanced, but they were happy to ride, the, ride the roadmap and uh, and buy what was available. You know, today the roadmap is no longer being updated and there are some divergence uh, points where different, some of these companies want different things, um. And so, they’re in there trying to pedal their influence and get the equipment companies to cater to what they want to do ultimately.
DYLAN MCGRATH: 所以，我认为答案明显是否定的，他们没有。嗯，我会稍微确定一下。你知道的，我认为太空机构总是在这里面扮演着某些角色。他们总是以某种形式突显存在感，但是现在情况有所不同了。我的意思是，半导体的国际技术线路图不再强制执行。我认为你提到的所有其他公司都是根据线路图购买可用的芯片，而且在大多数情况下，他们其实没有真正需要，他们也不想要任何特别的东西。他们只是想要可以获得最先进的芯片，或者某些时候甚至不是最先进的芯片，但是他们很乐意去购买这些市面上有售卖的。你知道，今天线路图不再更新，并且存在一些分歧，其中一些公司持有不同观点，想要不同的东西。因此，这些大公司在那里试图利用影响力促使设备公司满足他们最终想做的事情。
BRIAN SANTO: And they’re not just listening. Uh, going to conferences, listening in, finding out what’s available, offering their opinions, telling people what it is they’d like to see – they’re actually getting equipment vendors to make specific machines for the types of things they want.
BRIAN SANTO: 而且他们不只是听别人讲。 他们去参加会议，听取发言，找出可用设备，提供意见，告诉人们他们想看到的是什么——实际上他们就是在让设备供应商制作满足其需求的特定仪器。
DYLAN MCGRATH: Yeah it does appear that that is just starting to happen and what, what uh, at Semicon West this year, uh, applied materials made an announcement of a, several new endura PVD platforms that are, it seemed to be geared directly toward this trend and that, that is that these are for new emerging types of memories and I think uh, one of the things is you know, they cover many different types of memories and I think, think part of the reason for that is these companies are, are looking at different ways and different types of memory schemes to advance and, and increase the pace of their AI accelerators.
DYLAN MCGRATH: 是的，这个趋势似乎刚刚开始，今年在Semicon West，应用材料公司宣布了几个新的endura PVD平台，它似乎直面这一趋势——这些新兴的存储类型，我想其中一件事你应该知道，它们涵盖了许多不同类型的存储器，我认为部分原因是这些公司正在寻找不同方式和不同类型的存储方案，来推进和增加AI加速器的优化速度。
BRIAN: SANTO: Right so they have. A very specific applications, they’re AI accelerators and they’re willing to push that technology so hard, uh, they’re willing to underwrite the development of new, of, of new equipment.
BRIAN: SANTO: 对，他们确实是有这样做。 都指向了一个非常具体的应用——AI加速器，大公司们愿意如此努力地推动这项技术，他们愿意承保新设备的开发的费用。
DYLAN MCGRATH: Yeah, I don’t, I don’t have any specific knowledge of the, of the underwriting um, portion of it, but I think a promise if there, obviously. Obviously the prom of developing these tools, the promise of purchase orders and some of them have already been sold so they, we know that they have the money and they are highly incentivized to uh, create the best possible AI silicon that they can. It means a lot of revenue for them and they’re not afraid to spend and I think that applied in all the other equipment vendors ultimately will follow the money.
DYLAN MCGRATH: 是的，我对承保费用的部分倒没有任何具体了解，但我认为那显然是一个承诺，是开发这些工具的承诺，对采购订单的承诺，其中一些产品已经被预售。所以，我们知道他们有资金，且高度激励尽其所能地去创造最好的AI芯片。 这对超大型企业来说意味着大笔收入，他们并不害怕花钱，而且我认为适用于所有其他设备的供应商最终都会追随这笔投资。
BRIAN SANTO: So that’s where we are; that’s where the industry is today. The hyperscalers have some very specific needs they feel won’t be met unless they push the industry in ways it’s never been pushed before. As a practical matter, what does that mean for the semiconductor industry moving forward? Here’s Rick Merritt:
BRIAN SANTO: 这就是我们目前所处的位置，也是今天这个行业的现状。 超大规模公司有着非常具体的需求，他们认为除非其以前所未有的方式推动行业发展，否则这些需求是不会被满足的。那 实际上，这对半导体行业的发展又意味着什么？ 接下来连线到Rick Merritt。
RICK MERRITT: Well I guess the big message for our semiconductor readers is you know, embrace this. These guys have a lot of money and a lot of demands but embrace this smartly. Don’t try to do a custom everything for everyone. Continue to do your good marketing thing, work with these seven whales and figure out the few products that the most of them will buy. Otherwise you’ll drive yourself crazy.
RICK MERRITT: 嗯，我想对我们半导体读者来讲，最重要的信息就是接受这一现状。 这些家伙们资金丰厚，需求量大，也识时务地接受了这个现状。 不要试图去取悦每一位顾客。 继续做好营销工作，与这七家行业巨头一起，找出大多数客户会购买的少数产品。 否则你会把自己逼疯。
BRIAN SANTO: And here’s Dylan McGrath’s take on pretty much the same question:
BRIAN SANTO: Dylan McGrath在这里也提出了几乎相同的问题：
DYLAN MCGRATH: I think on the whole it’s a positive. I think that just about everyone would say this is all very positive. Um, the more companies that you have coming in, getting involved, spending money I guess is the main one, you know certainly the equine vendors are very happy to see it. There’s a whole new trend, whole new wave that’s driving the production of silicon which is in turn driving the sale of tools, uh and I think that everybody is pretty happy with the more the merrier, I think. Again, especially from the equipment guides perspective. I think you know there maybe some – there may be some ship vendors who aren’t thrilled with it, I’m sure that they would just assume these companies bought the products that they were making but, if these companies feel that they could do that and still differentiate themselves from their competition, I’m sure that that’s what they would do because they don’t necessarily want to be in the semiconductor business but they want to have the differentiation when it comes to AI that will put them above their competitors.
DYLAN MCGRATH: 我认为总体来说，这种情况是积极的。我想几乎每个人都会说这一切都非常积极。嗯，参与进来的公司越多，最主要的是参与进来的资金越多，供应商们很乐见的。这是一个全新的趋势、全新的浪潮，正推动硅的生产，这也反过来推动了工具的销售。我认为每个人都挺开心的，觉得越多越好。特别是从设备指南的角度来看。我想，你知道的，可能还有其它情况 ——可能船厂供应商对此并不感兴趣，我敢肯定他们会假设这些公司在购买其正在制造的产品，但是，如果这些公司认为他们也可以制造那些产品。并且从竞争中脱颖而出，我确信这是他们会做的事情，因为他们不一定想要从事半导体业务，但是他们希望在谈到AI时，能够跟其他厂商有所区别，这将使他们超越竞争对手。
BRIAN SANTO: Not too long ago, Junko caught up with Wally Rhines, the CEO Emeritus of Mentor Graphics, and asked him about the hyperscalers and how their decision to become more vertical would affect the EDA industry. Here’s what he had to say:
BRIAN SANTO: 不久前，Junko采访了Mentor Graphics的荣誉CEO Wally Rhines，向他询问了关于超大规模公司的情况，以及他们的决定变得更加垂直会如何影响EDA行业。 接下来是Wally Rhines的采访：
WALLY RHINES: The first thing is it’s a big financial opportunity for EDA because these companies have lots of resources, they tend to be very well funded and so they grow the EDA market, uh, for existing tools substantially.
WALLY RHINES: 首先，这从财务方面来说，对EDA来说是一个巨大机会，因为这些超大规模公司拥有大量资源，他们往往资金充足，因此对于现有工具，他们能极大地帮助EDA市场发展。
BRIAN SANTO: So basically there are some pretty profound changes within the semiconductor industry that are still in progress. The good thing is that many of these changes will be opportunities, for innovation, for growth, for more business.
BRIAN SANTO: 所以基本上半导体行业中仍有一些非常深刻的变化在酝酿中。 好的是这些变化许多都将是机遇、创新、增长，带来更多业务。
But that’s not the end of our story. Something Wally Rhines said about the hyperscalers caught our attention.
但这不是我们今天故事的结局。 Wally Rhines关于超大公司的言论引起了我们的关注。
WALLY RHINES: In many cases, different priorities, uh I don’t know if you remember the uh, presentation on the Google contact lens that was at [UNKNOWN] years ago. Now why didn’t Google have somebody else design a chip that when into that lens? I have to conclude it’s because Google doesn’t uh, want to get in the chip business. They want to own the information and they don’t want to make that chip available to other people. They want to own the system, the contact lens, the relationship with the doctor, the uh, program on the cell phone to control the insulin pump. And so a lot of the IOT applications I see, whoever is developing the silicon plans to make their money from the information.
WALLY RHINES: 在许多不同优先级的情况下，我不知道你是否记得，几年前在[UNKNOWN]上，谷歌的隐形眼镜上演示。 现在为什么谷歌没有其他人在进入隐形眼镜市场的时候设计芯片了？ 我必须得出结论，因为谷歌不想进入芯片业务市场。 他们希望拥有这些信息，并且不希望将该芯片和他人共享。 他们想拥有系统、隐形眼镜，和与医生的关系，利用手机上的程序来控制胰岛素泵。 因此，我看到很多物联网应用，无论是谁在开发芯片计划，谷歌都从信息中赚钱。
BRIAN SANTO: Did you catch that? When a company like Google has the means to build an entire system – it can own the system – in this particular example, the contact lenses, the relationships with the doctors, the program on the cell phone to control the insulin pump!
BRIAN SANTO: 你抓住重点了吗？ 当像谷歌这样的公司有能力建立整个系统—— 它可以拥有系统 ——在这个特例中，隐形眼镜与医生的关系，利用手机程序来控制胰岛素泵！
To engineers and technologists – to guys like Brad Booth – those guys have a very specific a job – trying to make a data center go faster – and their focus is on just getting that job done. But how they accomplish it matters, and part of the method of getting the job done is becoming vertical at a scale few companies have ever been vertical before. That’s business, and the business creates the Big Picture – it IS the Big Picture. And the Big Picture is that the hyperscalers, which already control so much in our daily lives, are poised to control even more.
对于工程师和技术专家 —— 对像Brad Booth这样的人来说 ——这些人有一份非常特定的工作—— 试图让数据中心运转得更快 ——这是他们工作的重心。 但是如何完成这项工作非常重要，而且完成工作的一部分方法正变得越来越垂直，以前很少有公司如此。 这就是生意，生意创造了Big Picture ——确实可以称之为Big Picture（宏图大业）。 而那些超大规模公司就是Big Picture，他们已经在我们日常生活中扮演了主导角色，并且做好了主导更多角色的准备。
Does anyone else find that alarming? That these hyperscale companies are getting so enormous it is conceivable they could gain control of our health care? Sure, there isn’t a health care system in the world that couldn’t be improved, but is letting Google take care of … all of it … really the best thing for anyone other than Google?
还有其他人发现这令人担忧的现状吗？ 这些超大规模公司规模如此巨大，可以想象他们可以控制我们的医保吗？ 当然，世界上还没有一个完美到不需要优化的医疗保健系统，但让谷歌去监管医保......亦或是其它所有的事物......对除谷歌以外的任何人来说真的是好事吗？
Or aybe there will be competition among Google and its hyperscale rivals to control your entire health system. I’m not sure that’s any better at all; I suspect it might be worse. Facebook just got fined $5 billion dollars for being cavalier with our personal information. FIVE BILLION DOLLARS. And to Facebook, that amount was inconsequential. Even more worrisome, they get to keep playing with our personal information.
或者谷歌及其超大规模企业对手之间，可能以控制整个医疗体系进行竞争。 我不确定那种情况是否更好；我倒怀疑那可能会更糟。 Facebook因为无视我们的个人信息权益而被罚款50亿美元。 整整50亿美元。 对于Facebook来说，这个金额可能并不肉痛。 更令人担忧的是，他们可能继续侵犯我们的个人信息。
Google, Facebook, the hyperscalers – they’re on the brink of gaining more power. There is no telling where this could go. But we can’t consider the consequences of what these companies are doing if we aren’t aware of what these companies are doing. Well, now we’re aware. What do we think about the consequences?
谷歌，Facebook，那些超大规模公司 —— 他们正处在获得更多权力的边缘。 没人知道今后会发生什么。 但是，如果我们不了解这些公司正在做的事，我们就无法思考这些事会带来的后果。 好吧，现在我们已经有所认知了。 我们又该如何看待这些后果呢？
That’s your Weekly Briefing for the week ending July 26th.
This podcast is Produced by AspenCore Studio.
It was Engineered by Taylor Marvin and Greg McRae at Coupe Studios.
由Coupe Studios的Taylor Marvin和Greg McRae设计。
The Segment Producer was Kaitie Huss.
The transcript of this podcast can be found on EETimes.com, complete with links to the articles we refer to. Be sure to join us next week for your July 26th Weekly Briefing on EE Times On Air. I’m Brian Santo.
此播客的文字版可在EETimes.com上找到，并附有我们所参考文章的链接。今天是7月26日，EE Times On Air每周新闻联播。 我是Brian Santo。敬请收听我们下周的节目。