-core things (public account: aichip001)
compiler | Cao Yubei
edit | Panken
There is a special chip company that has never designed a chip, but is in the center of the huge world of technology, and exists in mobile phones, computers, TVs, cars and even data centers that drive everything. It provides core underlying technical support for chip giants such as Apple , Qualcomm , Samsung and other chip giants.
It is Arm, a British chip design company that was first jointly founded by Apple, was later acquired by Japan's SoftBank Group, and almost sold to the US GPU chip giant Nvidia.
Arm licenses the instruction set architecture (ISA) to companies with chip design needs and charges usage fees from them. In this way, Arm's influence spreads to the world, and most mobile devices represented by smartphones are inseparable from Arm's chip technology.
On Tuesday, Nilay Patel, chief editor of the American technology media The Verge, and Rene Haas, the new CEO of Arm, completed a conversation. Nilay gradually moved from the most basic questions to talk about how Arm makes money, whether it is affected by the chip shortage, how to view competition with the x86 and RISC-V instruction set architecture, and progress in IPO listing.
▲Arm CEO Rene Haash
In February this year, on the same day Nvidia announced that it would give up its acquisition of Arm, the Arm board of directors announced the appointment of Rene Haas as the new CEO. Although Rene Haas has been the CEO of Arm for only half a year, he has actually worked at Arm for nearly 9 years. In his opinion, Arm is the Swiss in the electronics industry and will remain as neutral as possible and not try to pick winners.
He also shared his views on the US Chip Act, believing that this is not only important to the United States, but in 50 years, every continent should have world-class chip factories. "We should not worry about geopolitics' concerns about how the world works, because it is like the oxygen in the world."
The following is a complete compilation of the interview transcript of Xintiao:
1. "Almost all people you can think of are our customers."
Nilay: Arm is facing many changes now, and you also have plans to lead Arm to the market in the future, and recently received a revenue report from the previous quarter. Arm is a very interesting company, so I think we can start from the beginning. Arm is a key puzzle in the entire technology ecosystem. It affects almost everyone, but its image is opaque. Consumers do not have any direct relationship with Arm, but many other companies are in the middle to create a carrier identity. So, let's start from scratch, what exactly does Arm do?
Rene Haas: Just like you said, Arm is indeed not a well-known company, and the outside world does not understand us, but we think we are still very important.
First of all, we are in the semiconductor value chain of the semiconductor world. You can see Arm's technology in almost all semiconductor products or OEM products. We are in mobile phones, laptops, , and smart TVs. Just looking at my desk, there are almost dozens of things that contain Arm processor .
Arm does not directly build products or produce chips, but builds the designs inside the chips. We license our designs to those who want to build the final product. Our most famous product is CPU, which is the brain of almost all types of electronic devices.
Since Arm does not make chips, he directly authorizes his brain to those who are willing, so Arm's brain is everywhere. Judging from the data from last quarter, between all semiconductor companies and OEMs around the world, Arm Manufacturing has delivered 7.4 billion chips with built-in Arm CPUs, GPUs or other technologies. This is a very huge number, so we are in the semiconductor value chain, but we will not build any products. We are only responsible for design, especially the design and processing of microprocessors.
Nilay: Who are your customers?
Rene Haas: This question is very interesting. In the electronics industry, it seems easier to say who is not our customers because almost anyone you can imagine is our customers.
Our customers include Samsung, TSMC, Grofontein and other companies that actually produce chips, as well as Intel , AMD , Nvidia, Qualcomm, Amazon , Microsoft , Google . When you go to other parts of the world, we also have Alibaba , Tencent and ByteDance. Almost everyone is our customers.
Nilay: So how do you make money from these customers? Are they just purchasing a license to design? Or buy chip samples? Do they need to pay a patent licensing fee? How do these work?
Rene Haas: Our business model is divided into two parts. The first part of
is the preliminary licensing fee. Only after the cooperative customers pay us this amount can they obtain the right to use the technology, so they have the right to further design on our technology.
The second part is that if these designs are finally put into production and form the final product, we will charge a unit fee based on some mathematical algorithms related to the contract.
Therefore, it can be said that we have two sources of income, one is authorized licensing income, and the other is usage fee income.
Nilay: If I bought a Snapdragon chip from Qualcomm, do I need to pay you? Or does Qualcomm have to pay you? Or maybe someone else has to pay?
Rene Haas: You don't need to pay, but Qualcomm needs it. In the case you mentioned, Qualcomm needs to inform us of the number of chips you purchased. We and Qualcomm will have a pre-agreed license fee rate, and Qualcomm will pay the money according to the agreed fee rate.
Nilay: Samsung is designing and producing its own chips for their own smartphones, and the departments inside it must also sign conditions with each other. If I bought Samsung phone with Samsung Exynos chip instead of Qualcomm chip, how would that money go back to you?
Rene Haas: This is the wonderful thing about Arm's business model. Qualcomm uses our technology, so does Samsung. When the Galaxy phone is shipped, we will most likely receive payment from Samsung. If this product is partnered with Qualcomm, we will receive payment from Qualcomm.
In short, we will all get payment. This depends on the licensing situation between us and our partners, and usually we will work with the semiconductor departments of these companies. Take Samsung as an example. If we sign a contract with the chip department, the chip department will need to pay Arm.
Nilay: The last question at the top of the difficulty ladder. Apple offers a very unique license called the Architecture License. They completely design their own chips, I don't think they use a lot of your designs, but this is Arm's intellectual property. So when I buy a MacBook Air (M2), how do you get paid?
Rene Haas: It's all similar in business logic. We sign contracts with companies like Apple, which pay us copyright taxes like others.
Nilay: Can anyone get a license like Apple? Use your IP, but completely design your own stuff?
Rene Haas: Architectural license basically gives companies the right to make what we call Arm compatible processors. Anyone who does this can make subtle modifications to the microarchitecture, which is actually how to physically place it on the chip.
But what these companies can't do is modify the CPU, which will prevent the chip from running Arm's instructions. And this is crucial because ultimately, we must maintain software compatibility. Whether it is something we build ourselves or something that licensed partners have to build, as long as they are running Arm processors, they must meet the requirements of running Arm software.
We have built a very good CPU, and I think it is very difficult to build a CPU that is different or better than us, while still complying with Arm standards. Of course, this may also be my own bias.
There were many people doing this in the past, but now there are fewer people because, first, it is difficult to do, and second, it is difficult to find people to do this.It is very difficult to build such a talent team. Most companies will say on the road: "If I build an SoC (system-on-chip) and I only have so many valuable engineers, then a differentiated Arm CPU may not be a place worth investing in." Moreover, they have other places that need to be invested, such as cameras, modems or IO support.
Nilay: When the authorized party of Apple or other architecture delivers the product, do you have a team to verify that it is running Arm instruction set and they are not violating the rules? Or are you just using the honor system so that you won't get too fucked (Apple CEO) Tim Cook ?
Rene Haas: We have a set of requirements in this regard and a compliance suite that allows us to test to fundamentally verify that the content they build is in compliance with Arm standards. We will test it to check if the Arm instructions and code can be run, if the compiler is corrupted, or if the correct instructions cannot be recognized. So the short answer is that we do provide a set of compliance tests for all building designs based on architecture licensing.
Nilay: Apart from the Intel and AMD CPUs that people might encounter, because it's a little opaque, I think closed loop is where you really validate that this is the Arm thing that works in the Arm way, and I think most people don't understand that, so I want to ask this set of questions and feel the business.
Rene Haas: These are all very important issues. One thing that lets us be almost everywhere in history is that whether it is based on Arm CPU core authorization or Arm instruction set architecture authorization, they need to run software for Arm instruction set architecture without interruption.
Arm has some failed CPUs that allow extensions, i.e. allow customers to add custom instructions, and while this sounds innovative and cool, what is really related to the CPU architecture is to let developers know that it will run.
If the developer is writing a piece of code for the OEM and embedding Arm, the developer doesn't want to know, doesn't need to know, or even knows that they are designing a thermostat with a Bob chip built in, and Bob has some additional instructions. You need to take advantage of this because they may not know if other OEM devices have an Arm chip that contains these instructions.
plays fairly and ensures that the software dataset looks the same. Our founders did a great job of sticking to this and making it quick. You can see that this is really benefiting us now.
2. Stay neutral and don’t try to select winners
Nilay: Many of your customers have unshirkable competitive relationships in multiple fields. I've talked to a lot of executives like you. In my opinion, they do much more interpersonal communication than engineering and research jobs. Are you the same? Do you think you are a business politician who ensures fairness to balance competitive relationships, or do you still focus more on processor design and R&D?
Rene Haas: We are definitely stuck in all kinds of interpersonal communication, which is what we are going to do, but we also spent a lot of time and effort developing these CPU and software ecosystems to solve the problems of our partners and truly ensure that the product is at the forefront. What’s a little different about
is that when we deal with everyone, we have to maintain consistency in the way we handle partnerships, which is really around access to technology, access to errors, and access to people.
The world depends on Arm, as you said, we are a little opaque, and on the podcast, the world depends on Arm, and someone will listen to it and want to say, "I rely on you? I don't know you!" But we are really managing our relationship with our partners very seriously.
Nilay: Let me give you another simple example, I really find this example very inspiring. A few years ago, we talked to people running HDMI (HD Multimedia Interface) that they set the industry standard for plugging PlayStation (a home gaming console released in Japan in 1994) into their TVs.Of course, this is why HDMI doesn't work perfectly when your remote control can control everything on your TV. To be honest, they have raised their hands and surrendered: "This problem is too difficult to solve, no one wants to solve this, everyone is making a fuss." From a political perspective, I understand why this is their answer, because it is the lowest bet, right? I just wanted to plug my DVD player into the TV and let the remote control control the work, but the political nature inside was insurmountable, and it was not an industry standard agency, so it needed to take a lot of risks. Arm is a company that needs to report profits to shareholders. If you want to go public, you must increase profits for your shareholders. How will you deal with the friction and tension between these?
Rene Haas: We really try to stay neutral. We are called "Switzerland in the electronics industry" (known for its neutral attitude), which is a good comparison. We won't try to pick winners, we will get involved in various sub-ecosystems in the ecosystem. If you start from the bottom of the semiconductor chain, all chip manufacturers such as Global Foundries (Global Foundries), Samsung, TSMC, Intel, etc., you will realize that you have to work with all of them.
We must ensure that our technology can be built on every semiconductor process in the world, which requires the investment of all these partners. Then, all the way up, when you think of Android, Linux, Windows, and all the major operating systems we support, we have to make sure we are in it too.
We really try not to have an advantage in the competition between one side and the other. I'm not sure how we will eventually achieve this, but that goal means we're on the shoulders of a considerable ecosystem.
This is an ecosystem with design tools, manufacturing processes, software operating systems and middleware. We often cooperate with relevant industrial chain manufacturers and create them. We didn’t really work much with the standards body, but we did work a lot with all the vendors I just mentioned.
We make sure we understand everything they are trying to do from a roadmap perspective to make sure we are as compliant as possible. Ultimately, we license the technology to those who build chips, who want to have the widest range of opportunities that can be made.
3. The team size is nearly 6,000 people. How does Rene Haas make decisions?
Nilay: I want to know other basics of Arm, how many employees does Arm have?
Rene Haas: The last number of people I saw may be around 5,800, but if you add the contractor, it may be more than 6,000.
Nilay: How are their structures? Are they all chip designs, or are there lawyers? Is the ratio of patent agents to chip designers 1:1? How does this work?
Rene Haas: Of course, it’s not a lawyer or engineer 1:1, so I can only say that.
Nilay: Some companies are 1:1 lawyers and engineers.
Rene Haas: We are still far from that level. Our staff are mainly engineers, most of them in the UK. We have several different engineering sites in different regions of the UK, France and Nordic; there are several design centers in Arizona and Texas in the United States; there are also many design engineers in Bangalore and Noida in India.
In comparison, our legal department is very small. If you use thumbnail scales to display, the ratio of engineers and lawyers is quite large. We did a great job in how the licensing model works and in protecting intellectual property rights. Although there is no large legal department, there are indeed many engineers because these products are really hard to manufacture.
Nilay: Who reports to you? How is your team formed?
Rene Haas: I am the CEO, and we also have the Chief Financial Officer, Chief Personnel Officer and the head of legal reporting. The overall organization is based on the business department and forms a vertical business structure. We have automotive, Internet of Things (IoT), infrastructure and client businesses, as well as chief architects, sales executives and engineering executives, who also report to me.As I listed, there are a lot of direct subordinates. Arm is mainly people-oriented and focuses on engineering, people, legal and financial operations in its business.
Nilay: This question is a classic question I often ask others. You have worked at Arm for 9 years and have been CEO for 6 months. How did you make the decision? What is your framework?
Rene Haas: I had to tell a little lie and say I had dinner with Tony Fadel , I know you interviewed him not long ago, and I appreciate his habit of comparing opinion-based decisions with data-based decisions. What I want to say is that the whiter your beard and hair, the more you will be able to accept opinion-based decisions. Human history will repeat, because we are humans, and humans will repeat the same successes and mistakes.
And I am a mixture of opinions and data, the more experience I rely on intuition. Experience helps. For those fans: Why does Tom Brady still play quarterback at 45, although physically he plays with someone who is only half his age? Because it was difficult to lie to him, he could see everything. Business is far more complicated than professional football.
So, the short answer to your question is data and opinions, both of which I rely on, but now it may be that opinions help me more than data, because my intuition is more intuitive.
Nilay: My understanding of Tom Brady is that he doesn't eat any tomatoes and eats avocado ice cream every day. Are these on your menu too?
Rene Haas: He is a bit like Benjamin Button, and every time he sees him, he gets younger and younger. I also have strange eating habits. British people will know that I eat yogurt and granola every day for lunch. My assistant would almost preemptively say, "I put your yogurt in the refrigerator. You have your name on it."
isn't that I would classify myself as the super successful category, but I do see that from other leaders, which reduce the number of decisions you have to make, and I've always found it helpful to me personally.
I wear the same clothes. I ate the same food. This is something I like, and don't worry.
Nilay: Now let's talk about a big decision. As we all know, Arm is involved in this storm of SoftBank Vision Fund, which raised billions of dollars for Vision Fund. The direct acquisition of Arm is a big deal for SoftBank. They have invested in many companies and are trying to sell it to Nvidia when Vision Fund is a bit unstable. The industry basically lobbies against the deal as much as it can, saying: "We don't want Nvidia to have this core CPU technology, don't do that." The government also came forward to say it would block the deal, and SoftBank eventually gave up.
Then you come as the new CEO and say you want to go public, which is a big flip. Your ex had a firm objection to getting Arm listed because we talked about earlier: the pressure to increase revenues must be increased against the risks faced by a neutral and fair supplier model, after all, you can increase revenue through special transactions. These pressures will come, but it will be your decision, how did you make this decision?
Rene Haas: We announced this shift at the end of last year when Nvidia deals were basically falling apart. After I took over in mid-February and ended our fiscal year in March, we were finally able to talk about financial results, something that hasn't been done in a while.
During the Nvidia era, we were very low-key, and when we announced our revenue this year, we set a record of over $2 billion - $2.6 billion, and we have never done more than $2 billion in the past. Our operating margin is close to 40%, but almost everyone thinks we are losing money because we are very low-key. If you fast forward to the quarter, it's even higher than what we ended last year: over 50% and $700 million in revenue, of which $450 million is royalties .
I had several calls with analysts and reporters like, "Where does this come from? What interesting math are you doing? Is this some form of equation?" But in fact, we know we're doing well.
Shortly after SoftBank acquired Arm, we reorganized and created two business units, and I took over Classic Arm and started to turn to other markets. This fulcrum is not only a business model, but also a product. We know that some changes are happening in the data center, TSMC is doing a very good job in processing, and we are also making good progress in software workloads. We think that if we push our investments to specific instructions, such as SME and SVE, which are vector expansions for specific workloads on hyperscale computers, we can gain something in the hyperscale computer market.
Nilay: Let everyone know that this is the cloud computing market, that is your Google Cloud, AWS stuff.
Rene Haas: That's your cloud. AWS is our important partner. They announced the Graviton2, along with some rather eye-catching numbers, with a 40% increase in price performance compared to other architectures. Therefore, we diversify our business by not only developing different products, but also solving this problem through different parts of our business model strategy.
We know our business will get better. All the financial results you see now are great, the team is doing a great job in this, and these are really from work from a few years ago and you won't see royalty results for each unit overnight.
We develop IP, and then the IP must be handed over to the customer, they must make a chip, the chip must enter a product, and the product must be qualified. All of this can take 3 to 4 years and we feel good about where we are going to be, and we feel good about the areas we have been investing in, such as cloud, automobiles and the Internet of Things. These are the big long-term growth areas that I think we are very suitable for.
Nilay: You made some bets two years ago and now you have received a generous return. Do you think the pressure of investor in the quarter will change the way you run your company?
Rene Haas: I think at any time you are a public company, it's just part of how the world works. However, I can’t talk too much about what life would look like as a listed company, I’m subject to very strict order restrictions and stay in my area and can’t talk too much about it.
Nilay: One of the few lawyers in the company is about to run into the room.
Rene Haas: Haha, so I really can't say too much. But I can say that we are very confident in the long-term growth of the market involved.
4. For Arm, cars will be a big market
Nilay: You just expressed your confidence in cars, the Internet of Things and the cloud. The cloud market exists and is still growing, and Arm has taken measures to do so. I've talked to a lot of CEOs in the automotive industry and we spent a lot of time talking about smart homes and the Internet of Things on The Verge cast. But those markets have not really turned, and as Arm CEO, can you push this? Or is it more like "we just have to prepare design and products for the car to truly become a network computer on the wheel"?
Rene Haas: When people think of electronic devices in cars, they automatically turn to autonomous driving, etc. Think about what is inside a car and how many processes are in the car. The dashboard is all digital, computerized, the drivetrain, and anything related to the rearview mirror and brakes, all of which are quickly moving to Arm, operating on Arm. Many things that are happening in smart cars are moving towards Arm.
In the powertrain you have many old standalone electronic control units (ECUs) that they may have an old proprietary microcontroller and are not the same as the rest of the car. All of these ECUs are being redesigned, and each car has probably 50-70 ECUs inside, and each one may have an old proprietary microcontroller, they are not connected, there is no memory management unit. Therefore, after adding the automatic and advanced driving assistance system (ADAS), cars can become highly connected devices, which is an area that Arm wants to develop.
For us, cars are a bit like a sandbox of multiple technologies: instrument panel, power transmission, power transmission/ECU.By the way, autonomy is a huge opportunity for us.
Back to the data center field, as far as computing is concerned, what is really important in cars is performance and efficiency, you can't have a server in your trunk running electric car , and some cars today are like this. This will improve over time and we are very optimistic about the automotive market where we will also grow very fast.
Nilay: So, are you going to ask some of your engineers to find an ECU, engine controller or body control module that can work across cars so that Nvidia people can come and get licenses and sell it to Ford?
Rene Haas: This is actually happening. These are used for the automotive CPU efficiency, power and functional safety are important. You have to have all the safeguards to make sure it works safely. Some people do this in software through compute libraries, but most people prefer to do this in hardware because it is safer and more efficient to do it in hardware.
We have developed automotive processors with embedded functional safety and graphics processors . We didn't do this before, and before we basically just rolled out a generic thing for use anywhere you want. This is one of the things we have made a big change in the past few years.
Nilay: We are talking about the future now, let’s bring the topic back to the present. We have been in a chip shortage for a long time, and this "chip shortage" may not end. Intel and Nvidia have had poor quarterly results, and Intel says they will actually raise prices. Where would you be in the case of chip shortage? Is that what affects you now? Is that what you see coming and going? You're a little off from this actual part.
Rene Haas: We are out of the idea of not making anything, but we do have to do with chips, and our royalty model is related to components people order.
Go back to the previous discussion, we are very diverse in the terminal market. Another thing that is happening is that more and more CPUs are used in these SoCs. The application processor of a mobile phone may use one CPU, while the cluster used for computing is now 9 or 10 different CPUs. You have 10 CPUs, it's just an application processor (AP), and then consider things like touch sensors, monitors, and cameras. We see that although sales in certain markets such as smartphones have been weakening, we have been protected by it from the perspective of product direction.
Nilay: Have you heard of the episode we had with Willy Shih? He is here, telling us about toilet paper, and then talking about the manufacturing of chips and LCD displays. Since they have too many SKUs and sell too many types of paper, they decided to reduce the types of toilet paper, and this final decision solved a large part of the problem.
Rene Haas: That's right. So now think of a car that requires various diode , capacitor , resistor and thermal sensors. It is not only because it is difficult to obtain 3 nanometers, but also it is difficult to build an wafer factory.
You can tell from my hair color that I have been in this industry for a long time, I have been in the business cycle, people start to stop researching and developing and slow down projects, they gradually stop trying new things, and stop innovating.
I think it's because everything is digital, the super high demand for electronics and how it will change our lives. What are our indicators? Licensing and new designs have never been so good for us as it is all over the market.
5. "Five years later, every continent should have world-class wafer fabs"
Nilay: What you are talking about is one of the reasons why R&D is accelerating its pace. The world has begun to re-understand the prospects of globalization, and the world is no longer so peaceful. The United States has just passed the Chip Act to truly inspire chip manufacturing and various design investments. TSMC is under construction in the United States, and Intel has just broken ground at its new factory in , Ohio, . Have you seen that we all know that we rely heavily on a handful of fabs in Taiwan, which is a geopolitical hotspot.We need to transfer this critical technological dependency to our own country. Is this something you will be involved in, or are you just watching?
Rene Haas: We will definitely be involved in this. We will help as long as we can expand demand for this by talking to political officials in any country. This is not just an Arm issue, but an industry issue. Nothing to do with single point failure is a good thing, the epidemic has exposed a lot of things, people are starting to open their eyes wide and see what is the real problem. I am very grateful to the people who are pushing the chip bill, and that is really important.
This is not only important to the United States, but from now onwards to the next 50 years, every continent should have its own world-class wafer fab. We don't have to worry about geopolitical issues, the fab will then be like the oxygen that runs the world.
Nilay: Nowadays, cutting-edge process nodes are mainly controlled by TSMC, and they are far ahead of the 5nm and 3nm technologies. Few people can compete at this level. Your smartphone uses cutting-edge technology, while the ones used in cars are basically mature process nodes (40nm, 14nm). The process is severely restricted, and no one will invest in building these wafer fabs. What do you think of this result? How do you think of TSMC, which has such strict restrictions, that everyone wants to improve their construction capabilities on these things, but no one wants to improve their construction capabilities for the old things.
Rene Haas: I'm not a manufacturing expert, but what I can tell you is that we're seeing a lot of things about how to convert certain fabs into new jobs being done.
is like you said. People are usually a little reluctant to put a lot of money into it, and you might think that you can convert the fab we are building into some kind of process technology and turn it into a logic fab? Is there a storage fab that can be converted into a logic fab? But this is a very complicated question, because there are only so many factories in the world, and there are only so many people who know how to build these things. They are all listed companies and must make money. In short, this is a very complex matrix.
30 years ago, Japan's logic wafer factories were world-class and spread all over the country. Now this number has basically dropped to zero. Japan is looking for ways to change this situation to get more fabs back online. I think you'll see a lot of things like that in the United States there are other countries that have completed chip bills and things like that. I think Korean wafer fabs have huge capabilities, but it is also a complex and difficult problem.
6. "Intel CEO has done a lot of the right things"
Nilay: How relied on TSMC? It feels like a lot of companies realize they really rely on one company. It's an industry leader for a reason, but there are a lot of dependencies there. Have you considered this dependency when designing new processor types or architectures?
Rene Haas: We work closely with TSMC. All fabs are important to us and Arm will remain neutral.
Seriously, they are all very important. In an ideal world, we have many people who can build cutting-edge technologies. Generally speaking, in smartphones and data centers, it is usually the cutting-edge process, because people are really trying to achieve the maximum performance possible, not in the automotive and industrial fields. So, any fab that uses cutting-edge processes is very important to us.
Nilay: Intel has long been an integrated design and wafer fab enterprise. They were basically shocked by TSMC, and now, they have a new CEO who is breaking this situation and planning to open their factories to others. You just said that Intel is a customer, are you working with them?
Rene Haas: We hope TSMC can make a lot of Arm products, and they do. We hope Intel will do the same.
Nilay: Are you optimistic that Intel can do this?
Rene Haas: I think (Intel CEO) Pat did a lot of the right things. His work is very hard and done very well, and we are very willing to do more with him.
Nilay: Another political answer, but I have to ask. In terms of chip bills, people are excited about the passage and signing of the bill. Intel has received a lot of criticism for announcing cuts in capital expenditures and increasing dividends rather than putting it into fabs. Broadly speaking, do you think this is the right move? What opportunities should we look for for fabs in the United States?
Rene Haas: I wouldn't doubt Pat's decision on how he spends money. Going back to what you said, we are talking about single point of failure. I feel we need to act quickly, start these facilities as soon as possible, get them online, build more redundancy, and reduce the risk of single point failure. So we need more fabs.
Nilay: What time do you make your decision? If a bill is passed today, or if someone announces a new fab today, and we won’t see it for five years, how do you think about your decision-making timing?
Rene Haas: After six months as CEO, I am still learning and have a lot of responsibilities. One of the biggest responsibilities is to think about the problems in five years.
Because of our business operation, the process from designing products to being able to see revenue is quite long. We are considering where the opportunity lies, where investments need to be made, where threats lies, and where systems and software design are.
Most of my energy is spent thinking about what the world will look like in 2025 and 2026, not next quarter.
7. Please continue to pay attention to PC, Arm is making great progress
Nilay: There are some long-term bets that focus more on consumers than changing servers in the trunk of the car. The hottest trend related to Arm is that Apple has shifted from Intel x86 processors to self-developed M-series processors licensed based on Arm instruction set architecture to support their Macs, which improves battery life and performance while also helps Apple reduce procurement costs. Now, thanks to these chips, the Mac has basically become the best laptop you can buy. This reflects the great success of Arm architecture processors in the PC market. Can Arm help other industries do these things? Or just design and hope Qualcomm can make chips that allow Microsoft and Dell to do that too? From your point of view, how do you view your relationship with the industry?
Rene Haas: I think we can do more things and probably need to do more things. It's not because we need to help the industry, but because the construction of these products is really complicated.
Let's talk about architecture licensing first. Previously, many manufacturers thought you needed to obtain architectural permissions to build a better CPU to compete with Arm, but the IPC instructions per clock of the microprocessor are just one aspect that can really change the design, and also need to consider the memory subsystem, interconnection, cache size, how to interface with the rest of the SoC, and ensure that the SoC is in a multi-chip package with chip and chip interconnection.
Some customers are very good at these things and can figure them out so they don't need our too much help. But over time, more and more businesses will need Arm to do more, which is a trend we are paying close attention to.
Nilay: From your point of view, as long as they are all Arm chips, it doesn’t matter who wins, right? If Apple accounts for 100% of the laptop market, that's great for you. If Microsoft, HP and Qualcomm can pair competitive Windows on Arm laptops and take up 50% of the share, you can still get paid. Do you feel that no matter who wins, you will win?
Rene Haas: If the product is shipped with the Arm instruction set architecture, it is a good thing for us, and if an alternative device is used, it means it does not use the Arm instruction set architecture, which is not very good for us.
When I first took office, I gave a small nickname on the website with my name and worked with the public relations team to use the slogan "Where does calculations happen, Arm will appear." Whether it is thermostats or microwave ovens, Polycoms, smart TVs, laptops, etc., these are all based on Arm.
Nilay: You just said where computing happens, for most people, computing happens on their laptops and desktops. I heard that Windows on Arm (WoA) has been around for 13 years, and although it hasn't happened, are you just satisfied with letting Apple eat up this industry? Because there are still other people's instruction sets in this extremely large industry, such as Intel's instruction sets.
Rene Haas: I worked for a company that developed Windows RT and was the general manager of that product line. Before joining Arm, I worked at Nvidia for many years (at Nvidia for 7 years as Vice President and General Manager of Computing Products Business Unit), and we made tremendous progress in this regard because we felt the real tipping point was there. The
turning point does exist, and during my first news interview on my Windows RT laptop, I remember meeting with a group of reporters and the first question they asked me was: "Does it run iTunes?" We were like, "Ah, damn. No." It couldn't run because we didn't have a port and Apple didn't help us.
Fast forward to 10 years later, streaming audio has emerged and no one really cares about these apps. If you consider all the apps running on your PC or phone, it's hard to say "Well, which native apps don't really run on Arm?" They're almost all there.
Please continue to pay attention to PC matters. We are making a lot of progress. You can see what the features are through laptops in other ecosystems. I think it sounds a wake-up call for the industry in terms of the potential of these features and we will stick with it.
Nilay: That wake-up call is due to a very wealthy company spending a lot of money on customizing your instruction set, right? That's their bargaining chips. Are you saying that as Arm, we can develop products that perform as efficiently per watt as M-series chips and then sell these designs to Qualcomm? Or does it boil down to Qualcomm - I believe it also has an architectural license - doing it itself?
Rene Haas: I have 100% confidence in this. You do not need an architectural license to do this.
8. How to win in GPU?
Nilay: On the other hand, there are laptops, desktops, and even car manufacturers. I went to sit on a (American pure electric vehicle) Rivian and I was told that the GPU and Unreal Engine drive the car. Graphics performance has become a point of differentiation for various products in the industry, and Arm has a very good name new GPU, Immortalis. Apple is not using your GPU. If you consider large GPU vendors, they don't really think about performance per watt, are they? With a high-end Nvidia card, you just need to have a nuclear reactor in the basement to run that thing, which is great. Looks good. How will you win in GPU?
Rene Haas: When you calculate by unit, we are actually the number one supplier of GPUs on the planet. I remember that during the Nvidia acquisition, its CEO Jensen Huang raised a key point: We will be faithful to our review, and performance per watt is important. We won't risk doing a 100W TDP GPU and try to make a difference in that area. We will try it in the 1 watt range and optimize it within that range. Then in some areas, you can start doing more on multiple GPUs.
We are starting to see people thinking about doing machine learning extensions and possibly doing these things within the GPU. This is interesting because GPUs can benefit from some level of AI and machine learning to do shader drawing in a more efficient way. At the same time, you can start working or transfer ML workloads to the GPU. There are a lot of interesting innovations that will be able to be done on our GPUs and we are making a lot of investments in this area. It should be clear that we will stay in an area where performance and efficiency are important.
Nilay: Yes, you delivered a lot of GPUs. It is part of the SoC package for many customers. Customers with their own GPU or their own GPU expansion lead per watt performance.Is the same transaction not important at the end of the day because they are shipping the Arm instruction set architecture and you can still get paid? Or "we have to stay competitive here or they will leave completely"?
Rene Haas: GPU is slightly different from CPU because the API is decimated. Performance per watt is indeed important. We have done a lot of work to level the game and with Immortalis, we are at the forefront in many areas. Another thing that is very critical for these GPUs is the efficiency of the interface between the CPU and the GPU, which is also an area we invest heavily in, but we have to be alert in the GPU market, which, as you said, is very competitive. You let people do it themselves, you let other third parties make the product. At the same time, this is an area we are very concerned about, so we will invest. I think the machine learning AI vectors that will be used with GPUs in the future will be a huge opportunity for us.
Nilay: GPU is a huge investment. You have to spend a lot of money to compete and win the design. Are you making this investment because over time, can you win back these products and get more money from other custom GPUs? Or do you need a competitive GPU just to be competitive?
Rene Haas: I'm back to machine learning and artificial intelligence. If you have a heterogeneous computing system, you have a CPU, GPU, and NPU, I think at some point in time the compiler will get smart enough that they might point to some of the code that can run better on the GPU vs. CPU.
If you think of the entire cluster as a complete subsystem, I think this will bring long-term benefits. For us, especially when designing systems on chip, we are trying to make all kinds of major trade-offs. Sometimes I don't want that workload to run on the CPU because it consumes unnecessary power, and I may have transistors in the GPU, and I may be able to use them if I don't draw. We believe that GPU is a very critical strategic component.
9. There are only two options: Arm and RISC-V
Nilay: GPU competitors, which is obvious in one aspect. CPU competitors seem to be harder to identify, who are your competitors?
Rene Haas: In terms of CPU? I think there are probably only two choices, which are biased, but I even think there is no real choice.
From an instruction set architecture perspective, if you want to build a SoC based on x86, there are only two companies in the world that will do this for you: AMD or Intel. Then there is RISC-V, which is a completely different part because it is open source and there are many different versions.
RISC-V is a scalable processor, which means its advantages are its disadvantages. In my opinion, scalability leads to fragmentation. I think this hurts them in getting any kind of software ecosystem. It's really hard to say, "What is the main software ecosystem that runs consistently on a RISC-V processor?" No.
So, where did we see RISC-V in the system on chip today? It is located in a deep embedded part of the chip, and external programmers do not know its existence.
This is an analogy I can give. If you put something in the microwave and press the start key on the 30-second timer, the monitor might be Arm, because there might be a small piece of open source application code running on it. The timer that actually turns on the oven and turns off after 30 seconds is probably RISC-V.
These are indeed the only two choices. I mean, all of these are small companies, but these companies really don't have much appeal in market share.
Nilay: Do you see Arm and RISC-V will eliminate the outdated x86 world?
Rene Haas: x86 has been around for a long time and has a very strong installation foundation, and I certainly don't want to do anything to devalue what they are doing. I think the challenge with x86 is that it comes from only two companies, and that is a limiting factor in itself in terms of how far it can go.
Nilay: These two companies, Intel and AMD, have their own wafer fabs, and TSMC also have them. There is another manufacturing puzzle that over time these manufacturers seem to say, “Look, we’re really good at Arm.We will be good at RISC-V, we don't want to be good at x86 anymore. ”
Rene Haas: I would even doubt whether they are really really good at Arm. Needless to say, it wasn't easy.
Nilay: Are there any other competitors like RISC-V? I mean, RISC-V is another thing like Windows on Arm. I heard it's been a decade and it seems to be in some of the apps you're talking about.
Rene Haas: It's all about software. Ultimately, it's about having a rich ecosystem of developers that can leverage the software. For us, we think there are 15 to 25 million developers programming on Arm. Some people know they program on Arm because they take advantage of the instruction set, but some people don't know because it's abstracted. The compiler and the open source library are both there, which makes it easy to run.
As an example: smart TV. If you try to write a new menu for a smart TV and try to link to an Android version that is essentially the underlying kernel, it's Arm. It's all optimized for Arm. As far as self-reporting is concerned, it's a big advantage. These libraries are there and have been optimized. Suppose you tried it on RISC-V. If A company adds 17 new instructions to make their RISC-V looks a little different, so developers won't know that. How will developers take advantage of this? RISC-V may end up being reduced to the lowest common denominator.
Nilay: A few months ago, Qualcomm CEO Cristiano Amon participated in the show. I asked him the same Windows on Arm problem, and he kind of blamed the software developers, saying, "Look, Microsoft, Adobe and other large independent software vendors have to join in, they have to make these things quickly." "You just said it's all about software, did you take the time to say to those companies, "Look, you have to turn your attention to the future, that's Arm?" ”
Rene Haas: Are these companies you are talking about referring to Microsoft or Adob?
Nilay: Both are.
Rene Haas: Yes. We spent a lot of time on both, and that's where we ended up winning the battle. You've been watching for 10, 12 years like me, and it's closer than ever because not only do more work on the original apps, but more of these native apps have been written for Arm.
look at two different operating systems in the world. Other operating systems use their names in the code. See all the Microsoft applications that run on their systems only. They're all ported in this way, so all of these Microsoft applications can run on your phone. We're also similar, and I think the turning point comes when there are a lot of pretty good CPU products on the market that can compete with others. In your opinion, this product proves that you don't have to sacrifice performance and game-changing battery life in form factor.
Conclusion: What's next for Arm?
Nilay: I always ask people what's next for their company, but I'll ask you a more specific question. You announced that Arm will be available in March 2023, and as time approaches, can you achieve this goal?
Rene Haas: Unfortunately, we are at a stage where I can't reveal too much.
Nilay: Well, what's next for Arm?
Rene Haas: We're going to continue investing in the areas I mentioned. We think there's really strong growth in some of the topics we just talked about: those complex packaging systems and complex designs. We're trying to find the ability to provide more services to the industry, because I think there's a huge opportunity out there.
Source: The Verge
