DESIGN&ELEKTRONIK: Mike, your owner Softbank invests a lot of money in Arm, the result is very rapid growth. From 2015 to 2018, the number of your employees, including the joint venture in China, increased from around 4,000 to over 6,000, i.e. around 50 percent more in three years. Is this still manageable from an organizational point of view?
Mike Muller: We've already gone through several growth cycles where half of our employees worked for us for less than three years. That's when we learned how integration works. Nevertheless, for the first time I now have the feeling that we are a really large organization (laughs), which is also reflected in the division into the two groups IPG and ISG.
As you grow, the big challenge is that the organization increasingly reflects what the organization wants, not what the customer wants. We have adjusted our new business areas to ensure that the different needs of different markets are reflected. In the past, these were de facto consolidated by the development teams.
To answer your question: Yes, I think we are managing our growth well right now.
DESIGN&ELEKTRONIK: Before we turn to the individual markets, I'd like to discuss your expansion strategy in general. From a SoC's point of view, there is no area for which you do not offer standard IP, in addition to CPUs, GPUs, for example, wireless IP, system IP and security. And then, with the Pelion IoT platform and PSA, you also offer solutions based on your IP. arm is thus advancing into areas that were previously served by licensees with proprietary approaches. Why should an Arm licensee develop a SoC based on the Arm portfolio when he can no longer differentiate himself?
"We continue to enable our partners to innovate and differentiate. «
Mike Muller: I don't agree with you! We don't do complete SoC designs. Of course you are right that we offer IP for the most relevant IP blocks, but there are countless possibilities to bring your solution to the market. Take a look at all the different smartphones, for example. We continue to enable our partners to innovate and differentiate themselves.
DESIGN&ELEKTRONIK: Let's go briefly through poor's core markets. In mobile computing, you have a market share of over 90 percent and the number of cores per device is also approaching a natural limit. How do you intend to further increase your royalties per device in the future when organic market growth is only possible to a limited extent?
Mike Muller: That's right, the growth potential is certainly in additional accelerators such as for machine learning, if you look at the next five years, even with a saturated market you can continue to innovate thanks to a larger transistor budget, just see the arm-Always-On laptops that you can now buy as an example.
DESIGN&ELEKTRONIK: For the infrastructure market, i.e. servers, routers, gateways, base stations and top-of-the-rack switches, you have introduced a new brand "Neoverse" and noted that arm already has a market share of 28 percent - with very different distribution among the applications mentioned. How do you manage to finally become more than a niche provider in the x86-dominated server market?
Mike Muller: You're right, the closer we get to the Edge, the higher our market share will certainly be. The decisive factor for market success is the software base, with Linux we now have a mature and secure base. It takes years to achieve a mature software base, so you can't just take money into your hands and hope for timely success. The only way there is iteration over iteration and I think we have achieved that now. We have OEMs who sell arm-based solutions and have learned how to achieve cost savings with them.
DESIGN&ELEKTRONIK: If you say software base, today the server environment is clearly dominated by x86, so Intel, how long do you think it will take you to grow from today's estimated one percent market share to ten percent?
Mike Muller: I think that can happen very quickly, within two to three years.
DESIGN&ELEKTRONIK: On your Neoverse roadmap, you announced a 30 percent increase in computing power per product generation by 2021, five to ten percent more than for Cortex consumer products. If the microarchitecture will be identical, where do you want to get the additional five to ten percent increase? And what will be the general differences between Cortex and Neoverse CPUs?
Mike Muller: If you have an additional transistor budget, you can of course invest it differently, but you can also invest it in the core microarchitecture itself, but also in the components surrounding it, such as the size of the jump target buffer for dynamic jump prediction. Similar microarchitectures are not called identical implementations. In a server rack, you don't have the same limits on power consumption as in a smartphone. Other differences are the cache subsystem, the chassis, and the 3D memory that you can put into the chassis for the same reason.
"At the Leading Edge, mobile chips come before server chips. «
DESIGN&ELEKTRONIK: Are there also differences in chip production?
Mike Muller: That's an interesting question. Because today "premium chip manufacturing" takes place for mobile phones, which means that at the leading edge mobile chips come before server chips. Another story, of course, are the memories, where expensive technologies such as 3D stacking of chips are generally not used in the consumer market and are therefore driven by the server side.
DESIGN&ELEKTRONIK: In the automotive sector, most chip manufacturers have already licensed their IP, but their market share of around ten percent is clear. Do you therefore only have to wait until the cars become smarter and smarter in order to automatically grow with them, or are there other activities to develop further automotive-specific products? Which ones?
Mike Muller: Automotive is a conservative market with different dynamics in the individual segments. poor has so far primarily been found in microcontrollers, for example for window regulators or brake systems. This will continue, but segments such as infotainment, ADAS and autonomous driving will grow considerably faster. We have to address these special safety and security requirements, as we recently announced with our Lockstep solution.
DESIGN&ELEKTRONIK: Do I understand correctly that you are adapting products originally developed for the consumer market to automotive-specific requirements?
Mike Muller: Of course we want to reuse so many of the fundamental innovations that were developed for the consumer market, CPU, GPU and machine learning accelerators. However, the products delivered to Automotive are something completely different at heart because the basic technology is different and therefore there is a different roadmap.
DESIGN&ELEKTRONIK: In the area of embedded intelligence or IoT, the number of applications and chips sold is constantly growing, which can also be seen from the fact that almost 60 percent of the implemented CPUs are now Cortex-M. The number of applications and chips sold in the area of embedded intelligence and IoT is also increasing. On the other hand, the average microcontroller costs less than a dollar. Do you hope that the "old" Cortex generation M0/M0+, M3 and M4 will be replaced relatively quickly by the higher-quality M23 and M33 and that more money will be flushed into your cash registers? And also your market share with about 30 percent still offers growth potential....
Mike Muller: A large part of the microcontroller market is dominated by large volumes of old products, which we will supply for many more years to come. If an old product is good enough for the end product, it is good enough. I think the whole coming connectivity is driving up complexity and demand for more powerful controllers. However, the volumes of the old products are so high that the number of new chips is still very clear in comparison. Of course, we are talking about ten time frames in the embedded market, and once this has expired, there will also be an update for the controller. But that won't happen overnight.
DESIGN&ELEKTRONIK: Is the conclusion that you want to grow primarily by the amount of chips delivered rather than by more expensive IP per chip?
Mike Muller: One approach is that we want to generate growth in market share, especially for the expensive and complex controllers. On the other hand, cloud services, the Pelium-IoT platform, device management or data management offer us new sources of income beyond the traditional IP and license fee business.
DESIGN&ELEKTRONIK: Artificial intelligence and machine learning have the potential to penetrate almost all markets you address. So far you have announced a special ML processor in the so-called "Trillium" project. In which specific applications do you see this IP in addition to smartphones in the future?
Mike Muller: Our first target market is clearly the premium smartphone. However, a lot of IoT devices will use cameras and image recognition, for which our ML processor is also very well suited.
DESIGN&ELEKTRONIK: Your CEO Simon Segars has painted a shocking picture of the security industry for me. While from my point of view only 70 percent say security is important or essential, 25 percent say security is only good if it doesn't cost anything and doesn't bring any additional complexity. How can you help to convert these doubters?
Mike Muller: We have to make security simple. At the moment, it's far too expensive to build a system securely. We want to embed security as part of the standard platform. If you roll it out, you already have basic security. If we raise the base to a higher level, we will also eliminate reservations that still exist today because of the necessary investments. With our Platform Security Architecture (PSA), we have done a lot of work for our customers in advance.
DESIGN&ELEKTRONIK: I say: Your basic security is not end-to-end security....
Mike Muller: That's true. In fact, the entire value chain from the customer to the distributor is still struggling when it comes to end-to-end solutions. However, the more bad things happen, the more sensitive the industry will react because it no longer has any choice. When people wake up, ask yourself what has already been done and they will find that we at arm are already providing the basic security on the hardware side. Without investments in the software stacks above, this alone won't help.
DESIGN&ELEKTRONIK: While we are on our way to the fifth wave of computing, do you remember your first contact with a computer? What was that and what did you do with it?
Mike Muller: I was lucky enough to have access to a dial-up modem and a terminal in the mid-1970s. There I could dial into a computer service and upload programming jobs. There I learned programming, which was very unusual at the time.
DESIGN&ELEKTRONIK: In which language did you program at that time?
Mike Muller: It was the interpreter language APL. One strength of APL is recursion, the calling of a function within itself. This was already included in the original implementation of APL, unlike most programming languages at that time like Cobol or Fortran.
DESIGN&ELEKTRONIK: Mike, thank you for your time!
Disclosure: DESIGN&ELEKTRONIK travelled to San Jose by invitation from Arm, the costs were paid entirely by Arm. Our reporting is not influenced by this and remains as usual neutral and critical. The article has been written independently and is not subject to any specifications by third parties.