MIPS: RISC-V IP provider
"The best time has just begun"
MIPS Technologies has been through some turbulent times. In 2021, "MIPS Technologies" became "MIPS" and the focus shifted to RISC-V IP cores. In an interview with Markt & Technik, Sameer Wasson, CEO of MIPS, is convinced that the company is on a path of success.
Markt & Technik: MIPS is a well-known supplier of processor cores. And I think its own IP cores used to be really good, but the company still failed, even though there was only one serious competitor at the time: ARM. Today, MIPS relies on RISC-V, which also increases the number of competitors. Why do you think MIPS has a better chance today?
Sameer Wasson: I think that MIPS Technologies overlooked something important that many other companies have done: They didn't focus on what they do best. They played Arm’s game vs. focusing on what made them unique. I firmly believe that people and companies should do what they do best. And if they do that, with the right conviction and consistency in terms of direction and priorities, then success usually follows.
In my view, the MIPS cores of the time were data processing machines, whereas the Arm cores were developed and optimized for use in mobile phones. MIPS Technologies repeated the same and tried to become like Arm. In other words, the company tried to develop IP cores that were geared towards mobile phones or application/control processing rather than focusing on the network and data processing markets. MIPS Technologies' decision is easy to understand: at the time, the TAM for mobile phones was much larger than that for data processing. But it's always difficult when a company tries to push developments that are not really in their core competences. They were not focusing on what customers needed them for.
Maybe MIPS Technologies just couldn't sell the unique selling points of its IP cores?
I wasn't on board at the time, so I can't judge the individual features of the IP cores, but looking at the story from an outsider's perspective, I would say that the cores of that time were particularly well suited for data processing in general. An example that might make this point clear: the MediaTek modem has been based on MIPS cores, which means that even in the cell phone sector, the data processing part was MIPS-based.
If a company can do everything, it doesn't need to focus, but who can? So I think a company should never make the mistake of deviating from its core competencies and chasing after its competitor. Then you end up in the middle of nowhere.
What's more, I'm convinced that if you leave the decision to our customers' developers, they will choose the right technology for the individual problems.
What does that mean for the new MIPS?
That we make sure that we concentrate on what we can do, namely developing the best data processing machine in the field of modern data processing. And it's with this approach that we can compete with the other RISC-V IP core providers.
Of course, it's always tempting for a company to look at the market for server CPUs in data centers: if you could achieve just 5 percent of the Xeon market, for example, that would be a huge volume of sales. I can understand that, any company can try to do that. MIPS, on the other hand, concentrates on data processing problems and we will continue to implement this focus in the interests of our customers. Fortunately, unlike last time when it was just Arm and MIPS Technologies, the TAM for data movement and data processing has grown significantly.
If you used to look at the networking area, there were a few NIC (network interface card) cards and a few other things, but most of the growth in compute was in the application processing market. That situation has changed, especially when you go deeper into data areas and into the automotive industry, where there's a lot of data movement. And I think that these application areas are very, very interesting.
And that is exactly what we are focusing on. MIPS has a knowledge advantage here and we are well positioned. MIPS and Arm both have applications in one data center. Arm concentrates on the main CPU, we concentrate on the DPU (Data Processing Unit), on near memory compute and other places where we see heavy data movement and data processing needs.
I say quite deliberately: we can't do everything, but what MIPS can do, it can do better than anyone else. MIPS is not starting from scratch, we have a long history in this area and we have put together a team with which we can implement this. And you shouldn't forget: We already have a footprint in this area. We are already working with a very large hyperscaler that has been using our technology for years. And we will now expand this area. And another point: Arm is less of a direct competitor in this area, but rather proprietary cores and some DSPs or very special processors.
Sounds like an interesting approach, but the market is not doing well at the moment, not even for IP providers. How do you see the situation for MIPS?
I wish it was that easy to look at a financial report and say that things are going well or not. I think you have to look at several things today. For example, I'm convinced that the impact of COVID on the semiconductor industry is still and will continue to be felt for a number of years.
COVID has changed the way people think. With COVID, the regional aspect of supply chains became increasingly important. Accordingly, a lot of capital is being invested in the semiconductor industry. So large semiconductor manufacturers operating in multiple regions of the world will inevitably find that competition from local suppliers will have an impact. China is not the only example of this development, but it is the most prominent, partly because the geopolitical aspect is the most discussed. But I also expect similar developments in Europe, although it is already happening here. In India, this development will happen on a smaller scale over the next decade.
My point is this: I don't think the market has changed, but that there are no longer horizontal supply chains around the world, but more vertical regional approaches, and that leads to changed results.
A second point: the number of OEMs interested in developing custom solutions is increasing. It used to be limited to a handful of Smartphone & consumer products, however its now expanding to more Automotive, Data Centers and Industrial markets. I see overall TAM getting bigger for custom silicon which again is a huge opportunity for IP vendors. OEMs want more customization because of the rate of innovation is accelerating in this AI era. The impact to them falling behind on AI is larger in the long run vs. expanding investments in custom silicon. In most cases the traditional processor vendors are not able to keep up as they also have a legacy business. This is a classic innovation dilemma, which prompts OEMs to simply say: I'll do it myself, then I can also influence it according to my own needs.
Looking at these three points, I would say it's the most exciting time to be an IP provider. Because all these changes are making IP more and more valuable. They are being converted into chiplets, they are moving up the value chain, and instead of receiving 1 to 4 percent in royalties, 8 to 15 percent royalties may be possible in the future.
Our IP cores are also in greater demand because most people no longer want a soft core, they want a processor core that delivers exactly the application-specific computing power they need, plus flexibility in terms of IO, analog technology and memory - and that's exactly what we do.
So I don't see any problems for MIPS at the moment, quite the opposite. We have more customers than ever with whom we are working directly, we have more system houses who want to influence development directly. Arm is Arm, but from my own experience I know that the IP cores for the semiconductor manufacturers were mostly supplied in the form of black boxes around which development was carried out. With RISC-V, the semiconductor manufacturer can change the design and by providing customers with an interface for this, previous boundaries are being overcome or have already been overcome.
MIPS focuses on application-specific IP cores, but does that also mean that the company has to concentrate on certain areas?
Yes, of course, this is also reflected in our management team. Let me explain it with an example: Electrification in the automotive sector. MIPS has a great deal of expertise when it comes to moving data from the analog interface to the controller in a deterministic way. We can solve this problem excellently with a customized RISC-V core.
Of course you're right: MIPS can't handle all applications hence we will focus on four or five applications that we can serve better than other IP providers with our special know-how.
The personnel structure at MIPS is structured accordingly: We have IP specialists who know how CPUs are developed. In addition, we have brought people into the company who are familiar with SoCs, i.e. who know how an SoC has to be constructed and for this they also need to understand the overall system. Accordingly, we have also hired people who come from OEMs and have the relevant application knowledge - all of this together enables us to develop application-specific cores.
To date, MIPS has offered two different IP processor families. In your opinion, what are the differentiating features?
As previously mentioned: MIPS specializes in data processing and data movement. This means that, for example, our multi-threading architecture with hardware virtualization and with cache coherence and the ability to integrate heterogeneous elements are very fundamental differentiators that we build on.
Are there already plans for a new family?
I don't want to be too specific, but one thing is clear: we will clearly continue to drive application-oriented developments and I expect we will make an announcement this year.
If you look a little further into the future, what do you expect for MIPS?
Our goal is to make MIPS a leader in Application Specific Compute with an emphasis on Data Processing. I am convinced in the near future we will be talking about DPUs, i.e. Dataflow Processing Units. DPUs will become the most important component alongside CPUs and GPUs because they solve the problem of data movement.
There is a simple financial reason for this: the more GPU/CPU capacity a hyperscaler has, the more virtual machines it can run and the more capacity it can sell to customers. For these companies, it is therefore primarily a question of shifting as many tasks as possible from the GPUs/CPUs to other units, and DPUs are an important way of doing this.
