MediaTek has been covering major ground in the smartphone arena, and traversing quite a few other segments like smart devices, IoT, auto, entertainment, and of course the one thing that sort of ties it all — AI. We sat down with Finbarr Moynihan, GM & VP of Global Marketing at MediaTek at the Mobile World Congress to discuss AI, the chipmaker’s approach in the mid-tiers, their approach with customised chipsets, and more. Here are some snippets from the conversation.
Any conversation around tech these days refers to AI in some form or the other. Everyone is talking about AI. And we know that MediaTek has a very strong play in the AI arena. So what exactly is your approach to AI as a company?
When you talk about AI, you kind of have to separate out the traditional AI, which we’ve been doing for years. We’ve been using AI techniques for computational photography, display management, etc, for years, right on on pretty much all of the platforms at different levels. But I think the inflection point now is talking about generative AI, and we’re just at the start of that wave. It sort of, to me at least, still feels like magical technology. So there’s something magical about it when you see it in action, and we have some demos here at MWC that can go into that. Obviously, our focus is on enabling AI. We talk about ubiquitous AI, right? So we see a world where this technology has to become ubiquitous right now. What does that mean? Well, it’s spread across multiple categories of devices, not just in smartphones, but in automotive, IoT, and computing platforms. I think you’re gonna see this technology deployed for different use cases as we go forward. Also, I think from my perspective, ubiquitous means ultimately, it has to be deployed across all tiers of devices as well. Right. We’re at an early stage right now. And it’s really still in the high-end of the portfolio for now, such as our Dimensity 9300 flagship for example. Flagship and premium platforms have the capability for accelerating and running generative AI on the device without any cloud interaction… without any network connection. So that’s probably what we’re gonna see as the focus for this year in terms of the deployment, because of course, it’s not just the silicon and silicon capability, but you need to have enough on-device memory to run these large language models and these applications on the device. So what are we seeing right now? I think we’re still focused on delivering the best possible acceleration on the silicon and the right SDK for our customers, the OEMs, third parties, developers, and the ecosystem to create innovative applications that can take advantage of this generative AI capability.
Some of the stuff we’re demonstrating includes productivity stuff, taking a lot of data and summarising it so you can feed in a whole bunch of documents, text, whatever, and get a nice quick summary. It makes you more productive in a work environment. So there are definitely some use cases there. Similarly, we’re seeing things like smart systems that have a better understanding of your data, your use cases, your applications on your device, and how they all work together. And then there’s a whole host of applications that I think are still more in the consumer entertainment, fun side of things — video generation, taking your image and turning it into different avatars etc. Maybe these will catch on and become a new sort of social norm that people will share with friends, etc. So hard to predict where it’s going to land, but I think we’re seeing a rich array of stuff that people are beginning to try and experiment with. I think we’ll be sitting here a year from now and have a better idea. We’ll be sitting here in two years, and we’ll have a better idea even still, of what’s going to be the long-lasting kind of use cases for this technology.
AI has a whole different side and that’s also been in the news. Is there anything that MediaTek is doing to prevent misuse of AI technology?
I think it’s an area we’re certainly looking at. There are some discussions in the industry around how you sign things, how you identify things that are Gen AI generated etc. So my expectation is that’s going to be a necessary part of how this ecosystem will develop and go forward. Everything is moving very fast right now. And so trying to keep up with everything is probably challenging, but I think that’s going to have to be a subject of discussion going forward.
Moving on to the more mainstream stuff, the Dimensity 7000 and 8000 series of chipsets have been extremely popular. So what has been your approach and strategy with these tiers?
I think the approach and the strategy for the 7000 series and the 8000 series are actually different.
Talking about the 8000 series, MediaTek created that tier as we really wanted to focus on developing a truly optimised new chipset for what we call premium, and some people might call it the sub-flagship tier. Where one is obviously not able to deliver everything that one can on the flagship tier because it gets too expensive, but you still want to deliver that super high-end experience and offer a flagship-grade experience. But doing that in a really integrated and optimised chipset, rather than counting on a two-year-old flagship offering trickling down. They don’t quite hit the same price performance power capability. So I think for us the 8000 series over the years has been quite successful with that approach. This is a segment in its own right, which needs truly optimised solutions. And building those solutions to give those high-end experiences. And there the balance is all with how far can you push the performance without blowing the cost too much. So that’s kind of the strategy for the 8000 series.
I think the 7000 series is different. And that’s really about hitting the more mainstream part of the market. So again, you’ve got to be much more cost-conscious, but at the same time recognise that. You see this in India as well, as I’m sure you know, the consumer expectations of device capabilities are continuing to go up and the average selling prices of devices are continuing to increase. People are willing to spend a bit more for 5G, for more computing power, for better cameras etc. So the 7000 series is about really hitting that sweet spot of the mainstream market — the higher end of the mainstream market and delivering what MediaTek has traditionally done well for so long anyway. Getting that cost performance balance just right for the average consumer. And of course, we’d have entry products below that where cost is the driver. But yeah, we’ve been pretty happy with the success, with the boom of 5G and markets like India and Southeast Asia. I think the 7000 and the 6000 series will probably be the workhorses driving a lot of the volume for the next year or two there. Of course, we look forward to high-end devices selling as well, but I think the 7000 series is going to be quite important.
Speaking of the 7000 series, we know that a certain upcoming phone is currently in the news for its chipset. How do you cater to the demands of OEMs who approach you wanting to use something unique and special for the devices?
Well, a couple of years ago, we launched DORA — Dimensity Open Resource Architecture. At its most basic level, DORA was really about opening up the platform more in a more standardised fashion. So if you went back four or five years, the MediaTek software platform was probably a lot more closed in certain aspects and there was less opportunity for the OEMs to add differentiation and their own capabilities, particularly as we moved into the flagship and the premium stuff. And there was a need to architect the software solution to allow the OEMs to get in and differentiate where they wanted to differentiate. So some people want to differentiate on cameras, some people on power, some people on gaming, etc. So we’ve sort of built that flexibility into the platform architecture at the hardware and particularly at the software level. That allows OEMs to really kind of get in and then work either on their own or in close collaboration with us to find those areas where they want to differentiate and then go right ahead and do that.
