Due to my background in the semiconductor space, I engage in a lot of financial analyst conversations around semiconductor component companies. I used to write a specific note we called “the Circuit” exclusively for hedge funds that dove into the opportunities and challenges for some of the largest semiconductor companies in the world. I would write extensively about Qualcomm, Intel, Samsung semi, Marvell, Xilinx, NXP, Broadcom, ARM, etc., and it was interesting to track the influence of Apple on many of these companies through the years. Nearly every note I wrote for investors in these businesses had Apple mentioned somewhere, and Apple was a constant topic during calls I had with these investors as well.
Perhaps one of the biggest questions that still lingers is what Apple will design themselves vs. what they will still buy from a third party. On the heels of Apple’s worldwide developer conference, I thought I would share some key fundamentals to bear in mind regarding their silicon design strategy.
Apple’s Designers are Among the Best In the World
Most casual observers of Apple will have a hard time understanding this point in its entirety. Yes, we all know Apple’s designers are good and make chips that help the iPhone stand out, but I’m not sure many realize that Apple’s internal design team is among the best in silicon architecture in the world. There are few companies that have amassed a group this talented. Intel, Qualcomm, Samsung, and ARM all also have great designers, but as I’ll point out in a minute, there is a significant advantage Apple’s design team has over all those other companies.
Not only does Apple have many of the best silicon designers in the world, but Apple is a hot place to work if you are a great silicon designer working at another company. Perhaps this is evidenced most recently by Apple’s hiring of Qualcomm’s lead engineer of their communications business. What he will do is unclear at this time, but many have speculated the move means Apple is working on their own modem and baseband processor, which is something I have known for some time.
Furthermore, I’ve heard from many friends who work in many other semiconductor companies that recruiters from Apple have been hot in pursuit after top talent working at their respective companies. All proving that not only is Apple a hot place to work if you design silicon architectures, but that Apple is also committed to continuing to grow this team with the best talent they can assemble. Control the silicon and control the future. That would be my internal slogan for this group.
Apple’s Unique Advantage
If you find the time for a little retrospective, take the time to read one of the first posts I ever wrote for Tech.pinions right as we were starting it. It is simply called Why Apple Has a Strong Competitive Advantage (Published June 2, 2011). If I were re-writing that piece today, a fundamental pillar of Apple’s advantage would now include their silicon design team and strategy.
Apple, unlike all the other semiconductor companies making CPUs, GPUs, modems, image sensors, etc., only has to build and design a piece of silicon for their dedicated needs. This single truth will always give them an edge on their competition. Perhaps this point is made most clear in many benchmark results that highlight Apple’s A10 processor, which is nearly a year old, still being faster in key metrics than Qualcomm’s latest release the Snapdragon 835. Now to be fair, Apple has more space on the chip to work with since they do not put a world class modem on the same silicon die. Where Qualcomm fits the CPU, GPU, memory, and modem in a package smaller than Apple’s A10 with just the GPU, CPU, and memory (cache). So while it is a relevant comparison, it is also a bit unfair, but perhaps that is the point I’m making.
Apple does not have to design their chips for anyone else but themselves. A luxury Qualcomm and Intel do not have since they need to try to design an architecture to cover as many possible scenarios and customer needs as possible. Because of this, performance will always lack in key user experience ways to Apple’s hardware. Theoretically, Samsung could also be in a place to benefit from a similar advantage. However, their designs are still not as good as Qualcomm’s in many ways, so it makes sense for them to use Qualcomm in their high-end devices still. For now, at least.
Apple has the unique advantage to design silicon for specific use cases. They can have a vision in mind and not have to rely on anyone else to deliver that vision or core experience and simply design all the underlying silicon chipsets they need to achieve their vision. This is unprecedented for a consumer electronics company and will continue to yield significant advantage for Apple in core hardware experiences. Apple will simply be able to do many things others can not, or at the very least do it years before others can, because of this advantage.
This is Actually a Software Advantage
When you truly dig into it, you realize this appears to be a hardware advantage, but it is a software advantage. Not just because Apple can customize silicon to core iOS experiences, but they can also extend those silicon advantages in APIs and SDKs for developers to take unique advantage of as well. No other company making semiconductors controls the underlying software, and therefore they have to create their silicon designs for a massive amount of software flexibility and variables. Apple has such tight control over both that they can fine tune every chipset they design specifically for the application.
This point is why it was always inevitable Apple would start designing a dedicated chipset for artificial intelligence and machine learning. One of the biggest semiconductor trends at the moment is dedicated chips, sometimes a custom ASIC-like what Google did with Tensorflow, or by using custom FPGAs to offload the inference query in a request for information from the AI algorithm. Building a dedicated AI chipset fits with both Apple’s strength and their unique advantage.
Qualcomm has been doing some dedicated machine learning and AI chips on their architectures. The challenge is they don’t control the software so they have to make a design that works for Google, Microsoft, Samsung, etc., where Apple can design the chip just for Siri and the underlying machine learning techniques of iOS.
Being truly integrated with chipset design, software (APIs, SDK), and even tying future services ambitions to their silicon designs has now become the ultimate vertically integrated advantage that no other company can match today.
Design or Buy?
All of this leaves us with our initial question. What does Apple design themselves and what do they buy? My thesis here is that Apple looks at this two ways. The first is what silicon architectures are core to the iOS, and maybe someday Mac, experience. The CPU was obvious as well as the GPU which it now seems they are going to take more control over. The camera sensor, which they design was obvious. The display controller which they now design also seemed obvious. The power management controller (PMIC) could be an obvious one, and the modem as well comes to mind. All these things have a direct impact on the overall experience of a product. Things like RF, memory, etc., are crucial but don’t have as much objective impact on a computer. Apple will design the parts that are essential differentiators and yield objectively better experiences than if they used off the shelf parts.
The other area to design is when it comes to things that can not yet be done. Apple may have an idea for an experience and realize that no one else out there has a solution that will work. In this case, just design it! The S1 board in the Apple Watch and the W1 audio solution in Airpods are two great examples of Apple having a product vision, realizing no third party component solution could match that vision, so they just designed it themselves. This is a luxury not many companies have.
There is one variable I have not yet mentioned that I am watching very closely. This variable is the foundry that Apple uses to manufacture their silicon wafers. As companies like Intel, Samsung, TSMC, and Global Foundries, continue to push the limits of Moore’s Law they are doing so on their proprietary process technology. When Intel says they are making 14nm designs, they are doing so on a transistor design that is quite different than Samsung, TSMC, and Global Foundries transistor designs. Some of these transistor designs are much better than others, and some of them may push Moore’s Law even farther and more efficiently than others. I’m not saying that Apple’s current foundry (TSMC) does not design an excellent 14nm process, but that I do believe others are better. While Apple is unlikely to do anything more than print their designs on the process of one of these foundries, I am curious to see if they invest, or joint-venture with any of them to have more influence over how the process design goes forward and in ways that meet their particular needs. The foundry process they use is one of those variables they do not control but can have measurable benefits or consequences on the end product experience.