Dare to be Different

Differentation

Almost everything has to be differentiated from alternatives in some way. There are some things that are not, and we have a special word for that, "commodity." For some reason, the archetype of this is a trainload of pork bellies, because the very idea of a futures market in pork bellies seems like something out of a comedy program—and in fact, it is no more, since the futures market closed in July 2011. So let's use wheat.

If you purchase a trainload of wheat at take delivery, you get generic wheat from whichever farms were selling wheat at the time. Of course, if you want special organic wheat from a local farmer whose name you know, to make the special bread in your restaurant, then you can buy that, too. But that is not commodity wheat, that is differentiated by being organic and being grown by Peter on a farm 10 miles from the restaurant. In technology, we have a few types of component that are commodities, DRAM chips and DIMMs for example, but generally components are differentiated.

This applies up at the system level, too. Any electronic system, such as a smartphone or an electric car, is differentiated in many ways. At the high end of smartphones, Apple, Samsung, and Huawei build the market-leading phones. The middle market is now occupied mostly by Chinese companies that you've barely heard of, like Oppo, Vivo, Giomee, and TCL. Yes, LG is in there somewhere, too. At the top end, the companies distinguish themselves through features. At the low end, they are closer to commodities with similar screen size, all running the same Android, and not even that different in price. Inside the case they all use the same Mediatek (or maybe some Qualcomm) chips.

At the high end, to justify their high price, the manufacturers have to provide something that those almost-commodity phones cannot. Apple is the most differentiated, not just by having an aspirational brand-name with its own fan-club, but by developing its own software (iOS) and its own application processors (A9, A10...and presumably A11 in the next phone). Both Samsung and Huawei (HiSilicon) build their own chips, and customize Android to some extent. If they didn't, their phones would appear very similar to those Chinese mid-market smartphones but with an outrageous price tag.

If you want to make a profit, you have to have differentiation, since otherwise the only dimension left for competition is price. There are other aspects, such as distribution. Apple stores have the highest revenue per square foot of any retail (including places like Tiffany), so it is hard to remember that Apple's decision to open them (and so add another type of differentiation) was almost universally ridiculed at the time. I don't know much about mid-range smartphones in China and how (if at all) Oppo and Vivo and Giomee try and differentiate, or whether it is mostly price. On the other end of the scale are auto dealerships. I was recently at an automotive conference in Germany and the car manufacturers, such as BMW, Audi, and Mercedes, all realize that one of their big weaknesses is that their customers hate setting foot in their dealerships. That's not the kind of differentiation they want.

Nobody Wants to Design an SoC

Most people reading Breakfast Bytes probably work in the semiconductor ecosystem somewhere. (Although other people do read it occasionally. Hi Dad!) It is easy to assume that everyone "wants" to design a chip. After all, building a state-of-the-art SoC is the peak of high tech, what could be better than that. But the reality is that building a chip is about the most expensive differentiation you can think of. In fact, people will go a long way to avoid designing a chip. I remember in the late 1990s, pre-iPod, the company that led the market for mp3 players was Rio. Their player was built using an FPGA, and in low volumes that was a great choice. But Rio was way more successful than anticipated and the unit cost of an FPGA, compared to an equivalent SoC, was very high. I talked to one of their executives and he told me that they probably were not going to do it. They only had one engineering team, and while they had the money to hire more people, building a second team from a standing start is not easy. So the one team could cost-reduce the Rio 1. Or they could do a quick design using FPGAs again, and get the Rio 2 out around the time they expected the competition would have something similar to the Rio 1. Designing a chip is too hard, too expensive, and too risky.

Apple, in particular, faced a huge uphill challenge to build their own application processor: they didn't have a chip design team in house at all. They had to hire a lot of people, buy a couple of companies for their teams and expertise, and build it up from nothing. Now that's an expensive way to get differentiation. For sure, if they thought they could just buy chips from Qualcomm and write software, then that's what they would have done—after all, that's what they do on the Mac with chips from Intel. But they obviously decided that to deliver a unique user experience, they had to control everything from the chip, to the software, to other aspects (like the Gorilla glass, the stores, the advertising, even the white headphones).

Most IoT devices are likely to be built using standard chips that contain microcontrollers, memory, and radio interfaces. The market is so fragmented that it doesn't even make sense for a market leader like FitBit to design their own chip. The biggest challenge in IoT, at least consumer IoT, is knowing what consumers will consume. Cost-reducing the design, like the Rio I mentioned above, is only a problem that comes with success. Probably, too, the next issue is the same one that the Rio team faced, whether cost-reducing the design is more important than getting the next version out fast.

Automotive Differentation

Car manufacturers face a similar challenge. Their differentiation today is around things like brand, but largely around their competence at building engines. This ranges from the lower end of the market, where people want an engine with good fuel economy that will last for a couple of hundred thousand miles, to the high end where people like BMW and Porsche (let alone Ferrari and co) claim the high ground of the best engines of all. As we move forward into the electric era, both in the sense of electric traction and in the sense of ADAS/autonomy, they will face similar decisions to Apple in smartphones. Can they maintain a premium brand if they just use purchased electronic components that are available to all their competitors, too, and just use software to differentiate? In fact, software is not the expertise of car companies anyway, so maybe they should buy that in, too. Perhaps Google, NVIDIA, or someone will produce a sort of Android-style car software that all the car companies can just use. But you can immediately see the problem. If VW and BMW, and Porsche (to keep the German thing going) all have electric vehicles with the same electronics and same software, and differ only in the shape of the bodywork, can they maintain their position in the marketplace? After all, it is hard to be the ultimate driving machine when Volkswagen is using the same electronic systems.

I think that this goes to the heart of System Design Enablement. At the top level, it is about software, electronic hardware, and IP (both software IP, standard components, and chip-level IP). Car companies (and this applies to every other business, too) need to decide in which of those dimensions they need to do their own thing to get differentiation. Some decisions are easy, based on competence. I'll be extremely surprised if any car manufacturer builds its own 5G modems because it isn't competent to do that. Even smartphone companies buy them from specialists. On the other hand, they can't buy everything, they have to decide on what aspect of the user experience they want to compete (I'm leaving out things like competing on dealerships or advertising, not because they are not important, but they are out of scope).

Car companies know that software will be a large part of their differentiation going forward. Some of that will run in the cloud. But on the vehicle, it is not just software, it is electronic systems (with a software component). You can't just put a few PCs in the trunk and write software, or rely on the cloud to tell you whether to stop when someone steps into the road in front of the vehicle. So differentiation is more than software, it is optimizing the whole electronic system. That means everything from the overall system architecture (big central processor? processors at each camera?), networks (harness? ethernet? fiber?), down to boards, packages (more-than-Moore is more and more the way things will be done), chips. Power, in particular, is a huge driver (just as it is for smartphones). If you put a camera and an associated processor behind the rear-view mirror, in the sun, inside the cabin, then it had better not be dissipating the same power as a server.

Make Versus Buy

At some level, a lot of the detailed decisions sound like "make versus buy", as if it is driven by the cost of development versus the purchase cost. But really, it is to decide in which dimensions to differentiate and compete, and in which dimensions you can purchase commodity components. The answer is not going to be to buy all of them, except, perhaps, at the very low end of the market where price is the main dimension of competition. This applies quite generally, to mobile, to automotive, to medical, to servers and routers.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.