Four More Waves: 5G, Cars, Clouds, IoT

Earlier in the week, I did a sort of bait and switch, introducing the five waves and then talking about just one of them. In that post, The Five Waves: AI, 5G, Cars, Clouds, IoT, I just covered the first wave of AI and machine learning.

Today, let's take a look at the other four waves.

5G Mobile

5G is coming. It will arrive more slowly than the hype might make you think, but buildout will be taking place at full speed this year and next. Nobody wants to pay for a 5G handset if there is not a network of 5G basestations. I think that unlike with 4G and LTE (see my post 3G and 4G: The Internet Arrives), the expectations for how different 5G will be for the normal user have been exaggerated. The biggest difference may be that the network capacity is much greater and so it will be possible to get on the net or make a call in environments where it is not possible today—for example, at a sports game or at CES. Although 5G boosters talk about the increased speed to download a movie, it is actually rare for anyone to download a movie — we watch a movie, and LTE is already fast enough for that. So while it is true that downloading a movie will be faster, that's of little practical interest to the normal user.

Different generations of mobile have been associated with doing different things: 1G: making calls; 2G: making calls and texting; 3G: internet; 4G: video. For 5G, apart from capacity, the most important thing is likely to be connecting things other than smartphones, in particular cars. There are other things that could be connected, such as smart security cameras, industrial robots, or telemedicine. But I think these devices are much more likely to either be connected by wire or through a WiFi router of some sort. Only when you get to smart agriculture is a cellular network likely to be the connection of choice—but 5G is going to arrive most slowly in rural areas, and mmWave will never arrive.

So right now all the chips for 5G basestations and handsets are being created. On top of that will be IoT devices that require chips that leverage 5G connectivity. So, like the AI paragraph above, this is an area where there are lots of designs but production volumes will mostly come later. For example, in the US market, the Consumer Electronics Association expects 20M 5G handsets to be sold this year, but 146M non-5G handsets. But in 2021, that 20M is forecast to triple to over 60M handsets.

One trend that is now apparent in other markets was started in mobile: handset companies designing their own application processors (APs). In fact, all the market leaders in the smartphone market design their own APs. They don't design every chip in the phone, of course, but they design the big one that sits in the center of everything else and is most responsible for the users' experience.

Hyperscale Infrastructure

Cloud data centers, servers, high-performance networking, and storage are the components of hyperscale infrastructure. As a market, this is usually referred to as HPC for high-performance computing. I think that there are two interesting trends in this area. The first is that cloud data centers continue to be built out aggressively. Right now, with so many of us working from home, demand seems to be stronger than before. But data centers are a long-lead-time project, like a semiconductor fab. It is hard to respond to an unexpected short-term increase in demand.

The second trend is for cloud data center providers to design their own chips for some of what they offer. For what AWS is up to, see my posts The AWS Nitro Project and Xcelium Is 50% Faster on AWS's New Arm Server Chip. For some background on Google's TPU, see my posts Inside Google's TPU and Google TPU Software.

Automotive

There are two trends in automotive that are driving the semiconductor business. The first is the increase in semiconductor content in vehicles. I've seen outlooks forecasting that 50% of the cost of a vehicle will be electronics in a few year's time (that includes more than just the semiconductors, of course). The growth of automated driver assistance systems (ADAS) and the gradual transition towards more autonomous operation increase the amount of silicon required. It also moves it from analog design in mature processes towards digital design on leading-edge nodes. For more on this, see my post Automotive Industry Basics.

As with mobile and cloud data center, another trend is for the system companies to do their own chip design. In this market, "system companies" are the familiar automobile companies, knowns as OEMs in the terminology of the industry. One example is Tesla's Self-Driving Computer. For more details on that see my post Tesla Drives into Chip Design and The Tesla Full Self-Driving Computer.

Industrial

Industrial tends to be a catch-all name for all other markets. The internet of things (IoT) traditionally excludes mobile and automotive since they have their own category. Part of IoT is consumer, but today a lot of it is industrial. It turns out that the value proposition is much clearer for industrial IoT than for consumer IoT. It is easy to find articles on how many terabytes of data a modern aircraft generates on a flight, or how much data is required to do a good job of detecting failures ahead of time so that they can be addressed by preventative maintenance. There are also increasing numbers of systems using vision processing in industrial settings, examining fruit for ripeness or damage, or bolts for manufacturing issues. Robots all have semiconductors under the hood.

Having said that, the market is fragmented.

Summary

Let's let Lip-Bu summarize today's and yesterday's posts. Here's a recap of a couple of sentences of what he said on the call:

It's very unusual to have five major waves happening at the same time. You have the AI machine learning wave and you have 5G is starting to deploy and then you have...the really massively scaled infrastructure. And then we have autonomous driving and then the whole digital transformation of the industry group.

So there you are: the Five Waves.

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