Semiconductor 2018: Up and to the Right...But Memory Way Up

I was at a meeting of the ESD Alliance (probably still called EDAC back then) when someone memorably described the EDA industry as being like a water skier being towed behind the semiconductor industry speedboat. As long as the speedboat was going fast enough, skiing was good. If it slowed too much, then EDA would sink into the water.

Actually, outside of really big technology downturns like 2001 or 2007, EDA is fairly resilient. Semiconductor companies face some hard decisions, especially in the past when Moore's Law was in full swing. Even though the current design might not be selling well, the next design has to be done in any case, so R&D cannot be cut back much. The chart above shows in grey semiconductor R&D, in blue, EDA revenue, and the green line shows percentage of semiconductor R&D that went to the EDA industry. Two takeaways: there was a huge technology recession in 2001 and semiconductor R&D went "meh". Even a big global recession in 2008 barely caused a cut back. And throughout the entire period, EDA has been a pretty constant fraction of semiconductor R&D.

So how is the speedboat doing?

IC Insights

Well, the year isn't over but IC Insights updates their McClean Report towards the end of each year, which forecasts the final numbers. I'm just going to ignore the fact that these numbers are a mixture of actuals and a forecast for the last part of this year.

Note that you can't quite use this list for market share rankings since it includes both foundries and fabless companies, so there is some double counting going on (all of TSMC's revenue, in particular, shows up in other companies, and a little of Samsung's). Of course, it is margined up in those other companies. For example, Qualcomm and Broadcom's revenues are more than TSMC's despite TSMC having lots more customers than those two.

This table doesn't make a lot of sense without noting which companies are primarily memory. Another wrinkle is that during the year, Toshiba split its memory company off in a complicated transaction (where it bought some of it back again). For now, that company is simply called Toshiba Memory Corporation but it will have a new name sometime next year. But in this table, there is just one Toshiba covering all their semiconductor businesses. Anyway, the companies with primary memory revenue are Samsung, SK Hynix, Micron and Western Digital/SanDisk (a merger that happened in 2016).

The good news is that the top 15 companies grew an average of 18% between 2017 and 2018. However, the four big memory companies all grew at over 20%, with Micron over 30% and SK Hynix over 40%. Since Samsung and Western Digital/SanDisk both have other businesses than memory, I'm assuming that their memory businesses presumably grew faster than their overall numbers in the above table. Memory, for example, represented 84% of Samsung's business, and their memory business only grew at over 30% (non-memory, not close, at 6%).

The big star is NVIDIA, which grew 37%. I think this comes mostly from the adoption of GPUs in datacenters for neural network training, although their gaming business is strong and they are one of the leaders in automotive (although that has to be a market more for the future).

One challenged company is Qualcomm, whose business shrank. Only by 3% but when the overall market is growing by 18%, that's like shrinking by around 20% compared to competitors. I would say that Qualcomm's big challenges are their business in China, and the fact that a big focus in R&D has to be 5G but 5G is not really shipping yet. I will be going to Mobile World Congress in March, which I fully expect to be the 5G show (with added AI, since everything has added AI these days).

Intel might look challenged too, in the sense that Samsung passed it for the number #1 spot, but really this is just a symptom of Intel not having a big memory business. Intel grew at a respectable 14%, over twice as fast as Samsung's non-memory business did. They have 3DXpoint (jointly developed with Micron, although the partnership is going to be dissolved) which Intel calls Optane. Its big niche is (potentially) a new non-volatile level in the memory/storage architecture, especially in Intel-based servers, but that market has been slow to materialize. I understand there have been problems ramping up the 3D nature, while 3D NAND is up to 64 and 96 layers, which has tremendous economies of scale. Intel does have a small NAND business based out of their Dalian fab in China as well. It is probably growing fast like all the other memory on the planet, but is too small to move the needle on Intel's big overall revenue.

NXP also grew only 1%, which counts as shrinking in a market that grew 18%. Of course, they went through a lot of disruption when Qualcomm attempted to acquire them but was blocked by the Chinese. And while I make no claim to understanding the nuance of international politics, I don't understand how the US Government would let ZTE off the hook without making approval of the NXP acquisition part of the deal, whether officially or unofficially, at around the same time.

Staying with Europe, another surprise, to me at least, is ST which grew at 16%, the fastest growth of any non-memory company except NVIDIA. I don't really have a good feel for what ST's product line is driven by these days. They describe themselves as having one of the semiconductor industry's broadest product portfolios, which isn't always a good thing, but you can't argue with their results. They are big in analog (but I think TI is bigger). They are big in automotive (but NXP is #1). They have a fab in Crolles (near Grenoble) but I think that it is limited to 28nm FD-SOI (which was developed by ST and licensed to both GF and Samsung). They are trying to get money from the French and Italian governments to build two new 300mm fabs (surprise, one in France and one in Italy). They actually built the shell of a 300mm fab in Sicily, but then never moved equipment in.

The actual #16 player, so #15 if you eliminate TSMC and just look at dollars pulled from end-markets, is Mediatek (also fabless).

I'm not sure there are any big surprises in these numbers. Everyone knows the memory market is on fire, but the more general market has also grown quite respectably. Early next year, everyone will have their predictions for growth in 2019. One expectation is that memory prices will weaken and the market (in dollar terms, not bits) will shrink.

More Information

The official title of the full report is November Update to the McClean Report—A Complete Analysis and Forecast of the Integrated Circuit Industry. There is a summary research bulletin, and lots of other reports on the IC Insights website.

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