I wrote recently about how Samsung's extraordinary 50+% year-on-year growth took them to #1. I also cautioned that a lot seemed to be due to strong prices in memory, DRAM in particular. I was reminded of this when I came across an IC Insights report published a few days ago, Are Major DRAM Suppliers Stunting Demand? The Rankings I looked around to find the full rankings. This table is from Gartner in early 2018. I hate this table, actually. Sometimes, tables of semiconductor rankings include the foundries, and sometimes they don't. If you include them, then there is some double counting going on, since a chip sold by, say, Qualcomm counts as Qualcomm revenue, but also as (smaller, presumably) revenue to the foundry that actually built the chip. So it makes some sense to leave them out, but then you really need to add Apple in, as is sometimes done. Here's another table. This is purchases of semiconductors in 2017. Samsung isn't just the biggest manufacturer of semiconductors, it also buys so much (presumably a lot from itself, but also from companies like Qualcomm and Broadcom) that it is the biggest purchaser. But Apple is #2, so if it was split into two companies and counted as a semiconductor manufacturer and an OEM, it would be the third largest semiconductor company at nearly $40B ($38,754B seems ludicrously precise for an estimated value, 0.003% accuracy). I have no idea if and where Apple's semiconductor purchases show up as sales in the original table. Anyway, leaving the foundries and Apple out of the listing of semiconductor companies seems like a big distortion if you are trying to see what's going on. Late News : On Monday, the administration blocked the proposed Broadcom acquisition of Qualcomm (perhaps with NXP inside). If you don't know what CFIUS is, you should read Breakfast Bytes more often! I covered it in The Four Ts: Trade, Tax, Talent, and Technology Funding last July. The reasoning is interesting and unusual: there is an expectation that Broadcom would cut costs to pay the debt and keep its stock price up, meaning it would hand 5G to Huawei (and presumably Nokia) on a plate. Too early to say any more. Thanks for the Memory If we go back to the original table, with the manufacturing rankings, Samsung went up 52%, but the number is only so low because they are in a lot of semiconductor businesses. SK Hynix went up an even more amazing 79%, and Micro 78.1%. Western Digital went up 120% (Western Digital acquired SanDisk, so they are in the memory business, too, but they are in the SDD business that uses a lot of memory, so I don't know how the accounting is done there). Anyway, it is obvious that a lot of the 22% increase in the semiconductor market last year was due to memory—none of the companies that don't have a memory business got anywhere close to that, more like 10%. There are two ways you can make more money in memory. You can manufacture more memory, which is known as "bit growth". Or you can make the same amount of memory, and sell it for higher prices (or both, of course). The table above shows up in the report I mentioned at the start, from IC Insights, subtitle Skyrocketing DRAM Prices Potentially Open the Door to Chinese Competitors . Anyway, let's not worry about the Chinese competitors yet. Because there aren't any yet. The first part of that subtitle is that DRAM prices are skyrocketing. Anything consumed in large amounts like DRAM tends to be purchased on long-term contracts, and the spot market covers everything else (look at oil or electricity). The spot market price can swing around wildly. But if the spot market price remains high for a long time, eventually those contracts have to be renegotiated, and then the higher prices flow through into the larger market. The table above is hard to read since it compares apples and the rate of change of oranges. The red shows the skyrocketing price per Gb. The blue shows the year-on-year bit volume growth. So growth used to be in the 40% but now it's down to 10-15%. IC Insights' conclusion is that the DRAM suppliers are maybe stunting demand (well, they put a question mark on the end, and Betteridge's Law says that the answer to any question that ends with a headline is no). Of course, if DRAM was cheaper, they'd maybe sell more bits (Ferrari is stunting its unit growth by not pricing like Honda, too). I don't think in a traditional cartel sense that the memory companies could be reducing capacity by keeping their fabs running below capacity, but they may well be out of capacity. I was at a SEMI presentation a couple of years ago and they were saying that prices had not cratered in DRAM "this time" since the memory companies were not rushing to build overcapacity. However, if they didn't build enough, it's not like you can just order up a new fab overnight. It costs a huge amount, but more importantly, it has a long leadtime measured in years (land acquisition maybe, build the shell, purchase the equipment, ramp to volume, etc). My guess is that the memory companies didn't overinvest, but maybe they should have invested more, but you can't change that decision just because the prices are high. Plus, it makes little sense to invest in a new fab at anything other than the leading-edge node, since it risks getting stranded when the next change occurs. The general rule in fab operations is that you have to make a "bet" on what capacity you will need, whether there will be an upturn in demand, or a downturn. Then you have to live with that bet even if you are building a fab in a recession (it will be ready for the upturn), or suddenly something like smartphones come almost out of nowhere (back in the late 2000s). However, there is another thing going on. Most demand for DRAM indeed comes from smartphones (more than PCs and servers). They ship in billions of units per year. But they used to grow at insanely high rates and in the last quarter, some analysts have unit growth negative. The above chart just shows a year or so worth of data, so that's not the same as the decline in smartphone growth, which is more like a five-year trend. But if smartphone growth is flat, why would you expect memory growth to be high. I don't think the amount of DRAM in a smartphone is going up, if anything there seems to be a small trend in the opposite direction. China Now let's worry about the Chinese competitors. Who are they? (My first attempt to find more detail went hilariously wrong since Google decided dram must be a typo and so it found me Chinese drama companies. China is apparently remaking the global film industry, and has a lot of video streaming companies almost none of which I've ever heard of. YouTube is not one of them, as I wrote about earlier in the week. The largest is iQiYi, which I assume is pronounced ee-chee-ee and is not designed to be a great international brand name). YMTC is building a 3D NAND flash fab at Wuhan Donghu New Technology Development Zone. XMC was planning on capitalizing on the disruption caused by 3D NAND flash (see my post Memory in China: XMC ), but now it is switching to NOR. I suspect that this was "encouraged" in some way by the Chinese government. The alternative is that they couldn't build 3D NAND from a standing start, which wouldn't bode well for YMTC, who is trying to do the same. Innotron is building a fab for DRAM focused on mobile with LPDDR4 (low power). If they can build it, since getting on for 50 percent of global smartphone shipments are Chinese companies (even more are assembled there, but the sourcing decisions are made in places like Cupertino), this could be a game changer. Chinese companies are strongly encouraged to use Chinese sources, and this would take away a big chunk of the entire DRAM market from the current incumbents. JHICC is building a fab for DRAM for consumer electronics in Fujian. I think this would require selling outside China since, outside of mobile, I doubt there is enough demand. But that would require it gets patent licenses, too. Most of this info on Chinese memory fabs comes from the press release China Develops Its Domestic Memory Chips Fabrication; JHICC, Innotron Memory and Tsinghua Unigroup Emerge As Major Domestic Suppliers Says TrendForce . Another wrinkle is that Micron is suing some ex-employees now working for Chinese DRAM companies. Conclusion Gartner at least agrees with me (in fact goes further than I did). Andrew Norwood, a research VP, said: Memory pricing will weaken in 2018, initially for NAND flash and then DRAM in 2019 as China increases its memory production capacity. We then expect Samsung to lose a lot of the revenue gains it has made. I suspect that successfully manufacturing competitive low-power DRAM may turn out to be harder than expected. Obviously, the Koreans managed it, but over a longer time period. There is a lot more than getting the technology to work, and pouring money on the fire, to ramping a modern manufacturing process to volume. In another industry, try asking Tesla about that. I think that the other memory manufacturers are assuming this. They are certainly not lowering their prices just because it might happen, why should they? It reminds me a little of the early days of ASIC when often we, at VLSI Technology (along with LSI Logic), would be over twice the price of all the other semiconductor manufacturers trying to buy their way into the market. We'd tell them to go to the Japanese, and they'd be back begging us to do the order at an even higher price, now the timeframe was so short. We bet, and usually won, on the fact that the Japanese would have a hard time converting success in memory (a pure manufacturing business based entirely in Japan) to success in ASIC (more of a service business, requiring design centers all over the world). For years we were right. I'm at SEMICON China all this week, so maybe more will be revealed. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧