Price Check Q3 2016: DRAM Prices Down Over 20% Since Early 2016
by Anton Shilov on July 25, 2016 12:01 AM ESTAnalysts: PC DRAM Prices May Stabilize
As our frequent readers know, DRAM prices depend on supply and demand. While demand for DRAM depends on makers of computers and consumer electronics, supply depends on output of memory manufacturers. All sides will always try to cut down their costs, which is why the manufacturers introduce new manufacturing technologies every 18 to 24 months (producing more ICs per silicon wafer) and continuously try to improve their yields (higher percentage of usable ICs per wafer).
Since smaller fabrication processes can increase the production of memory ICs without needing to expand the manufacturing fabs, DRAM IC output as a result increases significantly every couple of years. In a bid to slowdown such natural decrease of costs and prices, according to analysts, leading DRAM manufacturers intend to slowdown transitions to smaller process nodes. This sounds slightly counter intuitive, as any manufacturer with a smaller node will have the advantage over the competition (anyone interested in this should go look at Game Theory).
Market observers from DRAMeXchange point out that while new process technologies are getting more expensive to develop, and fabs are getting costlier to equip, manufacturers of memory are cutting their research and development spending. This comes about as a result of decreasing revenues, which will inherently slow down adoption of new process technologies and therefore slow down reductions of DRAM IC costs and prices. Avril Wu, research director at DRAMeXchange, believes that in 2016 the total DRAM bit output will slow down and we will only see an increase of 23.1% year-over-year, compared to a 30% to 50% increase typically seen.
According to Mr. Wu, this is going to happen because of the slow transition to 10 nm class process technologies, as well as insufficient yields at 20/21 nm at other manufacturers (a claim that cannot be verified).
Samsung has already started mass production of DDR4-3200 ICs using its first-gen 10 nm-class fabrication technology (which some market observers call 18 nm) and plans to make LPDDR4 chips using the same process in the second half of the year. To stay competitive, Micron and SK Hynix will simply have to follow their rival with their 1x nm fabrication technologies. As a result, depending on yields, DRAM bit supply will increase, whereas DRAM prices will likely decrease if demand cannot catch up with supply.
18 nm and 16 nm process technologies will inevitably increase output of DRAM, but they are not the only factor that will affect production of computer memory in the coming years. Samsung’s giant manufacturing facility near Pyeongtaek, South Korea, will begin operations in 2017. The new fab is said to be two times larger than Samsung’s S1 plant in Giheung, South Korea, and is expected to accommodate up to five semiconductor production lines. The total cost of the initial fab required investment of 15.6 trillion won ($13.5 billion today) and in 2015 the company announced plans to expand it even further (by pouring in additional $8 to $9 billion). The fab will be used primarily for DRAM production, but Samsung could also use it to make other chips (i.e., NAND). We do not know exact production capacity of the new fab, but the S1 can process up to 170,000 wafers (300 mm wafers) per month and the addition of a fab twice as big will have a drastic effect on the DRAM supply. As a result, prices will be affected.
One predicition about DRAM supply gravitates around major purchasers of large about of ICs. One company gets mentioned a lot in this case: Apple. A good news story for DRAM makers in the short term should be that Apple is expected to release its next-generation iPhone in the second half of the year, and hence it will require a lot of memory (analysts predict that this will be LPDDR4). Quite naturally, manufacturers will have to shift production from PC DRAM to Mobile DRAM, which will lower supply of DDR3 and DDR4 amid increase of demand from PC makers in the second half of the year. As a result, memory makers will be able to sell more high-margin LPDDR4 memory (not only to Apple, of course) and either stabilize or even increase prices of mainstream memory.
Facebook
Twitter
RSS
37 Comments
View All Comments
invasmani - Monday, July 25, 2016 - link
The majority of American's that actively live in the northern states most impacted by the Keystone pipeline deal were in favor of it. It's the fault of Obama and democrats who tried to make it into a big environmental issue despite the fact that a pipeline is much safer than railroad transport.catboy - Tuesday, July 26, 2016 - link
tipoo is correct. Corporations scam Canadians with outrageously unfair prices just because they can. Here is a news report about that fact:http://www.cbc.ca/news/canada/country-pricing-a-ca...
Of course, that report is not about hardware, but Canadians get price gouged on hardware in the same way as they do on all other products.
I recommend for Canadians to stop buying products from Canadian sellers whenever possible. If enough Canadians do that, then corporations will end the practice of price gouging Canadians.
doggface - Tuesday, July 26, 2016 - link
Same thing happens in Australia. $160 in USD = $350-$400 in AUD.The manufacturers price items much lower in the US to get the buzz from websites like these. Then mark it up in lower volume countries.
Another example. $US600 for GTX1080 = $AUD1150
Sure. Our dollar is lower. But it doesnt account for the massive discrepency. We call it the Australia tax
Arnulf - Monday, July 25, 2016 - link
I find it intriguing that Samsung has been producing complex circuits on 14 nm node for some time now yet they are only switching to 18 nm for DRAM production (which should be more dense).DanNeely - Monday, July 25, 2016 - link
Somewhere around half a DRAM die is made up of analog components used to terminate signals on the database; and unlike the digital circuits in the ram cells themselves analog components scale minimally with process shrinks. As a result DRAM gets less of a benefit from process improvements than things like CPUs/GPUs/Flash that are almost all digital components on the die. The analog penalty has gotten worse with each new generation of DDR because to keep the data bus stable at higher frequencies the termination components need to be moved closer to the DRAM; leading to a steady migration of them from the mobo to the dimm to the dram chips themselves.p1esk - Monday, July 25, 2016 - link
Source? Everything is analog of the circuit level.DanNeely - Monday, July 25, 2016 - link
It was something I read back when the DDR4 spec was first released so I'm having trouble finding it (will look more later); but analog components are things that aren't transistors eg resistors, capacitors, inductors (also power transistors which need to be a certain size to carry the amount of current that they do or a lot of RF components; but neither of them are a factor here). Physical size is a major component in how they perform. ex Make a capacitor half as large and all other things equal it's capacitance is only half as great.It's one of the factors behind why the minimum size dimm goes up every time there's a new process. The lower capacity dram chips see the least shrinkage because the largest fraction of their die is signal termination components that don't shrink much.
yuhong - Tuesday, July 26, 2016 - link
I guess that is why they often eventually drop things like x4 configurations when moving older DRAM like 1Gbit DDR2 to newer processes.jardows2 - Monday, July 25, 2016 - link
I built my current computer in 2102, and purchased 8 (2x4) gigs of DDR3-1600 RAM. Thinking I could eventually upgrade to 16 gigs if needed, but then memory prices shot up. I can finally purchase the same RAM I did 4 years ago at a slightly lower price, instead of a significantly higher (at times was double what I paid) price!bananaforscale - Monday, July 25, 2016 - link
I built mine in late 2011 and bought 4x4 GB of DDR3-1600. Decided to upgrade the memory past winter, the price per GB was still about the same for the DIMMs I used, but doubling the amount and buying faster memory wasn't that much more expensive -> went from 16 GB 1600 to 32 GB 2133 (except the CPU only supports 1866 but whatevs) and distributed the old memory among less important hardware.