DRAM annual bit growth should stay in the 20 to 25% range while NAND can stay at around 40%. In 2019 or so, volumes for new NVMs should begin to be relevant. Might be less than ideal but there is nothing to panic about.
This year bit growth has been lower than that and pricing might get worse if Apple can meet demand for the OLED iPhone (the market severely underestimates demand for now) but supply will catch up with demand eventually.
1) Maybe demand has picked up unexpectedly fast for the producers so they haven't had time to expand production
2) Producers have started moving to 3D which means shutting down 2D capacity. Unfortunately 3D isn't working, so now nothing gets produced on the lines in transition.
3) Various temporary problems with production at the relatively few producers
However, none of these things means that RAM cannot be produced at 2015 prices again. #1 can be fixed by expanding production (new factories, ultimately). #2 Get 3D working or eventually abandon it altogether and move to 3D, #3 get rid of temporary problems.
The DRAM market has always been somewhat volatile and prices have gone up greatly in the past too.
San-Yun, I still don't agree or at least we are not talking about the same thing ;) What you are saying is that with shrinking, then through installation of a new tool, a relatively minor cost, then production could greatly increase on the same factory giviing a huge price decrease. I agree that this is true, but then you are talking about price decreases beyond the prices we were used for (per unit capacity) say in 2015. My claim is that prices on DDR can come back to where they were in 2015 without any innovation at all. Because whatever way they produced the chips back then, can just be replicated. If demand outstrips supply by a factor 3x then it means building 3x times the factories we had in 2015 using the same tools etc. Sure this will cost 3 times as much to produce as back then but revenue will also be three times as big (assuming same sales price as in 2015) so profit will be three times as big (and relative profit the same). As I said it could be that there are economies of scale which will drive down cost even more than they were in 2015 (still without innovation) whether that is the case nor depends on how optimized production were in 2015. But it is not a requirement anyway to reduce the prices to the 2015-level. The way things were produced in 2015 everything was profitable. There's no resource we have run out of and that is limiting us in doing how we did in 2015. So simply expand the full 2015 process until the demand is satisfied. Of course if innovation has come along since, even better, but it is not a requirement.
When DRAM makers make investment in fab (i.e. CapEx), they upgrade tools for advanced feature size in order to maximize productivity. For example, some tools at 22nm DRAM fab are replaced for 18nm DRAM production. It increases 33% production with about $500M CapEx.
If they build a new fab at 22nm without device shrinking, it will require about $3B fab investment and increase 100% production.
As a result, DRAM production without Moore's Law (i.e. building a new fab without device scaling) becomes TWICE expensive than DRAM production with tool upgrade (i.e based on device scaling).
Therefore, transition to 3D DRAM is not an option.
I don't see why DRAM prices would stay permanently high even if Moore's law runs out of steam. Presumably, the DRAM producers were making money when they were selling DRAM's at the low prices 1-2 years back. So even if someone proves that it is not possible to increase DRAM density any further (and that no other process improvements are possible), DRAM makers could simply expand production using the existing process. They can then mass-produce DRAM in line with rising demand, restoring the equilibrium. Prices might even come down due to economies of scale and also due to the fact that a fab can now be used and depreciated over a long period of time, rather than being obsolete in a few years. And further out, all patents on DRAM will expire and all the knowledge and IP will have become common knowledge, so that DRAM becomes fully commoditized and can be produced extremely cheat in China at a cost approaching cost of materials.
Of course there's a limit to that effect and at some point DRAM price will bottom (probably at a much lower level than the one we saw 1-2 years back) and from then on the prices will increase with the general rise in prices (inflation). Note again that the whole scenario assumes that no further technological improvements are possible.
Even though DRAM increases output along with more fab investment, the imbalance between supply and demand won't be easily solved because there is no more Moore's Law which increases # of bit cells EXPONENTIALLY. Without Moore's Law, fab investment just increases output LINEARLY, which is similar to Fig. 3.
So, customers are confused and frustrated by high memory price and hope memory vendors will find out solutions sometime soon. Technically, there is no ultimate solution in the industry except transition to 3D DRAM technology. As long as customers just wait for the downturn of memory price, due to limitation of 2D DRAM and the end of Moore's Law, the negative impact on electronics system market will be more severe.