Memory: Standard A Day Keeps Strangers Away
Commodity memory market is probably among the most predictable things in the technology industry. Manufacturers of dynamic random access memory (DRAM) just love standardization as it allows them to concentrate on efficient manufacturing and not on ensuring that their products actually work with certain controllers. Device makers love that too since they have a lot of sources for DRAM. This is something that is not going to change even in 2020.
As we do know, leading DRAM manufacturers will start commercial production of next-generation DDR4 memory in 2012. Actual mass transition to the new memory is projected to occur towards 2015, which means that by 2020 there will be DDR5 ramping up. As we also know, the 2133MHz - 4266MHz effective clock-speeds of DDR4 lead to change of topology of memory sub-system to point-to-point interconnect. As a result, DRAM manufacturers will need to increase capacities of memory chips by using multi-layer technique with through silicon via (TSV) technology. In case of server multi-layer DRAM IC approach only will not be viable for high-end machines and special switches will be installed onto mainboards to allow multiple memory modules to work on a single memory channel.
The point-to-point topology of DDR4 is rather questionable, but, they say, inevitable. Given the fact that JEDEC, the organization that standardizes memory and other technologies, makes major changes pretty rarely, it is highly likely that point-to-point topology will be a part of DDR5 as well. Naturally, it is possible to expect 4.20GHz - 8.40GHz effective clock-speeds as well as generally smaller form-factors for both chips and modules.
What we do not expect to happen is the emergence of a proprietary standard that will actually have chances on the mass market. In the past decade we saw a number of attempts to push proprietary memory technologies onto the market place, but whether they involved completely different memory chips (XDR) or simple modification of platforms (QBM), they all have failed.
There are a number of technologies that can potentially combine the advantages of flash memory and DRAM. We do expect some of them to find their place on the commercial market in the next ten years, but we hardly believe that they will actually replace DRAM and NAND flash as we know them today.