Anyone who designs in a memory product into their system has to be concerned about product lifecycles. For consumer products, this is rarely an issue…the system is replaced by a new and ‘better’ system in a few years, which can use a totally different memory. At the other end of the spectrum, Telecom and Military usually need products for more than a decade; indeed, suppliers to these systems may do one or two die revisions (‘shrinks”) before they get their first volume order from their customer…and a 15 year system life cycle is not unusual.

But for a kick-off of this discussion, recall that S DRAMs, which replaced Fast page, Static Column and ‘nibble mode” DRAMs in the later 1990s, are still in production today. When I worked at IBM from 1995 to 1999, we launched a 64M SDRAM using 0.45um design rules (that s 450nm, in today’s lingo). Today, there are still a few manufacturers who make upgraded versions of this popular part, but the mainstream SDRAM runners today are 128M, 256M, and 512M, all of which are multi-sourced…and, after successive cost-reduction die shrinks, are built today in technologies than go below 50nm, 9x finer line widths than they started with. While every vendors does a different roadmap, these original 64M/128M DRAMs from the second half of the 1990s, went through a series of 5-8 redesigns and die shrinks.

Together, these high-volume SDRAMs are still a market opportunity of more than several hundred million dollar a year. They SDRAMs were the “PC DRAMs” in their day, which began about 1998-99, to be replaced about three years after introduction by DDR1 in 2003, then by DDR2 in 2006, then by DDR3 in about 2009. For sure, DDR4 is coming. Each generation of architecture supports higher performance, denser chips (due to node migration progress) and reduced build cost and market price.

The first DDR3 chip was demonstrated by Samsung in February 2005, shortly after (or maybe a little before) the JEDEC DDR3 spec was issued. Qualification samples appeared in about a year from several vendors, and over the next 2 years, DDR3 gradually replaced incumbent DDR2 in new PC systems; some new PCs still sell with DDR2 DRAMs, though DDR1 is found only outside the PC space these days.

From first samples to EOL (end-of-life) production for PC DRAMs is about 14-15 years, and while each die revision often requires a new qualification at the customer, the datasheet is basically unchanged from generation to generation.

Outside the PC DRAM roadmap, things are still pretty good, with successive die revisions, but none is as long as for PC DRAMs. Eight to ten years lifetime is more the norm.

You cannot get VRAMs today, except in the after market (or one of the original vendors’ warehouses)…”they were so ‘90s”. Same goes for the early generation SGRAMs…8Mb and 16Mb, which were the predecessors to today’s wide-I/O G DRAMs, which started at 32/64Mb in 1999, and today are 1/2Gb…These are x32, very fast clocks for point-to-point designs like Graphics accelerators…Like PC DRAMs, the density of G DRAMs increases with advanced process nodes, but often the slower parts fall into disuse later on; faster can be just as cheap. Maybe the product can be had for 8-9 years, but probably only for a shorter period.

Rambus started with Concurrent DRAM in the mid-1990s, and then went to R RDRAM (Rambus Direct), and now XDR and XDRII. Since Rambus only develops and licenses the technology to DRAM makers, how long the products stay in production is up to the DRAM maker themself; with all the perpetual turmoil in the DRAM industry, sometimes companies disappear and take their product portfolio with them. You can still get RDRAM (after 8-9 years), and all versions of XDR today

RL DRAMs (Reduced latency DRAMs, or Fast Cycle DRAMs) were popular and essential for networking applications about a decade ago. It was a good part but was also a clumsy roll out for the industry, and most RL/FC DRAM makers disappeared I the dot.com bubble bursting in 2001; Micron in the sole survivor today (rumors of a second source are ever-present), and earlier this year, Micron announced a recommitment to supply 576Mb RL DRAM II, in their leading DRAM process technology …so, ten years and counting.

Every manufacturer always offers last time buys, as well, and gives 1-2 years notice before a product is being discontinued. After that, there are usually brokers and speculators who buy up surviving inventory and hold for resale later, if new demand turns up. These are usually still in factory-sealed containers and as good as “Right off the Factory Line.’

So, almost all DRAMs have a ten-year product availability cycle, covering several die revisions, with opportunity for last time buys and often continued availability in the aftermarket.