This resource page
includes notes and articles about DRAM modules which automatically save their
contents when electrical power drops to an integrated non volatile memory
from which the data is reloaded after normal power is restored.|
scope includes but is not limited to:- flash backed DRAM DIMMs, hybrid DIMMs
and NV DIMMs - in which the primary resource seen by the host is a random access
low latency, infinite endurance memory - which from the R/W point
performance of view is identical to a DRAM - but whose contents are
transparently saved to an integrated non volatile memory (such as flash) in the
event of power disturbance - and restored when normal line power resumes its
The scope excludes:-
SSDs - which are low latency flash SSDs which are in a DRAM form factor and
which use the native DRAM interface as the primary data transfer route.
|DRAM with integrated backup
and restore in a DIMM socket |
|by Zsolt Kerekes,
editor - -
October 21, 2014 (updated in 2017)|
|historical context and evolution|
as soon as the first bytewide CMOS static RAMs became available in the late
1970s - electronics designers realized that with the addition of a small lithium
battery (which could maintain the RAM's contents in the unpowered state for 5 to
10 years) they could easily create a unique type of storage device which was
easier and cheaper to integrate into small physical spaces than the other
options then available.
This enabled many useful functions to be
included in products which would - less than a decade later - be mostly
replaced by flash.
For applications which required more than a single
NV RAM memory device - the economics of system level integration were
And for applications which used large numbers of RAM
chips in a single non volatile array configuration - the early history of
non volatile RAM is really the early history of
RAM SSDs - as cards,
drives or rackmount products - rather than as specialized components.
flash entered the scene of keeping DRAM content alive
As RAM SSD
capacities grew in the 1990s - the physical space occupied in the racks by
batteries became a significant percentage of the SSD itself.
reduce the risk of data loss - the battery systems themselves started to get
more complex in order to prevent the loss of all data due to a single battery
So by the early 2000s - the leading RAM SSD makers optimized
the battery footprint problem by including a
hard drive in the SSD
rack to which data was backed
up when the incoming power went down. That also meant the batteries only
had to maintain the DRAM array in a powered state long enough to support the
backup and restore operations.
2007 - the
gap between the growing size of enterprise SSD capacity and the poor random
IOPS performance of a single backup hard drive - resulted in cold boot
restore times for high capacity RAM SSDs which could take from 30 minutes upto
over an hour.
asymmetry between R/W and power down/up characteristics - compared to the
needs of the big server users who formed the vanguard of enterprise SSD
adoption was an uncomfortable truth.
July 2008 - a
new solution to the power up restore ready time for large DRAM arrays was
illustrated by Texas
Memory Systems - who included a
RAIDed flash array in
their RamSan-440 (a
512GB 600K IOPS FC compatible RAM SSD).
And although - at 4U rack
space and $290K price - it was a long way from the flash backed DRAM DIMMs in
the market which would follow in later years - in another way it was the
ancestor of them all - inasmuch as it demonstrated that viable enterprise
storage products could be engineered by intertwining the most desirable
characteristics of each type of memory -
flash - into a useful new
type of hybrid.
enter flash backed DRAM in modern DIMM form
As has oftentimes happened in
- any good architectural ideas which first prove their worth in system level
soon trickle down and reappear at component levels. And so it was with the
flash backed RAM concept.
The main market events (as they appeared
from an SSD news point of
view) being represented in the following timeline.
In May 2009 -
AgigA Tech began sampling
the AGIGARAM - a DDR2 compatible module which implemented upto 2GB of non
volatile SDRAM (with integrated flash backup and restore) which plugged into
standard JEDEC DIMMs. This required a companion module to be located somewhere
close to the memory - which routed power to a bank of supercaps and power
AgigA Tech explained in
related architecture paper (pdf) that the supercaps were only required to
support the backup and restore operations to and from the onboard flash. The
backup took about 9 seconds for each gigabyte of RAM.
In October 2011 -
Viking - entered the
flash backed DRAM DIMM market with the launch of the ArxCis-NV - which plugged
RAM sockets and provided
uptto 8GB RAM backed up to SLC flash in the
event of a
power failure. As was the case with the earlier products from AgigA Tech -
the Viking ArxCis-NV required a separate "tethered" supercap module to
implement a functioning system.
In October 2013 -
SMART announced that its
DDR3 Non-Volatile DIMMs (NVDIMMs) were fully compatible with and automatically
recognized by the BIOS on new Supermicro X9DRH-iF-NV server boards.
SNIA formed the
which at the time of launch included 12 companies:- AgigA Tech, IDT, Inphi,
Intel, Micron, Microsoft, Pericom, Samsung, SK Hynix, SMART and Viking.
- AgigA Tech
announced that it was sampling the industrys first DDR4 Nonvolatile DIMM
(NVDIMM) to key OEMs and development partners.
In November 2014
- said in an
conference call - "We have shipped almost 0.5 million units of
NVDIMM over the past 5 years. This is over 20x more than the total
shipments from all of our competitors combined."
In August 2015 -
Micron unveiled high
capacity RAM-like products which would compete alongside and take business away
from the flash backed DRAM DIMM market. See
opening salvo for more about this.
In December 2015 -
"retiring and retiering enterprise DRAM" was named as one of the
the big SSD
ideas to have emerged 2015.
In September 2016 -
the issuance of U.S. Patent No. 9,436,600 (the '600 patent) relating to a
multi-channel hybrid memory architecture that uses non-volatile memory to
protect critical data in the event of power loss.
| SSD news|
nvm news from an SSD angle
changed in SSD year 2016?
how fast can your SSD
Big versus Small
in SSD controller architecture
utilization and the
enterprise SSD event horizon
in the event of sudden power disturbances in SSDs
|Did NV DIMMs kill the RAM
(those few companies still left after the transition to
In 2015 a reader asked - why is it
so hard to find suppliers of
sure enough - when looking at past product lines - they all seemed to have
reached a dead end.
Aha! There was a simple explanation.
The low capacity, high
speed RAM SSD market - which used to come in various shapes and sizes (PCIe
SSDs, 3.5" etc) - was killed off by the NVDIMM market.
is slightly more complicated than that because the NVDIMM and NV memory and
systems market is more granular and segmented due to new memory types. So there
is a larger choice of raw latencies and capacity densities.
about the direction of technology flow (from a systems perspective) see -
where are we
heading with memory intensive systems and software?
|Viking aims to design ReRAM
|Editor:- August 10, 2015 - Viking Technology and
a collaboration agreement to develop the next generation of Non-Volatile Dual
in-line Memory Module (NVDIMM) products leveraging ReRAM Storage Class Memory
from Sony Corporation. |
resilient are in-memory databases in NVDIMMs?|
|Editor:- October 9, 2013 - what happens if you
pull out the power plug during intensive in-memory database transactions?
(whose products include in-memory database software) yesterday
benchmarks using AGIGA
Tech's in which
they did some unthinkable things which you would never wish to try out for
yourself - like rebooting the server while it was running... The result?
Everything was OK.
"The idea that there must be a tradeoff
between performance and persistence/durability has become so ingrained in the
database field that it is rarely questioned. This test shows that mission
critical applications needn't accept latency as the price for recoverability.
Developers working in a variety of application categories will view this as a
breakthrough" said Steve Graves,
Here's a quote from the whitepaper -
"In these tests
eXtremeDB's inserts and updates with AGIGA's NVDIMM for main memory storage
were 2x as fast as using the same IMDS with transaction logging, and
approximately 5x faster for database updates (and this with the
transaction log stored on RAM-disk, a solution that is (even) faster than
storing the log on an SSD). The possibility of gaining so much speed while
giving up nothing in terms of data durability or recoverability makes the IMDS
with NVDIMM combination impossible to ignore in many application categories,
including capital markets, telecom/networking, aerospace and industrial