XLC Disk, Inc., founded in 2005 and based in Boulder, CO, is a fabless
focused on the SSD market.
Disk - editor mentions on STORAGEsearch.com
- editor's comments:- March 2011 - XLC Disk (first mentioned on
these pages 3 years ago - in March 2008) has been dipping in and out of
stealth mode in recent years and does not yet ship any products or license any
of its technologies.
The company - which is well
funded - is
apparently developing advanced
MLC technologies in
preparation for an exploding SSD market which some
believe could reach
$100 billion / year
The company refuses to comment on when or what MLC level
it will commercialize its products (x3 aka TLC, x4, x5?) - but the company's
CTO, Dimitri Sholokov, has told me that unlike other
flash SSD oems -
who anticipate deep MLC SSDs going mainly into consumer markets - his
company, XLC, is solving the technical problems needed to get
retention levels upto the standards needed for enterprise server apps.
a starkly differenct philosophy to another stealth mode SSD company -
Exabyte SSD - whose website is also
currently blank. Exabyte SSD's view is that only SLC will be reliable enough for
bulk storage SSDs
- and their approach is to develop the architecture and technology for power
management within such SSD
XLC Flash SSD Technology - March 2008 |
|editor's comments:- from my
limited contacts with the founders of this company I expect XLC Disk's products
to be flash SSDs - probably in
2.5" form factors.|
at their patent applications - it appears that XLC is a multi level cell flash
technology which offers 4x the capacity of SLC and 2x the capacity of MLC for
the same chip area.
It has been technically impossible for
researchers in long established chipmakers like
commercialize reliable flash memory with more than 4 levels so I wondered -
how is a well funded
start-up like XLC going to do better than that?
XLC's founder refers
to MLC as "dumb first generation multi level flash" and XLC as the "multi
level flash of the future."
From what I can gather XLC's
technology goes to the core of the multi level cell problem and the problems of
logic level discrimination and
Unless you're deeply versed in the internals of chip
design - these notes won't help you. But it's very hard making reliable MLC
flash - even when you control the whole process.
flash chip companies get over the limitations by using extensive redundant
data blocks and ECC. Until now no one has succeeded in storing more than 2
bits of data in a high density flash chip - and the reason lies in physics.
When the stored charges are related back to voltage levels - there's a wide
variation - not only over the whole area of the chip - but also in the same
part of the chip over time - due to the accumulation (or changes) in charge in
adjacent parts of the chip which may have occurred since the data was written.
In simple language - you don't always read out the same digital
value that you wrote in. Because the variation and tolerances in the end
to end read write process are greater than the thresholds between the logic
The problem is mitigated in today's MLC SSDs - because there
are only 4 levels to discriminate between. But when you try to increase this -
to say 8 - then the unreliability of the read back data - is very high. Adding
more ECC doesn't help - because you get into a situation where the unique data
capacity (after ECC) is lower than that achieved with traditional MLC.
solution is design distributed discriminator circuits in zones throughout the
chip - and instead
of relying on just traditional ECC - they add redundant calibration cells in
these zones which are prewritten with known data each time a block erase write
cycle is done.
When a read cycle occurs - the converted voltage is
compared - not to a fixed level or percentage of the rail - but to adjacent
levels written by the calibration controller. Variations in charge and voltage
in that zone of the chip affect djacent zones in a similar way. The
discriminator circuits compare the address selected data cell with nearby
calibrated cells. That means even if the voltage varies by 20% (which is more
than the difference in logic levels) the loc state is read back correctly -
because the read is effectively differential - the difference between adjacent
known calibrated cells.
Although the internal read and write processes
are slower than in SLC or MLC flash - the amount of data in a single cell can be
much more. For example an 8 level XLC can store 3 bits of data in a single cell
- which gives 2x the capacity of MLC and 4x as much as SLC. So from the
external chip point of view the data throughput is similar (or more) than SLC
even though the internal chip latency is slower. One by product of the process
is better endurance than MLC - approximately 2-3 times better (but still much
lower than SLC).
XLC won't say yet how many levels for XLC the first
commercial devices will have. They are characterizing the technology across a
range of foundry technologies. And that's another difference. XLC is less
process dependent than MLC.
XLC holds the promise of shattering the
cost per gigabyte gap which has separated
flash SSDs from
hard disk drives. I
look forward to the first unveiling of this technology, here on
STORAGEsearch.com April 1, 2008.
to 3 / 4 bit MLC flash
- ...Later:- the first draft of the article above (written in March
2008) was meant to be a work of fiction - and a spoof April 1st news story.
Although I had the sneaking feeling that some of the ways I had "invented"
of solving the multi level data problem might work at some time in the future...
add authenticity to the spoof - I contacted Jim Handy at
Analysis to elicit an analyst's comment on the new technology.
can imagine my delight when he said
SanDisk was already
working on something which sounded very similar. That shows that the SSD
industry is programmed on a "fast forward" rate that can surprise us
all - even editors who are embedded in the market.
An article by Lane Mason (Denali Software,
Inc) looks forward to
8-12x price reduction, in solid state storage, by the move to 4-bit MLC cells,
|Driving the need to develop
these extraordinarily complex SSD technologies is the certain knowledge and fear
that traditional SSD controller IP will fail to deliver working SSDs with future
shrinks of flash geometry.
That means a successful SSD company in generation X flash may wake up
to be told by its designers that its SSDs using generation Z flash - can't be
made to operate at all - let alone last for 1, 2 or 5 years.
the company's VPs to create / update their bios on Linkedin.)
care management & DSP IP in SSDs|
promises "enterprise" hybrid x4 SSDs |
|Editor:- April 1, 2010 - XLC Disk announced
details of a paper it will discuss later this month at the NV
Memories Worskhop (UC San Diego) called - Paramagnetic Effects on
Trapped Charge Diffusion with Applications for x4 Data Integrity.|
company says its findings could have applications in the enterprise storage
market by solving the
problems in x4 MLC SSDs within a new class of
Editor's comments:- I spoke to XLC's CTO - Dimitri Sholokov - he started by saying
he was a "great fan of the mouse site" - but got the impression we
were editorially biased
against the use
of MLC flash in enterprise apps.
"Sure - it's well known
there are many problems to be solved to get x4
integrity up to
the levels needed" said Sholokov. "Most designers are looking at
improving quantization errors and ECC. These are necessary to be fixed - but do
not go to the heart of the problem - which is charge diffusion - in the steady
state - and charge dumping and disturb effects. Even with perfect ADC and a
long chain EC - they only tell you that the data has already gone. You can't
get it back. Our technology shows a way to reduce this charge corruption before
it makes a data error."
XLC says it has experimentally
verified that attaching x4 flash chips distributed around a spinning platter
(5,400 RPM) - with the same diameter as a 2.5" hard drive - can in his
words "make the charge diffusion behave better." Orientation is
apparently important, and so too is the presence of a strong magnetic field.
Dimitri joked "Your readers may think this is kind of centrifugal force
stopping the charges falling out of their wells - but Newton's physics is not
known in these places - it's
paramagnetic force that tips the charge tunneling probabilities."
asked Dimitri if that meant his company was planning to set up a factory making
He said "No - that would be
ridiculous - then we risk being sued out of business like
Cornice - because of
some patent we didn't know about for gluing chips to a spinning platter. Let's
get serious - the hard
disk companies are the experts at making small mechanical things cheaply.
We would love it if they license our IP. We need them and they need us to get x4
into the enterprise."
management in flash SSDs|
|This is an introduction to the thinking behind
one of the many vital functions inside a flash
media defect quality in new flash memory chips has grown steadily worse in the
past 10 years as geometries have shrunk.
||This article enumerates
the scale of the problem and explains how intrinsically dodgy flash memory
is transformed into dependable flash SSDs which you can entrust with your data.
|SSD Data Recovery
|It's hard enough understanding the design
of any single SSD. And there are so
many different designs
in the market. |
Have you ever wondered what it looks like at the
other end of the SSD supply chain - when a user has a damaged SSD which
contains priceless data with no usable backup?