| leading the way to the
new storage frontier||...|
SSDs - boring right?
top SSD companies in Q1
AFAs - what's the next box?
can your SSD run backwards?
hold up times in 2.5" military SSDs
where are we
heading with memory intensive systems?
CPUs for use in SSDs in the Post Modernist Era
and risk reward with big memory "flash as RAM"
3D nand fab yield - the nth layer tax?are more dimensions of
analysis needed to get a clearer picture of future 3D nand successions?
editor - July 3, 2017|
If you're trying to predict and
anticipate how the supply of next generation nand flash will ramp up in the next
year compared to how you've seen memory successions before then the 3D nand
flash market has presented many problems of analysis and interpretation.
nand flash is still the dominant memory technology used inside over 99.9% of
SSDs and as the switch from planar 2D to 3D was pitched as the lowest risk
solution to continue supplying the growing need for more flash storage
- because memory geometry shrinks had gone about as far as they dared go
- there has been much uneasiness and disquiet recently in markets which use
these memory chips about the
which have been the most obvious sign that progress has not been entirely plain
Many blogs in the SSD market in recent quarters have focused
on the technology and business impacts (such as higher costs and lost
opportunities to make SSDs) with various degrees of informed speculation and
What's really going on?
One of the key secrets of
memory makers ever since the earliest days of semiconductor memory is that the
yields and manufacturability problems are kept closely guarded secrets because
they are competitively very sensitive. So the only way we're going to get an
accurate "from the horse's mouth" picture of everything which has
contributed to making these tranistions so bumpy is after the
underlying problems have been solved which the memory industry will signal in
the traditional manner by promises of lower prices and quicker deliveries.
of the best summaries I've seen of the flash industry's predicament is a blog
I recommended here in
at Yokkaichi - the NAND Industry at the Crossroads by William Tidwell,
Semiconductor Analyst who regularly
about such things on Seeking Alpha.
And I make no apology for repeating that here.
Another blog on the
current 3D successor generations which I think many of you would find
interesting is the newer (June 21, 2017) -
Hits Speed Bump by Mark LaPedus,
Executive Editor - Semiconductor
Mark's blog provides a snapshot of the current 3D
succession plans by the leading memory makers. And if - like me - you've already
seen more than enough of such pretty 3D memory pictures you might be tempted
to just click away. But I found 2 observations in that article particularly
First from Applied
Materials which is a strategic supplier of wafer level production,
inspection and testing technologies used by memory chip makers. AMC expects
wafer starts for 3D nand to be in the region 500,000 to 700,000 wspm (wafer
starts per month) by the end of 2017 - which is about 60% more than a year ago.
The other interesting comment came from Greg Wong, President -
Greg's point is that while going from 2D to 3D gave
the industry the same kind of
bit density improvement that the industry got accustomed to with decades of
Moore's Law geometry shrinks - the shape of the bits per chip roadmap curve -
as you add more layers of 3D to an industry which is already using double digit
layers of deposited
memory - is not going to be the same for successive N+ layer generations.
you add more layers to a 60 layer memory - for example - the next increment
can be any arbitrary level which makes sense from the manufacturing and
marketing points of view. The next steps don't have to be x2 jumps as they
often were in 2D - because the process challenges are different.
important thing which has been missing from most of the articles which I've seen
about 3D nand successor generations is any clear picture of how much longer each
3D chip spends in the fab compared to earlier 2D memory generations.
promise of the initial switch to 3D memory was you get more memory bits for a
similar number of wafer starts and invested wafer fabs.
that adding each memory layer adds more processing time. It would be useful
to have an idea of how that scales - and the risk reward factors of yield due
to rejects and rework - because that would give customers in the SSD industry
more realistic expectations of how they can interpret the other data they get
from the memory industry. (Investments in new fabs, and roadmaps on the number
of layers and virtual bit density TLC, QLC etc.)
A "time served
in the fab" model for incremnental layers would provide more confidence
in future generations of memory roadmap too.
"feed the fab"
- productivity issues
After writing the above I thought I'd better
look around and see if I can point readers to articles which discuss wafer fabs,
yield and time spent in the fab in more details. As I expected - due to the
business sensitivity of such insights - the papers I found were rather old or
talked about DRAM etc rather than flash. But they do give you a feel for some
of the strategic issues - which if you haven't had close contact with the
semiconductor business will tell you more about the tactics and strategies used
in the production flow.
- Enhancing the
Effectiveness of Cycle Time Estimation in Wafer Fabrication-Efficient
Methodology and Managerial Implications (pdf) - 2014 - includes this
"Estimating and shortening the cycle time
of each job is an important task to maintain a competitive edge in the DRAM
industry. For example, Samsung, implemented the short cycle time and low
inventory (SLIM) method to estimate the cycle times and WIP (work in progress)
levels for various manufacturing steps, so that a more effective control of the
factory was possible. As a result, the average cycle times of some DRAM products
were reduced from more than 80 days to less than 30 days, bringing Samsung a
benefit of about $1 billion.
I still wasn't satisfied with that. I know people in the industry today who can
tell you more if they choose to. But how to reach out to them?
Semiconductor Manufacturing (pdf) - 1997 - says this...
older-generation processing equipment on newer process flows may make the
achievement of world-class defect densities much more difficult than if newer
equipment is used. While yields may be lower when employing older processing
equipment, capital costs are lower as well, and so the strategy might turn out
to be economically competitive or even superior to the strategy that employs
solely new processing equipment."
This is relevant given the
prequel to the 3D memory direction story in which all the memory companies had
been losing money and reluctant to invest in entirely new equipment.
did this post on my
linkedin connection (and as it was July 4 I didn't expect to get many
answers quickly) but I was wrong about that.
nand fab yield - the nth layer tax?
How many extra days does
it take to make 60 layer nand flash compared to say 30 layers?
And how does that compare to planar?
I mean the
incremental time in the fab for manufacture and test. Let's call it the nth
Having a good grasp on these numbers will help memory market analysts
and systems customers get a more realistic feel for how the memory market can
ramp up to successor generations assuming any stable reference start point of
fab investment and wafer starts.
It's clear from the current shortages that we need more realistic
expectations about the manufacturing impacts of moving up the layers. On the
other hand the pressure of market demand and longer times in the fab have been
good incentives to bring forward QLC.
Here are some of the comments.
- Rob Peglar
Senior VP & CTO, Symbolic IO
"Zsolt - there is no single answer to this question. Highly
dependent on layer geometry, aspect ratio, materials used, deposition method,
bits per cell, etc. Dozens of factors, truth be told. Plus, it would be highly
unusual for the exact same characteristics and methods to be used in two
different products (# of layers). Sorry for not being of more help :-) but
that's the reality. No easy answers."
- Jon Haswell
- VP Firmware at SK hynix memory
"This whole question is at the heart of the
cooperative strategy of each NAND vendor. To the first order everybody knows how
to build 3D NAND now, the question is how to build it cheaper, trading off
geometry versus yield versus FAB time/steps. Nobody is likely to publish their
roadmap for this as they don't want to tip their hand, good or bad to their
Best the analysts can probably do is to plot the historical trends on
die capacity, die size and volume each vendor supplies to determine the
approximate yielded FAB capacity."
|rugged SLC like a legendary
blend of Don Quixote and Geronimo has staunchly defied many siren calls to
|Editor:- June 28, 2017 - planar SLC flash may
one day acquire the same romantic mythical status in the SSD market that
tournaments did in the early days of gunpowder. |
Old data warriors
drinking their murky brews on days which blur into each other unpunctuated by
the wake up calls of work related deadlines will say to one another - Dost
thou remember those days when we used to dress up in shiny armor and charge
wildly at each other on horseback while aiming our pointy lances?
had strong muscles and chivalry in those days. Now we just tap an icon to send
a poison cyber capsule or squeeze the virtual trigger for a remote drone
strike... Ah the good old days of SLC (memory you could trust to the care of
childlike simplistic controllers).
For anyone with a grasp of memory
roadmaps and generational progress it has been inevitable for a while that one
day SLC will come to an end - because when the cell sizes become sufficiently
small you need clever controllers to get data integrity. And if you've already
got them for any one purpose then you might as well use them for the next
obvious thing too.
Nevertheless - SLC like a legendary blend of Don
Quixote and Geronimo has staunchly defied many siren calls to surrender and
is still to this day the dominant memory technology in the hotter outposts
of the industrial and military markets.
That's why I draw your
attention to the
by the industrial SSD company Swissbit which
announced that in December 2016, for the first time 1/3 of all the flash
products sold by Swissbit were based on MLC. (Which means there's still time
for your squire to feed dress the horse and polish up the metal jackets.)
Knight of the Seven Kingdoms
|VCs in SSDs
the SSD news
trust SSD market data?
RAM really? - RAM in an SSD context
Can you tell me the best
way to get to SSD Street?
|"The Internet used to
be a one-way flow of information. Systems were only required to enable websites
to provide information feed to the consumer. Today, with more and more
user-generated content and uploads online, storage methods have also become
|Kevin Wang, VP
Sales - Longsys
in a press release -
needs in the IoT and IoV era (June 5, 2017)||
| IC Insights reports record breaking memory
Editor:- July 20, 2017 - A recent
research note about the memory market by IC Insights puts an interesting
spotlight on memory shipments.
Among other things IC Insights says:- "DRAM,
unit shipments are actually forecast to show a decline this year. Moreover,
NAND shipments are forecast to increase only 2%."
it comes to price expectations IC Insights says this.
though DRAM ASP growth is forecast to slow in the second half of the year, the
annual DRAM ASP growth rate is still forecast to be 63%, which would be the
largest annual rise for DRAM ASPs dating back to 1993 when IC Insights first
started tracking this data. The previous record-high annual growth rate for
DRAM ASP was 57% in 1997. For NAND flash, the 2017 ASP is forecast to increase
33%, also a record high gain. (In the year 2000, the predominantly NOR-based
flash ASP jumped 52%)."
For those who need much more information
IC Insights publishes a 250 page report ($4,090) which includes various free
monthly updates. ...read
Editor's comments:- One message to take away from this is
that as memories have been transitioning to the next multiple of 3D layers the
chip throughput from the industry's legacy wafer fabs has stayed the same or
gone backwards due to the extra time taken to reliably make those extra layers
to create higher bit density memories.
Related guides here on
StorageSearch.com with an SSD
Virtium says it's no longer a matter of if, but when, NVMe
supplants AHCI in industrial SSDs
Editor:- July 19, 2017 - Last
week I noticed a new blog from Virtium -
Taking a Seat at the Industrial-Embedded Table - and that prompted me to
ask - "Can Virtium tell me about any new industrial equipment or
application roles in which the availability of NVMe PCIe SSDs was the deciding
factor for their customer and for the feasibility of the project?"
spokesperson for Virtium said - "The concept of NVMe is new to the
industrial-embedded space, for the very reasons we cited in the blog that
the capacity needs simply didn't demand PCIe performance that be offset by the
penalties it would exact, namely software incompatibility and increased power
consumption. Industrial-embedded-system designers, therefore, are just now
starting to look at the NVMe-SSD "marriage" in a different light,
given how IIoT data is going to require a lot more storage. So, at this point
we can't cite any decided-upon designs that will definitely feature NVMe SSDs.
However, we can say many of the Virtium customers we have spoken with about this
concept agree that it's no longer a matter of if, but when, NVMe supplants AHCI
as the SSD interface in certain industrial-embedded designs."
is not your Grandfather's industrial SSD market
Storage Visions Conference 2017 - call for papers etc
July 18, 2017 - One of the big changes in the storage events calendar this year
is that the annual Storage Visions
Conference has moved from its traditional January (CES compatible) location and timing up to
October. Apart from the more convenient date that may also reflect that the
content of the conference has expanded and adapted in recent years to cover
all aspects of storage such as the enterprise and cloud instead of just the
consumer stuff which you'd expect from its proximity to a big consumer show.
2017 Storage Visions Conference will
be held Monday, October 16, 2017 in Milpitas, CA.
Listed among the many
you might be interested in attending include:- "Understand new applications
for storage class memory" and "Find new methods for reducing power
Speaking and panelist submissions are open through
August 15, 2017 and you can submit a speaking proposal at:
This linked page also includes contact details.
I've been watching the Storage Visions Conference grow in stature in our
industry (outlasting many others which sparng up in the early 2000s) and I
admire the dedication and work of its creator Tom Coughlin.
Having said that one of the odd things which puzzled me about SV
2017 (and which I mentioned to Tom this week) was the unclarity of the typeface
and dischordant contrast of the raw text compared to the background in the
banners and logos which have been designed to attract interest in the
included one example to the right of this here.
And you can see more
variations and judge the clarity or otherwise of the designs for yourself on
the SV 2017 banners
I'm color blind so maybe some of that is just me. But
isn't English words set against Ancient Egyptian heiroglyphics just text on
text? The background is too interesting and draws attention away from the text.
While the letters in the text are too hard to instantly recognize as their
contrast changes from top to bottom.
This is the kind of
readability issue which I've been warning
since the 1990s. And if you're interested in such web communication matters
see my seldom updated site Marketing
On the other hand - I hope that the seriousness of the
topics and the event itself will help you see past the idiosyncasy of this
year's Storage Visions banner. And if you want to spread the word then you now
know where to look.
Hey you're looking at a site which has
mice in its logo. So I
trust you can see beyond such superficial factors.
WDC secures advance warning of Toshiba nand asset changes
July 14, 2017 - Western
today it has obtained a court order to prevent Toshiba transferring its
interests in jointly held nand flash assets without giving prior notice to WDC's
(which shares some of these resources) "to ensure that the issue is
preserved for arbitration."
A follow up
in Japan Times clarifies that the notice period is "2 weeks"
and furthermore says "Toshiba lawyers have contended in court documents
that any sale of the chip division was not expected to close until early next
year, at the earliest."
forced memory sale - timeline of past stories
3D eTLC adoption - blog by Toshiba
Editor:- July 17,
2017 - The long saga of
naughty flash in
the enterprise (2004 - 2017) had already embraced 3D TLC by 2015 (in low
DWPD arrays) - but I have
to confess that although I had seen the prefix "e" used before for MLC
I didn't realize this hardening veneer of marketing
jargon had been
deemed worthy enought to have been been recycled unto the next generation of
memory too - as eTLC.
But there it was - eTLC -
large as life in a recent blog -
Flash in a 3D World What is it Good For? by Jeremy Werner
VP of SSD Marketing Toshiba
America Electronic Components - so I suppose it must be true.
other things Jeremy's blog mentions that early (2009) doubts by some
commentators about the efficacy of MLC in the enterprise were wrong! and in
recent years the market enabling factor for newer generations of MLC, TLC etc
is that there is now "Better understanding of workloads..." ...read
Which prompted me to say on linkedin
article placing naughty flash in context. As Jeremy says the enterprise didn't
understand the workloads in storage when flash SSDs began to be used in arrays.
Now we have a much more sophisticated market which can deploy any type of new
memory to a latency and cost per gigabyte role where it is most cost effective."
ahead it's tempting to see a pattern here. So in the years to come will we see
In reality this type of "enterprise" differentiation
of flash grades has long been meaningless at the technical level because it
makes much more difference to the system reliability what the
controller does and
the software and the
(NV)RAMcache that's available for data manipulation before it gets written
I guess that memory manufacturers have to choose
differentiate their many selections of flash and "e" has been useful
and is easy to deploy. And is less likely to be misunderstood than choosing
other letters such as for example "b" for "better" as in
the reliability difference in solo industrial SSDs
July 14, 2017 -
Reliability is one
of the concerns which got me interested in SSDs in the late 1980s, and the other
factor was raw speed
- sometimes - but not always - both in the same project. And different ways of
looking at reliability is one of the recurring themes which I notice in stories
about the industrial
Earlier this year I had noticed a statement in one of
customer case studies on
the web site of Cactus
Technologies which talked about having delivered 200,000 high
reliability flash storage cards to a customer "without a reported failure".
And from time to time I wondered what did that really mean?
week I asked Steve
Larrivee, VP Sales & Marketing at Cactus what was the time period
behind the story?
Steve said - "The 200,000 cards were delivered
over a 2 year period over 5 years ago without one reported failure."
comments:- I thought this was an impressive retrospective story and for
customers with applications where the reliability of each solo SSD is
critical it's a more convincing positioning statement about the design and
manufacturing capabilities of the SSD creator than any forward reaching
promises can be.
After our exchange of emails Steve wrote a new blog
about this -
Memory Failure Be Catastrophic to your business? - which included
additional anecdotal failure rates for the same application which happened when
the customer switched to a lower cost memory SSD design from a competing high
services marketing related to enterprise SSD systems
was it so hard to compile a simple list of military SSD companies?
WekaIO compares cloud storage pools to IBM FlashSystem
July 12, 2017 -WekaIO - a
today emerged from stealth and
details of its cloud-native scalable file system which the company says can
deliver performance comparable to
rackmount SSDs /
Editor's comments:- The notable thing for me in this announcement
was that WekaIO uses a performance benchmark compared against an
IBM FlashSystem 900 (the
decendant of the RamSan
world's fastest storage systems from
says "Utilizing only 120 cloud compute instances with locally attached
storage, WekaIO completed 1,000 simultaneous software builds compared to 240 on
IBM's high-end FlashSystem 900. The WekaIO software utilized only 5% of the AWS
compute instance resources, leaving 95% available to run customer applications."
an ambitious positioning statement and offers users a glimpse into the kind of
performance they can get by using flash assisted cloud services. Like other
modern SSD fabric software software - "WekaIO eliminates bottlenecks and
storage silos by aggregating local SSDs inside the servers into one logical
pool, which is then presented as a single namespace to the host applications."
Viking offers mid life kicker to SAS SSD array market
July 11, 2017 - Viking
it is shipping the industry's highest capacity 3.5"
dual port UHC-Silo SSD
series has 50TB planar MLC capacity, endurance rating of 1
DWPD for 5 years, idle
power consumption under 10W and active power usage of only 16W.
comments:- with 6Gps SAS rather than 12, and relatively low DWPD this is
aimed at the capacity and power optimized
of the rackmount SSD
In 2012 I speculated that SAS would become the new SATA for
array SSDs once the 2.5"
PCIe SSD market reached a competitive critical mass - because quite frankly
SAS can't compete on latency. Whereas there is a huge ecosystem which is
comfortable with managing arrays of dual port SAS drives. So as long as the
drives are big enough in capacity terms - there will be an appetite for them.
at the datasheet - at first glance - an apparent weakness of the new UHC-Silo
SSDs is the lack of any mention of
protection. But the best place to engineer such protection for enterprise
storage can be at the rack level. So in that respect saving space and cost at
the drive level can be regarded as a prudent
at the capacity claim in
"There is no higher capacity SSD solution
available today than the UHC-Silo SSD" said Hamid Shokrgozar,
President, Viking Technology."...
Seagate demonstrated 60TB
SAS SSD prototypes last
August but if Viking really is shipping 50TB now that suggests that those
wanting to extend the market life of SAS arrays and related software can look in
new controller enables low latency persistent PCIe memory
July 10, 2017 - IP-Maker
the release of its new PCIe Gen 3NVMe reference design based on the Xilinx VC709
evaluation kit. IP-Maker has implemented NVMe management as a full hardware IP,
able to process it in less than 1µS. The company says it has introduced the
new IP to support low latency persistent memories on the PCIe bus.
24 Gb/s SAS and other interesting milestones in Demartek's new
interface comparison guide
Editor:- July 7, 2017 - Demartek has published a new version of
Networking Interface Comparison Guide which includes timelines, speed
ratings and roadmaps for the most popular interfaces used in storage system.
Among other things the SAS section notes that "the first 24 Gb/s SAS
plugfest is expected to occur in 2H 2017."
60,000 wafers scrapped in memory fab
Editor:- July 6,
2017 - Given the current memory shortages it was interesting to see a
report in DIGITIMES
yesterday which said that 60,000 12" wafers had been scrapped recently in a
Taiwan based foundry owned by Micron.
that foundry makes DRAM the context of that wafers number is that it's about
the same as the worldwide wafer starts (for all manufacturers) of 3D nand in a
single month. The cause of the scrappage was corrected in a
Samsung announces more investment in memory fabs
July 5, 2017 - An
in the Guardian - mentions among other things an announcement that
invest nearly $18 billion in its chip business."
comment:- Haha for the very same amount
other lucky buyer will get Toshiba's memory business (and patents).
|SSD news in
July's of yore |
- July 2000 -
launched a 64MB PATA SSD on a chip as a replacement for hard drives in space
constrained and reliability sensitive appliances.
- July 2006 -
it was readying a 4GB PATA flash SSD which would be supported as a cache in
notebooks using the ReadyBoost feature of the Windows Vista operating
- July 2007 -
announced that IBM was saving 1,500 watts per server rack and getting much
faster performance by using 2.5" SATA SSDs in its new BladeCenter instead
of hard drives.
- July 2008 -
To speed up the power-up to boot time Texas Memory Systems
announced that its new ultrafast 512GB 4U RAM SSD used an internal flash
array instead of hard drives from which data was transparently restored to RAM.
- July 2010 -
revealed that its PCIe SSDs had been used to accelerate the production of many
3D movies including :- Alice in Wonderland, Avatar, How to Train a Dragon and
Clash of the Titans.
- July 2012 -
Signs that the SSD industry was getting volume came in an announcement that the
industry's most popular SSD controller - the SandForce range
- was shipping at the rate of over 1 million devices each month.
- July 2013 -
sampling 16nm nand flash memory chips. They were 128Gb MLC. In the next 4 years
which followed the flash industry chose to go up in layers and up in bits per
cell to get more bits per chip rather than down in nm geometries.
- the next box|
| Throughout the
the data storage market we've always expected the capacity of enterprise user
memory systems to be much smaller than the capacity of all the other attached
storage in the same data processing environment. |
new blog on StorageSearch.com
adapted memory systems - asks (among other things) if this will always be
Like many of you - I've been thinking a lot about the
evolution of memory technologies and data architectures in the past year. I
wasn't sure when would be the best time to share my thoughts about this one.
But the timing seems right now. ...read the
there yet? |
| After more than 20 years of writing guides to
the SSD and memory systems market I admit in a new blog on
we there yet? - that when I come to think about it candidly the SSD
industry and my publishing output are both still very much "under
|RAM has changed from being
tied to a physical component to being a virtualized systems software idea - and
the concept of RAM even stretches to a multi-cabinet memory fabric. |
RAM really? - RAM in an SSD context|
|I think it's not too strong
to say that the enterprise PCIe SSD market (as we once knew it) has exploded and
fragmented into many different directions.|
|what's changed in enterprise