Asian Memory Caper, Or, Are Your DRAMs Authentic?
Relatively recently, some Asian DRAM memory manufacturers
have been selling untested (“UTT”) DRAMs;
and the practice
seems to be getting traction at the lower portion of the module business. This
is being done by mostly Taiwanese DRAM makers, who undercutting the teir-1 guys
by selling un-tested, and un-marked, parts. The un-tested yield is high enough
that Asian houses are putting them straight into modules and selling them. Any
fall-out is re-worked, or returned by the customer and replaced. Apparently the
net cost under such a business model is sill cheaper than testing the IC’s.
I equate mis-marking of
DRAM units with counterfeiting. (I see this as a logical extension of all the
sub-$200 Rolexes being advertised in SPAM. When they stop one guy doing it,
it shows he was having some success and a half-dozen will pop up around the next
corner.) It risks driving everybody’s price down to the lowest-price Asian house
over the course of a couple quarters. Selling UTT parts saves on test costs,
which, as I’ve heard for Hynix for a 256Mb DRAM was/is around $0.25 per unit;
and from what I can tell, this appears to be at the low end.
All suppliers have
the capability to churn out top-notch material. It’s just a matter of how much
do they want to spend on manufacturing. (And money is cheap!) The “innovation”
here is identifying Tier-I companies that spend too much on testing. An untested
Tier-II DRAM is close enough in quality that the savings in test cost can
generate a decisive cost difference.
Winbond is one
Taiwanese chip company rumored to be producing a relatively small quantity of
UTT chips. But chatter in the industry suggests that Winbond may increase its
UTT output to two to three million 256Mbit-equivalent units over the next few
months. Winbond’s UTT chips are not labeled with the Winbond logo.
I can’t really
quantify the scope of the problem and so haven’t figured out how to invest
around this knowledge but it seems like there ought to be a couple companies who
will take at least a “marginal” hit from it.
Selling UTT
Parts is a Relatively New Trend
Selling parts that
are leading-edge, mainstream, and unmarked is new within the last few
months. Older-technology memory parts have been sold “blank” for at least a
couple years and sold to the private label market. The original purpose was to
make it easier for a company to put its own logo on them, which is what most
companies do with blanks (and then they test them!). As for timing of producing
UTT parts — whether it is seasonal or whatever – I think the practice is still
too new to be pinned down. My impression is customers who would accept such
untested material would want it all the time.
Selling packaged
but unmarked parts is not done by the Tier-I DRAM houses. (That I know with
confidence.) However, since the parts are unmarked, nobody knows for sure who is
supplying them. The rumors suggest it is mainly Taiwan fabs (including Winbond)
selling “blanks.” If/when the parts are mis-marked, it is after the fab sells
them blank. The big companies’ hands are most likely clean. (Risking a $1
billion fab facility to make a few extra pennies’ margin on a batch of
wafers would be really, really stupid.)
The innovation here
is that a strong design can make it cheaper to re-work modules (granted, at
Chinese labor rates) than pay for the testing to screen out failures. As soon as
the first Tier-I guy embraces this solution, there will be a revolution. Margins
could actually go up, since test cost is usually passed thru at cost.
Theoretically, the
practice of shipping UTT parts is all legit. There is nothing illegal or immoral
about shipping untested blanks; many UTT parts are sent to companies that put
their own label on them as Private Label parts. However making UTT parts is,
like many other manufacturing practices, open to abuse. The trick is that it’s
impossible to determine what percentage is being abused. The way the channel is
behaving, it seems like it’s not a trivial percentage
The really scary aspect of this practice is that some places in Asia are
apparently adding counterfeit tier-1 DRAM logos onto the IC’s (integrated
circuits). This could cause some confusion in the market over branding and
quality. Even without mis-marking in play, this approach creates incremental
price pressure. I haven’t attempted to quantify the scope of the problem, but it
seems like there ought to be a couple DRAM companies who will take at least a
“marginal” hit from it.
My understanding,
now, is the channel is filled by modules marked as though they were from tier-1
DRAM suppliers but in reality they are UTT parts from Asia. (When I say they
channel, I’m referring to what might be thought of as second-tier, maybe even
third-tier, module and systems houses. Their models are similar to the big guys,
although their scope, and profit margin, tends to be smaller.) The first
question after agreeing on the price is, “How do you want them marked?”
The major risk as I
see it is a batch of modules gets into a major user (think IBM, H-P, and / or
Dell) and fails (probably in Asia). The user goes publicly ballistic over the
combination of faulty material and the supplier’s inability to control the
quality of its material. The press runs with it and the unlucky DRAM supplier’s
stock gets hammered. Some time afterwards, it emerges that all the DRAM
suppliers have this risk and then they all go down.
There are some risks to the semiconductor business if this practice increases:
- A major OEM finding this
quality is sufficient and forcing a whole new cost regime on their
competitors. A parallel concern is that Dell, or whoever, does some modeling
and figures out, with a couple cheap QA steps added, that this really is a
lower-cost model. Then it gets really “interesting”; and / or - Somebody selling un-tested,
mis-marked parts as though they were fully-tested, original Tier-I material.
Does this practice
of shipping UTT chips spread from DRAM into CPUs and other components? Well,
Intel’s 85% share of the installed base of around 200 million PCs is “only” 170
million pieces (i.e., CPUs). Whereas DRAM ships a quantity of around 12 times or
more units at maybe 5% or less of the CPU’s margin. So, the biggest bang for the
test buck is in DRAMs. Low-margin parts will be the obvious candidates going
forward…
It’s very difficult
/(as yet, impractical) to quantify the scope of the problem and so quite
candidly, I haven’t yet figured out how to invest around this knowledge. But it
seems like there ought to be a couple companies who could take at least a
“marginal” hit from it.
I have been looking
at DRAM-maker Micron for the last few weeks, but for different reasons than this
UTT trend. It’s still a bit too early to buy MU until the risk / reward makes
sense.Â
Technically, MU
looks like it might be churning for a leg lower. It sold off after a ‘good’
earnings report too, which tells me to wait for the next leg lower and a break
of 10 beans. Fundamentally, obviously DRAM pricing is awful, but at the margin
I’m not sure pricing will get overly worse from these levels. I would probably
wait until we hit late second quarter to start to build a position; then we
should start to see some seasonal firmness in DRAM pricing starting in late July
/ early August. So, all things being equal I’d look to add MU late in the second
quarter.
Melanie
Hollands
melaniehollands@yahoo.com