The days of
PC domination of memory trends for memory performance, function or usage is
truly a business model of the 90’s. While the PC technology delivered ever-higher productivity
based on processor performance and memory capability, that was only remarkable for
the period when individual business productivity drove the computing sector of
the market.
But as the
Internet connected users at a level that enabled the mobile consumer market,
market conditions underwent a fundamental change. The Internet relies on a broad range of servers to provide
the communication, data, and information access; the consumer does not obsess
on the hardware as long as it is functional, attractive, and delivers the
requirements. The computing
functionality of the PC became relegated to secondary status as the target
application for the development of new memory technologies broadened to provide
equal priority between mobile and server applications.
The impact is that OEMs and end-market applications
now accept a higher degree of product differentiation among the memory
suppliers. Low-power mobile DRAM has
already become an important element of the product differentiation that system
designers demand for consumer mobile products. LPDDR2 and
LPDDR3 DRAM now use a different dedicated process than DRAMs used for PC and
Servers.
Additionally, volatile and non-volatile memory not only share the
system memory space in an ever-growing number of end products, but they also
share much of the capex in the remaining memory manufacturing companies. An extended excess of DRAM
supply has a quarter-to-quarter impact on NAND availability since Samsung,
Micron, and SK hynix have all demonstrated the manufacturing flexibility to
shift fab capacity between DRAM and NAND.
But this shared
manufacturing capability has a major downside—as a result of the general
decline in DRAM demand, both Samsung and Toshiba have now announced plans to
slow their capacity expansion in hope of also slowing any future excess NAND capacity.
During this attempt to
balance the supply and demand, while the industry is impatiently awaiting the
PC market’s resumed growth, NAND SSDs continue to thrive. And in that application, system designers
continue to support a degree of product differentiation among memory
manufacturers’ SSDs that OEMs have never before been willing to accept.
Other intelligence and
insight into the NAND and DRAM market are available in our recently released
quarterly update, OEM System Function Delineates Memory Types.
An interesting aspect of
this weakness in the current memory market is the potential impact that it may
have on another high-level industry topic—the 450mm fab. The history of the memory manufacturers
is one of constant consolidation from what was over 50 suppliers in 1995 down
to the four major memory companies today with complete in-house services of
design, fab R/D, production and test, and worldwide sales/marketing presence
with branded parts. Those four
companies are of course; Samsung, Micron, SK hynix, and Toshiba.
In the winnowing down
process of the past among memory manufacturers, the race to the next fab level
became a life-or-death struggle as any laggards were eventually priced out of
the market. Now we have reached
the point that Samsung and Toshiba are contemplating a slowing in memory
investments to preemptively correct for any potential oversupply, while Micron
is completing its acquisition of Elpida and Hynix is adjusting from its own recent
rescue by SK.
The impact on 450mm fab
introduction is that Samsung is one of the three companies (along with TSMC and
Intel) widely expected to be early adopters of 450mm fab technology. While we have no doubt that Samsung will
be the first memory supplier to convert to 450mm, we believe that the economics
are just not the same game for memory companies. In the first place memory die—and die in support of the
mobile markets in general—are usually much smaller than microprocessors with
onboard cache that are designed for data processing applications. Larger die of course result in a much
less efficient usage of the circular wafer surface. In addition, the trend in the past among memory suppliers
was to rush to the next higher wafer size and lower the manufacturing costs in
order to gain more market share from the competitors who were slower to make
that transition.
However the market dynamics
have changed to reflect Samsung’s current position. As high as Samsung’s market share in memory products is at
the moment, rushing into 450mm in order to squeeze out another competitor at
this point may in fact increase Samsung’s risk by creating too much exposure to
the cyclical nature of the memory markets.
In a recent discussion of
Intel’s recent investment in ASML, one observer commented that, “ Intel
realizes they need 450mm sooner than anyone can give it to them.”
We therefore believe that the
collapse of Elpida and the decisions of two other memory suppliers to delay
their own fab expansion plans may allow Samsung to be more flexible in its timetable
for 450mm. If that is the case, a
heavier burden in the pioneering of 450mm wafers may have now been shifted to
Intel in order to support Intel’s larger die sizes.
