Memory Blues: No End In Sight
By Jim Handy
May 27, 1998, TechWeb News
(my emphasis added-JS)
Probably the single most important question on the minds of memory manufacturers and users is the timing of the next change in each of the memory markets. DRAM prices have plummeted: when will they regain their ground? Flash memory has become fiercely competitive: will this change? The SRAM market bears only the vaguest resemblance to its 1995 self and is poised in 1998 to go through a weeding-out process: when will the bloodbath end? The short answer: certainly not in the near future.
Dataquest's compilation of the 1997 MOS memory market showed a 19 percent revenue decline from 1996 hot on the heels of a 36 percent decline from 1995 to 1996; the result is an $8 billion revenue drop to a total market of $31 billion. The only technology that did not decline was our "Other MOS Memory" category, a grab bag composed of such esoterica as FIFOs and CAMs that increased by 14 percent to a modest $655 million from a more modest $577 million in 1996. DRAM, SRAM, flash, EPROM, E2PROM and mask ROM all declined by percentages ranging from 3 percent to 37 percent, combining to make 1997 one of the most difficult years in the history of memory. Even the flash memory market, which had done nothing but grow in its entire history, fell by 3.2 percent. In light of this, will there be more bad news or have the market's troubles run their course?
The figure shows Dataquest's revised memory forecast, which ties into our overall semiconductor forecast. We have downgraded our outlooks for both the short and long terms mostly because of DRAM. It makes up nearly two-thirds of the market, and the significant factors influencing DRAM sales are affecting the memory market to nearly the same extent.
In the long term the DRAM market will be the major swing factor, causing the entire memory market to zoom up to $77 billion in 2001 before dropping to $61 billion in 2002. In the short term we see a very small increase in 1998 DRAM revenues of 1 percent over sales in 1997, reflecting persistent overcapacity and its effects on pricing coupled with an extremely aggressive shift by DRAM manufacturers from the 16-Mbit density to 64 Mbits, a move that will increase manufacturing efficiencies of this overcapacity and make the bad situation endure longer despite relatively healthy bit growth this year.
But Dataquest expects to see growth return in each segment-DRAM, SRAM and nonvolatile-over the long term, exceeding 10 percent. However, a conversion from EPROM to flash should cause EPROM revenues to decline at an average annual rate of more than 20 percent in favor of greater flash growth.
As we predicted for more than two years, 1997 became the least profitable DRAM year of at least the past five and possibly the entire last decade. Although Korean suppliers in February 1997 started to decrease their supply of 16-Mbit DRAMs, this had little effect outside the spot market.
Dataquest was a bit surprised to find that Taiwanese and Korean DRAM manufacturers did not stop their capital expansion in response to 1997's near-zero profitability. Generally, DRAM manufacturers stop spending when profits evaporate, but not this time. As a result of continued capital spending through the end of 1997, we have had to revise our forecast to reflect that our projected mid-1999 supply/demand balance would be delayed until the first half of 2000. Our revised forecast calls for a peak in 2001, with the subsequent downdraft occurring in 2002.
The DRAM Supply/Demand Quarterly Statistics report, in which we compare supply projections from supplier surveys against a demand model built from Dataquest/Gartner's extensive forecasts of consumption markets, indicates an oversupply this year and next. The result is that we expect to see important fab closures this year to compensate for the imbalance.
There is an estimated 20 percent to 25 percent overcapacity today. After continued capacity investment and improved manufacturing efficiencies, we now expect not to be passing over equilibrium until the middle of 1999 and entering undercapacity in 2000. Tight supply should last 15 to 24 months before overcapacity re-emerges later in 2001. This drives a cyclical downturn in the year 2002 in Dataquest's five-year DRAM forecast.
Future interfaces
Just as 1997 was the year that the SDRAM started to seriously displace EDO, 1998 will be the year for PC100 SDRAM to replace the 66-MHz part. The only question still to be answered is when Rambus will take hold. Dataquest continues to expect Rambus' penetration of the PC market to be postponed until 2001, with only minor penetration starting with the introduction of Intel's first Rambus-based core-logic chip set in 1999. Once Rambus moves into the PC market, though, it can be expected to gain ground as fast as did EDO and SDRAM in the past. Since the PC market should continue to account for about three-quarters of all DRAM sales, this will drive other DRAM users to Rambus so that they can realize cost savings over less widely produced DRAMs.
Although Intel recently has been mentioning support for Double Data Rate (DDR) in the graphics arena, the company has not mentioned this as a solution for the main memory space and continues to leave SLDRAM out of its memory road maps. These actions imply that Intel will have no near-term support for SLDRAM and only limited support of DDR, which could cause these two solutions to be relegated to a higher-priced niche rather than the PC mass market.
SRAMs
Worldwide SRAM revenue should total $4.7 billion in 1998, an 18 percent growth rate over 1997. This comes after two years of negative growth and reflects Dataquest's expectation that the worst is over for the SRAM market. Dataquest expects a similar rate of SRAM revenue growth in the long term. SRAM revenue will expand at a 17 percent rate from 1997 to 2002, based largely upon stable growth in all major application markets.
The SRAM market is still oversubscribed, but conversion of cellular phones from analog to digital has increased SRAM usage in these phones dramatically. There is a big difference between this market and the PC market, the last big SRAM stronghold. Cell-phone manufacturers are much more picky about their supplier base, as they use parts from top-tier suppliers and shy away from small fabless companies.
We still expect to see attrition in this market since many companies that thrived on an open market for PC cache chips will shortly be excluded from this previously highly penetrable market and will not be able to participate in the cellular-telephone market. Widespread and diverse opportunities exist in other telecommunications markets, such as network file servers and routers. However, it seems that no two of these systems use the same SRAM, making it difficult to participate without extremely good vendor-purchaser relationships.
The four nonvolatile memory technologies reflect differing market characteristics, although at times they compete for applications. Flash memory has seen the first year of negative growth in the technology's history, due to a somewhat sudden growth in the number of competitors in the market. This has taken its toll on EPROMs since they compete against flash on cost. A wide range of applications uses E2PROMs, among them industrial, telecom and consumer systems. Diversified applications mean a bright long-term outlook for the E2PROM market. In contrast, the mask ROM market depends to a great degree on video-game demand, and the shift away from ROM cartridges toward CD-ROMs in the past few years has taken a toll on this market.
Dataquest expects each to perform differently in the forecast window. Flash should restart its growth and benefit over the forecast horizon from the continued emergence of low-voltage, handheld systems for telecom and computing. Dataquest continues to expect flash to enjoy high growth in the long term. Driven by newly emerging applications, flash-memory revenue should have an 18 percent cumulative annual growth rate between 1997 and 2002, and worldwide revenue should be nearly $6.4 billion by 1999. The flash-memory market has moved away from high profitability and toward more competitive pricing as many competitors have joined the fold; however, unit growth remains strong. This new lower pricing has acted to enable the use of flash memory in a diverse array of new applications.
Our outlook for the EPROM market changed recently due to the flash oversupply. The EPROM market will continue to erode significantly as it yields to flash memory, gradually declining to $200 million by 2002, the end of this forecast's time frame.
E2PROM market revenue in 1997 rose above that of EPROM for the first time in history. The mask ROM market should be relatively flat through the forecast period. Worldwide E2PROM revenue should exceed $1 billion again in 1998, as it did in 1996, growing to nearly $2 billion by the end of our forecast window.
ROM price per bit will remain about one-fourth that of flash memory through the forecast period. Although a quick advance in either Toshiba-Samsung's NAND or Intel's multilevel-cell flash technology could alter this scenario, current trends signal that flash will be unlikely to displace ROM if price per bit is the major consideration.
Worldwide ROM revenue should total a little more than $800 million in 1998, a 21 percent decline from sales in 1997. Dataquest expects flat sales of this mature technology over the long term, with unit shipments also staying essentially flat. ROM revenue should decline at an average rate of about 3 percent between 1997 and 2002.
Dataquest's view
There's a lot of doom and gloom in nearly all sectors of the memory market after a year in which the revenues of nearly every technology shrank. We at Dataquest don't predict any boom for the near term but anticipate an industry cycle in 2001 driven by a lack of capital spending starting early this year. DRAM will have a swing up in 2001 followed by a compensating correction in 2002, while SRAM will change its dynamic but will continue to grow. Nonvolatile memories are already significantly shifting away from EPROM and toward flash, which is now being supplied by a far larger and more competitive contingent. E2PROMs have become a far more significant market, while mask ROMs will grow along a more modest trend than they enjoyed in the past simply by losing the significant boost in sales that they enjoyed from video games starting in the mid 1980s. Still, the market will grow to $77 billion in 2001, nearly half again as large as the booming 1995 market.
-Jim Handy is a senior analyst at Dataquest Memories Worldwide Service (San Jose, Calif.). Dataquest's Ron Bohn, Bruce Bonner, Evelyn Cronin, Clark Fuhs, Richard Gordon and George Iwanyc contributed to this report.
Copyright r 1998 CMP Media Inc.
|
