TECHNOLOGY DICTATES FATE OF MEMORIES
May 30, 2000
--------------------------------------------------------------------------------
Jim Handy
Since the memory market collapsed in early 1996, everybody has been eagerly hoping for a recovery. The question now is: "Are we there yet?" The answer depends on which market you participate in-DRAM, SRAM or flash. Those three markets make up the bulk of the semiconductor memory market, which accounts for 20 percent to 30 percent of all semiconductor sales.
Last year was a year of substantial changes in the memory market. Flash memory led the pack by recovering from its oversupply in February and increasing its revenues 71 percent from a stagnant $2.8 billion in 1996, '97 and '98, to a whopping $4.5 billion in 1999. Static RAM started surging in May, and although the market experienced a growth of 22 percent in 1999, it was surpassed by the strong growth in flash, which is now the second-largest memory market after dynamic RAM.
Speaking of DRAM, what happened there? Well, 1999 was the first year of DRAM revenue growth since 1995. For once, prices didn't fall as fast as unit growth increased. In spite of a 46 percent revenue growth from $15 billion in 1998 to $22 billion in 1999, DRAM manufacturers still are not dancing in the streets. Why not? The answer is simple: They are not yet profitable. DRAM prices fell so far during the price drops of 1996 to 1998 that production costs are having a very tough time catching up. Dataquest estimates that the sum of all DRAM manufacturers' losses in 1999 totaled $5 billion. This was not enough, though, to dissuade some manufacturers from adding capacity. Samsung Electronics Corp. opened its new fab nine and broke ground on its fab 10. Micron Technology Inc. (Boise, Idaho) ramped production at the fabs the company purchased from Texas Instruments Inc. (Dallas). Taiwanese investors continued to boldly add capacity in anticipation of taking market share from the rest of the world suppliers. Still, Dataquest believes that the capacity that has been added and the new commitments just named are insufficient to assure that demand will be satisfied for the rest of this year.
We believe that 2000 will end with a capacity/demand balance, turning into an undersupply in 2001.
This is reflected in the DRAM portion of our memory forecast. DRAM prices tend to do either of two things over time: They drop, or they remain flat. During shortages, prices remain flat at a level much higher than manufacturing cost, and the market is profitable. DRAM average price per megabyte has been essentially flat since May 1998. Although this is the first step toward recovery, we find that the average DRAM manufacturer is still not profitable, although the gap between cost and price is narrowing. We should expect the market to follow prior trends, and for prices to remain flat for the next 10 quarters before dropping again.
DRAM megabyte consumption tends to grow at a fixed rate of about 70 percent per year. The past two years have seen bit growth in the 90 percent range, and this implies that a correction may be in the offing, where we could see bit growth slow down. We have run exhaustive tests of the relationship between price drops and bit consumption, and find that the demand elasticity of the DRAM market is tiny enough to be difficult to measure.
In fact, the average main memory size of the PC seems to increase at a very steady rate of about 45 percent per year during times of either high or low prices. This number, when coupled with an average PC unit growth rate of 15 percent, drives a Megabyte growth of close to the 70 percent historical average.
If we couple a predictable bit growth of 70 percent with a flat price per megabyte, we come up with a very strong argument that DRAM revenues should grow 70 percent per year during the 2000-2002 shortage. Our forecast shows a slightly more conservative view, with prices dropping slightly, rather than remaining flat, and with bit growth slightly lower than normal, making up for the high bit growth of the past two years. This gives us revenue growth in 2000 of 58 percent, continuing to 61 percent in 2001, easing to 29 percent in 2002, with the market peaking at revenues of $76 billion. In 2003 Dataquest expects the start of the next DRAM cycle, with a 32 percent downturn to $52 billion, falling another 8 percent in 2004 to $48 billion.
Flash shortages
Flash, though a new market, is displaying a lot of the same mechanisms seen with DRAMs. For three years flash revenues were stable despite bit growth of more than 100 percent. The bit growth was offset by a price per megabyte decline of 50 percent per year. Now, with cellular handset unit shipments growing at a rate far exceeding even the most optimistic of 1999's forecasts, we find the flash memory market in a dire shortage situation. As I write this, Sharp Corp. has just announced an across-the-board flash-memory price hike of 30 percent. Though Dataquest doesn't forecast that such price hikes will be the norm (Sharp accounted for only 8 percent of the 1999 flash market), we do forecast a slight rise in flash pricing in future years.
With the high bit growth that the flash market experiences, we expect to see revenues ramp from $7 billion in 1999 to $13 billion in 2002. The market will once again get stuck at the $12 billion level in 2003 and 2004 as an oversupply takes its toll.
Finally, what about static RAM? The SRAM market, while poised for growth, will be outstripped by the more commodity-oriented behavior of the dynamic RAM and flash markets. This means that the SRAM yo-yo will neither rise nor sink as dramatically as flash or DRAM. SRAM will experience more moderate rates of growth and decline than will the leading two memory markets.
Dataquest forecasts that the SRAM market will grow at rates lower than 20 percent during the boom years and will shrink less than 10 percent during our forecast down years. This will flatten the market's swings, but will still allow the market to grow to a record level of nearly $7 billion by 2002.
Combine SRAM, DRAM and nonvolatile memory sales, and overall memory revenues nearly reach $100 billion in 2002.
Purchasers of DRAM, flash and SRAM should expect to experience allocation and shortages, with the smallest players feeling the squeeze the most. Fabless SRAM and flash manufacturers might also feel a pinch as foundries become booked to capacity. Other memory manufacturers will do well in the near term to plan conservatively, knowing that this boom will definitely occur, but that it will only last until the end of 2002.
Oh, and don't expect to see another 1995. That will never happen again.
JIM HANDY IS DIRECTOR OF MEMORY SEMICONDUCTOR RESEARCH AT DATAQUEST/GARTNER GROUP (SAN JOSE, CALIF.)
eetimes.com
Copyright c 2000 CMP Media Inc.
By Jim Handy |
