What will replace the PC in the DRAM market?
11/01/2000
 |
Sherry L. Garber, Semico Research Corp., Phoenix, Arizona
It is always safe to predict change in the volatile DRAM market. To forecast the kind of change and its consequences is more challenging.
The overall DRAM outlook is good. Revenue is forecast at a compound annual growth rate of 31%, unit shipments at 13%, and megabit growth at a healthy 61%. The only change will be the lack of a single type of DRAM able to satisfy the diverse applications in need of memory. The market will therefore segment by application and the DRAM vendor will have a much more complex manufacturing job to satisfy the demand.
This will lead to DRAM vendors specializing in specific DRAM types, with only a few DRAM vendors able to support the entire market.
One key force in the development of new DRAM products, reshaping both the IC and electronics markets and changing the demand for memory, is the Internet. While the desktop PC will continue to be a significant consumer of DRAM, it will cede its position to Internet servers, communications, and consumer products.
 Figure 1. Desktop PCs by price point. |
The shifting force of the PC
The desktop PC has been the primary driver for the DRAM industry since the '80s. Whatever DRAM was designed into the x86 platform was what other system designers with non-PC applications had to figure out how to use, since high-volume, inexpensive memory would become the selling commodity.
The market for the desktop PC will continue to grow, although in a moderate 10% range, through 2004. This growth rate, while significant, does not tell the whole story. Change in volume by price point is also important, because price point directly affects the amount of memory consumed.
 Figure 2. Average Mbytes/system by computer segment. |
Computer price points have always been important to the DRAM market — the high end of the desktop market is where new, more expensive DRAM has been introduced, and where the highest-performance, most expensive processor has been implemented. As the cost of DRAM declines with yield improvements, vendor competition, and volume increases, it eventually migrates to the other price points. To illustrate, the >$2000 range is where the Rambus DRAM (RDRAM) was introduced. Double data rate synchronous DRAM (DDR SDRAM) is starting in this area in 4Q00. As the cost of DDR SDRAM declines with the increase in production and number of vendors producing it, the product will be used in lower-cost desktops.
Figure 1 charts the change in shipped volume by price points. In 1996, the >$2000 desktop PC was 57% of the total desktop market. By 2000, this segment will shrink to 27% of the market and is forecast at 18% for 2004, a noteworthy prediction, since the high end has typically had twice the main memory of the low-end desktop PC. By comparison, the <$1000-1500 desktop PC had a 20% market share in 1996, will have 52% by 2000, and is projected to reach 62% by 2004. This segment generally has the least amount of main memory installed.
 Figure 3. DRAM demand by major application (128-megabit equivalent). |
One reason for this trend is that the introduction of new applications requiring a doubling of main memory size, such as Windows95, is just not happening. The high-end PC began shipping with 128 Mbytes in 4Q99 and is shipping with the same amount of main memory in 3Q00. The rise in DRAM prices has also kept main memory sizes static. The tight supply of DRAM that is driving average selling prices higher than a year ago, has caused PC OEMs to freeze the size of main memory until supply increases.
Figure 2 shows memory sizes for the computer segment. Note that desktop and notebook memory sizes are comparable as notebook functions mirror desktop capabilities.
New functions, new products
The convergence of applications will lessen the importance of the desktop PC. Convergence does not mean consolidation, but just the opposite; it enables new functions and capabilities that lead to new products — exactly the role of the Internet. The growing importance of non-PC products is exciting news for the DRAM market; it also brings some new challenges.
Figure 3 presents the consumption of 128-megabit-equivalent DRAMs by market category. The communications segment includes products such as LAN switches, hubs, routers, set top boxes, cellular phones, and central office and transmission equipment. Note that as cellular phones offer more features, the memory involved becomes DRAM.
Consumer products include arcade games, video games, digital TV, Internet appliances, gambling machines, kiosks, smart appliances, and future, yet-to-be-defined interactive devices. Industrial products include robotics, test equipment, industrial controls, and automation.
The category "other" includes military and automotive applications. DRAM is making its way back to the car market as more automotive PCs and expanded automotive entertainment systems are installed.
DRAMs get diversified
Diversification of application demand has increased the number of DRAM types over the last few years. There are currently five major types of DRAM shipping in volume: fast page mode (FPM), extended data out (EDO), SDRAM PC66/100/133, RDRAM, and DDR SDRAM.
The FPM and EDO are in demand for communications equipment, products that have a long life. Since continuity of service is essential, only memory products with long lives are designed into switches, hubs, routers, and central office switches destined to be in operation for decades.
 Figure 4. Worldwide DRAM revenue by type. |
Many production constraints affect the different types of DRAMs in demand this year. RDRAM is available in limited volume production from selected vendors; DDR SDRAM has been shipping in volume since 4Q99, with growing demand as chipsets become available; PC133 SDRAM is available in volume; and FPM and EDO DRAM are also in continuing demand. These availability options set the stage for a segmented DRAM market this year and beyond.
Adding to the complexity of the DRAM market is its dependence on other ICs. The type of DRAM is meaningless unless there is an interface or chipset that allows the memory to talk to the processor. All forecasts in this article take into account the availability of both processors and chipsets or other ASICs in forecasting types and volumes.
EDO DRAM will continue to ship though 2005, even as OEMs phase it out for newer types of DRAM. We predict SDRAM will continue to hold the majority of the market through 2003. The high cost of RDRAM and the limited supply of 800MHz products are hampering adoption in this segment. DDR SDRAM sampling is at a high level; working systems are available; and it appears to be ramping in production and demand. It will be early 2001, however; before there is significant demand for DDR SDRAM due to the lateness of some server chipsets. How fast the market transitions to DDR or another DRAM type is still to be determined.
The DRAM market is evolutionary. After five years of sampling and design-ins, SDRAM finally became the dominant type in 1998. SDRAM has progressed from PC66 to PC100 to PC133, and will likely progress to PC150. FPM and EDO are gradually declining as DRAM vendors raise prices on them to encourage design-in of newer types of DRAM. However, many FPM and EDO DRAMs are used in communication equipment that will be supported for many more years; thus, FPM and EDO DRAM will continue to be in demand well into the middle of this decade. Embedded DRAM is expected to help eliminate the need for some lower-density, older type DRAMs.
There are currently two high-volume densities, 64- and 128-megabit. Normally, the market moves to the higher density with price parity. But that is in markets where the density has quadrupled and all the DRAM vendors have the higher-density product. This is a good example of how important supply from a large number of vendors is to the success of a DRAM product. The allocation mode of the DRAM vendors is raising demand for 64-, 128-, and 256-megabit DRAMs. Production volumes of 256-megabit are available, and following the 256-megabit will be a 512-megabit. It appears that the half-density steps are here to stay. Moving to 512-megabit instead of 1-gigabit defers the need for 300mm wafers and new fabs.
The revenue mix between FPM, EDO, and SDRAM, and RDRAM and DDR SDRAM is shown in Fig. 4. Future types are included in the SDRAM, RDRAM, and DDR SDRAM segments. Many workstation and server manufacturers continue to use FPM and EDO DRAM, but expect to convert to DDR in 2001. FPM/EDO DRAM demand continues for 64-, 16-, and even 4-megabits through the end of the forecast period.
SDRAM will be the transitional product and may lose market share to DDR SDRAM if DDR goes into production faster than anticipated. RDRAM, although it has some market share, will continue to be a niche product. Some of the new DRAM types, such as fast cycle RAM (FCRAM) and virtual channel DRAM (VC DRAM) are also niche products; their success will require that they be designed into at least one high-volume application.
DRAM outlook
As DRAM diversification requirements become more demanding and expensive, a decline in the number of participants in the DRAM module market is expected — dropouts will exceed entrants. Additionally, the fragmentation of the DRAM market will result in companies that specialize in specific types. With lower volumes of a low-margin product, many vendors will find it difficult to weather the market.
Sherry L. Garber is a senior VP at Semico Research Corp., a marketing and engineering research company specializing in the semiconductor industry and related technologies. Semico Research Corp., 7319 N. 16th St. #202, Phoenix, AZ 85020; ph 602/997-0337, fax 602/942-8589, e-mail [email protected].
Further analysis of the DRAM market is available in the report 3Q00 DRAM Market Update: Tight Supply for Second Half. Contact Morry Marshall at ph 602/997-0337 or [email protected].