Tom Here is another good article. If you are going to buy a PC in the near future make sure it is PC/100 compliant and not PC/66.PC/100 compliance the key in'98
Farhad Tabrizi
In some quarters, the buzz about advanced DRAMs centers on Rambus vs. double-data-rate (DDR) DRAMs. However, the more practical OEM purchasing managers - those tuned to the near term - are focusing on how PC/100-compliant DRAMs will roll out this year. The PC/100 standard, initiated by Intel Corp. to support next-generation PC systems, established the parameters for the next-generation memory bus operating at 100 MHz.
The PC/100-compliant DRAM will become the prevalent device through 1998 and into 1999. By the end of this year, more than 50% of DRAM consumption will be PC/100. When this happens, the other DRAM device with 50% of the market - the 66-MHz SDRAM - will become a premium part. If purchasing managers don't already know, they will soon learn that PC/100 DRAMs are backward-compatible with 66-MHz SDRAMs.
Thus, it will make good sense to maintain one inventory, buying only PC/100-compliant devices and using them in place of 66-MHz or even non-PC/100-compliant, 100-MHz SDRAMs. Also, since PC/100 DRAMs will represent 50% of the volume by the end of this year, there will not be a premium for PC/100 DRAMs in 1999. Given this scenario, it will be wise for systems designers to begin planning now to exploit these emerging market conditions and convert existing 66-MHz-based designs to the higher-performance PC/100 DRAMs without incurring major design-cost penalties.
However, prior to launching PC/100-based designs, PC OEMs should understand several major design issues. The most prevalent of these deals with industry confusion over what the clock speed of an SDRAM means in a real memory application. A PC/100-compliant SDRAM is specified at 100 MHz, but some DRAM vendors already have 100-MHz and 125-MHz SDRAMs on the market. The PC/100 standard represents a significant jump in DRAM performance not reflected by the 100-MHz performance rating. The standard requires timing signals to be significantly faster than standard SDRAM specifications to allow the entire memory bus to operate at 100 MHz, not just the SDRAM device itself.
For instance, setup, hold, and clock-to-output timings are some of the more critical aspects of the current PC/66 and the new PC/100 SDRAM specifications. Clock-to-data delay in the PC/66 specification calls for 9 ns, compared with 6 ns in the PC/100. Setup time in the PC/66 specification is 3 ns, as opposed to 2 ns in the PC/100; and hold time after clock in the PC/66 is 1.5 ns, versus 1 ns in the PC/100.
Therefore, even if an SDRAM operates at 100 MHz or at 125 MHz, it may still not comply with PC/100 specifications. As the industry moves more and more into PC/100 designs, it is important for systems makers to closely analyze DRAM vendors' offerings to ensure PC/100 compliance.
Demand for PC/100 DRAMs is expected to be high early this year. On the supply side, this particular type of DRAM will be difficult to design and build since it is a tightly designed chip. To make this DRAM operate at its specified speed and parameters, there is room for only the smallest of design margins. Consequently, DRAM vendors targeting PC/100-compliant parts will probably struggle with their chip designs.
Aside from design issues, there is a lack of general understanding of how to use SDRAMs in a system design. This is especially true of such newcomers as consumer electronics applications, such as set-top-box designs. Unfortunately, in these instances an SDRAM is regarded and implemented as a garden-variety DRAM with little thought given to the implications it has on other aspects of the design. We're seeing such problems emerge as more consumer systems customers go from EDO DRAM to SDRAMs. Some system designers underestimate the major features SDRAMs provide.
Systems makers are now qualifying the 16-Mbit version of the PC/100-compliant SDRAM, and 64-Mbit versions will be qualified early this year. However, we expect the 64-Mbit SDRAM to be the more popular of the two, and, consequently, it will significantly cut into 16-Mbit demand. This makes sense because the number of memory slots in a PC are continually being reduced. Most PC makers today are shipping 32 Mbytes as the base memory in the form of a 4x64 module. This is an entry point at which a 64-Mbit SDRAM can be effective; hence, it can take off in the PC market as 4 or 8x64 modules.
From our perspective, workstation and server DRAM usage this year will evolve from EDO to SDRAM. The main reason is that these systems use very wide 512-bit or 1,024-bit bus architectures, and because EDO DRAM now comes with a premium, whereas SDRAMs are the dominant, low-cost technology. PC/100 DRAMs will be used from high- to low-end PCs throughout the year.
In particular, sub-$1,000 PC systems will rely on the dominant technology for main memory, which will be PC/100-compliant DRAMs. However, due to its costly nature and required royalty payments, Rambus is not expected to be a favorite for low-end PCs. Yet, workstation and server makers are keeping close tabs on Rambus since Intel is endorsing the DRAM technology. The big concern is whether Rambus will become 80% of DRAM production. However, what these OEMs fail to realize is that the Rambus DRAM (RDRAM) manufacturer cannot lower the cost enough for this technology to be a main memory commodity.
The average budget for a system's main memory is about 7% to 10% of the total cost. For instance, a $1,000 PC system should have about $70 worth of memory. This year, a $1,000 PC will have a 64-Mbyte memory, but it cannot be a Rambus-based system because it will cost considerably more than the allocated $70.
Because of power consumption concerns, portable PCs will remain with 66-MHz SDRAMs or EDO DRAM through this year and into 1999, when DDR DRAMs will be used for PC applications. But in that same time frame, DDR DRAMs are expected to share PC applications with RDRAM because of Intel's endorsement. However, RDRAM won't be a player in the server market since it is limited in terms of total Mbytes per system. The Rambus architecture doesn't lend itself to big memory systems. That's why it may be appropriate for high-priced PCs with small memory systems.
The graphics DRAM market will evolve from SDRAMs to 200-MHz DDR in 1999 to 300-MHz DDR in 2001. Interestingly enough, we see the main- and graphics-memory segments being divided, with distinct differentiation between the two. Graphics-memory applications will be x32 devices, for instance, 512-Kx32 or 1-Mx32 or 2-Mx32. However, main memory will remain at x4, x8, and x16.
By 2000, following DDR, we believe the 256-Mbit SyncLink DRAM (SLDRAM) will become the dominant main-memory technology for workstation, server, PC, and even portable-PC applications. In effect, SLDRAM will become a DDR upgrade since the SLDRAM interface is similar to DDR. Hence, the open standard will move to DDR and then migrate to SLDRAM.
-Farhad Tabrizi is director of strategic marketing for DRAMs at Hyundai Electronics America, San Jose. |
