Rambus technology versus SDRAM and embedded DRAM............................
Ronnie, I have some responses to your comments:
<<I think the risk of Rambus lies in that they can not control their own market. In other words, their revenue comes from other vendors who pay them loyolty when they use the Rambus interface. >>
True, but isn't this risk moderated significantly by Intel's support of the Rambus technology? Intel's support alters the entire terrain.
<<There are other alternative to their technolgy in that SDRAM interface will be probably easier to design and implement in general. In the future the Pincount will be a non-issue for chips due to the availability of BGA(ball-grid array).>>
SDRAM is easier to design and interface with than Rambus. However, companies are very creative when it comes to interfacing with Intel CPUs. If you don't interface to Intel, you don't sell memory chips in the PC market. The design and interface problems are not major sticking points.
<<And for this market, the embedded DRAM technology is certainly a great threat to Rambus.
Since once the DRAM is embedded on the chip, there is no need for rambus interface anymore, the controller can directly talk to DRAM with as many bits as they wish.>>
Embedded DRAM is a faster interface than non-embedded DRAM. However, there are significant barriers to embedding DRAM on the same die as a Pentium. First, the Pentium is a big chip. How much bigger are you going to make it -- if you make the die larger you will have to deal with problems such as higher power consumption, more heat, and lower yield in the manufacturing process.
Speaking of the manufacturing process, the process for making CPUs is very different from the process for making DRAM. DRAM requires lots of three-dimensional structures such as trench capacitors, so it requires processes such as deep etching, deposition of layers on top of that, and etchback. It uses a minimum number of metal layers to keep costs down. On the other hand, CPUs require lots of metal layers for all of the different interconnects. DRAM transistors are designed for low leakage and CPU transistors are designed for high speed. Lots of problems arise when you try to combine the different processing technologies on the same wafer. Costs increase because of the complex steps, and you lose some efficiency because the processes are not as fine-tuned for one type of technology.
Embedded DRAM may work better for microcontrollers, graphics chips, MPEG decoders, and other applications in which the CPU is specialized and not as large and complex as the CPUs used in personal computers. Also, these applications don't require expansion memory, which is difficult to do with embedded DRAM.
I think that the long term prospects for RMBS depends more on Intel's support than it does on competitive technology. Presumably, one of the main reasons (although not the only one) Intel supports Rambus is that Intel evaluated and believes in the Rambus architecture. I suspect that there are (anti) competitive reasons as well.
Your comments are welcome, as a contrarian view is very helpful to all of us. |
