Hi Mile,
a PS to my previous mail.
AMD's K6-3 will not support RDRAM. RDRAM is scheduled to come with the K7, but even then AMD has announced a parallel version with SDRAM.
I have a lot of respect for AMD for their technical prowess, but when it comes to execution they are a disaster.
The problems causing the 800MHz RMBS delay, although not pinpointed to as far as I am aware, are most certainly impedance and capacitance related due to the frequencies involved.
800MHz is RF, and RF is a black art and as such goes way beyond the know how of even the most experienced digital designers. To give you a comparison, the much respected GSM phone network works at 960MHz!!!
These problems are not insolluable, but they will not easy. RAMBUS, INTEL, AMD will have to hire teams of RF specialists to get it on the tracks.
This also explains the difficulties which are being experienced in the testing. Capacitance changes the response of a high speed circuit significantly. Capacitance changes (relatively) depending on where you measure it: In other words you get different readings measuring it at 1 or 2 or 4mm distances.
Due to these type of problems, RF devices are not dynamically tested on the wafer, but only after packaging. This obviously is detrimental to yield.
Having said all this gives you a clue in what field I worked in 16 years.
It is not my intention to sound negative, but the problems to still be overcome are probably significant. These type of problems were probably not apparent to RAMBUS themselves at the chip level, but will arise when attempting to put the packaged IC's down on a board and link them to the neighboring circuitry.
Sorry I have rambled on a bit, but I am starting to realise that RMBS is probably not as far advanced as I had all along assumed.
Alan |
