Hi Dan3; Re: Ultimately, Rambus could prove cheaper than 64 bit SDRAM and its evolutionary followers due to Rambus's efficient use of board real estate... I hadn't thought of analyzing it that way, but it makes sense, and we can take the analysis another few steps.
Boards cost well under 25 cents per square inch even for multilayer stuff in relatively small volume. So how much PCB does rambus actually save per chip or system? Assuming that the chip count stays the same, (i.e. a system who's minimum chip count is determined by memory size requirement rather than bandwidth requirement, and assuming that rambus memories are approximately the same number of bits as alternatives), we know that the extra area rambus puts onto silicon is around 5 to 10% (if I remember correctly).
So silicon has to get down in cost to make the extra silicon match the saved circuit board. Since the cost of PCB is more or less stable, this tells us how much silicon has to go down in price. This gives us the approximate price of the memory chip itself, and therefore, by applying the royalty figure, RMBS's patent income. (All this assumes that silicon ever gets that cheap.)
As densities increase, the extra cost of putting rambus interfaces in should decrease. But memory densities increase at a reasonably predictable rate. So we can then predict when the cheaper system cost will cross over to rambus...
By the way, I see rambus as a cost effective solution only for systems with small memories, not for the high end computers. The Sony deal was a great example of the rambus advantage, not workstations or servers.
-- Carl |
