Interesting comment, Gary J. Goetz; It is a fact that more of the electronics industry is structured the way it is in order to simplify the task of deciding who has to eat scrapped product. In fact, the consequences of bad design are always forced back on the designers' company, in gate array design. I had done several chips before I detected this fact, but it is evident in the sequence of steps foundry makes you go through before they put your design to fab. (For instance LSI Logic.) The Rambus business model allows themselves a certain distance from the fray.
Basically, they can (and did) spec a bus with so little margin for deviations that it just can't work reliably, but they don't get hit with anything other than a stock drop and lost future revenues. If I were the box makers, I would be beyond furious, as they are the ones that are eating product. I don't expect to see a lot of box makers formally dump RDRAM, but I will be surprised to see a lot of them announce products with it in the future.
My own speculation, regards the design problem, is that they just plain ran out of timing margin, as a combination of a lot of parts not quite running to simulation. I expect them to take up underclocking, like they did with the PC100, as a solution. The PC100 ended up underclocked by something like 20%, so you are likely to see RDRAM running at 600MHz, maybe even 700 by the end of the year, but not likely production 800 MHz stuff.
In the mean time, though I know that most people here are only worried about the alligators, there are still a few swamps out there. One is the swamp of alternatives to RDRAM. Everybody knows about most of them. Here are a couple more to consider. Each of these systems provides a very high bandwidth in a very small package, possibly on the same chip as the memory controller:
Multiple chips in one BGA package: #reply-11380493
Stacked chips are making progress in cell phones: #reply-11380477
-- Carl |
