In his blog Greg Papadopoulos talks about misconceptions of Moore's law. I'm really glad that he wrote down a description of what Moore's law really means as people mistating it bugs me also. Moore's law means that we have more transistors to work with, not that we'll use them better. He argues (deftly I believe) that these will be used for more SoC type designs, increasing speed through integration. We have to remember that the first ICs were designed to reduce the wiring between transistors, not to increase clock speed. Removing system buses is the ultimate expansion of integration.

One part of his entry I found particularly interesting (emphasis added):

...microsystems: my word for the just-starting revolution of server-on-a-chip. What's that? Pretty much what it sounds like. Almost the entire server (sans DRAM) is reduced to a single chip (or a small number of co-designed ones, just as the first micros often had an outboard MMU and/or FPU). These microsystems directly connect to DRAM and to very high speed serialized I/O that are converted to either packet or coherent-memory style network connections.

In his version of a system, the DRAM and the processing will still be distinct elements. While this makes sense from the perspective of the process required to make the two types of chips, it will start to become the bottleneck of any system. When access to your network card is more efficient than your system memory, there is a problem. That leads me to say that the next major revolution in semiconductors will be the ability to integrate large amounts of low power memory on the same die as processing elements.

One solution for this is MRAM. The article I linked to said in 2003 that they would be in production by 2005... so, I think there are still some technical hurtles. Last year, Freescale was sampling 4MB modules, but that's not yet enough to make it truly exciting.

What ever the solution is, it will have to allow for hundreds of megabytes of memory while still allowing for reasonable cost chips. The densities of DRAM have expanded at an impressive speed, no one can imagine having less than 32 MB today. This is an exciting opportunity, now if I only had an idea of how to solve the problem.


posted Nov 18, 2005 | permanent link