Is there two-transistor DRAM in your future PC?
One of the biggest problems with modern personal computer multi-core systems is that many applications are not optimized to use up the available processing units to as busy a level as possible so that the user gets the most bang for their buck. In fact, many users who have been early adopters of quad core systems report that their current applications are running even slower on their new quads than they were on their earlier dual or even single core systems. The reason is fairly simple. A single-threaded application running on a 3 GHz dual will run on one processor at the full 3GHz while the OS is running on the other processor at the same speed. When you apply the same process to a quad core, the OS will run on one core, the single-threaded application will run on the next core, and the other two cores will do pretty well nothing at all. But why do they run slower? Most quad cores run at a speed far below their dual core brethren. So while a 3GHz dual is fairly mainstream, a 3GHz quad will set you back well into four figure$$$.
In order to speed things up for all CPUs, regardless of number of cores, Intel has recently announced a technology that could feasibly multiply the available bandwidth within a processor itself in order to increase data throughput dramatically.
SRAM is the acronym for Static Random Access Memory, and is pronounced "Ess-Ram". SRAM is a type of memory that is currently faster and more reliable than the more common DRAM (Dynamic Random Access Memory ). The Static part of the nomenclature is derived from the fact that the data which resides in it exists in a static or steady format, and doesn't need to be constantly refreshed as DRAM requires. To store one bit of information approximately six transistors are utilized, thus it is more expensive than DRAM when it is analyzed in terms of cost per bit, since only one transistor and one capacitor is used to store a bit of data.
It turns out that this development was not widely heralded by Intel. A relatively low-level team at Intel research was able to shrink down the size of DRAM (dynamic RAM) cells to only two transistors and eliminated the capacitors entirely, thus no longer having to constantly refresh the memory contents. This development holds the key to being able to eliminate pricey and complex SRAM (static RAM) from a PC and replace them entirely with DRAM. SRAM requires six transistors per stored bit, so replacing it with a two transistor DRAM would immediately increase memory capacity threefold.
Intel has stated that under a physical clock of 2 GHz in a 65 nm manufacturing process, the two transistor DRAM could reach a bandwidth of 128 GB per second which is approximately 10 times greater than is available currently. The potential exists to bring the clock to the level of some of Intel's fastest Extreme processors which would then achieve a bandwidth of well over 200GB per second. To bring that amount into perspective it represents a maximum transfer rate which would move all the data in a filled-up large common PC hard drive in a couple of seconds. That's fast!
Since 45 nm is currently being utilized in Intel's Penryn CPUs and 32 nm will be introduced with their Nehalem line later in 2008, and when these processes are matched to high-k technologies, bandwidths could double even over these incredibly impressive figures! One thing is certain: The future of computing will be sizzling fast!