Intel makes on-CPU system RAM breakthrough Intel has developed an updated technology that could virtually solve questions of memory speed, the semiconductor firm has revealed at its recent Research@Intel event. Developers at the company have created dynamic RAM that behaves like typical system memory but needs just two transistors for each memory cell and which needs no capacitors. The invention makes dynamic RAM small enough that it can be embedded in a processor rather than put into a separate module, potentially eliminating several bottlenecks inherent to the static RAM used for cache on current processors.
A processor that complements or replaces existing cache would have much more on-chip bandwidth than any current processor and would be tied to clock speed, according to Intel. Where a 45-nanometer, quad-core Xeon currently has some of the fastest bandwidth available at between 18 to 20 gigabytes per second, a basic 2GHz processor made on an older 65-nanometer process could shuttle data at 128 gigabytes per second. Dynamic RAM also promises more storage in the same space and a lower price that could translate to the processors themselves.

The advancement could effectively overhaul Intel's approach to its chip designs as well that of programmers. Very fast memory access is considered critical to future many-core processors and may be necessary for the production equivalent of Intel's 80-core Teraflops Research Chip; developers writing for this and related processors may also depend on always having guaranteed access to cache, which would be virtually guaranteed with the cache becoming faster along with processors.

Intel hasn't said when or if it first expects to implement its two-transistor dynamic RAM into future chips, though mainstream processors with eight to 16 cores are due within the next few years.