Rambus has developed a memory technology it claims will yield as much as a fourfold performance gain over standard memory yet can be applied to today's DRAM core designs.
The technique, dubbed 'Micro-threading', involves partitioning the memory chip's storage into separate addressable areas, to which data requests can be sent in parallel, allowing the memory controller to send, say, four data requests in the time it would previously have taken it to send just one.
It's not a linear improvement. Rambus' numbers suggest that while the best cases might bring back data four times as fast as before, depending on the application, you might only get a doubling of performance. How well the host memory controller, be it on a computer motherboard or on a graphics card can take advantage of the gain is another matter. The performance gain may easily be countered by bandwidth limitations elsewhere in the system.
Rambus said the technique, which is currently the subject of a patent application, is available for licensing today. Micro-threading can be added to DRAM chips for a "relatively low incremental cost", the company said, without going into details.
Of course, the technology requires DRAM controllers to be 'aware' of the feature, so they can parallelise the data read requests, and sort the data as it's returned, and Rambus will be looking to license its system to memory controller makers too. ®
Toshiba, Elpida prep 'industry's fastest' DRAM
Infineon and Rambus kiss and make up
Philips claims 'super Flash' memory breakthrough
Samsung shows 'world's first' DDR 3 chip
Elpida licenses 'DVD on a chip' memory tech
Rambus sues four for GDDR 'infringement'
Samsung ships 256Mb XDR chips