IBM has announced it has come up with a way to triple the amount of memory stored on computer chips. If it can do that then it will double the performance of processors. It will replace one type of memory with another that uses less of the real estate on a silicon slice.

IBM has a new memory technology which allows a lot more to be stored on a single chip. Thus the microprocessor will be able to retrieve data from its own memory instead of using a separate memory chip.

Lisa Su, vice president for semiconductor research and development at IBM, summed up the new development when she said, ‘We kill ourselves in the semiconductor industry to try to get a little bit more performance in each generation. What we’re doing here is trying to merge two technologies . . . on the same chip to get significantly more memory.’

IBM’s solution means not using SRAM — static random access memory — to store information directly on computer chips but DRAM, which is dynamic random access memory.

The big plus for SRAM is that it is fast and not that complex, in chip terms, to manufacture. The downside is that, relatively, it takes up a lot of space. DRAM is the sort of memory you have on a separate chip so that the computer on which I write this has a gig of memory built in and that is DRAM. Probably the same as on your computer. Heretofore it has been thought that it was too slow to build DRAN onto the microprocessor.

What IBM has done is boost the speed of DRAM so that it is almost as fast as SRAM and still takes up much less space. So the new embedded DRAM — known now as eDRAM and you should get used to it — allows more memory on the chip itself which makes the microprocessor faster.

IBM says it has developed DRAM that takes 1.5 nanoseconds — or billionths of a second — to fetch data, compared with 10-12 nanoseconds for conventional DRAMs and 0.8 to 1 nanoseconds for SRAM. The DRAM can store three times as much data in the same space.

IBM expects to use the technology in new Power chips due next year. Another likely benefactor could be AMD, which is an IBM technology partner.

Roughly the equation is that IBM has tripled the memory which doubles the speed.

The company said the technology will be included in its server chips starting in 2008 and will expand to other products.

Earlier this week, Intel said it has developed a research chip capable of performing calculations at teraflop speeds but still using only a small amount of energy.

So this sort of keeps us in step with Moore’s Law — which is not in any way a law but a suggestion. The 1965 prediction by Gordon Moore, co-founder of Intel, was that the number of transistors on a chip should double about every two years so that they become twice as powerful every two years. This is a very wide statement but the essential truth is there — the speed of chips is now moving into areas we never dreamed of ten, even five, years ago.

Our illustration is of the IBM Harvard I of 1944, one of the first computers. Shrunk some since then.

Related:

Entry was posted on Wednesday, February 14th, 2007 at 6:01 am under Science and technology, Technology.

Read more entries by Gareth Powell.
Stumble It!