All computer data storage has compromises. Hard disks are capacious, but compared to silicon memory they're extremely slow, unreliable and power-hungry. SRAM is low power and very fast, but is more expensive than DRAM and harder to make in large sizes. DRAM takes more power and is slower than SRAM, but is much cheaper: both types lose their contents when power goes away. Flash memory doesn't do that -- but it has a limited life, isn't cheap and is very slow to accept data. So, despite the huge successes of the computer memory industry, the search always goes on for a successor technology that can compensate for some of the downsides of existing circuits while having advantages of its own. One long-term contender is MRAM, Magnetic Random Access Memory, which has been brewing in the labs for around thirty years. Despite this extraordinarily long gestation, many of the companies involved -- Motorola, IBM, HP, NEC and all the other major memory companies developing this technology -- promise commercial use in the next year or so. Motorola has gone as far as describing MRAM as 'universal memory', because of its potential to replace all other sorts of memory chip. All memory works by changing the long-term state of a circuit by a short-term signal. DRAM and Flash memory store the signal itself as an electrical charge that can be read back, SRAM uses the signal to turn on one of two transistors. With DRAM, the charge keeps leaking away and needs to be constantly checked and if necessary replenished, and SRAM forgets which transistor was turned on the moment the power goes away. Flash has much better ways of keeping the charge than DRAM, but that means it's much harder and takes longer to set the charge up in the first place. With MRAM, as with tapes or disk, the signal sets the magnetic state of the memory bit -- and that stays set indefinitely, without needing any power to maintain it. MRAM is made by sandwiching layers of magnetic metal compounds between two rows of wires, the one on top running at ninety degrees to the one underneath. By arranging a signal to flow through a top wire at the same time as one flows on one underneath, the point at which the wires cross can become magnetised. With the right magnetic material in the middle, its electrical resistance depends on its magnetisation -- the magnetoresistive effect -- so the same wires can be used to measure the resistance at the cross-over point and thus read back the contents of the memory. This is electrically very simple and can be made to happen very fast with very little power, ideal characteristics for computer memory. A less than ideal characteristic of MRAM is that it needs quite complicated physics using unusual compounds. While DRAM and SRAM use well-known techniques with silicon, MRAM adds much extra complexity with very precise application of new materials, while still needing the same basic silicon circuitry for interfacing and control. The economics of memory dictate very high volumes produced reliably, and MRAM has many unknowns that need to be solved. Currently, MRAM also needs quite a lot of power to write information and is also sensitive to high temperatures resetting the magnetic fields. Another problem with MRAM of late is cell size. The biggest market for memory is DRAM, which needs an area ten or so times smaller than MRAM per bit. The bigger the area of the chip needed to store a bit, the more expensive the whole chip is for a particular memory size. With the extreme cost pressure on IT at the moment, it's not clear that people would pay much more for a computer that turned on nearly instantly rather than taking thirty seconds to boot up. And, ironically, a lot of the development that MRAM needs to become a true universal memory can only be justified if it has already found a profitable mass market. It is most probable that MRAM will first appear in embedded applications, in devices where low power and permanent storage are at a premium. High speed wireless devices running from batteries are a likely candidate, especially those handing lots of streamed video or audio data. Another very interesting market is network routers, with very large addressing tables that are constantly updated and must be resistant to power outages and other glitches. Expect to see the first MRAM in 2004, although in small sizes -- one or two megabits -- for specialist uses. If the current trend for faster, more capable, more reliable computing operating at ever smaller amounts of power continues, the chances are good that MRAM will become increasingly attractive -- just don't expect it to replace gigabit DRAM sticks any time soon.