Rupert Goodwins' Diary
Tuesday 11/7/2006
Motorola Semiconductors — oops, sorry, Freescale — has announced a new magnetic memory chip. This is actually quite exciting; magnetic memory has some chance of becoming the much-awaited universal memory technology, which combines the best features of flash, dynamic and static memory designs to work fast, keep its contents when the power goes off and be easy to design into circuits.
Some chance, just not much. Even assuming that everything Freescale says is true — that the chip is as fast, retentive, reliable and usable as promised — it has one huge and probably insurmountable problem. It costs too much, and it always will.
A 4Mb chip — the biggest you can get – costs $25, which means a single megabyte costs $50. That's thousands of times more expensive per bit than the alternatives, and because the process that makes the chip is considerably more complex than that used for existing memories, it will never be possible to erase the differential completely. More complex chips have lower yields, and the lower the yield the higher the price.
Meanwhile, the existing memory technologies have huge markets, are making massive amounts of money and are being produced in staggeringly efficient ways. The market is carnivorously competitive, so the amount of development money being spent on keeping things that way is immense: it's just not possible to introduce something that's very different and just a bit better.
That's the reason that the last new memory technology to make an impact was Flash, introduced nearly 20 years ago. Even then, that was an adaptation of an existing technology, ultraviolet erasable programmable ROMs. Crucially, Flash did something that no other technology could manage — not just better or easier. These past 20 years have seen intensive research and development in all areas of semiconductors, with thousands of innovations and many discoveries, but nobody's come close to introducing a commercially significant new memory technology.
For MRAM to become a moneyspinner, Freescale has to find an application that it and it alone can do. There are a few — Flash's biggest problem is that it wears out after tens or hundreds of thousands of data writes, and MRAM is a lot more resistant to bit corruption from radiation — but none in the mass market. Not that there's anything wrong with making high-priced boutique circuits for specialist markets, except that it doesn't make you much money.
If I had to bet, I'd say that Freescale will continue to develop MRAM for another year or so while it builds up a portfolio of clients in aerospace and instrumentation, then flog the whole lot as a going concern to a company better set up for low-volume, high-price components. That's if it doesn't go the way of bubble memory — a marvellously weird magnetic chip technology that some company called Intel was enthusiastic about in the early 1980s, and that even found its way into a portable computer or two.
Shame. New cleverness in silicon makes my heart sing. I hope I'm very wrong, and that in five years' time we're MRAM'd to the gills. But I wouldn't put any money on it.