Original URL: http://www.theregister.co.uk/2006/07/17/freescale_rfid/

Freescale serves up alternative to flash

mmm-mmm-MRAM

By Guy Kewney

Posted in Mobile, 17th July 2006 13:12 GMT

Comment When Freescale recently announced a four megabit memory chip, my immediate reaction was to laugh.

Magneto-resistive RAM - a technology which several aspirant companies have abandoned - has been announced in commercial form by Freescale, which is prepared to sell you a four megabit memory chip.

It's been described as "an expensive solution to a problem nobody has" by experts on Flash memory, its main technology rival. But after some laughter, I decided to see what was going on - and came to the conclusion that there may be a market in RFID.

Scepticism has greeted the announcement from the few sources who actually investigated it. Once they realised that the memory chip in question is a mere four megabits in capacity, only 512 Kb - which compares dismally with the $20 1Gb Flash chips available to consumer markets - most comment has been to pour scorn on the investors who sold Sandisk short.

Sandisk, supplier of Flash RAM, is really not under threat just yet. It would take 2,000 MRAM chips to match the capacity of that Flash chip, and quite probably, just one of the MRAM chips would cost well over $25. So, say sceptics, why is Freescale bothering?

The main advantage of MRAM over Flash is that you can write more data, and much quicker, for the expenditure of far less power. So applications are, like those for Flash, ones where power may fail but data must not be lost, but where the time available before power fails is minimal.

Point-of-sale tags (like the London Underground "Oyster" ticket system) would appear to be the obvious example. The chip inside the Oyster card has no battery. It only functions for as long as power is transmitted by the card reader - and in that time, it has to "wake up" its radio, verify its access rights, and then write any changed data (like, how much cash is left) back to the RFID chip memory.

The Freescale announcement focuses on the "fast" feature of MRAM. It is, says Andreas Wild, European R&D director, "as fast as SRAM".

The capacity of the chip isn't as much of an issue as it sounds, says Wild. "It will scale. This technology is very conservative. It's standard CMOS, completely standard - that's the point of it - and then you add four new layers of processing to build the tunnelling electron engines."

Conservative means fabricated on virtually antique semiconductor processes - specifically, 0.18 microns (180 nanometres). It will, he assured us, scale right down to 65 nanometres.

That's a relief, because rival announcements from NEC and Toshiba spoke back in February of a 16 megabit chip. The reason this is only a quarter the capacity is "because it is nothing more than a lab curiosity until someone has a commercial-grade product with commercial-grade maturity and reliability", said Wild.

He's not giving too much away about what makes Freescale the first (it claims) to put this on the market, or why it took it so long. It isn't the only semiconductor on the trail. According to EE Times, Crocus Technology SA of Grenoble in France just raised $17m in capital (that was as recently as June) based on "key patents from Spintec, a CEA/CNRS research laboratory" (CEA is the French atomic energy commission, and CNRS is the national research council).

The Freescale announcement punctures the promise by Crocus's backers that it would be first to market. It may still be first to market with a different technology, but Freescale's announcement ensures that anybody doing prototype development will be ringing the Freescale phone number right now.

What, exactly, is tunnelling? What is magneto-resistive RAM? The first thought is that it's something dreamed up by Nobel prize-winner Professor Brian D Josephson - who used tunnelling in superconductor chips - and indeed, the technology is similar in that both use quantum effects. Professor Josephson said: "I can tell you the total of my knowledge of magnetic tunnelling..." and offered me a blank paragraph.

I guess he had real work to do. Of course, he could be more informative on the subject if he chose to be. Tunnelling is one of those phenomena of the Quantum universe, in which reality gives way to absurdity, and the measurements confirm the absurdity as the correct representation of reality.

An electron is not in any particular place. Rather, there is a probability of it appearing somewhere. Like a bookmaker, it's your job to fix the odds so that you get the result you were actually counting on. The quantum particles used in MRAM are persuaded to magically appear on the "wrong side" of a dielectric when a write pulse is applied to the magnetically polarised electrons - and by storing these magnetically polarised electrons, you generate a magnetic field which can be read.

The full explanation (for public consumption) is explained in a white paper from Freescale.

But with Freescale, you never can tell. I know Freescale is capable of wasting an hour of anybody's life by talking about Zigbee technology as being really really useful - and illustrating this by describing electric light bulbs which can be switched on by wireless messages, rather than by ordinary light switches. OK, it's just my opinion, but this is simply a silly idea and one of many technologies which Freescale is prepared to tout as a commercial product, without any real expectation of doing serious business with it.

I suspect it will use Zigbee capability to cross-sell Bluetooth, which solves all the problems Zigbee claims to solve in a more standard manner and for a fraction of the price, but that's just me, and a few other people who are already doing quite well with this tactic.

It wasn't until I started chatting to RFID people that I started getting an enthusiastic response to the MRAM idea. They love it!

And the problem they have is simple: it takes an uncomfortably long time to write data to other non-volatile memory technologies. And it takes power. RFID apps just don't have either to spare. And it's at this point that MRAM suddenly looks magic. To write Flash memory - even fast Flash - takes milliseconds. To write MRAM takes nanoseconds.

So we're looking at the ability to write thousands, even millions, more bits in the same time - and to write it with far less power available.

Of course, that's not the whole story. Any wireless device takes a while to turn on, wake up, and settle down before you can actually start using it, and there again, we're talking milliseconds. So the whole delay involved in using Flash or MRAM storage in an RFID device is not used to write data back to the Oyster card. But (say RFID designers) that makes it even more urgent: the window during which they can write and verify the data written to the card is tiny.

With MRAM, not only can they write a lot more data in that window, but the window is also bigger. The amount of power required to write Flash drops rapidly as the card user moves the card through the field. Users are always being urged to make contact with the pad. There's no actual contact needed, it's just that if you do make contact it takes much longer for you to get in and out of the field, and the voltage generated will be higher.

There's another very important feature, not emphasised in most of the comments: this uses a completely standard CMOS semiconductor technology as its base. The four semiconductor layers required to build up the MRAM cells come on top of standard CMOS layers, which means a non-volatile permanent scratchpad can be incorporated on existing chips.

That does leave one question to which the answer may be wrong, from Freescale's point of view. The question is: "Well, RFID - is that really going to be taking over the world?" and the answer, right now, has to be "nobody can be sure".

The main problem with most RFID applications is that there's an unacceptably high failure rate. In a corporate, administrative environment, it's possible to make incentives for workers, which ensure they patiently ensure that the tags are correctly read (for "incentive" - read, "they get fired if they don't").

But in point of sale applications you're asking your customers to spend extra effort. Take the example of Oyster. In London's underground system, the ticket gates beep once for a successful read, and twice for "there's a problem." You don't have to be Sherlock Holmes to realise that an awful lot of the beeps are double beeps.

London Underground's boss, Transport For London, can mandate the success of Oyster, because it controls pricing. Oyster journeys are cheaper than ticket journeys, not because the margins for the operator are better, but because Mayor Ken Livingstone has decreed that they shall be cheaper. Incentive again; it forces people to use the technology, even if occasionally they can't get into the network and have to go queue at a window to get the error corrected.

You can't run a small coffee-and-sandwich outlet on a street corner like that. If you are going to use RFID, it has to be at least as cheap as any other form of sales transaction (chip and PIN) and as reliable, or your customers simply won't go there.

RFID has yet to pass that test. Don't expect to see any actual four megabit MRAM chips in products this year, or next. And by the time the first products actually appear in commerce, they'll almost certainly be more like 16 megabytes, using smaller-scale silicon geometry, and with prices that make more sense. But if RFID does succeed, this MRAM breakthrough may be what makes that success possible. ®