FRUSTRATION is programmed into our electronic devices along with their operating systems. Every time you use your PC, it takes precious minutes to boot up and power down. Your PDA never holds quite enough data. Your cellphone’s battery fades before you can finish your list of calls and check your email. Your notebook computer crashes and sends the draft of your report to oblivion. And it’s all because of memory chips.
Memory is the Achilles’ heel of electronic devices: it slows them down and guzzles their power. If only there was one perfect form of computer memory – one that starts your PC instantly, lets batteries last all day, never loses your data and operates at high speed.
Computer designers have long shared that wish, and now many of them think it is about to come true. Motorola has recently demonstrated a dream technology that aims to combine all the best features of today’s computer memories into one tidy chip. That technology is called magnetoresistive random access memory (MRAM). It’s fast, hard-wearing, and miserly with power.
Advertisement
And MRAM is offering more than instant-on computers and an end to lost work. Insiders believe it will extend the lifespans of space missions, speed up digital devices and bring fantastic capabilities to cellphones and PDAs, including powerful data and video processing and as much as 20 times the normal battery life. With the potential market estimated at $40 billion a year, manufacturers are working hard to get MRAM onto the streets.
In many ways, MRAM represents a return to the dawn of electrical computing. In those early years, engineers stored 1s and 0s using magnetic materials. In the 1950s, the earliest memories were built from arrays of magnetisable metal rings strung on grids of criss-crossing wires, with each ring positioned at the intersection of two wires. The atoms inside these rings act like little bar magnets, with most of them oriented in random directions. To write data on a ring, you apply a current to the pair of wires passing through it. This generates a magnetic field strong enough to align all the atoms, in the same way that a compass needle points to the magnetic north pole. And the direction of the alignment specifies a 1 or a 0.
The metal rings were simple and cheap to make. And because they didn’t need a steady electric current to maintain their magnetic state, the data often outlived the gargantuan computers that generated it. However, the technology couldn’t keep pace with the needs of rapidly advancing processors. Engineers struggled to shrink the magnetic memory, and they eventually abandoned rings when smaller semiconductors and transistors arrived.
Yet even as transistors took off, engineers never lost their wistful fondness for magnetic memory. These dreams were revived in the early 1990s when researchers at IBM developed a device called a magnetic tunnel junction, and it is the tunnel junction that lies at the heart of today’s MRAM structure.
A magnetic tunnel junction (MTJ) is a sandwich of two magnetic metals separated by a layer of electrically insulating material just a few atoms thick. One of the metal layers is a permanent magnet, meaning its magnetic atoms are always aligned in the same direction and cannot be affected by an external magnetic field. But the magnetism of the other metal layer can be switched at will by exposing the sandwich to a magnetic field.
MTJs exploit a phenomenon known as giant magnetoresistance, in which layers of magnetic metals sandwiching an insulator can control the current flowing across the insulator. When the two layers are magnetised with opposite polarity, the device acts as an insulator with high electrical resistance. Yet when the magnetic layers have the same orientation, the resistance plummets and electrons tunnel through the insulator.
Researchers originally thought they could use MTJs to write or delete information stored on a transistor memory chip. They envisaged connecting each transistor to a tunnel junction to read or write the bit stored on the transistor, in much the same way that computer RAM uses transistors to read and write bits stored on capacitors (see “Chips with everything”). But engineers have always struggled with tunnel junctions. For a start, the external surfaces of the magnetic layers in an MTJ must be scrupulously clean, or the devices can easily fail. Also, the thin layer of insulation can break down if the current flowing between the layers is the slightest bit too strong.
Another problem is that to make such devices as fast and compact as other forms of memory, the MTJs have to be built onto the same silicon chips as the transistors they control. Engineers have found it devilishly difficult to stop metal atoms sloughing off the MTJs during manufacture, and the errant atoms migrate down into the chip’s silicon foundation and corrupt it.
Even when researchers managed to overcome all these difficulties, they were left with another problem. To keep hold of their data, the transistors had to be continually powered. While that’s fine for gadgets plugged into the mains, it saps the batteries in cellphones and laptops.
Then last October, Motorola announced that its engineers had worked out how to turn tunnel junctions into working memory for mobile devices. Their design turns the original concept on its head: while it still uses MTJs and transistors, it stores the 0s and 1s on the tunnel junctions and uses the transistors to control them. It is a crucial difference because the tunnel junctions retain their magnetic state, and hence their data, even when the power is off.
Unfortunately, just like metal ring memory, Motorola ran into trouble when it tried to shrink the devices. It faced two equally daunting problems: wandering magnetic fields and rising voltages. The more MTJs you cram together, the easier it is for the magnetic field of one junction to stray into the territory of a neighbour. That means they can begin flipping magnetic atoms that aren’t supposed to be flipped.
That problem is worsened by a more fundamental effect known as “coercivity”. The equations of electromagnetism reveal that as you shrink the magnetic layers in a tunnel junction, you need to apply an ever stronger field to flip the magnetic orientation of the variable layer. And that means supplying more current. So if you shrink MTJs down to the size of the transistors in today’s cellphone and PDA memory chips, the task of handling data gobbles more and more power. Not only would that drain batteries, it could also heat the tunnel junctions to absurd levels.
While engineers at Motorola, IBM, TDK and other companies have succeeded in making working tunnel-junction chips, their devices contained a mere 1 million transistors – the RAM in a typical PC has 250 times as many. Even worse, the MTJ chips were too slow and way too big to squeeze into cellphones or computers. And no one was able to build a device in which magnetic fields didn’t spill from one tunnel junction to another.
But that situation changed dramatically with a breakthrough by Leonid Savtchenko, a Motorola engineer and materials modeller. He predicted he could solve the magnetic leakage and heating problems by using a sandwich of metals in which the atomic magnets in each layer point in carefully chosen directions. This way he could turn the atomic magnets round gradually by applying a series of weak magnetic field pulses to the layers, rather than one strong pulse. Because the pulses were weak, they wouldn’t disturb the neighbouring tunnel junctions.
Motorola’s engineers then designed a way of applying just the right magnetic field, using a grid of wires rather like the ones in the old magnetic ring memory. To write a bit of data on a specific junction in an array, Motorola’s design sends two separate currents past the MTJ a few nanoseconds apart, along the two wires that intersect at that junction (see Graphic).
The arrival of the first current pulse nudges the memory cell’s magnetic orientation partly towards the opposite direction, and the second current turns it a little more. When the first current shuts off, the effect of the remaining current tweaks the spin direction a little farther, and cutting the second current completes the transition. Neither current is strong enough to flip a junction on its own, so the only bit that changes is the one where the two currents intersect.
To read the data, Motorola engineers have built their tunnel junctions on top of standard silicon transistors that act like switches. You simply close the transistor switch, which applies a voltage across the tunnel junction. If the magnetic layers are aligned, current flows and you read off a digital 1. Otherwise, no current flows – a 0.
Last October the company unveiled an MRAM chip that holds 4 million 1s and 0s. Although RAM chips found in PCs typically pack in 64 times that amount, Motorola’s design is big enough to run a basic cellphone. And the company has recently given prototypes of its chips to other electronics manufacturers to evaluate.
Motorola’s 4-megabit memory chip is not yet ideal, but it is important because it’s the first stage towards the computer engineer’s dream: building memory into a processor chip, the engine-room of a computer.
Every time you open a program on your PC, the processor loads the instructions it needs from the hard drive into a temporary storage area where it can access the data more easily. This storehouse is the computer’s RAM and is separate from the processor. Millions of times every second, the processor downloads data from the RAM, processes it and writes it back onto the RAM again. This continuous shuffling of data not only takes precious time and power, it is also risky. When you close a file stored on the RAM, the processor transfers the data to the hard drive and deletes it from the RAM to make room for more. But if the computer crashes before you save the data, then it is lost forever.
MRAM promises to put an end to all that shuffling. “Once you can put memory and processing physically together, then you can do away with disc drives and other forms of memory,” says Gary Prinz, director of the Nanoscience Institute at the US Naval Research Laboratory in Washington DC. “To do that, the memory has to update its data as fast as the processor requires, which magnetic memory can do.” What’s more, this way you could, in theory, build a complete computer onto a single chip and embed it into all sorts of everyday items, from toasters to pill bottles. Motorola hasn’t yet gone as far as combining a processor and memory onto the same chip but their magnetic memory chip represents the first crucial step.
But not everyone is impressed. Critics believe the MRAM designs are doomed because coercivity will prevent the electronics industry from making the tunnel junctions small enough and cheap enough. “You can’t make these things that small,” says Darrell Rinerson, president of Unity Semiconductor in Sunnyvale, California. “As you make magnetic elements smaller, they tend to be less stable. That makes them impractical for mass-market products.”
Rinerson is among a band of researchers betting on alternatives to MRAM that could yet consign silicon memory chips to museums (see “Return of the rings”).
However, Jack Judy of the University of Minnesota’s Center for Micromagnetic Technologies in Minneapolis thinks it is far too soon to write off Motorola’s MRAM cells. “They have a way to go, but people have ideas to get around those problems,” he says. “You might need to make a major change in the structure to make it work better, but there’s no doubt that clever engineers can solve these problems.”
Aerospace company Honeywell in Plymouth, Minnesota, certainly thinks MRAM is worth a gamble. It is so confident about the technology that it has bought rights to copy Motorola’s design and has plans to incorporate it into satellites. Perhaps MRAM is about to take off in more ways than one.
Chips with everything
Electronic gadgets can contain a plethora of chip-based memories, each type of which has serious limitations that counterbalance the advantages. Dynamic random access memory (DRAM) has been the workhorse of our computers for 30 years because it is fast, simple and dirt cheap.
Each DRAM memory cell stores the bit as charge on a capacitor, controlled by a transistor. The big disadvantage is that the capacitors leak their charge: every memory cell must be recharged several hundred times a second otherwise its data evaporates.
Static RAM (SRAM) does not need to be refreshed in this way, but only as long as the power stays on. The other disadvantage is that it takes as many as six transistors to make one SRAM memory cell. That makes SRAM bulky, and costly enough to limit it to specialised uses such as computer caches.
The flash memory used in digital cameras and pen drives holds data for years, even without power, and is as compact as DRAM. But it is pricey, and after a few hundred thousand rounds of writing and erasing data, the storage material itself degrades and it begins to lose data – not very desirable for workaday computers or space probes on long missions.
In contrast, MRAM seems nearly perfect. It’s as fast and compact as DRAM, as parsimonious with power as flash but works a thousand times faster, and it never wears out. “If you can build it cheap enough for the consumer market, you could replace all these other memories with MRAM,” says Hassan Kaakani of Honeywell in Plymouth, Minnesota.
Return of the rings
Gary Prinz at the US Naval Research Laboratory in Washington DC is harking back to the good old days of metal memory cells shaped like bagels. Prinz’s group has managed to store magnetic bits on rings smaller than tunnel junctions, and pack them more tightly together. “The potential density for ring-based magnetic memory is on the order of hundreds of gigabits per square inch,” he says compared to 100 megabits per square inch for MRAM. If all goes to plan, such devices could be ready for the market by early 2005, Prinz claims.
Others are convinced that the way forward is to dump silicon chips altogether. Integrated Magnetoelectronics (IME), a small company based in Berkeley, California, has decided to scrap semiconductors and build memory cells and their supporting electronics entirely from metal. The company believes its technology can keep magnetic fields tightly enclosed inside memory cells smaller than 50 nanometres across, and is easier to manufacture than silicon chips. The devices could also be laid on glass or plastic rather than just silicon, as conventional transistors require. But IME’s approach faces a formidable barrier. While tunnel junctions wait for technical breakthroughs, IME is waiting for the money to get off the drawing board.
