He's already set off one computer storage revolution. Now Stuart Parkin is reengineering RAM so we'll never have to boot up again.
Stuart Parkin is not a guy who needs caffeine. By the time I meet him for morning coffee inside IBM's Almaden Research Center, a tidy row of four steel-and-glass shoeboxes secluded in the foothills above San Jose, California, Parkin's already been up for hours, bouncing around his lab. We go down one floor to his office, with Parkin leaping stairs two at a time, as if mere walking holds him back. He's loudly humming some kind of double-time march. Everyone who meets Parkin notes his affinity for humming, a kind of continual soundtrack to the operation of his mind.
For the past decade, the 44-year-old British-born physicist has been Big Blue's chief innovator for magnetic storage research. In the early 1990s, Parkin developed a new type of read/write device for hard drives known as a giant magnetoresistive, or GMR, head. Now a component in every PC's hard drive, GMR heads have dramatically improved the reliability of data storage by helping to stabilize smaller magnetic bits, which had a tendency to unexpectedly flip polarity or vanish altogether.
In recent years, Parkin has shifted emphasis from static hard drive storage devices to dynamic memory technologies and is currently developing an innovative chip called MRAM. Based on a tiny, checkered grid of magnetized switches, magnetic random-access memory will eventually replace dynamic random-access memory (DRAM), the ubiquitous workhorse in almost all the 320 million PCs currently in use.
The crucial difference between MRAM and DRAM is that Parkin's chip operates without electricity, relying on magnetic polarity to store data. This distinction has important consequences for, among other things, a computer's boot-up time, which may explain why Parkin is in a hurry. He's on his way to delivering something that's very high on the wish list of PC users everywhere: "instant-on" computers.
Parkin walks me through MRAM basics, starting with a refresher course on DRAM. The bits in DRAM, he says, are nothing more than clumps of stored electrical charge. Hundreds of times every second, a bracing pulse of electricity is required to refresh the DRAM capacitors, or any data on the chip will be lost. "This charge continually leaks, or evaporates, from the capacitors," he says, "so power must be consumed to repeatedly refresh the memory."
A typical DRAM-based machine stores the operating systemand applications on the hard drive. As your computer slowly comes to life - a process that can take several tedious minutes - a working copy of the OS, as well as any programs you've tagged to launch at startup, are loaded from the hard drive into DRAM, where a microprocessor can get to them quickly. If your computer crashes or freezes, any data stored in DRAM is lost and you're forced to sit through another lengthy boot session.
Replace DRAM with MRAM, and your computer would work like other electronic devices: Flip the power switch on, and the machine is up and running immediately. When you switch off your PC, an MRAM chip would retain anything loaded on it, such as the OS and apps.
MRAM is a type ofnonvolatile memory, which means the chip is based on a solid-state design (no moving parts), and data on the chip doesn't have to be periodically refreshed. Another example of nonvolatile memory is flash memory, used in devices like MP3 audio players and digital cameras.
Why not use flash memory to make an instant-on PC? Because there's a hitch: Flash memory cells get damaged each time they write a bit. After about 10,000 read/write cycles, they crap out. Thus, flash will prevail in consumer electronics, but its lack of long-term reliability makes it a poor choice for desktop memory.
"The holy grail in memory chips is something that is nonvolatile, consumes low power, and is cheap," says Jim Handy, semiconductor memory analyst with Dataquest. "But we just haven't been able to find a fusion of all those attributes. The nonvolatile memories we have now are slow as molasses." Handy says DRAM is vulnerable to any technology with "the right attributes of price and speed."
Parkin has already built prototype MRAM chips that store about 1 Kbyte of data. That's not much, but he's confident he can ramp up memory capacity in the next few years and produce a chip that is smaller and faster, stores more data, and costs less to manufacture than DRAM. Parkin won't say exactly how long it will take IBM to do this, but analysts outside the company believe it will happen in less than five years.
To wholly supersede the present memory standard in a mere half-decade may sound wildly optimistic. The worldwide DRAM market was $21 billion in 1999, reports Dataquest. By year-end 2000, it's forecast to increase 44 percent to $30 billion, with annual revenues growing steadily at 40 percent through 2002. But several big names behind commercial R&D labs - including Motorola and Honeywell - are sure a big change is coming, and are funding MRAM projects. Intel, Hewlett-Packard, Toshiba, Siemens, and Bosch are also experimenting with MRAM - all of them determined to leap ahead of IBM and capture the potentially huge profits if millions of computer owners race to upgrade their PCs to instant-on machines.
I ask Parkin how, with all the competition, IBM can capture the MRAM market. "We have 10 to 15 patents that relate to MRAM," he tells me. "Two are very important: one for the magnetic memory cell itself, and another relating to the chip's architecture. We've also been very successful in moving leading-edge materials into mainstream products, including GMR in disk drives, and copper interconnect wiring for chips."
Parkin was born in Watford, England, but his father's marketing business kept the family on the move - first to Bowden, near Manchester, then to Scotland. As a teenager, Parkin says he turned inward, describing himself as a recluse,a bookish, overachieving youth with a natural curiosity for science. Cambridge was the obvious choice for college, recalls Parkin, because of its strong reputation in physics. He earned his PhD in 1981 while doing research at the university's Cavendish Laboratory, where J. J. Thomson discovered the electron a century ago. After Cambridge, Parkin went to France for a year of postdoc work in superconductors at the Université de Paris Laboratoire de Physique des Solides in Orsay. In 1982, he joined IBM's San Jose facility, which was about to refocus R&D attention on magnetic physics - Parkin's specialty.
As Parkin tells it, IBM was searching for a way to revive its data storage business, a market Big Blue had dominated for 10 years after inventing the first commercial hard drive in 1956. Almaden researchers were interested in structures made of thin layers of magnetic materials, believing these substances might let them pack in more bits of data per square inch of disk and improve the read/write heads in existing hard drives. A process called molecular beam epitaxy (MBE), developed at Bell Labs in the mid-'70s, used an atomic-scale spray gun to lay down precise films of materials for fabricating electronic devices such as transistors. In 1984, IBM gave Parkin and a colleague $1 million and asked them to build a modified MBE gun to do the same thing with magnetic materials.
Big Blue wanted Parkin to create new kinds of layered magnetic materials and analyze them for use in mass storage. "But the MBE system was a big, cumbersome instrument, very time-consuming to use," says Parkin. A year later, he turned the machine over to another scientist, abandoning the MBE experiments but using $50,000 left over from the original project to construct something called a sputtering chamber. A vacuum cylinder about the size of a washing machine, the chamber can slam atoms of gas into various substances, breaking off tiny shavings that cling to whatever material is nearby. This is basically a brute-force way to very quickly deposit layers of elements on a surface, which can then be tested for its ability to hold magnetic data.
At a 1988 meeting in Le Creusot, France, three years after he'd begun the sputtering chamber experiments, Parkin heard a talk by Albert Fert, a University of Paris physicist who was measuring magnetoresistance, the change in electrical resistance of a metal in the presence of a magnetic field. Along with Peter Grünberg, a colleague from the Institute of Solid State Research in Jülich, Germany, Fert had discovered that an iron-and-chromium sandwich produced a whopping magnetoresistance - so big it was dubbed giant magnetoresistance (GMR).
After Fert's talk, Parkin decided to try the same experiment back at IBM using his sputtering machine. "I'd been brainwashed like everyone else to think that the sputtered films wouldn't work," he recalls, "but with the chamber I could try things quickly."
He spent the next two years doing trial-and-error runs until he was able to perfect extremely sensitive magnetic field detectors that existing fabrication methods could produce for storing data. IBM quickly commercialized Parkin's GMR effect in its high-end disk drives, packing 3.3 Gbytes onto a 2.5-inch platter. Now the entire disk industry uses GMR read/write heads, which have helped gross IBM $8.1 billion and a 27 percent share of the storage market.
Parkin won't say exactly what financial rewards he reaped directly from IBM's GMR windfall, but acknowledges that he's "happy with the personal remuneration and benefits." His work also helped nab him the company's top R&D honor, IBM Fellow, which he was awarded in June 1999 by Big Blue's CEO, Lou Gerstner. The new title fits comfortably on Parkin's impressive résumé, which includes the American Institute of Physics Prize for the Industrial Application of Physics, the International New Materials Prize by the American Physical Society, and the Hewlett-Packard EuroPhysics Prize from the European Physical Society.
"Being made an IBM Fellow was probably linked to the success of my work on GMR," allows Parkin. And while he says that his research efforts have always been supported by IBM, the status change has provided some new leverage. "By the time you're a Fellow, hopefully people listen to you anyway - but it does give you more of a voice."
Magnets are ideal for storing data because they don't need power to maintain a state of either positive or negative polarity. Magnets are also inherently binary: Positive and negative can be represented as a 1 or a 0. Magnetic storage devices like hard drives and floppy disks already exploit this relationship. But until now, scientists couldn't re-create the process in RAM chips because it's so hard to control magnetism in such tiny structures. The smaller the magnet, the more easily it can shift polarity without warning.
During lunch, Parkin arranges his fork and knife into a model of a magnetic memory cell. MRAM places millions of tiny magnetic sandwiches on a silicon substrate. "You've got these tiny, parallel wires going in one direction on top of the sandwiches, and the other wires going perpendicular to them below," he says. Like threads in a loom, the read/write wires form the array's warp and weft. At each point where top and bottom wires cross, one of these little magnetic sandwiches represents a single bit.
To write a bit on an MRAM chip, a current passes through a wire above the sandwich, flipping the polarity on one of the magnets. To read a bit, a current travels through the sandwich, measuring the resistance of each magnet: Low equals 0 (or negative), high equals 1 (or positive). Once code is written into the bit, the magnetic sandwiches are stuck in their 1 or 0 state until the system erases or rewrites data. Meanwhile, the sandwiches retain their magnetic state, even when there's no current flowing.
Parkin imagines arrays of these sandwiches doing exactly the same job as the working RAM in a desktop PC. It seems like a simple notion, and in fact it harks back to an earlier technology, the magnetic core memory in early computers of the '60s and '70s - whose binary 1s and 0s were represented in magnetic fields created by iron rings so large that a 500-Kbyte memory bank took up an entire room. But it wasn't until solid-state physicists learned to make magnetic materials as they do semiconductors - mass-produced at micron scale with minimal contamination of the material - that MRAM became feasible. Prior to his work, Parkin says, no one would have believed you could control a layer of magnets just a few atoms thick.
Over the next five years, further iterations of Parkin's MRAM chips will become increasingly smaller and able to hold more data. But with each successive prototype, a range of microfabrication obstacles must be resolved. And IBM will also have to hammer out the best methods for mass production. After that, MRAM chips could be available to consumers at a cost comparable to DRAM, since MRAM doesn't need all the supporting circuitry to connect with a power source. But because it cuts out the delays associated with transferring electricity between the power source and the chip, MRAM would be up to 30 times faster.
"MRAM combines the desirable memory attributes of speed, density, and nonvolatility; thus, many applications are possible," says Parkin. "One very important application is what we call 'pervasive computing' - the notion of many small handheld devices combining and supplementing the functions of cellular phones, personal assistants, and Internet appliances. I suspect we'll see MRAM used in pervasive applications first, since these devices typically benefit from significant amounts of nonvolatile memory. But in time, MRAM's performance and cost attributes should evolve to the point where it is used in all computers."
While Parkin and his five-member development team perfect their new chips, Saied Tehrani, chief researcher for Motorola's MRAM project, says he's working on a similar, but not necessarily competing, technology. "We're not a stand-alone memory company and don't plan to compete with the DRAM makers," he notes. Instead, Motorola is steering away from the instant-on computer idea and looking at ways to integrate logic and memory MRAM chips for embedded systems in portable communications devices. "We see MRAM as a universal memory, with the positive attributes of technologies like flash and static RAM," says Tehrani, "and we anticipate a rollout of some kind in about five years."
A more direct threat to IBM's instant-on, no-boot chip is the MRAM work under way at Honeywell. "MRAM is perfect to fill the need for nonvolatile memory," says Theodore Zhu, head of Honeywell's MRAM technology group. His company has produced larger magnetoresistant chips with 16 Kbytes of memory for various military apps (MRAM can survive radiation exposure). But Honeywell broke its promise to unveil a 1-Mbit chip by the end of 1999, which would have put the company far ahead of IBM. Perhaps that's because Honeywell's primary customer is the military, says Parkin, "so they tend to build devices that wouldn't be commercially viable."
Another contender in the MRAMrace is the US Naval Research Laboratory in Washington, DC. "We've arranged magnetic material in tiny rings - like washers stacked on top of each other," says NRL physicist Gary Prinz. "You can run a wire right down the middle and switch the direction of the magnetic field in the rings." Prinz claims his design overcomes many of the roadblocks other MRAM structures encounter, such as high electrical resistance and imprecise control of magnetic switching.
The NRL work is impressive from a fabrication standpoint, says Parkin, "but it doesn't seem to have commercial potential. The electrical resistance is too low, and without some resistance, you don't get a signal."
Every year since 1994, IBM has topped the US Patent and Trademark Office's annual tally of new patents - 2,756 in 1999 - and the company has spent more than $5 billion on R&D every year for the last decade. For Parkin, this means almost limitless resources for pursuing the instant-on computer.
While IBM VP and director of research Paul Horn won't reveal how much Big Blue has earmarked for the MRAM project, a clue can be found in Parkin's machine shop, packed with multimillion-dollar, one-of-a-kind tools for producing new experimental sandwiches. In one area, he shows off his newest toy, a $1 million, computer-controlled vacuum chamber capable of cranking out preliminary MRAM materials even faster than his prototype sputtering chamber. When we return to his lab, Parkin softly quizzes a colleague, then bustles over to a computer near the old sputtering machine, where a data curve from a new compound is appearing onscreen. "I can't tell you about this one yet," he says, gesturing to the plot, "but it looks very good."
After spending the day with Parkin, I've become accustomed to his caginess, his reluctance to speculate when I ask him about the wheres and whens of MRAM: "It's impossible to know," he admits. "It's like an exploration of new territory, demanding a lot of intuition and instinct. Occasionally, things come along that are unexpected, and that's what you want, isn't it?" Then Stuart Parkin starts humming again.
