(Post 16/05/2006) Magnetic recording on hard disk drives is following an evolutionary path quite similar to that of the mainstream semiconductor industry. Through scaling, both are pushing the limits of lithography - makers of the heads that read and write the data from the disk say even more so than their microprocessor-making counterparts. Both employ thin-film deposition and etch technology to fabricate complex structures on wafers, and both are seriously considering how it will be possible to maintain the same pace of advancement as in the past.

The semiconductor industry has so far been closely following Moore's Law, which holds that the density of ICs will double every 18 months. The disk-drive industry lacks a catch phrase like "law" but is on a similar path, best defined in terms of the number of bits per square inch, called the areal density. After magnetoresistive (MR) heads were introduced in 1991, areal density increased by ~60% each year; in recent years, the increase has been closer to 100% per year. In 1996, for example, state of the art was 1.3 Gb/in.2. Today, 30 Gb/in.2 drives are being shipped, 60 Gb/in.2 have been announced and 100 Gb/in.2 designs are in the R&D phase, said Jim Porter of Disk/Trend Inc. (Mountain View, Calif.), a market research firm.

The way in which disk drives work is fairly simple, at least in theory: Bits are stored on the hard disk through magnetics, with a "1" or "0" being determined by the direction of the magnetic field at a given address. The disk is usually an aluminum or glass substrate coated with an electroless nickel layer, an insulator and a magnetic material, typically an alloy of cobalt, chrome, platinum, boron and/or tantalum. An overcoat of carbon and a lubrication layer finish off the disk.

All drives today are longitudinal, in that the grains of magnetic material stretch out across the surface of the disk, much like bar magnets. To go to even higher areal densities, researchers are looking at several alternatives, including perpendicular recording, patterned media and thermally assisted writing. Of these, perpendicular - where magnetic grains are stacked on their ends - has received the most attention and is most clearly poised to lead the way to the Holy Grail of the terabit level. That's 1 Tb/in.2, or ~1.55 Gb/mm2.

1. A finished device, with the inductive write head integrated with the GMR head. The head flies over the disk at speeds up to 100 miles/hr. (Source: Ed Grochowski, IBM Almaden)

The pixie dust solution

The reason longitudinal recording is running out of steam is mostly a factor of the thermal stability of very small magnetic grains. Called the superparamagnetic limit, the phenomenon causes loss of magnetization over a period of time; after 10 years or so the data might be gone for good. "There's a certain energy barrier to flipping a grain's magnetization. When you have grains that are small enough to spontaneously demagnetize, you call them superparamagnetic," said researcher Michael Mallary of Maxtor (Shrewsbury, Mass.). To avoid the effect, the magnetic energy of each grain must be high. "The way we've done so in the past is by increasing the coercivity of the material and also by working with large grain sizes, because the larger the grain the more volume it has, and the more energy it has against flipping." High-coercivity magnetic materials more strongly resist any change to their magnetization, but it works both ways - they also are more difficult to write on.

Unfortunately, as the size of the write head has been scaled down, so too has the size of the grain in order to maintain adequate media signal-to-noise. "As we scale down the bit size and scale down the grain size to match it, we're approaching the superparamagnetic limit of longitudinal recording. And we can't raise the coercivity any higher; we're already at the limits of the material," Mallary said.

There's one recent advancement that helps get around the superparamagnetic limit of small grains: pixie dust. At least that's the term IBM researchers came up with to describe the 3-atoms-thick layer of ruthenium that is at the heart of a new antiferromagnetically coupled (AFC) media introduced last year. IBM says AFC media could enable an areal density of 100 Gb/in.2. Where conventional disk media stores data in only one layer of magnetic alloy (e.g. CoPtCrB), AFC media is a multilayer structure in which two magnetic layers are separated by a thin layer of the non-magnetic metal, ruthenium. This ruthenium layer causes the magnetization in each of the magnetic layers to be coupled in opposite directions - anti-parallel - which constitutes antiferromagnetic coupling. The end result is that the magnetic thickness of the media can be reduced, increasing data density.

2. A close-up cross section of the integrated device, indicating the complexity of the structure. (Source: Ed Grochowski, IBM Almaden)

Perpendicular advantages

To achieve densities beyond 100 Gb/in.2 , it will probably be necessary to employ a different approach, and presently perpendicular media is the most promising. "People are talking about longitudinal recording possibly getting up to 70-100 Gb/in.2," Mallary said. "After that we're really in a bind because now the track density is so high, the lithography limits are so difficult, the tracking limits are so difficult and we have fringing problems on top of that... everything is becoming ridiculous. We need a way out of the box. Perpendicular recording can possibly offer us this alleviation."

Although the idea of perpendicular technology has been around since the 1970s, it has only recently received serious attention. Earlier this year, IEEE hosted the first North American Perpendicular Magnetic Recording Conference (NAPMRC), and the organizers had this to say: "... perpendicular magnetic recording will sustain the current growth in technological advances for the next several generations of disk drives. The technology is technically the closest alternative to conventional longitudinal recording, while it is capable of deferring the superparamagnetic density limit beyond that achievable with longitudinal."

"It might be possible to go to higher areal densities than with longitudinal recording before getting to the point where the trade-off between media noise and thermal stability limits the ability to further decrease the size of a stored bit," added Mason Williams of IBM's Almaden Research Center (San Jose).

Maxtor's Mallary said that perpendicular technology offers four main advantages over traditional longitudinal approaches. "With perpendicular you can make the media a great deal thicker. In longitudinal, the thicker you make the media, the more it tends to demagnetize itself. Perpendicular doesn't have that problem; demagnetization is independent of thickness. Because you can make it thicker, the grains can have a smaller diameter for a given volume. Therefore, you can get more grains per bit."

Second, it's possible to grow all grains in a perpendicular disk pointed in the same direction with very good orientation. "This is a key feature that you can't achieve in longitudinal. This means you can write sharper transitions, that the regions between up and down magnetization can be better defined than if all these grains were pointed in different directions. Now we can write a bit that has adequate signal-to-noise and fewer grains," he said.

Perpendicular media also does not tend to demagnetize itself as it does in longitudinal. "In longitudinal, when you start spacing your transitions very close together they start demagnetizing each other worse and worse," he explained. "The opposite happens in perpendicular. Putting transitions close together alleviates demagnetization instead of aggravating it"

Finally, in longitudinal recording, the magnetic material that is being switched is adjacent to the write gap. "It's working in the fringe field in the write gap, where in perpendicular when you have what's called a soft underlayer, the magnetic material is actually in the write gap of the head. Therefore, you can impose a higher magnetic field on perpendicular than you can on longitudinal. Therefore, you can write higher-coercitivity media. That allows you to improve the energy-per-unit volume of the grain, and that allows you to make smaller grains," Mallary said. (The magnetically soft underlayer below the storage layer acts to image the recording head in such a way that the field lines go down from the recording head into the soft underlayer and then back up.)

At the conference, the consensus was that areal densities up to ~1 Tb/in.2 "are likely to be realized" using perpendicular technology. However, even those committed to the technology see plenty of challenges. "Terabit per square inch is feasible with a move to perpendicular," Mallary said. "But that assumes that material, skills and art can match up to theoretical perfection."
"It's one of those things where there's a conjecture that, if we can do all the right things with material and head design, there might be more extendibility to perpendicular recording," IBM's Williams added. "There are people who are publishing papers claiming that there might be an advantage at some point in the future if everything can be appropriately developed, if we can make media that have properties that the people who do modeling can imagine, but we haven't necessarily seen those properties yet." Among the future technologies predicted to replace perpendicular recording once it runs out of steam are heat-assisted magnetic recording (HAMR) and magnetic recording based on patterned media.

3. In a GMR spin valve, one of the magnetic layers is pinned so that its polarization doesn’t switch and the other is free to switch. (Source: Ed Grochowski, IBM Almaden)

Head advances

Of course, the recording media is only half the story in that it must work in combination with a recording head to write the data onto the disk and then subsequently read it. In the past, both read and write functions were performed with a single inductive head, where a small copper coil was used to induce a magnetic field on the disk. The same coil was also able to sense the direction of the field during read operations.

Today, writing is still done with the same type of inductive head, but the reading of the data is now done with a giant magnetoresistive (GMR) head that measures changes in resistance of magnetic metal conductors when a magnetic field is applied. GMR technology is implemented in a spin valve where two thin films of magnetic material are separated by a spacer. In a spin valve, one of the magnetic layers is pinned so that its polarization doesn't switch and the other is free to switch. Figure 1 shows a finished device, with the inductive write head integrated with the GMR head. Figure 2 is a close-up cross-section of the integrated device, indicating the complexity of the structure. Figure 3 provides more detail on the GMR spin valve.

A new type of head called the tunneling MR head (TMR) is also the subject of much research, and areal densities of 100 Gb/in.2 have been demonstrated. The use of a TMR film at this density demands that the junction resistance (resistance per square micron of junction area) be reduced. However, until recently, insulation layers made by conventional film fabrication technologies have shorted, causing deterioration in the MR ratio.

At last year's MMM-Intermag Conference for magnetic recording, TDK Corp. and NEC Corp. independently reported that they have fabricated the insulation layer on an extremely flat film to minimize insulation layer shorts. How flat? Minimizing the interlayer coupling, which increases sensitivity to changing magnetic field, make the "layers very, very flat - less than 3 Å RMS values," said Hari Hegde of Veeco (Woodbury, N.Y.). Veeco plans to soon introduce a new Nexus ion beam deposition system, and is working on ion beam etch and Nexus PVD systems for advanced heads.

Head manufacturing is also challenging process technologies in many other areas, including lithography and metrology. "We're beginning to outstrip the semiconductor industry in terms of the need for finer lithography," Mallary said. "In the past, we've been able to ride the coattails of the semiconductor industry in an effort to reduce the dimensions of these heads. But now our demands are actually exceeding semiconductor industry capabilities."

Two major changes are a move to CVD of some dielectrics and to dry etching, said Ed Ostan of Unaxis (St. Petersburg, Fla.). "Typically, the dielectric in the head for most manufacturers is aluminum oxide. There's continued research in looking at the silicon-based dielectrics - silicon nitride, silicon oxynitride, silicon dioxide - involved with doing CVD deposition in a production mode. Certainly, new dielectric materials as well as new techniques to deposit the dielectrics are key to meeting these new areal densities." Ostan added that many of the new magnetic films being researched are "more dry-etchable than nickel iron," which could mean a shift from ion milling to dry etching in some cases.

"Some GMR stacks are nine or 10 layers, some as thin as 6-7 Å," said Chris Morath, director of metrology and advanced systems development at Rudolph Technologies (Flanders, N.J.). "It's just impossible to imagine any metrology technique out there today that can qualify that whole stack simultaneously. We're able to do some combination of layers, and they can use other techniques to qualify the magnetoresistive properties."

The Table lists the various layers commonly used in read-write heads and their typical thicknesses

The structures are so complex that manufacturers are using focused ion beam analysis to analyze them and use that analysis for process control, said Brian Miller, product line manager for structural nanofabrication at FEI Co. (Hillsboro, Ore.). "We do anywhere from 25-35 separate measurements on the head. If you cross-section the device, you can measure the height, the width at any angle, the write gap, the etch depth, the angle of the etch, and even cut all the way down to the reader and measure how the reader is offset with respect to the writer."

Peter Singer
Semiconductor International