MRS: Integrating magnetic devices

Moore's Law has driven semiconductor industry dynamics for years. It allows IC manufacturers to reconcile to apparently conflicting goals, increasing profits while still offering better performance. The data storage industry has seen similar improvements. According to J. A. Katine of IBM's Almaden Research Center, speaking at this week's Materials Research Society Fall Meeting in Boston, storage density roughly doubles every year. To support that increase, the drive heads must shrink by more than 30% each generation. For data density of 40 Gb/in², critical dimensions are in the neighborhood of 200 nm (0.2 microns). Within two or three years, head CDs will be even smaller than those seen by IC manufacturers.

At nanometer-scale feature sizes, magnetic behavior depends on edge effects, crystal orientation, surface energy, and other properties that can be ignored at larger scales. Head manufacturing needs tighter process control and a better understanding of the relationships between processing and properties.

As Katine explained, manufacturing methods are either additive or subtractive. Additive methods like electroplating or liftoff start with a template, then build the structure on top of it. Electrodeposition can create multilayers from a multi-component solution by simply changing the deposition voltage. It's impossible to electroplate insulating barrier materials, and difficult to control the thickness and purity of layers tightly enough.

Liftoff processing uses sputtering to deposit a series of layers on the substrate. Windows in a sacrificial layer expose the device regions. Removing the sacrificial layer takes away excess material. High aspect ratio features are difficult to build this way because sputtering will coat the sidewalls of the deposition trench. Accordingly, it's not clear that liftoff deposition will work for feature sizes below 100 nm.

Because of these limitations, Katine believes that only subtractive processes will be commercially viable for nanometer-scale magnetic devices. Subtractive methods resemble those used in IC manufacturing: a mask exposes areas to be patterned. Etching or milling removes material from the exposed regions. After mask removal, an additional layer may be deposited into the resulting trenches. The deposit/pattern/etch cycle repeats as many times as needed to build the device. These methods are time- and capital-intensive, compared to additive processes, but allow better process control

Removal of magnetic materials in subtractive processes is problematic. Wet chemical etching causes corrosion, while reactive ion etching is often impossible because many magnetic materials do not form volatile etch products. Both focused ion beam milling and argon ion milling can redeposit milled material. Katine prefers argon ion milling because it causes less edge damage. However, erosion of the mask can be a problem if the milling depth is close to the device width.

Making magnetic devices by lithographic methods suggests the possible integration of magnetic and semiconductor devices on a single substrate. While such a combination would reduce device size and improve performance, it would also raise new processing concerns. As Dexin Wang of NVE Corporation explained, ICs often have significant topography, leaving a difficult substrate for deposition of magnetic layers. The materials used in magnetic devices are very different from, and often chemically incompatible with, semiconductor materials. Processing conditions can create problems, too. For example, the plasma steps used to manufacture giant magnetoresistive (GMR) heads can damage ICs. The damage can be annealed out at high temperatures, but the GMR structure can't tolerate the resulting diffusion.