IBM changes directions in magnetic memory

Big Blue teams up with TDK to develop a magnetic type of memory chip called STT-RAM; 65-nanometer prototype expected within four years.

At IBM, manipulating magnetic fields are out, and spin-polarized electrons are in.

IBM has linked with Japan's TDK to develop so-called spin torque transfer RAM (random access memory) or STT-RAM. In STT-RAM, an electric current is applied to a magnet to change the direction of the magnetic field. The direction of the magnetic field (up-and-down or left-to-right) causes a change in resistance, and the different levels of resistance register as 1s or 0s.

Under the current plan, IBM and TDK, an integral player in magnetic recording components for hard drives, will develop a 65-nanometer prototype within the next four years.

Grandis, a Silicon Valley start-up, is also trying to commercialize STT-RAM. Grandis is making samples for potential customers in its current facility and hopes to hit the market late next year.

Previously, IBM had been working on a more conventional type of magnetic memory called MRAM. However, the company has been having trouble shrinking the transistors on these chips.

"As you make that device smaller (MRAM), you need to increase the magnetic field, and to continue to write (data), it becomes impractical," said Bill Gallagher, senior manager of exploratory nonvolatile memory at IBM. "To scale beyond 65-nanometer, we have to find a new mechanism to write information."

The 65-nanometer process is currently used to make processors today, though most other chips are made on older processes like 90-nanometer or even 13-nanometer. (65 and 90 nanometers refer to the average feature size inside the chip. A nanometer is a billionth of a meter. Smaller features lead to faster, and often more energy-efficient, chips.) IBM has made MRAM prototypes, but on older manufacturing processes.

IBM has not released MRAM chips commercially. Freescale Semiconductor has released MRAM chips commercially, but has recently expressed doubt about the technology's longevity. At the Flash Memory Summit last week, Freescale's David Bondurant said MRAM "may not go beyond" 65 nanometers.

STT-RAM and phase-change memory are probably the two leading candidates of nonvolatile memory to go commercial in the future, Gallagher said. STT-RAM is faster, but phase change is denser.

STT-RAM may also last longer, he said. In phase-change memory, microscopic bits on a chip are heated up to several hundred degrees Celsius. Heating changes the crystalline material to amorphous. The difference on how the host computer interprets the crystalline and amorphous surfaces is registered as a 1 or a 0 in the computer.

Intel and STMicroelectronics have worked extensively in phase change. Some sources in the memory business believe that the joint venture between the two, Numonyx, will soon announce plans on how it intends to bring phase change memory to market. (Phase change has been discussed as a memory technology since 1970 but has not yet come to market.)

Why does IBM care about memory chips when it makes its money mostly from services and servers? Big Blue also makes chips and licenses intellectual property. Developing nonvolatile memory will give it a necessary ingredient for making chips for itself or others.

"You can't make a system on a chip without some kind of nonvolatile memory," Gallagher said.

Featured Video