ETH researchers have measured the timing of individual write operations in a new type of magnetic storage device with a resolution of less than 100 picoseconds. Their results are relevant for the next generation of main memories based on magnetism.
At ETH Zurich’s Department of Materials, Pietro Gambardella and his colleagues are studying the memories of tomorrow. They should be fast, store data reliably for a long time and be cost-effective. So-called magnetic “random access memories” (MRAM) achieve this squaring of the circle by combining fast switching via electric currents with permanent data storage in magnetic materials. Several years ago, researchers were already able to show that a certain physical effect – the spin-orbit torque – enables particularly fast data storage. Now, Gambardella’s group, together with the R&D center IMEC in Belgium, has succeeded in resolving the exact dynamics of a single such storage event over time – and making it even faster with a few tricks.
Magnetising with single spins
In order to store data magnetically, one has to reverse the direction of magnetization of a ferromagnetic (i.e. permanent magnetic) material to represent the information as a logical value, 0 or 1. In older technologies, such as magnetic tapes or hard disks, this is achieved by magnetic fields generated in current-carrying coils. Modern MRAM memories, on the other hand, directly use the spins of electrons, which are magnetic, similar to small compass needles, and flow directly through a magnetic layer as an electric current. In Gambardella’s experiments, electrons with opposite spins are spatially separated by the spin-orbit interaction. This in turn creates an effective magnetic field that can be used to reverse the direction of magnetization of a tiny metal point.
“We know from previous experiments in which we scanned a single magnetic metal spot stroboscopically with X-rays that the reversal of magnetization occurs very quickly, in about a nanosecond,” says Eva Grimaldi, a postdoc in Gambardella’s group. “But these were average values averaged over many reversal events. Now we wanted to know exactly how a single such event occurs and show that it could work on an industry-compatible magnetic storage device”.
Time resolution through a tunnel crossing
To achieve this, the researchers replaced the isolated metal point with a magnetic tunnel junction. Such a tunnel junction contains two magnetic layers separated by an insulating layer only one nanometer thick. Depending on the spallation – along the magnetization of the magnetic layers or in the opposite direction – the electrons can tunnel through this insulating layer more or less easily. This results in an electrical resistance that depends on the orientation of the magnetization in one layer to the other and thus represents “0” and “1”. From the time dependence of this resistance during a reversal event, the researchers were able to reconstruct the exact dynamics of the process. In particular, they found that the reversal of magnetization occurs in two stages: an incubation stage, during which the magnetization remains constant, and the actual reversal stage, which takes less than a nanosecond.
“For a fast and reliable storage device, it is important to minimize the temporal fluctuations between the individual reversal events,” explains Gambardella’s doctoral student Viola Krizakova. Based on their data, the scientists therefore developed a strategy to keep these fluctuations as small as possible. To this end, they changed the current pulses used to control the magnetization reversal so that two additional physical phenomena were introduced. The so-called spin-transfer torque and a short voltage pulse during the reversal phase now reduced the total time for the reversal event to less than 0.3 nanoseconds, with temporal fluctuations of less than 0.2 nanoseconds.
“All this combined, we have found a method that allows data to be stored in magnetic tunnel junctions with virtually no errors and in less than a nanosecond,” says Gambardella. The collaboration with the IMEC research center also made it possible to test the new technology directly on an industry-compatible wafer. Kevin Garello, a former postdoc from Gambardella’s laboratory, produced the chips with the tunnel contacts for the experiments at ETH Zurich and optimised the materials for them. In principle, this would make the technology ready for immediate use in a new generation of MRAMs.
Gambardella emphasizes that MRAM memories are particularly interesting because, unlike conventional main memories such as SRAM or DRAM, they do not lose their information when the computer is switched off, but are still as fast. However, he admits that the market for MRAM memories does not currently demand such high write speeds because other technical bottlenecks such as power losses due to high switching currents limit access times. In the meantime, he and his colleagues are already planning further improvements: They want to reduce the size of tunnel crossings and use other materials that use electricity more efficiently.