Hot Electronics Get Magnetic Cool For Less Money

“Conventional electronic CMOS devices are great for logic operations but not so good for working memory,” explains Bernard Dieny, who leads this project funded by the European Research Council (ERC). “Magnetic storage is much better, because of its ability to keep the written information even when the electrical supply is switched off. In HYMAGINE we deposited MTJ memory structures (MRAMs) directly onto commercial CMOS semiconductor wafers and tested the results with great success.”

Basic MTJs have two magnetic layers separated by a thin layer of magnesium oxide. In one magnetic layer the magnetic polarity is fixed, in the other ‘free’ storage layer it can switch. The junction uses ‘spin transfer torque’ to write information, whereby electrons flowing in the device are ‘spin polarised’ and can switch the polarity in the storage layer between two (binary) states. Read operations rely on measuring the resistance through the MgO layer, which is higher with opposed polarities and lower with aligned polarities.

“When testing read/write operations in our junctions we investigated several important properties,” says Dieny. “First we demonstrated that CMOS/MTJ hybrids can operate at industry-standard speeds of around 1GHz. We found they consume a fifth of the energy needed by conventional all-CMOS systems, so they use significantly less power.

“A further critical property is the ‘endurance’ of the junction, which is the number of read/write voltage cycles it can support before failure becomes likely. Standard flash memory such as USB drives will support 100 000 cycles, but we found our hybrids have an endurance of 1015 cycles – almost unlimited for practical purposes!”

Material matters

As endurance is such a critical property for eventual take-up, the HYMAGINE team investigated the physical mechanisms causing device failure. They found that electrons tunnelling through the MgO layer are trapped at lattice defects. Trapping and untrapping of electrons can lead to high stresses in the layer leading to early material breakdown.

“We established that the density of defects, such as incorporated water molecules, must be kept low. Already a number of equipment suppliers are adapting their vacuum equipment to reduce background H2O pressures with an eye on growing markets for MTJ devices,” explains Dieny. “We also found that the endurance of a newly manufactured device can be predicted using a measure of voltage background noise. This is a significant result for chip-makers who can use such measurements as quality control steps in volume manufacturing.”

HYMAGINE also developed advanced computer-based modelling and design tools for CMOS/MTJ hybrids and incorporated these into widely-used industry-standard software packages. Building on this work, a new company eVaderis was set up to offer spintronic design services, and eventually devices to the semiconductor world.

Encouraging crosstalk

“There is too little communication between the ‘microelectronics’ and ‘magnetism’ communities in the semiconductor world, and this is holding back spintronic applications,” says Dieny. “This is why we launched annual summer schools in Grenoble on MRAM technologies – bringing researchers and engineers together to learn more about spintronics.”

Dieny is also taking spintronics further in a new ERC project called MAGICAL, which will add communications and sensor functions to low power CMOS/MTJ hybrids. “If the ‘Internet of Things’ is to advance, then low power devices are a must,” he explains. “Wearable computers, solar-powered sensors, connected pacemakers – they all demand low power solutions, and magnetism-based devices can offer these as HYMAGINE showed.”

Bernard Dieny’s achievement in the field of MRAMs was recognised with the award of the Adrien Constantin de Magny Prize by the French Académie des Sciences in 2015.