Spin-transfer-torque magnetic random access memory (STT-MRAM) is the most promising emerging non-volatile embedded memory. For most applications, a wide range of operating temperatures is required, for example −40 °C to +150 °C for automotive applications. This presents a challenge for STT-MRAM, because the magnetic anisotropy responsible for data retention decreases rapidly with temperature. In order to compensate for the loss of thermal stability at high temperature, the anisotropy of the devices must be increased. This in turn leads to larger write currents at lower temperatures, thus reducing the efficiency of the memory. Despite the importance of high-temperature performance of STT-MRAM for energy efficient design, thorough physical understanding of the key parameters driving its behavior is still lacking. Here we report on CoFeB free layers diluted with state-of-the-art non-magnetic metallic impurities. By varying the impurity material and concentration to modulate the magnetization, we demonstrate that the magnetization is the primary factor driving the temperature dependence of the anisotropy and thermal stability. We use this understanding to develop a simple model allowing for the prediction of thermal stability of STT-MRAM devices from blanket film properties, and find good agreement with direct measurements of patterned devices.

No abstract

No abstract

Nanocrystalline (NC) textured Mn0.5Zn0.6Fe1.9O4 (MZFO) films, grown at room temperature on both isostructural and non-isostructural substrates, show magnetization values significantly suppressed from epitaxial MZFO films. X-ray absorption spectroscopy and x-ray magnetic circular dichroism measurements indicate larger ratios of Fe3+ to Fe2+ ions on the tetrahedral sites in the NC films compared to the epitaxial films. The magnetization loops of the NC films are shifted by 200−400 Oe at low temperatures. No such effect is observed in the epitaxial films. We hypothesize that the presence of a more structurally disordered, possibly magnetically frustrated, matrix exchange biases the crystalline regions.

No abstract