The switching voltage (Vc) variability of STT-MRAM has been investigated with respect to the effects of temperature, magnetic tunnel junction (MTJ) diameter, bias pulse-width (PW) and backhopping (BH). Arrays of STT-MRAM devices having 1 transistor + 1 MTJ bitcell structure with ±10 nm MTJ diameters were tested under an applied ramped voltage waveform sweep at temperatures ranging from −10 °C to 125 °C with pulsewidths of 40 ns to 10 µs. We show that the Vc distributions as a function of temperature and PW are highly skewed. BH of the MTJ bits increases with increasing temperature, longer PW and smaller MTJ diameter. We also demonstrate the switching and BH performances of 40 Mb embedded MRAM macro having different MTJ diameters.

Role of temperature, MTJ size and pulse-width on STT-MRAM bit-error rate and backhopping

Designing advanced single-digit shape-anisotropy MRAM cells requires an accurate evaluation of spin currents and torques in magnetic tunnel junctions (MTJs) with elongated free and reference layers. For this purpose, we extended the analysis approach successfully used in nanoscale metallic spin valves to MTJs by introducing proper boundary conditions for the spin currents at the tunnel barrier interfaces, and by employing a conductivity locally dependent on the angle between the magnetization vectors for the charge current. The experimentally measured voltage and angle dependencies of the torques acting on the free layer are thereby accurately reproduced. The switching behavior of ultra-scaled MRAM cells is in agreement with recent experiments on shape-anisotropy MTJs. Using our extended approach is absolutely essential to accurately capture the interplay of the Slonczewski and Zhang-Li torque contributions acting on a textured magnetization in composite free layers with the inclusion of several MgO barriers. 

/pdf/spin-and-charge-drift-diffusion-in-ultra-scaled-mram-cells-25nco80f.pdf

Spin and charge drift-diffusion in ultra-scaled MRAM cells

We investigate Edge-AI Inference (EAI) architectures based on 22nm FD-SOI embedded-MRAM (eMRAM) using quantized neural networks (QNN) for inference applications in harsh industrial conditions having strong magnetic field and wide operating temperature (-40∼125 °C). We achieved best case test accuracy of 98.99% with Quantized-Convolutional Neural Network (QCNN) and 89.94% with Quantized-Multi-layer Perceptron (QMLP) surpassing prior reported results in literature on MNIST dataset. By exploiting BER resilience of QNN, we show that eMRAM based EAI offers a superior magnetic immunity of ≈ 700 Oe at 125 °C (≈ 98% accuracy) without the use of ECC and significant energy saving of ≈ 14% for QCNN and ≈ 11% for QMLP. A detailed analysis on the tradeoff between retention time, write energy and inference accuracy is presented.

MRAM-based BER resilient Quantized edge-AI Networks for Harsh Industrial Conditions

We present an accurate extension of the coupled spin and charge transport model used to compute the spin-transfer torques in nanovalves to enable the anlysis of ultra-scaled magnetic tunnel junctions. The charge current behavior is reproduced with low conductivity locally dependent on the angle between the magnetization vectors, while for the spin current proper boundary conditions at the tunnel barrier interfaces are introduced. We demonstrate that the experimentally measured voltage and angle dependencies of the torques acting on the free layer are accurately reproduced, and that using our extended approach is key to accurately capture the interplay of the Slonczewski and Zhang- Li torque contributions acting on a textured magnetization. The described approach is successfully employed for switching simulation of recently proposed ultra-scaled MRAM cells. 

Spin Torques in ULTRA-Scaled MRAM Devices

This chapter briefly describes the development history of MRAM, explains the flip mechanism and flip conditions of STT-MRAM from the perspective of the LLG equation, and further extends to the electrical and magnetic properties of MRAM. Then the manufacturing process related to MRAM is described, including but not limited to photolithography, etching, dielectric isolation, contact, passivation and testing technologies. Finally, the development status of STT-MRAM is given as a guideline for IC designers working on MRAM. 

Spin‐Transfer Torque Materials and Devices for Magnetic Random‐Access Memory (STT‐MRAM)

We demonstrate highly reliable and mass-production ready 22nm FD-SOI 40Mb embedded-MRAM for industrial-grade (-40~125°C) applications. This technology having 5x solder reflows compatibility stack has passed JEDEC standard qualification (ECC-OFF) with total reliability failures below the product life-time bit-failure-rate requirement for industrial-grade. Using design-process co-optimization, we show the extended performance to meet -40~150°C product operation for Auto-Grade-1 applications with stable read performance, ~47% reduced read power, data retention of 20yrs (0.1 PPM), read disturb rate of <1 PPM for ~1M cycles with 500Oe field, 1M endurance cycles, and stand-by magnetic immunity (SMI) of ~1400Oe at 25°C and ~500Oe at 150°C (0.1 PPM). With magnetic shield-in package solution, we demonstrate ~4kOe SMI at 25°C for 48hrs of field exposure.

JEDEC-Qualified Highly Reliable 22nm FD-SOI Embedded MRAM For Low-Power Industrial-Grade, and Extended Performance Towards Automotive-Grade-1 Applications

V. Kriegerstein

Papers

JEDEC-Qualified Highly Reliable 22nm FD-SOI Embedded MRAM For Low-Power Industrial-Grade, and Extended Performance Towards Automotive-Grade-1 Applications