Effect of Nanomagnet Geometry on Reliability of Energy-Efficient Straintronic Spin Neuron and Memory: A Size-Dependent Study

In this letter, a model of strain-induced nanomagnet at room temperature is established to explore the effect of size error on the reliability of straintronic memory and spin neurons. The results show that the relationship between input voltages and 180° magnetization switching probability (activation function) of the nanomagnet has a size dependence. The slope of the activation function is related only to the aspect ratio of the nanomagnet, and the thickness error will not affect its slope. The slope of the activation function will decrease as the aspect ratio increases. At the same time, the offset of the activation function increases with the increasing energy barrier (increase in thickness or decrease in aspect ratio) of the nanomagnet. We found that a slight dimensional error will also significantly reduce the reliability of straintronic memory. Increasing the fabrication accuracy or changing the input voltages can solve this problem. We further studied the relationship between the size error and the reliability of spin neurons by using different size spintronic neural networks for handwritten digits recognition under the same input voltages. We found that the size error has a minor effect on the spin neuron's neurologic computing function. A thickness error of 3 nm or a width error of 8 nm will not affect the recognition accuracy of the spintronic neural network. In other words, straintronic spin neurons are more tolerant to dimensional errors and more reliable than strain memory, and its fabrication accuracy requirements can be lower. These findings can provide important guidance for the design of straintronic spin neuron and memory.