In recent years, there has been an increasing need for dedicated devices that act as components of the brain for use in non-von Neumann-based architectures known as artificial neural networks (ANNs). Furthermore, to transition from the large, power-consuming supercomputer paradigm used to complete brain like tasks, such as pattern recognition, to a "wearable" edge-computing-based artificial intelligence (AI) paradigm, compact and efficient hardware is required. In this report, we describe one such approach towards hardware-based ANNs using recently reported spintronics technology. We give a systematic explanation on a procedure of an ANN-based associative memory operation using spin-orbit torque-controlled devices. On the basis of previously obtained results, we then elaborate on the specific roles and requirements of spintronics devices in the demonstration system. We also describe several challenges regarding efficient and reliable operation of the devices that is clarified by additional measurements of endurance properties of the device as a function of operation current. The results indicate that limitation of the maximum operation current to prevent the decay of device properties provides negligible device-to-device variation after hundreds of iterations, allowing for a reliable associative memory operation.