Multilevel Nanoarray Spin–Orbit Torque Device for Process-in-Memory Applications

The advancement of data-driven technologies has increased energy and time consumption in data transfer between processors and memory units, limiting further improvement in device performance. This challenge can be addressed by introducing process-in-memory (PIM) architecture, which alleviates data transfer overhead through in-memory computation. In this work, we propose multilevel nanoarray spin-orbit torque (SOT) devices for PIM applications. In a Hall bar structure with multiple ferromagnetic islands, the SOT switching current varies depending on the size or shape of each island. Discrete multilevel states can then be precisely controlled by modulating input current, thus demonstrating analog PIM functionality. Furthermore, the same device also enables logic operations, with pulse currents as digital inputs and multilevel resistances as digital outputs, thereby demonstrating its suitability for digital PIM applications. Notably, multilevel SOT switching can operate with nanosecond current pulses, without requiring external magnetic field, highlighting its potential for ultrafast, energy-efficient PIM platforms.