Spin as State Variable for Computation: Prospects and Perspectives

Recent experiments on spin devices like magnetic tunnel junctions (MTJ's), domain wall magnets (DWM) and spin valves have led to the possibility of having very high density on-chip memories and logic. While the possibility of having on-chip spin transfer torque memories is close to reality, several questions still exist regarding the energy benefits of spin as the state variable for logic computation. Latest experiments on lateral spin valves (LSV) have shown switching of nano-magnets using spin-polarized current injection through a metallic channel such as Cu. Lateral spin valves with multiple input magnets connected to an output magnet using metal channels or domain wall magnetic strips can be used to mimic "neurons" . The spin-based neurons can be integrated with CMOS and other devices like domain wall magnetic strips or PCM’s to realize ultra low-power data processing hardware based on neural networks (NN), and are suitable for different classes of applications like, cognitive computing, programmable Boolean logic and analog and digital signal processing. In this short course, I will first discuss the advantages of using spin (as opposed to charge) as state variable for both memory and logic and then present how a cellular array of magneto-metallic neurons, operating at terminal voltages ~20mV, can do efficient analog computation for applications such as image sensing and processing.