Georgia Institute of Technology

Information processing utilizing the spin degree of freedom of the electron may pave the way for the next generation computer. In this paper we have analyzed a non-local spin torque device to obtain its performance and energy dissipation as a function of the length of the interconnection between the transmitter and the receiver nanomagnet. Impact of the spin relaxation in the interconnect material on the overall performance and energy dissipation is analyzed. It is shown that spin relaxation lengths of several microns (~8 μm) are highly desirable in order to reduce the energy dissipation substantially and to increase the performance of the circuit. Nanomagnets with dimensions on the order of 15nm × 20nm × 20nm are critical for logic application since the reduction in the volume of the nanomagnet reduces the amount of electrical injection current needed to rotate the magnetization from parallel to anti-parallel orientation or vice-versa. It is shown that for a Cobalt nanomagnet of volume 6000 nm3 the switching time due to spin-torque can be ~1-2 ns with an energy dissipation of 107kBT for electrical injection current of 25μA and spin relaxation length of 8μm. Using stochastic wire length distribution models, the upper bound on the circuit size of the spin-based logic is also obtained for a fixed percentage of gates being dedicated as spin boosters. It is shown that for a spin relaxation length of 8μm and Rent’s exponent p = 2/3, the maximum number of gates in the spin based logic block can be ~2000 if the maximum number of spin boosters is limited to 200. Some possible approaches are quantified to reduce the delay of the circuit while maintaining low energy operation.