This paper demonstrates a proof-of-principle implementation of a weighted spin torque nano-oscillator (WSTNO) as a programmable building block for neuromorphic computing systems. The WSTNO, a spintronic circuit composed of magnetic tunnel junctions (MTJs), uses non-volatile magnetic memories as synapses and a non-linear spin torque nano-oscillator (STNO) as a neuron. The system demonstrates non-linear output based on weighted input sums using three MTJs, achieving an output power exceeding 3 µW and frequencies of 240 MHz. The MTJs are fabricated from a multifunctional stack in a single process, enabling reduced footprint and compatibility with CMOS integration.

T. Böhnert, Y. Rezaeiyan, M. S. Claro, L. Benetti, A. S. Jenkins, H. Farkhani, F. Moradi, R. Ferreira