Variability tolerant reading of nanomagnetic energy minimizing co-processor

A recently proposed novel nanomagnetic co-processor harnesses the quadratic Hamiltonian of a system of coupled nanomagnets in order to solve quadratic optimization problems. The key principle here is “Let physics do the computation” in the sense that the relaxation physics of a grid of nanomagnets directly solve the optimization problem [1]. More interestingly, our preliminary research suggests that the performance of this co-processor is independent of problem size. Relaxed magnetic states of the co-processor translate the output of the original problem. Since the framework of this coprocessor is similar to that of STT-MRAM memories, we can leverage the heterogeneous integration with CMOS technologies in order to access, control and read the cells. This paper focuses on an efficient reading of the magnetic cells and also addresses the effects of resistance variations. Based on the framework, the read mechanism should be able to distinguish between the inplane single domain and vortex domain states. Unlike STT-MRAM, the circular magnets do not have shape anisotropy but, here we show that with an additional pre-amplifier circuit, we are able to improve sense margin by at least 73%. A comparative analysis of variation tolerance between two reading schemes was done which shows that an additional pre-amplifier aids in handling the effects of process variations too.