丁香园AV

Synopsis

Neuromorphic computing applications, such as recognition, classification, and prediction, often demand substantial circuit areas and energy consumption. Stochastic computing offers a promising alternative to classical computing schemes, boasting low area and processing power requirements. However, the challenge lies in the impractical costs associated with random number generation using CMOS-based circuits. Recent attempts have turned to other methods, with spintronics emerging as a standout candidate due to the inherent stochasticity of magnetic tunnel junctions (MTJs). This talk will provide an overview of recent research, focusing on the controllable stochastic properties of MTJs.