The key to the recent success of machine learning lies in the power of computing hardware, a huge amount of data to learn from, and bio-inspired algorithms. However, the existing electronics hardware is failing to process and store the ever-growing learning data and algorithm weights in a fast, cheap, and energy-efficient way. The major bottleneck lies in the low speed and high energy cost incurred when data is transferred between memory storage and logic processing units. One promising approach to solving this bottleneck is a high-density on-chip cache based on magnetic memory. However, R&D of new memory materials across academia and industry remain a trial and error process, lacking frameworks that link material-level chemical design, device-level electrical properties, and application-level energy and delay.
In this talk, Dr. Li will talk about his recent works on two bottom-up frameworks that could bridge materials, devices, and applications to guide memory materials, devices, and circuits research in the future.