Summary
- Researchers developed a dual-IMC (in-memory computing) scheme to improve data-processing abilities for AI models.
- The dual-IMC accelerates machine learning processes and enhances energy efficiency in traditional data operations.
- The scheme eliminates the von Neumann bottleneck by storing both neural network weights and input in memory arrays.
- Testing on resistive random-access memory devices showed benefits such as greater efficiency, optimized computing performance, and lower production costs.