<ul data-eligibleForWebStory="false"><li>Researchers from various institutions have published a technical paper on using racetrack memory for in-memory computing of CNN inferences in embedded systems.</li><li>Racetrack memory is a non-volatile technology that offers high data density fabrication, making it suitable for in-memory computing, but challenges exist in integrating arithmetic circuits with memory cells.</li><li>The paper proposes an efficient in-memory CNN accelerator designed for racetrack memory, including fundamental arithmetic circuits for multiply-and-accumulate operations and co-design strategies to enhance efficiency and performance while maintaining model accuracy.</li><li>The work aims to address the challenges of building efficient in-memory arithmetic circuits on racetrack memory within area and energy constraints, catering to embedded systems for CNN inference.</li></ul>

Accelerator Architecture For In-Memory Computation of CNN Inferences Using Racetrack Memory

Discover more