Bringing together calculation and memory: the issues
Computing in/near memory

By definition, "memory bound" applications are those whose execution time is determined by memory data waiting times.

There are many examples of such applications. For example, an object classification algorithm in augmented reality typically processes millions of sample images or video clips and performs classification on high-definition video streams in real time. Applications in AI, machine learning and data analysis are limited by memory. Many fields such as radio astronomy, materials science, chemistry, health sciences, etc. generate huge quantities of data.

Bringing calculations closer to memory can boost performance considerably. The how questions are as follows:

• ...