Memory accesses are a major contributor to energy consumption in modern hardware accelerators. One promising approach to reducing this cost is in-memory computing (IMC), which performs certain computations directly within the memory system rather than transferring data to a separate processing unit. In this seminar, we examine where IMC can provide meaningful energy and performance benefits. We also discuss the inherent limitations and challenges associated with IMC, such as reduced flexibility and potential overheads. We will find that the NN accelerator in certain hearing aids is an ideal case to profit from the benefits of IMC and explore a proposed architecture to do so.
27/3/2026 11:00 - 12:00
ESAT B91.200
