A vast majority of modern computing systems enforce a strong separation between processing (CPUs, GPUs, DSP, DLAs) and memory (DRAM, Flash, etc) units. With the explosion of data-intense applications such as high fidelity real-time multimedia processing and machine learning the boundary between these two elements has become a performance bottleneck. Additionally, the energy cost of moving data back and forth often exceeds that of the computation alone.
By implementing parts of the computation within the memory devices, Processing-In-Memory (PIM) architectures address these fundamental issues. Moreover, memristive devices and Resistive RAM (ReRAM) cross-bars constitute a promising alternative to implement both energy-efficient non-volatile storage media and PIM kernels.
Here at the Chair for Software for Systems on Silicon (SSS), we are developing models and simulation tools to enable the architectural exploration of novel memristor-based PIM solutions. To validate and calibrate our models, a comparison against real HW prototypes is of utmost importance. The goal of this project is to create a custom measuring system for this purpose, consisting of current sensing devices, ADCs and an MCU.