Background: Biomedical engineers in particular require fast and powerful data processing systems to process computerized tomography and magnetic resonance imaging scans and other medical imaging technologies. However, current computer data processing technologies are unable to satisfy such requirements. A promising approach to addressing these limitations is processing in memory (PIM). Unfortunately, several issues, such as the compatibility and interconnection of PIM with legacy systems, still remain.
Objective: This paper proposes a standard memory bus-based PIM interface for medical image processing and a PIM platform. The proposed PIM interface can overcome problems of compatibility with legacy systems.
Methods: We will adapt an embedded system based on a commercial application processor (AP) to a medical image system to verify the functions and the performance of the proposed PIM interface. Using the PIM platform, we apply the proposed PIM interface and the AP to execute an image processing program, measure the image processing times, and compare the results of the measurements.
Results: Experimental results show that while the functions of the proposed PIM interface are normal, the processing time of PIM is more than 81% faster than that of the AP.
Conclusion: The experimental results prove that the proposed PIM interface is able to solve problems of compatibility with legacy systems. We foresee that not only the medical image processing field but also a number of academic fields and professional sectors will use PIM in their data-intensive applications.
Keywords: Processing in memory; medical image processing; memory centric computing; near data processing.