The field of memristive computing and in-memory processing is witnessing significant advancements, driven by the need to overcome the limitations of traditional computing architectures. Researchers are exploring innovative solutions to improve the efficiency, scalability, and performance of memristive devices and in-memory computing systems. One notable direction is the development of advanced algorithms and techniques for optimizing the deployment of deep neural networks on in-memory computing hardware, such as the use of load-balancing strategies and adaptive learning-based methods. Another area of focus is the improvement of phase change memory (PCM) technology, including the reduction of write energy consumption and the enhancement of cell endurance. Furthermore, researchers are investigating new coding mechanisms and data storage methods to minimize write energy and improve the overall performance of PCM-based systems. Noteworthy papers in this area include:
- SMART-WRITE, which proposes a method that integrates neural networks and reinforcement learning to dynamically optimize write energy and improve performance.
- MDM, which introduces a post-training deep neural network weight mapping technique to reduce parasitic resistance nonidealities in memristive crossbars.
- WIRE, which presents a new coding mechanism to reduce write energy consumption in phase change main memories.
- PIMfused, which enables fused-layer dataflow for end-to-end CNN execution in near-bank DRAM-PIM, optimizing cross-bank data transfers and improving performance.