Global RTC (Revolutionary Technology Center) in SK hynix R&D is showing its Research Leadership and Future Value Position in Emerging Memory-Centric Computation Device at the 16th International Memory Workshop (IMW) from May 12^th to 15^th, 2024 in Seoul, Korea. It is sponsored by the IEEE Electron Devices Society and brings the memory community together in a workshop environment to discuss the memory process and design technologies, applications, market needs and strategies. Be sure to take a look at the papers that received the most views on the Global RTC website.

“Improvement of MAC Accuracy using Oxygen Diffusion Barriers in Resistive Synaptic Cell Arrays” by Dr. Youngjae Kwon. HfO₂-based resistive synaptic cell (RSC) arrays with 16-level analog properties are successfully demonstrated for the MAC operation. Al₂O₃ layers are introduced to improve weight variations, and increase MAC accuracy of RSC arrays. Fully integrated 256Kcell 1T1R cross-bar arrays show the high MAC accuracy and feasibility of RSC array-based A-CiM device for energy efficient AI edge applications.

“Realistic Noise-aware Training as a Component of the Holistic ACiM Development Platform” by Dr. Jihun Kim. AI inference accuracy can be reduced by the non-idealities of resistive cell array in A-CiM device. Novel noise-aware training process is introduced to mitigate the realistic cell array variation for securing high accuracy. The results provide a standard guide line for how each non-idealities contribute to inference accuracy, and how the neutral network should be trained to optimize with the A-CiM device.

“Modeling and Demonstration for Multi-level Weight Conductance in Computational FeFET Memory Cell” by Dr. Won-Tae Koo. Advanced weight conductance model in FeFET synapse cell is generated by combining a ferroelectric threshold switching & conduction behavior. The model is successfully demonstrated through the processed FeFET cell that shows multi-level conductance values (≥16-level/cell) in low conductance range with stable retention. The FeFET cell can be one of candidates for power efficient A-CiM applications.