Customized Memories for Memory-Centric Computing

4:05 pm - 4:30 pm

In the era of AI and IoT, the volume of generated data is increasing exponentially. Storing, processing, and analyzing these data require substantial computational power and storage capacity, which are enabled by advancement in semiconductor devices, integrated circuits, and systems. Consequently, CPU, GPU, and memory technologies are rapidly evolving to meet these demands. Memory technologies are central to this evolution, playing critical roles in both computation and data storage. Among the mainstream memories, DRAM, SRAM, and Flash continue to attract significant investment and research, aiming for higher performance and greater efficiency. Meanwhile, researchers are also exploring emerging memory technologies such as ReRAM, FeRAM, and MRAM with great innovation and enthusiasm.

Memory-centric computing is emerging as a viable approach to complement traditional processor-centric computing. This paradigm enables AI computations, such as multiply-and-accumulate (MAC), to be carried out in close proximity to or even within memory units. By reducing data movement, this approach enhances energy efficiency and accelerates processing speed. Various memory technologies can support this methodology, tailored to specific application requirements. Notably, 3D Flash memories are distinguished as excellent candidates for AI computing and data search applications. They offer high storage capacity, significant on-off current ratio, tunable read currents with tight distribution, low read disturbance, and robust massively parallel operation, making them ideally suited for the demands of AI workloads.

In this presentation, the speaker will outline the current status, challenges, and future prospects of customized memories for memory-centric computing. Additionally, he will highlight several applications that showcase the potential of these technologies in a data-driven world.

Key Technologies Covered

Memories for computing and data storage
Memory-centric computing, near-memory computing, in-memory computing
AI related computation, multiply-and-accumulate (MAC), vector database search
Retrieval augmented generation (RAG), recommendation system, etc.

Featured Speakers

Dr. Keh-Chung Wang

Head of Emerging R&D, Macronix International Co., Ltd.

Dr. KC Wang earned his B.S. degree in Physics from National Taiwan University and his Ph.D. degree in Physics from California Institute of Technology. Currently, he serves as the Head of Emerging R&D at Macronix International, leading efforts in developing new memory technologies and exploring innovative system applications.

Prior to joining Macronix in 2015, Dr. Wang held various positions at esteemed organizations, including Rockwell, Conexant, OpNext, HRL, UMC, and ASTRI. With four decades of experience in semiconductor device research, integrated circuit design, and management, he has made significant contributions to the field. Notably, he and his colleagues at Rockwell pioneered in developing Gallium Arsenide HBT technology and successfully transferred it to production. The technology achieved widespread use in microwave power amplifiers for cellular phones.

From 1975 to 1985, Dr. Wang conducted research in experimental nuclear and neutrino physics at Caltech and UC Irvine.

Dr. Wang is an IEEE Life Fellow. He received several prestigious awards, including Rockwell's Engineer of the Year Award in 1994 and Chairman's Team Award in 1995, and R&D 100 Award in 1996. He has actively participated in multiple technical, advisory, and community-service committees and served as a guest editor for the Journal of Solid-State Circuits. Dr. Wang has co-authored more than 250 journal and conference papers in the fields of physics, electronic devices, circuits, and systems.