Electronic Design Automation for Next-Generation In-Memory Computing Systems

The exponential growth in the availability of digital data has powered the emergence of data-driven applications like large language models, computer vision, and digital twin. These applications have incredibly high computing demands that exceed the capabilities of today’s high-performance computing systems. Unfortunately, the limitations of scaling silicon technology, and the von-Neumann bottleneck suggest that these demands cannot be addressed through traditional means. To address the pressing challenge of computational scalability and efficiency, new computing paradigms are being explored. One promising solution to this computational challenge is to perform in-memory computation using emerging non-volatile memory (NVM) devices. This approach enables energy efficient execution of computationally expensive operations and promises substantial improvements in throughput. However, the NVM technology is still in its infancy stage. To fully unleash the promises of in-memory computation systems, we need novel electronic design automation (EDA) based solutions tailored to data-intensive applications. In this talk, I will discuss the role EDA can play to close the application-to-devices gap in the in-memory computational stack. In the first part of the talk, I will discuss different in-memory computing styles in analog and digital domains. In the second part of the talk, I will present my representative projects that are focused on improving the scalability, energy efficiency, and robustness of different in-memory computing paradigms. In the final part of the talk, I will discuss future research directions for designing next-generation in-memory computing systems.

Speaker(s): Muhammad Rashed,

Room: EC 550, Bldg: Engineering Center, 115 Library Drive, Rochester, Michigan, United States, 48326, Virtual: https://events.vtools.ieee.org/m/403709