Public Research Seminar by Microelectronics Thrust, Function Hub, HKUST (GZ) - Reconfigurable Compute-In-Memory on Ferroelectric Memory Technology: A Cross-Layer Research

Current computing systems are mainly constructed on the von Neumann architecture, where data needs to be transferred to a processing unit from memory components. The latency associated with accessing data from the memory units is a key performance bottleneck for a range of data-intensive applications in the convergence of big data and AI. Several solutions have been proposed to mitigate and overcome this bottleneck, with a prominent one being placing memory and logic units in close physical proximity. While significant progress has been made along those lines at both technology and architecture levels, a transformative approach would be to perform computing functions precisely where the data are stored using memory devices. This is known as compute-in-memory.

In this talk, I will begin by discussing the most recent advancements in the CMOS-compatible ferroelectric memory technologies on aluminum nitride platform. Second, I will present a reconfigurable compute-In-memory system on field-programmable ferroelectric diodes, allowing for on-chip memory, parallel search, and neural network operation. Last, I will discuss the conceptualization and demonstration of a programmable parallel search architecture - analog content-addressable memory (ACAM) on complementary Si-CMOS ferroelectric field-effect-transistor memory. The deployment and acceleration of attention-based deep neural network on ACAM will also be presented.