Alessio Antolini

Applications, algorithms, architectures, and analog circuits for in-memory computing based on PCM memories (in collaboration with STMicroelectronics)

 

In-Memory Computing is gaining interest to improve energy efficiency in network nodes for IoT applications by reducing the need for wireless data transmission. Specifically, analog in-memory computing utilizes non-volatile phase-change memories (PCM) to perform matrix-vector product operations. The proposed thesis topics are:

• Study and development of new algorithms to program PCM memory cells to stable and precise conductance values.

• Design of analog circuit blocks (AD/DA converters, voltage references, charge pumps, etc.) to enhance performance and reduce the energy consumption of PCM memories. Tools required: Cadence Virtuoso with STMicroelectronics Design Kit.

• Implementation of neural network inferences on a PCM cell matrix, integrating the digital architecture with AIMC circuits.

• Study and design of a DIMC architecture based on SRAM cells, evaluating performance in terms of TOPS/W. Tools Required: CAD Cadence Virtuoso with STMicroelectronics Design Kit.