This PhD project, part of the REACT MSCA Doctoral Network, aims to develop an energy-efficient compute-in-memory (CIM) architecture using gain-cell memory for real-time edge learning, addressing power, latency, and memory bandwidth issues with reliable fault detection.
InformationSelf-awareness in humans is an innate capability, arising from the brain’s ability to process a multitude of sensory inputs. Emulating this functionality in electronic systems, commonly referred to as neuromorphic computing holds the potential to create highly intelligent machines capable of supporting a wide range of everyday applications, from autonomous vehicles to smart navigation systems. However, realizing neuromorphic computing in practice presents significant challenges, particularly in the areas of energy efficiency, reliability, and security.
The
REACT Marie Skłodowska-Curie Actions (MSCA) Doctoral Network addresses the above mentioned challenges by developing a neuromorphic platform that is inherently self-aware in terms of energy consumption, secure operation, and system reliability.
Are you enthusiastic about the opportunity to join the REACT Doctoral Network, be involved in impactful research whilst providing your career an international perspective and boost?
As part of the REACT initiative, 15 early-stage researchers (ESRs) will be trained through a comprehensive, interdisciplinary program spanning material science, device physics, computer architecture, hardware prototyping, compiler design, simulation and emulation tools, as well as cybersecurity, reliability, and system verifiability.
The objective of this PhD project is to develop a gain-cell memory-based compute-in-memory (CIM) architecture to enable energy-efficient online learning at the edge. Conventional edge devices are limited by constraints in power, latency, and memory bandwidth, which pose significant challenges to real-time learning when using traditional von Neumann architectures. By leveraging gain-cell memory, which offers high density and low leakage, and integrating mixed-signal computation directly into memory arrays, the proposed approach significantly reduces data movement and energy consumption. To ensure dependable operation under real-world conditions, efficient fault detection and recovery mechanisms will also be evaluated and integrated, addressing the susceptibility of analog and in-memory computing to noise, process variation, and soft errors. The primary objective is to design a CIM system capable of performing key learning operations, such as vector-matrix multiplication and weight updates, within the memory itself, thereby enhancing energy efficiency and computational throughput. This architecture will be optimized for lightweight, adaptive learning tasks commonly encountered in edge scenarios, such as sensor fusion.
REACT offers a uniquely structured training environment, combining academic excellence with industrial collaboration. ESRs will benefit from close mentorship by leading researchers and industry experts, while also developing essential skills in scientific writing, research ethics, time management, and entrepreneurship. By the conclusion of the REACT project, participants will be well-equipped to pursue impactful careers across academia and industry, with the REACT program serving as a strong foundation for their future success. More information about the project can be found here
https://project-react.eu/.