ENERGIZE is one of the four projects selected through a joint EU-South Korea call in February 2024, focused on advancing critical but pre-commercial semiconductor technologies, including heterogeneous integration and neuromorphic computing. The project aims at establishing an internationally leading consortium to tackle the challenges of energy-efficient computing in the age of big data and AI.
Addressing the Energy Challenge of AI
The explosive growth of data collection—from mobile devices and autonomous vehicles to IoT and Industry 4.0—demands highly efficient circuits and algorithms for data processing and storage. Artificial Intelligence (AI), especially through Artificial Neural Networks (ANNs), is central to addressing these challenges.
However, training large-scale ANNs on conventional processors results in significant energy consumption, contributing to environmental concerns. As traditional computer architectures struggle to keep pace with the requirements of future AI applications, including multimedia data analysis and simulations of complex systems, the need for innovative computing technologies has become critical.
Neuromorphic hardware, inspired by the human brain, represents a promising solution to these challenges. Unlike traditional Von Neumann architectures, where computing and memory are physically separated, neuromorphic chips integrate computation and memory directly within the same hardware. This approach, known as computing-in-memory, eliminates the energy-intensive data transfer between processors and memory while enabling highly parallel processing.
Neuromorphic hardware has the potential to solve AI tasks with energy efficiency improvements of several orders of magnitude compared to conventional hardware.
The Bet on Two-Dimensional Materials
Research on neuromorphic hardware is still in its early stages and substantial effort is still being dedicated to identifying the most promising materials and approaches for developing the new hardware.
The project ENERGIZE places its bet on 2D materials as a foundation for neuromorphic hardware.
These materials promise exceptional properties, including low switching voltage, short response times, and high power-efficiency. Furthermore, their unique surface characteristics, such as high crystallinity and the absence of dangling bonds, make them particularly suited for seamless integration with existing semiconductor technologies, including back-end-of-line (BEOL) processes.
To demonstrate the potential of 2D materials for neuromorphic computing, the project ENERGIZE proposes a holistic approach that integrates
- material growth
- single device fabrication and characterization
- multiscale simulations and modelling
- development and characterization of large-scale synapse arrays and neuromorphic systems.
This multilevel interconnected research will enable a full investigation of innovative concepts based on 2DMs. In particular, it will allow to understand how each device and array based on 2DMs operates, and how it can be refined for an optimal performance.
Specifically, ENERGIZE aims at demonstrating:
- wafer-scale growth of 2D materials for neuromorphic devices,
- reliable fabrication and characterization processes of two- and three-terminal devices,
- development of arrays of 2D devices compatible with existing technologies,
- efficient inference and training of neural networks in crossbar arrays,
- standardized benchmarking methods for neuromorphic devices and circuits.
International Collaboration for Strategic Impact
ENERGIZE’s multilevel interconnected research is enabled by the complementary expertise of the project partners, which covers the entire semiconductor value chain. This integrated approach has the potential to serve as a blueprint for the emergence of a European – Korean collaborative network for the development of advanced neuromorphic hardware – a field of immense strategic relevance both for Korea and the EU.