Digital electronics is suffering from fundamental challenges in improving its energy efficiency. At the current pace of growth of AI, if supported by digital electronics, by 2027 it will reach a carbon footprint the size of Brazil and consume nearly 150 TWh per year. For reference that’s enough energy to power every single smartphone on Earth for 40+ years
Therefore, research into alternative computing paradigms, as well as a new generation of experts in such paradigms is critically needed. This is confirmed by two of the major industrial players in the field, HPE and NVIDIA, as well as leading startups (Spincloud Neurobus, Albora), directly involved in the consortium.
Uniting brains
The doctoral network will focus on exploring new computing paradigms, both in electronics and photonics that will provide next-generation computing architectures.
To achieve this goal, Da Ros brings together experts in electronics and photonics but also in computer science/machine learning, theoretical physics/dynamics systems and neuroscience.
The latter is especially relevant as the novel computing paradigms are brain-inspired, also called neuromorphic, as brains are the most energy efficient computers we know of.
Beyond the research aspect, the key focus of the doctoral network is to train a class of 15 doctoral candidates in the 5 disciplines:
- Electronics
- Photonics
- Computer science
- Nonlinear dynamics
- Neuroscience
And provide them with experience both in the academic and industrial sectors.
Great minds think alike
MINDnet tackles a very active research field, so it will leverage on the existing ecosystem of European research projects on the topic. It is unique in its multidisciplinary dimension, thus providing the holistic perspective required to push the state of the art in neuromorphic computing.
At DTU, two doctoral candidates will focus on photonic implementations, especially based on the concept of reservoir computing, a machine learning architecture particularly suited for implementations using photonic hardware.
We will leverage the strong collaborations across the MINDnet consortium to merge our photonic expertise with our collaborators’ unique understanding of machine learning, nonlinear dynamics and neuroscience, to develop energy-efficient and scalable photonic hardware for computing.
