Information Technology (IT) has developed enormously over time, enabling many important new functionalities. Its energy use however is a growing problem, strikingly visible in the form of power-hungry data centers. Energy efficiency is also essential in furthering the performance of processors and battery-powered devices, as well as in facilitating privacy-preserving and cyber-secure applications.
To address this problem, three interrelated aspects are key. First, the energy cost per information-processing step is orders of magnitude higher than required theoretically and considered attainable. Second, current computer architectures necessitate massive, dissipative transfer of data between processors and memory; on-chip, off-chip and/or to the cloud. Third, novel computing paradigms can be a means to reduce complexity and the amount of data processing needed for a certain task.
In an interdisciplinary, holistic approach, NL-ECO aims at scientific and technological breakthroughs on above-mentioned aspects. Our approach is based on the conviction that the future is in heterogeneous computing, combining the best of electronics, photonics, spintronics, thereby extending the existing CMOS-technology toolbox. Within a framework formed by the interconnected concepts of in-memory computing, light-wave-based and neuromorphic computing, we will investigate adaptable (quantum/nano-) materials, optimized circuitry and heterogeneous architecture configurations, as well as efficient scheduling and mapping for reduced data-processing and -transfer. In pursuing this, NL-ECO will present a launch path for knowledge utilisation and outreach activities.
For our ambitions we have an enticing benchmark example, namely the brain. While consuming a mere 20 Watts, it outperforms multi-megawatt supercomputers on specific tasks such as learning. The brain has amazing capabilities to learn by adaptation of its hardware. It makes use of co-location of memory and processing in an intricate 3D networked structure, and is very efficient in selecting the level of accuracy needed to perform a certain task. Such concepts will be rich sources of inspiration for NL-ECO.
