Neuromorphic pERception framework for event-baseD radars (NERD)
2024 – 2028 750k Euros (total)
Radar sensing is traditionally used in weather forecasting, air traffic control, surveillance and security, but it has the potential to replace cameras in several edge applications thanks to advancements in neuromorphic technologies. The main advantage of radar over cameras is its ability to operate under all weather and light conditions, accurately measure speed and distance, and preserve users’ privacy. However, current Frequency Modulated Continuous Wave (FMCW) radars produce huge amounts of data to support radar-based applications at the edge, and the current artificial intelligence approach based on conventional neural networks is too expensive in energy and memory.
To these fundamental roadblocks, nature has found better solutions in, e.g., bats’ echolocation. Bats are much more efficient than traditional radar-based systems in detecting, identifying, and fast adapting with a tight energy budget. This is due to the unique architecture of the sensory pathway and processing substrate in their brains.
In NERD, we aim to augment our radar perception systems with bio-inspired principles and signal processing architectures. This can unlock the effective integration of radar into mobile embedded devices, allowing efficient and pervasive radar-based applications at the edge. Concretely, we propose three major innovations.
1) At the radar front-end, novel bio-inspired event-based and sparse sampling methods compress and encode temporal relations in the radar signals into sparse streams of spikes (i.e., binary events mimicking neurons’ action potentials.
2) In terms of data processing, novel spiking neural network pipelines are designed to perform advanced radar signal processing tasks such as echolocation, tracking and classification of objects.
3) Finally, new scalable and efficient neuromorphic hardware architectures are developed to achieve online learning and adaptation. These will provide extreme energy efficiency with robust cognitive-like computing capabilities.
With sensing & computing closely integrated, our new perception approach will open new perspectives for radar sensing and AI at the edge, in key applications such as autonomous vehicles and robotics, human-computer interfacing, and future living spaces with contactless healthcare monitoring.
Project Funded by:
Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO)
Self-healing Neuromorphic Systems (SNS)
2024 – 2028 1.2M Euros (total)
As the world witnesses the emergence of applications powered by Artificial Intelligence (AI) in almost every edge device, there is an urgent need for ultra-low-power (ULP) edge AI processors to offload the computing closer to the source of data generation to address the limitations (e.g., latency, bandwidth) of cloud or centralized computing. This can only be realized if we can make edge AI processors at least 100 times more energy-efficient while offering sufficient flexibility and scalability to deal with AI, which is a fast-moving area. SNS aims to achieve these targets by taking a holistic approach with innovations at all design stack levels to realize novel energy-efficient hardware with radically new brain-inspired concepts that equip edge AI processors with autonomous self-healing capabilities, compensating for external and internal disturbances.
To achieve this ambitious goal, SNS brings together a top Dutch consortium consisting of both academic and industry partners and enables synergy between five scientific and engineering disciplines, with machine learning taking inspiration from neuroscience for self-healing circuits and engineering disciplines such as electrical engineering and microelectronics device technology strongly interacting for system architecture and integration.
Project Funded by:
Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO)
Innatera B.V.
Stitching IMEC Nederland
In-material sensing, learning and computing (IMAGINE)
2024 – 2028 1.2M Euros (total)
Artificial Intelligence (AI) empowers many applications in virtually all aspects of life. More and more devices, machines, and sensors connect to the Internet of Things (IoT), which creates large amounts of data. Even fast networks sometimes struggle to transfer data to a data center or the ‘cloud’ for analysis, demanding incredible power. Solving this high energy consumption and connectivity crisis is only possible by developing intelligent hardware for edge-computing systems, allowing for decentralized data processing close to a machine, device, or IoT sensor. To create these systems, we must embed multiple functionalities in fast, compact devices that can learn from local data and consume little power. IMAGINE aims to design a novel and innovative CMOS-compatible energy-efficient framework for sensing and computing at the edge. It is capable of exploiting the in-material properties of emerging silicon-based nanodevices at ultra-low power. Specifically, IMAGINE aims at embedded and pervasive computing for vision edge-AI applications, which is essential in the automotive industry. We will realize dopant-atom network processing units (DNPUs) and employ them to execute in-material sensing, learning, and computing necessary to run deep learning workloads with a fraction of the energy required today (>100X improvement). IMAGINE hence aims to co-design device technology, analog, and digital electronic circuitry, algorithms, and systems to empower many applications, from reactive systems and autonomous driving to industrial inspection. IMAGINE is a collaboration between UT, TU/e, Toyota, and IMEC, which will contribute to ensuring a leading position in the Netherlands in the rapidly emerging field of edge-AI technology.
Project Funded by:
Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO)
Toyota Motor Europe
Stitching IMEC Nederland
Combining SYnthetic Biology & Neuromorphic Computing for CHemosensory perception (SYNCH)
2024 – 2028 1.2M Euros (total)
Like the human brain, neuromorphic computing relies on data sampled from the outside world.
In the human brain, sensory neurons provide information about light, touch, sounds, taste, and smell.
Using synthetic biology, we want to equip artificial microelectronic systems with one of these senses: the sense of smell, or olfaction.
Biological olfaction outperforms traditional chemical technologies in detection limit, specificity, response time, coding capacity, robustness, size, and power consumption.
This outstanding performance is mainly due to the unique architecture of the olfactory pathway that has evolved over millions of years in all living species, from tiny insects to large mammals, and has produced membrane proteins with specialized channels that selectively recognize odor molecules.
These naturally selective interactions are entry points to biological signaling.
Today, the field of synthetic biology can engineer an artificial version of such entry points with the goal to fully control and leverage the abilities of biological systems. Moreover, neuromorphic electronic circuits can replicate biological neural systems’ computational properties using microelectronic devices and emerging nanomaterials’ physical properties.
Thus, we aim to develop an integrated platform for chemical sensing by integrating, for the first time, neuromorphic electronic systems coupled with synthetic biological mediums.
SYNCH aims to combine synthetic biological pathways, neuromorphic electronic circuits, and bio-constrained multichannel decoding algorithms to develop an efficient and accurate neuromorphic chemical perception system inspired by the natural olfactory system. The goal is to enable real-time decoding of biological interactions with small molecules through temporal spiking data, facilitating fast, distributed, and online chemosensing identification and classification.
SYNCH is a collaborative initiative involving three prominent institutions: CAU Kiel University in Germany (Jan Steinkühler), the University of Bern in Switzerland (Kevin Max), and Eindhoven University of Technology (F. Corradi). The project’s objective is to advance the field of olfactory sensing using synthetic biological assemblies. In this endeavor, CAU Kiel University will investigate these assemblies’ creation, specifically for olfactory sensing purposes. Meanwhile, the University of Bern will contribute by focusing on computational modeling to enhance the understanding of the processes involved.
Project Funded by Volkswagen Stiftung
Memory technologies with multi-scale time constants for neuromorphic architectures (MeM-Scales)
2016 – 2019 3.9M Euros (total)
Novel neuromorphic chip technology to support learning at multiple timescales. Neuromorphic computing is an umbrella term given to a variety of efforts to build computation emulating the neural structure of the human brain. The EU-funded MeM-Scales project will work on the development of an innovative platform that will serve as a basis for future products combining extreme power efficiency with human cognition capabilities. The focus will be on building novel memory and device technologies, and autonomous learning algorithms that support on-chip learning over multiple timescales, for both synapses and neurons. This new kind of computing technology has the potential for use in advancing distributed environmental monitoring, implantable medical diagnostic microchips, wearable electronics and human-computer interaction.
Framework of key enabling technologies for safe and autonomous drones’ applications (COMP4DRONES)
2016 – 2019 29.4M Euros (total)
Advances solutions for safe and autonomous drones. The use of drones today is expanding as it reduces costs and offers environmental benefits. However, existing technologies could make drones’ usage harmful for humans, vehicles and properties. SESAR JU, in charge of EU research in air traffic management, suggested that further investments and motivations are needed for the safe use of drones. The ECSEL JU-funded project COMP4DRONES will work on safe software and hardware drone solutions aligned with SESAR objectives. Coordinated by Indra, COMP4DRONES brings together 49 partners from 8 countries aiming to build an ecosystem that will support the systemization and safety of drone platforms, reliable communications, cost-efficient and safe design of drones. COMP4DRONES will deploy applications in five domains: transport, construction, surveillance and inspection, logistics and agriculture.
NEUral computing aRchitectures in Advanced Monolithic 3D-VLSI nano-technologies (NEURAM)
2019 – 2023 4.1M Euros (total)
We propose to fabricate a chip implementing a neuromorphic architecture that supports state-of-the-art machine learning algorithms and spike-based learning mechanisms. With respect to its physical architecture this chip will feature an ultra low power, scalable and highly configurable neural architecture that will deliver a gain of a factor 50x in power consumption on selected applications compared to conventional digital solutions; and fabricated in Fully-Depleted Silicon on Insulator (FDSOI) at 28nm design rules. In parallel the project will be validating the modules to realise RRAM synapses both planar and in a 3D monolithic structure. We will complete this vision and develop complementary technologies that will allow to address the full spectrum of applications from mobile/autonomous objects to high performance computing coprocessing, by realising (1) a technology to implement on-chip learning, using native adaptive characteristics of electronic synaptic elements; and (2) a scalable platform to interconnect multiple neuromorphic processor chips to build large neural processing systems. The neuromorphic computing system will be developed jointly with advanced neural algorithms and computational architectures for online adaptation, learning, and high-throughput on-line signal processing, delivering
1. an ultra-low power massively parallel non von Neumann computing platform with non-volatile nano-scale devices that support on-line learning mechanisms
2. a programming toolbox of algorithms and data structures tailored to the specific constraints and opportunities of the physical architecture;
3. an array of fundamental application demonstrations instantiating the basic classes of signal processing tasks.
The neural chip will validate the concept and be a first step to develop a European technology platform addressing from ultra-low power data processing in autonomous systems (Internet of Things) to energy efficient large data processing in servers and networks.