ICNCE 2026

Information technology is undergoing a profound transformation: traditional, command-based data processing and CMOS technology, in which algorithms act like recipes to produce deterministic results, are increasingly being complemented and, in some areas, replaced by AI-based approaches. These approaches rely on machine learning and the automated extraction of knowledge from data, yielding results that are inherently probabilistic. Data processing in such systems is based on artificial neural networks (ANNs), inspired by neuroscience, whose operation relies on threshold-like computational elements and artificial synapses. A key challenge of this development is the rapidly growing energy demand of modern computing systems.
In our projects, we investigate memristive devices and systems for neuromorphic computing (NC) that have the potential to operate far more energy-efficiently than conventional computer architectures. The focus is placed on the development and analysis of redox-based memristive elements.
The lecture outlines the physical principles underlying these devices and illustrates how materials science and electrical engineering draw inspiration from the remarkable energy efficiency of the human brain—a topic intensively studied in neuroscience. Insights gained from this research are incorporated into the development of future computer hardware, which can subsequently be exploited in computer science and AI applications. We will demonstrate how such an interdisciplinary approach—bridging neuroscience, materials science, electrical engineering, and computer science—can make a decisive contribution to the realization of energy-efficient neuromorphic systems.
In addition, selected applications of AI will be presented, and their opportunities and risks at both the individual and societal levels will be briefly discussed. Finally, we will reflect on the question of whether future AI systems could develop some form of consciousness, thereby bringing the discussion back to the field of brain research.

Speaker:
Rainer Waser received his PhD in Physical Chemistry from the Technical University of Darmstadt in 1984. He subsequently joined the Philips Research Laboratories in Aachen before being appointed Professor at the Faculty of Electrical Engineering and Information Technology at RWTH Aachen University in 1992. In 1997, in parallel with his university appointment, he assumed the position of Director of the Institute for Electronic Materials at Forschungszentrum Jülich. His scientific work focuses on the physicochemical fundamentals of functional oxide materials — initially on their dielectric and ferroelectric properties, and later in particular on redox-based resistive switching phenomena. His research made substantial contributions to the development of memristive devices, which are now regarded as key candidates for non-volatile memory technologies, computing-in-memory architectures, and neuromorphic hardware systems. Together with Professor Matthias Wuttig and Regina Dittmann, he coordinated the Collaborative Research Center SFB 917 on “Resistively Switching Chalcogenides for Future Electronics,” funded by the German Research Foundation (DFG) from 2011 to 2023 and involving 14 institutes. In 2014, he received the Gottfried Wilhelm Leibniz Prize awarded by the DFG for his interdisciplinary contributions to the physics and materials science of emerging memory technologies. In 2019, he initiated the BMBF-funded structural transformation project “Neuro-inspired Artificial Intelligence Technologies for the Electronics of the Future in the Rhineland Region (NEUROTEC)” and – together with Max Lemme – he organized the ICNCE 2024.

This tutorial is an introduction to neuroanatomy and neurophysiology for engineers and computer scientists. We will work our way down from the whole brain to individual neurons and synapses, focusing on where things are and how they work. No prior training in biology is required.

Speaker:

Abigail Morrison is the group leader of “Computation in Neural Circuits” at IAS-6, Jülich Research Centre and a professor at the Faculty of Computer Science at RWTH Aachen. She holds a master’s degree in artificial intelligence and received her PhD in computational neuroscience in 2006 from the University of Freiburg, Germany. Between 2006 and 2009 she was a scientific researcher at the RIKEN Brain Science Institute in Wako-Shi, Japan; she subsequently held a junior professorship at the University of Freiburg, Germany, as well as a group leadership at the Bernstein Center Freiburg, Germany from 2009 to 2012. In 2012 she moved to the Jülich Research Centre, where, in addition to leading her computational neuroscience group in IAS-6, she was the scientific lead of the Simulation and Data Laboratory Neuroscience at the Jülich Supercomputing Centre until 2023. She was appointed to a professorship at the Ruhr University of Bochum in 2012 until moving to RWTH Aachen in 2020. Her neuroscientific research interests include learning, representation and self-organization in spiking neural networks, with a secondary technical focus on neuromorphic computing and the software ecosystem for high-performance neural network simulation.

Spiking neural networks (SNNs) occupy a unique intersection amid computational neuroscience and neuromorphic engineering: they are both models of biological neural computation and practical candidates for energy-efficient inference on neuromorphic hardware. Yet training them to perform well on real-world tasks has long been a major impediment, owing to the non-differentiable nature of spike generation.
This tutorial provides a self-contained introduction to SNNs and modern methods for training them. We begin with a concise recap of spiking neuron models, focusing on leaky integrate-and-fire neurons, which are widely used in network modeling and provide the mathematical basis for the rest of the session. We then turn to surrogate gradient methods, which resolve the non-differentiability of spikes by substituting a smooth approximation during the backward pass while preserving spike-based dynamics in the forward pass. We discuss both the practical recipe and the theoretical underpinnings.
A recurring practical obstacle when training deep SNNs is initialization: networks initialized naively often operate in saturated or quiescent regimes that impede gradient flow from the outset. We present fluctuation-driven initialization, a principled approach that places networks in a dynamically active, fluctuation-driven regime at the start of training, dramatically improving convergence in both rate- and temporal-coded settings.
Finally, we examine the question of on-chip and biologically plausible learning. We introduce online local learning rules, derived from the surrogate gradient framework, that approximate backpropagation through time while respecting locality constraints compatible with neuromorphic hardware and synaptic plasticity mechanisms observed in the brain.
Throughout, theory is motivated by practical implementation considerations, and connections to neuromorphic hardware deployment are highlighted. Attendees will leave with a coherent picture of the state of the art in SNN training and the tools needed to apply these methods in their own research.

Abstract: Running AI on edge devices requires high efficiency under strict energy and latency constraints. In-memory and near-memory computing reduce data movement and improve efficiency. However, beyond efficiency, ensuring the trustworthiness of AI systems remains a critical concern. This talk introduces Bayesian electronics, where the intrinsic randomness of emerging nanodevices is used to encode probability distributions directly in hardware, enabling on-device uncertainty estimation. I will also present on-chip learning as a way to reduce reliance on cloud-based training, enabling local adaptation, improved privacy, and greater robustness. Recent hardware advances, including hybrid synaptic architectures that decouple inference from learning, will also be discussed.

Bio: Elisa is a Senior Scientist at CEA-Leti. Her primary research interests focus on the development of new technologies for highly energy-efficient, memory-centric computing. In 2022, she was awarded an ERC Consolidator Grant for her project “Heterogeneous integration of imprecise memory devices to enable learning from a very small volume of noisy data.” She has been a member of the VLSI Technical Program Committee (TPC) since 2023. Elisa received her Ph.D. in Electrical Engineering in 2010 through a joint program between the Università degli Studi di Udine (Italy) and the Grenoble Institute of Technology (INPG, France).

Abstract:
Edge AI is entering a new era where functionality is defined not just by efficient inference, but by on‑device learning, continual adaptation, and user personalization. For wearable systems, this shift exposes a fundamental limitation of today’s architectures: energy is no longer spent on computation, but on moving data back and forth across memory hierarchies. As a result, this keynote presents a vision for brain‑inspired, memory‑centric edge AI architectures that rethink where and how edge AI systems operate. Inspired by architectural principles of biological intelligence, such as local learning, event‑driven operation or and tight memory-compute integration, this talk advocates that hardware-software co‑design concepts can overcome the energy and latency barriers of conventional computing systems to conceive the next generation of edge AI wearables.

In particular, the future of edge AI will not be built around bigger models, but around co-designed system architectures that learn continuously, adapt locally, and can operate with brain‑like efficiency. Therefore, by bringing computation to memory and leveraging event‑driven sensing and fine‑grained adaptive learning engines, these new edge AI architectures can unlock ultra‑low power and real‑time adaptation to the specific user context. At scale and considering system manufacturing costs, advances in heterogeneous integration and open design frameworks can enable collaborative, privacy‑preserving intelligence across networks of edge AI wearable devices.

Bio: David Atienza is a Full Professor of Electrical and Computer Engineering, leads the Embedded Systems Laboratory (ESL), and serves as the Associate Vice President for Research Centers and Platforms at EPFL, Switzerland. His research interests focus on system-level design methodologies for energy-efficient multi-processor system-on-chip (MPSoC) architectures for next-generation computing systems and edge AI embedded systems (particularly smart wearables and medical devices) in the Internet of Things (IoT) era. In these fields, he is a co-author of more than 450 publications and 14 patents, and has received several recognitions and awards, among them, the 2024 Test-of-Time Best Paper Award at the IEEE/ACM CODES+ISSS Conference for the most influential paper in system co-design in the last 15 years, the IEEE/ACM ICCAD 10-Year Retrospective Most Influential Paper Award in computer-aided design in 2020, the DAC Under-40 Innovators Award in 2018, and an ERC Consolidator Grant in 2016. He served as President of IEEE CEDA in 2018-2019 and as Chair of the EDAA from 2022 to 2024. He is a Fellow of IEEE and of ACM, and a Member of the Academia Europaea.

Abstract: Various applications demand more and more powerful machine inference in resource-scarce distributed devices. To allow intelligent applications at ultra-low energy and low latency, one needs 1.) custom AI processors, exploiting parallelism and data reuse under strong resource limitations; 2.) efficient ML models, optimized for the target hardware platform; 3.) data-efficient scheduling techniques and algorithm-to-hardware mapping tools. This talk will zoom into such a future of cross-layer optimized AI platforms for edge computing.

Bio: Marian Verhelst is a Full Professor at KU Leuven’s MICAS Laboratories and Research Director at imec Leuven, where she specializes in AI compute chips, hardware acceleration, analog in-memory computing, neuromorphic hardware, and edge AI accelerators. She received a PhD from KU Leuven in 2008, and worked as a research scientist at Intel Labs from 2008 till 2010. Marian is a scientific advisor to multiple startups, member of the board of ECSA, and served in the board of directors of tinyML. She serves as President of the Technical Advisory Board of Italy’s Chips-IT initiative and received the laureate prize of the Royal Academy of Belgium in 2016, the 2021 Intel Outstanding Researcher Award, and the André Mischke YAE Prize for Science and Policy in 2021. Beyond her research, Marian is a dedicated science communicator as a permanent cast member of the monthly Nerdland, a popular Dutch-language science and technology podcast.

Abstract: While artificial neural networks represent a highly successful mapping from neuroscience AI, and clearly capture important aspects neuronal information processing, data from experimental neuroscience strongly suggests that the ANN abstraction omits important biological computational principles. This includes such features as the pulsed nature of neural communication, the diversity and function of neuronal morphology, and the inherently time-continuous mode of operation. To investigate these computational principles, we need to be able to train large and complex networks of spiking neurons for specific tasks. I will show how effective online and approximate learning rules enable the supervised training of large-scale networks of detailed spiking neuronal models, how we can integrate extended temporal delays in a principled and efficient manner, and how these spiking neural network models can be integrated with brain-derived decision-making circuits to operate continuously. As I will argue, this approach opens up the investigation of both network and neuronal architectures based on functional principles, while at the same time demonstrating the potential power and energy efficiency of AI-solutions based on neuromorphic computing as embodied by spiking neural networks.

Bio: Prof. Dr. Sander M. Bohté heads the CWI Machine Learning group, and is also a part-time full professor of Cognitive Computational Neuroscience at the University of Amsterdam, The Netherlands. He received his PhD in 2003 at CWI on the topic of “Spiking Neural Networks” and then spent a Post-doc wat the University of Colorado in Boulder. In 2004, he rejoined CWI as scientific staff. In 2016, he co-founded the CWI Machine Learning group, where his research bridges the field of neuroscience and deep learning, addressing topics like local learning, online learning and learning in deep spiking neural networks.

Abstract: Deep neural networks (DNNs) outperform cortical network models in behavioural performance. DNNs also outperform cortical neuron models in terms of their statistical response properties. Yet, the alignment of DNNs with cortical circuits, cortical structures, and synaptic plasticity is rather limited. I present a functional bottom-up model of cortical pyramidal neurons and cortical circuits that is inspired by self- and cross-attention networks in artificial intelligence (AI). I show how cortical layer 2/3 pyramidal neurons can sustain a key-value memory that is queried by cortical layer 5 neurons. The AI-inspired self-attention is implemented as a gain modulation in cortical pyramidal neurons. Crucially, synapses in the cortical key-value memory are frozen, so that the network is learned purely online, without need of backpropagation through time (BPTT). I show how this online plasticity, in principle,could be implemented via dedicated cortical error neurons. I also show how reward-prediction errors (RPEs), calculated in basal ganglia, help in task-switching. Via gain modulation, the RPEs increase the signal-to-noise ratio in the cortical networks. RPEs further gate the recruitment of hippocampal memories and their inclusion in the cortical key-value memories. Overall, the model aligns well with cortical structures involved in cognition, and offers a mechanistic explanation of inference and learning.

Bio: Prof. Dr. Walter Senn is a computational neuroscientist recognized for his models on cortical computation that connect biological and artificial intelligence. He has a PhD in Mathematics from University of Bern and Freiburg i.Br., and was for research stays at Moscow Lomonossov University (with Prof. Y. Sinai), at the NIH and NYU (with Prof. Rinzel) and at the Hebrew University in Jerusalem (with I. Segev). Since 2006 he is Full Professor for Computational Neuroscience at the Institute of Physiology, University of Bern, where he is Co-Director since 2010.  Senn builds mathematical models of cognitive functions such as learning, memory, attention, perception and lately awareness and consciousness. His models capture morphological and biophysical properties of neurons, synapses and networks, characterized by a close qualitative match to the cortical architecture. He was pioneering models of spike-timing-dependent synaptic plasticity, and algorithms for reinforcement learning in populations of spiking neurons with delayed reward. He also introduced a theory of `learning by the dendritic prediction of somatic spiking’, with extensions to reward-based leaning in multi-compartment neurons. He developed ideas on how learning through error-backpropagation could be implemented by cortical microcircuits in the brain, and with dendritic errors enabling local plasticity rule. His recent work is devoted to the neuronal least action principle from which dynamic laws of neuron and synaptic plasticity are derived in a similar way as the law of motion is derived from the least action principle in physics. The theory links cortical microcircuits with behavioral error minimization, links to artificial intelligence and serves as a basis to design neuromorphic hardware.

Abstract: Biological intelligence is remarkably data- and energy-efficient. We strive to understand and replicate this in artificial intelligence. A key step toward this goal is to develop robust, scalable learning algorithms for physical and biologically inspired neural networks.Achieving this requires rethinking fundamental principles of representation learning and credit assignment. In this talk, I will present recent work from our group on learning algorithms suited for recurrent, noisy, and resource-constrained substrates. These include, but are not limited to, spiking and neuromorphic systems. I will show how local, biologically plausible learning rules can emerge from self-supervised objectives. I will also discuss applied work on training spiking networks for ultra-low-power brain-machine interfaces.

Bio: Friedemann Zenke studied physics at the University of Bonn. He also studied at the Australian National University in Canberra. He pursued his Ph.D. in computational neuroscience with Wulfram Gerstner at the École polytechnique fédérale de Lausanne (EPFL). After his Ph.D., Friedemann joined Surya Ganguli’s group at Stanford as a postdoctoral researcher. He then moved to the University of Oxford as a Sir Henry Wellcome fellow with Tim Vogels. Friedemann is now a senior research group leader at the Friedrich Miescher Institute for Biomedical Research (FMI) and an assistant professor at the University of Basel, Switzerland. His group addresses fundamental theoretical questions about biological neural networks and learning algorithms.

Abstract: Many real-life problems — from combinatorial optimization and constraint satisfaction to inference in energy-based AI models — share a common mathematical structure: finding configurations that minimize a complex, non-convex energy landscape. Boolean satisfiability SAT (or MaxSAT), canonical NP-complete (or NP-hard) problems, exemplify this class: the task is to find assignments of Boolean variables satisfying the maximum number of logical constraints. Despite their ubiquity, no efficient algorithm is known for these problems in the worst case. We show that this broad class of problems admits a unified treatment through continuous-time dynamical systems (CTDS) in the form of ordinary differential equations. Our analog solver operates as an energy-based reasoning machine; however, instead of performing static gradient descent, it generates a deterministic, history-weighted gradient flow on a dynamically evolving energy landscape. Unsatisfied constraints accumulate exponentially growing tension via auxiliary variables, continuously reshaping the effective landscape and enabling escape from spurious local minima — a mechanism fundamentally distinct from both standard gradient descent and stochastic Langevin dynamics. Hardness in this framework appears as transient chaos governed by a chaotic repeller. We will briefly discuss the mathematical foundations of analog, continuous-variable computation in continuous time — its advantages, and limitations. We then present CTDS variants, some inspired by experimentally observed many-body neuronal interactions and discuss implications for energy-based AI reasoning, where the same history-weighted dynamics may offer a principled alternative to stochastic sampling for inference in learned energy landscapes.

Bio: Zoltán Toroczkai is Professor of Physics and Concurrent Professor of Computer Science and Engineering at the University of Notre Dame. His research bridges nonlinear dynamics, complex systems, discrete mathematics, and the physical foundations of computing, with a particular focus on analog approaches to hard optimization problems. Together with Mária Ercsey-Ravasz, he demonstrated that NP-hard constraint satisfaction problems can be mapped onto continuous-time analog dynamical systems in which optimization hardness emerges as transient chaos (Nature Physics, 2011; cover article). He has also made foundational contributions to the network science of brain cortical connectivity (Science, 2013; Neuron, 2018). Prior to Notre Dame, he was a Director-Funded Fellow and Deputy Director of the Center for Nonlinear Studies at Los Alamos National Laboratory. He is a Fellow of the American Physical Society, holds an Erdős number of 2 and has authored over 100 peer-reviewed publications in Nature, Science, PRL, PNAS, and related journals.

Abstract: A large part of the current effort in AI hardware is directed at accelerating linear operations, especially matrix-vector multiplications. Yet the expressive power of artificial neural networks does not arise from linear operations alone. Neural networks are nonlinear function approximators, and their ability to represent complex input-output relations critically depends on nonlinear transformations. This motivates a complementary hardware paradigm in which nonlinear processing itself is implemented directly in physical devices.

Here, I discuss reconfigurable nonlinear-processing units (RNPUs) as silicon-based physical computing primitives for hardware-native nonlinear computation. RNPUs are multi-terminal nanoelectronic devices whose nonlinear input-output characteristics can be tuned by electrical control signals, enabling a single physical substrate to implement many different computational transformations. This approach connects to broader efforts toward adaptive and intelligent matter [1,2] and builds on material-learning concepts [3], but places nonlinear processing at the center.

We have shown that silicon-based RNPUs can perform benchmark classification tasks and that their functionality can be programmed through machine-learning-inspired optimization [4,5]. More recently, we demonstrated gradient descent in materia using homodyne gradient extraction, enabling direct physical optimization of device functionality [6]. We further showed that RNPUs can perform efficient real-time processing of temporal signals, including room-temperature analogue speech recognition [7]. Recent work also demonstrates that RNPUs can serve as physical nonlinear building blocks for Kolmogorov-Arnold Networks, where learnable nonlinear edge functions are implemented in hardware rather than emulated digitally [8].

Together, these results establish RNPUs as a silicon-compatible route toward hardware-native nonlinear computation beyond conventional linear-acceleration paradigms. Recent work by Kareem et al. clarifies the underlying charge-transport mechanism in silicon RNPUs, identifying space-charge effects as the physical origin of their strong and tunable nonlinear response [9]. These insights provide a route toward planar silicon implementations with engineered doping profiles, reducing reliance on etched structures and interface traps while strengthening compatibility with CMOS fabrication.

References
[1] C. Kaspar et al., Nature 594, 345 (2021).
[2] H. Jaeger et al., Nat. Commun. 14, 4911 (2023).
[3] S.K. Bose, C.P. Lawrence et al., Nat. Nanotechnol. 10, 1048 (2015).
[4] T. Chen et al., Nature 577, 341 (2020).
[5] H.-C. Ruiz Euler et al., Nat. Nanotechnol. 15, 992 (2020).
[6] M.N. Boon, L. Cassola et al., Nat. Commun. 16, 10272 (2025).
[7] M. Zolfagharinejad et al., Nature 645, 886 (2025).
[8] M. Escudero et al., arXiv:2602.07518 (2026).
[9] J. Kareem et al., arXiv:2605.13477 (2026).

Bio: Wilfred G. van der Wiel (Gouda, 1975) is full professor of Nanoelectronics, co-director of the BRAINS Center for Brain-Inspired Computing, and co-chair of the Department of Electrical Engineering at the University of Twente, The Netherlands. He holds a second professorship at the Institute of Physics, University of Münster, Germany. His research focuses on unconventional electronics for efficient information processing. Van der Wiel is a pioneer in material learning at the nanoscale and in the development of reconfigurable nonlinear-processing units (RNPUs), realizing computational functionality directly in nanomaterial substrates through principles analogous to machine learning. He has authored more than 125 journal articles, receiving over 15,000 citations.

Abstract: The demand for accurate perception of the physical world leads to a dramatic increase in sensory nodes. However, the transmission of massive and unstructured sensory data from sensors to computing units poses great challenges in terms of power‐efficiency, transmission bandwidth, data storage, time latency, and security. To efficiently process massive sensory data, it is crucial to achieve data compression and structuring at the sensory terminals. In‐sensor computing integrates perception, memory, and processing functions within sensors, enabling sensory terminals to perform data compression (reduce quantity) and data structuring (improve quality). We will overview the framework of bioinspired in-sensor intelligence for artificial vision. We will examine optoelectronic devices that can compress and structure multidimensional vision information, and demonstrate a few vision sensors for different scenarios, including visual adaptation, motion perception, event-driven vision sensors for spiking neural network, and in-sensor spectrometers.

Bio: Prof. Yang Chai is the Chair Professor of Semiconductor Physics of the Hong Kong Polytechnic University. He is an IEEE Distinguished Lecturer, an IEEE Fellow, an Optica Fellow, IOP Fellow, and AAIA Fellow. He is the Director of Research Institute for Artificial Intelligence of Things, the Director of Joint Research Center of Microelectronics, and the Associate Dean of Faculty of Science (Research) at the Hong Kong Polytechnic University. He is also the Chair of Semiconductor Nanotechnology Alliance, the Vice President of the Physical Society of Hong Kong, and an Associate Editor of ACS Nano. He is a receipt of the Falling Walls Science Breakthroughs in Engineering and Technology for his work on “Breaking the Wall of Efficient Sensory AI Systems”, the BOCHK Science and Technology Innovation Prize in the field of AI and Robotics, The Croucher Senior Fellowship, The Ho Leung Ho Lee Foundation Science and Technology Innovation Award, and NSFC Distinguished Scholar. His current research interest mainly focuses on emerging electronic devices. He is the founder of SenseMind.

Abstract: Biological synapses do more than change their strength: they also regulate how changeable they should remain. This “plasticity of plasticity,” or metaplasticity, is thought to help neural systems balance memory retention with flexibility. While the underlying biological mechanisms remain only partly understood, this idea offers a powerful inspiration for neuromorphic AI: a useful learning system should not only update its synapses, but also decide which synapses should stay plastic, which should consolidate, which information should fade, and when new evidence is worth acquiring.

This talk develops that idea through a Bayesian view of metaplasticity. The central proposal is that uncertainty can act as a local synaptic control signal. Rather than treating uncertainty as an external diagnostic computed after training, we embed it into the learning rule itself. Uncertain synapses remain adaptable, confident synapses are protected from unnecessary change, and outdated information can relax through controlled forgetting. This creates a bridge between bio-inspired principles, continual learning, and neuromorphic hardware constraints, where memory is finite, synaptic writes are costly, and stochasticity is unavoidable.

I will first introduce MESU, Metaplasticity from Synaptic Uncertainty [1]. Inspired by the idea that synapses may carry not only a weight but also an “error bar” on that weight, MESU derives a continual-learning rule in which each parameter update is scaled by synaptic uncertainty. A bounded-memory Bayesian formulation adds controlled forgetting, allowing networks to avoid both catastrophic forgetting and excessive rigidity, without relying on explicit task boundaries. MESU therefore turns synaptic uncertainty into a mechanism for balancing plasticity, consolidation, and adaptation.

I will then present BiMU, Binary Metaplasticity from Uncertainty [2]. BiMU brings the same principle to binary Bayesian neural networks, where the connection to neuromorphic systems is especially direct: synapses have minimal precision, updates are expensive, and stochastic sampling is cheap. By preventing binary posterior saturation, BiMU keeps synapses from becoming permanently frozen and preserves useful epistemic uncertainty. This uncertainty can then drive active continual learning: the system requests labels and performs updates only when stochastic network samples disagree.

Together, MESU and BiMU suggest a neuromorphic view of learning in which finite precision, stochasticity, and forgetting are not necessarily limitations, but computational resources.

[1] D. Bonnet, K. Cottart, T. Hirtzlin, T. Januel, T. Dalgaty, E. Vianello, D. Querlioz, "Bayesian continual learning and forgetting in neural networks", Nature Communications, 16(1), 9614 (2025).

[2] K. Cottart, T. Ballet, D. Bonnet, D. Querlioz, “Active Continual Learning with Metaplastic Binary Bayesian Neural Networks”, ICML, 2026.

Bio: Damien Querlioz is a CNRS Research Director at the Centre de Nanosciences et de Nanotechnologies of Université Paris-Saclay and CNRS. His research focuses on novel usages of emerging non-volatile memory and other nanodevices, in particular relying on inspirations from biology and machine learning. He received his predoctoral education at Ecole Normale Supérieure, Paris and his PhD from Université Paris-Sud in 2009. Before his appointment at CNRS, he was a Postdoctoral Scholar at Stanford University and at the Commissariat à l’Energie Atomique. Damien Querlioz is the coordinator of the interdisciplinary INTEGNANO research group, with colleagues working on all aspects of nanodevice physics and technology, from materials to systems. In 2016, he was the recipient of an ERC Starting Grant to develop the concept of natively intelligent memory. In 2017, he received the CNRS Bronze medal. He has also been a co-recipient of the 2017 IEEE Guillemin-Cauer Best Paper Award and of the 2018 IEEE Biomedical Circuits and Systems Best Paper Award.

Abstract: Brain networks exhibit a prominent hierarchical organization at all levels, from anatomical organization to activity patterns. In contrast, artificial deep neural networks are organized sequentially using the end-to-end error backpropagation algorithm for training. This algorithm propagates information forward and backward through the entire network, updating its parameters in tiny steps. While this approach scales well to large networks, it treats them as monolithic “black boxes,” requiring substantial resources. In this talk, I will revisit results on the hierarchical organization of biological neural networks. Based on these results, I will identify principled mechanisms for the next generation of learning algorithms that efficiently use available resources while maintaining robust performance on challenging AI tasks.

Bio: David Kappel is an assistant professor at the Center for Cognitive Interaction Technology (CITEC) Bielefeld University, where he leads the group of Sustainable Machine Learning since 2024. Previously, he was a postdoctoral researcher at the Ruhr-Universität Bochum, the TU Dresden and the University of Göttingen. David Kappel received his Ph.D. in computer science from Graz University of Technology in 2018. His research interests focus on efficient algorithms and models for synaptic plasticity, neural dynamics, Bayesian inference and hierarchical learning models.

Abstract: This talk explores our success and failures with neuromorphic language models and their deployment on Intel Loihi 2. We have achieved the first billion parameter LLMs running on neuromorphic hardware at 2-watts, moving state-of-the-art reasoning from the datacenter to the edge. We have spent a painful amount of time working out when neuromorphic LLMs do and don't make sense, and our findings have flipped some of our assumptions about how sparsity should be used in these models.

Bio: Jason Eshraghian is an Assistant Professor and Fulbright Scholar in the Department of Electrical and Computer Engineering at the University of California, Santa Cruz. He is the developer of snnTorch, a Python library with over 500,000 downloads for training spiking neural networks. He is a dual-appointed IEEE CAS and EMBS Distinguished Lecturer, an Associate Editor of APL Machine Learning, the Chair of the IEEE Neural Systems and Applications Technical Committee, has been the recipient of seven IEEE Best Paper Awards, a Scientific Advisory Board Member of BrainChip, and leads the Neuromorphic Agents Team at Conscium.

Abstract: Modern AI spans an increasingly diverse set of model architectures and computational workflows. Different types of models, such as large language models, multimodal systems, diffusion models, and world models, exhibit distinct computational characteristics. Different workflows such as training, reinforcement learning, and inference also have different demands for compute infrastructure. Despite this diversity, today's AI infrastructure, from hardware to core frameworks and libraries, remains largely homogeneous. To maintain computational efficiency, we impose uniformity of primitives - we force diverse mathematical structures of different models into rigid, dense matrix multiplications (GEMM).

Bio: Natalia Vassilieva is Vice President and Field CTO at Cerebras Systems, leading customer engineering. Her expertise spans hardware and software, with a strong understanding of how algorithms run efficiently on different architectures. Prior to joining Cerebras, Natalia led the Software and AI group at Hewlett Packard Labs and served as the Head of HP Labs Russia. She has directed research in LLM training, NLP, computer vision, and information retrieval. Earlier in her career, she served an Associate Professor at St. Petersburg State University and as a lecturer at the Computer Science Center in St. Petersburg, Russia.

This talk presents a first-principles analysis of the relationship between modern AI workloads and hardware topologies. We examine how different model architectures and computational workflows result in distinct computational patterns, and how those patterns translate into requirements for compute, memory, and communication. We compare alternative hardware approaches, from HBM-centric accelerators to ultra-high-bandwidth SRAM-based architectures, and explore the trade-offs they make across diverse workloads.

The central argument is that no single hardware architecture can efficiently serve the full spectrum of modern AI. As workloads diversify, forcing them onto uniform compute platforms leads to increasing inefficiency, cost, and wasted resources. The next era of AI progress will depend on tighter co-design between algorithms and hardware, enabling increasingly heterogeneous computing systems that align computational structure with the hardware best suited to execute it. Ironically, AI itself may accelerate this transition by reducing the cost of developing specialized hardware and software, making heterogeneity not only technically desirable, but economically practical.

Abstract: Brain-inspired event-based neuromorphic computing is a promising technology for energy-efficient bio-inspired AI. It also enables continuous learning based on local learning algorithms. For maximum energy efficiency, a brain-like in-memory realization is desirable. The Heidelberg BrainScaleS platform is an example of a neuromorphic architecture that combines true in-memory computing with hardware support for continuous local learning. In addition, its analog implementation of the core neuronal operations enables complex neuron models without an energy or area penalty.

For real-world applications as well as neuroscience, certain minimum network sizes are required. To realize the necessary upscaling, BrainScaleS pioneered wafer-scale integration. For future generations of BrainScaleS, this will not be feasible due to the high mask costs of modern semiconductor processes. This talk presents alternative solutions and how they can be utilized by the current BrainScaleS platform in EBRAINS. EBRAINS is a European research infrastructure for neuroscience. Its BrainScaleS neuromorphic service democratizes access to analog neuromorphic computing.

Bio: Johannes Schemmel has been the head of the “Electronic Visions” research group at Heidelberg University since 2000. Since 2008, he has also led the ASIC-Laboratory of the University. Following the untimely passing of Prof. Karlheinz Meier in 2018, he served as the acting chair. Since 2024, he has held the chair of “Neuromorphic Computing Architectures” at Heidelberg University. His research focuses on highly parallel mixed-signal circuits for information processing. In recent years, the analog implementation of biologically inspired neural networks has been his primary research direction. The most prominent example of this work is the Heidelberg neuromorphic BrainScaleS system, which is part of the European EBRAINS research infrastructure for Neuroscience.

Abstract: Language models with recurrent depth, also referred to as universal or looped when considering transformers, are defined by their capacity to increase their computation through the repetition of layers. Recent pretraining efforts have demonstrated that these architectures can scale to modern language modeling tasks while exhibiting advantages in reasoning tasks. This makes their recurrence in depth an exciting additional axis for scaling model performance, separate from the established ‘verbalized’ chain-of-thought paradigm. In this talk, we’ll discuss recent advances, connect to classic methods and explore neuroscience motivations.

Bio: Jonas is a ML researcher in Tübingen, Germany, where he leads the research group for safety- & efficiency- aligned learning. He is a Max-Planck research group leader and Hector-Endowed Principal Investigator at the ELLIS Institute Tübingen. Before this, he spent time at the Universities of Maryland, Siegen and Münster. When it comes to efficient learning, he studies how to build systems that do more with less, from weight averaging techniques to recursive computation approaches that extend model capabilities. with a particular interest in how these systems reason, and whether we can enhance their reasoning abilities while maintaining efficiency. How do we build mechanisms that let these models learn to be intelligent systems? At the core of his research is also the intersection to safety: Can we make models that reason well without sacrificing safety? How do computational constraints affect safety guarantees? Can we design systems where intelligence and safety reinforce each other?

Abstract: Oxide-based memristors are emerging as key enabling technologies for neuromorphic hardware and unconventional computing paradigms. These devices can emulate biological synaptic and neuronal functionalities, while also serving as compact computational units for in-memory processing and reservoir computing architectures [1].
In this talk, I will present an overview of our recent work on two-terminal oxide memristors and three-terminal ionic transistors, also known as electrochemical random-access memories (ECRAMs). The discussion will focus on materials and device engineering, switching mechanisms, and strategies to exploit their intrinsic properties for brain-inspired computing primitives and unconventional information processing.
In the first part, I will demonstrate how the programmable and nonlinear characteristics of non-volatile resistance switching Pt/HfO₂/TiN memristors can be leveraged to implement a memristor-driven circuit enabling single-node reservoir computing [2]. This approach efficiently addresses nonlinear classification tasks and real-time information processing. The second part of the presentation will focus on volatile electrochemical memristors based on Ag/SiOₓ/Pt structures [3,4], and how their properties can be tailored by inserting an ultrathin Al₂O₃ layer (1–2 nm), deposited by atomic layer deposition at the SiOₓ/Ag interface (Figures 1a,b). I will discuss how the interplay between switching kinetics and relaxation dynamics governs device operation, and how engineering multiple relaxation time scales which are crucial for brain-inspired temporal information processing.
Finally, I will introduce our ongoing work on three-terminal ECRAM devices based on WO₃/HfO₂/WO₃ stacks. These devices exhibit bulk, gate-controlled ionic switching, enabling analog, linear, and energy-efficient conductance modulation, which is highly promising for future scalable neuromorphic systems.

[1] A. Mehonic et al., APL Materials 12, 109201, 2024
[2] M. Escudero et al., Adv. Intell. Syst. 8, 2500508, 2026
[3] M. Dutta et al., Adv. Electron. Mater. 10, 2400221, 2024

Bio: Sabina Spiga is Research Director at the National Research Council of Italy (CNR), Institute for Microelectronics and Microsystems (IMM). She received her Degree in Physics from the University of Bologna in 1995 and earned a Ph.D. in Materials Science from the University of Milan in 2002. Her research focuses on the development of memristive devices that exploit ionic and electronic phenomena at the nanoscale to emulate synaptic and neuronal function in hardware, enabling advances in neuromorphic and unconventional information processing.
Spiga has served as Principal Investigator for CNR in several national and Horizon 2020/Horizon Europe projects, including MeM-Scales, Neuram3 and Neurotech, and is currently involved in the European IPCEI ME/CT initiative. She is presently the Editor-in-Chief of the Journal of Physics D: Applied Physics.

Abstract: Electronic devices that emulate the excitability of biological cells hold promise for bioelectronic systems capable of detecting, processing, and responding to physiological signals directly at the interface with living tissue[1]. Conventional silicon-based hardware, however, faces challenges in biointegration due to mechanical rigidity, circuit complexity, and relatively high power consumption. Organic mixed conductors provide an alternative materials platform in which ionic and electronic transport coexist, enabling efficient coupling with biological signals and low-voltage operation. In this presentation, I will discuss the development of organic electrochemical spiking circuits based on organic mixed conductors, including organic electrochemical neurons and other bioinspired excitable systems. These devices exploit ion-mediated processes to reproduce key features of biological excitability, from neuronal spiking[2-5] to more complex action-potential waveforms[6], while enabling event-driven sensing with high temporal resolution and low energy consumption. When integrated with bioelectronic interfaces, such circuits can detect physiological activity and enable responsive stimulation in a closed-loop fashion[7]. These capabilities open new opportunities for adaptive neuroprosthetics and implantable bioelectronics.

References:
[1] P. C. Harikesh, et al., Nat. Electron. 7, 525-536 (2024).
[2] P. C. Harikesh, et al., Nat. Commun. 13, 901 (2022).
[3] P. C. Harikesh, et al., Nat. Mater. 22, 242-248 (2023).
[4] J. Ji, et al., Nat. Commun. 16, 4334 (2025).
[5] P. C. Harikesh, et al., Sci. Adv. 11, eadv3194 (2025).
[6] D. Gao, et al., Nat. Commun. DOI: 10.1038/s41467-026-72584-5 (2026).
[7] C.-Y. Yang, et al., Nat. Sens. 1, 63-72 (2026).

Abstract: Polymer-based ECRAMs have been recently developed to the point of demonstrating outstanding performance at the device level. Indeed, all solid-state ECRAMs switch with frequencies exceeding 50MHz even when moderately scaled, they use less than 100fJ per switching event and can be switched billions of times at temperatures up to 90°C. Furthermore, with a judicious choice of materials, the read current can be as low as a few nAs, as needed for scaling. The next development step involves the fabrication and testing of arrays of ECRAMs for hardware accelerators. This involves thinking about access devices and array design. I will present our progress in fabricating and testing oxide-based transistors as access devices. Oxides are attractive because of their record-low off currents, which are advantageous for state retention. We have devised a fabrication process that allows to monolithically integrate oxide electronics with polymer-based ECRAMs thus enabling hybrid systems. Addressing ECRAMs is inherently more complex than resistive-RAM devices because they are 3-terminal devices. I will show that they can be operated in a 2-terminal configuration, with advantages in term of array size, energy efficiency and speed.

Bio: Alberto Salleo is the Hong Seh and Vivian W. M. Lim Professor in the School of Engineering. He currently serves as the Deputy Director for Science and Technology at SLAC National Accelerator Laboratory.
Salleo earned a Laurea in Chemistry from the University of Rome La Sapienza and an MS and PhD in Materials Science from UC Berkeley. He was a post-doc and then a staff scientist at Xerox PARC from 2001 until 2005. Salleo joined Stanford as an Assistant Professor in 2006. He rose through the ranks and was eventually promoted to Full Professor in 2019. He served as Chair of the Materials Science Department between 2019 and 2025. Alberto won an NSF Career Award as well as the SPIE Early Career Award and the Gores Award for Teaching, Stanford’s highest teaching honor. He has been a Clarivate Highly Cited Researcher in Materials Science since 2015. Alberto is a Knight of the Italian Republic, a Fellow of the Materials Research Society, the European Academy of Sciences, the National Academy of Inventors and the AAAS, as well as a member of Academia Europaea.

Abstract: Neuromorphic devices are transforming healthcare by enabling innovative, low- power, and efficient solutions for biomedical applications. Inspired by the neural architecture and computational principles of the human brain, these systems are particularly suited for wearable and implantable technologies capable of real-time, closed-loop interaction with biological tissue while operating directly at the edge. In this talk, I will introduce the key features of neuromorphic circuits for healthcare applications, highlighting how they enable energy-efficient edge computing for continuous sensing, adaptive processing, and autonomous decision-making. I will present examples of neural computational primitives for biomedical signal processing and discuss their integration into closed- loop systems combining artificial and biological intelligence. By leveraging neuromorphic sensing, spiking neural networks, and edge AI, we can develop next- generation healthcare technologies, from wearable monitoring platforms to implantable neuroprosthetic devices, that improve patient care, support personalized medicine, and revolutionize chronic disease management.

Bio: Elisa Donati received her B.Sc. and M.Sc. degrees in Biomedical Engineering from the University of Pisa and her Ph.D. in Biorobotics from the Sant’Anna School of Advanced Studies. She is currently an Adjunct Professor at the Institute of Neuroinformatics (INI), University of Zurich and ETH Zurich, where she leads the Neuromorphic Closed-loop Systems research group. She is also affiliated with the Neuroscience Center Zurich. Her research lies at the intersection of neuroscience, neuromorphic engineering, and biomedical technologies, with a focus on closed-loop neural interfaces, neuromorphic sensing and processing, and spiking neural networks for real-time sensory encoding and biomedical signal analysis. Her work aims to develop adaptive and energy-efficient neuro-inspired systems for next-generation biomedical and neuroprosthetic applications. Elisa is involved in several international and national research initiatives funded by organizations including the Royal Society (UK), the Swiss National Science Foundation (SNSF), and NATO.

Abstract: The explosive growth of transformer-based AI models and the push toward adaptive intelligence at the edge have exposed fundamental limits of conventional von Neumann hardware, where data movement—not computation—dominates energy and latency. This talk presents recent progress from our group on memory-centric co-design spanning devices, circuits, and architectures to address these challenges for two key AI primitives.
First, we reformulate attention score computation as massively parallel in-memory similarity search using Flash-based Content-Addressable Memory (FlashCAM). High-uniformity amorphous oxide semiconductor Flash devices (>95% yield, 4 V memory window) with optimized speed–retention–endurance characteristics have been realized and integrated into 16×16 CAM arrays. A custom PCB measurement platform with Arduino/Jetson control has been developed to demonstrate matchline discharge dynamics that directly encode similarity scores.
Second, we introduce a family of CMOS-compatible non-filamentary memristors (graphene- to metal-insulator-metal stacks) engineered for BEOL monolithic 3D integration and edge continual learning. Latest devices achieve 100 ns switching at 2.5 V while maintaining >100 s retention, high uniformity via via-hole structures, and low cycle-to-cycle variation that enables verification-free programming. We experimentally validate a deterministic outer-product parallel programming scheme on 6×6 subarrays within 32×32 crossbars, achieving O(1) weight updates. Supported by generalizable compact models and macro architectures that emulate floating-point operations for BF16-quantized LoRA adapters, these primitives enable accurate in-situ LLM fine-tuning with minimal accuracy loss.
Together, these results demonstrate practical hardware pathways that dramatically reduce data movement for attention mechanisms and enable efficient on-device adaptation, offering a cohesive device-to-architecture framework for next-generation AI accelerators.

Abstract: This talk focuses on event-based computing and the broader need to extend current computing architectures toward a new paradigm shift. This shift requires a rethinking of the entire event sensing and processing stack, from sensor design to algorithms and system-level integration.

Recent results in event-based sensing are presented, along with an overview of the key challenges involved in making these systems widely usable and capable of complementing, and in some cases replacing, conventional frame-based cameras.

A unified approach for jointly leveraging frames and events is introduced, enabling more robust and efficient perception systems. The main technical and scientific challenges required to reach this goal are also discussed, including sensor design, multimodal data fusion strategies, algorithmic frameworks, and supporting computational architectures.

Finally, the talk outlines the components that still need to be developed across the full event sensing stack in order to enable scalable, real-world deployment of this paradigm.

Bio: Ryad Benosman is a French scientist and researcher working at the intersection of artificial intelligence, computational neuroscience, computer vision, and neuromorphic engineering. He is internationally recognized for his pioneering contributions to event-based vision, a bio-inspired computational paradigm that departs from conventional frame-based imaging by encoding visual information asynchronously, in direct analogy with the dynamics of biological retinas. His work has played a significant role in shaping the theoretical and algorithmic foundations of event-driven perception systems, now a central topic in neuromorphic sensing and intelligent robotics.

He has a background in pure mathematics, with formal training in mathematical foundations, which has strongly influenced his approach to modelling, abstraction, and the design of computational systems in perception and artificial intelligence.

His research also extends to biologically inspired vision restoration systems, including computational models and hardware–algorithm co-design for retinal prostheses and optogenetic stimulation frameworks. This work focuses on bridging neural computation, sensory encoding, and clinical translation in visual neuroprosthetics. He has contributed to research and development efforts in optogenetic vision restoration at GenSight Biologics, including work spanning system architecture, computational modeling, and translational studies documented in associated scientific publications and preclinical research. He is also a co-founder of Pixium Vision, a company developing advanced retinal implant and vision restoration technologies.

In addition to his academic contributions, he has held senior research leadership positions in industry, including at Meta (formerly Facebook), where he contributed to advanced research in machine perception and artificial intelligence. He is the co-founder of Prophesee, a leading company in event-based vision sensors, and has been involved in Grey Matter Labs, focusing on neuromorphic computation and brain-inspired AI architectures, as well as other ventures in computational sensing and intelligent hardware systems. More recently, he has also co-founded TempoSense, a company working on next-generation event-based sensing and compute systems.

Earlier in his career, following his doctoral studies, he contributed to foundational work in omnidirectional and panoramic vision systems, advancing early approaches to wide-field visual sensing for robotics and autonomous perception.

His research spans neuroscience, robotics, artificial intelligence, and neuromorphic hardware and chip design, with a consistent focus on biologically inspired approaches to perception and intelligent systems. When not working on redefining machine perception, he is usually focused on pushing the boundaries of how machines understand and interpret the world.

Abstract: As artificial intelligence and machine learning technologies are projected to consume increasingly significant amounts of energy, potentially leading to costly, inefficient, and unsustainable systems. Drawing inspiration from the brain’s exceptional energy efficiency, this series of studies seeks to demonstrate that the execution of simple behavioral tasks by biological neuronal networks is dependent upon the precise regulation of energy homeostasis within individual neurons. Furthermore, we aim to show that Artificial Neural Networks incorporating energy-dependent constraints demonstrate enhanced efficiency in learning behavioral tasks. This presentation will cover the theoretical framework of the project and present recent findings regarding the metabolic dependence of cultured neurons, specifically as observed through gene expression and synaptic connectivity. We will also provide electrophysiological data from neural cultures concerning changes in energy availability, alongside advances in computational modeling for Artificial Neural Networks that utilize energy-dependent learning mechanisms.

Bio: Pedro E. Maldonado, PhD, is a Full Professor within the Faculty of Medicine’s Department of Neuroscience at the University of Chile, where he directs the Neurosystems Laboratory. He serves as a Principal Investigator at the National Center for Artificial Intelligence (CENIA) and is an Associate Researcher at the Millennium Institute of Biomedical Neuroscience (BNI). Dr. Maldonado’s academic background includes a Bachelor in Biology and a Master in Biological Sciences from the University of Chile (1986). He earned his Doctorate in Physiology from the University of Pennsylvania in 1993. He subsequently conducted postdoctoral research at the University of California, Davis, Neuroscience Center before joining the University of Chile’s Medical School in 1997. His research primarily explores the neural foundations of visual perception, mechanism of active sensing and brain-inspired artificial intelligence.

Abstract: What does it mean when a brainlike system ‘computes’? This is the question of the semantics of neuromorphic computing. In classical digital computing, several mutually connected workouts of computational semantics have matured to textbook standard. These formal frameworks allow one to characterize, analyse and prove, for instance, whether a computer program actually does what the user meant it to achieve; whether two different programs actually compute ‘the same’ task; which tasks can be ‘programmed’ at all; or what hardware requirements must be met to implement a given program. In brief, semantic theory allows one to analyse how abstract models of computational processes interface with reality – both at the bottom level of the physical reality of computer hardware, and at the top level of real-world user tasks. Neuromorphic computing theory can learn a lot from looking at the digital world, but also needs to find its very own view on semantics.

Bio: Herbert Jaeger studied mathematics and psychology in Freiburg (Germany), got his PhD Computer Science / AI in Bielefeld (Germany) and then did a postdoc fellowship at the (then) German National Research Institute for Mathematics and Computer Science (GMD) in Sankt Augustin (Germany), where he subsequently founded the research unit ‘Modeling Intelligent Dynamical Systems’ (MINDS); then from 2001 to 2019 he served as professor in the CS department of Jacobs University Bremen (Germany). Since 2019 he has been Professor for Computing in Cognitive Materials at the University of Groningen. Current research focus: mathematical foundations for a theory of computing on the basis of non-digital physical substrates. Jaeger retired in June 2025 and now has almost enough time for his research.

Abstract: Unconventional sensing and perception: using event-driven technologies for robots
Biological sensory systems have developed to best capture the properties of surrounding objects and environment that are useful for acting in the world. The physical properties of tactile, visual and auditory sensory organs, and the way neurons encode the characteristics of each stimulus allow our brain to make sense of the world and take appropriate decisions on how to behave.
This is done by a very efficient system that spares the slightest bit of information, to avoid consuming too much energy for each single action. As such, artificial systems have much to learn from biology, to develop cheap solutions that can run in a very small device and at minimum energy cost.
Since the first prototypes of neuromorphic vision sensors and computing devices, part of the community focused its efforts in deploying neuromorphic devices in practical applications, to exploit their intrinsic compression, low latency, high temporal resolution, high dynamic range.
The quest to find the best strategy to exploit neuromorphic engineering is still open, but a lot of progress has been made. In this talk, I’ll describe possible approaches towards the development of neuromorphic perception for robots and discuss the relevance of the development of neuromorphic sensing for touch and other modalities.

Bio: Chiara Bartolozzi is senior researcher tenured at the Istituto Italiano di Tecnologia. She earned a degree in Engineering (with honors) at University of Genova (Italy) and a Ph.D. in Neuroinformatics at ETH Zurich, developing analog subthreshold circuits for emulating biophysical neuronal properties onto silicon and modelling selective attention on hierarchical multi-chip systems. She is currently principal investigator of the Event Driven Perception for Robotics group (www.edpr.iit.it), mainly working on the application of the “neuromorphic” engineering approach to the design of sensors and algorithms for robotic perception. Chiara has participated in a number of EU funded projects, she coordinated the H2020 MSCA-ETN “NeuTouch” and FP7 FET “eMorph”. She co-organised the Neuromorphic Colloquium, a series of online events to build up educational material for the next generation of neuromorphic researchers available athttps://neurotechai.eu/educational/. She is in the scientific board of the Capocaccia Workshop on Neuromorphic Intelligence. She is Editor for NPJ Robotics, IOP Neuromorphic Computing and Engineering, Frontiers in Neuroscience, IEEE JETCAS and TCASI.

Abstract: The shift toward edge computing has enabled real-time data processing closer to the source of data collection, reducing latency and improving overall efficiency. Yet this shift imposes strict constraints on power consumption, physical footprint, and computational performance. These constraints cannot be met by conventional hardware approaches alone. At the same time, logic and memory technologies face increasing complexity as continued scaling necessitates consideration of multiple physical processes to sustain device, circuit, and system reliability. This complexity drives the need for Design-Technology-Co-Optimization (DTCO) and System-Technology-Co-Optimization (STCO) approaches, in which systems, circuits, and devices are co-designed to improve performance and address critical development challenges.

Non-volatile memory (NVM) devices show strong potential for enabling energy-efficient, massively parallel computing, owing to their CMOS-compatible operating voltages and analogue behaviour. Incorporating such emerging memory technologies at the back-end-of-line (BEOL) of CMOS circuits or within 3D array configurations opens significant opportunities for next-generation memory and computing systems. Realising this potential, however, requires addressing a set of critical obstacles: device variability, endurance limitations, fabrication compatibility, scalability, adequate compact models supporting circuit design, and system-level integration. These challenges are further compounded as SRAM and embedded DRAM scaling approach fundamental limits, increasing the relevance of Compute-In-Memory approaches and the emerging memory devices that underpin them.

This presentation highlights how DTCO and STCO provide the appropriate framework for navigating this complexity. By co-designing devices, circuits, and architectures in a holistic manner, and exploiting extended regimes of device behaviour, including low-power operation at low voltages, these approaches are of key importance for successfully integrating emerging memory technologies with CMOS circuits, enabling next-generation systems built on BEOL and 3D integration.

Bio: Prof. Dr. Erika Covi is an Assistant Professor at the Technical University of Munich (Germany), where she leads the Nanoelectronics Circuits and Systems (NCAS) Group. She received her Ph.D. in Microelectronics from the University of Pavia (Italy) in 2014. Following her doctoral studies, she was a researcher at the National Research Council (CNR) of Italy and Politecnico di Milano (Italy). She was later Senior Scientist at NaMLab gGmbH in Dresden, Germany, and Assistant Professor at the University of Groningen (the Netherlands).

Her research focuses on the intersection of emerging memory devices, circuit design, and brain-inspired computing, with an emphasis on design-technology co-optimization (DTCO). Her work explores how the intrinsic physical properties of novel memory technologies can be leveraged to develop energy-efficient computational systems by integrating emerging memory devices with CMOS circuits. She has been awarded with the ERC Starting Grant in 2021 and the ERC Proof of Concept in 2025.

Prof. Dr. Covi has co-authored approximately 50 publications in international journals and conferences, and she has served on the organizing committee of around 10 international conferences. She is a Senior Member of IEEE and serves on the Board of Governors of the IEEE Circuits and Systems Society.

Abstract: Artificial Intelligence (AI) has an energy problem, which is both economically and environmentally unsustainable. The fundamental cause of this problem is the primitive thermodynamic computing principles used in digital processors. The brain offers a fantastic alternative by employing complexity within every neuron, wherein information is processed parallelly by multiple kinetic degrees of freedom, thereby making the system highly efficient. In this talk, I will discuss ways to engineer complexity into the kinetics and material physics within electronic components. I will talk about how an extreme form of complexity, known as Edge of Chaos, enables interesting behaviors within a single component (e.g., neuron-like bursting), which would typically take hundreds of traditional digital components to simulate.

Bio: Suhas Kumar obtained a PhD from Stanford University in Electrical Engineering in 2013. He now heads the R&D department at Rain AI, a semiconductor research startup in the San Francisco Bay area. His research focuses on building brain-inspired electronic devices and designing novel algorithms to use such devices. His research is aimed at reducing AI’s energy consumption by addressing underlying fundamental
scientific challenges. In the past, Suhas led research groups in the semiconductor industry and at US national labs, in addition to helping create two research-based startups. Suhas has over 100 publications in scientific conferences and journals, in addition to 25 granted patents.