Project A01 – New Photoswitches for Integration in Adaptive Nanosystems | Ravoo
Bastian Stövesand
Molecular photoswitches have proven to be versatile tools in application-oriented disciplines such as supramolecular chemistry, where they are used to construct and functionalize organogels, non-covalent polymers, switchable catalysts, surfaces, and nanostructures. However, many of these applications rely on organic solvents, which limits access to systems in aqueous environments, such as host–guest interactions, photopharmacology, or hydrogels.
In this part of the project, classical photoswitches such as azobenzenes, arylazopyrazoles, and diazocines will be functionalized with water-soluble substituents. Subsequently, these compounds will be investigated with regard to their surface activity and their behavior in host–guest chemistry. In addition, the structural motif of diazocine will be further modified to improve its spectral properties and to enable photoisomerization using longer-wavelength light. The synthesis and characterization will be the main focus of this project.
Project A03 – Stereospecific Photoswitching Emissive Materials with Chiral Memory | Fernández
Alfonso Jose Schwalb Freire
The aim of this project is to design and develop advanced emissive materials capable of enhancing circularly polarized luminescence (CPL) through stereospecific photoswitching processes. Upon irradiation with light, these materials are expected to undergo controlled structural transformations that selectively generate enantioenriched molecular states, thereby amplifying their chiroptical emission properties.
A key feature of these systems is their ability to retain a form of “chiral memory,” preserving the induced asymmetry over a measurable timescale. Importantly, this chiral bias is reversible: upon thermal stimulation, the system relaxes back to its thermodynamically favored racemic state. By integrating light-driven stereochemical control with thermally induced reset mechanisms, this project seeks to establish dynamic, switchable platforms for tunable chiral photonic applications.
In this project we work on developing theoretical models for suspensions of refractive light-driven particles with a symmetry breaking in shape and/or refractive index profile. While models such as Active Brownian Particles (ABP) only consider the close-range repulsive interaction between particles, refractive particles can also be subject to long-range interactions mediated by the light field as well as feedback effects created by external control of the light field via optical means. This makes them a viable building block for adaptive matter.
When it comes to theoretical descriptions of classical particle systems, there are two general approaches:
Microscopic models, that describe the dynamics of each individual particle. Examples of this are Langevin dynamics1, Brownian dynamics1 or Ornstein-Uhlenbeck particles2
Field theories, that describe the dynamics of order parameter fields like density and polarization. Example of this are (Active) Phase Field Crystal (PFC) models3, Dynamical Density Functional Theory (DDFT) models4, (Active) Model B5 or (Active) Model H6
While microscopic models are typically easier for formulate based on experimental observations of individual particles, they give less insight into collective phenomena like phase transitions emerging in many-particle systems. Field theories on the other hand are harder to formulate, but give a much better theoretical grip on collective effects.
Besides the analytical methods connected to the models mentioned above we also make heavy use of numerical methods. At the level of particle dynamics this includes molecular dynamics simulations of many-particle systems as well as ray optic simulations to determine the interaction between refractive particles and a light field. At the level of field theories this mainly includes solving of partial differential equations by various means such as finite-difference or finite-element methods.
The numerical methods described above can require very large amounts of computational resources. To increase efficiency and make simulations of larger systems computationally feasible we are also invested in modern approaches to high-performance computing like offloading computations to GPUs, novel approaches to safe parallelism (Rust) or JIT-compilation via domain specific languages.
1. T. Schlick, Molecular Modeling and Simulation (Springer New York, 2002)
2. L. L. Bonilla, Physical Review E 100 (2019)
3. M. te Vrugt, J. Jeggle, R. Wittkowski, New Journal of Physics 23, 063023 (2021)
4. M. te Vrugt, H. Löwen, R. Wittkowski, Advances in Physics 69, 121–247 (2020)
5. R. Wittkowski et al., Nature Communications 5 (2014)
6. A. Tiribocchi, R. Wittkowski, D. Marenduzzo, M. E. Cates, Physical Review Letters 115 (2015)
Project B01 - Propulsion of Light-Responsive Nano- and Microsystems | Denz
Matthias Rüschenbaum
In our project, we design and investigate artificial light-propelled micro-swimmers. The propulsion of these particles is based on refraction of light with a net force resulting from asymmetric particle shapes and symmetry-broken refractive index profiles. Our task is the experimental realization of these particles and their subsequent light-controlled propulsion. For fabrication, two-photon polymerization (TPP) in a direct laser writing (DLW) setup is employed. Together with numerical simulations of our B01 team partners, the geometries, refractive index profiles as well as the complexly structured light used are optimized for e.g. highest velocities, followed by investigations of a high number of these particles, thereby creating artificial colloidal swarming structures. We also envisage mixtures of particles with different features including light-activated shape-changes. The ultimate goal of this project is the realization of memory effects and intelligent behavior in dense solutions of light-propelled micro-swimmers through structured illumination-based delay and added feedback
Project B02 - Adaptive Polymer Morphologies Through Reversible Block Fragmentation | Gröschel
Yorick Post
In our contribution to CRC 1459, we will develop dissipative block copolymer nanostructures and explore their application as sensors, actuators, and memory in synthetic adaptive and intelligent systems. Block copolymers consist of two or more covalently linked incompatible segments capable of self-assembling into complex nanostructures, such as micelles, vesicles & cubosomes. Driven by differences in chemical affinity, like solubility, the resulting nanostructure is primarily defined by the lengths of the various blocks. We are designing block copolymers that can actively alter their composition through energy-driven block fragmentation. By incorporating and tuning orthogonal and multivalent fragmentation mechanisms the stability, lifetime and self-assembly behavior of the block copolymer aqueous solution can be adjusted. The resulting out-of-equilibrium nanostructures can only exist under a sufficient energy supply. These nanostructures will be stabilized into dynamic steady states by regulating the energy supply through the implementation of intrinsic feedback mechanisms.
Proejct B04 - Multistimuli Sensing with Memory and Feedback Function using Photoswitchable Proteins and Coordination Chemistry | Wegner
Alice Casadidio
We aim to create responsive matrixes such as hydrogels, which are sensitive to various inputs (from different colors of visible light to redox and pH). By combining photoswitchable proteins and coordination chemistry, we aim to achieve the formation of such materials, pursuing the development of memory due to multiple stimulation. Furthermore, we aim to investigate whether is possible to integrate signals in order to generate feedback. Together, these building blocks will give us access to processing molecular information through hydrogels.
Project B04 - Multistimuli Sensing with Memory and Feedback Function using Photoswitchable Proteins and Coordination Chemistry | Wegner
Saskia Frank
We will develop a hydrogel with multistimuli responsive crosslinks that can sense diverse input signals, process them following a chemically defined logic and respond with an output signal. Crosslinks will be mediated by photoswitchable proteins that convey responsiveness towards light of different wavelengths and metal coordination complexes which are sensitive towards changes in pH, redox potential and the presence of small soluble molecules. Some signals can also be integrated into the hydrogel matrix, thus creating a memory, in which I am particularly interested in my project.
Project B04 - Design and Development of Materials with a Pulsatile Response | Wegner
Gereon Otte
Our aim is to develop new and innovative adaptive materials capable of responding to complex optical signals. To this end, we utilise a wide range of different photoswitchable proteins that work together within a closed-loop negative feedback system. Such materials provide a further tool for advanced light-sensitive applications in the fields of synthetic biology, biomaterials and biosensing.
Project B04 - Multistimuli sensing with memory and feedback function using photoswitchable proteins and coordination chemistry | Wegner
Bas Wennemar
We aim to create new photoswitchable proteins to realize non-invasive spatiotemporal control of cell-material interactions. These proteins can be used in a wide variety of applications in the field of biomaterials, biosensing, and fundamental cell biology studies.
Project B05 - Investigating Molecular Forces in Focal Adhesions, Hemidesmosomes and Adaptive Hydrogels | Grashoff
Theresa Mösser
The aim of our project is the generation of a conceptually novel, biosynthetic material in which mammalian cells are utilized as information-processing elements that sense, integrate, and feedback on mechanical stimuli. Together with the Strassert and Trappmann groups, we manufacture and characterize 3D hydrogels, in which mammalian cells bestow the hybrid material with a mechanical memory. This system is monitored by fluorescent biosensors that are purified and integrated into the cell-matrix hydrogel, so that mechanical signals can be quantified with fluorescence lifetime imaging. By gradually increasing the complexity of the hybrid material, responsive and adaptive features will be incorporated.
Project B05 - Role of Cellular Mechanotransduction in Cell Adhesion and Migration | Trappmann
Inka Schröter
The aim of our project is the generation of a conceptually novel, biosynthetic material in which mammalian cells are utilized as information-processing elements that sense, integrate, and feedback on mechanical stimuli. Together with the Strassert and Grasshoff groups, we manufacture and characterize 3D hydrogels, in which mammalian cells bestow the hybrid material with a mechanical memory. This system is monitored by fluorescent biosensors that are purified and integrated into the cell-matrix hydrogel, so that mechanical signals can be quantified with fluorescence lifetime imaging. By gradually increasing the complexity of the hybrid material, responsive and adaptive features will be incorporated.
Photonic memory and computing devices are an emerging technology to be used for machine learning and artificial intelligence applications. We use chalcogenide phase-change materials (PCM) to implement memory functionality in the integrated photonic circuits. In our research, light from a waveguide couples evanescently to a PCM patch which influences the transmission through the waveguide depending on the phase state of the PCM. By decreasing the size of the PCM volume, the switching speed can be increased while the needed switching energy can be decreased. Additionally, we use plasmonic nanoantennas to enhance the interaction between the electric field and the PCM. Therefore, we develop a method to fabricate nanometer-sized PCM cells embedded in the gap of gold dimer nanoantennas on a waveguide. These devices are promising candidates for the use in material learning, machine learning in materio, and pattern recognition applications.
Project C03 - Self-Assembly of Hybrid Nanostructures for Neuromorphic Electronics | van der Wiel
Marc Beuel
Previous research1 has shown that disordered networks of functionalized nanoparticles can be configured to behave like Boolean logic gates and classifiers. The functionalized nanoparticles act as single-electron transistors, i.e. strongly nonlinear periodic switches, while the organic ligands act as tunable tunnel barriers that add memory functionality to the network. The purpose of this research project is to build on these results and enhance the functionality by introducing various novel organic ligands and magnetic nanoparticles to enhance the addressability and to introduce memory to tackle new time-dependent problems and realize artificial neural networks that will pave the way for new applications of energy-efficient in-materio computing.
Reference:
1. Bose, S., Lawrence, C., Liu, Z. et al. Evolution of a designless nanoparticle network into reconfigurable Boolean logic. Nature Nanotech10, 1048–1052 (2015)
Project C03 - Kinetic Monte Carlo Model for Computing Functionalities in Nanoparticle Networks | Heuer
Jonas Mensing
The theoretical underpinning of the experimentally studied nanoparticle networks is investigated by developing a physical model and subsequent simulations. For this purpose, a highly optimized parallel C++ code is being developed in order to model the charge transport within the electrically tunable network stochastically, i.e. with a Kinetic Monte Carlo approach. Requirements for computing functionalities such as Boolean logic and memory functionalities based on intelligent materials are examined. Therefore, statistical and data-driven tools are being developed to investigate the effects of different materials and system sizes. A close comparison with corresponding experiments on nanoparticle networks will be performed.
Project C04 - Spin Wave Systems for Reservoir Computing | Pernice
Dmitrii Raskhodchikov
Today, scientists are actively looking for a replacement for conventional electronics. One of the promising areas is spintronics - where information is transmitted not directly by electrons, but by means of spin. This approach allows operations to be completed faster and significantly improves the energy efficiency of devices. The propagation length of spin-waves in amorphous yttrium iron garnet (YIG) is sufficient for the use of this material in electronics and it combines well with existing materials both during operation and in production. The quanta of spin-waves are magnons: the dynamic eigen-excitations of a magnetically ordered body. Analogous to electric currents, magnon-based currents can be used to carry, transport and process information. The use of magnons allows the implementation of novel wave-based computing technologies free from the drawbacks inherent to modern electronics, such as dissipation of energy due to Ohmic losses. We will realize adaptive magnonic networks in a complex system comprised of a large number of coupled spin-waveguides, which transform the input of electrical data into spatiotemporal patterns in a high-dimensional space using nonlinear interference of spin-waves.
Project C05 - Artificial Intelligence for Intelligent Matter in Nanophotonics | Risse
Jonas Konrad
Photonic platforms are a promising alternative for electronics with regards to implementing
artificial neural networks: light-based processing enables massively parallel, energy-efficient
computation that can be realized within the material itself. This doctoral work aims to contribute to
that vision through three interconnected sub-projects:
Introducing Random Label Prediction Heads. Modern Optical Neural Networks (ONNs) are
currently constrained by physical scaling and are generally less prone to the massive overfitting
observed in large digital models. However, as photonic architectures grow in complexity, an open
and largely unexplored question emerges: to what extent do ONNs memorize individual training
samples, and to what extent do they extract genuinely generalizable features? In digital deep
learning, this question has received significant attention, yet for photonic systems – where physical
noise, fabrication imperfections, and limited programmability shape the learned representations –
no equivalent diagnostic framework exists. We argue that there is a pressing need for hardware-
agnostic tools that can quantify memorization in ONNs before these systems scale to regimes where
such effects may become practically relevant.
To address this gap, the first sub-project introduces random label prediction heads (RLP-heads) as a
general-purpose probe for analyzing memorization and representation capacity. An auxiliary
prediction head is attached to a trained network and tasked with learning deliberately random label
assignments. Because memorizing random labels requires the network to retain high-fidelity,
sample-specific information rather than generalizable patterns, the RLP-head's performance serves
as a direct, principled measure of how much individual training data is encoded within the
network's representations – whether those representations are formed in photonic or electronic
layers.
This approach is hardware-agnostic, applying equally to electronic, hybrid, or fully photonic
architectures and does not degrade primary task performance. RLP-heads do not presuppose a
trade-off between memorization and generalization; rather, they provide a memorization-based lens
through which to study the well-established overfitting–generalization trade-off. For the ONN
community specifically, this means that the classical question of whether a network's limited
capacity is allocated efficiently toward generalization or lost to overfitting can now be analyzed in
terms of memorization; a perspective that, to our knowledge, has not yet been systematically
applied in the photonic domain.
Photonic ANNs: Leveraging Intrinsic Randomness for a Diffusion-Based UNet. Photonic
neural network processors are never perfectly identical: fabrication imperfections, waveguide
variations, and coupling tolerances make each device physically unique. The second sub-project
reframes these variations not as defects but as a computational resource. Specifically, the intrinsic
randomness of a photonic processor is leveraged to supply the stochastic input required by a
diffusion-based UNet architecture. Diffusion models rely on iterative denoising, a process that
fundamentally depends on the availability of random noise. Rather than generating this noise
algorithmically, the photonic hardware provides it natively through its physical imperfections –
effectively turning fabrication-induced variability into a functional component of the generative
pipeline. This approach eliminates the need for a separate random number generator and tightly
couples the stochastic process to the physical substrate, resulting in an implementation where the
material itself both computes and supplies the randomness essential to the model. The concept
directly embodies the CRC 1459 vision of intelligent matter: a photonic device that does not merely
execute a pretrained algorithm but contributes an intrinsic physical property – its unique
randomness – as an integral part of the computational process.
Nonlinear Building Blocks Based on MZIs. Photonic platforms excel at linear matrix–vector
multiplications, yet without nonlinearities, cascaded layers collapse into a single linear
transformation. The third sub-project addresses this bottleneck by realizing nonlinear building
blocks, e.g., the inversion function, and nonlinear activation functions using Mach–
Zehnder interferometer (MZI) based circuits. By exploiting the sinusoidal phase-to-power transfer
characteristic of MZIs, nonlinear input–output mappings are engineered and integrated
monolithically with the linear mesh, advancing the goal of fully photonic neural network
architectures without intermediate electronic conversion.
Processing artificial neural networks requires powerful hardware. Today, they are usually calculated on conventional computers based on the von Neumann architecture. In this architecture, the processing unit is separated from the memory. Moving data from the memory to the processing unit, however, is costly in terms of time and energy. Hence, collocating processing and memory in so-called in-memory computing systems can make these calculations significantly more efficient. A novel and promising approach are systems based on integrated photonics. Here, optical signals are passed through waveguides and the weight of artificial synapses are represented by analog memory cells. I am working on mixed electro-optical memory cells based on phase-change materials. In such devices, a thin film of phase-change material is located on top of a photonic waveguide. The transmission through the waveguide can be tuned based on how much of the material is crystalline, and how much is amorphous. This can be probed with short laser pulses, taking advantage of the high data modulation rate and low latency of photonic systems. The state of the phase-change material can be changed by local heating with an electrical current. To optimize the performance of the electro-optical devices, I also study new materials with optimal properties. Utilizing both the electrical and optical domain in a single system is beneficial for learning, contributing to the development of intelligent matter.
