Publications 2026
Script list Publications
(1) Integration of Nanophotonic Structures with Optoelectronic Devices on the Si Platform
L. Augel, F. Berkmann, J. Schlipf, O. Skibitzki, I.A. Fischer
APL Photonics 11(2), 020901 (2026)
DOI: 10.1063/5.0291192, (OASYS)
Light-matter-interaction in nanostructures as well as the integration of optical and electro-optical technologies on the Silicon platform are research fields that receive strong and growing attention and that are relevant for a broad range of application areas from imaging to quantum computation. Here, we put a focus on their intersection with the discussion of device concepts, challenges and opportunities for the integration of nanophotonic structures into optoelectronic devices on the Silicon platform. The incorporation in particular of metasurfaces into optoelectronic devices offers unique possibilities not only for adding device functionality and improving efficiency at the same time, but also for material integration, provided that the challenges associated with device integration can be addressed.
(2) Exploring Variability and Quantization Effects in Artificial Neural Networks using the MNIST Dataset
A. Blumenstein, E. Perez, Ch. Wenger, N. Dersch, A. Kloes, B. Iniguez, M. Schwarz
Solid-State Electronics 232, 109296 (2026)
DOI: 10.1016/j.sse.2025.109296
This paper investigates the impact of introducing variability to trained neural networks and examines the effects of variability and quantization on network accuracy. The study utilizes the MNIST dataset to evaluate various Multi-Layer Perceptron configurations: a baseline model with a Single-Layer Perceptron and an extended model with multiple hidden nodes. The effects of Cycle-to-Cycle variability on network accuracy are explored by varying parameters such as the standard deviation to simulate dynamic changes in network weights. In particular, the performance differences between the Single-Layer Perceptron and the Multi-Layer Perceptron with hidden layers are analyzed, highlighting the network’s robustness to stochastic perturbations. These results provide insights into the effects of quantization and network architecture on accuracy under varying levels of variability.
(3) Mid-Infrared Intersubband Transitions in p-Type SiGe Parabolic Quantum Wells
M. Faverzani, D. Impelluso, S. Calcaterra, C. Zucchetti, D. Chrastina, C. Tassi, G. Capellini, P. Biagioni, G. Isella, M. Virgilio, J. Frigerio
Advanced Optical Materials 14(3), e03060 (2025)
DOI: 10.1002/adom.202503060, (IHP- Roma Tre University Joint Lab)
The design, fabrication, and comprehensive characterization of hole-doped Ge-rich SiGe parabolic quantum wells engineered to exhibit intersubband transitions in the mid-infrared spectral range around 120 meV are reported. The heterostructures are grown on Si substrates by low-energy plasma-enhanced chemical vapor deposition, enabling finely controlled compositional profiles and high crystalline quality. Thorough structural analysis confirms the formation of parabolic potential wells despite the presence of entropic interdiffusion. Photoreflectance spectroscopy is employed to investigate interband optical transitions in these heterostructures, whereas intersubband transitions are studied by Fourier-transform infrared spectroscopy that revealed characteristic constant-energy TM-polarized absorption features up to room temperature. At higher doping levels, a more structured spectral response is observed due to valence-band non-parabolicity. Tight-binding band structure simulations, incorporating many-body effects, accurately reproduce the observed spectral features. These results highlight the potential of SiGe parabolic quantum wells as a versatile and scalable platform for the development of Si-compatible mid-infrared optoelectronic devices based on intersubband transitions.
(4) Hole-Doped Ge-Rich SiGe Parabolic Quantum Wells for Mid-Infrared Photonics
M. Faverzani, D. Impelluso, S. Calcaterra, G. Capellini, P. Biagioni, G. Isella, J. Frigerio
50th International Conference on Infrared, Millimeter, and Terahertz Waves (IRMMW-THz 2025), (2026)
DOI: 10.1109/IRMMW-THz61557.2025.11319846
(5) Deposition of CeOx/SnOx-Based Thin Films via RF Magnetron Sputtering for Resistive Gas Sensing Applications
A. Kalra, C.A. Chavarin, P. Nitsch, R. Tschammer, J.I. Flege, M. Ratzke, M.H. Zoellner, M.A. Schubert, Ch. Wenger, I.A. Fischer
Physica B: Condensed Matter 723, 418098 (2026)
DOI: 10.1016/j.physb.2025.418098, (iCampus II)
Cerium oxide-tin oxide (CeOx/SnOx) thin films with varying Sn content were deposited using RF magnetron sputtering and investigated for hydrogen sensing applications. Structural, compositional and morphological properties were characterized using X-ray diffraction (XRD), X-ray photoelectron spectroscopy (XPS), atomic force microscopy (AFM), transmission electron microscopy (TEM) and energy-dispersive X-ray spectroscopy (EDX). Gas sensing measurements showed effective hydrogen detection at room temperature, with the sensitivity strongly influenced by Sn content and oxygen vacancy concentration. Higher Sn concentration enhanced the sensing response, which was correlated with microstructural features obtained from AFM and EDX, as well as with the presence of Ce3+ and Ce4+ oxidation states identified by XPS. This study highlights the potential of CeOx/SnOx thin films for possible back-end-of-line integration and provides proof-of-principle for room-temperature hydrogen sensing.
(6) Variability in HfO2-based Memristors under Pulse Operation
D. Maldonado, C. Acal, H. Ortiz, F. Navas-Gomez, A. Cantudo, Ch. Wenger, E. Pérez, J.B. Roldán
Microelectronic Engineering 304, 112445 (2026)
DOI: 10.1016/j.mee.2026.112445, (AVMMSafe)
We have studied device-to-device variability in TiN/Ti/HfO2/TiN devices under pulse operation. We measured extensively memristive devices that are CMOS integrated with different pulse trains, changing the pulse width and amplitude for groups of more than one hundred devices. The statistical parameters of the measured current were extracted to better understand the device physics under the pulse operation regime. An analytical model to describe synaptic depression and potentiation behavior in the device conductance is introduced, it fits accurately the means of the current data for all the pulse trains under study. In addition, an explanation of the measurements is enlightened with kinetic Monte Carlo simulations that allow the study of resistive switching at the atomic level. Finally, the probability distribution functions of the measured currents in some of the pulses within the pulse series employed are analyzed to extract the probability distribution that works better. A proposal for the implementation of device-to-device variability in the Stanford models is introduced.
(7) Electroluminescence and Franz-Keldysh Modulation Observed in Sn/Ge Multi-Quantum Wells
M. Oehme, D. Marian, M. Wanitzek, C. Spieth, D. Schwarz, F. Bärwolf, M.A. Schubert, A. Daus, M. Virgilio, G. Capellini
Advanced Optical Materials 14(9), e03561 (2026)
DOI: 10.1002/adom.202503561, (IHP- Roma Tre University Joint Lab)
Thanks to their tunable electronic and optical properties, GeSn alloys have emerged as promising materials for next-generation silicon-compatible optoelectronic devices operating at extended near- to mid-infrared wavelengths. In this work, we experimentally and theoretically investigate the optoelectronic properties of p-type/intrinsic/n-type (PIN) diodes based on ultrathin α-Sn/Ge multiple quantum well (MQW) structures grown by molecular beam epitaxy with different Ge barrier thicknesses. Thorough structural analysis confirms the high crystalline quality of the MQW and clearly defined interfaces with Sn contents in the wells exceeding 4 at.%. Room-temperature electroluminescence measurements reveal two direct transitions: an MQW stack independent higher-energy emission, related to direct radiative recombination in intrinsic Ge, and a Ge barrier-dependent lower-energy peak, attributed to transitions involving weakly quantum-confined states, whose energy shifts with the barrier thickness in agreement with theoretical prediction. In line with this observation, the devices exhibit clear signatures of the Franz–Keldysh effect when operated as photodetectors, in a wavelength range extending well beyond the direct gap of Ge, as captured by theoretical modeling. These results demonstrate the potential of α-Sn/Ge MQWs as an integrable material platform for complementary metal—oxide–semiconductor (CMOS)-compatible electro-optical modulators spanning the near- to mid-infrared range.
(8) Influence of Illumination Conditions on Photoluminescence Enhancement in an Al/Si/Ge Metasurface
P. Oleynik, D. Ryzhak, J. Schlipf, C.A. Chavarin, Y. Yamamoto, F. Berkmann, M. Ratzke, I.A. Fischer
Optics Express 34(1), 78 (2026)
DOI: 10.1364/OE.577751
Strong field enhancement supported by metasurfaces at resonance can be used to control and enhance the spontaneous emission rate of emitters. This is particularly relevant for emitters with comparatively low quantum yield such as germanium. Here, we investigate the µ-photoluminescence response obtained from a hybrid metasurface comprising a square lattice of Al/Si/Ge pillars. We explore how variations in excitation energy, excitation intensity and number of excited meta-atoms affect the spectral dependence of the photoluminescence signal and, in particular, the contribution of the metasurface to it. Our metasurface exhibits a magnetic dipole collective lattice resonance, whose contribution to the photoluminescence signal increases with increasing number of excited meta-atoms. Measuring only one metasurface under different illumination conditions can potentially be an alternative approach to probe the transition between finite-size effects and collective effects.
(9) Titanium Nitride Plasmonic Nanohole Arrays with Polymer Coating: Optical Properties and their Humidity-Induced Modifications
A. Sengül, S. Reiter, Z. Lotfi, J. Efremenko, A. Laroussi, A.A. Corley-Wiciak, M. Ratzke, V.M. Mirsky, Ch. Wenger, I.A. Fischer
Optical Materials Express 16(2), 184 (2026)
DOI: 10.1364/OME.578871
The use of titanium nitride (TiN) for the fabrication of plasmonic structures such as nanohole arrays (NHAs) can enable their integration into optoelectronic devices on the silicon (Si) platform, for example, for the realization of on-chip chemical sensors and biosensors based on refractometric transduction. With a corresponding functionalization of the TiN nanohole arrays, these ultra-compact devices can be utilized in the development of various affinity sensors and sensor systems, such as cost-effective electronic noses for the early detection of gases in the food industry or agriculture. In this work, we focus on two types of coating for functionalization of TiN nanohole arrays: electrochemically synthesized poly-N-methylaniline and layer-by-layer deposited polyacrylic-acid/poly-allylamine (PAA/PAH). Our investigation comprises the experimental characterization of the optical properties of TiN nanhole arrays coated with polymer layers of different thicknesses as well as a comparison with simulation results. We demonstrate the potential of our setup sensing applications by measuring changes in optical properties of TiN nanohole arrays coated with PAA/PAH upon exposure to air of different humidity.
(10) End-to-End Design Flow for Resistive Neural Accelerators
M. Uhlmann, T. Rizzi, E. Perez-Bosch Quesada, B. Al Beattie, K. Ochs, E. Perez, P. Ostrovskyy, C. Carta, Ch. Wenger, G. Kahmen
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 45(3), 1504 (2026)
DOI: 10.1109/TCAD.2025.3597237, (Neutronics)
Neural hardware accelerators have demonstrated notable energy efficiency in tackling tasks, which can be adapted to artificial neural network (ANN) structures. Research is currently directed towards leveraging resistive random-access memories (RRAMs) among various memristive devices. In conjunction with complementary metal-oxide semiconductor (CMOS) technologies within integrated circuits (ICs), RRAM devices are used to build such neural accelerators. In this study, we present a neural accelerator hardware design and verification flow, which uses a lookup table (LUT)-based Verilog-A model of IHP’s onetransistor-one-RRAM (1T1R) cell. In particular, we address the challenges of interfacing between abstract ANN simulations and circuit analysis by including a tailored Python wrapper into the design process for resistive neural hardware accelerators. To demonstrate our concept, the efficacy of the proposed design flow, we evaluate an ANN for the MNIST handwritten digit recognition task, as well as for the CIFAR-10 image recognition task, with the last layer verified through circuit simulation. Additionally, we implement different versions of a 1T1R model, based on quasi-static measurement data, providing insights on the effect of conductance level spacing and device-to-device variability. The circuit simulations tackle both schematic and physical layout assessment. The resulting recognition accuracies
exhibit significant differences between the purely application-level PyTorch simulation and our proposed design flow, highlighting the relevance of circuit-level validation for the design of neural hardware accelerators.
(11) RRAM-Based Spectral-Domain Convolution Accelerator for Reliable and Energy-Efficient CNN Inference
J. Wen, A. Baroni, Ch. Wenger, M. Krstic, L.M. Bolzani Pöhls
IEEE Transactions on Very Large Scale Integration (VLSI) Systems 34(3), 809 (2026)
DOI: 10.1109/TVLSI.2025.3644141, (6G-RIC)
The growing computational demands of convolutional neural networks (CNNs) have motivated the use of spectral-domain inference as an alternative to costly spatial-domain convolutions. In this work, we propose an resistive RAM (RRAM)-based spectral-domain convolutional layer that exploits in-memory computing (IMC) for low energy consumption and high parallelism. Both the two-dimensional Fourier transform and the pointwise multiplications are directly implemented on RRAM crossbar arrays, while Hermitian symmetry is leveraged to further enhance the energy efficiency of the transform and subsequent spectral processing. To ensure robustness, measured RRAM device data are incorporated into system-level simulations to evaluate the impact of device variability on inference accuracy. Furthermore, we introduce a layer-wise mapping framework that adaptively selects between spatial- and spectral-domain execution based on the trade-off between energy efficiency and accuracy. Simulation results show that the proposed design achieves up to a 2.18× improvement in energy efficiency across convolutional layer configurations. For VGG-8 on CIFAR-100, the proposed architecture with the layer-wise mapping scheme reduces the energy–delay product (EDP) by 45% while incurring negligible accuracy loss. This work presents the first complete RRAM-based spectral-domain convolutional layer that accounts for device variability, providing a promising solution for edge CNN inference.
(12) RRAM-Based Spectral-Domain Convolution Accelerator for Reliable and Energy-Efficient CNN Inference
J. Wen, A. Baroni, Ch. Wenger, M. Krstic, L.M. Bolzani Pöhls
IEEE Transactions on Very Large Scale Integration (VLSI) Systems 34(3), 809 (2026)
DOI: 10.1109/TVLSI.2025.3644141, (INSEKT)
The growing computational demands of convolutional neural networks (CNNs) have motivated the use of spectral-domain inference as an alternative to costly spatial-domain convolutions. In this work, we propose an resistive RAM (RRAM)-based spectral-domain convolutional layer that exploits in-memory computing (IMC) for low energy consumption and high parallelism. Both the two-dimensional Fourier transform and the pointwise multiplications are directly implemented on RRAM crossbar arrays, while Hermitian symmetry is leveraged to further enhance the energy efficiency of the transform and subsequent spectral processing. To ensure robustness, measured RRAM device data are incorporated into system-level simulations to evaluate the impact of device variability on inference accuracy. Furthermore, we introduce a layer-wise mapping framework that adaptively selects between spatial- and spectral-domain execution based on the trade-off between energy efficiency and accuracy. Simulation results show that the proposed design achieves up to a 2.18× improvement in energy efficiency across convolutional layer configurations. For VGG-8 on CIFAR-100, the proposed architecture with the layer-wise mapping scheme reduces the energy–delay product (EDP) by 45% while incurring negligible accuracy loss. This work presents the first complete RRAM-based spectral-domain convolutional layer that accounts for device variability, providing a promising solution for edge CNN inference.
(13) Influence of Strain and Initial Reactions on Epitaxial SiGe Growth
Y. Yamamoto, W.-C. Wen, F. Bärwolf, J. Schlipf, O. Fursenko, S. Karki, M.H. Zoellner, J. Murota, B. Tillack
Materials Science in Semiconductor Processing 209, 110572 (2026)
DOI: 10.1016/j.mssp.2026.110572
Heteroepitaxial growths of SiGe (0 – 40%) on fully relaxed SiGe virtual substrate (VS) with various Ge content (0 – 44%) are investigated to clarify the influence of strain and the initial surface on the SiGe growth. The deposition is performed using H2-SiH4-GeH4 based chemical vapor deposition (CVD) on planarized SiGe substrates, achieved by chemical mechanical polishing (CMP). The CMP interface is separated from the heteroepitaxial interface by depositing a homoepitaxial SiGe buffer layer on the SiGe VS, followed by a top heteroepitaxial SiGe growth. No relaxation of the top SiGe is confirmed for all combinations of the top SiGe and the SiGe buffers. With increasing tensile strain, an increase in the Ge content and a decrease in the growth rate are observed. The increase in the Ge content is more pronounced for the top SiGe growth with higher Ge content. It seems the surface reaction of SiH4 on the SiGe growth front is reduced by higher tensile strain. The top SiGe with higher Ge content shows higher interface pile-up of Ge, and the pile-up is pronounced on the SiGe buffer with higher Ge content as well as higher growth temperature, indicating non-equilibrium reaction at the initial stage of the SiGe growth. These results obtained here enable precise control of heteroepitaxial SiGe growth by CVD.
L. Augel, F. Berkmann, J. Schlipf, O. Skibitzki, I.A. Fischer
APL Photonics 11(2), 020901 (2026)
DOI: 10.1063/5.0291192, (OASYS)
Light-matter-interaction in nanostructures as well as the integration of optical and electro-optical technologies on the Silicon platform are research fields that receive strong and growing attention and that are relevant for a broad range of application areas from imaging to quantum computation. Here, we put a focus on their intersection with the discussion of device concepts, challenges and opportunities for the integration of nanophotonic structures into optoelectronic devices on the Silicon platform. The incorporation in particular of metasurfaces into optoelectronic devices offers unique possibilities not only for adding device functionality and improving efficiency at the same time, but also for material integration, provided that the challenges associated with device integration can be addressed.
(2) Exploring Variability and Quantization Effects in Artificial Neural Networks using the MNIST Dataset
A. Blumenstein, E. Perez, Ch. Wenger, N. Dersch, A. Kloes, B. Iniguez, M. Schwarz
Solid-State Electronics 232, 109296 (2026)
DOI: 10.1016/j.sse.2025.109296
This paper investigates the impact of introducing variability to trained neural networks and examines the effects of variability and quantization on network accuracy. The study utilizes the MNIST dataset to evaluate various Multi-Layer Perceptron configurations: a baseline model with a Single-Layer Perceptron and an extended model with multiple hidden nodes. The effects of Cycle-to-Cycle variability on network accuracy are explored by varying parameters such as the standard deviation to simulate dynamic changes in network weights. In particular, the performance differences between the Single-Layer Perceptron and the Multi-Layer Perceptron with hidden layers are analyzed, highlighting the network’s robustness to stochastic perturbations. These results provide insights into the effects of quantization and network architecture on accuracy under varying levels of variability.
(3) Mid-Infrared Intersubband Transitions in p-Type SiGe Parabolic Quantum Wells
M. Faverzani, D. Impelluso, S. Calcaterra, C. Zucchetti, D. Chrastina, C. Tassi, G. Capellini, P. Biagioni, G. Isella, M. Virgilio, J. Frigerio
Advanced Optical Materials 14(3), e03060 (2025)
DOI: 10.1002/adom.202503060, (IHP- Roma Tre University Joint Lab)
The design, fabrication, and comprehensive characterization of hole-doped Ge-rich SiGe parabolic quantum wells engineered to exhibit intersubband transitions in the mid-infrared spectral range around 120 meV are reported. The heterostructures are grown on Si substrates by low-energy plasma-enhanced chemical vapor deposition, enabling finely controlled compositional profiles and high crystalline quality. Thorough structural analysis confirms the formation of parabolic potential wells despite the presence of entropic interdiffusion. Photoreflectance spectroscopy is employed to investigate interband optical transitions in these heterostructures, whereas intersubband transitions are studied by Fourier-transform infrared spectroscopy that revealed characteristic constant-energy TM-polarized absorption features up to room temperature. At higher doping levels, a more structured spectral response is observed due to valence-band non-parabolicity. Tight-binding band structure simulations, incorporating many-body effects, accurately reproduce the observed spectral features. These results highlight the potential of SiGe parabolic quantum wells as a versatile and scalable platform for the development of Si-compatible mid-infrared optoelectronic devices based on intersubband transitions.
(4) Hole-Doped Ge-Rich SiGe Parabolic Quantum Wells for Mid-Infrared Photonics
M. Faverzani, D. Impelluso, S. Calcaterra, G. Capellini, P. Biagioni, G. Isella, J. Frigerio
50th International Conference on Infrared, Millimeter, and Terahertz Waves (IRMMW-THz 2025), (2026)
DOI: 10.1109/IRMMW-THz61557.2025.11319846
(5) Deposition of CeOx/SnOx-Based Thin Films via RF Magnetron Sputtering for Resistive Gas Sensing Applications
A. Kalra, C.A. Chavarin, P. Nitsch, R. Tschammer, J.I. Flege, M. Ratzke, M.H. Zoellner, M.A. Schubert, Ch. Wenger, I.A. Fischer
Physica B: Condensed Matter 723, 418098 (2026)
DOI: 10.1016/j.physb.2025.418098, (iCampus II)
Cerium oxide-tin oxide (CeOx/SnOx) thin films with varying Sn content were deposited using RF magnetron sputtering and investigated for hydrogen sensing applications. Structural, compositional and morphological properties were characterized using X-ray diffraction (XRD), X-ray photoelectron spectroscopy (XPS), atomic force microscopy (AFM), transmission electron microscopy (TEM) and energy-dispersive X-ray spectroscopy (EDX). Gas sensing measurements showed effective hydrogen detection at room temperature, with the sensitivity strongly influenced by Sn content and oxygen vacancy concentration. Higher Sn concentration enhanced the sensing response, which was correlated with microstructural features obtained from AFM and EDX, as well as with the presence of Ce3+ and Ce4+ oxidation states identified by XPS. This study highlights the potential of CeOx/SnOx thin films for possible back-end-of-line integration and provides proof-of-principle for room-temperature hydrogen sensing.
(6) Variability in HfO2-based Memristors under Pulse Operation
D. Maldonado, C. Acal, H. Ortiz, F. Navas-Gomez, A. Cantudo, Ch. Wenger, E. Pérez, J.B. Roldán
Microelectronic Engineering 304, 112445 (2026)
DOI: 10.1016/j.mee.2026.112445, (AVMMSafe)
We have studied device-to-device variability in TiN/Ti/HfO2/TiN devices under pulse operation. We measured extensively memristive devices that are CMOS integrated with different pulse trains, changing the pulse width and amplitude for groups of more than one hundred devices. The statistical parameters of the measured current were extracted to better understand the device physics under the pulse operation regime. An analytical model to describe synaptic depression and potentiation behavior in the device conductance is introduced, it fits accurately the means of the current data for all the pulse trains under study. In addition, an explanation of the measurements is enlightened with kinetic Monte Carlo simulations that allow the study of resistive switching at the atomic level. Finally, the probability distribution functions of the measured currents in some of the pulses within the pulse series employed are analyzed to extract the probability distribution that works better. A proposal for the implementation of device-to-device variability in the Stanford models is introduced.
(7) Electroluminescence and Franz-Keldysh Modulation Observed in Sn/Ge Multi-Quantum Wells
M. Oehme, D. Marian, M. Wanitzek, C. Spieth, D. Schwarz, F. Bärwolf, M.A. Schubert, A. Daus, M. Virgilio, G. Capellini
Advanced Optical Materials 14(9), e03561 (2026)
DOI: 10.1002/adom.202503561, (IHP- Roma Tre University Joint Lab)
Thanks to their tunable electronic and optical properties, GeSn alloys have emerged as promising materials for next-generation silicon-compatible optoelectronic devices operating at extended near- to mid-infrared wavelengths. In this work, we experimentally and theoretically investigate the optoelectronic properties of p-type/intrinsic/n-type (PIN) diodes based on ultrathin α-Sn/Ge multiple quantum well (MQW) structures grown by molecular beam epitaxy with different Ge barrier thicknesses. Thorough structural analysis confirms the high crystalline quality of the MQW and clearly defined interfaces with Sn contents in the wells exceeding 4 at.%. Room-temperature electroluminescence measurements reveal two direct transitions: an MQW stack independent higher-energy emission, related to direct radiative recombination in intrinsic Ge, and a Ge barrier-dependent lower-energy peak, attributed to transitions involving weakly quantum-confined states, whose energy shifts with the barrier thickness in agreement with theoretical prediction. In line with this observation, the devices exhibit clear signatures of the Franz–Keldysh effect when operated as photodetectors, in a wavelength range extending well beyond the direct gap of Ge, as captured by theoretical modeling. These results demonstrate the potential of α-Sn/Ge MQWs as an integrable material platform for complementary metal—oxide–semiconductor (CMOS)-compatible electro-optical modulators spanning the near- to mid-infrared range.
(8) Influence of Illumination Conditions on Photoluminescence Enhancement in an Al/Si/Ge Metasurface
P. Oleynik, D. Ryzhak, J. Schlipf, C.A. Chavarin, Y. Yamamoto, F. Berkmann, M. Ratzke, I.A. Fischer
Optics Express 34(1), 78 (2026)
DOI: 10.1364/OE.577751
Strong field enhancement supported by metasurfaces at resonance can be used to control and enhance the spontaneous emission rate of emitters. This is particularly relevant for emitters with comparatively low quantum yield such as germanium. Here, we investigate the µ-photoluminescence response obtained from a hybrid metasurface comprising a square lattice of Al/Si/Ge pillars. We explore how variations in excitation energy, excitation intensity and number of excited meta-atoms affect the spectral dependence of the photoluminescence signal and, in particular, the contribution of the metasurface to it. Our metasurface exhibits a magnetic dipole collective lattice resonance, whose contribution to the photoluminescence signal increases with increasing number of excited meta-atoms. Measuring only one metasurface under different illumination conditions can potentially be an alternative approach to probe the transition between finite-size effects and collective effects.
(9) Titanium Nitride Plasmonic Nanohole Arrays with Polymer Coating: Optical Properties and their Humidity-Induced Modifications
A. Sengül, S. Reiter, Z. Lotfi, J. Efremenko, A. Laroussi, A.A. Corley-Wiciak, M. Ratzke, V.M. Mirsky, Ch. Wenger, I.A. Fischer
Optical Materials Express 16(2), 184 (2026)
DOI: 10.1364/OME.578871
The use of titanium nitride (TiN) for the fabrication of plasmonic structures such as nanohole arrays (NHAs) can enable their integration into optoelectronic devices on the silicon (Si) platform, for example, for the realization of on-chip chemical sensors and biosensors based on refractometric transduction. With a corresponding functionalization of the TiN nanohole arrays, these ultra-compact devices can be utilized in the development of various affinity sensors and sensor systems, such as cost-effective electronic noses for the early detection of gases in the food industry or agriculture. In this work, we focus on two types of coating for functionalization of TiN nanohole arrays: electrochemically synthesized poly-N-methylaniline and layer-by-layer deposited polyacrylic-acid/poly-allylamine (PAA/PAH). Our investigation comprises the experimental characterization of the optical properties of TiN nanhole arrays coated with polymer layers of different thicknesses as well as a comparison with simulation results. We demonstrate the potential of our setup sensing applications by measuring changes in optical properties of TiN nanohole arrays coated with PAA/PAH upon exposure to air of different humidity.
(10) End-to-End Design Flow for Resistive Neural Accelerators
M. Uhlmann, T. Rizzi, E. Perez-Bosch Quesada, B. Al Beattie, K. Ochs, E. Perez, P. Ostrovskyy, C. Carta, Ch. Wenger, G. Kahmen
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 45(3), 1504 (2026)
DOI: 10.1109/TCAD.2025.3597237, (Neutronics)
Neural hardware accelerators have demonstrated notable energy efficiency in tackling tasks, which can be adapted to artificial neural network (ANN) structures. Research is currently directed towards leveraging resistive random-access memories (RRAMs) among various memristive devices. In conjunction with complementary metal-oxide semiconductor (CMOS) technologies within integrated circuits (ICs), RRAM devices are used to build such neural accelerators. In this study, we present a neural accelerator hardware design and verification flow, which uses a lookup table (LUT)-based Verilog-A model of IHP’s onetransistor-one-RRAM (1T1R) cell. In particular, we address the challenges of interfacing between abstract ANN simulations and circuit analysis by including a tailored Python wrapper into the design process for resistive neural hardware accelerators. To demonstrate our concept, the efficacy of the proposed design flow, we evaluate an ANN for the MNIST handwritten digit recognition task, as well as for the CIFAR-10 image recognition task, with the last layer verified through circuit simulation. Additionally, we implement different versions of a 1T1R model, based on quasi-static measurement data, providing insights on the effect of conductance level spacing and device-to-device variability. The circuit simulations tackle both schematic and physical layout assessment. The resulting recognition accuracies
exhibit significant differences between the purely application-level PyTorch simulation and our proposed design flow, highlighting the relevance of circuit-level validation for the design of neural hardware accelerators.
(11) RRAM-Based Spectral-Domain Convolution Accelerator for Reliable and Energy-Efficient CNN Inference
J. Wen, A. Baroni, Ch. Wenger, M. Krstic, L.M. Bolzani Pöhls
IEEE Transactions on Very Large Scale Integration (VLSI) Systems 34(3), 809 (2026)
DOI: 10.1109/TVLSI.2025.3644141, (6G-RIC)
The growing computational demands of convolutional neural networks (CNNs) have motivated the use of spectral-domain inference as an alternative to costly spatial-domain convolutions. In this work, we propose an resistive RAM (RRAM)-based spectral-domain convolutional layer that exploits in-memory computing (IMC) for low energy consumption and high parallelism. Both the two-dimensional Fourier transform and the pointwise multiplications are directly implemented on RRAM crossbar arrays, while Hermitian symmetry is leveraged to further enhance the energy efficiency of the transform and subsequent spectral processing. To ensure robustness, measured RRAM device data are incorporated into system-level simulations to evaluate the impact of device variability on inference accuracy. Furthermore, we introduce a layer-wise mapping framework that adaptively selects between spatial- and spectral-domain execution based on the trade-off between energy efficiency and accuracy. Simulation results show that the proposed design achieves up to a 2.18× improvement in energy efficiency across convolutional layer configurations. For VGG-8 on CIFAR-100, the proposed architecture with the layer-wise mapping scheme reduces the energy–delay product (EDP) by 45% while incurring negligible accuracy loss. This work presents the first complete RRAM-based spectral-domain convolutional layer that accounts for device variability, providing a promising solution for edge CNN inference.
(12) RRAM-Based Spectral-Domain Convolution Accelerator for Reliable and Energy-Efficient CNN Inference
J. Wen, A. Baroni, Ch. Wenger, M. Krstic, L.M. Bolzani Pöhls
IEEE Transactions on Very Large Scale Integration (VLSI) Systems 34(3), 809 (2026)
DOI: 10.1109/TVLSI.2025.3644141, (INSEKT)
The growing computational demands of convolutional neural networks (CNNs) have motivated the use of spectral-domain inference as an alternative to costly spatial-domain convolutions. In this work, we propose an resistive RAM (RRAM)-based spectral-domain convolutional layer that exploits in-memory computing (IMC) for low energy consumption and high parallelism. Both the two-dimensional Fourier transform and the pointwise multiplications are directly implemented on RRAM crossbar arrays, while Hermitian symmetry is leveraged to further enhance the energy efficiency of the transform and subsequent spectral processing. To ensure robustness, measured RRAM device data are incorporated into system-level simulations to evaluate the impact of device variability on inference accuracy. Furthermore, we introduce a layer-wise mapping framework that adaptively selects between spatial- and spectral-domain execution based on the trade-off between energy efficiency and accuracy. Simulation results show that the proposed design achieves up to a 2.18× improvement in energy efficiency across convolutional layer configurations. For VGG-8 on CIFAR-100, the proposed architecture with the layer-wise mapping scheme reduces the energy–delay product (EDP) by 45% while incurring negligible accuracy loss. This work presents the first complete RRAM-based spectral-domain convolutional layer that accounts for device variability, providing a promising solution for edge CNN inference.
(13) Influence of Strain and Initial Reactions on Epitaxial SiGe Growth
Y. Yamamoto, W.-C. Wen, F. Bärwolf, J. Schlipf, O. Fursenko, S. Karki, M.H. Zoellner, J. Murota, B. Tillack
Materials Science in Semiconductor Processing 209, 110572 (2026)
DOI: 10.1016/j.mssp.2026.110572
Heteroepitaxial growths of SiGe (0 – 40%) on fully relaxed SiGe virtual substrate (VS) with various Ge content (0 – 44%) are investigated to clarify the influence of strain and the initial surface on the SiGe growth. The deposition is performed using H2-SiH4-GeH4 based chemical vapor deposition (CVD) on planarized SiGe substrates, achieved by chemical mechanical polishing (CMP). The CMP interface is separated from the heteroepitaxial interface by depositing a homoepitaxial SiGe buffer layer on the SiGe VS, followed by a top heteroepitaxial SiGe growth. No relaxation of the top SiGe is confirmed for all combinations of the top SiGe and the SiGe buffers. With increasing tensile strain, an increase in the Ge content and a decrease in the growth rate are observed. The increase in the Ge content is more pronounced for the top SiGe growth with higher Ge content. It seems the surface reaction of SiH4 on the SiGe growth front is reduced by higher tensile strain. The top SiGe with higher Ge content shows higher interface pile-up of Ge, and the pile-up is pronounced on the SiGe buffer with higher Ge content as well as higher growth temperature, indicating non-equilibrium reaction at the initial stage of the SiGe growth. These results obtained here enable precise control of heteroepitaxial SiGe growth by CVD.
