Research articles

Device imperfections are a major challenge that limit the potential of analogue neural networks. Nanyang Ye and colleagues propose a training-time noise injection approach to improve their robustness without hardware modifications, which comes with theoretical guarantees.

Communications systems delivering signals at millimetre-wave/sub-THz frequencies suffer capacity fading over both long distances fibre transmission and also over short distance wireless transfer. Hai-Han Lu and colleagues report a combined wired/wireless system with sufficiently low bit error rates and error vector magnitudes as to satisfy the needs of future 5 G communications technologies.

Renewable energy sources could be used to drive alkaline electrolyzers for large-scale hydrogen production. Here Yanghong Xia and colleagues show that the fluctuating nature of renewable energy sources compromises the efficiency and consistency of electrolyzer performance. They also show that a strategy of multi-mode self-optimization can address this problem, extending the range of operation as well as improving efficiency at low loads.

Kristof Jacobs reports a method for microscopic defect analysis of semiconductor devices during electrical stimulation, without interference by probe needles usually located in close proximity to the objective lens. His approach re-routes the probes that need to be electrically contacted during the measurement with a low-profile redistribution layer realized by conductive inkjet printing.

Hamed Nikfarjam and colleagues report an integrated microelectromechanical system (MEMS) hardware capable of performing simultaneous sensing, compute and produce an action signal all within the same physical layer. The researchers demonstrate that different classification algorithms can be run on the MEMs hardware, expanding MEMS applications to more advanced fields such as surgical robotics and wearable devices.

Athanasios Oikonomou and colleagues develop a robust and computationally efficient multi-fidelity modeling framework that can learn the jet dynamics of melt electrowriting from videos and physical models. This learning strategy paves way for the next generation of self-calibrating electrohydrodynamics-based additive manufacturing technologies.

Aerial imaging of underwater environments is challenging, even more so in the presence of surface waves. Fitzpatrick et al. develop and perform lab-scale proof of principle demonstrations of a sensor fusion framework which couples airborne sonar imaging and water surface mapping to overcome ocean waves.

Hichem Guerboukha and colleagues describe a conformal (or curved) leaky waveguide in the THz range for antenna applications in wireless communications. The authors describe rich wave behavior at THz frequencies. They also experimentally show application potential for their antennas as multi-beam high-gain transmitters of wireless links with low bit error rates.

Nguyen and colleagues report a microfluidic platform with a ladder shaped design which identifies bacterial susceptibility to antibiotics in less than 5 h. The device could assist physicians and veterinarians to make more targeted and rapid prescriptions for antibiotic infections.

Coal fly ash (CFA) is an attractive cement diluent additive but contains heavy metals which could slowly leach out. Deng, Meng and colleagues report a rapid and water-free process to remove the heavy metals from CFA using flash joule heating. This approach reduces the energy consumption for CFA purification as well as other industrial wastes decontamination.

Data-Driven Quantum Approximate Optimization Algorithm for Power Systems Quantum Approximate Optimization Algorithms can enhance the monitoring, operation, and control of Distributed Energy Resources. Li and coworkers reduce the computational effort required for training these algorithms by efficiently obtaining algorithm parameters.

Zhang and colleagues report an orbital angular momentum detection device for vortex microwave photons. Specifically, the device is able to distinguish the intrinsic orbital angular momentum mode of the vortex microwave photons with the diffraction pattern of the coupled vortex electrons at room temperature, occupying a much smaller area compared to a particle accelerator system. The device enables application of microwave photons in practical scenarios.

Numerical tools for flood forecasting and for designing coastal protection schemes require accurate real world data on speed and volume of overtopping waves on sea walls. Margaret Yelland and colleagues here describe the validation and field deployment of arrays of capacitance sensors, termed Wirewall, as a tool for acquisition of detailed data of coastal overtopping.

Alanine aminotransferase (ALT) activity is an important indicator to determine drug induced liver injury. However, its measurement is limited to centralized laboratories. Muratore, Zhou, and colleagues designed an ALT assay biosensor platform using silicon nanowire field effect transistor arrays. The biosensor could monitor ALT enzyme activity electrically, with comparable precision to spectrophotometric detection. This biosensor platform will enable routine liver function monitoring in settings both inside and outside of the clinic.

Mingshan Liu and colleagues fabricate p- and n-channel vertical-type GeSn nanowire MOSFETs and their CMOS components down to 25 nm. The mobility in n-FETs increased 2.5-fold compared to a Ge reference device, a step toward extending Moore’s law beyond the silicon era.

Li et al. achieve polarization-independent wavelength demultiplexing with a single device, measuring polarizationdependent losses of 0.5–1.8 dB, minimum insertion loss of 0.5 dB, and crosstalks under −30 dB for an etched diffraction grating.

Maximilian Feil and colleagues report an optical method to characterize interface defects for wide-bandgap transistors. The approach will help to optimize passivation techniques and push the performance of transistors to their theoretical limit.

Over-reliance on automation in transportation systems is known to cause accidents. To address this, here, Tomohiro Nakade and colleagues describe a collaborative strategy for autonomous steering, in which the driver can take over from the automation without its full deactivation.

Shimahara and colleagues report a simulator based on an iterative deep neural network for identifying optimal conditions for ultrashort pulse laser drilling in dielectrics. The approach enabled a search across vast parameter space and the discovery of 20% more energy-efficient processing parameters than any of the experimental data on which the simulator was trained.