Neural net predicts semiconductor properties; ferroelectric titanium dioxide thin film; bidirectional pixels.
Researchers from the Institute of Science Tokyo, Yokohama City University, and National Sun Yat-sen University devised a tandem neural network capable of quickly inferring key physical parameters of semiconductor materials from simple transistor measurements.
The approach uses two machine learning models linked in series. The first model tries to estimate material properties from transistor measurements. Then, a pre-trained forward network that uses the material estimates produced by the first model to reconstruct the original transistor characteristics. By using the output of this second model as part of the training input given to the first model, the overall system learns to find solutions that are both mathematically plausible and physically consistent.
The tandem neural network (TNN) was trained using 1,000 amorphous indium–gallium–zinc oxide (a-IGZO) transistor datasets, covering six physical parameters including defect densities, trap-state characteristics, and electron mobility. The TNN was able to infer all six parameters from a single current–voltage curve in under one millisecond with high accuracy.
In tests using real transistors fabricated in the lab under different conditions, the model was able to reproduce their measured behavior without any additional optimization steps or adjustments.
“Compared with conventional device-simulation-based methods requiring hundreds of iterative calculations and taking tens of hours to several days, the proposed approach achieved a speedup of over six orders of magnitude,” said Keisuke Ide, an associate professor at Institute of Science Tokyo, in a press release. “We expect our approach to be applicable not only to semiconductors but also to a wide variety of inverse problems involving multivaluedness.” [1]
Researchers from the University of California Berkeley, Lawrence Berkeley National Laboratory, and SLAC National Accelerator Laboratory transformed titanium dioxide (TiO₂) into a ferroelectric material by reducing its thickness to less than 3nm.
“We were quite surprised when we discovered that as the thickness of TiO₂ films dropped below 3nm, the material becomes ferroelectric, a phase in which it shows spontaneous electric polarization that can be switched by applying an electric field,” said Sayeef Salahuddin, professor of electrical engineering and computer sciences and of materials science and engineering at UC Berkeley and a senior faculty scientist at Lawrence Berkeley National Laboratory, in a press release. “More importantly, this new ferroelectric behavior remains stable in films as thin as about 1nm, approximately two unit-cells thick.” Salahuddin suggested that other common dielectric materials of this class might develop new electronic behaviors in atomic-scale dimensions.
The TiO₂ films retained their ferroelectric properties when deposited on different substrates. “Our work demonstrates that this ferroelectricity is stable on both crystalline, or silicon, and amorphous carbon film substrates. This indicates the feasibility of integration with silicon-based technologies and beyond,” said Koushik Das, a graduate researcher in the College of Chemistry and the Department of Electrical Engineering and Computer Sciences at UC Berkeley, in a press release. “The ultrathin TiO₂ films can be grown at a low temperature, less than 400°C, using atomic layer deposition (ALD), a technique already used in state-of the-art chip fabrication. Furthermore, we can produce thin films with uniform thickness across all surfaces and polarization properties conducive to enabling new functionality for 3D integrated electronics.” [2]
Researchers at ETH Zurich developed bidirectional pixels that can both steer light and analyze it, an advance the team hopes could lead to devices that function as a camera and a display at the same time.
The pixel uses wave-shaped sculpted surfaces that can control and analyze the light intensity, oscillation phase, and polarization. For steering, the pixel transforms the incoming light into a surface wave propagating along the surface of the chip. At a different position within the pixel, the surface wave is scattered back out of the material as a light wave. Through interference of the light waves, patterns and images can be created. What these images will look like and what kind of surface pattern is needed for a specific image can be calculated using Fourier analysis.
“We can also, however, apply the principle of interference and Fourier analysis in the opposite direction to analyze light using the Fourier pixel”, said Sander Vonk, a postdoctoral researcher at ETH Zurich, in a statement. “Thanks to the fact that the relevant surface profiles of the pixels can be determined using Fourier analysis, we can combine the control and analysis of amplitude, phase, and polarization on a single pixel.”
The oscillation phase of the light can be made visible by superimposing the light wave and a reference wave on the pixel. The researchers then capture the interference pattern of the scattered light from both waves with a camera. The pattern can be used to calculate the phase of the light. A similar method is used to analyze the light’s polarization state.
Next, the team plans to create a matrix made of many Fourier pixels to create more complex devices. [3]
[1] M. Kimura, K. Ide, K.-J. Zhou, et al. Tandem Neural Network Rapidly Solves Multivalued Inverse Problems: Application to Oxide-Semiconductor Characterization. Advanced Intelligent Systems. https://doi.org/10.1002/aisy.70437
[2] K. Das, K. Reidy, S. Husain, et al. Ferroelectricity in atomic-scale titanium dioxide dielectric films. Science 392, 280-284 (2026). https://doi.org/10.1126/science.aec9417
[3] Y.M. Glauser, S.J.W. Vonk, D.B. Seda, et al. Fourier pixels for bidirectional light control. Nature 655, 87–92 (2026). https://doi.org/10.1038/s41586-026-10681-7
 |
 |
 |
 |
 |  |
 |  |
 |  |
Leave a Reply