Harnessing Machine Learning to Solve Complex Inverse Problems in Semiconductor Analysis

In a groundbreaking advancement poised to revolutionize semiconductor analysis, a team of researchers from the Institute of Science Tokyo has pioneered a tandem neural network (TNN) architecture that dramatically accelerates the inference of vital material parameters from transistor measurements. This novel machine learning framework addresses one of the most vexing challenges in semiconductor physics—the multivalued inverse problem—achieving unprecedented speed and accuracy that eclipses traditional computational methods.

Semiconductor devices underpin nearly every aspect of modern electronics, from smartphones to sophisticated computing systems. The performance and reliability of these devices hinge on intricate material properties such as defect densities, trap states, and charge carrier mobilities. While engineers can rapidly measure transistor behavior through current-voltage characteristics, deciphering the precise material attributes driving this behavior remains an elusive inverse problem. Conventional approaches rely heavily on iterative simulations and optimization techniques, often consuming hours or even days to converge on plausible solutions.

At the heart of the difficulty lies the phenomenon of multivaluedness: distinct combinations of semiconductor material parameters can yield virtually indistinguishable transistor responses. This ambiguity complicates the task of tracing device performance back to its physical origins. The research team recognized that conventional simulation frameworks are ill-equipped to navigate this degeneracy efficiently. Their insight was to harness the power of tandem neural networks, an innovative deep learning configuration capable of simultaenously enforcing physical consistency and mathematical plausibility in parameter estimation.

The tandem neural network developed by the team couples two interconnected models operating in series. The initial inverse model proposes candidate sets of material parameters based on observed transistor behavior. This output then feeds into a forward model pre-trained to reconstruct transistor characteristics from material parameters. By integrating the forward model’s reconstructions into the training loss function of the inverse network, the system effectively teaches itself to generate solutions that not only fit the measurement data but also align with the physical laws governing transistor operation.

Training this sophisticated architecture required an extensive dataset of over 1,000 simulated transistor measurements, specifically focusing on amorphous indium-gallium-zinc oxide (a-IGZO) transistors—a material of significant interest for flexible and transparent electronics. The dataset spanned a parameter space roughly 1,000 times broader than previous studies, encompassing six key variables: defect density, electron trap states, and carrier mobility among them. Astonishingly, the TNN inferred all six parameters from a single transistor current-voltage curve in less than one millisecond, achieving near-perfect accuracy across the range.

Beyond simulations, real-world validation was paramount. The researchers fabricated a series of a-IGZO transistors under five distinct processing conditions in their laboratories. When subjected to the TNN analysis, the model reproduced the measured device characteristics with exceptional fidelity, requiring no parameter tuning or iterative refinement. This immediate, high-precision performance marks a quantum leap compared to traditional simulation-dependent approaches that demand repetitive calculations, often spanning hours or days.

Assistant Professor Keisuke Ide, one of the principal investigators, highlighted the transformative implications of this speedup: “Our tandem neural network achieves a computational acceleration exceeding six orders of magnitude relative to conventional device simulation methods. This means what once took days can now be done almost instantaneously, radically changing the landscape of semiconductor diagnostics and research.”

The implications of this advancement extend well beyond academic curiosity. In industrial manufacturing, instantaneous and ultra-accurate assessment of transistor material properties can serve as an integral part of quality control, allowing defects and performance variants to be detected and addressed on the production line in real-time. This rapid feedback loop could markedly reduce waste and improve yield in semiconductor fabrication facilities.

Moreover, the TNN framework paves the way for autonomous laboratories where artificial intelligence designs, executes, and interprets experiments with minimal human oversight. By swiftly and reliably unraveling complex material-property relationships, AI-driven research tools can accelerate innovation cycles, optimizing device designs and material formulations in ways previously unattainable due to computational bottlenecks.

Interestingly, the tandem neural network architecture is not conceptually limited to semiconductors alone. The researchers underscore its potential applicability across a broad spectrum of inverse problems characterized by multivaluedness, spanning materials science, chemistry, optics, and possibly even biological systems where multiple underlying parameter sets may produce indistinguishable observational data.

The Institute of Science Tokyo, newly established in 2024 through the merger of Tokyo Medical and Dental University and Tokyo Institute of Technology, serves as a nexus for such cutting-edge interdisciplinary research. This project exemplifies the institute’s mission to advance scientific understanding and technology in ways that create tangible value for society.

As the demands for faster, smaller, and more efficient semiconductor components intensify, novel computational tools that can disentangle the complex interplay of material parameters swiftly and accurately become indispensable. The tandem neural network approach represents a critical milestone on this journey, ushering in a new era where machine learning not only supplements but fundamentally transforms physical sciences research.

This rise of AI-powered physics inference is poised to fuel accelerated material discovery, optimized manufacturing, and ultimately, enhanced electronic devices that power tomorrow’s technological landscape. With this tandem neural network, researchers have cracked open the door to solving multivalued inverse problems at blistering speeds, a feat that was once considered beyond reach.

News Publication Date: 27-May-2026

Web References: https://doi.org/10.1002/aisy.70437

Keywords

Artificial intelligence, machine learning, semiconductor analysis, inverse problems, multivaluedness, twin neural networks, material characterization, oxide semiconductors, transistor diagnostics, amorphous indium-gallium-zinc oxide, computational speedup, autonomous laboratories