Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Dual-mode ferroelectric transistors for high-performance generative-adversarial-network-based imaging

Nature Sensors volume 1, pages 222–231 (2026)Cite this article

Subjects

Abstract

Artificial intelligence has the capacity to accelerate the interpretation of medical images and support diagnostic decision-making. Delivering these benefits directly within clinical imaging systems, however, requires compact computing hardware capable of executing both generative and discriminative operations with low latency and high energy efficiency. Generative adversarial networks, which combine image synthesis with classification, offer particular promise for improving diagnostic accuracy and augmenting limited datasets, but their implementation in hardware remains challenging owing to competing requirements for rapid weight updates and long-term retention stability. Here we report a reconfigurable neuromorphic platform based on ferroelectric field-effect transistors that enables dual-mode operation via dynamic switching between ferroelectric polarization and charge-trapping mechanisms. This architecture supports both high-speed, energy-efficient synaptic programming and robust non-volatile retention within the same device. Implemented in a 6 × 6 analogue crossbar, the system performs convolution and deconvolution operations representative of generative-adversarial-network workflows, enabling realistic synthesis of mammographic lesions. This approach establishes a compact hardware foundation for embedded diagnostic intelligence in next-generation imaging instruments.

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on SpringerLink
  • Instant access to the full article PDF.

USD 39.95

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Dual-mode FeFETs for GAN hardware configuration.
Fig. 2: Device performance of dual-mode FeFETs.
Fig. 3: Crossbar array modulation.
Fig. 4: Hardware implementation of GAN-based operation.
Fig. 5: Architecture and performance evaluation of GAN for breast lesion detection.

Similar content being viewed by others

Data availability

The data that support the plots within this article and other findings of this study are available from the corresponding authors upon reasonable request.

Code availability

All codes are available from the corresponding authors upon reasonable request.

References

  1. Lee, J.-G. et al. Deep learning in medical imaging: general overview. Korean J. Radiol. 18, 570–584 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

  2. Litjens, G. et al. A survey on deep learning in medical image analysis. Med. Image Anal. 42, 60–88 (2017).

    Article  PubMed  Google Scholar 

  3. Zhu, Y. et al. MRI-based prostate cancer detection with high-level representation and hierarchical classification. Med. Phys. 44, 1028–1039 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

  4. Wang, G., Ye, J. C., Mueller, K. & Fessler, J. A. Image reconstruction is a new frontier of machine learning. IEEE Trans. Med. Imaging 37, 1289–1296 (2018).

    Article  PubMed  Google Scholar 

  5. Kaissis, G. A., Makowski, M. R., Rückert, D. & Braren, R. F. Secure, privacy-preserving and federated machine learning in medical imaging. Nat. Mach. Intell. 2, 305–311 (2020).

    Article  Google Scholar 

  6. Yao, P. et al. Fully hardware-implemented memristor convolutional neural network. Nature 577, 641–646 (2020).

    Article  CAS  PubMed  Google Scholar 

  7. Li, Y. et al. Anomalous resistive switching in memristors based on two-dimensional palladium diselenide using heterophase grain boundaries. Nat. Electron. 4, 348–356 (2021).

    Article  CAS  Google Scholar 

  8. Kim, M.-K., Kim, I.-J. & Lee, J.-S. CMOS-compatible compute-in-memory accelerators based on integrated ferroelectric synaptic arrays for convolution neural networks. Sci. Adv. 8, eabm8537 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  9. Osuala, R. et al. Data synthesis and adversarial networks: a review and meta-analysis in cancer imaging. Med. Image Anal. 84, 102704 (2023).

    Article  PubMed  Google Scholar 

  10. Alrashedy, H. H. N., Almansour, A. F., Ibrahim, D. M. & Hammoudeh, M. A. A. BrainGAN: brain MRI image generation and classification framework using GAN architectures and CNN models. Sensors 22, 4297 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

  11. Liu, G., Lan, S., Zhang, T., Huang, W. & Wang, W. SAGAN: skip-attention GAN for anomaly detection. In 2021 IEEE International Conference on Image Processing (ICIP) (eds Delp, E. J. III et al.) 2468–2472 (IEEE, 2021); https://doi.org/10.1109/ICIP42928.2021.9506332

  12. van Tulder, G. & de Bruijne, M. Combining generative and discriminative representation learning for lung CT analysis with convolutional restricted Boltzmann machines. IEEE Trans. Med. Imaging 35, 1262–1272 (2016).

    Article  PubMed  Google Scholar 

  13. Kendall, J. D. & Kumar, S. The building blocks of a brain-inspired computer. Appl. Phys. Rev. 7, 011305 (2020).

    Article  CAS  Google Scholar 

  14. Müller, J. et al. Ferroelectricity in simple binary ZrO2 and HfO2. Nano Lett. 12, 4318–4323 (2012).

    Article  PubMed  Google Scholar 

  15. Chien, Y.-C. et al. A MoS2 hafnium oxide based ferroelectric encoder for temporal-efficient spiking neural network. Adv. Mater. 35, 2204949 (2023).

    Article  CAS  Google Scholar 

  16. Cheema, S. S. et al. Enhanced ferroelectricity in ultrathin films grown directly on silicon. Nature 580, 478–482 (2020).

    Article  CAS  PubMed  Google Scholar 

  17. Zhou, Y. et al. A reconfigurable two-WSe2-transistor synaptic cell for reinforcement learning. Adv. Mater. 34, 2107754 (2022).

    Article  CAS  Google Scholar 

  18. Lee, H. S. et al. MoS2 nanosheets for top-gate nonvolatile memory transistor channel. Small 8, 3111–3115 (2012).

    Article  CAS  PubMed  Google Scholar 

  19. Wang, X. et al. Ultrasensitive and broadband MoS2 photodetector driven by ferroelectrics. Adv. Mater. 27, 6575–6581 (2015).

    Article  CAS  PubMed  Google Scholar 

  20. Lee, Y. T. et al. Nonvolatile ferroelectric memory circuit using black phosphorus nanosheet-based field-effect transistors with P(VDF-TrFE) polymer. ACS Nano 9, 10394–10401 (2015).

    Article  CAS  PubMed  Google Scholar 

  21. Huang, H. et al. Ferroelectric polymer tuned two dimensional layered MoTe2 photodetector. RSC Adv. 6, 87416–87421 (2016).

    Article  CAS  Google Scholar 

  22. Wang, X. et al. Two-dimensional negative capacitance transistor with polyvinylidene fluoride-based ferroelectric polymer gating. npj 2D Mater. Appl. 1, 38 (2017).

    Article  Google Scholar 

  23. Luo, Z.-D. et al. Dual-ferroelectric-coupling-engineered two-dimensional transistors for multifunctional in-memory computing. ACS Nano 16, 3362–3372 (2022).

    Article  CAS  PubMed  Google Scholar 

  24. Li, D. et al. The ambipolar evolution of a high-performance WSe2 transistor assisted by a ferroelectric polymer. Nanotechnology 29, 105202 (2018).

    Article  PubMed  Google Scholar 

  25. Liao, J. et al. Van der Waals ferroelectric semiconductor field effect transistor for in-memory computing. ACS Nano 17, 6095–6102 (2023).

    Article  CAS  PubMed  Google Scholar 

  26. Wang, X. et al. Van der Waals engineering of ferroelectric heterostructures for long-retention memory. Nat. Commun. 12, 1109 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  27. Zha, J. et al. Electronic/optoelectronic memory device enabled by tellurium-based 2D van der Waals heterostructure for in-sensor reservoir computing at the optical communication band. Adv. Mater. 35, 2211598 (2023).

    Article  CAS  Google Scholar 

  28. Jiang, X. et al. Ferroelectric field-effect transistors based on WSe2/CuInP2S6 heterostructures for memory applications. ACS Appl. Electron. Mater. 3, 4711–4717 (2021).

    Article  CAS  Google Scholar 

  29. Liu, X. et al. Post-CMOS compatible aluminum scandium nitride/2D channel ferroelectric field-effect-transistor memory. Nano Lett. 21, 3753–3761 (2021).

    Article  CAS  PubMed  Google Scholar 

  30. Kim, K.-H. et al. Tuning polarity in WSe2/AlScN FeFETs via contact engineering. ACS Nano 18, 4180–4188 (2024).

    Article  CAS  PubMed  Google Scholar 

  31. Zhang, X.-W. et al. MoS2 field-effect transistors with lead zirconate-titanate ferroelectric gating. IEEE Electron Device Lett. 36, 784–786 (2015).

    Article  CAS  Google Scholar 

  32. Lu, Z. et al. Nonvolatile MoS2 field effect transistors directly gated by single crystalline epitaxial ferroelectric. Appl. Phys. Lett. 111, 023104 (2017).

    Article  Google Scholar 

  33. Mbisike, S. C., Eckart, L., Phair, J. W., Lomax, P. & Cheung, R. Amplification of pyroelectric device with WSe2 field effect transistor and ferroelectric gating. J. Appl. Phys. 131, 144101 (2022).

    Article  CAS  Google Scholar 

  34. Ko, C. et al. Ferroelectrically gated atomically thin transition-metal dichalcogenides as nonvolatile memory. Adv. Mater. 28, 2923–2930 (2016).

    Article  CAS  PubMed  Google Scholar 

  35. Chen, L. et al. A van der Waals synaptic transistor based on ferroelectric Hf0.5Zr0.5O2 and 2D tungsten disulfide. Adv. Electron. Mater. 6, 2000057 (2020).

    Article  CAS  Google Scholar 

  36. Jeon, H. et al. Hysteresis modulation on van der Waals-based ferroelectric field-effect transistor by interfacial passivation technique and its application in optic neural networks. Small 16, 2004371 (2020).

    Article  CAS  Google Scholar 

  37. Xiang, H. et al. Enhancing memory window efficiency of ferroelectric transistor for neuromorphic computing via two-dimensional materials integration. Adv. Funct. Mater. 33, 2304657 (2023).

    Article  CAS  Google Scholar 

  38. Yap, W. C., Jiang, H., Liu, J., Xia, Q. & Zhu, W. Ferroelectric transistors with monolayer molybdenum disulfide and ultra-thin aluminum-doped hafnium oxide. Appl. Phys. Lett. 111, 013103 (2017).

    Article  Google Scholar 

  39. Migliato Marega, G. et al. A large-scale integrated vector–matrix multiplication processor based on monolayer molybdenum disulfide memories. Nat. Electron. 6, 991–998 (2023).

    Article  CAS  Google Scholar 

  40. Burr, G. W. et al. Neuromorphic computing using non-volatile memory. Adv. Phys. X 2, 89–124 (2017).

    Google Scholar 

  41. Duong, N. T. et al. Coupled ferroelectric-photonic memory in a retinomorphic hardware for in-sensor computing. Adv. Sci. 11, 2303447 (2024).

    Article  CAS  Google Scholar 

  42. Turashvili, G. & Brogi, E. Tumor heterogeneity in breast cancer. Front. Med. 4, 227 (2017).

    Article  Google Scholar 

  43. Yersal, O. & Barutca, S. Biological subtypes of breast cancer: prognostic and therapeutic implications. World J. Clin. Oncol. 5, 412–424 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

  44. Moreira, I. C. et al. INbreast: toward a full-field digital mammographic database. Acad. Radiol. 19, 236–248 (2012).

    Article  PubMed  Google Scholar 

  45. Davis, J. & Goadrich, M. The relationship between precision–recall and ROC curves. In Proc. 23rd International Conference on Machine Learning—ICML ’06 (eds Cohen, W. W. & Moore, A.) 233–240 (ACM, 2006); https://doi.org/10.1145/1143844.1143874

  46. Saito, T. & Rehmsmeier, M. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PLoS ONE 10, e0118432 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

  47. Demirkiran, C., Nair, L., Bunandar, D. & Joshi, A. A blueprint for precise and fault-tolerant analog neural networks. Nat. Commun. 15, 5098 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  48. Moon, S., Shin, K. & Jeon, D. Enhancing reliability of analog neural network processors. IEEE Trans. Very Large Scale Integr. Syst. 27, 1455–1459 (2019).

    Article  Google Scholar 

  49. Gong, N. et al. Signal and noise extraction from analog memory elements for neuromorphic computing. Nat. Commun. 9, 2102 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  50. Gulrajani, I., Ahmed, F., Arjovsky, M., Dumoulin, V. & Courville, A. Improved training of Wasserstein GANs. In Proc. 31st International Conference on Neural Information Processing Systems (eds Guyon, I. et al.) 5769–5779 (Curran Associates, 2017).

Download references

Acknowledgements

This work is supported by the National Research Foundation (NRF), Prime Minister’s Office, Singapore, under its Competitive Research Program (NRF-CRP24-2020-0002, K.-W.A., NRF-F-CRP-2024-0006, K.-W.A.). L.-J.L thanks the NRF for support (NRF-P2025-002, L.-J.L.). L.-J.L. thanks Nexstrom Pte Ltd, Singapore for MoS2 films.

Author information

Author notes

  1. These authors contributed equally: Lingqi Li, Chenyang Li, Haofei Zheng.

Authors and Affiliations

  1. Department of Electrical and Computer Engineering, National University of Singapore, Singapore, Singapore

    Lingqi Li, Haofei Zheng, Heng Xiang, Jianze Wang, Wei Zhang, Jiali Huo, Jing Gao, Xuanyao Fong & Kah-Wee Ang

  2. Department of Materials Science and Engineering, National University of Singapore, Singapore, Singapore

    Chenyang Li, Yi Wan, Wanqing Meng & Lain-Jong Li

  3. Nexstrom Pte Ltd, Singapore, Singapore

    Chenyang Li

  4. Department of Mechanical Engineering, The University of Hong Kong, Hong Kong, China

    Wanqing Meng

Authors
  1. Lingqi Li
    View author publications

    Search author on:PubMed Google Scholar

  2. Chenyang Li
    View author publications

    Search author on:PubMed Google Scholar

  3. Haofei Zheng
    View author publications

    Search author on:PubMed Google Scholar

  4. Heng Xiang
    View author publications

    Search author on:PubMed Google Scholar

  5. Jianze Wang
    View author publications

    Search author on:PubMed Google Scholar

  6. Wei Zhang
    View author publications

    Search author on:PubMed Google Scholar

  7. Jiali Huo
    View author publications

    Search author on:PubMed Google Scholar

  8. Jing Gao
    View author publications

    Search author on:PubMed Google Scholar

  9. Yi Wan
    View author publications

    Search author on:PubMed Google Scholar

  10. Xuanyao Fong
    View author publications

    Search author on:PubMed Google Scholar

  11. Wanqing Meng
    View author publications

    Search author on:PubMed Google Scholar

  12. Lain-Jong Li
    View author publications

    Search author on:PubMed Google Scholar

  13. Kah-Wee Ang
    View author publications

    Search author on:PubMed Google Scholar

Contributions

This project was supervised and directed by K.-W.A. and L.-J.L. L.L. and W.M. designed the experiments. C.L. and Y.W. synthesized the MoS2 films. L.L. and H.Z. conducted the device fabrication and electrical measurements. H.Z. conducted high-resolution transmission electron microscopy image analysis. J.W., W.Z. and X.F. contributed to the circuit-level simulations. L.L. and H.X. contributed to the simulation of GAN. C.L., H.Z., J.W., W.Z., J.H., J.G., Y.W. and X.F. contributed to discussion and data analysis. L.L., W.M., L.-J.L. and K.-W.A. wrote the paper. All authors discussed the results and approved the final version of the paper.

Corresponding authors

Correspondence to Wanqing Meng, Lain-Jong Li or Kah-Wee Ang.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Sensors thanks Yann-Wen Lan and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Area ratio dependent on pulse ID-VG behavior.

The area ratio (AR) is defined as the ferroelectric-to-dielectric area proportion. Pulse ID-VG curves measured from representative devices across a sweep width range from 50 µs to 100 ms for area ratios of: a, AR = 0.4; b, AR = 0.3; c, AR = 0.1; d, AR = 0.06; e, AR = 0.04; f, AR = 0.03.

Extended Data Fig. 2 Cycle-to-cycle transfer characteristics for different area ratios.

Dual-sweep DC measurements over 50 cycles at VD = 100 mV with area ratios of: a, AR = 0.4; b, AR = 0.3; c, AR = 0.1; d, AR = 0.06; e, AR = 0.04; f, AR = 0.03. A total of 50 dual-sweep cycles from representative devices is shown, where the light-blue curves represent individual cycles and the dark-blue line indicates one representative cycle.

Extended Data Fig. 3 Benchmarking with state-of-the-art 2D-FeFETs.

a, Benchmark of state-of-the-art 2D-material FeFETs showing on/off ratio and memory window (MW)18,19,20,21,22,23,24,25,26,28,29,30,31,32,33,34,35,36,37,38. b, Benchmark plot comparing the memory window efficiency and ferroelectric thickness among the reported 2D-FeFETs and our devices18,19,20,21,22,23,24,25,28,29,31,32,34,35,36,37.

Extended Data Fig. 4 Response current uniformity under generator operating conditions.

Comparison of 9 individual device currents to the average crossbar array current: measured after programming voltage (VP) a, 4.8 V and b, 6.0 V under generator conditions (tPW = 100 ns), with the read voltage (VD) swept from 0.05 V to 0.562 V in 2 mV increments.

Extended Data Fig. 5 Response current uniformity under discriminator operating conditions.

Current responses of 9 devices in the array under discriminator conditions: a, at 4.6 V and b, at 6.0 V bias voltages, with tPW = 100 µs. VD was swept from 0.05 to 0.306 V in 8-bit steps of 0.01 V.

Extended Data Fig. 6 Schematic of device-based deconvolution for pattern ‘N’.

a, Original 5×5 pattern padded to 7×7 size with zeros. b, The input matrix split into two matrices: one with positive values and another with negative values multiplied by –1. c, Two sets of 3×3 devices, programmed with the corresponding voltages and serving as kernels, are transposed and simultaneously used to perform operations with the input pattern. d, Pattern ‘N’ is derived from the final current response of the differential circuit formed by the two sub-arrays.

Supplementary information

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Li, L., Li, C., Zheng, H. et al. Dual-mode ferroelectric transistors for high-performance generative-adversarial-network-based imaging. Nat. Sens. 1, 222–231 (2026). https://doi.org/10.1038/s44460-025-00024-w

Download citation

  • Received:

  • Accepted:

  • Published:

  • Version of record:

  • Issue date:

  • DOI: https://doi.org/10.1038/s44460-025-00024-w

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing