Yuheng Chen Email & Phone Number

Lafayette, In, Us

West Lafayette, In, Us

West Lafayette, Indiana, United States

1. Authentication in chip encryption through deep engine-based processing of tampered optical responses• Designed a RAPTOR (Residual, Attention-based Processing of Tampered Optical Response)discriminator for identifying adversarial tampering of an optical, physical unclonable function based on a random array of gold nanoparticles embedded in semiconductor packaging.• Extracted features of gold nanoparticles from 1000 dark-field images in just 27 ms and verified their authenticity using RAPTOR in 80 ms with 97.6% accuracy under difficult adversarial tampering conditions.2. Advancing photonic design with topological latent diffusion generative model• Developed topology optimization (TO) based deep generative model: Topological Latent Diffusion Model (TLDM), to realize high-quality photonics device inverse design. • Took advantage of efficiency prediction model-embedded conditional U-net and demonstrated substantial efficiency improvement compared with state-of-the-art generative model benchmarks.3. Quantum-inspired latent polynomial unconstrained binary optimization (PUBO) for device co-design• Mapped device optimization problem into latent PUBO energy model to enforce the combinatorial optimization problem to the data optimization problem.• Introduced variational neural annealing implemented through recurrent neural networks (RNNs) to solve PUBO in energy state, significantly outperformed simulated annealing and quantum annealing on sampling time and device efficiency.4. Multimodal model for prompt-guided integrated photonics design• Utilized stable diffusion model for a device feature description text-device topology image multimodal dual-training. • Combined ChatGPT API with the packaged trained model to realize an interactive LLM-empowered prompt-guided photonics device design interface.

Milpitas, California, United States

1. Neural network solution for KLA AcuShape# chip architecture software design workflow optimization• Developed neural network to realize geometry critical dimension (CD) and spectra prediction/validation for Samsung V-NAND device top-down and cross-sectional metrology, substituting RCWA numerical method in the previous workflow.• Designed bidirectional training networks to accomplish software library generation and interpret sophisticated datasets, achieving an 80% smaller training dataset and better linearity performance. 2. Investigate effective device degree-of-freedom (DOF) reduction via generative AI for spectra metrology• Built up a variational autoencoder-based model to effectively reconstruct and sample large DOF geometry critical dimension (CD) within chip spectra metrology process into decreased DOF in compressed latent space.• Perturbed compressed DOF CDs during high-quality spectra output regression optimization and decreased 70% of regression iterations/simulation resource usage.

Hefei, Anhui, China

Self-organized lithography-free nanodevice fabrication with tunable optical anisotropy• Implemented lithography-free nanofabrication method as team leader, realizing 3-fold aspect ratio promotion in self-organized metal co-deposition ion etching.• Demonstrated outstanding tunable optical anisotropy feature in polarization, fitting well with RCWA simulation.

Tempe, Arizona, United States

Integrated 2D semiconductor light-emitting devices with plasmonic nanostructures • Realized first experimental transfer and emission characterization of 2D TMDs (Transition-metal dichalcogenides) on plasmonic nano-terrace morphology.• Generated 12-fold light emission enhancement with flexible manipulation feature, LSPR (Localized surface plasmon resonance) enhancement mechanism verified through FDTD simulation.

Doctor Of Philosophy - Phd, Electrical And Computer Engineering

Prof. Vladimir M. Shalaev, Prof. Alexandra Boltasseva, and Prof. Alexander V. Kildishev's Group

Optical Engineering

Prof. Shaojun Fu's Group, Honored Graduate