I am currently a senior scientist at Flexcompute Inc, working to bring elegromagnetic simulations to a new level of speed and scale. Previously I was a postdoc at Stanford University's EE department, in the group of prof. Shanhui Fan. I'm interested in all things photonic, my longest-term passion has been photonic crystals, while a more recent one is automatic differentiation for physical simulations.

When I'm not squeezing Maxwell's equations simulations out of my computer, you can most probably find me at home, but if I'm not there then try looking around some mountain or national park, or somewhere playing soccer or volleyball.

Selected works

(See a full list of publications here)

Photonic crystals optimization through automatic differentiation
We developed a package for the efficient simulation and optimization of photonic crystals, which allows the user to efficiently compute gradients of all output quantities (e.g. eigenmode frequencies and field profiles) with respect to all input parameters (e.g. hole positions and shapes). What I find particularly exciting is the general idea of using automatic differentiation for the simulation of physical systems. Packages like TensorFlow and PyTorch, which have become extremely sophisticated in the past decade largely because of machine learning, are, in their core, just autodiff libraries. We can use these to "backprop" through a physical simulation, and perform really complicated optimizations with a large number of free parameters. This could be a game changer for next-generation devices, in photonics and beyond.
Momchil Minkov, Ian Williamson, Lucio Andreani, Dario Gerace, Beicheng Lou, Alex Song, Tyler Hughes, and Shanhui Fan (2020)

Doubly-resonant photonic crytal cavity for second-order nonlinearities
We have designed a microscopic photonic crystal resonator that confines light at a given wavelength and at half that wavelength. This serves to strongly enhance second-order optical nonlinear frequency conversion processes, which find applications in both classical and quantum technologies. Photonic crystal cavities have always been an attractive candidate for this problem, but it turned out really hard to design a device that confines light at two different wavelengths that are so far apart from each other. However, we recently had a breakthrough idea based on bound states in the coninuum, which allowed us to achieve the feat.
Momchil Minkov, Dario Gerace, and Shanhui Fan Optica (2019)

Wave physics as an analog Recurrent Neural Network
Analog machine learning hardware platforms promise to be faster and more energy-efficient than their digital counterparts. We have identified a mapping between the equations describing the propagation of physical waves and the computation done in recurrent neural networks. This mapping indicates that wave systems can be trained to learn complex features in temporal data, using standard training techniques like automatic differentiation. As a demonstration, we show that an inversely-designed inhomogeneous medium can perform vowel classification on raw audio data simply through wave propagation.
Tyler Hughes*, Ian Williamson*, Momchil Minkov, and Shanhui Fan Science Advances (2019)

Experimental band structure along a synthetic dimension
'Synthetic dimensions' are internal degrees of freedom of a system that can be coupled to form higher-dimensional lattices in lower-dimensional physical structures. Here, we provide a direct experimental measurement of the Bloch bands along such a synthetic dimension. By dynamically modulating an optical resonator, we create a lattice of coupled modes in the frequency dimension. We then show in theory and experiment that time-resolved transmission measurements provide a direct readout of the band structure. We also realize long-range coupling, gauge potentials and nonreciprocal bands by simply incorporating additional frequency drives, enabling great flexibility in band structure engineering.
Avik Dutt, Momchil Minkov, Qian Lin, Luqi Yuan, David Miller, and Shanhui Fan Nature Comm. (2019)
  • PDF

  • Inverse design for nonlinear nanophotonic devices
    Inverse design is the idea of using computational optimization techniques to create devices based on certain specifications. One way to achieve this is through gradient-based optimization, which requires an efficient way of computing the gradient of an objective function with respect to a large number of degrees of freedom. In photonics, this is commonly done using the adjoint variable method (reverse-mode differentiation), but this had previously only been applied to linear devices. Here, we present an extension of this method to modeling nonlinear devices in the frequency domain, with the nonlinear response directly included in the gradient computation.
    Tyler Hughes*, Momchil Minkov*, Ian Williamson, and Shanhui Fan ACS Photonics (2018)

    Zero-index bound states in the continuum
    Photonic metamaterials with an effectively zero refractive index can find applications like enhanced light emission, directional transmission, cloaking, and enhanced optical nonlinearities. One particularly interesting approach for achieving such materials is using a photonic crystal slab with properly engineered photonic bands. This all-dielectric implementation eliminates the absorption losses that are significant in all other zero-index paradigms, but it still suffers from strong radiative losses. Here, we have designed the bands of a photonic crystal slab to be simultaneously zero-index and non-radiative, which could enable, for the first time, large-scale integration of zero-index materials in photonic devices.
    Momchil Minkov, Ian Williamson, Meng Xiao, and Shanhui Fan Phys. Rev. Lett. (2018)
  • PDF

  • Training of photonic neural networks through in situ backpropagation
    Integrated optics has gained interest as a hardware platform for implementing machine learning algorithms, since matrix-vector multiplications, which are used heavily in artificial neural networks, can be done efficiently in photonic circuits. In our work, we introduce a method that enables highly efficient, in situ training of a photonic neural network (as opposed to using a model on a digital computer). Essentially, we derive the photonic analogue of the backpropagation algorithm, and show how the gradients may be obtained exactly by performing intensity measurements within the device. This can enable highly efficient training of large-scale photonic neural networks.
    Tyler Hughes, Momchil Minkov, Yu Shi, and Shanhui Fan Optica (2018)


    (Some interesting things that have happened)