Photonics

Photonics

Optimisation of parameters for two wavelengths

    • sagar
      Subscriber

      I am optimizing my structure's parameters, diameter, height, etc to gate maximum absorption. Is it possible to optimize the parameters of the photonic structures for two wavelengths in a single optimization process? In optimizations, I am only able to maximize the absorption for a single wavelength. Please suggest if it is possible with lumerical fdtd? 

    • Guilin Sun
      Ansys Employee

      The Particle Swarm optization Lumerical developed is for one FOM. When you have two results to be optiized, it is doable, provided that you design a FOM to contain both of them in one value. For example if you want mimize 1-T1/T2, where T1 is the transmission for wavelewngth 1 and T2 is the transmission for wavelength 2.

       

      You can design other FOM to better realize your optimization goal, as one value of FOM.

       

       

    • sagar
      Subscriber

      This 1-T1/T2 will be minimized when T1~T2, not likely when both are maximum. I want to do the optimization in such a way that T1 and T2 would maximum corresponds to two wavelengths and transmission corresponding to other wavelengths will be a minimum. The actual goal is to build a structure that supports resonance for two wavelengths. 

       

      Sometime after optimization, I get diverged FOM; there is no saturation value. Should I go for more generation? 

    • Guilin Sun
      Ansys Employee

      Unfortunately many optimizatio algorithms do not provide vectorial FOMs. The Particle Swarm Optimization Lumerical developed is a scalar optization, so you cannot direct optimize two FOMs. The suggestion I proposed can have both T1 and T2 about the same. You may need to figure it out where the loss goes, reflection or absorption? You can also invent your own FOM formulation and try if it works.

      You could also borrow 3rd party vectorial optimization algorithm, or write your own.

      If the FOM does not converge, it means you will need more number of generations. Yes, it needs to reach more parameter space in order to converge.

       

Viewing 3 reply threads
  • You must be logged in to reply to this topic.