THz diffractive lenses have recently gathered a lot of attention as an effective alternative to conventional THz lenses which are bulky, thick, expensive and suffer from strong wavefront (geometric and chromatic) aberrations. It has already been shown that employing a direct binary search technique on the actual height profile of diffractive elements (DEs) can yield designs with better or comparable performance metrics with respect to kinoforms. Such search techniques are however, limited by their exponential time complexity in an unstructured design solution space. If the solution can be proven to exist, we can further perform a gradient descent optimization along with the binary search to overcome the time complexity required to arrive at the desired solution.
The semi-analytic and measured efficiency for all diffractive lenses under both narrowband and broadband focusing is > 80% based on our modified DBS employed design prediction. The modified DBS is observed to converge much faster for both 1D and 2D diffractive lens cases (>10X) with respect to conventional direct binary search based design prediction. For 2D diffractive lenses, the figure of merit is initially high and takes a longer time to converge to the desired solution, which can be understood from the fact that the number of “dielectric” pixels in a 2D lens is much greater than its 1D analogue. Furthermore, since the direct binary search is an iterative algorithm, it convergence depends a lot on the initial random pixel height profile, which is not the case in the modified DBS method.