Numerical Algorithms for Simulation of Multisection Lasers by Using Traveling Wave Model

Abstract

Abstract. Sequential and parallel algorithms for the simulation of the dynamics of high-power semiconductor lasers is presented. The model equations describing the multisection broad-area semiconductors lasers are solved by the finite difference scheme, which is constructed on staggered grids. This nonlinear scheme is linearized applying the predictor-corrector method. The algorithm is implemented by using the ParSol tool of parallel linear algebra objects. For parallelization we adopt the domain partitioning method, the domain is split along the longitudinal axis. Results of computational experiments are presented, the obtained speed-up and efficiency of the parallel algorithm agree well with the theoretical scalability analysis.

Effective Numerical Integration of Traveling Wave Model for Edge‐emitting Broad‐area Semiconductor Lasers and Amplifiers

“…The finite difference schemes of the Crank-Nicolson type for the full TW model were described in our previous publication [7]. We remind here only some most relevant aspects of these schemes and propose some modifications in approximation of the polarization functions p ± .…”

“…In contrast to [7], we treat the lateral boundaries of the field functions in a different manner. We assume, that the outer lateral regions of the laser device are rather simple, i.e., they are not pumped, there is no direct counterpropagating field coupling, and the impact from the polarization equations can be neglected there:…”

“…For more details see [7]. Let us assume that we know the grid functions E ± h and p ± h at the time layer t n and the grid function N h at t n−0.5 .…”

“…The simulation of a laser during typical 3 ns transient time which, usually, is just enough to simulate the switching on of the laser or the relaxation of the device towards some new attractor after a change of parameters requires ∼ 0.5 · 10 5 time iterations and can be performed on a single processor computer in, approximately, 4-5 hours [13]. Some speed-up of computations can be achieved by using problem-dependent relations of the grid steps, including also variable steps in lateral direction of the device [7,9]. All these grid optimizations, however, are not sufficient once one-or a few-parameter studies should be performed (they can correspond to simulation times ∼1000 ns) The required results can only be computed in acceptable time by using parallel computers and parallel solvers.…”

“…Seeking to optimize the performance of the numerical schemes for TW model discussed earlier in [7,9] we investigate here also a related 1+1 dimensional Schrödinger equation describing a diffractive beam propagation in linear gain or/and index guided media. The results obtained in solving this simplified auxiliary problem by using artificial boundary conditions [1,14] and special non-uniform space grids with discretization steps from the moderate-to-large sizes let us to choose a proper strategy allowing to resolve the full TW model with a required precision in reasonable time.…”