Artificial-intelligence-based design optimization of low-noise transimpedance amplifiers for optical detection

Photodetectors based on transimpedance amplifiers are important in building blocks for many electro-optical instrumentation systems, including optical ultrasound detection. We present an artificial-intelligence (AI)-based design optimization of transimpedance amplifiers for optical detection, by a Genetic Algorithm (GA). We explore the influence of the GA parameters: population size, number of generations, and mutation rate. We optimize a photodetector for the acousto-optic detection of nonlinearities in resonant ultrasound transducers, comparing the results to a Monte Carlo (MC) optimization, by using the systematic search as a reference. A photodetector optimization based on AI, with a population size of 1000 chromosomes, 10 generations, and 5% mutations, requires 104 evaluations. For these GA parameters, we find that, in 95% of the simulations, the merit differs in less than 0.12% from the maximum value (from the systematic search). To achieve the same statistical result, the MC optimization requires a much larger number of evaluations: 3.4 × 105. For both design optimization algorithms, we demonstrate that the performance scales with the size of the initial population following a power law. The exponent for the GA is larger than that of MC (1.2 vs 0.88). We experimentally validate the photodetector AI-based optimization, obtaining measurements in agreement with the specified values. We also apply the AI-based design optimization to two designs from the literature: a DC-coupled general-purpose detector and a photodetector for the research of quantum processing. We conclude that the AI-based design optimization is a promising method that enables low-noise transimpedance amplifiers for ultrasound detection.