In this paper, we propose an optimized transmission scheme with energy-harvesting for a diffusion-based molecular communication system composed by nano-devices fed by piezoelectric nanogenerators. To this end, we firstly derive a system model that analytically describes the mean and the variance of the aggregated noise at the output of the receiver and the achievable Bit Error Rate. Then, we formulate an optimization problem that minimizes an objective function defined as a linear combination of the probability that the voltage across the ultra-nanocapacitor of the transmitter goes under a target value and the number of enqueued packets. We solve this problem by considering the actual energy budget, a target Bit Error Rate, and the need to achieve the simplicity of the transmitter as constraints. Finally, we use computer simulations to validate the formulated analytical models and demonstrate the unique ability of the proposed approach to guarantee BER = 5% and BER = 10% for communication distances up to 47 μm and 50 μm , respectively, while registering better results against baseline scenarios.