Reference Plant · FabONE · explainable ML

The FabONE process, as an explainable neural-network simulator.

A neural network — trained in your browser on the FabONE graphene-photonics process — predicts the functional yield, cost and device performance from the physical process parameters. It is not a black box: move a lever and you see the prediction change and why. The model exposes its sensitivity (gradients), what it learned per parameter (partial dependence), the network itself (live activations and weights), a 2D response surface, and where the yield is lost step by step. Every parameter range is grounded in the graphene-photonics literature.