The one standout graph to me from their paper showed that adding these sources in either serial or parallel configuration did not result in a linear increase of

The one standout graph to me from their paper showed that adding these sources in either serial or parallel configuration did not result in a linear increase of either current or voltage. 4x the units resulted in a 3.n multiplier. That looks to me like the beginning of a limit value that remains tiny. Of course, with only two points we can't tell if it's a curve or a straight line, and that doesn't happen by accident. In addition, the paper (paraphrased) says repeatedly that they've made thousands and thousands of these devices, and yet their entire serial/parallel exploration is a few combinations in the 4x regime. Nothing is said about throwing 1000+ of them in a row/parallel to see if a stronger signal emerges from the noise.