The average solar flux during testing was 770 W/m2, and the greatest solar power generation was attained with the graphite counter electrode and en­riched medium-density CNT active electrode. The efficiency of that cell was 1.8 x 10-5. Compared to the all-CNT construction, an improvement of more than a factor 10 was attained. If a cell were constructed with the graphite counter electrode and the low-concentration CNT enriched active electrode, an increase of power by a factor 2 is anticipated. This can be deduced by com­parison of the medium density enriched cell to the low density enriched cells with the regular construction. As the graphite counter electrode lowered the output resistance by a factor ~3, the power output may be larger by a factor 3 as well. Further improvements may be obtained by changing the aspect ratio of the solar cell. In the design reported here, we used effectively square films.

Changing the cell design by making the cells wider, will lower the resistance further. An aspect ratio of10 can then reduce the film resistance by a factor of 10, causing a reduction of R, which will improve P. Our thin cell results indicate that the largest resistance is due to the nanotube film, we therefore believe the efficiency increase with this improvement may be as large as 10­fold. We believe the greatest efficiency increase may be obtained by using CNT source material with a greater fraction of semiconducting nanotubes. The films we used had 90% semiconducting nanotubes and 10% metallic nanotubes. As we argued above, the semiconducting nanotubes provide the photo-generated current, but the metallic nanotubes short the load. If one were to use 99% semiconducting films, the amount of nanotubes could be increased by a factor 10, while still maintaining the same number of metallic nanotubes. As metallic nanotubes are more conductive than semiconducting nanotubes, we assume that the number of semiconducting nanotubes can be increased by this factor 10 without affecting Rq. Future generation cells can then reasonably be expected to deliver 100 times more power, due to the increase of ns by a fac­tor 10 (equation 4). However, at a certain abundance factor of semiconducting to metallic nanotubes, this argument will not hold any longer. Combining all of these improvements may lead to an efficiency of 0.8-5%, where the lower bound is a conservative estimate that every improvement will only contribute half we argued above. We hope the studies reported here will motivate fur­ther development of methods to create highly-enriched semiconducting CNT source material cost effectively at a large scale.