There isn’t a single physical ISO-per-pixel-area limit where images suddenly stop being useful. In a hypothetical perfect sensor, the remaining limit is photon shot noise: light arrives randomly, and the noise grows roughly with the square root of the number of photons, so signal-to-noise improves as you collect more total photons.

Pixel area by itself is not the key limit. Smaller pixels look noisier per pixel, but if you downscale or compare images at the same output size, that noise averages out. What matters more is total light gathered by the whole sensor, plus exposure and scene brightness. Sensor saturation/full-well capacity would also matter if you want ISO to be defined meaningfully, because ISO depends on how signal maps to clipping and output brightness.

So the practical answer is: with zero electronic noise, image quality is still limited by shot noise, and “useful” becomes subjective. It depends on the scene, output size, viewing distance, and denoising. Extremely low photon counts can still produce recognizable images, but they become progressively grainier rather than hitting a hard cutoff.