NEP as function of absorbed power for all measured devices. Top row: Data of all measured MKIDs using radiation coupling via a small pinhole to the small radiator source. The left figure gives the data for all KIDs identical to Fig3.e in the main text. The centre and right plots give the NEP divided by the photon noise limit, showing more clearly that the measured NEP can be below the photon noise limit for 10 aW < P_abs < 100 aW. Bottom row: Similar data, but now using the large radiator coupled without pinhole to the detector chip. We observe that the region where the experimental NEP is below the photon noise limit scales with P_abs, and not with radiator temperature.