This paper introduces an embedded solution for the detection of hardware trojans (HTs) and counterfeits. The proposed method, which considers that HTs are necessarily inserted on production lots and not on a single device, is based on the fingerprinting of the static distribution of the supply voltage (V dd) over the whole surface of an integrated circuit. The measurement of this fingerprint is done through an array of sensors sensitive to the local V dd value and fingerprint extraction is based on a novel variation model of CMOS logic performance. This model takes into account not only process variations but also the impact of the design (layout, supply routing, and so on). In addition to the fingerprinting process, this paper introduces an adaptive distinguisher to deal with the difficult problem of fixing the p-value on large sets of statistical tests. The efficiency of the whole detection methodology is experimentally demonstrated on a set of 24 FPGA boards.