A hardware Trojan detection method based on kernel maximum margin criterion and an improved detection method are proposed on basis of the statistical model of power side-channel signal.The methods can map the raw power side-channel signal into a higher dimensional space

where it had a higher separability

and then it is projected onto a low-dimensional subspace

so that non-linear characteristics of differences in the raw data are found

and nonlinear characteristics extraction and recognition of power side-channel signal are achieved.The detection experiment against the Trojan circuit in AES encryption circuit shows that

the number of samples beyond the detection boundary by the method (792) is more than Karhunen-Love Transform (400)

which gets a better detection result.