Winner-take-all (WTA) circuits are commonly used in neuromorphic design for selecting the highest input among a group of alternatives. Device fabrication introduces mismatch in circuits due to random errors in doping as well as errors in physical dimensions. Mismatch biases the circuit and results in incorrect decision making. In this paper, we analyze and characterize behavior of a sub-threshold WTA circuit due to mismatch introduced in the threshold voltage and scaling current of a transistor. Further, we propose two mismatch compensation techniques using floating gate programming. We compare the I-V characteristics before and after compensation in both techniques.