A statistical temperature sensor

Abstract

The physical effect which is used in this type of temperature sensor is the temperature coefficient mismatch of transistors. The basic cell for evaluating the effect is a bi-stable cell. For instance this could be an SRAM cell. After powering up the circuit such a cell provides either a 1 or a 0 at its output. In the case that such a bi-stable cell is used as a temperature-sensor, a well-designed cell bases its decision on the mismatch of two NMOS transistors (N1 and N2). The threshold voltage of NMOS transistors is decreasing over temperature. The amount of change is mainly defined by the coefficient KT1 (see BSIM). It turns out that there is a mismatch on KT1 between the different transistors. Thus it may happen that the startup value (or initial value) may change over the temperature. If a huge number of cells is used an accurate value of the temperature can be given. Once a startup vector at a well-defined temperature is defined, the temperature difference can be derived by measuring the Hamming distance (HD) between the reference vector and the output at the actual temperature.