This thesis describes the design of a digital backend for a resistor-based temperature sensor realized in a 180nm CMOS process. The sensor’s analog frontend outputs a bitstream at 500kHz, which is then decimated and linearized with the help of 3 different polynomials, whose coefficients are obtained after a calibration process. A polynomial engine, with storage registers for the coefficients, a decimation filter, and an SPI communication module for off-chip communication of results/coefficients were implemented. The polynomial engine was implemented in fixed-point representation using two evaluation methods: Horner’s rule and scaling method. The implemented digital backend does not degrade the resolution, accuracy, and energy-efficiency performance of the sensor. It has an area of 0.068mm2 (compared to the 0.12mm2 of the sensor). It consumes 28µW (compared to the 66µW of the sensor). With the digital backend, the sensor achieves a resolution of 501µK at 25℃, which is only slightly worse than with off-chip (floating point) calculations (476 µK). The sensor’s inaccuracy, determined from 32 samples, also maintained its original order of magnitude: 0.05℃ (3σ) over the military temperature range (-55℃ to 125℃).

This thesis describes the design and characterization of thermistor-based electrothermal filters (ETFs) intended for use in high-accuracy CMOS temperature sensors. ETFs have been previously realized by placing an on-chip heater in close proximity to an on-chip thermopile, which then picks up the thermally-delayed signals generated by the heater. This delay is a well-defined function of absolute temperature and can be used as the basis for highly accurate temperature sensors. In this work, the thermopiles are replaced by a thermistor, resulting in greater sensitivity and higher resolution. Measurements show that the new ETFs can achieve 3.6m℃ resolution in a conversion time of 1s and 0.2℃(3ó ó) untrimmed inaccuracy from 30℃ to 60℃.