As one of the most important ambient parameters, temperature sensing has attracted much attention and research. CMOS temperature sensors are fabricated in standard CMOS technology, which make these satisfy the requirements of low-power, low-cost and micro-size for emerging applications. Moreover, CMOS smart temperature sensors can directly communicate with a microcomputer in a digital format with less complexity. In terms of advantages of CMOS temperature sensors and their huge potential market, it makes sense to investigate the CMOS temperature sensors. In CMOS temperature sensor design, one of the biggest issues is the process variations. Current solution such as trimming strongly increases the production cost. This thesis focuses on designing a low power, wide range CMOS temperature sensor with reduced calibration cost. In this thesis, a novel on-chip process compensation method is proposed to reduce the sensing errors induced by process variations. By exploiting piecewise base recombination current, BJT process spread can be reduced over a wide temperature range. Besides, weighted combinations of various types of resistors are applied to compensate the overall resistance variation. After the piecewise compensation and resistor compensation, the inter/intra- die variations in VBE are significantly reduced. Measurement results from 2 batches demonstrate that the standard deviation (STD) of VBE decreases from 3.4 mV to 1.5 mV. The temperature is extracted based on the measured VBE and ∆VBE . Using the calibration parameters from one batch, this work presents a maximum untrimmed batch-to-batch inaccuracy of ±1.5 ℃ (3σ) from -40 ℃ to 125 ℃. Since we only require one batch for multi-batch calibrations, the practical calibration cost can be potentially reduced, which demonstrates a low-cost solution for CMOS temperature sensors. We also design a 12-bit incremental ADC as sensor readout. Circuit techniques such as split-length transistors are used to design an opamp with reduced power and increased stability. Chopping techniques are used to cancel the offset of the opamp in the integrator. DEM techniques are used to eliminate the mismatches errors to get precision current ratios and voltage gain. After piecewise first-order linear fitting, this sensor presents the untrimmed temperature error of ±0.15 ℃ over the temperature range from -40 ℃ to 125 ℃. This results shows that the proposed compensation method can be applied to design a high accurate CMOS temperature sensor without trimming.