A BJT-Based CMOS Temperature Sensor Achieving an Inaccuracy of $\pm 0.45{{}^{\circ}\mathsf{C}}(3\sigma)$ from -50°C to 180°C and a Resolution-FoM of 7.2pJ.K<sup>2</sup> at 150°C
Bo Wang, Man‐Kay Law, Amine Bermak
Abstract
Integrated temperature sensors for industrial digital transformation such as turbine and bearing monitoring should exhibit low power consumption and high energy efficiency with moderate inaccuracy over a wide sensing range (e.g., > 150°C) to achieve autonomous operation under a limited energy budget. Even though resistor-based temperature sensors can achieve a superior sub-pJ.K <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> resolution-FoM [1], they typically require a 2-point trim together with a high-order nonlinearity correction (6 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> -order in [2]), inevitably burdening the processing cost. In contrast, BJT-based temperature sensors in bulk or SOI CMOS can achieve accurate sensing at high temperature with only 1-point trim and simple digital processing [3], [4]. However, they can suffer from a degraded energy efficiency at high temperature for ensuring the sensing resolution and/or accuracy (e.q., ~3× increase in bias current for improving the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3\sigma$</tex> -inaccuracy from <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\pm 0.6{{}^{\circ}\mathsf{C}}$</tex> to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\pm 0.4{{}^{\circ}\mathsf{C}}$</tex> in [3]). This paper describes a BJT-based temperature sensor capable of wide-range operation from −50°C to 180°C. By employing a nonlinear readout and the proposed subranging, double-sampling, and constant-biasing techniques, this work achieves a high resolution-FoM over the entire sensing range (9.7pJ.K <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> at room temperature and 7.2pJ.K <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> at 150°C), corresponding to a 6-to-10× improvement when compared with prior BJT-based wide-range designs [3], [4]. We further employ dynamic error-correction [5] and switch-leakage compensation to effectively suppress the mismatch- and leakage-induced errors, resulting in a high precision of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\pm 0.46{{}^{\circ}\mathsf{C}}(3\sigma)$</tex> .