Wide-range CTAT and PTAT sensors with second-order calibration for on-chip thermal monitoring

Abstract

This paper presents wide-range complementary to absolute temperature (CTAT) and proportional to absolute temperature (PTAT) sensors with second-order calibration for on-chip thermal monitoring. Particularly, a current mirror and an n-well resistor are used in these sensors to eliminate the second-order term of the temperature coefficient to linearize the transfer function between the output voltage and the temperature. The proposed CTAT and PTAT sensors are implemented on silicon using a typical 0.18 μm CMOS process. The core area of the CTAT and PTAT sensors is 0.0125 mm² and 0.0074 mm², respectively. In the range from $-55°\mathrm{C}$ to $155°\mathrm{C}$, the worst deviation of the CTAT and PTAT temperature sensors is measured to be $-3.75°\mathrm{C}\mathrm{to}+3.34°\mathrm{C}$ and $-3.73°\mathrm{C}\mathrm{to}+3.85°\mathrm{C}$, respectively. The maximum non-linearity reduction of the CTAT and PTAT sensors is 59.84% and 87.48%, respectively, by the proposed second-order calibration. Notably, the overhead area of the CTAT and PTAT sensors is only 0.64% and 1.08%, respectively.

Introduction

It is well known that more transistors have been integrated in a chip by using advanced CMOS technologies. When many transistors are integrated and realized on a single die, the power density as well as the temperature will rise inevitably. Moreover, the chip with high switching activities will generate massive heat (namely hot spot) [1]. Another critical issue is the range of the temperature. The harshest temperature range is demanded by electrical vehicles (EV) and military equipments, which are $-40°\mathrm{C}$ to $125°\mathrm{C}$ and $-55°\mathrm{C}$ to $125°\mathrm{C}$, respectively [2], [3], [4]. Temperature sensors are then required to monitor the temperature of those chips in these applications in such a wide temperature range to ensure that no hazards would be caused by over heat problems.

Many prior temperature sensors relied on threshold voltage detection to estimate the temperature [5], [6], [7], [8]. The accuracy in a very wide range is still a challenge regarding on-chip temperature sensing. Vaz et al. proposed a temperature sensor for human body temperature monitoring, where a large resistor is used to cancel temperature non-linearity components of the saturation current [7]. However, this design was meant for low temperature range. Several logic-based design temperature sensors have also been proposed [1], [9]. Chung and Yang proposed an all-digital temperature sensor, where serious process and non-linear effects still exist [1]. The reason is that the process variation and second-order effects of MOSs in all digital circuits are hard to be detected and then calibrated. A few time-domain temperature sensors were also reported [10], [11], [12], [13]. Chen et al. proposed a time-domain temperature sensor based on a successive approximation algorithm [10]. The successive approximation algorithm, however, costs a large area overhead. Meanwhile, the high linearity and second-order effects calibration methods were reported [14], [15], [16], [17], [18]. Lin et al. proposed a non-linear calibration method by isolating body effect. This design, however, did not consider the process variation [14]. Jeong and Ayazi disclosed a process offset cancellation method by applying an identical circuit to realize temperature sensors [17]. The reported process compensation and second-order effects calibrated methods usually need a large area to realize compensation and calibration circuits [10], [11], [12], [13], [14], [15], [16], [17], [18]. Many low-cost calibration methods were also proposed [19], [20], [21]. Fisk and Hasan showed a method using the correlation between pinch-based resistance and substrate bipolar temperature gradient to carry out process-compensation [20]. Besides, Souri et al. also proposed high accuracy and low-power temperature sensors [22], [23], where silicon on insulator (SOI) process was used to attain a much wider temperature range and low power consumption [22]. Another prior design used dynamic threshold MOSTs (DTMOSTs) and an inverter-based second-order zoom analog-to-digital converter (ADC) to achieve high accuracy and low power [23]. However, the SOI process, DTMOSTs, and the other prior calibration technologies must pay high die area and cost [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26].

In this study, we propose complementary to absolute temperature (CTAT) and proportional to absolute temperature (PTAT) sensors with low-cost second-order calibration to resolve all the mentioned problems. By using the proposed second-order calibrated CTAT and PTAT sensors consisting of a current mirror and an n-well resistor, the linearity can be enhanced.