CMOS Temperature Sensor with Programmable Temperature Range for Biomedical Applications

A CMOS temperature sensor circuit with programmable temperature range is proposed for biomedical applications. The proposed circuit consists of temperature sensor core circuit and programmable temperature range digital interface circuit. Both circuits are able to be operated at 1.0 V. The proposed temperature sensor circuit is operated in weak inversion region of MOSFETs. The proposed digital interface circuit converts current into time using Current-to-Time Converter (ITC) and converts time to digital data using counter. Temperature range can be programmed by adjusting pulse width of the trigger and clock frequency of counter. The proposed circuit was simulated using HSPICE with 1P, 5M, 3-wells, 0.18μm CMOS process (BSIM3v3.2, LEVEL53). From the simulation of proposed circuit, temperature range is programmed to be 0 °C to 100 °C, it is obtained that resolution of the proposed circuit is 0.392 °C with -0.89/+0.29 °C inaccuracy and the total power consumption is 22.3 μW in 25 °C.


INTRODUCTION
Engineering is an innovative field that its origin ideas leading to everything, including biology and medical area. Application of engineering in biology and medical area is then called biomedical engineering. The purpose of this field is combining the design and problem solving skills of engineering with medical and biology sciences to advance health care treatment, including diagnosis, monitoring, and therapy [1]. In the recent years, there is a lot of research focus on biomedical engineering. This topic is becoming interesting and challenging due to the increase of system complexity, human population and its distribution.
One of the most important components in biomedical engineering is sensor. This component is very important because it is directly connected to physical phenomenon [2,3]. In many biomedical applications, sensor plays a crucial role to collect data from many objects such as human body temperature, environment humidity and oxygen levels in the air. The often utilized sensor in biomedical applications is temperature sensor, because temperature is a highly important parameter to monitor, identify, or control many conditions in biomedical field, such as diagnosing human disease, temperature monitoring in operating rooms and preventing bacteria growth in some places. By knowing the useful of temperature sensor in many biomedical applications, the availability of a temperature sensor which can be used in various biomedical applications is highly recommended. The problem is different biomedical application has different temperature range to be measured. It means that the sensor must have wide temperature range. However, wide temperature range causes the resolution of its digital data decreases. To keep the resolution high, high-bit analog to digital (ADC) must be applied. Nevertheless, it will decrease the speed and increase the power consumption. Furthermore, in biomedical applications, not only high sensitivity temperature sensor is needed but also low voltage and low power temperature sensor is strongly required [4], [5], [6].
A low power and low voltage temperature sensor has been proposed and reported [7], [8]. This sensor in IJECE ISSN: 2088-8708 947 general achieves low power consumption, but it has small sensitivity. The other high sensitivity temperature sensors are also proposed [9], [10]. However, these sensors are not low voltage and it consumes high power. The other problem is that many conventional on-chip temperature sensor circuits use BJT (V be ) to sense temperature [4], [9], [11], [12]. The problem of BJT is not able to be implemented in the same chip in many standard CMOS processes. This means that cost of fabrication will increase. The last problem is that many cores of temperature sensor circuits use external bias circuits or high value resistor [8], [9], [13]. These require large chip areas. A low voltage and low power temperature sensor circuit with digital output for health care monitoring system has been proposed in [14]. This sensor can achieve low voltage, low power, high sensitivity and high resolution. However, this sensor is special for health care monitoring system whose temperature range only 33°C to 45°C, and it can not to be used for other biomedical applications with different temperature range. This paper proposes a new temperature sensor circuits based on previous work [14] with addition of programmable temperature range digitalization. In this paper, the analysis of the circuits, simulation results and the measurement of fabricated temperature sensor core are reported in detail. This paper is organized as follows. The previous research and improvement are presented in Sect. 2. In Sect. 3, the simulation results of the designed circuit and measurement of fabrication temperature sensor core are presented. Finally, Sect. 4 concludes this paper.

PREVIOUS RESEARCH AND IMPROVEMENT
In this section, the previous research [14] and its improvement are presented in detail. Fig. 1 shows the block diagram of the proposed temperature sensor in previous research. This proposed temperature sensor consists of some sub circuits: sensor core circuit [3], voltage to current converter (VIC) circuit, 1/x circuit, current to time converter (ITC) circuit [15], and counter.
Where V θ is the thermal voltage, k(= 1.38 × 10 23 J/K) is the Boltzmann's constant, q(= 1.60 × 10 19 C) is the electron charge, T K is the absolute temperature, n is slope factor, η is DIBL (Drain Induced Barrier Lowering) coefficient, µ is carrier mobility, C d is capacitance of the depletion layer, C ox is capacitance of the oxide layer.
In (1), if V ds ≥ 4V θ , 1 exp(V ds /V θ ) is satisfied, as the results exp(V ds /V θ ) can be ignored. Furthermore, because it is well-known that ηV ds is small value, (V gs V th ) ηV ds can be satisfied. From these conditions, (1) can be rewritten to (5).
And the threshold voltage with the body effect of MOSFET is generally given by (7).
Where V th0 is the zero-bias threshold voltage, γ is the body effect coefficient, 2Φ F is the surface potential parameter. If V sb is small enough, (7) can be approximated as follows.
From Fig. 2 (a) it can be known that V sb1 = 0 and V sb2 = V P T AT 1 , then V P T AT 1 can be expressed as by (9) It is assumed that γ = 0.61 and 2Φ F = 0.7, the denominator of (9) is calculated as follows Slope factor n is known to be a value of approximately 1.5, therefore, it can be considered that n' is relatively close to the value of n. As a result, V P T AT 1 can be expressed by equation (11).
From (11), it can be found that V P T AT 1 of the proposed circuit is directly proportional to T K . Moreover, it could be understood that the output voltage of Fig. 2 is proportional to absolute temperature. The sensitivity of the sensor can then be increased using cascade connection of the circuit. Fig. 2 (b) shows the cascade connection of circuit Fig. 2 (a). Using (11), V P T AT n of Fig. 2 (b) is as follows.
Using (11) and (12), VPTAT2 is then given by From (13), since the output voltage can be expressed by summation of the logarithmic term, this method can be used to increase sensitivity effectively.  Figure 3 shows Voltage to Current Converter (VIC). From the figure it can be inferred that V m = V in , and hence the following equation can be obtained.
I R is then copied by two current mirrors (M vi1 , M vi3 , M vi4 , and M vi5 ) to I P T AT . If the ratio of W/L between M vi1 Where I ref is current source which have no temperature dependency, V b is supplied by reference voltage circuit, and I P T AT is the output current of voltage to current converter. The temperature dependence of I a is very small because it utilizes translinear principle. The operating principle of ITC could be described as follows, When V tgr shown in ITC circuit is low, M x3 , M x4 , and M x5 are OFF, and the circuit is in the idle mode. Therefore, V p reaches V dd , as the result, T out becomes Low. When a short single pulse is applied to the gate of M x5 , not only M x5 but also M x3 and M x4 become ON, ISSN: 2088-8708 because V p becomes Low. As the result, I a flow in C, and C store the charges that are proportional I a . During this period, T out is high. The capacitor C is continuously charged, and V p is increasing. When V p reaches the threshold voltage of Inv 1 (V inv1 ), T out is low, M x3 and M x4 become OFF and M x2 becomes ON, respectively. Therefore, the charge of C is discharged through M x2 , and then the circuit returns to the idle mode. Using (16), T out can be given by From (17), it can be inferred that T out is proportional to I P T AT . T out can be converted to the digital value by counting up the period of high level in T out by a counter.

Programmable Temperature Range Digitalization
In the sub section 2.3 it has been explained that digital value of the measured temperature can be obtained by counting up the high level period of T out by a counter. However, if T out is directly connected to counter, it will be ineffective. Because it is known from equation (17) that T out is proportional to absolute temperature. It means that the high level of T out will appear as far as the temperature is larger than absolute zero. In the other hand, it is well known that in the biomedical applications, the measured temperature is much higher than absolute zero, for example the critical human body temperature is 35°C (hypothermia) to 41.5°C (hyperpyrexia). Thus, if T out is directly connected to counter, the temperature range of digital conversion result will be difficult to be adjusted. Moreover, the digital conversion will not reach high resolution.
In T out -counter direct connection, the temperature range can be adjusted using clock frequency of the counter. The higher clock frequency of counter, the more narrow temperature range that can be convert to digital data. Oppositely to the temperature range, the resolution of digital data is higher. However, for narrow temperature range biomedical applications, T out -counter direct connection is still not effective way to be used. Because T out pulse width is start to appear right after absolute zero temperature, and increases proportionally as temperature increases. Therefore, if the minimum temperature of the biomedical applications is much larger than absolute zero, the unnecessary conversion will exist. This problem will also make the adjustment difficult, because the counter will be overflow many times before start to converts desired T out pulse width. Figure 5 shows the desired conversion and unnecessary conversion in T out pulse width. To make easier temperature range adjustment and more effective conversion, unnecessary conversion must be eliminated. In other word, the counter counting up only in desired conversion region in T out pulse width. To perform this function, the connection between T out and counter is modified. This architecture is shown in Fig. 6. Using architecture in Fig. 6, the temperature range can be programmed or adjusted not only by clock frequency of the counter but also by pulse width of the trigger (V tgr ). Since V tgr is connected to Rst using an inverter, the counter will be in reset condition and it does not count up as far as the V tgr is high. Thus, the conversion only in desired temperature range can be performed by keeping high V tgr from t 0 to t 1 shown in Fig. 5. The pulse width of V tgr is written in equation (18). The clock frequency of the counter then can be calculated using equation (19).
Where t 1 is pulse width of minimum temperature in temperature range, t 2 is pulse width of maximum temperature in temperature range, and N is resolution of the counter (bit). Generating pulse and clock with various IJECE ISSN: 2088-8708 951 width and frequency is an easy thing that can be performed in programmable devices, like microcontroller and microprocessor. In other word, the temperature range of the proposed temperature sensor is easily programmable, so that it can be used in various biomedical applications.

SIMULATION RESULTS AND MEASUREMENT OF FABRICATED TEMPERATURE SENSOR CORE
The performance of the proposed circuit was evaluated using HSPICE with 1P, 5M, 3-well, 0.18-m CMOS process (BSIM3v3.2 LEVEL53). All simulations use 1.0 V supply voltage. Figure 7 shows the simulation result of sensor core circuit in the temperature range of -40°C to 160°C. V P T AT 1 is output of single sensor core, and V P T AT 2 is the output of cascade connection of two sensor cores. From this simulation, it was obtained that sensitivity of single sensor core is 0.392 mV/C with 0.78 % nonlinearity. This sensitivity could be increased using cascade connection like shown by V P T AT 2 , its sensitivity is 0.783 mV/°C, with 0.89 % nonlinearity. In order to verify the performance of temperature sensor core, this temperature sensor core is fabricated using 0.6 µm CMOS process. Figure 8 shows the average measurement results of 10 different chips. From this figure it can be known that the output voltage of the temperature sensor core is proportional to temperature. The average sensitivity of 10 measured chips is 0.8343 mV/°C. Figure 9 shows the accuracy of fabricated temperature sensor core. From this measurement it was obtained that its inaccuracy is -1.144/+1.059 mV or 2.70% nonlinearity. Figure 10 shows the relationship between temperature and pulse width. In these simulations, temperature ISSN: 2088-8708 was changed from -40°C to 160°C in step of 10°C. As a result, it was obtained that the pulse width was proportional to temperature with 0.276 µs/°C sensitivity and 2.14 % nonlinearity. Based on the data from these simulations, temperature range programmability of this proposed circuit is verified by programming its temperature range to be 0°C to 100°C. Using equations (18) and (19) the pulse width of V tgr and clock frequency of the counter (CP) are 70.9 µs and 9.59 MHz, respectively. Figure 11. D out -temperature characteristic Figure 12. Inaccuracy of the proposed temperature sensor Figure 11 shows digital output of the simulation result after the temperature range of the sensor is programmed to be 0°C to 100°C. From the simulation it was obtained that digital code D out is proportional to temperature. The digital code of this simulation was 0 to 255. This means that 1 LSB is equal to 0.392°C. Figure 12 shows the accuracy of the proposed integrated temperature sensor circuit. The accuracy measurement was done using two calibration points (0°C and 100°C). From this measurement it was obtained that inaccuracy of the proposed circuit is -0.98/+0.29°C. From these simulation results it can be verified that the temperature range of the sensor can be programmed well.

CONCLUSION
In this paper, the integrated temperature sensor circuit is constructed using sensor core, voltage-to-current converter (VIC), 1/x circuit, current-to-time converter (ITC) and counter. Sensor core is formed using CMOS circuit operated in weak inversion region and it is insensitive to device parameter of fabrication process. The output of the sensor is then digitized using proposed programmable temperature range digital interface.
The performance of the proposed circuit was evaluated using HSPICE with 1P, 5M, 3-wells, 0.18-m CMOS process (BSIM3v3.2 LEVEL53). As a result, sensitivity of temperature sensor core is 0.783 mV/°C, with 0.89 % nonlinearity in -40°C to 160°C. This temperature sensor core has been fabricated using 0.6 µm CMOS process. As a result of 10 different chips measurement is the average sensitivity of fabricated chip is 0.8343 mV/°C in 10°C to 60°C, with 2.70% nonlinearity.
Temperature range of the sensor is then programmed to be 0°C to 100°C using pulse width of V tgr and IJECE ISSN: 2088-8708 953 clock frequency of counter (CP). As the results of simulation, resolution of the sensor that its temperature range has been programmed is 0.392°C with -0.98/+0.29°C inaccuracy and total power consumption is 22.3 µW in 25°C. The future work of this research is designing mask layout of the proposed digital interface, fabrication of whole circuit in one prototype chip and evaluation of the characteristic.