A-Smart-Single-Chip-Micro-Hotplate-Based-Gas-Sensor-System-in-CMOS-Technology-外文文献.doc

原文
A Smart Single-Chip Micro-Hotplate-Based Gas Sensor System in CMOS-Technology

Abstract
This paper presents a monolithic chemical gas sensor system fabricated in industrial CMOS- technology combined with post-CMOS micromachining. The system comprises metal -oxide-covered (SnO2) micro-hotplates and the necessary driving and signal-conditioning circuitry. The SnO2 sensitive layer is operated at temperatures between 200 and 350℃. The on-chip temperature controller regulates the temperature of the membrane up to 350℃ with a resolution of 0.5℃. A special heater-design was developed in order to achieve membrane temperatures up to 350℃ with 5 V supply voltage. The heater design also ensures a homogeneous temperature distribution over the heated area of the hotplate (1–2% maximum temperature fluctuation). Temperature sensors, on- and off-membrane (near the circuitry), show an excellent thermal isolation between the heated membrane area and the circuitry-area on the bulk chip (chip temperature rises by max 6℃ at 350℃ membrane temperature). A logarithmic converter was
included to measuring the SnO2 resistance variation upon gas exposure over a range of four orders of magnitude.An Analog Hardware Description Language (AHDL) model of the membrane was developed to enable the simulations of the complete microsystem.Gas tests evidenced a detection limit below 1ppm for carbon monoxide and below 100 ppm for methane.

Key Words: metal-oxide gas sensors； analog IC design；AHDL；CMOS compatible micromachining； microhotplates
1. Introduction
There is a strong interest in CMOS-based microsensors and, in particular, in micro-hotplate-based gas sensors, since miniaturization and the possibility of monolithic integration of transducer and circuitry offer significant advantages such as lowpower consumption, potentiall low costs, and the possibility of applying new dynamic sensor operation modes.
Tin dioxide (SnO2) is a widely used sensitive material for gas sensing in ambient air. Tin dioxide is heated to temperatures between 200 and 350℃ and changes its conductivity upon gas exposure.To achieve such high operation temperatures with the lowest possible power consumption, micro-hotplate struc- tures have been developed for metal oxide gas sensors during the past few years. A micro-hotplate consists of a thermally isolated area with a heater structure, a temperature sensor and contact electrodes for the sensitive layer. Many micro-hotplates were termed“CMOScompatible”,but only a few have been produced in standard CMOS processes.
Micro-hotplate-based gas sensors systems were implemented in a multi-chip approach.the micro-hotplates were placed on one chip, and the necessary driving and signal conditioning circuitry was hosted by a separate Application Specific Integrated Circuit (ASIC). It was established that integrating micro-hotplates and the corresponding circuitry on the same substrate is feasible without over-heating the electronics.Additional features of the monolithic solution include low power consump- tion, low noise, reliable electrical signals, high yield, and low production costs as compared to a multi- chip approach. The integration of multiple gas sensors in sensor arrays, the use of multi-component analysis algorithms, such as principal component regression (PCR) or artificial neural networks(ANN), and the application of new dynamic sensor operation modes [14, 15] help to overcome the problems associated with poor selectivity and drift of individual gas sensors [16].
2. System Description
2.1. Membrane
Figure 1 shows a micrograph of the microsensor system.The chip was fabricated using a standard doublepoly,double -metal, 0.8μmCMOS-process as provided by [17].
The chip hosts a micro-hotplate (close-up in Fig. 2), which is thermally isolated from the rest of the chip by placing it on a very thin membrane with low heat- conductivity. The membrane consists of the dielectric layers (silicon oxide/nitride) and features an additional N-well island (no connection to the bulk), which remains after applying a potassium-hydroxide (KOH) wet-etch with an electrochemical etch stop [18]. After wet etching, the SnO2 sensitive layer is deposited on the membrane as a nanocrys- talline thick film (crosssection in Fig. 3).
The membrane is 500 by 500 μm in size with a heated area (micro-hotplate) of 300 by 300 μm. The heated area exhibits a 6 μm-thick N-well silicon island underneath for homogeneous heat distribution and membrane stiffening. The membrane features a polysilicon ring-heater, which provides symmetric heat generation and dissipation. Four resistive polysilicon temperature sensors (Fig. 2) are integrated on the membrane to monitor the heat distribution. The polysilicon temperature sensor located in the center of the membrane provides input for the temperature control circuitry. The experimentally assessed maximum temperature variation across the membrane is 1–2% up to 350℃, and the membrane power efficiency is 4.8℃/mW.
2.2. Circuitry
A schematic of the chip is shown in Fig. 4 as a block diagram.The circuitry can be grouped in three functional units: (i) Membrane temperature control loop, (ii) bulk chip temperature measurement, and (iii) SnO2 resistance measurement. The chip is biased with an external current source (Hewelett Packard 3245A Universal Source), which is temperature-independent. This bias current is then copied and used for biasing the membrane temperature sensor.

Fig. 1. Micrograph of the integrated chemical sensor chip. Fig. 2. Close-up of the microhotplate.
F

Fig. 3. Cross-section of the microhotplate. Fig. 4. Block diagram of the monolithic sensor system.

Fig. 5. Proportional temperature controller.
The proportional temperature controller (Fig. 5) is implemented with an operational amplifier [20, 21] and an internal stabilization capacitor of 8 pF. The membrane temperature is controlled from room tem- perature up to 350℃. The operational amplifier drives a power transistor, which provides the current to the polysilicon heater (RHEATER). The inputs of the operational amplifier consist of the control voltage (VCONTROL) and the voltage drop on the polysilicon temperature sensor (RTEMP SENSOR), which provides the feedback signal for the temperature controller. The polysilicon temperature sensor is biased with a temperature-independent current (IBIAS). The aging effects in the polysilicon temperature sensor cannot be compensated by the electronics on chip. The dominant pole of the temperature control system is determined by the thermal time constant of the membrane, which is approximately 20 ms. An on-chip stabilization capacitor was included to cancel the low frequency pole of the operational amplifier.
The main advantage of the proportional temperature controller is that it occupies minimum silicon area. The most important trade-off in the design of the proportional controller is between its stability (phase margin) and the steady-state error. When the open-loop gain of the controller increases, the phase margin and the steady-state error decrease and vice versa. A good trade-off between stability (phase margin) and steadystate error was found to be an open-loop gain of 76 dB, with a phase margin of approximately 88◦.
The main source of error for the temperature controller is the steady-state error introduced by the offset voltage of the operational amplifier, which is independent of the open-loop gain of the system (i.e., the steady-state error introduced by the offset voltage of the operational amplifier does not decrease when the open loop gain increases). During calibration of the sensor, the offset voltage is compensated for at room temperature (i.e., the steady state error is compensated for at room temperature), but the temperature dependence of the offset voltage remains. The steady-state error measured over the sensor system operation temperature range (ambient temperature between −40 and 120℃) is less than 1% of the membrane temperature.
The bulk chip temperature is assessed via the baseemitter voltage difference between a pair of diodec- onnected vertical PNP transistors (parasitic transistors available in the CMOS process, collectors tied to substrate) working at different current densities (Fig. 6).
The resistance of the SnO2 sensitive layer can vary over a wide range (up to six decades) and is measured from 1 k_hm to 10 M_hm using a logarithmic converter(Fig. 7), which was implemented with a voltageto- current converter and a pair of diode-connected vertical PNP transistors. The dc level of the logarithmic converter can be changed with the reference current (IREF). The bulk chip temperature is used to compensate for the temperature dependence of the logarithmic converter.
The circuitry is located at 1 mm distance from the micro-hotplate owing to packaging design consid- erations. A partial epoxy cover shields the circuitry and the bond wires from the environment, but en- ables free access of the analyte gas to the gas-sensitive material on the micro-hotplate.

△ VBE = VBE1 - VBE2=VT .㏑(I40μA /I10μA)
△VBE= VBE1 - VBE2=VT .㏑(ISnO2/IREF)
Fig. 6. Bulk chip temperature measurement circuitry. Fig. 7. Logarithmic converter

Fig. 8. Membrane modeling using AHDL.

3. Simulation Results

The ADHL model was used for simulating the stability of the temperature controller. Figure 9 shows the Bode plots (magnitude and phase) of the output of the control system (voltage drop over the polysilicon temperature sensor). The phase margin is around 88◦. The dominant pole at 8Hz is resulting from the thermal time constant.
Transient simulations were done using a sinusoidal variation of the control voltage with an amplitude of 800mVaround a common mode of 1Vand a frequency of 0.1 Hz (Fig. 10). Such low frequency of the control voltage was selected in the simulations because this frequency is commonly used in the temper- ature modulation mode of hotplate-based gas sensors (speed of the chemical reaction at the interface). The simulations show, that the circuitry can drive the membrane temperature to 350℃ without distorting the shape given by the control voltage. When the control voltage is lower than the voltage drop across the polysilicon temperature sensor at ambient temperature (around 1 V at 27℃), the membrane temper- ature equals ambient temperature. The simulations also show, that the simulated membrane temperature is 350℃ for a control voltage of 1.8 V, which results in an error of 2% in comparison to the measured membrane temperature (Fig. 11).
4. Experimental Results
4.1. Electrical Measurements
The tracking mode performance of the temperature controller was measured and is shown in Fig. 11. The measurement was done at room temperature, and the control voltagewas increased in steps of 10mV.Acontrol voltage of 1.80 V produced, e.g., a membrane temperature of 343℃. The measured sensitivity was about 0.63℃/mV. The controller showed an excellent performance with a resolution of 0.5℃, which was assessed at the center temperature sensor of the membrane (temperature inhomogen- ities over the membrane were not taken into account).
The dynamic behavior of the temperature controller was assessed by subjecting the membrane to a temperature step from 27 to 300℃. The temperature controller was found to be very stable showing no overshoot or ringing.
The performance of the temperature controller in the stabilization modewas also assessed (Fig. 12). The ambient temperature was ramped from −40 to 120℃ in steps of 5℃. A control voltage of 1.77 V was applied, which produced a membrane temperature of 330.9℃.The steady-state error measured over the operating temperature range is less than 1% of the membrane temperature.
The performance of the temperature sensor on the bulk chip was also measured. The ambient temper- ature was swept from −40 to 120℃ in steps of 5℃, and the temperature controller was switched off. A twopoint calibration at −20 and 85℃ was performed. The measured sensitivity was about 128 μV/℃, and the resolution was 1.5℃.
4.2. Chemical Measurements
Gas test measurements were carried out with various concentrations of carbon monoxide (CO) and methane(CH4) in air. The CO concentrations ranged between 10–100 ppm (Fig. 14), and the CH4 concentrations between 750–10000 ppm (Fig. 15). The measurements were performed at relative humidities (r.h.) of 0 and 50%.
The sensor was operated between 250 and 350℃ in steps of 50℃. The gas flow rate was 0.2 liters per minute for all measurements. The exposure time of 30 min to the analyte gas is sufficient to achieve stable steady-state signals (response time approximately 30 s). A recovery of the sensor is reached with- in 15 min. The exposure time to pure carrier gas was set to 30 min to be on the safe side.The sensor resp- onses to CO and CH4 show different temperature dependence, which facilitates the quantitative detec- tion of these compounds in a mixture using multi-component analysis methods.
Detectability of less than 1 ppm CO and less than 100 ppm CH4 enables the use of this chip in a wide range of applications.
5. Conclusion and Outlook
A monolithic gas sensor system fabricated in standard 0.8μmCMOS process combined with post-CMOS micromachining was presented. It comprises metal-oxidecovered (SnO2) micro-hotplates and the necessary driving and signal-conditioning circuitry on the same chip.
The on-chip proportional temperature controller can accurately adjust the membrane temperature up to 350℃ with a resolution of 0.5℃. The steady-state error measured over the operating temperature range is less than 1% of the membrane temperature. The temperature controller is very stable and does not show any overshoot or ringing. The steady-state error in the operation temperature range can be further decreased if a proportional-integral-derivative (PID) temperature controller is used. The integral part will decrease the steady-state error, but it will also slowdown the time response of the controller. This problem can be addressed with the derivative part of the controller.
The modeling of the membrane in AHDL can be used for design and simulation of the temperature controller without the need of changing the simulation environment.
The logarithmic converter enables resistance measurements of the SnO2 sensitive layer between 1 k_hm and 10 M_ with an error of ±1%.
Gas tests yielded limits of detection below 1 ppm CO and below 100 ppm CH4. The problems associated with poor selectivity and drift of individual gas sensors will be overcome by monolithic implementation of gas sensor arrays, which are based on the indivi