An Integrated Microstructure with Temperature Control for Gas Sensors G. C. Cardinali, L. Dori, M. Fiorini, P. Maccagnani,
Abstract | This
paper
describes
an
integrated
Contacts
mi-
crostructure used to control temperature in a gas sensor system. An innovative technological approach has been pursued to integrate gas sensors on silicon substrate, compatible with IC fabrication.
and V. Liberali
A mixed analog-digital
electronic interface controls temperature operation, to maximize the sensitivity to substances to be detected.
Sensing Layer Passivation Layer
Temperature Sensor
Silicon Nitride Membrane
Heater
I. Introduction
Microintegrated gas sensor systems are becoming more and more important in electronic market, as their fabrication is feasible using conventional integration technologies and costs are reduced. In integrated gas sensors an electrical parameter of the sensing layer changes as the results of chemical reactions between the substances to be detected and the sensor itself. To optimize their response, usually integrated sensors are operated at a controlled temperature obtained through a heating element. To reduce power consumption, the sensing layer can be deposited on a thin membrane realized on a silicon substrate using the bulk-micromachining technique. This paper describes an integrated microstructure suitable for use as heating element. Innovative technological steps in fabrication have been developed and optimized, to improve sensor reliability over a very large number of temperature cycles. The electronic interface for temperature control has been designed in a conventional 1.2 m CMOS technology. II. Design of the Microstructure
Fig. 1 shows the microstructure of the gas sensor. The sensing lm is placed on top of a layer stack constituted (in order from bottom) by the membrane, the heater, the passivation layer and the metal contacts. A thin lm applied as mechanical support is an important feature in microsensor technologies. Essential membrane requirements can be summarized as follows: 1. mechanical stability; 2. low stress material; 3. low thermal conductivity; 4. high chemical inertness under the KOH etchant (the same lm must work either as etch-stop allowing the membrane to form or act as masking G. C. Cardinali, L. Dori, M. Fiorini, and P. Maccagnani are with CNR { LAMEL Institute, Via Gobetti 101, 40129 Bologna, Italy, E-mail:
[email protected] V. Liberali is with the Department of Electronics, University of Pavia, Via Ferrata 1, 27100 Pavia, Italy, E-mail:
[email protected]
Silicon Substrate
Fig. 1. Integrated microstructure
layer during the silicon bulk micromachining process); 5. lm deposition process and material compatible with IC technology. Many solutions have been proposed to realize thin lm membranes [1]{[5]. They use insulators like SiO2, Si3N4 , SiC either as a single layer or as a sandwich of dielectric layers. To ful ll the requirements mentioned above without increasing fabrication process complexity, we used silicon nitride (Si3N4 ) realized with a modi ed low pressure chemical vapor deposition (LPCVD) process, to obtain a low stress membrane lm. To optimize the mechanical properties of the micromachined membranes, their dimensions and fabrication processes have been considered with care. A number of membranes, with sizes ranging from 1 2 1 to 2 2 2 mm2, were realized varying the parameters of the silicon nitride deposition process: temperature and gas ratio (diclorosilane/ammonia). A test was performed to investigate the mechanical stability of the membrane by applying a dierential pressure. The membrane de ects in a blister-like shape and de ection is a function of membrane size and intrinsic lm stress. Test results show that an increase in membrane area or in the gas ratio leads to an increase in de ection. On the other hand, the deposition temperature does not aect the de ection magnitude. Experimental characterization of fabricated devices shows that a 1:5 2 1:5 mm2 membrane, 250-nm thick, breaks at a maximum overpressure of 100 kPa, while it de ects about 35 m when a pressure of 40 kPa above the atmospheric pressure is applied.
TABLE I Insulation Resistance of Passivation Layers
S
LIDI
SO SOLIDIS IDIS
IS
LID
IS
Platinum
LID
Nitride
S
SOLIDI
LID
IS
IS LID
Fig. 2. Geometry of the double spiral shaped heater Temp [C] 555 495 435 375 325 265 205 145 85
S
LIDI
SOL SO SOLIDIS
25
IDIS
SO SOLIDIS
IS
LID
SOO S SOLIDIS
SOLIDI
S
LID
IS
IS
LID
IS
IS
LID
LID
O SO S SOLIDIS
To set and measure the operating temperature of the gas sensitive layers, a heating resistor and a temperature sensor must be integrated on the micromachined membranes. Both components can be fabricated by selectively etching the same conductive lm deposited on top of the membrane surface. Sputtered platinum is the most widely used and characterized material for this type of application. It exhibits high stability in temperature cycles and a quite high temperature coecient which ensures intrinsic temperature uniformity of the heater and provides a simple way to obtain a temperature measurement by monitoring resistance variations. The temperature sensor could be the heater itself, but design characteristics of the temperature control electronics suggest separate realization of these components would be more suitable. Indeed, at the operating temperature T = 400 C, the optimum resistance value for the heater is 100 , while the value for the temperature sensor is around 1 k . Following the scheme reported in Fig. 1, a passivation layer is needed to electrically insulate the heater from the sensing lm. The resistance of the passivation layer must be high enough to ensure that leakage currents are not aecting the measurement of the sensible lm conductance at working temperature (150 C T 430 C) and at the applied bias (few volts). The dielectric must also have long term stability, to avoid lifetime reduction of the device due to Sn or Pt diusion across the layer stack. Table I shows the resistance values of passivation layers with dierent thickness. A 1500 nm SOG + LTO insulation layer has been chosen for this application, because it has very good insulation properties and it is less expensive than titanium nitride (TiN). As shown in Fig. 1, the heater is placed at the center of the membrane, beneath the gas sensitive layer. The region where the sensing layer conductance is measured by means of an interdigitized shaped contact, is commonly referred to as the active area. The ratio between the edge of the square membrane and the edge of the active area is a key gure of the device design. The greater the ratio, the smaller the losses due to conduction heat
Silicon
LID IS
Low Temperature Oxide Spin-On Glass
SOL SO SOLIDIS
resistance few k
few M
few k
G
G
G
O SO S SOLIDIS
a LTO: b SOG:
thickness 100 nm 210 nm 300 nm 235 nm 1500 nm 1750 nm
SOO S SOLIDIS
Passivation layer Si3N4 Si3N4 LTOa Si3N4 + TiN SOGb + LTOa SOGb + LTOa + TiN
Fig. 3. Simulation of temperature distribution in the double spiral shaped heater
transfer. Simulation results, carried out by means of the three-dimensional nite-element analysis program SOLIDIS [6], and experimental observations show that a reasonable trade-o between membrane size and conduction losses is obtained when this ratio ranges from 2 to 3. An active area with a 700 m side results from fabrication constraints. Starting from this active area and considering a 1500 m membrane side, the heating power needed to raise the temperature of the active area to T = 400 C is calculated at about 100 mW. To de ne heater geometry, the temperature distribution inside the active area was considered. An average gradient of 0:4 C/m was obtained with the double spiral shaped resistor shown in Fig. 2. The spiral layout has a variable pitch, to account for the distribution of heat losses inside the device active area. Fig. 3 shows the simulated temperature distribution, obtained with SOLIDIS. A minimum gradient of about 0:2 C/m has been obtained. III. Temperature Control
The main target in the design of control circuitry was achieving the best trade-o between performance and costs. To this end, the electronic interface was realized
VDD Rm
VREF1 (1.1 V) temp. cycle
R1
M_OUT
VREF2 (1.2 V) Rref(400)
A
R2
Rh
Rref(150)
Fig. 5. Scheme of the \on/o" temperature control loop
Fig. 4. Waveforms for timing and measurement control
using a 1.2 m CMOS technology and it was made as
exible as possible to easily conform to dierent sensor characteristics. Control circuitry can be easily integrated with signal processing interface [7]. The strong dependence of sensor behavior on the operating temperature requires a high degree of exibility, to optimize system performance for dierent sensors. The control logic must allow the timing and duty cycles of heating steps to be changed, as well as the number and timing of measurements. The timing section employs a 32 kHz quartz oscillator to generate a master clock signal with maximum stability, which is divided by 216 to obtain a 0.5 s clock. Binary numbers corresponding to the duration of each temperature cycle and of measurement cycles are stored to control the programmability. Since the time interval of each cycle will not exceed 128 s, an 8-bit counter is suitable for timing. The counter value is compared with the stored data to drive toggle ip- ops. Fig. 4 shows the clock at 0.5 s (1), together with signals for temperature control (2) and measurement acquisition (3) with the programmed duty-cycle. To operate in temperature pulsed mode, temperature stabilization is needed during heating cycles. As explained before, the heating system is constituted by two sputtered platinum resistors: the rst is the heating resistor Rh , while the second is the measuring resistor Rm . Their nominal resistance values are 100 and 1000 , respectively. The operating temperatures are 400 C and 150 C. To reach 400 C during the heating cycle, at least 100 mW must be supplied to the heating resistor, that means that a current of 33 mA must be supplied by the control system. To furnish the heating current, a special output stage is added to a conventional single stage mirrored ampli er. The output stage consists of an open-drain PMOS transistor with W=L = 1000. The heater is connected to the drain to optimize the voltage swing. Temperature is controlled through an \on/o"
switching circuit. The basic idea is to have a reference resistor Rref which has the same resistance as that expected for the measuring resistor at a given temperature. Hence Rm is compared with the reference resistor and the power is switched on (o) if Rm is smaller (higher) than Rref . Fig. 5 shows the circuit implementing this control scheme. The temperature cycle is selected by switching between two reference resistors, according to the control logic signal. Let us suppose that the system has to perform the cycle at T = 400 C. Since the temperature is lower, the value of the sensing resistance Rm is smaller than its reference value, Rref (400). The output of the comparator assumes the high state, and power is switched onto the heater. When the temperature exceeds 400 C, the value of Rm becomes greater than the reference temperature and the comparator switches o the heating element. In this control scheme the output PMOS transistor behaves like a switch. A quite large output dynamic range is achieved, and the supplying power capability is considerably increased. The temperature accuracy of the proposed control circuit depends on the matching between the reference and the measuring resistors. To account for deviations of real parameters from design values, the two reference resistors are discrete components (e. g. two multi-twist trimmers) external to the integrated circuit. The system can be set to cycle between two given temperatures simply by adjusting the values of these resistors. As computer simulations con rmed, a temperature control accuracy of about 2 C is achievable if a tight thermal link exists between the heater and the temperature sensor. This condition is satis ed in the device considered, due to the small size of sensor microstructures. IV. Experimental Results
To evaluate structure performance under dierent operating conditions and to optimize the whole system, the basic components have been fabricated and characterized. The blister test was used to establish the maximum pressure that the Si3N4 membrane can sustain without
600 500
∆ TH
400
[°C] 300 200 100 0
0
50
100
150
P [mW] H
Fig. 7. Temperature increase in active area versus heating power Fig. 6. Voltage applied to the heating resistor
breaking. This is useful in estimating the long-term membrane stability under operating conditions. The stress was applied by ramping a dierential pressure up to one quarter of the breaking pressure, at a frequency of 100 cycle/min. The number of applied cycles was about 200,000 (equivalent to 1 year in pulsed temperature mode with cycles of 150 s). To evaluate the magnitude of the stress applied, we point out that the membrane, when heated at T = 450 C, de ects 1 4 2 m while during the dierential pressure test it was de ected by tens of microns. All the tested devices exhibited excellent mechanical stability and none broke during the tests, proving the long-term reliability of the membrane. Further mechanical and electrical investigations were carried out on the complete structure (sensing layer included). In particular, an accelerated test was performed where the device was stressed pulsing the temperature between 200 C and 500 C each 2 s for 10 days. The results of the accelerated test have con rmed both the excellent mechanical stability of the whole system and the considerable endurance of the passivation layer. The thermal behavior of the sensor microstructure has been thoroughly investigated by means of electrical and thermographic measurements. Fig. 6 shows the voltage applied to the heating resistor for the \on-o" temperature control. Heating power depends on the duty-cycle, and it has been veri ed to be independent on the voltage supply. Fig. 7 shows the active area temperature as a function of the heating power. A total power of about 100 mW is required to heat the active area at 450 C and 35 ms is the time interval needed to switch from 200 C to 500 C. The thermal variation of the platinum resistance has been investigated by electrical measurements performed on the real device and on test structures, maintained at a uniform and well-controlled temperature (60:1 C). A positive temperature coecient
= 2:71 6 0:06 C01 has been evaluated. V. Conclusion
This paper presented an integrated microstructure for gas sensor. Experimental characterization activities allowed us to de ne and, in some cases, to optimize the dierent technological steps involved in the realization of the mechanical support and of the temperature control with a programmable electronic interface. Experimental measurements con rmed that the designed microstructure and temperature control are suitable for a variety of gas sensors operated in pulsed temperature mode. Acknowledgment
This work has been supported by the European Union under ESPRIT Project 7500 { MEPI (Demonstrator D{114). References
[1] M. Sekimoto, H. Yoshihara, and T. Ohkubo, \Silicon nitride single-layer X-ray mask," J. Vac. Sci. Technology, vol. 21, pp. 1017{1021, 1982 [2] E. I. Bromley, J. N. Randall, D. C. Flanders, and R. W. Mountain, \A technique for the determination of stress in thin lms," J. Vac. Sci. Technology, vol. B{1, pp. 1364{1366, 1983 [3] M. G. Allen, M. Meheregany, R. T. Howe, and S. D. Senturia, \Microfabricated structures for the in situ measurement of residual stress, Young's modulus, and ultimate strain of thin lms," Appl. Phys. Letters, vol. 51, pp. 241{ 243, 1987 [4] I. George, P. Carmeli, B. Bonvalot, M. Wagener, A. Girard, A. Zarudiansky, and J. Suski, \Thin membranes with optimized thermomechanical properties for microsystem applications," in Proc. EUROSENSOR VIII Conf., Toulouse, France, 1994, vol. 2, pp. 38{42 [5] S. Moller, J. Lin, and E. Obermeier, \Material and design considerations for low power microheater modules for gas sensor applications," in Proc. Fifth Int. Meeting on Chemical Sensors, Rome, Italy, 1994, vol. 1, pp. 433{436 [6] J. G. Korvink, The SOLIDIS User Manual: Version 1.0, PEL-ETH Zurich, 1995 [7] V. Liberali, F. Maloberti, and D. Tonietto, \Sigma-delta processing in multisensor systems for carbon monoxide detection," in Proc. IEEE Int. Symp. on Circ. and Syst., Atlanta, GA, USA, 1996, vol. IV, pp. 376{379