A Frequency-Based Signature Gas Identification Circuit for SnO2 Gas ...
A Frequency-Based Signature Gas Identification Circuit for SnO2 Gas Sensors Kwan Ting Ng1,2 , Farid Boussaid1 , and Amine Bermak2 , 1
School of Electrical, Electronic and Computer Engineering The University of Western Australia, Perth, WA, Australia 2
Department of Electronic and Computer Engineering Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong [email protected]
Abstract—This paper presents a gas identification circuit for tin oxide (SnO2 ) gas sensors. The proposed circuit uses 2 gas sensors with different characteristics to achieve gas identification. A spike train is generated during operation, with the frequency of spike occurrence being gas dependent but concentration invariant. As a result, the spike firing frequency can be used to achieve gas identification. The calibration of this readout technique requires only a single exposure to the target gases to extract the sensor resistances. The low complexity processing is suitable for on-chip implementation. The functionality of this circuit has been validated with real data from our in-house fabricated sensors.
I. I NTRODUCTION The use of gas sensors to identify dangerous gases for safety purpose is inevitable. Besides, high efficiency, simple architecture and low costs gas identification systems are increasingly imminent. Tin oxide (SnO2 ) sensor is one of the most commonly used gas sensors due to its low fabrication cost, high sensitivity to a large variety of gases and CMOS compatibility [1]. This enables sensors and processing circuits to be integrated onto a single chip. However, there is a major drawback when using SnO2 as the sensing material for gas identification. The selectivity of the SnO2 sensors is poor because they are highly sensitive to a wide range of gases. Same output signals can be observed even if the sensed gases are different. This is because the sensor response does not depend only on the type of the sensed gas, but also depends on the gas concentration. Therefore, it is crucial for the gas identification system to be robust against the interference by the gas concentration. A common approach for gas identification systems is to employ an array of different gas sensors to compensate for the poor selectivity of SnO2 . The gas sensor array is usually constructed by SnO2 sensors with different additives such as catalysts and dopants. Classifiers such as principal component analysis (PCA) are applied on the extracted sensor information [2]. The number of sensors used in the array directly affects its identification accuracy. The discrimination between different gases is more observable by using a large number of sensors. Consequently, most of the existing gas identification systems employs large sensor arrays [2][3][4]. The downsides
978-1-4244-5309-2/10/$26.00 ©2010 IEEE
of large sensor arrays are the expensive and high complexity fabrication, which results in low yield. Another approach to improve the sensor selectivity is to modify the sensor characteristics by temperature. Since SnO2 sensor response has a large dependence on operating temperature, multiple signals can be extracted from a single sensor with this approach. An example is to perform gas sensing with dynamic temperature modulation, which is to apply a periodically varying voltage to the sensor heater [5]. As a result, thousands of signal responses can be obtained from a single sensor and can be used for gas identification. Nonetheless, the gas identification systems using dynamic temperature modulation is slow and unreliable due to the slow response of SnO2 . Sampling the sensor resistances at multiple constant temperatures can also be used to enhance the selectivity [6]. However, the time for SnO2 to reach thermal equilibrium is long hence it is not amenable to perform real time readout. Manufacturers recommend users to heat up a commercial sensor for at least 2 hours prior to sensing. This is not suitable for sensors used in gas identification systems which are always used for dangerous gas monitoring. It is vital to perform fast identification for safety reasons therefore temperature modulated sensors are inadequate. To address the tradeoff between sensor selectivity, identification efficiency and fabrication costs, we propose in this paper an on-chip gas identification circuit that requires only 2 different SnO2 gas sensors. The circuit generates a spike train whose frequency is gas dependent but concentration invariant. The advantage of the proposed approach is that it can perform real-time gas identification with a small dimension sensor array. This paper is organized as follows. Section II describes the working principle of the proposed gas identification technique. The VLSI implementation of the circuitry is presented in Section III. Experimental results are discussed in Section IV, and finally a conclusion is drawn in Section V. II. G AS IDENTIFICATION BASED ON FREQUENCY SIGNATURE
Tin oxide gas sensor is a conductometric gas sensor whose resistance changes due to adsorption and reaction of reduc-
2275
ing/oxidizing gas on the sensor surface. The resistance Ri for sensor i is approximated by [7] Ri ≈ αi ζ −γi
(1)
Where ζ is the sensed gas concentration, αi and γi are constants which depend on the type of gas and sensing material. The block diagram of the proposed technique is shown in Fig. 1. It uses 2 different SnO2 gas sensors with individual readout circuit to convert the sensor information into electrical signals in order to generate a concentration independent signature for the sensed gas.
III. VLSI I MPLEMENTATION The two major building blocks for the proposed circuit are the sensor readout circuits and spike generation circuit. The schematic of the readout circuit is shown in Fig. 2. The readout circuits for both sensors are identical, with sensor i being modeled as variable resistance Ri .
Fig. 2.
Schematic of the readout circuit
To readout the sensor resistances, we firstly convert the resistances into current then perform a log compression by diode. If the gain A is large enough, the current passing through the sensor is given by Fig. 1.
Block Diagram of the proposed technique
Vref (3) Ri This current is copied and fed into a forward biased diode. Its forward bias voltage Vxi is given by Ii =
Prior to sensing, in the calibration phase, the 2 sensors are exposed once to the target gases at a high concentration. The sensor resistances are stored in the on-chip memory as references. The gas concentration of this exposure should be as high as possible so that the reference signals are large. During the sensing period, the spike generation circuit returns a spike train with its period tperiod given by:
tperiod =
ln R1 (ζ) − ln R1 (ζH ) γ1 ≈ ln R2 (ζ) − ln R2 (ζH ) γ2
(2)
Where Ri (ζ) and Ri (ζH ) are the resistances for sensor i during sensing and calibration respectively. As seen from (2), ζ has been eliminated from the expression, resulting in the period of the spike train being independent of the gas concentration. Equivalently, the frequency of the spike train only depends on the gas and sensor dependent term, γ1 and γ2 , regardless of its concentration. The frequency of the spike train is chosen as the signature of the sensed gas and stored in the counter and memory. Therefore, gases can be identified simply by reading the frequency of the spike train from the counter and memory. One should note that ζH has also been removed from the expression of tperiod in (2). In other words, the reference concentration is not required to be precisely controlled. This further reduces the complexity of the calibration process.
Vref (4) IES · Ri Where VT and IES are the thermal voltage and reverse bias saturation current of the diode respectively. At the reference extraction stage, Vxi for both sensors are stored in the on-chip memory. They are used as input signals at sensing for the spike generation. The schematic of the the spike generation circuit is shown in fig. 3. Vxi = VT ln
Fig. 3.
Schematic of the spike generation circuit
During sensing, Vx1 and Vx2 become the readout voltages that correspond to the sensor resistances. They are to be
2276
compared with the reference signals Va1 and Va2 respectively. The difference between Vx1 and Va1 is amplified by a constant voltage gain Ao . By using (4), Vout is given by Vout = Ao · VT ln
RSen1 (ζ) ζH = Ao · VT · γ1 ln RSen1 (ζH ) ζ
(5)
Similarly, The difference between Vx2 and Va2 is amplified by a constant transconductance gain Gm, resulting an output current Icomp given by Icomp = Gm · VT ln
RSen2 (ζ) ζH = Gm · VT · γ2 ln RSen2 (ζH ) ζ
(6)
Initially, Vcomp is connected to ground through the global reset transistor with a high reset signal GR. To initiate the identification process, GR is toggled low, disconnecting Vcomp with ground and leave it as a floating node. Subsequently, Icomp charges the capacitor up linearly from ground with the rate given by Icomp dVcomp (7) = dt C When Vcomp reaches Vout , the output of the comparator toggles from low to high. After propagating through the inverter chain, AR becomes high, causing the capacitor to be discharged to ground. When Vcomp is discharged, the comparator output returns to low. AR thereafter returns to low which disconnects Vcomp from ground. The inverter chain is used to extend the pulse width of AR, ensuring that sufficient time is available for Vcomp to be fully discharged to ground. This resetting process iterates until the end of the integration period. The timing diagram is illustrated in fig. 4. GR
Vcomp Vout AR
tperiod
Fig. 4.
time
Timing diagram
A spike train is observed at the signal AR. The pulse width of the spike depends on the propagation delay of the inverter chain, whereas the period of the spike train is given by tperiod =
Vout · C Ao · C γ 1 = · Icomp Gm γ2
(8)
As seen from (8), the period of the the spike train is concentration independent. Equivalently, the frequency of the
spike train can be used to identify the gas. A spike driven counter and memory is attached to the output of the readout circuitry in order to record the spike firing frequency [8]. As the firing frequency is linearly proportional to the spike count when the integration period is fixed, the spike count stored in the counter and memory can be used to represent the sensed gas. IV. R ESULTS AND D ISCUSSION The circuit was designed using AMIS 0.35µm standard CMOS technology and the sensors were fabricated in an inhouse fabrication facility. The microphotograph of the sensors is shown in Fig. 5. The size of a sensor is approximately 190µm × 190µm. Platinum (Pt) is chosen as the catalyst for both sensors whereas boron (B) and phosphorus (P) are used as dopants. As a result, the drift behavior of the 2 sensors are similar. This enhances the robustness of the proposed gas identification circuit because the effect of drift has been reduced by the ratiometric sensing.
Fig. 5.
Microphotograph of sensors
The sensors were tested with gases such as propane, ethanol, carbon monoxide and hydrogen between concentration of 40ppm and 200ppm. The test was run in a laboratory chamber for 108 cycles in total by exposing sensors to gas and dry air alternatively. Dry air exposure was used to ensure that the sensor surface has been cleaned before the next measurement. The sensor steady state resistances were extracted from the sensors and used to verify the functionality of the proposed circuit. Sensor resistances sampled at 200ppm for all gases are assigned as the reference resistances Ri (ζH ) because it is the highest available concentration throughout the test. An example of the results is plotted in Fig. 6. 30µs were chosen as the integration period in this example to demonstrate the spikes clearly. A longer integration period is used in the actual sensing to obtain a larger signal and to increase the diversity between different gas signatures. 4 inverters with minimum sizes are used in the inverter chain. The pulsewidth of the spike is 6ns, which is sufficient enough to fully discharge Vcomp since a small capacitance of 1pF was chosen. The spike train is fed into the spike driven counter and memory. The spike counts for different gases are plotted in Fig. 7 with 200µs as the integration period. The results show clearly that each gas is represented by a spike count, which is independent on the gas concentration. The mean and the standard error of mean (SEM) are summarized in table I.
2277
TABLE I S UMMARY OF THE SPIKE COUNTS FOR DIFFERENT GASES
Mean SEM
Fig. 6. Demonstration result of the readout circuit. Integration time is 30µs
Propane 91 0.5
CO 106 0.7
Ethanol 78 0.9
Hydrogen 64 2.5
As seen from fig. 8, the spike counts for hydrogen is relatively constant at low concentration. However, significant drifts are observed at high concentrations. This is because the sensor resistances at high concentrations are very close to the resistances at the reference concentration ζH (200ppm). By (5) and (6), the output signals Vout and Icomp are small after subtraction. The spike count is proportional to the ratio between Vout and Icomp , hence misleading results are obtained. One way to resolve this is to extract the reference resistance Ri (ζH ) at a higher concentration. The output signals will be significantly enhanced and further improve the identification accuracy. V. C ONCLUSION A gas identification technique with simple calibration utilizing only 2 SnO2 gas sensors is presented in this paper. The proposed scheme extracts reference resistances by exposing the sensors once to high concentration gases prior to sensing. During operation, the sensor resistances are compared with the references and converted into a spike train that has a unique firing frequency for each monitored gas regardless of its concentration. The number of spikes within the integration period is stored in an on-chip memory and to be used as the signature of the sensed gas. The functionality of the circuitry has been validated using data obtained from an in-house fabricated sensors and has shown satisfactory result.
Fig. 7.
R EFERENCES
Counts of gases at different concentrations
The SEM for hydrogen is relatively high as shown in table I. It is because the spike count for the sample at concentration of 180ppm is 81, which is much higher than all other hydrogen samples. The measurement of hydrogen was repeated 3 times. The results are shown in fig. 8.
Fig. 8.
Results for identifying H2 in 3 repeated tests
[1] W. Gpel and K. D. Schierbaum, “Sno2 sensors: Current status and future prospects,” Sensors and Actuators B: Chemical, vol. 26, no. 1-3, pp. 1 – 12, 1995. [2] M. Shi, A. Bermak, S. Chandrasekaran, A. Amira, and S. BrahimBelhouari, “A committee machine gas identification system based on dynamically reconfigurable fpga,” Sensors Journal, IEEE, vol. 8, no. 4, pp. 403–414, April 2008. [3] S. Marco, A. Ortega, A. Pardo, and J. Samitier, “Gas identification with tin oxide sensor array and self-organizing maps: adaptive correction of sensor drifts,” Instrumentation and Measurement, IEEE Transactions on, vol. 47, no. 1, pp. 316–321, Feb 1998. [4] B. Guo, A. Bermak, P. Chan, and G.-Z. Yan, “Characterization of integrated tin oxide gas sensors with metal additives and ion implantations,” Sensors Journal, IEEE, vol. 8, no. 8, pp. 1397–1398, Aug. 2008. [5] A. Far, F. Flitti, B. Guo, and A. Bermak, “Gas identification system based on temperature modulation tin-oxide sensors and bio-inspired processing,” in Electronics, Circuits and Systems, 2008. ICECS 2008. 15th IEEE International Conference on, 31 2008-Sept. 3 2008, pp. 1010–1013. [6] J. Zakrzewski, W. Domanski, P. Chaitas, and T. Laopoulos, “Improving sensitivity and selectivity of sno2 gas sensors by temperature variation,” Instrumentation and Measurement, IEEE Transactions on, vol. 55, no. 1, pp. 14–20, Feb. 2006. [7] N. Yamazoe and K. Shimanoe, “Theory of power laws for semiconductor gas sensors,” Sensors and Actuators B: Chemical, vol. 128, no. 2, pp. 566 – 573, 2008. [8] K. T. Ng, C. Shoushun, F. Boussaid, and A. Bermak, “Compact gray-code counter/memory circuits for spiking pixels,” in Electronic Design, Test and Applications, 2008. DELTA 2008. 4th IEEE International Symposium on, Jan. 2008, pp. 506–511.
2278
School of Electrical, Electronic and Computer Engineering The University of Western Australia, Perth, WA, Australia 2
Department of Electronic and Computer Engineering Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong [email protected]
Abstract—This paper presents a gas identification circuit for tin oxide (SnO2 ) gas sensors. The proposed circuit uses 2 gas sensors with different characteristics to achieve gas identification. A spike train is generated during operation, with the frequency of spike occurrence being gas dependent but concentration invariant. As a result, the spike firing frequency can be used to achieve gas identification. The calibration of this readout technique requires only a single exposure to the target gases to extract the sensor resistances. The low complexity processing is suitable for on-chip implementation. The functionality of this circuit has been validated with real data from our in-house fabricated sensors.
I. I NTRODUCTION The use of gas sensors to identify dangerous gases for safety purpose is inevitable. Besides, high efficiency, simple architecture and low costs gas identification systems are increasingly imminent. Tin oxide (SnO2 ) sensor is one of the most commonly used gas sensors due to its low fabrication cost, high sensitivity to a large variety of gases and CMOS compatibility [1]. This enables sensors and processing circuits to be integrated onto a single chip. However, there is a major drawback when using SnO2 as the sensing material for gas identification. The selectivity of the SnO2 sensors is poor because they are highly sensitive to a wide range of gases. Same output signals can be observed even if the sensed gases are different. This is because the sensor response does not depend only on the type of the sensed gas, but also depends on the gas concentration. Therefore, it is crucial for the gas identification system to be robust against the interference by the gas concentration. A common approach for gas identification systems is to employ an array of different gas sensors to compensate for the poor selectivity of SnO2 . The gas sensor array is usually constructed by SnO2 sensors with different additives such as catalysts and dopants. Classifiers such as principal component analysis (PCA) are applied on the extracted sensor information [2]. The number of sensors used in the array directly affects its identification accuracy. The discrimination between different gases is more observable by using a large number of sensors. Consequently, most of the existing gas identification systems employs large sensor arrays [2][3][4]. The downsides
978-1-4244-5309-2/10/$26.00 ©2010 IEEE
of large sensor arrays are the expensive and high complexity fabrication, which results in low yield. Another approach to improve the sensor selectivity is to modify the sensor characteristics by temperature. Since SnO2 sensor response has a large dependence on operating temperature, multiple signals can be extracted from a single sensor with this approach. An example is to perform gas sensing with dynamic temperature modulation, which is to apply a periodically varying voltage to the sensor heater [5]. As a result, thousands of signal responses can be obtained from a single sensor and can be used for gas identification. Nonetheless, the gas identification systems using dynamic temperature modulation is slow and unreliable due to the slow response of SnO2 . Sampling the sensor resistances at multiple constant temperatures can also be used to enhance the selectivity [6]. However, the time for SnO2 to reach thermal equilibrium is long hence it is not amenable to perform real time readout. Manufacturers recommend users to heat up a commercial sensor for at least 2 hours prior to sensing. This is not suitable for sensors used in gas identification systems which are always used for dangerous gas monitoring. It is vital to perform fast identification for safety reasons therefore temperature modulated sensors are inadequate. To address the tradeoff between sensor selectivity, identification efficiency and fabrication costs, we propose in this paper an on-chip gas identification circuit that requires only 2 different SnO2 gas sensors. The circuit generates a spike train whose frequency is gas dependent but concentration invariant. The advantage of the proposed approach is that it can perform real-time gas identification with a small dimension sensor array. This paper is organized as follows. Section II describes the working principle of the proposed gas identification technique. The VLSI implementation of the circuitry is presented in Section III. Experimental results are discussed in Section IV, and finally a conclusion is drawn in Section V. II. G AS IDENTIFICATION BASED ON FREQUENCY SIGNATURE
Tin oxide gas sensor is a conductometric gas sensor whose resistance changes due to adsorption and reaction of reduc-
2275
ing/oxidizing gas on the sensor surface. The resistance Ri for sensor i is approximated by [7] Ri ≈ αi ζ −γi
(1)
Where ζ is the sensed gas concentration, αi and γi are constants which depend on the type of gas and sensing material. The block diagram of the proposed technique is shown in Fig. 1. It uses 2 different SnO2 gas sensors with individual readout circuit to convert the sensor information into electrical signals in order to generate a concentration independent signature for the sensed gas.
III. VLSI I MPLEMENTATION The two major building blocks for the proposed circuit are the sensor readout circuits and spike generation circuit. The schematic of the readout circuit is shown in Fig. 2. The readout circuits for both sensors are identical, with sensor i being modeled as variable resistance Ri .
Fig. 2.
Schematic of the readout circuit
To readout the sensor resistances, we firstly convert the resistances into current then perform a log compression by diode. If the gain A is large enough, the current passing through the sensor is given by Fig. 1.
Block Diagram of the proposed technique
Vref (3) Ri This current is copied and fed into a forward biased diode. Its forward bias voltage Vxi is given by Ii =
Prior to sensing, in the calibration phase, the 2 sensors are exposed once to the target gases at a high concentration. The sensor resistances are stored in the on-chip memory as references. The gas concentration of this exposure should be as high as possible so that the reference signals are large. During the sensing period, the spike generation circuit returns a spike train with its period tperiod given by:
tperiod =
ln R1 (ζ) − ln R1 (ζH ) γ1 ≈ ln R2 (ζ) − ln R2 (ζH ) γ2
(2)
Where Ri (ζ) and Ri (ζH ) are the resistances for sensor i during sensing and calibration respectively. As seen from (2), ζ has been eliminated from the expression, resulting in the period of the spike train being independent of the gas concentration. Equivalently, the frequency of the spike train only depends on the gas and sensor dependent term, γ1 and γ2 , regardless of its concentration. The frequency of the spike train is chosen as the signature of the sensed gas and stored in the counter and memory. Therefore, gases can be identified simply by reading the frequency of the spike train from the counter and memory. One should note that ζH has also been removed from the expression of tperiod in (2). In other words, the reference concentration is not required to be precisely controlled. This further reduces the complexity of the calibration process.
Vref (4) IES · Ri Where VT and IES are the thermal voltage and reverse bias saturation current of the diode respectively. At the reference extraction stage, Vxi for both sensors are stored in the on-chip memory. They are used as input signals at sensing for the spike generation. The schematic of the the spike generation circuit is shown in fig. 3. Vxi = VT ln
Fig. 3.
Schematic of the spike generation circuit
During sensing, Vx1 and Vx2 become the readout voltages that correspond to the sensor resistances. They are to be
2276
compared with the reference signals Va1 and Va2 respectively. The difference between Vx1 and Va1 is amplified by a constant voltage gain Ao . By using (4), Vout is given by Vout = Ao · VT ln
RSen1 (ζ) ζH = Ao · VT · γ1 ln RSen1 (ζH ) ζ
(5)
Similarly, The difference between Vx2 and Va2 is amplified by a constant transconductance gain Gm, resulting an output current Icomp given by Icomp = Gm · VT ln
RSen2 (ζ) ζH = Gm · VT · γ2 ln RSen2 (ζH ) ζ
(6)
Initially, Vcomp is connected to ground through the global reset transistor with a high reset signal GR. To initiate the identification process, GR is toggled low, disconnecting Vcomp with ground and leave it as a floating node. Subsequently, Icomp charges the capacitor up linearly from ground with the rate given by Icomp dVcomp (7) = dt C When Vcomp reaches Vout , the output of the comparator toggles from low to high. After propagating through the inverter chain, AR becomes high, causing the capacitor to be discharged to ground. When Vcomp is discharged, the comparator output returns to low. AR thereafter returns to low which disconnects Vcomp from ground. The inverter chain is used to extend the pulse width of AR, ensuring that sufficient time is available for Vcomp to be fully discharged to ground. This resetting process iterates until the end of the integration period. The timing diagram is illustrated in fig. 4. GR
Vcomp Vout AR
tperiod
Fig. 4.
time
Timing diagram
A spike train is observed at the signal AR. The pulse width of the spike depends on the propagation delay of the inverter chain, whereas the period of the spike train is given by tperiod =
Vout · C Ao · C γ 1 = · Icomp Gm γ2
(8)
As seen from (8), the period of the the spike train is concentration independent. Equivalently, the frequency of the
spike train can be used to identify the gas. A spike driven counter and memory is attached to the output of the readout circuitry in order to record the spike firing frequency [8]. As the firing frequency is linearly proportional to the spike count when the integration period is fixed, the spike count stored in the counter and memory can be used to represent the sensed gas. IV. R ESULTS AND D ISCUSSION The circuit was designed using AMIS 0.35µm standard CMOS technology and the sensors were fabricated in an inhouse fabrication facility. The microphotograph of the sensors is shown in Fig. 5. The size of a sensor is approximately 190µm × 190µm. Platinum (Pt) is chosen as the catalyst for both sensors whereas boron (B) and phosphorus (P) are used as dopants. As a result, the drift behavior of the 2 sensors are similar. This enhances the robustness of the proposed gas identification circuit because the effect of drift has been reduced by the ratiometric sensing.
Fig. 5.
Microphotograph of sensors
The sensors were tested with gases such as propane, ethanol, carbon monoxide and hydrogen between concentration of 40ppm and 200ppm. The test was run in a laboratory chamber for 108 cycles in total by exposing sensors to gas and dry air alternatively. Dry air exposure was used to ensure that the sensor surface has been cleaned before the next measurement. The sensor steady state resistances were extracted from the sensors and used to verify the functionality of the proposed circuit. Sensor resistances sampled at 200ppm for all gases are assigned as the reference resistances Ri (ζH ) because it is the highest available concentration throughout the test. An example of the results is plotted in Fig. 6. 30µs were chosen as the integration period in this example to demonstrate the spikes clearly. A longer integration period is used in the actual sensing to obtain a larger signal and to increase the diversity between different gas signatures. 4 inverters with minimum sizes are used in the inverter chain. The pulsewidth of the spike is 6ns, which is sufficient enough to fully discharge Vcomp since a small capacitance of 1pF was chosen. The spike train is fed into the spike driven counter and memory. The spike counts for different gases are plotted in Fig. 7 with 200µs as the integration period. The results show clearly that each gas is represented by a spike count, which is independent on the gas concentration. The mean and the standard error of mean (SEM) are summarized in table I.
2277
TABLE I S UMMARY OF THE SPIKE COUNTS FOR DIFFERENT GASES
Mean SEM
Fig. 6. Demonstration result of the readout circuit. Integration time is 30µs
Propane 91 0.5
CO 106 0.7
Ethanol 78 0.9
Hydrogen 64 2.5
As seen from fig. 8, the spike counts for hydrogen is relatively constant at low concentration. However, significant drifts are observed at high concentrations. This is because the sensor resistances at high concentrations are very close to the resistances at the reference concentration ζH (200ppm). By (5) and (6), the output signals Vout and Icomp are small after subtraction. The spike count is proportional to the ratio between Vout and Icomp , hence misleading results are obtained. One way to resolve this is to extract the reference resistance Ri (ζH ) at a higher concentration. The output signals will be significantly enhanced and further improve the identification accuracy. V. C ONCLUSION A gas identification technique with simple calibration utilizing only 2 SnO2 gas sensors is presented in this paper. The proposed scheme extracts reference resistances by exposing the sensors once to high concentration gases prior to sensing. During operation, the sensor resistances are compared with the references and converted into a spike train that has a unique firing frequency for each monitored gas regardless of its concentration. The number of spikes within the integration period is stored in an on-chip memory and to be used as the signature of the sensed gas. The functionality of the circuitry has been validated using data obtained from an in-house fabricated sensors and has shown satisfactory result.
Fig. 7.
R EFERENCES
Counts of gases at different concentrations
The SEM for hydrogen is relatively high as shown in table I. It is because the spike count for the sample at concentration of 180ppm is 81, which is much higher than all other hydrogen samples. The measurement of hydrogen was repeated 3 times. The results are shown in fig. 8.
Fig. 8.
Results for identifying H2 in 3 repeated tests
[1] W. Gpel and K. D. Schierbaum, “Sno2 sensors: Current status and future prospects,” Sensors and Actuators B: Chemical, vol. 26, no. 1-3, pp. 1 – 12, 1995. [2] M. Shi, A. Bermak, S. Chandrasekaran, A. Amira, and S. BrahimBelhouari, “A committee machine gas identification system based on dynamically reconfigurable fpga,” Sensors Journal, IEEE, vol. 8, no. 4, pp. 403–414, April 2008. [3] S. Marco, A. Ortega, A. Pardo, and J. Samitier, “Gas identification with tin oxide sensor array and self-organizing maps: adaptive correction of sensor drifts,” Instrumentation and Measurement, IEEE Transactions on, vol. 47, no. 1, pp. 316–321, Feb 1998. [4] B. Guo, A. Bermak, P. Chan, and G.-Z. Yan, “Characterization of integrated tin oxide gas sensors with metal additives and ion implantations,” Sensors Journal, IEEE, vol. 8, no. 8, pp. 1397–1398, Aug. 2008. [5] A. Far, F. Flitti, B. Guo, and A. Bermak, “Gas identification system based on temperature modulation tin-oxide sensors and bio-inspired processing,” in Electronics, Circuits and Systems, 2008. ICECS 2008. 15th IEEE International Conference on, 31 2008-Sept. 3 2008, pp. 1010–1013. [6] J. Zakrzewski, W. Domanski, P. Chaitas, and T. Laopoulos, “Improving sensitivity and selectivity of sno2 gas sensors by temperature variation,” Instrumentation and Measurement, IEEE Transactions on, vol. 55, no. 1, pp. 14–20, Feb. 2006. [7] N. Yamazoe and K. Shimanoe, “Theory of power laws for semiconductor gas sensors,” Sensors and Actuators B: Chemical, vol. 128, no. 2, pp. 566 – 573, 2008. [8] K. T. Ng, C. Shoushun, F. Boussaid, and A. Bermak, “Compact gray-code counter/memory circuits for spiking pixels,” in Electronic Design, Test and Applications, 2008. DELTA 2008. 4th IEEE International Symposium on, Jan. 2008, pp. 506–511.
2278
