IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 42, NO.

7, JULY 2007 1543

A 141-dB Dynamic Range CMOS Gas-Sensor

Interface Circuit Without Calibration With
16-Bit Digital Output Word
Marco Grassi, Member, IEEE, Piero Malcovati, Senior Member, IEEE, and Andrea Baschirotto, Senior Member, IEEE

Abstract—In this paper, we present the design and the charac-

terization of a wide-dynamic-range interface circuit for resistive
gas-sensors able to operate without calibration. The circuit is
based on resistance-to-frequency conversion, which guarantees
low complexity. The state-of-the-art of this measurement method
has been improved first by separating the Resistance value Con-
trolled Oscillator circuit (RCO) from the sensing device, thus
leading to higher linearity performance, and then by exploiting a
novel digital frequency measurement system. Measurement results
on a silicon prototype, designed in a 0.35- m CMOS technology, Fig. 1. Typical gas-sensing system block diagram.
show that the circuit achieves, without calibration, a precision in
resistance measurement of 0.4% over a range of 4 decades and
better than 0.8% over 5 decades (dynamic range, DR 141 dB).=
Furthermore, after calibration, it reaches a precision of 0.4% for development is a pattern recognition system [1]. Of course,
resistance values ranging between 1 k and 1 G , thus leading improving a gas-sensing system requires the optimization of
to a DR of 168 dB. The prototype chip consumes less than 15 mW each sub-module, having at the end the advantage of deciding
from a 3.3-V supply.
whether to integrate all the parts in a single chip or to keep the
Index Terms—Gas-sensor interface circuit, mixed analog-digital gas-sensor array separated, in order to optimize the system with
IC, resistance-to-frequency conversion. respect to the operating temperature condition [3].
Furthermore, the exploitation of innovative algorithms on
I. INTRODUCTION which the latest versions of the feature extraction circuits are
based [7], may lead to significant power saving, since they
ANDHELD and wireless gas-sensing systems are getting
H more and more popular especially due to worldwide
legislative initiatives promoted in order to keep air pollution
are able to extract information also from the dynamic sensor
response, thus requiring a shorter array query time for achieving
the same performance [8], [9]. Moreover, by using specific
under control and to avoid human exposure to dangerous
patterns to modulate the temperature of the sensor, a further
environments. For these reasons, the development of integrated
efficiency improvement is possible [10]. This latter operation
interface circuits being mandatory for embedding gas-sensors
mode exploits the additional information given by sensors that
in such portable devices, manufacturers are now dropping the
are operated at different temperatures, taking into consideration
employment of expensive ad hoc instruments or multi-chip
also temperature transients and edge profile.
board solutions, while also academic institutions and govern-
This work is outlined as follows. Section I is a short in-
ment-funded consortia are working in the research field in
troduction to portable sensing devices. The development of
order to investigate the microsystem or micromodule approach
integrated solutions for gas-sensor interfacing is called manda-
[1], [2], [3], [4]. This is also justified by the fact that last
tory for such instruments. The feasibility is then justified by
micromachined resistive gas-sensors can be fabricated with
advantages given by recent improvements in research and
embedded integrated low-power heater/thermometer and small
development of basic building blocks. Section II describes
enough for reaching the operating temperature in few tens of
the interface circuit requirements. Front-end circuit design
milliseconds [5], [6]. A gas-sensing microsystem consists of
constrains, referring once again to Fig. 1, are given by sensor
four main parts, as shown in Fig. 1: an array of gas-sensors, a
array electrical behavior and technology process on the left
temperature controller circuit (heater), an electronic read-out
side, while on the right side the interface circuit must meet
block, and a data processor, that actually in the project under
pattern recognizer input signal specifications. Section III is an
overview of the developed integrated interface circuit, while
Manuscript received November 20, 2006; revised March 14, 2007. This work Section IV reports the major design details of the interface cir-
was carried out within the Italian Government PRIN Project 2005092937.
M. Grassi and P. Malcovati are with the Department of Electrical Engi- cuit. Here the performance of the most significant sub-circuits
neering, University of Pavia, 27100 Pavia, Italy (e-mail: [email protected], is underlined. Section V reports the experimental results for the
[email protected]). developed silicon prototype, as well as an example of chemical
A. Baschirotto is with the Department of Innovation Engineering, University
of Salento, 73100 Lecce, Italy (e-mail: [email protected]). measurement. Finally, Section VI draws a few conclusion about
Digital Object Identifier 10.1109/JSSC.2007.899087 the overall interface design goals and results achieved.
0018-9200/$25.00 © 2007 IEEE
1544 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 42, NO. 7, JULY 2007

TABLE I
FRONT-END CIRCUIT REQUIRED SPECIFICATIONS

chemical sensor employed in Section V for chemical measure-

ments1 is reported in Fig. 3 as an example.

A. Interface Circuit Requirements

The latest developed dynamic pattern recognition techniques
Fig. 2. Gas-sensing system channel. for gas-sensing, being able to extract important information not
only from the static value of the resistance of the sensor, but also
from its transient behavior, through a morphological descriptor
of the sensor trajectory in a phase space [1], [8], [11], [12],
require an electronic interface circuit sufficiently fast to pre-
serve the transient data, e.g., able to process all the information
coming from the sensor array with a throughput close to 10 Hz.
For most chemical targets, in order to detect every kind of gas
of interest with enough accuracy, usually in the order of tens of
parts per milion (ppm) after pattern analysis for CO, NO , CH
or % Vol. for alcohols, it is important to measure the sensor re-
sistance value, hereinafter , with a precision better than 1%
Fig. 3. Cross-section scheme of the employed Si-micromachined tin-oxide
sensors (CNR-IMM, Lecce-IT). (signal-to-noise-and-distortion ratio, SNDR 40 dB) [1], [10],
[13]. Furthermore, considering the state of the art in sensor man-
ufacturing [5], [6], the sensor resistance value may vary across
II. FRONT-END CIRCUIT SPECIFICATIONS several decades, being the effect of three variable components:
AND STATE OF THE ART the baseline , which typically ranges, for undoped
tin-oxide devices, from a moderated value, like 100 k , to a
It is important, in order to define the specifications required much higher one, like 100 M ; the deviation from the baseline
for the electronic read-out circuit in terms of performance, to , due to technology spread and sensor age as well as tem-
keep in mind the sensor response behavior and the input data perature and finally the resistance variation , due to gas
format needed by the digital pattern analyzer, especially in terms concentration, negative for oxidizing gas types, which may be as
of rate and resolution for every gas-sensing system equivalent large as a couple of decades from baseline [1], [10], [13]. Con-
channel, actually consisting of a given sensor from the array sidering the three reported resistance value range contributions
with associated heater and the interface circuit with embedded together with the required precision, a dynamic range (DR) of
A/D converter [1], [10], as shown in Fig. 2. The developed about 140 dB is required [13]. The interface circuit has thus been
read-out chip has been designed in order to match chemoresis- designed in order to satisfy the above reported requirements,
tive sensing devices requirements. which are briefly reviewed in Table I. Calibration operation or
They are actually sensors based on metal-oxide thin films additional operating modes may be exploited in order to raise
(MOX), usually realized with the sol-gel technique and the dynamic range performance even over 160 dB, paying in
deposited by spin coating on silicon micromachined sub- terms of additional complexity, or in order to distribute more
strates, equipped, as mentioned, with an integrated platinum efficiently the required DR between SNDR and resistance mea-
heater-thermometer and interdigitated electrodes. surement range.
Since these types of gas-sensors are based on direct analyte
adsorption and charge transfer processes between the gas B. Possible Interface Circuit Implementations
molecules and the MOX surface, causing a simple electrical
The interface circuit has to deliver, without significant dis-
resistance variation of the gas-sensing element, they are char-
tortion, the information coming from the sensor to the digital
acterized by an effective resistive electrical behavior [6]. This
pattern recognition block. A good trade-off solution in terms of
particular relationship between gas concentration and resis-
accuracy and response speed over a dynamic range of sensor re-
tance value for MOX devices has been the main reason for their
sistance values up to more than 160 dB [14] would be the use
employment in the development of the presented gas-sensing
system. The cross-section of the tin-oxide micromachined 1Courtesy of Research Center CNR-IMM, Lecce-IT.
GRASSI et al.: A 141-dB DYNAMIC RANGE CMOS GAS-SENSOR INTERFACE CIRCUIT WITHOUT CALIBRATION WITH 16-Bit DIGITAL OUTPUT WORD 1545

of smart auto-ranging architectures, e.g., composed by a pro-

grammable transresistance amplifier that converts the current
flowing through a constant voltage-biased sensor into a voltage,
followed by an oversampled A/D converter. In this wide-range
solution reported in [15], an auto-ranging circuit for scale se-
lection was included together with a DSP to perform coarse and
fine calibration in order to cancel any inter-range offset and gain
errors. This kind of approach, inspired to high-precision labora-
tory instruments, is often mandatory to meet extremely severe
Fig. 4. Possible first enhancement for the resistance-controlled oscillator
gas detection specifications, which require an accuracy [16], (RCO) design: sensor decoupling from oscillator circuit.
[17] up to 0.1% for sensor resistance read-out and, as a conse-
quence, a calibration-free circuit cannot be taken into consider-
ation. Although, taking into account novel portable gas-sensing
specification requirements overviewed in previous subsection,
with particular emphasis on indoor environment and, as un-
derlined, on power saving and system complexity reduction,
uncalibrated solutions are more promising in this application
context, even if it is not straightforward to reach the required
Fig. 5. Block diagram of the developed interface circuit.
dynamic range. Therefore, our research is oriented to a sim-
pler solution, where extremely high dynamic range ( 160 dB)
is not demanded [10]. We recall that a wide-range interface
circuit without calibration may be, for example, realized per- The sensor resistance value may be extracted by measuring
forming a compression of the sensor current value . Actu- the period of the output bitstream by means of a frequency
ally, can be compressed exploiting the differential voltage meter. The maximum accuracy reached by this solution [20] is
drop across two matched diodes [18]. Simulation results of such 0.5% for sensor resistance in the 500 k –1 G range, but the
a circuit are satisfactory for many indoor environmental targets, read-out channel still needs to be reconfigured controlling the
since the achieved precision is near to 1% over an uncalibrated voltage reference value and reprogramming the value of
range of 4 decades DR 115 dB , but higher dynamic range (actually needs to be selected from a capacitor array in
interface circuits must be developed in order to completely fulfil the range 0.5–62.5 pF with 2-pF step), thus leading to a circuit
the specifications of the novel portable gas-sensing systems de- complexity comparable to the smart interface circuit mentioned
scribed above. Another approach without calibration available in [15] and [16] based on calibrated multi-scale approach.
in the literature exploits oscillator circuits for resistance to fre- Therefore, we developed a low-complexity, high dynamic
quency conversion [19]. This technique allows the measure- range resistance to number converter, designed for gas-sensing
ment of a larger resistance range than the direct linear con- applications, which does not require calibration [13], [21]
version, since it is not limited by voltage swing constraints in by introducing several improvements in the Resistance value
the used analog circuits. Besides, particular care in this kind of Controlled Oscillator circuit (RCO) of Fig. 4. The circuit is
front-end circuit design must be taken in order to cancel the ef- based on an oscillator approach, designed in 0.35- m CMOS
fect of the parasitic capacitance of the sensor layer. An addi- technology with a power supply of 3.3 V, consuming 15 mW at
tional drawback of this technique is that the sensor is exposed maximum sensor current. The block diagram of the developed
to the large voltage swing of the resistance-controlled oscillator, circuit, consisting of a V-to-I converter, an integrator circuit
thus leading to a linearity performance again of the order of 1% with input current polarity inverter, a Schmitt trigger, and a
over 4 decades exploiting a single scale DR 115 dB . Also digital unit, is shown in Fig. 5. The whole circuit in more
this approach, like the one exploiting range compression, should detail and every sub-block will be described in the following
be improved in terms of precision and resistance value range to subsections.
meet the requirements, e.g., introducing calibration operation or
multi-range approach. A. Overview of the Interface Circuit
As reported in the schematic of Fig. 6, the resistive sensor
III. THE DEVELOPED OSCILLATOR-BASED is biased with an accurate reference voltage ap-
FRONT-END CIRCUIT plied exploiting a low-noise, high-gain (100 dB), low output re-
A possible solution to the problems that limit the performance sistance amplifier in V-to-I configuration. A couple of high-lin-
of oscillator-based sensor read-out circuits, described in the pre- earity cascoded current mirrors alternately push or pull the cur-
vious section, is to bias the sensor with a voltage that is constant rent into or from the virtual ground of a reset-
and, as a consequence, independent from the oscillator circuit, table Miller integrator, which has also the aim of keeping con-
using a voltage buffer followed by a V-to-I converter, as shown stant the voltage drop across the output current mirrors while
in Fig. 4. The sensor current drives a push-pull CMOS current varies, thus raising the linearity of the response ( is actually
mirror that feeds a capacitor , which behaves as an integrator, the scaling ratio of the current mirrors and the reference voltage
followed by a Schmitt trigger that actually bounds the voltage has been set to the nominal value V after simulation
swing across . level system evaluation).
1546 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 42, NO. 7, JULY 2007

Fig. 6. Detailed schematic of the developed interface circuit based on enhanced RCO.

The two-stage operational amplifier (op-amp) employed in to achieve the desired accuracy, i.e., 256 for a minimum
the integrator exhibits a DC gain of 60 dB and a unity-gain band- guaranteed 8-bit equivalent precision, as implemented in the re-
width of 280 MHz. By comparison of the output of the inte- ported interface system. As a consequence, if we need to mea-
grator with two constant boundary voltages and ( sure resistance values in the range of at least 2 decades cen-
V, set by design once more on the basis of tered around , over a single read-out scale, the minimum
system level simulations), by means of two high-speed con- number of bits for each of the two counter registers is given by
tinuous-time comparators 2 ns , the control sig-
nals for the switches that shunt the mirrored current are gener- (3)
ated. An additional flip-flop finally guarantees that comparators
switch alternately and feeds a counter, whose transition period More in detail, in the implemented circuit, an extra number of
is consequently directly proportional to the resistance value internal bits is employed for the oscillator-dependent counter to
: allow resistance measurement 3 decades below . Then, if
the applied value is lower than midrange, the slower of
(1)
the two counters reaching will be the reference one and the
measurement time is constant. By contrast, the slower counter
B. The Novel Frequency Measurement Technique will be the one controlled by the oscillator if the measured
is larger than midrange, leading to a measurement time propor-
The frequency measurement is carried out in the digital do-
tional to the resistance value itself. Therefore, the output digital
main using a novel principle. The implemented technique al-
word will be composed of 15 bits, which represent the value
lows to handle a dynamic range which is two times, in terms of
reached by the faster of the two counters during the measure-
decades, with respect to previously published frequency mea-
ment and by an additional bit which indicates if the converted
surement methods [20] [21]. This circuit works as follows. The
resistance value is either higher or lower than .
ratio between a reference counter, whose clock frequency is
set at the midrange of the given oscillator frequency interval
C. Available Front-End Circuit Resistance Value Measurement
and a counter, whose clock frequency
Ranges
is the output signal of the resistance dependent oscillator, repre-
sents the digitized value of . In order to extend the dynamic range of the system, three dif-
Therefore, the measured resistance value is given by ferent current scaling ratios ( 1:200, 1:20, 1:2) are avail-
able at the input of the Miller integrator by means of a pro-
(2) grammable set of current mirrors, allowing a full-range mea-
surement in terms of resistance value varying from 1 k to 1 G
where and indicate the reference and the oscillator by exploiting different overlapped scales. Programmability in
frequency-dependent counter values at the end of the conver- the integrator capacitance has also been introduced in order to
sion, respectively, is the logarithmical center of the scale, improve accuracy and conversion speed for the highest resis-
while and represent the front-end circuit gain and offset tance values, choosing a value for among two: pF or
contribution terms. These coefficients may be uncompensated pF. The nominal value for the reference external clock
in gas-sensing because they do not affect linearity in gas con- is 2.5 kHz, thus leading to a constant conversion time
centration extraction. The measurement ends when the slower for of 100 ms for every - combination, while
of the two counters reaches a given value which is sufficient for higher resistance values the time needed for conversion is,
GRASSI et al.: A 141-dB DYNAMIC RANGE CMOS GAS-SENSOR INTERFACE CIRCUIT WITHOUT CALIBRATION WITH 16-Bit DIGITAL OUTPUT WORD 1547

Fig. 7. Op-amp employed for V sensor bias voltage buffering.

TABLE II
SUMMARY OF AVAILABLE FRONT-END CIRCUIT SUB-RANGES WITH stage Miller-compensated amplifier, whose topology is shown
f =
2.5 kHz. FOR GENERIC f : R ,R , AND R in Fig. 7. Low output resistance at the buffered value node
MUST BE NORMALIZED TO 2.5 kHz/f
is guaranteed by a nMOS (800/0.35) source follower which has
also the function of V-to-I conveyor and by the feedback, as
previously shown in Fig. 6.
Having low output resistance is in fact mandatory to meet
linearity specifications, in particular when the sensor draws
the highest current values. The developed buffer exhibits a
measured maximum distortion of only 0.3% in the 6-decade
for example, setting to 1:200 and pF, microsec- output current extended range of interest [1 nA–1 mA]. This
onds, if is expressed in Ohms, as reported in Table II. feature means a worst case SNDR performance of 50 dB
Notice that, as appears in Table II, not all sub-ranges over 6 decades of current values, i.e., 170 dB dynamic range
are symmetrical with respect to . In fact [15]. The highest relative measured output resistance value
these bounds, differently from the mentioned ge- 0.003 occurs when the sensor draws the
ometrical bounds, take also into account further physical limita- maximum current 1 mA 3 1k .
tions and allow over-ranging capabilities of the interface circuit. Thus, this boundary condition exhibits the maximum measured
The lower limit is typically given by the fact that the maximum distortion.
oscillator frequency is 2.5 MHz The employed amplifier achieves a DC-gain of 100 dB and
because of the comparator speed limitations and finite inte- a unity-gain bandwidth of 100 kHz, while consuming 2 mW.
grator op-amp bandwidth, while the maximum current to be Transistors reported in Fig. 7 are all 0.35- m CMOS standard
measured should be 1 mA k , due to the devices and dimensioned in order to minimize flicker noise
buffer linearity specifications, in order to fulfil gas-sensing without degrading the gain [15]. Therefore, the residual total
requirements. For example, in range (later called also mode) RMS noise across the biased sensor is always below 80 V
“01” the lower limit constraints are coincident since we have for every value of current drawn by the sensor, even in
2.5 MHz when 1 mA. The upper limit fast-acquisition mode (100 Hz), described in the last section.
in resistance measurable value comes instead when the cur-
rent lost towards the bulk from the input analog pad is no B. The Current Mirrors
longer negligible for the accuracy of interest: this happens The cascoded current mirrors of Fig. 6, embedded in the de-
for 1 G , which means in terms of sensor current, veloped interface circuit and operated in push-pull configura-
1 nA. tion in order to feed with alternate current the following Miller
integrator, have been designed to grant a maximum linearity
IV. INTERFACE CIRCUIT DESIGN AND OPERATING DETAILS error of 0.25% over 5 decades of current for all the three pos-
Additional details on the operating modes and on the perfor- sible programmable configurations . The state-of-the-art oscil-
mance of each block of the developed circuit will be briefly re- lator-based approach employing current mirrors [20] has been
ported in the following subsections. improved by using cascode configuration and by keeping con-
stant the voltage of at least the output of current mirrors by con-
A. The Voltage Reference Buffer in V-To-I Configuration necting such net to Miller integrator op-amp virtual ground.
The amplifier used for applying the bias voltage to the Current mirrors may also operate properly in the subthreshold
sensor is a low-noise pMOS input pair, high-gain (100 dB), two- region for currents down to 1 nA 1 G , with
1548 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 42, NO. 7, JULY 2007

Fig. 8. Schematic of the two-stage op-amp embedded in the integrator block.

a precision always better than 0.5% exploiting suitable scaling

ratios .

C. The Resettable Miller Integrator

The integrator block in Fig. 6 consists of a two-stage compen-
sated amplifier with a selectable feedback capacitor, that may be
chosen among two possible values: pF or pF,
while an additional disable/reset switch is provided. As men-
tioned in the circuit overview, the employed op-amp exhibits
a gain of 60 dB, with a phase margin larger than 70 and a
gain-bandwidth product of 280 MHz for pF. This
value guarantees an op-amp settling time fast enough to fulfil
linearity specifications even when the oscillator is driven at the
maximum possible frequency 2.5 MHz, when sensing Fig. 9. Schematic of the continuous-time comparator embedded in the
minimum sensor resistances. The output current driving capa- enhanced Schmitt trigger.
bility is 700 A, while the intrinsic slew rate performance is
240 V s (70 V s for pF). The quiescent power con-
sumption is about 3.7 mW, while the peak power consumption the SR latch will inhibit it and the system will wait for the other
is 6 mW. comparator response, while the Miller integrator input current is
The schematic of the operational amplifier used in the inte- inverted. By contrast, when the comparator connected to lower
grator is reported in Fig. 8, where detailed transistor geomet- reference voltage 1.15 V toggles, it is inhibited and the
rical ratios are also shown. An additional buffer, which may first comparator is enabled again, while the Miller integrator
be enabled by test controls, with the aim of driving an analog current is again inverted, leading to a triangular waveform at
output pad for externally probing of the integrator output tri- the integrator output. The complete comparator switching time
angular waveform, has been embedded in the interface circuit is, for an output frequency above 100 kHz, only 2 ns, including
chip. Its topology is identical to the one shown in Fig. 8 with an latch regeneration. For lower integrator output frequencies, the
additional output stage. switching time is longer, accordingly with more relaxed timing
requirements.
D. The Enhanced Schmitt Trigger Circuit The comparator includes a positive feedback loop, as shown
The integrator, as shown in Fig. 6, is followed by a discrim- in Fig. 9, in order to rise the unbalancing speed, while the output
inator/bounder circuit which has the aim of limiting the os- signal is extracted by means of suitable current mirrors and
cillator output swing and controlling the switches that shunt squared by a digital inverter. By properly dimensioning the pos-
the integrator input mirrored current. This circuit consists of itive feedback loop in the active load of the input pair of the in-
two high-speed continuous time comparators, and an additional tegrator, it would be also possible to introduce hysteresis in the
NAND Set-Reset (SR) asynchronous latch, not reported in Fig. 9 threshold voltage of the comparators [22], [23]. In this particular
for simplicity. design, switching threshold value is not an issue, thanks again to
This SR latch also has the function of forcing the comparators the fact that comparators are forced to switch alternately. Each
to switch alternately. For instance, once the comparator con- comparator is characterized by an average power consumption
nected to the reference voltage 2.15 V toggles its state, of 2.5 mW.
GRASSI et al.: A 141-dB DYNAMIC RANGE CMOS GAS-SENSOR INTERFACE CIRCUIT WITHOUT CALIBRATION WITH 16-Bit DIGITAL OUTPUT WORD 1549

Fig. 10. Simplified block diagram of the implemented frequency measurement circuit.

Fig. 11. Microphotograph of the read-out circuit silicon prototype.

E. The Embedded Digital Frequency Measurement and Each equivalent sample is obtained, for each sub-
Control Logic Circuit decade of a given scale, by means of an external 0.02% or
The frequency measurement logic developed, whose simpli- 0.2% precision CerMet resistor, called , connected at
fied block diagram is shown in Fig. 10, consists of two counters: one end to the buffered interface circuit pin and at
the clock of the first one is generated by the oscillator circuit and the other end to a precision voltage source . Sweeping
thus has a period proportional to the value of the under the voltage source output in the range [ 9 V; 0 V],
measurement, while the second counter is driven by an external we obtain an equivalent resistance value range of one decade:
reference clock. The measurement starts on the rising edge of [ ; ], while for extending the measurement
an external signal ENABLE, which initially resets both counter over more decades the CerMet resistor must be changed.
registers and unlocks integrator output. The measurement ends High-precision resistors of 10 k , 100 k , 1 M , 10 M ,
when the slower of the two counters reaches a defined value, i.e., 100 M , and 1 G values have been employed. The data
256, in the developed interface circuit. At this point, both acquisition is performed in an automated way: the chip pro-
counters are inhibited and a multiplexer updates the front-end totype, mounted on the board, is characterized by means of a
circuit digital output bus with the content of the faster counter developed Labview test environment, which executes about a
register, actually corresponding to the least significant 15 bits thousand measurements sweeping the equivalent value of
of the measured value, of which at least 8 are meaningful. by controlling the voltage source through a GPIB serial bus.
The most significant bit is instead the flag which indicates which Furthermore, the routine acquires the digitized values with a
has been the faster counter. Finally, an End Of Conversion signal logical analyzer, and generates an output data file to be analyzed
(EOC) is provided with an additional delay with respect to the in Matlab. For each scale, values higher than midrange
output data in order to facilitate external data acquisition. are stepped linearly, while values lower than
are stepped hyperbolically. We could say that, for higher
than midrange, the input resistance is simply stepped linearly,
V. DEVELOPED INTERFACE CIRCUIT CHARACTERIZATION while for lower than midrange it is the input conduc-
The proposed circuit, whose chip microphotograph (1.05 tance to be stepped linearly. This particular
0.35 mm is shown in Fig. 11, consumes less than 15 mW from shrewdness has been necessary due to the intrinsic behavior
a 3.3-V supply. of the novel frequency measurement technique, for which the
To characterize the interface circuit, mounted on a customized output digital code always results to be linear with respect to
test board, we applied at its terminals all needed bias voltages the faster among the two counters in the measurement. Before
and supplies, generated by commercial regulators, as shown in proceeding with the accurate linearity test of the developed
Fig. 12. Furthermore, emulated values for in the ranges front-end circuit, the following scope plots are reported in
and/or sub-ranges of interest have been connected to the chip Fig. 13 and Fig. 14, as an example of a qualitative evaluation
sensing input in order to process the digital converted values of the interface circuit behavior over a 5-decade range, actually
and retrieve the linearity performance of the read-out circuit. the full “10” range [10 k –1 G ]. These plots represent the
For each scale of interest, about 1 kSample of different syn- analog and digital output test signals (the triangular buffered
thesized resistor values are converted. integrator output waveform and the asynchronous latch output)
1550 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 42, NO. 7, JULY 2007

Fig. 12. Simplified schematic of the testing measurement setup.

Fig. 13. Integrator buffered output test signal and regenerated latch output test Fig. 14. Integrator buffered output test signal and regenerated latch output test
signal in mode “10” for R = 1 G , with one second scope persistence signal in mode “10” for R = 10 k , without scope persistence (f 
(f  26 Hz), in order to underline noise contribution. 2.6 MHz).

for the highest and lowest resistance value of applied , exhibited an accuracy up to 0.8% over 5 decades in mode “01”
respectively. DR 141 dB , as shown in Fig. 16 and 0.4% over 4 decades
In Fig. 13, one second scope persistence was used, in order to in mode “00” DR 128 dB , as reported in Fig. 17, by setting
underline the noise contribution in time domain. Furthermore, the reference clock frequency to 2.5 kHz. We recall that
Fig. 15 represents the worst-case noise persistence histogram the conversion time for the two lower decades will result con-
plot (5000 acquisitions) of the front-end circuit digital output in stant and equal to 100 ms, while for the upper decades the time
mode “11” for applied resistance value G . needed will range in [100 ms–10 s] for every available scale.
Since in the plot the local quantization is equiva- Furthermore, an ad hoc, faster operating mode for this circuit
lent to LSB 391 k , the absolute error in resis- is available if the user is interested in a range of only
tance measurement for G , due to noise, is 2 decades, very useful in gas-sensor query once the baseline
ERR LSB 1.5 M , thus leading to a system value has been compensated by selecting the reference clock
worst-case noise sensitivity of 0.15%, taking into account that frequency to get , where is the sensor
maximum input resistances lead to maximum noise. resistance in the absence of gas, after choosing the best working
range.
A. Linearity Characterization Results Without Calibration For example, if the sensor resistance in absence of an oxi-
Linearity characterization was performed following the pro- dizing gas is M , setting the external clock to
cedure described above, and in particular, the front-end circuit 25 kHz, the conversion time will be constant and equal
GRASSI et al.: A 141-dB DYNAMIC RANGE CMOS GAS-SENSOR INTERFACE CIRCUIT WITHOUT CALIBRATION WITH 16-Bit DIGITAL OUTPUT WORD 1551

Fig. 15. Noise persistence histogram plot (5000 acquisitions) of the front-end Fig. 17. Relative error in resistance measurement over [1 k –5 M ] “00”
circuit digital output in mode “11” for applied resistance value R = 1G . without performing any calibration operation (3.8 decades).

Fig. 16. Relative error in resistance measurement over [1 k –100 M ] “01”

without performing any calibration operation (5 decades). Fig. 18. FFT plot in fast conversion mode applying at the interface circuit input
6
a conductance value of 50 S with an amplitude of 25 S and a frequency of
3.2 Hz.

to 10 ms (100 Hz) for resistance values lower than 1 M in mode

“10”, leading to a fast-measurement range [10 k –1 M ]. We B. Linearity Characterization Results With “Soft” and
recall that the lower limit is given by the maximum rec- “Hard” Optional Calibration Operations
ommended oscillator frequency 2.5 MHz, while the
maximum current to be measured should be 1 mA If we decide to employ calibration, there are two more modes
k , in order to respect linearity specifications. In particular, of operation for the presented interface circuit. The first, called
in the example, the limitation comes from the frequency. The Soft Calibration, allows us to measure resistance values ranging
upper limit G in resistance measurable value from 1 k to 1 G with a 0.8% accuracy, leading to an equiva-
comes instead, from the input analog pad finite resistance to- lent dynamic range of 163 dB. In this case, the different scales,
wards bulk. that are mostly overlapped, should be linked by correcting the
The performed fast mode conversion test exhibits a SNDR of different gain error contributions by adding reconfigurability to
53 dB and a spurious-free dynamic range (SFDR) of 52 dB, that the value, actually trimming in (2). The latter, called
lead to a worst-case resolution of 8.5 bits. Fig. 18 shows the fast Hard Calibration, requires also to split the range of the measure-
Fourier transform (FFT) plot for the interface circuit operated in ment in decades and, for each of these, to perform gain error cor-
the described fast mode condition. The throughput obtained is rection to improve the linearity response over the whole range,
100 Hz, while the input is an equivalent sinusoidal conductance always adjusting the value for each decade interval. By
value centred at 50 S with an amplitude of this approach, the measurement accuracy obtained is 0.4% over
25 S and a frequency of 3.2 Hz. 6 decades, leading to a DR equal to 168 dB.
1552 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 42, NO. 7, JULY 2007

Fig. 19. SNDR and DR in resistance measurement over [1 k –1 G ] per- Fig. 21. System transient response to NO in 30% R.H. wet air.
forming soft calibration.

Fig. 20. SNDR and DR in resistance measurement over [1 k –1 G ] per- Fig. 22. System transient response to methanol in 30% R.H. wet air.
forming hard calibration.

is oxidizing and thus leading to negative resistance variation)

The best trade-off in terms of complexity and dynamic range, from 4.2% to 12.5%. The coatings used in these particular mea-
verified in measurements on the silicon prototype seems to be surements were kept at a temperature of 400 C by means of an
the soft calibration approach, whose measurement results are re- open-loop temperature control.
ported in Fig. 19. Hard calibration results are plotted in Fig. 20.
VI. CONCLUSION
C. Example of Chemical Measurements A low-complexity 141–168-dB dynamic-range CMOS
From the chemical point of view, the gas-sensing system per- gas-sensor interface circuit with digital output obtained with
formance has been evaluated for different chemical compounds: a novel resistance-to-frequency measurement technique has
in particular, CO, NO , CH and Methanol, either in dry or in been developed and characterized. From the measured per-
wet air, while for the sensors, undoped tin-oxide coatings were formance achieved from the silicon prototype, summarized in
used. Table III, the front-end circuit fulfils, without calibration, all the
As an example, Fig. 21 shows the system transient response requirements in terms of accuracy for modern environmental
varying NO concentration in 30% relative humidity (R.H.) air monitoring applications, while, performing a soft calibration
from 10 ppm to 30 ppm. Please notice that for this kind of gas on the sensor bias voltage value or on the reference frequency,
the sensor resistance value variation is positive and the order the circuit becomes suitable also for newer available dynamic
of magnitude of in presence of gas is close to 1 G . Fi- pattern recognition algorithms, thanks to increased output data
nally, in Fig. 22, we report the system response to methanol, rate. Moreover, a larger dynamic range can be achieved by
applying different volumetric concentrations of this gas (which exploiting partial sub-range overlap. Chemical measurement
GRASSI et al.: A 141-dB DYNAMIC RANGE CMOS GAS-SENSOR INTERFACE CIRCUIT WITHOUT CALIBRATION WITH 16-Bit DIGITAL OUTPUT WORD 1553

TABLE III
MEASURED INTERFACE CIRCUIT SPECIFICATION SUMMARY

results have been reported to validate the real effectiveness of [15] M. Grassi, P. Malcovati, and A. Baschirotto, “A 160 dB equivalent
the system. dynamic range auto-scaling interface for resistive gas-sensors arrays,”
IEEE J. Solid State Circuits, vol. 42, pp. 518–528, 2007.
[16] M. Grassi, P. Malcovati, and A. Baschirotto, “A 0.1% accuracy 100
ACKNOWLEDGMENT Ohm-20 MOhm dynamic range integrated gas-sensor interface circuit
Chemical measurements were carried out thanks to Dr. with 13 + 4 bit digital output,” in Proc. ESSCIRC, 2005, pp. 351–354.
[17] P. Malcovati, M. Grassi, F. Borghetti, V. Ferragina, and A. Baschirotto,
P. Siciliano’s group at CNR-IMM, Lecce, Italy. “Design and characterization of a 5-decade range integrated resistive
gas-sensor interface with 13-bit A/D converter,” in Proc. IEEE Sensors,
REFERENCES 2005, pp. 472–475.
[1] “Portable system for ambient gas monitoring with smart A/D [18] D. Barrettino, M. Graf, W. H. Song, K. Kirstein, and A. Hierlemann,
“Hotplate-based monolithic CMOS microsystems for gas detection
front-end improving sensors resolution and accuracy,” PRIN
and material characterization for operating temperatures up to 500 C,”
2003091427, funded by Italian Government [Online]. Available:
IEEE J. Solid-State Circuits, vol. 39, no. 7, pp. 1202–1207, Jul. 2004.
http://ims.unipv.it/prin03 [19] A. Flammini, D. Marioli, and A. Taroni, “A low-cost interface to high
[2] “General olfaction and sensing on a European level,” “GOSPEL” value resistive sensors varying over a wide range,” in Proc. IEEE IMTC,
Network, funded by European Community [Online]. Available: 2003, pp. 726–731.
http://www.gospel-network.org [20] M. Malfatti, D. Stoppa, A. Simoni, L. Lorenzetti, A. Adami, and A.
[3] A. Baschirotto and P. Malcovati, “Technology-driven alternatives for Baschirotto, “A CMOS interface for a gas-sensor array with a 0.5%-
smart sensor interfaces,” in Sensors Update, H. Baltes, G. Fedder, and
J. Korvink, Eds. Weinhein, Germany: Wiley-VCH, 2003, vol. 13, pp.
6
linearity over 500 k -to-1 G range and 2.5 C temperature control
accuracy,” in IEEE ISSCC Dig. Tech. Papers, 2006, pp. 294–295.
45–81. [21] M. Grassi, P. Malcovati, and A. Baschirotto, “Wide-range integrated
[4] D. Barrettino, M. Graf, S. Hafizovic, S. Taschini, C. Hagleitner, A. Hi- gas-sensor interface based on a resistance-to-number converter tech-
erlemann, and H. Baltes, “A single-chip CMOS micro-hotplate array nique with the oscillator decoupled from the input device,” in Proc.
for hazardous-gas detection and material characterization,” in IEEE IEEE ISCAS, 2006, pp. 4395–4398.
ISSCC Dig. Tech. Papers, 2004, pp. 312–313. [22] V. Ferragina, P. Malcovati, F. Borghetti, A. Rossini, F. Ferrari, N. Ratti,
[5] G. Sberveglieri, “Recent developments in semiconducting thin-film and G. Bertuccio, “Implementation of a novel read-out strategy based
gas-sensors,” Sensors and Actuators B, Chemical, vol. 23, no. 2–3, pp. on a wilkinson ADC for a 16 x 16 pixel X-ray detector array,” in Proc.
103–109, 1995. IEEE ISCAS, 2005, pp. 5569–5572.
[6] S. Capone and P. Siciliano, “Gas-sensors from nanostructured metal [23] P. Bastia, G. Bertuccio, F. Borghetti, S. Caccia, V. Ferragina, F. Ferrari,
oxides,” Encyclopedia of Nanoscience and Nanotechnology, vol. 3, pp. D. Maiocchi, P. Malcovati, D. Martin, A. Pullia, and N. Ratti, “An in-
769–804, 2004. tegrated reset/pulse pile-up rejection circuit for pixel readout ASIC’s,”
[7] R. M. Dowdeswell and P. A. Payne, “Odour measurement using con- in IEEE Nuclear Science Symp. Conf. Rec. (NSS ’04), 2004, vol. 3, pp.
ducting polymer gas-sensors and an artificial neural network decision 1415–1417.
system,” IEEE J. Eng. Sci. Ed., vol. 8, no. 3, pp. 129–134, 1999.
[8] E. Martinelli, G. Pennazza, C. Falconi, A. D’Amico, and C. Di Na-
tale, “Comparison between two alternative feature extraction methods Marco Grassi (S’04–M’06) graduated in electronic
for chemical sensor array,” in World Scientific AISEM Proc., 2003, pp. engineering from the University of Pavia, Italy, in
334–339. 2002, and pursued the Ph.D. degree in electronic
[9] A. Burresi, A. Fort, S. Rocchi, M. B. Serrano Santos, N. Ulivieri, and engineering at the same university in 2006. In 2001,
V. Vignoli, “Temperature profile investigation of SnO sensors for CO he worked at Texas Instruments Inc., Dallas, TX, as
detection enhancement,” IEEE Trans. Instrum. Meas., vol. 54, no. 1, an internship co-op in the Data Converters Group, and
pp. 79–86, 2005. in 2002 he joined the Integrated Microsystems Labo-
[10] “Interface and control circuits for high-selectivity gas-sensors operated ratory of the University of Pavia as a Ph.D. candidate.
with temperature pattern,” PRIN 2005092937, funded by Italian Gov- He participated in high-resolution and high-speed
ernment [Online]. Available: http://ims.unipv.it/prin05 mixed-signal circuits design and characterization.
[11] J. W. Gardner, E. L. Hines, and C. Pang, “Detection of vapours and Since 2004, his research activity mainly focused on
odours from a multisensor array using pattern recognition self-or- smart sensor interfaces, especially in terms of resistance measurement techniques
ganising adaptive resonant techniques,” Meas. Contr., vol. 29, pp. development for gas-sensors arrays. For this application in 2004, he designed and
172–178, 1996. characterized a prototype chip in the first national government funded project,
[12] A. Carpenter and S. Grossberg, “A massively parallel architecture for while in 2005–2006, he developed a second integrated solution for gas-sensors
a self organizing neural pattern recognition machine,” Comput. Vis. interfacing in a new national project and participated in European GOSPEL net-
Graph. Image Process, no. 37, pp. 116–165, 1987. work activities providing gas-sensors arrays interfacing solutions. In 2007, he
[13] M. Grassi, P. Malcovati, and A. Baschirotto, “An uncalibrated 141 dB worked on a large format X-ray detector read-out for a space applications chip
dynamic-range CMOS gas-sensor interface with 16-bit digital output,” and on the development of an integrated temperature profile synthesizer for gas
in Proc. ESSCIRC, 2006, pp. 235–238. sensors. He is now interested in healthcare and homeland safety with particular
[14] M. Grassi, P. Malcovati, and A. Baschirotto, “A high-precision wide- focus on elder people, working as a Post-Doctoral Researcher at the Electrical
range front-end for resistive gas-sensors arrays,” Sensors and Actua- Department of University of Pavia in the European NetCarity project. He is the
tors B: Chemical, Eurosensors XVIII Special Issue, vol. 111–112, pp. author of three papers in international journals and author of more than 10 pa-
281–285, 2005. pers presented at international conferences (with published proceedings).
1554 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 42, NO. 7, JULY 2007

Piero Malcovati (M’95–SM’05) received the Laurea Andrea Baschirotto (M’95–SM’01) graduated in
degree (summa cum laude) in electronic engineering electronic engineering (summa cum laude) from
from University of Pavia, Italy, in 1991. In 1992, he the University of Pavia, Italy, in 1989. In 1994, he
joined the Physical Electronics Laboratory (PEL) at received the Ph.D. degree in electronics engineering
the Federal Institute of Technology in Zurich (ETH from the University of Pavia.
Zurich), Switzerland, as a Ph.D. candidate. He re- In 1994, he joined the Department of Electronics,
ceived the Ph.D. degree in electrical engineering from University of Pavia, as a Researcher (Assistant Pro-
ETH Zurich in 1996. fessor). In 1998, he joined the Department of Innova-
From 1996 to 2001, he was an Assistant Professor tion Engineering, University of Salento, Italy, as an
in the Department of Electrical Engineering of the Associate Professor. About his research activity, he
University of Pavia. Since 2002, he has been an Asso- founded and he is leading the Microelectronics Group
ciate Professor of electrical measurements at the same university. His research at University of Lecce, which is collaborating with several companies and re-
activities are focused on microsensor interface circuits and high-performance search institutions (including IMEC, Infineon, University of Pavia, RFDomus,
data converters. He has authored or co-authored more than 30 papers in interna- STMicroelectronics). His main research interests are in the design of CMOS
tional journals, more than 100 presentations at international conferences (with mixed analog/digital integrated circuits, in particular for low-power and/or high-
published proceedings), 5 book chapters, and 3 industrial patents. speed signal processing. He participated in several research collaborations, also
Dr. Malcovati was guest editor for the Journal of Analog Integrated Circuits funded by National and European projects. He is and has been responsible for
and Signal Processing for the special issue on IEEE ICECS 1999. He served as some national and regional projects for the design of ASIC. Since 1989, he
Special Session Chairman for the IEEE ICECS 2001 Conference, as Secretary has collaborated with several companies on the design of mixed-signal ASICs,
of the Technical Program Committee for the ESSCIRC 2002 Conference, and such as STMicroelectronics, Mikron, ACCO, ITC-IRST, IMEC, RFDomus (he
as Technical Program Chairman of the IEEE PRIME 2006 Conference. He was was in its Advisory Board), and Conexant. He has authored or co-authored
and still is a member of the scientific committees for several international con- more than 190 papers in international journals and presentations at interna-
ferences, including ESSCIRC, DATE, ECCTD, SENSORS, and PRIME. He is tional conferences and 6 book chapters, and holds 25 U.S. patents. In addi-
an associate editor of the Journal of Circuits, Systems, and Computers. tion, he has co-authored more than 120 papers within research collaborations
on high-energy physics experiments. He has long-term experience in microelec-
tronics for what concerns teaching, researching, and industrial designing. He has
been teaching regular academic courses since 1997. He organized the full ed-
ucational courses for electronics engineering (Bachelor, Master, and Ph.D.) at
the University of Salento. He has given industrial courses since 1996 (at Bosch,
STMicroelectronics, ITC-IRST, Conexant, and Mikron, among others). He is a
speaker at the MEAD Summer courses held at EPFL, Lausanne, Switzerland. He
has given short courses or tutorials at important conferences including ISSCC,
ISCAS, and PRIME.
Dr. Baschirotto was Associate Editor IEEE TRANSACTIONS ON CIRCUITS
SYSTEMS—PART II for the period 2000–2003, and of IEEE TRANSACTIONS ON
CIRCUITS SYSTEMS—PART I as an Associate Editor for the period 2004–2005.
He was the Technical Program Committee Chairman for ESSCIRC 2002
and the Guest Editor for the IEEE JOURNAL OF SOLID-STATE CIRCUITS for
ESSCIRC 2003. He was the General Chair of PRIME2006. He is a member
of the Technical Program Committees of several international conferences,
including ISSCC, ESSCIRC, and DATE.