Biosensors & Bioelectronics Vol. 7 No. 10

Sensor Arrays Based on Biological Systems M.R. Haskard and D.E. Mulcahy The Microelectronics Centre, School of Electronic Engineering, The University of South Australia, The Levels Campus, P.O. Box 1, Ingle Farm, SA 5098 Australia. Tel [61] 8-302-3308. Fax: [61] g-302-3384. Indexing terms: Microelectronic sensor, Sensor arrays, Taste, Smell. Instrument and control systems employ one or more discrete analogue sensors, yet examination of biological systems, particularly the human olfactory system, suggest an alternative strategy. Microelectronic technology allows this strategy of large arrays of primitive binary sensors to be implemented. Single and multiple species can be discriminated. Introduction The use of sensor arrays for gas/odour or solute/taste detection is an exciting current development area in microelectronics as evidenced by recent literature (Nakamoto et al, Gardner et al.). Most employ a small number of sensors and certainly less than 100. This communication proposes a method where arrays of many thousands of primitive sensors are used, giving a direct digital output. Work undertaken by Freeman and others on the human olfactory systems suggests that the number of receptor neurons excited depends on odour concentration. Because of turbulence, different receptor neurons are excited each inhalation, yet for a given past history, the same spatial pattern is generated in the olfactory bulb. The bulb can also be primed for a response, perhaps initially sampling the odours present in the inhaled air and then focusing in on a specific

odour. The features of:

a> a large sensor array b)

dependence of number of sensors excited on odour concentration

c)

a fixed stationary pattern for each odour

d)

independence of individual receptor neuron

e) ability to extract general information on an odour present

and specific

can be incorporated in the design microelectronic based sensor arrays.

of

Sensor Arrays An m by n array of microelectronic sensors can be constructed, each sensor having a threshold beyond which the odour concentration causes it to switch. Even though the sensors are notionally identical there is a Gaussian type production spread so that sensors switch over a small limited range of odour concentration (cl to c2 in Figure 1). This range would depend upon the quality of the process, so that for a ‘sloppy’ process the odour concentration range will be considerably greater. Over the range the resultant pattern of switched sensors would give a picture of odour concentration. This pattern could be analysed using artificial neural network methods. The sensors envisaged are simple, such as

689

Biosensors & Bioelectronics

-8TLIQIDARDg

Vol. 7 No. 10

Figure I. Gaussian distribution and accumulated response for a sensor array.

Number devices switched

,,

Number devices switched

c2

) “2

-------

n o

___

___

Odour concentration

c1

cO

c2

Odour concentration

+30

impedance (odour) or ISFET (taste), integrated with a threshold circuit as shown in Figure 2. Sensors in the array would have a sensitising coating, so that the array can respond to different odours/tastes. Thickness, composition and other parameter spreads can be interpreted simply as circuit variations, again increasing the range of operation of the array. Sampling the array pattern of switched sensors with time will reveal additional information if the process at an odour (or taste) coating interface are significantly time dependent. The sampling need not be at fixed rate, but can be varied logarithmically or in any other fashion and this process can be used to increase the range of

odour sensitivity and selectivity. An important factor is the number of sensors required to achieve a given accuracy for a specified confidence level. Consider again the n by m array and one analyte to be sensed. If p is the probability estimate of the number of sensors that have switched out of the total n.m independent sensors, then, taking a 95% confidence interval, the accuracy of the range, as a percentage, is given in Table 1. Thus using the 5 and 95 percentiles of the range, an array consisting of 1000 sensors will have an accuracy of better than a 5%. The range of detection can be extended in one of two ways. Firstly, the common threshold voltage is changed so that the narrow Gaussian

Figure. 2 Primitive conductance odour and ISFET taste sensors.

“dd

“dd

Pre-charge

Pre-charge

,

Thresh01 circuit

d

-

-

0

1

Threshold circuit

Sensor

Cnd

690

Cod

--c

Biosensors & Bioelectronics

Vol. 7 No. 10

distribution is a window that can be moved across a wider range of concentrations, as depicted in Figure 3. Secondly, both the range of detection and distribution can be modified in a pseudo-random or systematic way. For example, use of a resistor string to generate cell threshold voltages will give a rectangular distribution and a linear accumulative response. The accuracy of the Table 1: Percentage Accuracy of Range for a 95% Confidence Interval Against the Number of Sensors in an Array

potentials does not matter greatly. The m by n array can be replicated to form a much larger array, for example, an 2m by 2n array comprising four m by n arrays. Here economies in the silicon area can possibly be achieved by sharing threshold circuits on a row/column/array basis. Each sub array may have a different coating applied so that the full array can now respond to mixtures. Many coatings have poor selectivity and will respond to several components in the mixture. However, correct selection of the coating types will provide a different time varying pattern of switched devices for different mixtures. Even though the full array is sampled there will be occasions when only a portion of the resulting Figure 3. Extending the range by using a window technique.

pattern will be processed, either to give a speedy overview of the mixture composition or to extract specific information on one component of it. Implementation With today’s CMOS (complementary metal oxide silicon) technology and coatings (e.g. conducting organics, lipids, immobilised enzymes) interdigitised conductance/capacitance sensors or ISFETs (insulator-silicon field effect transistor) occupy areas of only a few tens of square microns. Inverters, Schmidt triggers and simple operational amplifier comparators have been used as threshold circuits. Variation of the backgate voltage has also been used to modify circuit threshold levels. Combined with digital sequential sampling circuits, sensor arrays of 100 by 100 elements can be accommodated on a modest-sized silicon integrated circuit of 30 sq mm area. Conclusions Using knowledge gained from studying the human olfactory system, coupled with silicon microelectronic technology, it is now possible to make very large arrays of primitive sensors that replicate some of the properties of biological sensing systems. The pattern of the number of switched devices indicates not only the nature of the analytes present, but their concentrations. Further, this output pattern is in a form ideally suited for artificial neural network processing. Acknowledgement We wish to acknowledge the assistance of Dr Brenton Dansie in calculating array sizes. References Freeman, W.J. The Physiology of Perception. (1991). Scien@c American, 264,2,34-41. Gardner, J.W., Shurmer, H.V. and Tan, T.T. (1992). Application of an Electronic Nose to the Discrimination of Coffees. Sensors a&Actuators B, 6,71-75.

Nakamoto, T., Fukunishi, K. and Moriizumi, T. (1990). Identification Capability of Odour Sensor Using Quartz-resonator Array and Neural-network Pattern Recognition. Sensors and Actuators B, 1,473-476.

691

Sensor arrays based on biological systems.

Biosensors & Bioelectronics Vol. 7 No. 10 Sensor Arrays Based on Biological Systems M.R. Haskard and D.E. Mulcahy The Microelectronics Centre, School...
244KB Sizes 0 Downloads 0 Views