Introducing a standard microprocessor to the mix was the next evolutionary step, from which came the first smart sensors (Fig. 1c). This provided a more powerful system that could utilize software for testing, calibrating, and linearizing the sensing system. Analog-to-digital conversion was required on both sides of the processor, however, and that added cost. Moreover, microprocessors themselves make the sensing systems more expensive than many applications can afford.
Don Pullen, applications engineer with Texas Instruments Inc.’s Linear Products Div., Dallas, sees the most recent evolutionary step as one in which the processor and its ADC and DAC cohorts are replaced with a signal processor dedicated to sensing operations (Fig. 1d). Raising the level of integration and replacing the standard microprocessor with an intelligent device specifically designed for the application gives sensor designers the advantages of programmability, low power consumption, and superior linear performance.
These sensor-signal processors can condition the analog signal from the transducer using a combination of hardware and software. Sensor linearization, for example, often includes piece-wise linear approximation of the sensor’s output characteristics by selecting a number of segments.
A good example is the nonlinear-response thermistor, used to measure temperature by monitoring the change in its resistance. The first step in linearizing the output is to introduce a resistor in parallel with the thermistor. This procedure results in a good approximation of a linear response. But even better accuracy can be achieved by using a simple algorithm that approximates the curve by a series of straight lines. Software-implemented multipoint approximation can even be adjusted to modify the characteristics of individual sensors to which the processor is attached, says Pullen.
Smart sensors can be self-calibrating, or at the very least, make calibration easier. During final test, multiple sensing units can be multiplexed with a host computer that exercises each sensor individually and downloads the appropriate calibration constants to the device. Moreover, field calibration of the sensor can be simplified by adding a keyboard interface or LCD interface to the sensor. Then a handheld computer can calibrate the sensor. Software also provides a means of running diagnostics on both the sensor and the transmitter in the sensing system. Autodiagnostics range from watchdog timers to periodic self-testing routines during operation, says Pullen.
An outstanding example of a next-generation smart-sensing system on one chip can be seen in technology developed by Monolithic Sensors Inc., Rolling Meadows, Ill., to guage low pressures. Made entirely of silicon, it combines an air-pressure sensor, CMOS analog sensor circuitry, and CMOS digital interface circuitry on one device.
MSI creates the sensor with the aid of micromachining technology. The device is essentially a capacitor consisting of a pair of diaphragms. The capacitance value changes as one diaphragm plate flexes under air pressure (Fig. 2). Corrugations machined in the silicon allow the diaphragm to flex. Its movement is both linear and quite large, says Warren Graber, an MSI applications engineer. Capacitance can change by as much as 25% during operation, which makes for an easy-to-detect and accurate reading. In addition, linearity is one of the sensor’s primary advantages.
Being fabricated completely of silicon means that all components have the same coefficient of expansion, which curbs thermal-sensitivity problems. The sensor can also be built with mechanical stops for an over-pressure capability of 100 times its rated pressure. Perhaps most important, however, is that since it’s made of silicon, all of the associated circuitry can be put on one chip.
On the analog side, the variable capacitor is used in a frequency oscillator that varies with pressure. In addition, a reference frequency is created by a fixed capacitor that drives another oscillator. The frequency information is converted to digital samples that are proportional to the ratio of the two capacitors. A digital CMOS interface provides output as two 8-bit words. In typical smart-sensor fashion, the digital circuits allow calibration data to be stored in EPROM. The sensor can be calibrated in the factory and needs no trimming or periodic recalibration. Total worst-case error for the system is less than 5% and digital compensation can improve accuracy to within 0.5%. The device is packaged in a 24-pin plastic DIP that has two airhose fittings so the sensor can operate with a differential air-pressure input.
Smart sensing as implemented in MSI’s technology is but one part of the new outlook for sensors. New materials that provide superior performance to conventional solutions also are under investigation in laboratories around the world.
Matshushita Electric Industrial Co. Ltd.’s laboratories, located in Tokyo, Japan, are currently tackling pressure-sensing technology. Using an iron-base amorphous magnetic-alloy ribbon, researchers have developed an oil-pressure sensor that measures up to 20 megaPascals (MPa), with full-scale accuracy coming within less than 2%. The sensor operates at temperatures between -30 and 100 [degrees] C, which makes it acceptable for suspension and brake systems in automobiles.
Amorphous magnetic alloys have many attractive properties for harsh-environment sensor applications, including high mechanical strength, corrosion resistance and, most important, a variable magnetic permeability that’s caused by stress. The research team’s cylindrical sensor is built of titanium and has two hollow chambers at either end that are unconnected. One chamber is filled with air and serves as a reference; the other is filled with oil. A strip of iron-base amorphous alloy is annealed to the cylinder. Two coils encircle the reference and detecting chambers, respectively, to detect the change in permeability of the amorphous alloy when pressure is applied. To measure an output, the sensor is used in a simple electronic circuit that has two inductors (the coils), two compensation resistors, an ADC, and a DAC. The circuit is driven by a 32-kHz voltage source.
Although the sensor is still a few years from commercial use, the researchers have been able to model its response mathematically. This means it’s a likely candidate for integration into a smart-sensor system because the mathematical model can be used for in-situ calibration and self test.
Significant progress is also being made in another new field: optical sensors. Research interest is particularly strong for applications such as pollution monitoring, in which the concentrations of the material to be identified are as low as parts per billion. Electronic sensors can seldom monitor at this level of precision.
Researchers at the Georgia Institute of Technology, for example, have developed a generic device called an integrated-optic interferometer. This planar waveguide consists of an approximately 1-[in.sup.2]. glass substrate (Fig. 3). It’s coated with a thin film that has a slightly higher index of refraction than glass. The thin film is made of a material that reacts in a detectable but reversible way with the contaminant the device is supposed to sense. The basic operating model calls for the thin film’s index of refraction to vary proportionally with the amount of contaminant it absorbs.
In a generic interferometer, changes in refraction are measured by having a tiny laser emit light that’s split into two beams. One beam propagates through the glass-based substrate to establish a reference point. The other beam propagates through both the substrate and the thin film. If the film’s refraction index changes, the beam will undergo a phase shift relative to the reference. The phase shift is generally proportional to the concentration of the contaminant. Researchers can develop calibration curves to measure the amount of contaminant down to parts per billion.
Because the sensor is so small and its operation relatively simple, it can be easily taken to the location of the suspected contaminant. A single sensing device can also be used to detect more than one contaminant. This is accomplished by applying several types of thin films to the glass substrate. Each film is reactive to a different contaminant.
Reversibility of the chemical reaction is an important characteristic for the sensor, but that’s largely a matter of finding the right materials for the thin film. One of the first working integrated-optic interferometer prototypes, for example, was developed to measure the presence of ammonia. A film of dodecylanilinium salt dissolved in ethanol was spin-applied tot the substrate. Early devices could detect ammonia in the air down to a few parts per million. With full integration of the laser light source and a planar waveguide, the research team was able to reduce sensor noise. Sensitivity dropped to the 100 parts per billion range.
The device doesn’t react at all to water vapor, which is important because it will be used primarily in the field. Moreover, this one research project that doesn’t have to endure an aggressive cost-reduction phase to reach commercial viabilitty: The device’s components cost less than $100. The technology has been licensed to Photonic Systems Inc., a “technology incubator” company on the Georgia Tech campus. Work is underway on a integrated-optic interferometer that can detect multiple contaminants.
Medical applications have also shown a strong affinity toward advanced sensor technology using light sources. In a collaboration between Sandia National Laboratory and the University of New Mexico, both of Albuquerque, N.M., infrared light is used in a non-invasive glucose sensor for diabetes patients. Although glucose was chosen for the first system, the technology, which is based on infrared spectroscopy and statistical techniques, is readily applicable to other measurements.
Near-infrared light has the capacity to penetrate biological tissue, and this sensor operates on that principle. Besides, being non-invasive, the procedure allows for continuous blood-glucose monitoring, highly desirable for diabetics undergoing surgery or in childbirth. Another scenario made possible by continuous glucose monitoring using light is the development of a monitor that could be worn with a programmable insulin pump. This would, in effect, create an artificial pancreas. Small pumps that can meter pumped insulin into the bloodstream have been researched at Sandia.
In the prototype monitor, wavelengths of light passing through the patient’s finger are absorbed by components in the blood. The spectral characteristics of the remaining light are recorded by a spectrometer and evaluated by versions of algorithms originally developed at Sandia to analyze the aging process of nuclear-weapons material. This relatively new branch of analytical chemistry is called chemometrics. Two statistical methods–Partial Least Squares and Principal Component Regression–were tested, as were three instrument configurations. The researchers concluded that near-infrared testing is viable and are seeking commercial partners through a company, Rio Grande Medical Technologies Inc., Albuquerque, NM.
Quite a different approach to the same problem–glucose sensing–is being taken by researchers at Fujitsu Ltd’s research laboratories in Japan. The key component in this group’s biosensor research is a miniature Clark oxygen electrode that’s micromachined in silicon. Its operation is based on the fact that the sensor detects the changes in oxygen concentration that result from glucose oxidation, which is caused by a glucose oxidase enzyme situated in the sensor. The amount of oxygen is almost linearly proportional to the current, measured in nanoamperes, that is generated by the Clark oxygen electrode.
Two types of electrodes were fabricated. In the first, the electrode resides on a 2-by-15-by-400-mm undoped silicon substrate. The electrolyte is contained in two etched 0.2-by-0.7-mm V grooves. The anode is formed in the grooves and the cathodes occupy the area between the grooves. Both anode and cathode consist of a 400-nm-thick gold film with a 40-nm chromium adhesive layer. Several electrolytes were tested, including potassium-chloride suspended in polyvinylpyrrolidone (PVP) and poly vinyl-4-ethylpyridinium bromide (PVEP). A gaspermeable membrane resides on top of the electrolyte and steam is used to activate the electrolyte by infusing water into the electrolyte. The glucose-oxidizing enzyme adheres to the membrane.
The prototype sensor exhibited a response time of one minute and showed good linearity for glucose concentrations between 56 micromoles and 1.1 millimoles at 38 [degrees]C and a pH of 7.0. It had a lifetime of about 10 days before the enzyme matrix detached from the gas-permeable membrane. Using the same miniature Clark electrode structure, biosensors were also fabricated to test for the presence of carbon-dioxide, L-lysine, and hjypoxanthine, a chemical that can be monitored to determine the freshness of fish. The Fujitsu researchers found several problems with the first electrode structure, however, including electrochemical crosstalk between electrodes. As a result, they’re working on a new electrode structure that incorporates several architectural changes.
At the cutting edge of sensor technology, researchers are beginning to design circuits that can recognize images and sounds. Work at the California Institute of Technology, Pasadena, for example, attempts to model, to a limited degree, the ear’s cochlea in both function and structure. The ultimate goal of the research is to create devices that interpret and understand sound, pinpoint the direction from which a sound is coming, and perhaps even provide a cochlear prosthesis.
At Cal Tech, researchers John Lazarro and Carver Mead developed a chip architecture that computes all outputs in real time using analog continuous-time processing. As one might imagine, the project is conceptually complex. One key difference between biological ears and the chip is that the analog mechanical processing done by the ear is accomplished electronically in the chip. For example, sound energy in the eardrum is coupled into a mechanical traveling-wave structure called the basilar membrane, which converts time-domain information into spatially encoded information. In the chip, this is approximated by a cascade of second-order circuits with exponentially increasing time constants, says Lazzaro, who is now at the University of California at Berkeley.
In addition to the basilar membrane, the chip also models outer hair cells, inner hair cells, and spiralganglion neurons (Fig. 4). The outer-hair-cell circuits control local damping of the basilar-membrane circuit. Taps along the basilar membrane connect to circuit models of the inner-hair cells. Outputs from the inner-hair cells connect to circuits that model spiral-ganglion neurons. These neuron circuits form the primary output of the chip that models the auditory response.
In both physiological systems and the IC, the response to sound is a series of voltage spikes, which correspond roughly to biological neurons firing. After the chip was fabricated, it was subjected to two tests. The first compared with the chip’s response to an 1840-Hz tone with that of a cat. The second consisted of 2000 30-dB clicks. In both cases, the chip’s response was qualitatively the same as the response from the auditory fiber of a cat. The tests were relatively simple ones, however, and more modeling of physiological systems is needed before commercial versions can be implemented. The chip can encode 25 dB of dynamic range, for example, compared to 125 dB for human ears.
The importance of the research results so far is two-fold. First, the chip does roughly model the response of the biological system it mimics. Second, the computations are done in real time by using an s architectural model similar to the physical cochlea. The model includes autocorrelations in time and cross-correlations between auditory fibers.
Research into silicon retinas and position-sensing devices is further along than auditory systems. Work is well underway in several universities including Cal Tech, the University of California at Berkeley, and the University of Pennsylvania, to name just three (ELECTRONIC DESIGN, May 3, p. 33).
Recent investigative research at the University of Pennsylvania focused on 2D motion detection, a capability that would have use in robotics, surveillance, and other systems. By combining two models that seel to describe motion detection in biological systems, the University of Pennsylvania researchers were able to design an analog chip that could detect motion as fast as 6 meters/s in two dimensions. It also responds well over a range of six orders of magnitude in illumination.
The chip was a 5-by-5 array of photoreceptors, each 100 by 100 [mu]m. Although the output current of the sensors is basically linear, it’s converted to a logarithmic response so that the sensors can respond to a large input range without saturating. Motion sensing requires edge detection. This is implemented by a circuit that functionally models a retina to some degree. The inputs of every pixel are averaged in a complex operator called Difference og Gaussians, which discards the local average input and shows only disontinuities. As in the silicon cochlea, many of the computations are executed in the analog domain.
Motion detection is a difficult problem, but easier ones, such as character reading, are already being solved in commercial applications that generally include neural networks. The Synaptics Inc, I1000 chip, for example, combines an imager with a single-font character-recognition system to read the computer-coded characters on band cheeks. “The first commercial applications to use these silicon eye and ear sensors,” says Lazzaro, “will solve simple tasks for lower-power, high-volume, size-sensitive applications.”