Lecture #2

Basic Intent
This lecture is intended to overview the basic sensor terminology as it is generally used on product data sheets and in the technical literature. This overview is presented in general terms, as well as in the description of a specific sensor, an off-the-shelf accelerometer, ADXL50A. Then, a summary of the basic electronic circuits we will be covering this quarter is presented.

Where do Sensors Come From?

There are a lot of sensors out there, and we'll do our best to cover a significant fraction of them this quarter. In all cases, the sensors we're going to be looking at are the result of some sort of invention, and it is interesting to think about how such things come to be. In this day and age, there are researchers at companies, universities, and out there on their own competing to invent, design, build and sell the sensors which are going to be the next big wave in the industry. This competition has been going on for decades, and there are several basic facts at work: 1. Many of the obvious, easy inventions have already been tried out and completed or abandoned. As a potential sensor inventor, it is important to realize that there are a lot of people out there thinking about similar problems, and every day isn't going to be filled with really new ideas. Put in the harshest terms: most of the things you might think of have already been thought of. 2. Some of the people working in this business have great resources at their disposal. For example, University professors get to hire crowds of the smartest people on the planet (students), pay them a pittance, and watch them invent and test ideas. Or other professors (like me) get to teach classes where they ask their students to come up with cool ideas, work out some details, and then write the whole thing up in a term paper Other examples include researchers at national labs or big industrial labs, where access to state of the art materials and equipment is key to finding the next really important sensor. It is VERY HARD to compete with all of these people. So, access to resources (people, equipment,etc) is a big advantage. 3. There are some really important problems out there. If the Auto Industry says it is interested in buying 20 million gyroscopes at a price of $10 each every year, you can be sure that hundreds of people are trying to meet that challenge. So, there are some industries which get to set the challenges that are the focus of inventor interest.

In light of this, I have some basic advice to people who might be interested in being inventors or product developers. Simply, I would discourage most people from getting into the game of trying to invent the next really important sensor. However, there is another game that I think is much easier to play, and much more likely to be lucrative. In a nutshell, I think it is much easier to be the inventor who uses the next cool sensor in a product that the sensor was not originally intended for. Here's an example : The Auto industry wants to use gyros for automatic skid control. Because of this, there are a lot of people working on ways to develop inexpensive gyros to detect auto skidding. There are a lot of other applications of gyros with this basic performance potential. I think the smart play is to assume that the auto industry will get what it wants sooner or later, and focus your efforts on thinking about other applications of a decent, small $10 gyro. To win this game, you need to think about some of the things that the big players want to have available, and think about other things you could do with them. Who are the big players? 1. Auto Industry. There are 15 million cars sold a year in the US. Most components cost $5. If the auto industry wants something, they usually get it. Right now, they want to detect skids, detect the location and orientation of the passengers, and they want to provide navigational assistance to drivers. 2. Medical Industry. You only have to watch ER or any other hospital-based drama on TV to be shown several examples each hour of sensor-based medical technology. The big things this industry wants are blood chemistry sensors, low-cost, high speed diagnostics, implantable therapeutic devices, and DNA testing. 3. The Department of Defense (DoD). The cold war might be over, but the DoD still has the largest basic research budget in the world. The DoD loves sensors, and is very interested in technologies that allow us to wage war without putting personnel at risk. Night vision, navigation, "smart" soldier, smart munitions, bullets, etc, Stealth technologies for Sonar and Radar, and many more. The DoD usually gets what they want, and they have the $$$ to pay for people to work on their problems, so things will emerge. So, read up on the cutting edge in these communities, and make assumptions about things coming out sooner or later, and be prepared to use those things in new ways

Introduction to Sensor Terminology

For our purposes, a Sensor is a device which converts a physical phenomena into an electrical signal. As such, sensors represent part of the interface between the physical world and the world of electrical devices, such as computers. The other part of this interface is represented by Actuators, which convert electrical signals into physical phenomena. Why do we care so much about this interface? In recent years, enormous capability for information processing has been developed within the electronics industry. The largest example of this capability is the personal computer. In addition, the availability of inexpensive microprocessors is having a tremendous impact on the design of products ranging from automobiles to microwave ovens to toys. In recent years, versions of these products that utilize microprocessors for control of functionality are becoming widely available. In automobiles, such

capability is necessary to achieve compliance with pollution restrictions. In other cases, such capability simply offers an inexpensive performance advantage. All of these microprocessors need electrical input voltages in order to receive instructions and information. So, along with the availability of inexpensive microprocessors has grown an opportunity for the use of sensors in a wide variety of products. In addition, since the output of the sensor is an electrical signal, we tend to characterize sensors in the same way we characterize electronic devices. The data sheets for many sensors are formatted just like electronic product data sheets. However, there are many formats out there, and nothing at all like an international standard for sensor specifications exists. We will encounter a variety of interpretations of sensor performance parameters, and sometimes a lot of confusion will emerge. It is important for you to realize that this confusion is not due to our inability to explain the meaning of the terms - it is a result of the fact that different parts of the sensor community have gotten comfortable using these terms differently. It is important to realize the function of the data sheet in order to deal with this variability. The data sheet is primarily a marketing document. It will be designed to highlight the positive attributes of the sensor, emphasize some of the potential uses of the sensor, and might neglect to comment on some of the negative characteristics of the sensor. In many cases, the sensor has been designed to meet a particular performance specification for a specific customer, and the data sheet will concentrate on the performance parameters of greatest interest to this customer. In this case, the vendor and customer might have grown accustomed to unusual definitions for certain sensor performance parameters. As a potential new user of such a sensor, it is initially your problem to recognize this situation, and interpret things reasonably. So, expect that you will encounter odd definitions here and there, and expect that you will find that most sensor data sheets are missing some information that you might be most interested in. That is the nature of the business.

Sensor Performance Characteristics Definitions

Transfer Function: The functional relationship between physical input signal and electrical output signal. Usually, this relationship is represented as a graph showing the relationship between the input and output signal, and the details of this relationship may constitute a complete description of the sensor characteristics. For expensive sensors which are individually calibrated, this might take the form of the certified calibration curve. Sensitivity: The sensitivity is defined in terms of the relationship between input physical signal and output electrical signal. The sensitivity is generally the ratio between a small change in electrical signal to a small change in physical signal. As such, it may be expressed as the derivative of the transfer function with respect to physical signal. Typical units:

Volts/Kelvin. A Thermometer would have "high sensitivity" if a small temperature change resulted in a large voltage change. Span or Dynamic Range: The range of input physical signals which may be converted to electrical signals by the sensor. Signals outside of this range are expected to cause unacceptably large inaccuracy. This span or dynamic range is usually specified by the sensor supplier as the range over which other performance characteristics described in the data sheets are expected to apply. Typical units: Kelvin Accuracy: Generally defined as the largest expected error between actual and ideal output signals. Typical Units: Kelvin. Sometimes this is quoted as a fraction of the full scale output. For example, a thermometer might be guaranteed accurate to within 5% of FSO (Full Scale Output) Hysteresis: Some sensors do not return to the same output value when the input stimulus is cycled up or down. The width of the expected error in terms of the measured quantity is defined as the hysteresis. Typical units: Kelvin or % of FSO Nonlinearity (often called Linearity): The maximum deviation from a linear transfer function over the specified dynamic range. There are several measures of this error. The most common compares the actual transfer function with the `best straight line', which lies midway between the two parallel lines which encompasses the entire transfer function over the specified dynamic range of the device. This choice of comparison method is popular because it makes most sensors look the best. Noise: All sensors produce some output noise in addition to the output signal. In some cases, the noise of the sensor is less than the noise of the next element in the electronics, or less than the fluctuations in the physical signal, in which case it is not important. Many other cases exist in which the noise of the sensor limits the performance of the system based on the sensor. Noise is generally distributed across the frequency spectrum. Many common noise sources produce a white noise distribution, which is to say that the spectral noise density is the same at all frequencies. Johnson noise in a resistor is a good example of such a noise distribution. For white noise, the spectral noise density is characterized in units of Volts/Root(Hz). A distribution of this nature adds noise to a measurement with amplitude proportional to the square root of the measurement bandwidth. Since there is an inverse relationship between the bandwidth and measurement time, it can be said that the noise decreases with the square root of the measurement time. Resolution: The resolution of a sensor is defined as the minimum detectable signal fluctuation. Since fluctuations are temporal phenomena, there is some relationship between the timescale for the fluctuation and the minimum detectable amplitude. Therefore, the definition of resolution must include some information about the nature of the measurement being carried out. Many sensors are limited by noise with a white spectral distribution. In these cases, the resolution may be specified in units of physical signal/Root(Hz). Then, the actual resolution for a particular measurement may be obtained by multiplying this quantity by the square root of the measurement bandwidth. Sensor data sheets generally

quote resolution in units of signal/Root(Hz) or they give a minimum detectable signal for a specific measurement. If the shape of the noise distribution is also specified, it is possible to generalize these results to any measurement. Bandwidth: All sensors have finite response times to an instantaneous change in physical signal. In addition, many sensors have decay times, which would represent the time after a step change in physical signal for the sensor output to decay to its original value. The reciprocal of these times correspond to the upper and lower cutoff frequencies, respectively. The bandwidth of a sensor is the frequency range between these two frequencies. These definitions are adapted from those in Fraden, and will be used in this manner throughout the course.

Sensor Performance Characteristics of an Example Device

To add substance to these definitions, we will identify the numerical values of these parameters for an off-the-shelf accelerometer, ADXL50A from Analog Devices. View the ADXL50A Data Sheets with Adobe Acrobat. Transfer Function The functional relationship between voltage and acceleration is stated as

This expression may be used to predict the behavior of the sensor, and contains information about the sensitivity and the offset at the output of the sensor. Sensitivity The sensitivity of the sensor is given by the derivative of the voltage with respect to acceleration at the initial operating point. For this device, the sensitivity is 19 mV/g. Dynamic Range For ADXL50A accelerometer, the stated dynamic range is +/- 50g. For signals outside this range, the signal output is saturated at either 0.25V or 4.75V. The device can withstand up to 2000g without damage. Hysteresis There is no fundamental source of hysteresis in this device. There is no mention of hysteresis in the Data Sheets. Temperature Coefficient In this device, temperature can introduce a change in sensitivity. The change is less than 1% over the range from -40 to +85 degrees Celsius. There is also a shift in offset of up to 35 mV. Linearity In this case, the linearity is the difference between the actual transfer function and the best straight line over the specified operating range. For this device, this is stated as less than 0.2% of the full scale output. Figure 5 in the Data Sheets show the expected deviation from linearity. Accuracy

The accuracy is essentially limited by the nonlinearity and the temperature coefficients. Altogether, the device is accurate to within 3% over the full scale signal range and over temperatures from -40 to +85 degrees Celsius. Noise Noise in this device comes from the electronic measuring circuit, and is expressed as 125 uV/sqrt(Hz). This noise density should be used to calculate the actual noise for a particular measurement. For example, if the output is filtered by a 10Hz low-pass, the RMS Noise would be

Resolution The resolution is the minimum detectable signal fluctuation. This is given by the voltage noise density divided by the sensitivity

Again, for a real experiment with a 10Hz bandwidth, the resolution would come to 20mg. Bandwidth The bandwidth of this sensor depends on choice of an external capacitor. For C = 0.022 uF, the Bandwidth is approx. 1300Hz. For C = 0.007 uF, B/W = 10kHz.

Introduction to Sensor Electronics

The electronics which go along with the physical sensor element are often very important to the overall device. The sensor electronics can limit the performance, cost, and range of applicability. If carried out properly, the design of the sensor electronics can improve the characteristics of the entire device. As for the rest of this course, the intent is not to prepare you to design sensors in great detail. Nevertheless, it is important to include some discussion of sensor electronics. We will focus on basic techniques for processing the signals most typically produced by a sensor. Most sensors do not directly produce voltages. Instead, most sensors act like passive devices, such as resistors, whose values change in response to external stimuli. In order to produce voltages suitable for input to a microprocessors and their analog to digital converters, the resistor needs to be `biased' and the output signal needs to be `amplified'.

Types of Sensors
Resistive Sensor Circuits

Figure 1: Voltage Divider

Resistive devices obey Ohm's law, which basically states that when current flows through a resistor, there will be a voltage difference across the resistor. So, one way to measure resistance is to force a current to flow and measure the voltage drop. Current sources can be built in number of ways (see Horowitz and Hill for loads of good examples). One of the easiest current sources to build is to take a voltage source and a stable resistor whose resistance is much larger than the one you're interested in measuring. The reference resistor is called a load resistor, and the two resistor configuration is sometimes called a resistive bridge. Analyzing the connected load and sense resistors as shown in Figure 1, we can see that the current flowing through the circuit is nearly constant, since most of the resistance in the circuit is constant. Therefore, the voltage across the sense resistor is nearly proportional to the resistance of the sense resistor. As stated, the load resistor must be much larger than the sense resistor for this circuit to offer good linearity. As a result, the output voltage will be much smaller than the input voltage. Therefore, some amplification will be needed. Capacitance measuring circuits Many sensors respond to physical signals by producing a change in capacitance. How is capacitance measured? Essentially, all capacitors have an impedance which is given by

where `f' is the oscillation frequency in Hz, is in rad/sec, and `C' is the capacitance in Farads. The `i' in this equation is the square root of -1, and signifies the phase shift between the current through a capacitor and the voltage across the capacitor. Now, ideal capacitors cannot pass current at DC, since there is a physical separation between the conductive elements. However, oscillating voltages induce charge oscillations on the plates of the capacitor, which act as if there is physical charge flowing through the circuit. Since the

oscillation reverses direction before substantial charges accumulate, there are no problems. The effective resistance of the capacitor is a meaningful characteristic, as long as we are talking about oscillating voltages. With this in mind, the capacitor looks very much like a resistor. Therefore, we may measure capacitance by building voltage divider circuits as in Figure. 1, and we may use either a resistor or a capacitor as the load resistance. It is generally easiest to use a resistor, since inexpensive resistors are available which have much smaller temperature coefficients than any reference capacitor. Following this analogy, we may build capacitance bridges as well. The only substantial difference is that these circuits must be biased with oscillating voltages. Since the resistance of the capacitor depends on the frequency of the AC bias, it is important to select this frequency carefully. By doing so, all of the advantages of bridges for resistance measurement are also available for capacitance measurement. However, providing an AC bias is a substantial hassle. Moreover, converting the AC signal to a dc signal for a microprocessor interface can be a substantial hassle. On the other hand, the availability of a modulated signal creates an opportunity for use of some advanced sampling and processing techniques. Several good examples are described in the textbook, and there are several more in any good circuits book, such as Horowitz and Hill. Generally speaking, voltage oscillations must be used to bias the sensor. It can also be used to trigger voltage sampling circuits in a way that automatically subtracts the voltages from opposite clock phases. Such a technique is very valuable, because signals which oscillate at the correct frequency are added up, while any noise signals at all other frequencies are subtracted away. One reason these circuits have become popular in recent years is that they may be easily designed and fabricated using ordinary digital VLSI fabrication tools. Clocks and switches are easily made from transistors in CMOS circuits. Therefore, such designs can be included at very small additional cost - remember that the oscillator circuit has to be there to bias the sensor anyway. So, capacitance measuring circuits are increasingly implemented as integrated clock/sample circuits of various kinds. Such circuits are capable of good capacitance measurement, but not of very high performance measurement, since the clocked switches inject noise charges into the circuit. These injected charges result in voltage offsets and errors which are very difficult to eliminate entirely. Therefore, very accurate capacitance measurement still requires expensive precision circuitry. Inductance measurement circuits Inductances are also essentially resistive elements. The resistance of an inductor is given by

(where L is the inductance), and this resistance may be compared with the resistance of any other passive element in a divider circuit or in a bridge circuit as shown in Figure 1 above. Inductive sensors generally require expensive techniques for the fabrication of the sensor mechanical structure, so inexpensive circuits are not generally of much use. In large part, this is because inductors are generally 3-dimensional devices, consisting of a wire coiled around a form. As a result, inductive measuring circuits are most often of the traditional variety, relying on resistance divider approaches.

Limitations

Limitations to resistance measurement

Lead Resistance - The wires leading from the resistive sensor element have resistance of their own. These resistances may be large enough to add errors to the measurement, and they may have temperature dependencies which are large enough to matter. One useful solution to the problem is the use of the so-called 4-wire resistance approach (Figure. 2). In this case, current (from a current source as in Figure. 1) is passed through the leads and through the sensor element. A second pair of wires is independently attached to the sensor leads, and a voltage reading is made across these two wires alone.

Figure 2: Lead Resistance Compensation with 4-wire measurement. Note that E represents a voltage measurement. It is assumed that the voltage measuring instrument does not draw significant current (see next point), so it simply measures the voltage drop across the sensor element alone. Such a 4-wire configuration is especially important when the sensor resistance is small, and the lead resistance is most likely to be a significant problem.

Output Impedance - The measuring network has a characteristic resistance which, simply put, places a lower limit on the value of a resistance which may be connected across the output terminals without changing the output voltage. For example, if the thermistor resistance is 10k and the load resistor resistance is 1 M, the output impedance of this circuit is approximately 10k. If a 1k resistor is connected across the output leads, the output voltage would be reduced by about 90%. This is because the load applied to the circuit (1k) is much smaller than the output impedance of the circuit (10k), and the output is `loaded down'. So, we must be concerned with the effective resistance of any measuring instrument that we might attach to the output of such a circuit. This is a well-known problem, so measuring instruments are often designed to offer maximum input impedance, so as to minimize loading effects. In our discussions we must be careful to arrange for instrument input impedance to be much greater than sensor output impedance.

Limitations to measurement of capacitance

Stray Capacitance - Any wire in a real world environment has a finite capacitance with respect to ground. If we have a sensor which has an output which looks like a capacitor, we must be careful with the wires which run from the sensor to the rest of the circuit. These stray capacitances appear as additional capacitances in the measuring circuit, and can cause errors. One source of error is the changes in capacitance which result from these wires moving about with respect to ground, causing capacitance fluctuations which might be confused with the signal. Since these effects can be due to acoustic pressureinduced vibrations in the positions of objects, they are often referred to as microphonics. An important way to minimize stray capacitances is to minimize the separation between the sensor element and the rest of the circuit. Another way to minimize the effects of stray capacitances is mentioned later - the virtual ground amplifier.

Filters
Electronic filters are important for separating signals from noise in a measurement. During this course, we'll look at a few simple filters, and I'll expect you to be able to work through simple circuits with some of these filters in them.

Low pass - A low-pass filter (Figure 3.) uses a resistor and a capacitor in a voltage divider configuration. In this case, the resistance of the capacitor decreases at high frequency, so the output voltage decreases as a the input frequency increases. So, this circuit effectively filters out the high frequencies and passes the low frequencies.

Figure 3: Low-pass Filter The mathematical analysis is as follows : Using the complex notation for the impedance, let

Using the voltage divider equation in Fig. 1

Substituting for Z1 and Z2

The magnitude of Vout is

and the phase of Vout is

High-pass - The high pass filter is exactly analogous to the low pass, except that the roles of the resistor and capacitor are reversed. The analysis of a high-pass filter is as follows :

Fig. 4: High-pass Filter Similar to low-pass filter,

The magnitude is

and the phase is

Band-pass - By combining low-pass and high-pass filters together, we can create a bandpass filter that allows signals between two preset oscillation frequencies. Its diagram and the derivations are as follows:

Fig. 5: Band-pass Filter Let the high-pass filter have the roll off frequency 1 and the low-pass filter have the roll off frequency 2 such that

Then the relation between Vout and Vin is

The operational amplifier in the middle of the circuit was added in this circuit to isolate the high-pass from the low-pass filter so that they do not effectively load each other. The op-amp simply works as a buffer in this case. In the following section, the role of the op-amps will be discussed more in detail. To further understand the purpose and theory of the follower op-amp configuration, see Operational amplifiers.

Operational amplifiers

Op-Amps are electronic devices which are of enormous generic use for signal processing. The use of op-amps can be complicated, but there are a few simple rules and a few simple circuit building blocks which we need to be familiar with to understand many common sensors and the circuits used with them. An op-amp is essentially a simple 2-input, 1-output device. The output voltage is equal to the difference between the non-inverting input and the inverting input multiplied by some extremely large value (105). Use of op-amps as simple amplifiers is uncommon.

Figure 6: Non-inverting Unity Gain Amplifier One really valuable concept for use of op-amps is that of feedback. For instance consider the circuit shown in Figure 6. This is called the follower configuration. Notice that the inverting input is tied directly to the output. In this case, if the output is less than the input, the difference between the inputs is a positive quantity, and the output voltage will be increased. This adjustment process continues - until the output is at the same voltage as the non-inverting input. Then, everything stays fixed, and the output will follow the voltage of the non-inverting input. This circuit appears to be useless, until you consider that the input impedance of the op-amp can be as high as 109 ohms, while the output can be many orders of magnitude smaller. Therefore, this follower circuit is a good way to isolate circuit stages with high output impedance from stages with low input impedance. This op-amp circuit can be analyzed very easily, using the op-amp golden rules: 1. No current flows into the inputs of the op-amp 2. When configured for negative feedback, the output will be at whatever value makes the input voltages equal. Even though these golden rules only apply to ideal operational amplifiers, the op-amps can in most cases be treated as ideal. Let's use these rules to analyze some more circuits...

Fig. 7: Inverting Amplifier

Fig. 7 shows an example of an inverting amplifier. We can derive the equation by taking following steps. 1. The + terminal is ground. Therefore, the terminal is also ground. (Rule 2) 2. Since the current flowing from Vin to Vout is constant (Rule 1), Vout/R2 = -Vin/R1 3. Therefore, voltage gain = Vout/Vin = -R2/R1

Fig. 8: Non-inverting Amplifier Fig. 8 illustrates another useful configuration of an op-amp. This is a non-inverting amplifier, which is slightly different expression than the inverting amplifier. Taking it step-by-step,
1. 2. 3. 4.

V- = Vin (Rule 2) Since V- comes from a voltage divider, V- = (R1/(R1 + R2)) Vout Therefore, Vin = (R1/(R1 + R2)) Vout Vout/Vin = (R1 + R2)/R1 = 1 + R2/R1

The op-amp rules are simple enough that I'll expect you to be able to use them to work through simple circuits and figure out what the voltages are doing.

Summary
This lecture has overviewed the basic characteristics of sensors that you can expect to find specified in sensor data sheets. The details of those definitions are discussed for the case of a resistance thermometer, and numerical values are produced for a typical device. Finally, some background on electrical measurement of sensor outputs is given. Some details regarding the behavior of simple passive filters and operational amplifiers are also given.

Lecture #3
Basic intent
This lecture is intended to provide an overview of piezoresistive devices. Some examples are worked out using this sensing technique.

Piezoresistance
A piezoresistor is basically a device which exhibits a change in resistance when it is strained. There are two components of the piezoresistive effect in most materials - the geometric component and the resistive component. The geometric component of piezoresistivity basically comes from the fact that a strained element undergoes a change in dimension (Figure 1). These changes in cross sectional area and length affect the resistance of the device.

Figure 1: Geometric change (strain) from applied force A good example of the geometric effect of piezoresistivity is the liquid strain gauge. It sounds weird, but there was a great many liquid strain gauges in use many years ago. Imagine an elastic tube filled with a conductive fluid, such as mercury (really!). The resistance of the mercury in the tube can be measured with a pair of metal electrodes, one at each end. Since mercury is essentially incompressible, forces applied along the length of the tube stretch it, and also cause

the diameter of the tube to be reduced, with the net effect of having the volume remain constant. The resistance of the strain gauge is given by , where is resistivity of the mercury, L is length of the conductive fluid, A is the cross-sectional area, and V is the volume. Taking the derivative gives

We define a quantity called the gage factor K as:

Since , we have K = 2 for a liquid strain gage. This means that the fractional change in resistance is twice the fractional change in length. In other words, if a liquid strain gauge is stretched by 1%, its resistance increases by 2%. This is true for all liquid strain gauges, since all that is needed is that the medium be incompressible. Liquid strain gauges were used in hospitals for measurements of fluctuations in blood pressure. A rubber hose filled with mercury was stretched around a human limb, and the fluctuations in pressure were recorded on strip-chart recorders, and the shape of the pressure pulses could be used to diagnose the condition of the arteries. Such devices have been replaced by solid state strain gauge instruments in modern hospitals, but this example is still interesting to use as an introductory example. Metal wires can also be used as strain gauges. As is true for the liquid strain gauge, stretching of the wire changes the geometry of the wire in a way which acts to increase the resistance. For a metal wire, we can calculate the gage factor as we did for the liquid gauge, except that we can't assume that the metal is incompressible, and we can't assume that the resistivity is a constant:

L L = A r 2 L 2 L R = 2 L + 2 3 r r r L 2 L L 2 3 R r 2 = + r r L L L R 2 2 r r r 2 R L 2r = + R L r
R= Where r is radius of a wire with circular cross-section. Then, R R = 1+ L L Since 2r r L L is defined as Poisson's ratio, v, we have
R K = R = 1+ L L L L

2r r L L L L

material prop .

For different metals, this quantity depends on the material properties, and on the details of the conduction mechanism. In general, metals have gage factors between 2 and 4. Now, since the stress times the area is equal to the force, and the fractional change in resistance is equal to the gage factor times the fractional change in length (the strain), and stress is Young's modulus times the strain, we have

or

So the fractional change in resistance of a strain gauge is proportional to the applied force, and is proportional to the gage factor divided by young's modulus for the material. Clearly, we would prefer to have a large change in resistance to simplify the design of the rest of a sensing instrument, so we generally try to choose small diameters, small young's modulus, and large gage factors when possible. The elastic limits of most materials are below 1%, so we are generally talking about resistance changes which are in the 1% - 0.001% range. Clearly, the measurement of such resistances is not trivial, and we often see resistance bridges designed to produce voltages which can be fed into amplification circuits.

Fig. 2: Thin Film Strain Gage For many years, there has been an industry associated with the fabrication and marketing of thin metal film strain gauges and the necessary tools and equipment for attaching these gauges and the wires to various mechanical structures. A photograph of a thin film strain gauge is shown in Fig. 2. This particular strain gauge consists of a metal wire patterned so that it is primarily sensitive to elongation in one direction. Strain gauges are available from several vendors, and literally hundreds of patterns of the metal film can be selected, with different patterns providing sensitivity to strain in particular directions. In recent years, much use has been made of the fact that doped silicon is a conductor which exhibits a gage factor which can be as large as 200, depending on the amount of doping. This creates an opportunity to make strain gages from silicon, and to use them to produce more sensitive devices than would be easy to make in any other material.

Fig. 3: Silicon Strain Gage Another aspect of the utility of silicon is that recent years have seen the development of a family of etching techniques which allow the fabrication of micromechanical structures from silicon wafers. Generally referred to as Silicon Micromachining, these techniques use the patterning and processing techniques of the electronics industry to define and produce micromechanical structures. An example company manufacturing these microdevices is MicroStrain.

Fig. 4: AFM Thermo-mechanical Data Storage. Micromachining can be used to fabricate piezoresistive cantilevers for a wide variety of applications. Past research between IBM and Stanford focused on the development of piezoresisitve cantilevers for a data storage applications. In this design, a 100 micron-long piezoresistive cantilever is dragged along a polycarbonate disk at 10 mm/s, bouncing up and down as it passes over sub-micron indentations in the surface of the disk. This idea is essentially a high-performance phonograph needle (for those of you old enough to remember...) The devices shown in figure 4 illustrate cantilevers developed for this data storage application. A great deal may be said about these techniques, but for now, we simply state that these techniques are capable of producing diaphragms and cantilevers of silicon with thickness of microns and lateral dimensions of hundreds of microns up to millimeters (see Fig. 3). The mechanical properties of these structures are exactly what we would expect from the bulk mechanical characteristics of silicon. Since these microstructures can have sensitive strain gauges embedded in them, it is easy to see that a number of useful sensing devices can be built. Particular examples include strain gaugebased pressure sensors, where an array of strain gauges can be positioned around the perimeter of a thin diaphragm, and be connected into a bridge configuration to automatically cancel out other noise and drift signals from the gauges. As we discuss the particular sensing devices, we'll see many examples of strain gauge-based microsensors. Another issue associated with strain gauges is the accuracy of the resistance measurement. Generally, accuracy would be improved by using larger currents and producing larger voltage changes. However, the practical limit to the amount of current that may be used comes about due to power dissipation in the resistive element. For this reason, the technologies for bonding thin film strain gauges has been optimized to maximize the thermal conduction from the thin film to the substrate. Improving the thermal conductance enables the use of more current in the measurement. Many strain gauges, and particularly doped silicon strain gauges, are sensitive to temperature changes. In some cases, this is a useful effect - especially if your application also needs to measure temperature. Generally, this is not the case, so it is necessary to compensate for this sensitivity. The easiest way to do this is to fabricate reference resistors from the same material,

and locate them so that they do not sense the strain signal. A bridge configuration can be easily arranged to retain the strain sensitivity while canceling the temperature sensitivity of an array of strain gauges. Such arrangements are very important, and easily produced, so they are very common. So, the applications of strain gauges are in sensors where medium-to-large amounts of strain are expected to occur (0.001% - 1%), where very low-cost devices are needed, where miniature silicon devices are necessary, and where signals are expected at frequencies from DC to a few kHz. The frequency limitation comes about because the bonding configuration of these devices generally leads to large stray capacitance, which tends to filter out rapidly varying signals.

Example Calculation: Piezoresistive Cantilever

Figure 5: Piezoresistive Cantilever The calculation of the sensitivity of a piezoresistive cantilever is presented here to provide an example of the strain gauge calculations. As shown in Figure 5, we use a piezoresistive cantilever to sense variations in the shape of a surface which is passed beneath. This technique has been demonstrated as the Atomic Force Microscope (AFM) by several recent graduate students in the group of Cal Quate. In AFM, attractive forces between a sharp tip and a sample surface cause slight cantilever deflections. If the cantilever is thin enough, forces associated with atomic interactions between individual atoms can be measured. The figure is taken from the thesis of Marco Tortonese, in which fabrication and operation of an AFM based on piezoresistive cantilevers is described in detail.

The load-deflection relationship for a simple cantilever beam is

where

Here, L is the length, T is the thickness, and w is the width. Since F = kZ, we have stiffness :

For a deflection Z, the cantilever has an angle of deflection of approximately , and, therefore, a radius of curvature of approximately

The strain in the upper surface of the cantilever is caused by the difference in arc length for the upper and lower surfaces.

The strain is given by

For a typical AFM cantilever (as shown in Fig. 4), we have parameters T = 4m, L = 100 m, w = 4 m, E = 2 x 10^11 N/m2, and F = 10-7 N. Therefore,

Since doped silicon has a gage factor of about 100, we would expect a change in resistance dR/R of 0.03% for this example. In fact, the cantilever does not take on a circular deflection, and the strain is largely concentrated at the base. If we place our strain gauge at the base, we can expect a strain enhancement of order 5-10x, thereby increasing the resistance change. With a decent circuit it is possible to measure resistance changes as small as 1 part in 10^6, so this is indeed a reasonable measurement. It is not simple, but it is possible. In many cases in AFM, forces as small as 10^-10 N are measured, which requires a careful electrical circuit design. It is this difficulty which allowed this to be the topic of a PhD thesis at Stanford University.

Lecture #4

Basic Intent
This lecture is intended to overview basic techniques for sensing temperature, and study some product examples. Following that, some techniques for the measurement of flow will be briefly highlighted.

Thermometers
There are a number of well-known historical technologies for the measurement of temperature. Everyone is familiar with the mercury thermometer, in which a reservoir of mercury is sealed in a glass container under vacuum.

Figure. 1: Filling of a Mercury Thermometer

When the reservoir is heated, the mercury expands, rising through a long thin column, upon which a graded ruler has been etched. What sort of sensitivity can be expected for such a system?

Well, the thermal expansion coefficient of mercury is well known to be about 30 PPM/K. If we assume that dimensions of the container do not change appreciably, (Is this a good assumption?) then the mercury in the column expands linearly with temperature. For a typical thermometer with a bulb at the bottom, acting as a reservoir for the mercury

Where V is the volume of the mercury in the reservoir, V is the change in volume, T is the temperature {Kelvin, K}, and T is the change in temperature. When the temperature rises, the change in volume of the mercury causes it to rise into the column (radius = R) by L.

Therefore,

If we want 1 mm/K at room temperature, and we have a reservoir volume of 0.1 cm3, we need :

Clearly, the sensitivity depends very strongly on the diameter of the column. Historically, makers of thermometer tubes worked very hard to control column diameter. Nevertheless, it was important to calibrate each thermometer with an ice point and a boiling water reference. Other traditional techniques for temperature measurement based on thermal expansion are very popular even today. For example, a great many home thermostats still rely on differential expansion in a bimorph to close a switch. Also, toaster ovens that are a few years old still feature bimetal temperature switches to operate the timing feature of the oven. Bimetal switches are fairly inexpensive, can operate reliably for many cycles, and may still be the correct choice for temperature sensing applications. In many cases, bimetal temperature switches are not accurate enough, or do not allow operation over a broad temperature range. For high temperature applications, thermocouple thermometers are often used. The thermocouple is an interesting device from a physics standpoint, but its operation can be easily understood from a thermal model. All metal wires may be considered as tubes filled with a fluid of electrons. The electrons are more or less free to move about in the tube, and certainly move in a preferential direction when a voltage is applied (voltage acts like a pressure). If the density of the electrons is non-uniform, the non-uniform charge distribution exerts a force on the electrons, tending to even them out. This charge effect is similar to a finite compressibility in a normal fluid. Now, suppose a tube filled with electrons is heated at one end. The effect of heat is to increase the average thermal velocity (because 1/2mv2 is proportional to 1/2kBT) of the heated electrons. In this situation, the effect of heat is to increase the average velocity of the electrons on the heated end of the wire. Because of this velocity increase, `warm' electrons will leave the heated end of the wire faster than they can be replenished by `cold' electrons from the other end. This

will lead to a non-uniform density distribution which gradually increases until the electrostatic pressure (because of the charges) is large enough to balance it. If one attached the leads of a voltmeter to this wire, there would be an excess of electrons at the cold end, and a net voltage difference across the wire. Since the cold end has an excess of electrons, it is repulsive to additional electrons, and is therefore at a `low' voltage with respect to positive charges. The amount of voltage difference is approximately proportional to temperature, and depends on materials properties (e.g. electron mobility and thermal conductivity). Tables of thermocouple properties are widely published. Now, it is generally inconvenient to attach the leads of a voltmeter across the wire, especially since one end is at the point of temperature measurement. Instead, it is common to use a pair of wires made from 2 different materials. The wires are joined at the end which is to be the point of temperature measurement (the `junction'), and the voltage is measured across the other two ends. If the materials have different thermocouple effects, there will be a voltage difference across those wires. Tables of the thermocouple voltages for a set of standard pairs are also widely published. A good pair of materials for a thermocouple includes any materials which can each survive the environment of the measurement, do not react with each other, and have suitably different thermocouple coefficients. The voltages generated by such effects are fairly small. A good thermocouple exhibits a voltage signal of only 10 uV/Kelvin. Therefore, accurate measurements of small temperature changes require very well-designed electronics. For measurements which require accuracy of +/- 10 K, and need to be carried out at temperatures near 1000K, thermocouples are definitely the way to go. There are also many examples of thermometers based on resistance changes. We have already seen several examples of resistance changes which are considered to be a problem in measurements of other quantities (piezoresistive strain gauges, for example).

Figure 2. RTD probes. Platinum wires are commonly used for resistance thermometry. Even though platinum is quite expensive, it is favored for these applications for several very good reasons. Do you know what they are? For narrower temperature ranges, a large assortment of resistance thermometers exists. Ordinary carbon resistors can be used, but companies such as Thermometrics offer a very broad collection of resistance thermometers in different shapes, sizes and characteristics. An important parameter for a resistance thermometer is the temperature coefficient, generally denoted as Alpha. The temperature coefficient is defined as the fractional change in resistance per unit change in temperature. This definition is convenient because of the way it emerges from expressions which are associated with voltage divider-based resistance measurement.

Figure. 3: Negative Temperature Coefficient (NTC) thermistor.

Figure. 4: Positive Temperature Coefficient (PTC) thermistor. For a voltage divider with a thermistor of average resistance R1 and a load resistor with resistance RL, the voltage at the output is given by

Now, if RL>>R1, we have

If the temperature of R1 changes by 1 K, the resistance changes by (Alpha R1), so the voltage changes by Alpha(Vin R1/RL). The definition of the temperature coefficient as a fractional change

in resistance per unit change in temperature produces a result in which the fractional change in voltage per unit change in temperature is given by alpha as well. When using a thermistor to measure small temperature changes, noise can impose a limitation. For example, all resistive elements exhibit voltage noise known as Johnson noise with density given by

The resolution of a thermistor limited by Johnson noise would be given by

We can see from this that the resolution is improved by reducing the temperature, the bandwidth, and the load resistor, and by increasing the temperature coefficient and the bias voltage, Vin. Of all of these parameters, it may be easiest to increase the bias voltage. However, there is a problem to watch out for. The bias in this system also causes power to be dissipated in the sense resistor. We may assume that the sense resistor is attached to the object of interest with a finite thermal conductance G. There will be a temperature difference between the sense resistor and the object of interest given by

So, we can see that increasing the bias voltage can easily lead to problems. Since thermal conductance between resistance thermometers and the surface are generally of order 10-2 to 1 W/K, we see that power dissipation in the mW range can cause substantial errors. Because of this, manufacturers of resistance thermometers generally provide extensive information on electrical measurement conditions to be used with their products. Thermometrics, for instance, provides a lengthy (35 page) tutorial on their products in their standard catalog. In addition, there are a number of solid-state thermometer technologies which have recently become commercially feasible. National Semiconductor, among other companies, is marketing a family of thermometers based on diodes. The leakage current in a forward-biased diode is generated by thermal excitation of electrons over an energy barrier. At temperatures near room temperature, the voltage across a current-biased diode is small and linearly dependent on temperature. National Semiconductor makes a series of such devices which produce very nice voltage outputs over a decent range, and at low cost. These devices look like 3-terminal transistors, require 5V and ground, and produce an easily measured voltage output.

Finally, a number of commercially-available `non-contact' temperature sensors are available which are based on measurement of infrared radiation. This approach can be expensive, and can suffer from some calibration errors. The infrared radiation from any object is given by

If a positive difference in temperature exists between the object and sensor, a net power will be incident upon the sensor. For a slightly higher than room temperature object (+10K), this comes to ~5 mW/cm2. This isn't much power, but is detectable with a good sensor. The difficulty lies in that the emissivity may be very difficult to know with any precision, since it depends critically on surface finish, thin coatings of residue, and may depend on temperature. Also, the radiation emitted by the object of interest may be absorbed by the atmosphere. The atmosphere is reasonably transparent in some regions of the IR (3-5 microns, 8-12 microns), and basically opaque in others (6-7 microns, 14-20 microns). Therefore, all of the radiation emitted by the surface cannot be expected to arrive at the detector. Finally, the radiation emitted by the surface is distributed more or less uniformly over the entire field of view. Therefore, the amount available for detection is substantially less, unless the instrument being used has a very large primary optical element. All things considered, this is a difficult approach to use, and can be expensive. Good IR detectors must be cooled, required an expensive package and operating system. Nevertheless, these products are becoming popular for manufacturing applications, and device and packaging costs can be expected to continue falling as the market grows. Also recent progress has made uncooled IR detectors seem to be promising for a number of consumer applications (e.g., night driver assistance). For any thermometer, there are issues associated with measurement of changing temperature that need to be considered. Consider a general situation in which a thermometer is attached to an object with a thermal conductance of G (W/K). The thermometer is a physical object, and has a heat capacity C (J/K). Assume that some power, P, is being applied to the thermometer (perhaps from bias currents). Finally, assume that the temperature of the object is oscillating in time according to

where is the frequency of oscillation. We assume that the temperature of the thermometer is also oscillating, and that the frequencies are the same, so the temperature of the thermometer may be expressed

From energy balance, we know that the energy into the thermometer equals the change in energy of the thermometer:

We can plug in our expressions for To and Ts, and we have

We may separate the static and oscillating parts to give equations

So, we should expect a finite temperature offset due to the bias power (P/G), and an oscillation amplitude which varies with frequency. At low frequency, Ts2 is nearly equal to To2, as we would hope for. At higher frequencies, the thermal time constant associated with the heat capacity of the thermometer can cause a reduce oscillation and a phase lag. These issues are important to keep in mind for measurements of time-varying temperatures. Parts after this may be repeated: We build a resistance measuring circuit in the form of a voltage divider with a load resistor (RL) in series with the thermistor (Rs). As always, the voltage at the output is given by

The resistance of the thermistor is

So,

Now, this is clearly a mess, and it is hard to see what the actual response function will look like. To make some sense of it, we will do a Taylor series expansion of Vout(Rs) = Vin Rs/(Rs + RL) about dT = 0, and extract the offset, the slope and the nonlinearity of this response.

As mentioned in an earlier lecture, the Taylor series expansion is :

For our expression for Vout(Rs), we have :

Since Rs - Ro = Ro dT, we have

The sensitivity is defined as the derivative of the voltage with respect to the temperature evaluated at dT = 0, so we have

The linearity of this system is essentially the maximum fractional error between the true response and the linear response. A good approximation to this quantity may be calculated by taking the ratio of the quadratic term in the expansion to the linear term.

which simplifies to

We see that the linearity is improved (by reducing it) by taking RL>>Ro, and by keeping either dT or small. Simply put, if we need a certain dT and a certain linearity, we can select the thermometer and the load resistance to meet our needs.

Lecture #5
Basic Intent
To cover the basic operating principles of flow sensors.

Flow Sensors
There are three basic approaches to the measurement of flow. The first of these categories involves the use of thermal effects to measure fluid motion. In general, this approach uses a heat source to deposit heat into the fluid, and a thermometer to measure the temperature of the fluid. If the heat source is upstream of the sensor, flow increases heat transport and causes the sensor temperature to increase. Another possible arrangement is to heat a thermistor with a fixed power, and measure its temperature. In this case, fluid flow acts to cool the thermometer.

These approaches can be analyzed to predict the sensitivity of such a system. In general, the fluid flow around a real physical system is very difficult to model, and the resulting performance generally needs to be calibrated. In addition, non-linear effects in turbulent flow can cause severe nonlinearities. Nevertheless, such an arrangement is easy to assemble, and inexpensive thermistors enable such systems to be produced at low cost. When properly calibrated, such systems are capable of excellent performance and are in wide use in industry. A slightly more complicated approach relies on Bernoulli's Equation, which is:

This roughly states that the Pressure + the kinetic energy density + the gravitational potential energy density is a constant throughout a fluid. This principle is applied by measuring pressure at a pair of points in a fluid. When water flows through a pipe with a varying diameter, the total flow rate in each region is a constant (since the fluid must all get through the tube). Therefore, changes in tube diameter are compensated for by changes in fluid velocity. By measuring the pressure in regions with different diameter, it is possible to measure fluid velocity. Now, the textbook section (P. 362, 3rd edition) that describes this technique is not completely accurate. Whereas the drawings and the text discuss measuring pressure drop across a flow impedance (much like a resistor - relies on dissipation in the impedance to produce a pressure drop), the equations all describe a Bernoulli principle, in which a velocity change is being measured. It is important to understand this distinction:

Bernoulli's equation techniques only work if the flow is not turbulent (non-dissipative). Then the pressure difference is between the wide and narrow regions. (see equations in text) Pressure drop techniques only work if the flow is turbulent (dissipative). In that case, the pressure difference is across the dissipative region. (see figures and words in text)

Fig. 1: Capacitive Pressure Sensor Fig. 1 (Fig. 11.14 in the text, 3rd edition) does show a proper configuration for a Bernoulli flow sensor. Here, flow induces a pressure difference across a silicon diaphragm which separates a wide channel from a narrow channel. The last technique for flow measurement is based on measurement of Doppler effects in sound transport. Since sound is carried by pressure waves in a medium (the fluid), its transport laterally across a channel is affected by the motion of the fluid. It is possible to measure the change in sound frequency due to fluid motion (direct Doppler effect), or listen for changes in the travel time from transmitter to receiver. High sensitivity techniques generally measure frequency shifts, since excellent accuracy may be obtained by use of analog or digital signal processing techniques to measure small frequency shifts. To review, flow may be measured by thermal, Bernoulli, or Doppler techniques. Thermal techniques are generally least accurate and least expensive, Bernoulli techniques can work well, but are accurate only for non-turbulent flow, and Doppler techniques are potentially most accurate, but are also generally most expensive. Some possible pictures to be used in the development of these pages:

Fig. : Float-type flow meter.

Fig. : Turbine flow meter.

Fig. : Nozzle flow meter in duct.

Fig. : Orifice, flow nozzle, and venturi flow meters.

Fig. : U-tube manometer.

Fig. : Wall pressure tap for static pressure measurement.

Fig. 1: Static pressure probe.

Fig. :Simultaneous measurement of stagnation and static pressures.

Lecture #6
Basic Intent
This lecture is intended to offer an overview of the basic types of sensors used to detect optical and near infrared radiation. At the completion of this lecture, the student should be familiar with some of the more common types of photosensors, some of the principles which govern their performance, and some applications.

Photosensors
Detection of light is a basic need for everything from devices to plants and animals. In the case of animals, light detection systems are very highly specialized, and often operate very near to thermodynamic limits to detection. Device researchers have worked on techniques for light detection for many years, and have developed devices which offer excellent performance as well.

Clearly, a major sponsor of light detection device research has been the military. Devices for light detection are of fundamental importance throughout military technology, and the maturity

and widespread availability of inexpensive photosensors is a direct result of this DOD research investment over many years. Light is a quantum-mechanical phenomena. It comes in discrete particles called photons. Photons have a wavelength , a velocity , a frequency , energy

, and even a momentum . Among all of this, it is important to remember the relationship between energy and wavelength. In all cases, the energy of the photon determines how we detect it. Light detectors essentially may be broken into two categories. The so-called Quantum detectors all convert incoming radiation directly into an electron in a semiconductor device, and process the resulting current with electronic circuitry. The Thermal detectors simply absorb the energy and operate by measuring the change in temperature with a thermometer. We will start be examining the Quantum detectors, since they offer the best performance for detection of optical radiation. In all of the quantum detectors, the photon is absorbed and an electron is liberated in the structure with the energy of the photon. This process is very complicated, and we will not examine it in detail. It is important to recognize that semiconductors feature the basic property that electrons are allowed to exist only at certain energy levels. If the device being used to detect the radiation does not allow electrons with the energy of the incident photon, the photon will not be absorbed, and there will be no signal. On the other hand, if the photon carries an amount of energy which is `allowed' for an electron in the semiconductor, it can be absorbed. Once it is absorbed, the electron moves freely within the device, subject to electric fields (due to applied voltages) and other effects. Many such devices have a complicated `band structure' in which the allowed energies in the structure change with location in the device. On example of such a `band structure' is that offered by a p-n diode, such as shown in the textbook. In a diode, the p-n junction produces a step in the allowed energy levels, resulting in a direction in which currents flow easily and the opposite direction in which current flow is greatly reduced. A photo-diode is simply a diode, biased against its easy flow direction (`reverse-biased') so that the current is very low. In a photon is absorbed and an electron is freed, it may pass over the energy barrier if it possesses enough energy. In this respect, the photodiode only produces a current if the absorbed photon has more energy than that needed to traverse the p-n junction. Because of this effect, the p-n photodiode is said to have a `cutoff wavelength' - photons with wavelength less than the cutoff produce current and are detected, while photons with wavelength greater than the cutoff do not produce current and are not detected.

Fig. 1: Connection of a photodiode in a photovoltaic mode Photodiodes may be biased and operated in two basic modes: photovoltaic and photoconductive. In the photovoltaic mode, the diode is attached to a virtual ground preamplifier as shown in Fig. 1 (from the textbook), and the arrival of photons cause the generation of a voltage which is amplified by the op-amp. The primary feature of this approach is that there is no dc-bias across the diode, and so there is no basic leakage current across the diode aside from thermallygenerated currents. This configuration does suffer from slower response because the charge generated must charge the capacitance of the diode, causing an R-C delay.

Fig. 2: Photoconductive operating mode In the photoconductive mode, the diode is biased, and the current flowing across the diode is converted to a voltage (by a resistor), and amplified. A photoconductive circuit is shown in Fig. 2 (from the book). The primary advantage of this approach is that the applied bias decreases the effective capacitance of the diode (by widening the depletion region), and allows for faster response. Unfortunately, the dc bias also causes some leakage current, so detection of very small signals is compromised. In addition to making optical detectors from diodes, it is also possible to construct them from transistors. In this case, the `photocurrent' is deposited in the base of a bipolar junction transistor. When subjected to a collector-emitter bias (for npn), the current generated y the photons flows

from the base to the emitter, and a larger current is caused to flow from the collector to the emitter. For an average transistor, the collector-emitter current is between 10 and 100x larger than the photocurrent, so the phototransistor is fundamentally more sensitive than the diode.

Photodiodes and phototransistors are very widely available. Most semiconductor device manufacturers also offer photodiodes and transistors, so there are nearly 100 suppliers. More than 10 manufacturers specialize in photosensors. As a result, optimized photodiodes and transistors are available at very low cost. These devices are also available in packages which are designed for particular applications. For example, it is common to use a light emitting diode and a detector mounted in a pair so that passing objects can interrupt the optical beam between them. "Opto-interruptors" which consist of such emitter-detector pairs are available in a wide variety of configurations. "Proximity

detectors" which are situated side-by-side sense the presence of a reflecting surface by causing reflected light to strike the detector.

Fig. 3: Incremental Encoders Other applications of optical detector-emitter pairs include measurement of the rotation rate of electric motors. In this case, a disk is mounted on the shaft of the motor with a large number of slits cut through it. The detector emitter pair is mounted so that the slits cause an oscillation in the signal - and the rotary position can be determined by counting the peaks in signal. This is called an optical encoder, and it is widely used in electric motors, as shown in the Fig. 3.

Fig. 4: Spectral Response of an Infrared Photodiode Most phototransistors and photodiodes have their peak sensitivity in the near infrared (see Fig. 4; from the book). The peak sensitivity occurs near the cutoff wavelength (near 1 um) and extends to shorter wavelengths. The location of this peak sensitivity is due to the energy of the `band gap' in silicon, and is not easily adjusted.

Table 1: Band Gaps and Longest Wavelengths Photosensors can be made from other electronic materials with different band-gaps, as shown in Table 1 (from the book). None of these materials are as widely available as silicon, and costs for detectors made from InSb can be substantially higher. There is another important consideration to keep in mind when selecting photosensors. In addition to the photocarriers in the device, thermally-generated carriers can be produced. The distribution of energies generated by thermal processes is dependent on the thermodynamics of the device, and on the temperature. Because of this relationship, increasing the temperature causes an increase in the number of thermally generated carriers. Conversely, reducing the band gap of a room-temperature device will also cause an increase in the number of thermallygenerated carriers. Silicon detectors work well at room temperature, but heating to more than 100C starts to cause substantial increases in `dark current'. Detectors made from materials other

than silicon may offer increased cutoff wavelength, but may also require cooling below room temperature. In general, there is a nearly linear relationship between the maximum operating temperature and the cutoff energy for the detector. By selecting a material with a cutoff energy 1/5 that of silicon (such as InSb), it is necessary to cool the device to about 1/5 of the maximum operating temperature of silicon (cooling to 77K is optimal for InSb). This tradeoff between cutoff and operating temperature imposes severe cost issues for operation of devices at fairly long wavelengths.

Fig. 5: Operating Ranges for Some Infrared Detectors If cooling is affordable, a large selection of materials and devices with `engineered band-gaps' is available. The tremendous interest in devices with cutoff wavelengths near 10-20 um is a direct result of the DOD interest in infrared detectors for `night vision'. It turns out that the peak of the infrared spectrum for objects at room temperature is in this region, and so the maximum contrast in thermal detection is available by producing devices with sensitivity in this region. There is a simple relationship between the temperature of an infrared source and the peak wavelength of the blackbody spectrum.

where the wavelength is in microns, and the temperature is in Kelvin. So, for room temperature, the max. wavelength is near 10 microns.

Of the materials most studied, the clear winner is Mercury Cadmium Telluride (`MCT'). It may be formulated to have cutoff between 10 and 20 microns, and offers excellent properties for infrared detection. In particular, it offers low dark current, high absorptivity, and low carrier scattering. Unfortunately, it is difficult and expensive to manufacture. As should be expected for anything containing mercury, its fabrication process is an environmental nightmare, and the basic material is not compatible with electronics. As a result, it is `bump-bonded' onto silicon substrates for readout and signal processing. In addition, it must be operated at or below 77K, which imposes operational complications. A commercial imaging system based on MCT detector arrays generally costs near 100K. Most military applications (ballistic missiles, aircraft imaging systems, satellite systems) can afford this set of costs and complications, but commercial and civilian application are generally cost-constrained. Therefore, recent research activities have focused on other materials which might be less expensive to make and operate. InSb does not offer sensitivity in the 10-20 um region, but is more easily made than MCT, is electronics compatible, and can be operated near 100K. Research to extend the operation to higher temperatures is underway throughout academia and industry. Overall, the relationship between cutoff and operating temperature is pretty strict. MCT, which has been the focus of billions of dollars of materials research effort, has only been slightly extended to higher temperatures. There is not tremendous hope that InSb or other materials will benefit from a large change in operating requirements.

The other type of optical detector, the thermal detector, does offer some hope for this problem. Thermal detectors operate by absorbing the infrared radiation and measuring the temperature rise of the detector with a thermometer. Generally, the performance of thermal detectors is limited by the availability of sensitive and small heat capacity thermometers. An important advantage of thermal infrared detectors is due to the absence of any relationship between the wavelength of the absorbed radiation and the response of the detector. Any energy which is absorbed causes a response in the detector. Therefore, it is possible to use a thermal infrared detector at room temperature to detect radiation from room temperature blackbodies. However, it is important to note that if the conditions allow use of a quantum detector, such a detector will outperform a thermal detector by several orders of magnitude. Thermal detectors come into their own in situations which simply don't allow quantum detectors. Since the thermometer is mounted within the infrared detector structure, it is connected to a temperature reference by a finite thermal conductance. This finite conductance imposes dynamic constraints on the system behavior, and we may analyze the situation as follows: Assume we have a thermometer which is a thermistor with a temperature coefficient given by:

Fig. 6: Voltage Divider This thermistor is mounted in an electrical circuit with a load resistor which has a resistance of RL, and is biased by a dc voltage of Vin. The electrical circuit is shown in Fig. 6. As for all voltage dividers, we have:

Vout = Vin Vout = Vin

Rs Rs + R L Rs RL for R L >> Rs

The sensitivity of this system is given by:

Sensitivity =

Vout R s = Vin T RL

Fig. 7: Thermal Circuit However, we must consider the thermal characteristics of this system as well. In this case, we model the thermometer as a finite heat capacity attached to an object by a finite thermal conductance. Infrared power is deposited into the thermometer, causing the temperature of the thermometer to oscillate. This thermal situation may be modeled as a thermal circuit as shown in the figure below. By energy balance, the energy gained is equal to the change in energy of the thermometer:

Now, we assume that the power and the thermometer temperature oscillate:

Now, we insert these expressions into the energy balance equation, and we have:

We may take the constant and oscillating parts to be independent, and we have :

These reduce to:

So the sensitivity of this device to changes in infrared absorbed power is:

To improve the sensitivity, is it important to choose a thermometer with a large temperature coefficient and a small heat capacity. We can see from this expression that the response of the detector will have a simple 1-pole response, which is to say that it is frequency-independent below the cutoff frequency and decreases as 1/f above the cutoff. Its response is exactly the same as that of an electrical low-pass filter. There are several different infrared detectors that are based on this detection concept. In fact, almost every well-established thermometer has also been optimized as an infrared detector. In the next lecture, we will look at infrared detectors which are based on piezoelectric materials, resistance thermometers, the expansion of ideal gases, and on thermocouples.

Lecture #8
Basic Intent
This lecture will provide an overview of the use of capacitance measurements in sensors, and describe the fundamentals of accelerometers. At the end of the lecture, the student should be familiar with capacitance measuring systems, limiting factors of the measurement, and obtainable performance levels. Also the student should be familiar with the fundamentals of accelerometer operation, including the relationship between the mechanical characteristics of the sensor and its performance, and the limitations of the performance of most accelerometers.

Capacitive sensing

Fig. 1: Two Objects in Space First, what is capacitance? Any two metallic objects, positioned in space, can have voltage applied between them (Fig. 1). Depending on their separation and orientation, the amount of charge that must be applied to the two elements to establish a certain voltage level varies. The capacitance is defined as the ratio of the charge to the voltage for a given physical situation. If the capacitance is large, more charge is needed to establish a given voltage difference. In practice, capacitance between two objects can be measured experimentally. Predicting the capacitance between a pair of arbitrary objects is very complicated, because it is necessary to know the electric field throughout the space between the objects. The field distribution is affected by the charge distribution, which is, in turn, affected by the field distribution. Iterative analytical techniques are generally required, and accurate calculations are very costly.

Fig. 2: Parallel Plate Capacitor

However, for simple geometry, the capacitance may be estimated very easily. For a pair of parallel plates (Fig. 2), separated by a distance which is much smaller than the lateral dimensions of the plates. The capacitance is given by:

Clearly, the capacitance is increased by increasing the area of the electrodes or by reducing the separation between the plates. In addition, the capacitance can be increased by filling the gap with a medium with a large dielectric constant. If you were to try to make a capacitor by placing electrodes close together, you could take electrodes with area 1cm x 1cm, and the best you could hope for would be a separation no smaller than 1um. This would amount to a capacitance of about 1000 pF, which isn't very big but still about the biggest you would ever expect to find in a real sensor. More generally, capacitive sensors have capacitance closer to 100 pF or less.

Fig. 3: Change in Capacitance due to the Lateral Movement of the Plates Capacitance measurement is used to detect the motion of a sensor element. A simple example would involve the motion of one electrode in the plane parallel to the electrodes. Assume a pair of rectangular electrodes, as shown in figure 3, with dimensions Length (L) and Width (W). If one of the electrodes moves laterally a distance x, the capacitance changes from to

So, in this case, the capacitance signal changes linearly with displacement. To implement such a sensor, it is necessary to guarantee that the lateral motion does not also affect the separation between the electrodes, d. Also, this approach is difficult to use for measurement of very small lateral displacements, since a small lateral displacement would represent a very small fractional change in the capacitance of the sensor. For example, a 1um lateral displacement would cause only 10 PPM change in the capacitance of the capacitor geometry worked out earlier.

Lateral displacement capacitive transducers are useful for many applications, though. For example, rotary capacitive transducers were in wide use as tuning elements for AM/FM radios in recent years, and rotary variable capacitors are still available as adjustable circuit elements from electronic part suppliers.

Fig. 4: Change in Capacitance due to the Change in Plate Separation The most common use of capacitive detection for sensors is based on signals which are coupled to changes in the electrode separation, d. As shown in Fig. 4, consider a pair of electrodes with area A and separation d. A physical signal causes the separation to increase by a small quantity, . The capacitance changes from to

Now, the relationship between the displacement and change in capacitance isn't obviously linear, but for small changes in separation, we can approximate the capacitance through use of a Taylor series expansion. In general, any function, F(d) can be approximated in the neighborhood of some nominal value d(0) as follows :

For the expression above, this expansion takes the form :

So, for

, the capacitance change is linear with respect to displacement. The nonlinearity . As long as we aren't concerned about errors of this

appears as a correction term of order order, the signal is very nearly linear.

If the initial separation between the capacitive electrodes is a few microns, a 1% change in the capacitance would indicate a displacement of a few tens of nanometers, which is a very small deflection. Such a measurement should be considered well within the capabilities of capacitive sensing. Nice features associated with such a measurement include good sensitivity to very small deflections and no natural sensitivity to temperature. Precision fabrication is required, since it is necessary to produce electrodes which are very close to one another and highly parallel. So, capacitive sensing is generally used for situations in which a precision measurement is required, and the expense associated with the sensor fabrication is acceptable.

Fig. 5: Differential Capacitor One technique for reducing the effect of the nonlinearity relies on the use of a differential capacitor, as is shown in Fig. 5. In this case, the capacitance measuring circuit is set up to measure the difference between the two capacitances, which is expressed as:

In this case, the nonlinearity associated with the nonlinearity appears as a cubic term.

term is subtracted away, and the first

term, which should be substantially smaller than the squared

Why do we care so much about linearity in capacitive sensors? Generally, capacitive measuring techniques are only applied in cases where precision measurement is necessary - otherwise, a strain gauge based measurement would suffice. One example of such a measurement is the measurement of acceleration for inertial navigation applications. A common problem in navigation situations is due to vibrations of the vehicle. In inertial navigation, one is generally taking the output of an accelerometer, and integrating twice with respect to time to obtain displacement. Because of the nature of this integration, offset errors in the output of the accelerometer accumulate as errors in position as t^2. Therefore, inertial navigation applications are especially concerned about offset errors. If an accelerometer with a small nonlinearity in the form of a includes a vibration, there will be a displacement of the form the output of the sensor of the form term is used in a situation which . There will be a term in

. Note that this expression includes an oscillating term and a static term. Generally, this phenomenon is referred to as vibration rectification - the process of generating a dc offset signal from a vibration signal. As described above, inertial navigation is one application which is particularly concerned about such phenomena, and so cancellation of nonlinearities in capacitive sensing is very important for such applications. The textbook gives several good examples of capacitive sensing circuits and applications. The switched capacitor sensing circuit shown in the book is a particularly good example of the use of a square-wave oscillation and FET switches to sample and rectify a waveform in order to convert a capacitance to a voltage. Such circuits are becoming very common because it is a simple matter to design an entire circuit of this type as an integrated circuit on a single silicon chip. Among the kinds of sensors which can use capacitance measurement to detect physical signals are pressure sensors, accelerometers, position detectors, level sensors, and many others, all of which are shown as examples in the book. Several of these examples will be studied in more detail later in the course. In general, capacitance detection is a good way to measure displacement. If implemented carefully, very small displacements may be measured. Measurement of such small displacements requires fabrication of precise mechanical structures with small gaps between the electrodes of the transducer. In addition, well designed and very well-packaged circuitry is necessary to carry out precision measurements. As a result, capacitance detection is best suited to applications

which require better performance than can be obtained from a strain gauge, and where the added cost of the capacitance detection is allowed. Capacitive detection has the advantage that it is not directly sensitive to temperature. However, the output of a capacitive transducer is not immediately linear. If linearity is important, differential capacitance schemes are advisable.

Accelerometer overview
Accelerometers are devices that produce voltage signals in proportion to the acceleration experienced. There are several techniques for converting acceleration to an electrical signal. We will give an overview for the most general such technique, and then look briefly at a few others.

The most general approach to acceleration measurement is to take advantage of Newton's law, which states that any mass that undergoes an acceleration is responding to a force given by F = ma.

Fig. 6: A General Accelerometer The most general way to take advantage of this force is to suspend a mass on a linear spring from a frame which surrounds the mass, as shown in Fig. 6. When the frame is shaken, it begins to move, pulling the mass along with it. If the mass is to undergo the same acceleration as the frame, there needs to be a force exerted on the mass, which will lead to an elongation of the spring. We can use any of a number of displacement transducers (such as a capacitive transducer) to measure this deflection.

For the case shown in Fig. 6, the sum of the forces on the mass are equal to the acceleration of the mass (X is the position of the frame, x is the position of the mass):

We make assignments

and we have:

Now, since X is the position of the frame, we can impose an acceleration on this problem by forcing X to take the form . If we assume all the time varying quantities also oscillate, we need . Substituting these into the above equation, we have :

Canceling, assigning

, and rearranging gives :

2X0 k ib 2 m m A0 Z0 = k ib 2 m m A0 Z0 = ib 2 02 m b b = = 2 km 2m A0 Z0 = 2 02 i 2 2
Z0 =

Note: (damping ratio) is just another definition of a damping term.

Fig. 7: Amplitude Response of Vibration-measuring Instruments Things to note about this expression:
1. If b = 0 (no damping), this expression blows up at . This means that the signal at the resonance of an undamped accelerometer can lead to infinitely large signals. This is one of the reasons that accelerometer designers generally impose finite damping on the system. 2. If , this expression simplifies to :

In this case, the displacement of the mass is proportional to the acceleration of the frame. This is the response we would hope for from an accelerometer.
3. If , the expression simplifies to :

. This is the case for high frequency signals, during which the mass remains stationary, and the accelerometer frame shakes around it. In this case, the displacement between the mass and the frame is the same size as the motion of the frame. This mode of operation is

generally referred to as `seismometer mode'. Seismometers are instruments which attempt to measure ground motion, rather than ground acceleration.
So, the general accelerometer consists of a mass, a spring, and a displacement transducer. The overall performance of accelerometers is generally limited by the mechanical characteristics of the spring (linearity, dynamic range, cross-axis sensitivity) and the sensitivity of the displacement transducer.

Many different displacement transducers can be used in accelerometers. It is generally easy to design a mechanical system which is well enough behaved that the performance of the accelerometer is limited by the displacement transducer. Examples of displacement transducers which may be used include:

capacitive transducers strain gage transducers optical transducers (laser interference measurement) resistive transducers electromagnetic transducers

In each case, the transducer is configured to measure the displacement of the mass relative to the frame. How large are the displacements being measured? Suppose we want a device to measure 1 milli-g accelerations with a bandwidth of 20 kHz . Then, we need a displacement transducer capable of resolving displacements given by:

. This is a very small displacement! Such a device is not easily made, and measurements with this level of accuracy are difficult to carry out. We can see that the difficulty comes in large part from the very large bandwidth (20 kHz) requested from this device. If we were to reduce this request to 1 kHz, the required resolution would increase by a factor of 400, which would make the displacement detection problem much easier.

Conclusions
We have looked at the use of capacitance detection for use in physical sensors. For applications which require accurate measurement of small signals, capacitance detection is a good selection. The performance of a capacitive detection system is

obtained at a cost, and this cost is generally more than 10x higher than a device made with piezoelectric or piezoresistive technology.

We then looked at the design and operation of accelerometers. Depending on the capabilities of the detection system, it is possible to have miniature devices for measurement of signals from the milli-gs down into the micro-gs. The best of these sensors is based on capacitance detection of a very small displacement and is on the market for a few thousand dollars. On the other end of the market, there are devices available for a few tens of dollars based on piezo technologies. In the next lecture, we will look at a particular accelerometer which has recently entered the market, and examine some of the tradeoffs in its design, performance, and cost.

Lecture #10
Basic Intent
This lecture is intended to provide an overview of piezoelectric devices. Some examples are worked using this sensing technique.

Piezoelectricity
Piezoelectricity is the name of a phenomenon which sounds as if it might be similar to piezoresistivity, but there is really very little in common between these two. Piezoelectricity refers to a phenomenon in which forces applied to a segment of material lead to the appearance of electrical charge on the surfaces of the segment. The source of this phenomenon is the distribution of electric charges in the unit cell of a crystal. The textbook describes the example of the quartz crystal, in which forces applied along the x axis of the crystal lead to the appearance of positive and negative charges on opposite sides of the crystal along the z axis (Fig 1). The strain which is induced by the force leads to a physical displacement of the charge in the unit cell.

Fig 1. Piezoelectricity This polarization of the crystal leads to an accumulation of charge according to the following expression : Q (charge) = d F In fact, the force is a vector quantity, and the d (piezoelectric coefficient) is a 3x3 matrix. Forces along the x axis produce charges along the x, y, and z axes, with the charge along the x axis given by the d11 coefficient of the matrix, the charge along the y axis given by the d21 coefficient, and so on. Piezoelectric coefficients have been tabulated for several materials (primarily those materials with large coefficients), and a table of coefficients is in the textbook. The detailed properties of this phenomenon and the materials properties which are responsible for it are not of our concern in this course. We will need to perform calculations based on these materials and the piezoelectric effect, since this phenomena has been very useful in the development of sensors. Typical values of the piezoelectric charge coefficients are 1-100 pico-coulombs/N. As an example, assume we have a 1 cm x 1cm slab of 1 mm thick PZT material. A 1 N force is applied along the z axis, which is the 1mm dimension. What voltage appears across electrodes on the large surfaces?

The capacitance of this device is

If we want to produce a larger voltage, we need to reduce the capacitance of this structure. The easiest way to do this is to reduce the area. Out of curiosity, how much has the length of the crystal changed under this load of 1 N?

An interesting corollary effect is that this effect is reversible - which is to say that application of voltages results in dimensional changes of the crystal. The same coefficients are used when describing the conversion of a force to a charge as the other way of an electric field to a strain.

Fig 2. Piezo leakage One interesting effect to take note of is that piezoelectrics are not generally very good dielectrics. In particular, piezoelectric materials are somewhat leaky (Fig 2). This means that a charge placed on a pair of electrodes gradually leaks away. Because of this phenomenon, there is a time constant for the retention of a voltage on the piezoelectric after the application of a force. This time constant depends on the capacitance of the element, and the leakage resistance. Typical time constants are of order 1 sec. Because of this effect, piezoelectrics are not very useful for the detection of static quantities, such as the weight of an object.

Fig 3. Equivalent electrical circuit for piezomeasurement circuit. Rx = Resistance of piezo Cx = Capacitance of piezo Cc = Capacitance of cable Ca = Capacitance of amplifier circuit Ra = Resistance of amplifier circuit Another important aspect to the use of piezoelectrics is the fact that they are fabricated using a process which relies on the crystallization of the lattice in a particular arrangement. This is accomplished by heating the crystal to above the Curie temperature while applying voltages to the electrodes. If the crystal is ever heated to near the Curie temperature, it can become `depoled' which can result in a loss of piezoelectric sensitivity. For various materials, this Curie temperature can be as high as 600C or as low as 50C. The need to stay below this temperature can impose serious constraints on the applicability of these sensors. Overall, piezoelectric elements have several important advantages over other sensing mechanisms. First and foremost is the fact that the device generates its own voltage. Because of this, the sensor element does not need to have power applied to it in order to function. For applications where power consumption is a significant constraint, piezoelectric devices can be very valuable. In addition to this, the piezoelectric effect has some interesting scaling laws which suggest it is useful in small devices. The primary disadvantage of piezoelectric sensing is that it is inherently sensitive only to time varying signals. Many application require sensitivity to static quantities, and piezoelectric sensing simply does not work for such applications.

Fig 3. Useful frequency range Nevertheless, if you have a time-varying signal, you should give serious thought to the use of piezoelectric sensing elements. One technology for piezoelectric materials involves the use of poly-vinyl di-fluoride films which are treated during manufacture to have a piezoelectric coefficient. The primary advantage of this process is that the films can be made at extremely low cost, and they are popular for low-cost sensing applications. One company in particular has pioneered the development of this material AMP Inc.. In addition to the successful commercialization of a number of products based on these piezo films, they offer unmounted film elements which are suitable for use in the construction of simple test devices.

Piezoelectric Accelerometer
A final example is that of a piezoelectric accelerometer. In this case, we consider an accelerometer which consists of a 10 gm mass resting on a slab of piezoelectric material. Assume the piezo slab has dimensions of 1 square cm in area, and 1mm thick. It is made of PZT material with its z axis perpendicular to the large faces, which are coated with metal contacts. What voltage is expected? where is the resonance?

The voltage is given by:

an acceleration of 1 milli-g exerts a force given by Newton's law :

which would produce a voltage of ~10^-4 V. This is measurable. As we discussed in the piezoelectric sensor lecture, this situation can be improved in several ways. For instance, the piezoelectric element can have a smaller area and a larger thickness. The resonant frequency can be calculated (an exercise for the reader) and is of order 200 kHz for the above geometry. So this device can be good for applications which require sensitivity to very high frequency vibration signals. It has the added feature that it does not require excitation voltages. On the other hand it does require a good preamplifier.

Fig 4. Charge amplifying circuit In any case, piezoelectric accelerometers are on the market, and are primarily offered for vibration measurement. For moderate signals (milli-gs), fairly small devices with simple circuits are quite sufficient, so these devices can be in the 10-100 dollar range.

Fig 5. Commercial accelerometer