Just some notes for my upcoming exam.

“Creep” = drift in output under a steady, applied load. Silicon is highly valued in mechanical microsensor design because it acts as an almost perfect spring and exhibits virtually zero creep or hysteresis.

Noise = random errors. Some are systematic (we can account for them), and some are not (just random).

2. Micromachining

  • Silicon is a really strong material, stronger than steel.
    • It doesn’t deform, but it breaks when it’s overloaded (it’s “brittle”).
  • Monocrystalline silicon is an almost perfect spring, showing no hysteresis or creep, i.e., ideal material for sensors
  • Photolithography is essentially a 2D process, making true 3D mechanical structures difficult to create.

3. Strain gauges and capacitive sensors

Core transducer concepts

  • Modulator Type Transducers:
    • no energy flow from one physical domain to the other;
    • the blank energy flow in the receiving domain is modulated by information from the other domain
    • e.g., resistive strain gauges
  • Generator Type Transducers:
    • transduction relates to an energy flow from one domain to another;
    • energy can be stored in the transducer.
    • For these, the energy function is key to analyzing them.
    • e.g., capacitive, magnetic, and piezoelectric transducers

Piezoresistive Strain Gauges

Strain gauges are metal or semiconductor resistors bonded to a deforming structure, where deformation induces changes in resistance.

Piezoresistive vs. Geometric Effect: for semiconductor strain gauges, the piezoresistive effect is about 100 times larger than the purely geometric effect.

Strain gauges are typically connected in a Wheatstone bridge because it reduces temperature dependence and provides an output voltage directly proportional to the change in resistance

Stress and strain calculus:

  • Stress is force per unit area
  • Strain is relative elongation
  • is Young’s Modulus.
  • The spring constant is

How to find the neutral line ?

Resistance is dependent directly to length and inverse to area.

Piezoresistivity: semincoductor strain gauges. Greek = to press. The silicon adds a 100x larger effect than metal (geometric effect).

Capacitive Sensors

Capacitance of parallel plates is direct to area, and inverse to distance between the plates (the electrodes).

The displacement on the horizontal causes the overlap to increase on C2 and decrease on C1, so that’s how we sense force in the inplane direction.

Some examples are microphones. something to do with the pressure change. Another one is the accelerometer.

Where the laser beam is, the top part is the mass with each finger attached to the mass. And each finger is in-between two fixed electrodes. If the red mass moves to the right, the capacitance of the green fingers increases, and the capacitance of the blue fingers decreases, since the distance is larger.

Capacitors are used for both actuation and detection. Example: Gyroscope

Core transducer concepts

  • Modulator Type Transducers:
    • no energy flow from one physical domain to the other;
    • the blank energy flow in the receiving domain is modulated by information from the other domain
    • e.g., resistive strain gauges
  • Generator Type Transducers:
    • transduction relates to an energy flow from one domain to another;
    • energy can be stored in the transducer. For these, the energy function is key to analyzing them.
    • e.g., capacitive, magnetic, and piezoelectric transducers

With this one, think that charge in electrical domain corresponds to displacement in the mechanical domain. And the same for voltage and force.

Again remember the transverse comb structure and the C = eps Area(A) / Displacement(x)

And the lateral comb.

piezoresistive sensors are more sensitive than metal sensors. The relative change in capacitance / resistance drives that.

Force and pressure sensors

Optical readout: the LED in the center transmits the light coming from the reflective layer to the photodiodes. Therefore, if the surface is pressed upon, then the deformation will be caught by the reflective layer, and the photodiodes will catch that deformation.

Touch-mode sensor concept: the membrane actually touches the electrode, and tests have shown that we get a fairly linear characteristic of the Capacitance in relation to Pressure. Useful for example when you have car tyres and you don’t want to measure the small pressures, but the tyre pressure around the values that we have.

With pressure sensors, the pressure is mostly felt by the resonator, as the ends want to rotate away from each other. The resonators are those small thin beams that connect one ridge to the other. Think also of a string of a guitar — when we apply pressure, it vibrates.

For thermal pressure sensors, the heat flow from the heater to the silicon substrate depends on the distance. And the temperature of the heater is a measure for the pressure. On the triangle graph, if there’s no pressure, the measurements to the left and right should be equal. As soon as the pressure starts to come from one side, the air flow moves and so we can detect from which side and how much.

Acceleration Sensors

Force feedback is applied so the mass is stable, at equilibrium. For example, if the mass moves and it enters oscillations, it will take time until it stops. So apparently we compensate with a closed loop.

  • position sensing convert displacement to voltage
  • the compensator reduces the shift for higher frequencies, because those tend to make the system oscillate.

It’s used in this configuration for 3 reasons:

  1. Differential change in capacitance from which we can obtain the position of the mass.
  2. We can reduce the spring constant K to increase sensitivity ⇒ we get more displacement for a certain acceleration
  3. We can apply bias voltages to the static fingers so we can apply a feedback force that really drives the mass back to the centre position.

The mass fluctuates on the crystal ⇒ electrical voltage.

Piezoelectric accelerometers have reduced sensitivity for lower frequencies.

If the deformation remains in the crystal, that causes reduction in sensitivity below a certain frequency i.e. leakage.

this counters the seismic noise.

Reducing the spring constant K increases the sensitivity ⇒ we get a larger displacement for a certain acceleration

the hot bubble of air moves, and the two temperature sensors TC will understand from where. It is not that accurate, performance not that good. It is almost impossible to damage and accelerometer like this.

Piezoresistive readout is not that sensitive, so use it when you don’t need high precision.

However, for high precision accelerometers, we find electron tunneling readout. The electrons can travel between the very narrow gap between the tip and proof mass. Sub nanometer displacement of the proof mass can be detected. Huge temp dependence + low freq drift make it not very usable.

The magnetic field between the two electrodes keeps the mass floating. Works only in theory. High noise.

Gyroscopes

omega = angular velocity, v = linear velocity of the ball.

The farther away we are from the center of rotation, a higher velocity we need to stay in place.

In an angular rate sensor like this, we need a moving structure that moves away and towards the rotation axis. Then it has to adjust the velocity ⇒ the coriolis acc.

It’s always at least 2 masses that we need — we need to distinguish the rotation from accelerations. Only one mass — the coriolis motion could be mistaken for acceleration. They vibrate in opposite directions, then linear accelerations will have the same effect on both of them, so the Coriolis will have an opposite effect. the comb structures are used to electrostatically apply a force ⇒ we can generate the motion in the x direction, and if there’s a angular velocity, we get the motion in the y direction.

If I understood correctly: in the case above the Force is applied from the z direction, and the masses move away and towards each other on the x axis. That leaves the linear velocity on the y axis.

  • The Drive Force (from the comb structures) is applied along the X-axis, causing the masses to move towards and away from each other.
  • The input Angular Velocity / Rotation () is around the Z-axis.
  • This specific combination results in the Coriolis Force pushing the masses along the Y-axis, which is what the sensor measures.

How to increase the sense amplitude?

  • Increase the drive amplitude ()
  • Match the resonance frequency of the sense mode to the drive frequency
    • Gain equal to
    • Bandwidth reduced to

Here the idea is as follows: You have the second order system on the left which you are trying to keep at its resonant frequency amplitude (the drive force causes it to vibrate). Then, the velocity together with the angular velocity () activates the second order system.

There are 2 possibilities:

  1. the resonant frequency of the sense mode is much higher so we are in the flat area of the response. It introduces shift.
  2. we can tune the resonant frequency to be the exact same as the drive mode ⇒ we get the maximum amount of motion.

In the mode-split case we have 2/4 masses that are vibrating. In that case the drive and sense mode have completely different spring suspensions ⇒ hard to match the resonant frequencies. Sensitivity is low. Really small signals we need a large mass and low spring constant to get the highest detection amplitudes.

The ring type structure on the right deforms continuously in an electrical shape, since it’s symmetrical in all directions. Bandwidth is decreased since we divide by the quality vector, but we gain sensitivity.

with 2 masses we distinguish angular velocity & linear acc. With 4 masses we can distinguish between angular velocity & angular acc.

To sense tilt, this structure has an electrode underneath, and with the capacitive combs we see, we detect movements along that axis too.

Optical Sensors

Stimulated emission: 1 in, 2 out. As longs as the atom is excited. So we need to create a condition where most of the atoms are excited ⇒ produce double photons.

LASER = Light Amplification Stimulated Emission of Radiation. Exactly the process above.

The right mirror is partially reflective since we want some light to return into the system, and some light to get out of the system. The flashlamp excites it.

Compared to a LED, the laser is very concentrated in one direction. The LED spreads the light randomly. We can also see the wavelength. The ends of the Semiconductor Diode Laser are polished so the light can reflect perfectly inside.

With photodiodes, they start to conduct enormous amounts of current once we reach and they don’t allow reverse current flow unless we reach a Breakdown point.

You need to be with the curve in the green band or over it if you want to absorb light in the semiconductor. So that’s somewhere around 100 ?

A real photodiode lies below the ideal curve because not every absorbed photon produces a collected electron-hole pair, and not every incident photon is absorbed: there are reflection losses at the surface, absorption in inactive regions (e.g. the top contact or surface layers) where generated carriers recombine before being collected, and incomplete internal quantum efficiency, all of which reduce the responsivity below the ideal (100% quantum efficiency) value.

If we shine light in the middle, we expect to measure 4 identical currents. But If I move it towards one quadrant, then that signal gets much stronger than the other. So in electronics, we can steer/refocus the beam towards the middle.

Image sensors have millions of this. We can use them to very accurately steer motors.

A small camera chip looks at the surface and sees some kind of contrast. That is a direct measure of the speed of the mouse. The LED makes it easy to distinguish the surface.

In all three there’s a very small tip that goes over a surface and barely touching it. In the Atomic Force Microscope, we use a laser and a photodiode to readout the position of the sample tip.

Better than 1 resolution in the height measurement of that tip. Because only very small rotations of that sample result in large distances on the photodiode.

Never two consecutive changes. The electronic systems are not able to detect with that much accuracy. At least 2 0’s in between.

When we see a hill with a laser, no light will reflect because half of the reflective face is out phase. If there’s no hill, then all light gets reflected. For some reason the hill distance must be .

That’s how they make the light/dark connection. By measuring which one of the beams has the higher intensity, we will know that we need steering correction.

They use a cylindrical lens ⇒ the shape looks different depending on the case. And if the laser is out of focus, we know in which direction we need to steer from the shape. Again an application of the four-quadrant diode.

Flow Sensors and Controllers

Everything is stainless steel. The slits drop the pressure and increase the flow. The flow tube mesures flows up to 10mm/min. So we need to increase the flow rate. The split allows the 10mm/min to flow through the Sensor tube. We keep the split ratio always constant. That’s done with the laminar flow elements — tiny slits where each slit ~ 10 mm/min. So we can put additional 10 mm/min devices in paralel with the sensor tube ⇒ we can incease the flow rate by increasing the number of slits.

Using Wheatstone Bridge, we get an output differential voltage as a function of delta T (differential temperature). If we connect 10 thermocouples in series, we get 10 times the Seebeck coefficient. Thermocouples in series ⇒ Thermopiles. We use them to increase the amount of voltage per temperature. The nice thing with thermopiles is that we do not have offset issues.

In the middle of the chip we see a membrane — a thinner region. Bottom right we have a cross-sectional view. In the middle of the membrane we apply the heater and to the left and right we have the sensors. When we apply power to the heater, in the middle there’s the higher temp. And with flow we can detect the movement. We can see the thermopiles on the sides.

Because of the square root, it’s not nice and linear. The gas type always needs to be known in advance so calibration can take place beforehand. I believe they reached linearity in the end which wins speed.

Micromachined Coriolis mass flows sensors

Coriolis is only dependent on the mass flow.

This is the basic structure of a coriolis flow sensor. A loop structure which vibrates. When flow passes through, we get resulting vibration which is always a true measure of the mass flow.

Moreover, these sensors need to operate under vacuum conditions (packaging issues)

Again a loop, tiny tube. Let the tube vibrate. Two magnets ⇒ magnetic field. Applied alternating currents. When flow passes, the size of the tube will move according to the Coriolis force up and down. It’s proportional to the massflow, independent of the properties of the gas or liquid. Apparently there’s two capacitors and the time difference between the readings is the measure for the flow.

Does not need vacuum packaging. The quality vector of ~30 gives enough sensitivity.

has 2 functions

  1. actuate the flow tube at its resonance freq
  2. readout the two capacitors with charge amplif. to obtain the flow sensor signal.

Coriolis sensors respond faster than thermal devices. Again, coriolis only measures mass flow!

The higher the mass, the lower the freq. The square root eq from above holds this.

The thermoflow sensor has a wheatstone bridge. We apply power to these heaters ⇒ temp will rise. In the middle will be the highest and symetrically temp will drop (delta t ~ 0 when no flow). One side will decrease, one will increase, and that delta T will be the measure for the flow.

Combinations of flow and other sensors

With Surface Channel Technology we can integrate many sensors on one chip.

This enables us to calculate so much more. If we combine the eq. for coriolis and thermal sensor, we can get heat capacity. Coriolis + Pressure ⇒ viscosity.

With 4 types of info, we can solve 4 eq with 4 unknowns. We can solve how much of each gas/liquid is in the mixture.

Wobbe index has strong correlation with viscosity from pressure+density+coriolis. From viscosity we can say something about the percentage of the inputs.

Noise

Johnson

averaging out decreases the noise. The bandwidth term is the reason — averaging out results in surpressing the high frequencies.

Shot noise occurs because the intervals of the individual photons are unevenly distributed.

It scales linearly with the square root of I_dc. Also white noise (no frequency). This noise is not dependent on temperature. So changing temperature makes it clear which one it is (Johnson or shot).

Some electrons block others. And so we see these impulses.

At high frequencies, it drops down with 1/f^2. At low freq it looks like white noise.