2.3 The fabrication process for a MEMS Pirani sensor
This section is fairly long, but is best read in one go. If you run out of time, reschedule your study to allow you to start again from here.
Thin layers of material are added to the surface by a variety of means, depending on the material to be deposited, and what is already on the wafer.
The sensor starts off, as so many microsensors do, with a silicon wafer, shown in cross section in Figure 3(a). This wafer is a few hundred micrometres thick (1 micrometre = 1 µm = 10−6 m), but because all the interesting stuff happens in a few layers typically less than a micrometre thick, the wafer's thickness is shown at a reduced scale relative to the layers that are put on it. A very important feature of the silicon wafer is that it has been sliced with high precision from a large single crystal of silicon. The planes on which atoms are most densely packed are precisely oriented with respect to the polished surface.
Deposition is the way in which material is added on to the structure: etching is the way unwanted parts of the new material are taken away. Just as there are several ways of depositing material according to what they are composed of, so there are different ways of etching them.
The first step is to deposit a layer, 100 nm (0.1 µm) thick, of polycrystalline silicon – Figure 3(b). It will not form any part of the resulting device – its function is to act as a temporary spacer that can be etched away during another process in the making of this device, which we'll come to later.
This layer is aptly called a sacrificial layer. The choice of its material is all to do with its behaviour during the manufacturing stages. Different materials are chosen for this role according to what other materials are present. Specifically, the sacrificial layer must be of a material that can be deposited, patterned and finally etched away using processes that do not harm any of the other materials present. It must also be robust enough to survive unscathed the processes used for the deposition and patterning of all the other materials laid down during its brief stay on the wafer.
The next step is to define the places where this layer is to perform its function. This is done by photolithographic patterning; in this example, a long thin rectangle (25 µm × 1000 µm) of polysilicon remains – Figure 3(c).
Now the whole wafer is coated again, this time with a layer of silicon nitride – Figure 3(d). This is the first layer to have any function in the final device. It has two purposes: one mechanical, the other electrical. Its mechanical role is that of a strong support for the conducting track that will form the heater. Its electrical role is that of an insulator. It will prevent current flowing from the conducting track to the underlying silicon wafer, which would short out the whole device. This is a good choice of material because it is a superb insulator, having a resistivity of about 1014 Ω m, and a dielectric strength (see Box 2) of about 108 V m−1.
Box 2 Dielectric strength
The dielectric strength of (an insulating) material is defined as the maximum electric field it can stand without breaking down – that is to say, without it changing in such a way that it starts to conduct electricity. In catastrophic breakdown, once this happens, runaway heating caused by the current flowing results in the material reaching such a high temperature that it decomposes.
The strength of an electric field is given in units of volts per metre; so a gap of 0.5 m between two conducting plates, where one plate is at a voltage of 1000 V relative to the other, will have an electric field strength of 2000 V m−1 across it. If the gap were reduced to 0.5 mm, a voltage of only 1 V between the plates would produce an electric field of equal strength in the gap. So, a 1 µm thick layer of silicon nitride with a dielectric strength of 108 V m−1 will be able to withstand 100 V across its thickness before it breaks down.
One of the interesting things about the effects of scale is that the dielectric strength of materials generally increases as their thickness decreases. For gases this effect can be dramatic, and the reason is fairly easy to visualise: as the gap between the electrodes is reduced, eventually the point will be reached when it is of the same order of magnitude as the average space between the molecules of the gas. Electrical breakdown of gases is characterised by the generation of a spark – the visible evidence of an avalanche of electrons, driven on by the electric field, growing ever larger as it knocks more and more electrons off atoms in its path. This can occur only if there are enough gas atoms in the way of each electron to produce a multiplying effect before it reaches the other electrode.
In solid materials, electrical breakdown is more akin to mechanical failure, in that it tends to occur first where the electrical or mechanical stress is intensified locally owing to a defect in the structure of the material. This then causes the defect to grow, which then further increases the stress, and so on in another kind of runaway effect. If the dimensions of a piece of material are reduced, so is the probability of there being a defect above a given size within that piece of material; hence the material's overall ability to resist the imposed stress, whether it be in the form of a mechanical tension or an electric field, will be increased.
Dielectric breakdown is not always catastrophic. In defect-free thin films, a sufficiently strong electric field can cause the material to lose grip of some of its electrons, allowing it to conduct without the avalanching associated with catastrophic breakdown. When the field is reduced in strength, the material returns to its insulating condition.
Mechanically, too, silicon nitride is outstanding. In thin-film form on a silicon substrate, it can be made to have a slight tensile intrinsic stress (see Box 3), which is a good thing for suspended structures such as this (think of a guitar string – a slight tension keeps it firmly in a well-defined position relative to the rest of the guitar).
Box 3 Intrinsic stress
Intrinsic stress (which is called residual, frozen-in, or internal stress by mechanical engineers) is ever-present in thin-film materials, and it usually has a significant effect – for instance by making the film tend to peel off, or craze, or by causing curvature, or even buckling, of a structure. It can even affect the film's electrical and chemical properties. It can be either compressive (the film wants to expand laterally but is prevented from doing so by the substrate to which it is attached), or tensile, or a mixture, varying from tensile on one side to compressive on the other. It is often possible to exercise some control over the magnitude and even the sign of this stress, by adjusting the conditions under which the film is deposited.
Figure 4 shows how the residual stress in the silicon nitride depends on the ratio of the two gases used in the process. Increasing the proportion of dichlorosilane (SiH2Cl2) relative to the ammonia (NH3) in the gas mixture causes the nitride film to become silicon-rich. At a ratio corresponding to the lowest residual stress, the film is no longer what is known as stoichiometric silicon nitride, which has the formula Si3N4, but something more like Si5N4. (The word stoichiometric comes from the Greek stoicheio – element, and metria – measuring.) In other words, the ratio of silicon to nitrogen in the material has been shifted a little, so that there's more silicon present than the chemistry says is the preferred form. In addition to this, small amounts of hydrogen – up to 8 atom per cent – find themselves incorporated in the film, making the composition approximately Si5N4H.
With a tensile strength of about 6 GPa, the bridge or beam that this material eventually forms will be extremely rugged, much more so than the metal. Another good property of silicon nitride in this context is its low thermal conductivity (about 3 W m−1 K−1 – compared with about 1.4 W m−1 K−1 for window glass and about 240 W−1 K−1 for aluminium). Combined with the tiny cross-sectional area, the flow of heat by conduction down the bridge is kept small relative to the cooling effect of the gas – thus increasing the sensitivity of the sensor.