Click here to go to our main page on semiconductor wafer processing
Click here to go to our main page on MMICs
Click here to go to our page on solid-state switches (for info on switch FETs)
Click here to go to our page on semiconductor technology tradeoffs
Click here to go to our page on curve tracer measurements
Welcome to the best web page on microwave FETs! This tutorial will provide an "outstanding understanding" of FETs as they are applied in microwave engineering; all of this information is directly applicable to understanding monolithic microwave integrated circuits (MMICs).
New for September 2012! Let's start this page off with some information on how the phase "drain bias" is often misused. We've been guilty of this ourselves but will work to clean up our act. Below an email from Chris is paraphrased:
Gates are "biased", as are grids and bases, but plates, collectors and drains are supplied with voltage. I think the (mis)use of the word "bias" has been started by non-native English speakers, who has no conception of the English word "bias". As we all know setting the quiescent current in an amplifying device (Idq for a FET) is set by the normal, no-signal, voltage (or current) of the control element ( grid, gate, base etc). If we set zero current thru the main current path, varying the voltage on the drain, or plate etc, will not change that current, unless you send it into breakdown, which would be a dumb idea. Thus the plate, drain etc. cannot be responsible for the device quiescent current, and cannot be called a "bias" terminal. For the drain terminal, "bias voltage" should perhaps be substituted with the words "supply voltage " or just "voltage".
"Drain bias" is a misnomer, and should be avoided as much as "tech", as it is factually incorrect.
Chris, now that you have explained this we completely agree, and will hunt down and remove all references to "drain bias" on the site. Thanks! - UE
We are putting this important chapter of Microwaves101 onto the site in several installments because this topic is so huge in microwave engineering. The first installment is up and running as of Summer of 2003, and includes basic theory, all the terminology you will ever need, how FETs are made, and a discussion of I-V and transfer curves. Future installments will describe bias networks, including self-biasing analysis, small-signal equivalent-circuit modeling, FET power handling, and a large-signal discussion including drain efficiency, power-added efficiency, and load line analysis using the method first described by Steve Cripps (he was the first person to note that the reactive part of a FET's output equivalent circuit accurately describes its load pull contours). It is truly amazing how many brilliant microwave guys are named Steve. Contact us if there is any specific info we've omitted that you'd like to see here!
Here is a clickable table of contents for this page:
How FETs are made (moved to a separate page)
What's a FET? In microwaves we are almost always referring to a MESFET, which stands for metal-semiconductor field effect transistor. A FET is a three terminal device capable of both microwave amplification and switching. The FET's three terminals are denoted as gate, source and drain. With respect to a bipolar transistor (BJT), the gate of a FET corresponds to the base of a BJT, the drain corresponds to the collector and the source corresponds to the emitter terminal. This is useful knowledge since every curve tracer we've ever seen in a lab has its three terminals labeled collector, base, and emitter, not drain, gate, and source. Pay attention, in case your boss puts you on the spot someday!
Used as an amplifier, the gate is most often configured as the input terminal, the source is grounded and the drain is the output. The output current (IDS) is controlled by the input voltage (VGS). This configuration is called common source since the source is common to the input and output ground connections. It is also possible (but unusual) to ground the gate and create a common-gate amplifier. Such an amplifier does not provide the voltage gain of the common-source amplifier, but it has the interesting property of being easier to impedance match than a "normal" common-source amplifier. We won't get into that here.
The figure below shows a cross-section of the channel of a field-effect transistor and explains some FET terminology. The drain and source are connected by the FET channel, which is formed by creating a mesa of N-type semiconductor (for an N-channel FET) on top of a semi-insulating substrate (typically GaAs). In microwaves we are almost often dealing with N-channel FETs. P-channel FETs are possible but are never used at microwave frequencies, because they would have far worse performance compared to N-channel FETs. Go ask a device guy why that is and he will explain to you something about the electron mobility of the device, but who really cares? The drain and source contacts are connected to the channel with ohmic metal contacts that form low-resistance connections to these terminals. The gate connection to the channel is formed between the drain and source by a Schottky metal contact to the channel. The rectifying property of the gate contact means that when it is reverse biased with respect to the channel it conducts almost zero DC current (IGS) to the channel, but its electric field can be used to effectively displace the electrons within the channel. Thus an AC voltage incident on the gate terminal causes a variable resistance between the source and drain of the FET. When the gate reaches pinch-off voltage the electrons below the gate are depleted to the point where essentially no current can flow from drain to source.
The source connection is the "source" of electrons in the channel, and the drain is where they are "drained off". Remember that we are talking about electrons flowing here, and you will see that the direction of current flow is positive from drain to source.
FET channel cross-section
FET geometry refers to the physical dimensions of a FET. FET dimensions are always described in microns or millimeters, never in mils, with the exception that overall chip dimensions (length, width and thickness) as often given in mils as well as microns. This is because the next higher assembly (artwork for a thin-film network for example) is often dimensioned in inches.
Gate length is often confused with gate width. Just remember when you look at a gate finger, gate length is the short dimension and gate width is the long dimension. This is illustrated in the figure below. Gate length has a major effect on maximum frequency of operation: one-micron gates start to suck wind at C-band, half-micron gates are good through X-band, quarter-micron gates are good into Ka-band, and 0.15 micron gates can work up through W-band. What is the limit on gate length? We aren't there yet, some companies are experimenting with 50 to 100 nanometer gates!
Gate width refers to the unit width of the gate as it passes between the source and drain across the mesa (the semiconducting area of a FET). Wider gates mean more DC and RF current, and therefore more power capability. Gate width must be sized appropriate to frequency: if the gate width starts to become an appreciable fraction of a wavelength, the RF performance of the FET starts to suffer. At X-band, power FETs often have 150 um wide gates. At Ka-band the the gate width is typically 75 micron maximum. At W-band perhaps 40 micron fingers is the upper limit.
Gate width versus gate length
A gate finger refers to a single gate structure. Gate periphery is the total size of a FET. Most FETs have multiple gate fingers, so the periphery is equal to the number of gate fingers times the unit gate width. In the example figure there are four gate fingers. Many of the FET parameters can be directly scaled with gate periphery, for example the saturated drain current is proportional to gate periphery.
The gate bus-bar is the electrical contact that is used to connect all of the multiple gate fingers together. The drain bus bar serves a similar purpose.
Via holes are what connect the source (or individual sources) to the chip backside metal, which is considered RF and DC ground. When individual sources are grounded with separate via holes, these vias are referred to as ISVs which stands for individual source vias. ISVs are only used on very thin FETs, perhaps with two mils (50 microns) maximum thickness. ISVs provide very low-inductance grounding to the source connection, providing the most gain and efficiency for power amplifiers, which becomes more important for power FETs operating at millimeter-wave frequencies. On four-mil GaAs and thicker, ISVs are not usually possible, because the source contact pad is typically smaller than the minimum diameter of an etched via hole.
FET geometry lesson
Mushroom gate or tee gate refers to a technique of providing very short effective gate length, while providing low gate resistance. Gate resistance is a parasitic element that affects the maximum available gain of a FET, and is inversely proportional to the cross-sectional area of metal along the gate finger. A picture of a tee-gate is shown below. This type of structure involves extra process steps and is therefore used only in higher-frequency applications where short gates are required, such as X-band through millimeter-waves.
A tee gate
Why use GaAs?
The FET is built on top of a semi-insulating substrate, most often GaAs. When we say "semi-insulating" this is perhaps misleading. In its pure form, GaAs is remarkable insulator, which is what makes monolithic microwave integrated circuits (MMICs) practical. Here is one advantage GaAs has over silicon. Pure silicon is a better conductor than pure GaAs, so it tends to dissipate electrical fields that are needed to support transmission modes and hence needs some "help" to be used as a MMIC. We'll discuss that later.
GaAs is referred to as a "compound semiconductor", because it is a crystal of more than one element. Silicon is a semiconductor all by itself. GaAs wafers are available in up to six inch diameter, but more often FET and MMIC manufacturers use four-inch material.
What does III-V semiconductor refer to?
Three-five material refers to compound semiconductors made from one element from Group III on the periodic chart (gallium in the case of GaAs) and one from Group V (arsenic in the case of GaAs). Thanks to Luis, we just corrected that statement in March 2011, it was backwards for the past five years! Other three-five (or III-V in Roman numerals) semiconductors include indium phosphide and gallium nitride. TriQuint Semiconductor derived their name from the III-V material that their business is based on (GaAs, of course!)
The short answer is "yes", although I have seen a person eat a MMIC and live just to prove this wrong. Gallium arsenide may not kill you, but it has a nasty habit of breaking down into gallium and arsenide if left in your town dump or incinerator. Traces of arsenide in water cause cancer, while small doses are quite lethal. Remember the play "Arsenic and Old Lace"?
In the United states the GaAs industry has strict controls on what they can flush into the wastewater stream. Even though they saw, etch, and polish tons of GaAs wafers each year, GaAs foundries have some clever ways of separating out the bad stuff before it goes down the drain, so give them credit for being good citizens.
The EPA's specification on arsenic in drinking water is 50 parts per billion. Check out your water bill next month, we often go over this limit. Could it be the 130 million cell phones (65,000 tons) that get discarded every year are putting us over the top? Damn right, and wireless trash contributes a plethora of poisonous substances to your community, associated with cancer and a range of reproductive, neurological and developmental disorders, including:
- Metals such as arsenic, antimony, beryllium, cadmium, copper, lead, nickel and zinc,
- Brominated flame retardants,
- Dioxins and furans (produced during incineration).
May 21, 2011... Doug wants to point out that zinc and copper are essential nutrients for humans, and not toxic (in appropriate quantities of course).
We predict that someday your wireless service provider will have to take care of safely disposing your obsolete or broken phone equipment. Until then it's your responsibility. One option is to check out Charitable Recycling, a U.S. charitable organization (duh) that pays a dollar to charity for every unwanted cell phone turned in. Then they fix up the phones and donate them to needy people all over the world. Hopefully these needy people don't burn your little gift to keep their mud hut warm! You can go to their site for information about local collection sites and the charities they support. Perhaps a better option is the Rechargeable Battery Recycling program available at many places where you buy batteries (like Sears, Home Depot, Ace Hardware and Radio Shack).
Bandgap is a material property that takes some knowledge of semiconductor physics to understand. Who cares? You might. The higher the bandgap, the higher the breakdown voltage the material can support. High breakdown is a huge advantage for power amplifiers, remember Ohm's law and you will see that voltage swing is proportional to power. GaAs is a medium bandgap technology at 1.5 electron-volts, you can get 20 volts breakdown on a good day with a GaAs MESFET. InP is a low-bandgap device at 0.75 electron-volts, it only supports a few volts breakdown. The "great white hope" (is that politically incorrect or what?) of microwave semiconductors is gallium nitride, which is a wide bandgap semiconductor at greater than 3 electron-volts bandgap energy. GaN FETs have exhibited over 100 volts breakdown voltage. DARPA is a big fan of GaN technology and is spending tens of millions of taxpayer dollars trying to develop this technology beyond a laboratory curiosity that blows up in a few hours of operation into the "next big thing" for solid-state microwave power.
Ohmic metal on a FET is often called "source-drain metal". This is because it forms the contacts for these two terminals of the FET. The drain and source contacts are considered "ohmic" because they behave resistively, that is, they pass current in either direction, obeying Ohm's law where current is proportional to voltage. Ohmic metal is usually the first layer of metal applied when a FET or a MMIC is fabricated. It is alloyed at high temperature.
What is a Schottky contact? It is a diode junction formed between certain metals and semiconductors. You can read about the illustrious Mr. Schottky here. Metals that form Schottky contacts to N-type GaAs include aluminum, gold, silver, titanium and platinum. Often a layered structure of metals is used in FETs, such as titanium/platinum/gold (Ti/Pt/Au). Gold reacts with GaAs so it is a bad Schottky metal. Platinum keeps the gold away from the GaAs (it acts as a barrier metal in this case). Titanium is what makes the gate "stick" to the GaAs.
What does semiconductor refer to?
Based on differences in bulk resistivity, five classes of materials are in common usage, namely conductors, semiconductors, semi-insulators, insulators and superconductors. There is no IEEE standard for these categories, and if someone can supply a good reference on this point we'd appreciate it! Click here for the Microwaves101 version of the definition of these categories.
GaAs bulk resistivity can be tailored over a huge range, 10-6 to 1022, so that GaAs can be anywhere from a conductor to an insulator. By doping Chrome is usually added to the melt to raise resistivity, but this trick has its limitations (it does not stay stable through wafer processing steps). High purity, undoped GaAs can be 107 to 108 ohm-cm.
Intrinsic versus extrinsic GaAs? intrinsic refers to the pure crystal, extrinsic refers to the doped material where conduction is due to donor or acceptors.
Thermal conductivity of FETs
The thermal conductivity (TC) of GaAs varies at 1/T (T in Kelvin). It is approximately 0.55 W/cm-C at room temperature.
This discussion got so big we had to put it on a separate page! Click here to check out our latest description of microwave semiconductor processing.
The operation of any three-terminal device is well described on a three dimensional surface plot as shown below. For an FET, the output characteristics VDS and IDS are shown to be a function of the input voltage VGS. A typical FET response is shown below.
Three-dimensional FET IV characteristics
The three-dimensional characteristics are most often collapsed onto 2-D plots of IV curves, as shown below. Here the output drain current/voltage relationship is plotted at discrete gate voltages. Such a plot can be produced from a FET using any Tektronics curve tracer. Depicted on this plot are some definitions:
IMAX: the drain-source current when the gate is forward biased for maximum channel current. This is typically measured at up to 1.0 volts on the gate (higher potentials will conduct tons of current across the gate Schottky contact which tends to roast your FET) and perhaps 1.5 or 2 Volts drain-to-source. To get to IMAX the gate must be raised to its Schottky barrier height (voltage), which is approximately 0.7 volts. This is the intrinsic gate bias. The other 0.3 volts will drop across the intrinsic source resistance RS. Still, you might want to limit the measurement current with a current-limiting resistor....
VPO: pinch-off voltage. This is where the drain-source terminals start to look like an open circuit, and no appreciable current flows even at high drain-source potentials. In practice there is always some residual current and the actual VPO measurement must make an allowance for this. For example, the pinch-off voltage could be measured at 2.5% of IDSS and VDS=2 volts.
VBR: the gate-drain breakdown voltage, which is indirectly measured on the IV curves. At high drain-source potential and near pinch-off, the IV curves tend to bend up. As shown in the picture the breakdown voltage VDS is approximately 10 volts (VGS=-4 volts and VDS=6 volts combined). Stay away from this bias region if you want your FET to have a long and happy life!
Knee voltage: the voltage at which the curves transition from "linear" to "saturation". In the linear region, IDS depends on both VGS and VDS (from VDS=0 Volts to approximately VDS=2 Volts). In the saturation region, IDS depends mainly on VGS and not VDS. This is the right side of the curve, beyond VDS=2 volts..
FET IV characteristics and definitions
Another very useful plot of the FET's characteristics is called the FET transfer characteristic. Here we see the variation in drain current (thanks Karel!) due to variation in gate voltage, at some fixed drain voltage in the saturation region (beyond VDS=2 Volts). This is analogous to looking at a cut in the Y-Z plane in the surface plot above. Plotted below is the transfer characteristic of a FET. This type of plot is extremely useful in designing self-biasing networks which are described below. Here we see why a FET is an effective amplifier: for a quiescent point of "1 Volts, a peak-to-peak voltage swing of +/-0.5 Volts on the gate terminal provides a variation in drain current from 50 to 250 mA.
FET transfer characteristics
Where did we get the nice transfer curve shown in the above plot? We have developed a model that allows the user to fit a continuous transfer curve to measured data, with separate coefficients to fit the regions above and below VGS=0 Volts. The equations are shown below. By using two different exponent terms, is possible to control the ratio of IMAX/IDSS, which is impossible in simpler models. Soon we will put a spreadsheet in the download area that contains these equations for you to use when fitting your own FET. Check back later! By the way, the third equation is missing a minus sign. Find it and you've won a pen knife!
With respect to their intended operation, FETs can be divided into three categories: low noise, power and switch FETs. Low noise FETs are optimized to provide the lowest possible noise figure at very low voltage and power (perhaps 1.5 volts and 10 ma). Power FETs possess higher breakdown voltage than low noise FETs and can therefore operate at higher voltages, and are much larger in periphery than low noise FETs.
Switch FETs are intended to operate passively (no drain current and no gain); the gate voltage is merely used to switch the device from a resistive element to a small capacitive element. Switch FETs can be configured in series with a transmission line (drain and source act as input or output), or in shunt, with the source grounded. One beautiful thing about switch FETs is that you don't really care how high the gate feed resistance is, because the RF signal doesn't traverse the gate terminal. This opens up many possibilities when you design the gate structure. One type of switch FET is called a "meandered gate" FET. This means that multiple FET gates are hooked up in series, rather than through a single bus-bar.
A "depletion mode" FET is one where the gate is mainly used to reduce the current within the channel (most common microwave FETs are depletion mode). An "enhancement mode" FET does not conduct drain-to-source until the gate is slightly forward biased. Think of this as a depletion-mode FET with a zero-volt pinch-off voltage. There is a big limitation to enhancement-mode FETs: you can't exceed the turn-on voltage of the Schottky contact, which is typically 0.7 volts, so the gate etching process has to be well-controlled to produce a FET with pinch-off of zero volts or perhaps a few tenths of a positive volt. Etch too far and you will end up with no FET at all, just a capacitor between drain and source!
Bias networks are what are used to put a FET at the intended quiescent operating point. For example, you might want to operate a FET in a power amplifier at 6 volts VDS and at 50% of the saturated drain current (IDSS/2). This is the quiescent point.
FET DC characteristics, such as IDSS and VPO vary from lot to lot, and even within a wafer. This complicates the life of the amplifier designer, since VGS needs to be set to achieve either a fixed fraction of the saturated drain current, or a fixed current. One amplifier might need VGS to be "1.05 volts, another might need VGS =-2.1 volts to perform as designed. What's a designer to do?
There are at least three ways to bias up a FET amplifier to get to the intended quiescent operating point. The most obvious is to have separate DC power supplies for the gate and drain connections, with the gate supply being adjustable, and ground the source. Grounding the source directly will provide the most gain from the FET, which is why this is a good idea if efficiency is a concern. In practice, the "adjustable" gate bias supply is often a fixed supply of perhaps "5 Volts, with an adjustable resistor-divider network being employed to supply the needed gate voltage.
Another method of biasing a FET is with an active bias network. This is an analog circuit that attempts to eliminate any manual adjustments to the FET Q-point, by using a small FET to "calculate" the required gate bias for the FET in the circuit and supply it to the larger FET which is the active device in the amplifier. Such a circuit is often called a "current mirror" and won't be covered here at this time. An active bias network, if designed properly, does not reduce the overall efficiency of a power amplifier by much. However, a negative supply voltage is still required, although it need only be at a fixed voltage such as "5 Volts.
The third way to bias a FET is to employ a "self-biasing" network, in which a resistor of a strategic value is placed between the source connection and ground. The resistor is bypassed with a capacitor so that the FET source connection sees a zero-Ohm connection to ground at the operating frequency. When drain current flows through the FET and then through the source resistor, the source voltage rises above ground. The gate voltage is either held at a fixed voltage or grounded, resulting in a fixed negative gate-source voltage, which is (hopefully) the intended Q-point. For example, if the gate was grounded, and the FET was drawing 200 ma of drain current through a 10-ohm source resistor, the gate-source bias would be "2 volts. The major advantage of the self-bias scheme over other bias schemes is that only a single positive voltage supply is needed to power up the amplifier. The downsides to using self-bias schemes are that amplifier efficiency is lost due to the voltage drop of the source resistor. Also, the FET cannot be RF grounded at all frequencies as well as if it was DC grounded with via holes, so gain and efficiency can be degraded as a result. Self-bias networks are often used in LNAs, but not power amplifiers, for these two reasons.
Self-bias network with raised gate voltage
The self-bias network is used to eliminate the need for a negative voltage to a FET-based amplifier. One of the ways to make a design less susceptible to normal variations in FET transfer characteristics (the gate voltage needed to induce a fixed drain current) is to raise the gate bias above ground. Such designs are often called "raised gate bias" designs. Duh! Below we will prove to you that raised gate bias designs are advantageous in this regard. We also will supply you with an Excel spreadsheet where you can analyze your particular design to make it immune to variations in pinch-off voltage and IDSS.
These comments are from James, a semiconductor guy with some RF test experience, submitted in July 2005. We'll distribute them around the page in appropriate places when we get a chance...
The importance of gate inductance can't be over-emphasized for mm wave operation. In most cases the gate finger inductance will limit finger width to small fractions of a wavelength. It is controlling the gate inductance, too, which is the primary driver for the use of tee gates.
All FETs run up against a simple geometrical constraint: the channel has to scale down at least in proportion to the gate length, or gate control of the current is lost. This makes it hard to get power densities out of short gate devices.
InP actually has a bandgap of around 1.35 eV, but most InP based FETs have InGaAs based channels, which do have bandgaps around 0.75eV.
Most contacts (ohmic or Schottky) to compound semiconductors are far from perfectly stable over time. Some Schottky contacts might not be stable in the presence of hydrogen from, say, plated packages.
The thermal conductivity of GaAs is basically miserable, and ternary layers are far worse.