Why Is Saturation Mode Preferred for Biasing MOSFETs?

[salil87]

Hi
Why do we always bias a mosfet in Saturation mode? Is it becaus gm is more in saturation or is it because gm is more linear in saturation?

Thanks
Salil

 

[analogdesign]

We bias a MOSFET in Saturation mode (as opposed to subthreshold) for a number of reasons.

The key reasons are you get higher gm and the device is much faster in saturation. In many processes the output impedance is higher in saturation as well.

The key benefit of subthreshold operation is low power. While gm is low compared to saturation, gm/I is quite high!

 

[jim hardy]

There do exist circuits that use MOSFETS in linear region.

Audio enthusiasts sometimes use them in place of vacuum tubes for home built power amplifiers.

 

[analogdesign]

There do exist circuits that use MOSFETS in linear region.

Audio enthusiasts sometimes use them in place of vacuum tubes for home built power amplifiers.

Hi Jim,

Do you mean subthreshold region? Typically the term "linear region" refers to the region where the MOSFET acts as a voltage-controlled resistor.

 

[Electropioneer]

We use Mosfets in Saturation mode, for exactly the same reasons we use BJ Transistor in saturation mode.
If you wish to use MOSFET as a simple switch (equivalent to mechanical switch) then it is in your best interest to have minimal loss of energy across that switch. That's why you would want to get gm as high as possible (saturation), because energy wasted on mosfet rises exponentially (P=R*I^2) if conductance decreases slightly. For high current applications that can mean tens and even hundreds of watts wasted.

I want to point out that there are many applications where you would want to use mosfets in linear region. Mosfets sound good in audio amplifiers, we use them in voltage/current regulators, electronic loads etc...

 

[analogdesign]

We use Mosfets in Saturation mode, for exactly the same reasons we use BJ Transistor in saturation mode.
If you wish to use MOSFET as a simple switch (equivalent to mechanical switch) then it is in your best interest to have minimal loss of energy across that switch. That's why you would want to get gm as high as possible (saturation), because energy wasted on mosfet rises exponentially (P=R*I^2) if conductance decreases slightly. For high current applications that can mean tens and even hundreds of watts wasted.

I want to point out that there are many applications where you would want to use mosfets in linear region. Mosfets sound good in audio amplifiers, we use them in voltage/current regulators, electronic loads etc...

I think you're a bit confused in your terminology. Saturation in MOSFETs is analogous to the Forward Active Region in BJTs, not Saturation. Typically Saturation is avoided in BJT design and desired in MOSFET work. An unfortunate choice of words, certainly, as they describe conceptually different operating regions of conceptually similar devices..

Also, what you described is using a MOSFET as a switch in the linear region, not saturation. The power equation you gave is based on Ohm's law which is not applicable when the MOSFET is operating in saturation. It is a voltage-controlled resistor in the linear mode and in that case your equation is correct.

Electronic loads are usually operated in the saturation region for high output impedance. However, the other applications you mentioned (such as series voltage regulation) use MOSFETs in the linear region as you indicated.

 

[Electropioneer]

Thanks for the reply analogdesign.
First of all, i know what you are saying, but I believe you misunderstood me.

1) If you look at I-V diagrams of both types of transistors, then yes MOSFET's saturation region corresponds to active region of BJT. However, that is inherently bad analogy since MOSFET is voltage controlled, and BJT is current controlled device. Difference will be obvious in a minute.

2) You would never want to use any transistor as a switch in linear region.

3) Ohm's law functions as is. It doesn't matter if you are talking about transistors of any type, diodes, resistors, switches, coils, capacitors or connectors. Power that any device will dissipate is equal to product of voltage across that device and current that flows trough it. For example a simple diode has a constant voltage drop of 0.6v across PN junction, no matter the intensity of current, so applying the ohm's law you would conclude that resistance of diode changes with changing current, and we call that dynamic resistance. Most transistors are the same.

In case of Mosfets:
If you start increasing Vgs from 0 the Ids starts to rise. Most MOSFETs start opening around 4 V. As Vds increases Id rises in linear fashion, so we call this mode linear mode. If you would design an amplifier you would keep mosfets in this mode. However since channel between Drain and Source is not well established D-S resistance is high, and mosfets dissipate lot's of heat in this mode. As i said, P=Rds * Id^2.
If you keep increasing Vds above 6-7 V mosfet goes into saturation, DS channel becomes wide, and Rds(on) becomes small, sometimes even few mOhm and less for state of the art components used in High power SMPS devices for example. In this mode mosfet dissipates little heat, that's why it acts as a good switch. How much heat?? Well P=Rds(on) * Id^2. This equations still works, as I said.
Some mosfets designed for switching open at two and saturate around 4 V. There are many types.

Now for the BJT:
If you repeat the same process but instead of concentrating on Base Emitter voltage (equivalent to Vgs) you concentrate on Ib, current that goes into base (let's take NPN for ex.).
Raising Ib from 0 transistor starts to open, however Collector - Emitter current is limited by transistor construction, and is equal to BETA*ib, where beta is amplification factor. Here the same thing happens, Ice rises linearly with Ib, so we call this Linear mode. In this mode voltage across C-E is high, so is resistance, and just like MOSFET, in this mode BJT dissipates lots of heat because P=Vce*Ice=Rce * Ice^2.
When you increase Ib so much that Collector current is smaller than BETA*Ib transistor reached saturation point, and Voltage across C-E decreases to about 0.1V or less. CE resistance becomes small and power dissipation decreases significantly but is still equal to P=Rce * Ice^2. And if we would to use BJT as a switch we would saturate is even further. For low frequency application, for ex. you want to turn on some diode, some light, relay, motor you would drive bjt into deep saturation. For high frequency application (like SMPS, PWM control, oscillators) you would keep transistor above saturation threshold, so dissipation would be small but switching from saturation to off mode would still be fast. When transistor is deep in saturation it takes time to drive it back down.

So you see, as i have said previously, you would use MOSFET in saturation for exactly the same reason you would use BJT in saturation.

4)
I gave few examples where MOSFET is used in linear mode. In electronic load example mosfet is most certainly used in linear mode, since the point is to load the circuit to an extent you wish.

Another even more popular example would be in analog integrated circuits, MOSFETs are used as Active Resistors.
The thing with resistors is that it is really hard to produce them in silicon substrate to be used in an integrated circuit. However resistors are the basic components, can't do much electronics without them. So instead of producing resistors where they are needed we put a transistor and make it work in linear region. It's not hard actually just short circuit base and collector (or gate and source). This is the simplest example. There are number of simple blocks of circuits that use MOSFETs in linear mode, especially in analog ICs, like current mirrors, temperature stabilizers etc etc.

 

[analogdesign]

Thanks for the reply analogdesign.
First of all, i know what you are saying, but I believe you misunderstood me.

I appreciate the detailed reply. I disagree with much of what you wrote, but I suspect it is a confusion based on terminology. Throughout your reply you appear to be confusing the linear and saturations regions of MOSFETs.

1) If you look at I-V diagrams of both types of transistors, then yes MOSFET's saturation region corresponds to active region of BJT. However, that is inherently bad analogy since MOSFET is voltage controlled, and BJT is current controlled device. Difference will be obvious in a minute

This I mostly agree with. In my opinion forward active is pretty similar to saturation in MOSFETs with the big caveat that the equation for the gm is very different (i.e the current output of a BJT is exponentially related to the small-signal input voltage, while in a MOSFET the relationship is quadratic).

2) You would never want to use any transistor as a switch in linear region.

You are 100% wrong in this, but I suspect it is a question of terminology. THe convention is that the linear region is when Vgs is greater than Vt and Vds is less than (Vgs-Vt). In this case the MOSFET is operating as a linear resistor. To use this MOSFET as a switch, Vgs is made as large as possible to minimize Ron (Ron = 1/gm). The desired value of Ron for the switch determines its size. This is not a very linear resistor as the source voltage varies, so sometimes you put PMOS and NMOS devices in parallel to make a switch called a transmission gate.

I believe you know this, but confused about the term "linear region" when applied to MOSFETs. You are aware of course, that in CMOS digital logic at a given instant the devices are either in cutoff or they are deep in the linear region, correct? You don't have large static current in digital CMOS because every device in linear region has a corresponding device in cutoff which blocks the DC current path.

) Ohm's law functions as is. It doesn't matter if you are talking about transistors of any type, diodes, resistors, switches, coils, capacitors or connectors. Power that any device will dissipate is equal to product of voltage across that device and current that flows trough it. For example a simple diode has a constant voltage drop of 0.6v across PN junction, no matter the intensity of current, so applying the ohm's law you would conclude that resistance of diode changes with changing current, and we call that dynamic resistance. Most transistors are the same.

I disagree with you here. In my opinion you're confusing the definition of power with Ohm's law. To be clear, here is Wikipedia's definition of Ohm's law:
"Ohm's law states that the current through a conductor between two points is directly proportional to the potential difference across the two points."

In your own example, you state that "For example a simple diode has a constant voltage drop of 0.6v across PN junction, no matter the intensity of current, so applying the ohm's law you would conclude that resistance of diode changes with changing current, and we call that dynamic resistance."

By your own example the diode does not follow Ohm's law. The dynamic resistance is not a real, physical resistance, but a mathematic convention we invent to describe the behavior of the device.

If the current through a device is not directly proportional to the voltage across it, the device does not follow Ohm's law. End of story. So while it is obvious you understand how a diode works, I don't think you are applying Ohm's law correctly here.

I gave few examples where MOSFET is used in linear mode. In electronic load example mosfet is most certainly used in linear mode, since the point is to load the circuit to an extent you wish.

Another even more popular example would be in analog integrated circuits, MOSFETs are used as Active Resistors.
The thing with resistors is that it is really hard to produce them in silicon substrate to be used in an integrated circuit. However resistors are the basic components, can't do much electronics without them. So instead of producing resistors where they are needed we put a transistor and make it work in linear region. It's not hard actually just short circuit base and collector (or gate and source). This is the simplest example. There are number of simple blocks of circuits that use MOSFETs in linear mode, especially in analog ICs, like current mirrors, temperature stabilizers etc etc.

I respectfully submit you are a bit out of your element when you start discussing ICs. Your assertions that "The thing with resistors is that it is really hard to produce them in silicon substrate to be used in an integrated circuit" and "So instead of producing resistors where they are needed we put a transistor and make it work in linear region" are obsolete.

Since at least the mid to late 70s integrated resistors have been commonplace. I am an analog IC designer by profession and I have never worked on a chip that didn't have some combination of P and N-doped polysilcion resistors and N and/or P-type diffusion resistors available for use. For high-current applications people also use interconnect metals as resistors.

The main use of a MOSFET in the linear region as a resistor is in situations where you can use feedback to vary the value of the resistance in an analog way. And example of this is the zero setting resistor in the Miller Compensation network of a two-stage op amp. In that case, you can arrange the feedback such that across process and temperature variation the resistance of the MOSFET in the linear region tracks the gm of the common-source device in the second stage, as desired for stability. I should point out that when I'm designing a Miller op amp I never do this. It is simpler and typically good enough to just use a poly resistor of value 1/gm of the expected gm of the common-source device.

There are number of simple blocks of circuits that use MOSFETs in linear mode, especially in analog ICs, like current mirrors, temperature stabilizers etc etc.

This statement is what leads me to believe that you are confused between linear and saturation modes in MOSFETs. The entire point of a current mirror is that the output current is constant even as the output voltage varies over a large range. This is the output resistance of the mirror and ideally it is infinite. By definition a MOSFET operating in the linear range has a current that is proportional to its output voltage so it is easy to see linear mode MOSFETs are not appropriate in a current mirror.

I assure you that the devices in a well-functioning current mirror are operating in saturation. In fact, in order to give them a high output resistance we typically operate the devices deep in saturation. What that means is we choose a gate voltage such that Vgs-vt is large. This has the tradeoff that it limits the output swing of the current mirror but it makes it act more ideal.

I suspect we are much closer to each other in our understanding that it appears, but it is very important in engineering to get the terminology clear.

 

[Electropioneer]

You are 100% wrong in this, but I suspect it is a question of terminology. THe convention is that the linear region is when Vgs is greater than Vt and Vds is less than (Vgs-Vt). In this case the MOSFET is operating as a linear resistor. To use this MOSFET as a switch, Vgs is made as large as possible to minimize Ron (Ron = 1/gm). The desired value of Ron for the switch determines its size. This is not a very linear resistor as the source voltage varies, so sometimes you put PMOS and NMOS devices in parallel to make a switch called a transmission gate.

I believe you know this, but confused about the term "linear region" when applied to MOSFETs. You are aware of course, that in CMOS digital logic at a given instant the devices are either in cutoff or they are deep in the linear region, correct? You don't have large static current in digital CMOS because every device in linear region has a corresponding device in cutoff which blocks the DC current path.

You are right about this. I see my mistake. I was using the wrong terminology and I apologize if I confused someone.

By your own example the diode does not follow Ohm's law. The dynamic resistance is not a real, physical resistance, but a mathematic convention we invent to describe the behavior of the device.

If the current through a device is not directly proportional to the voltage across it, the device does not follow Ohm's law. End of story. So while it is obvious you understand how a diode works, I don't think you are applying Ohm's law correctly here.

Sure it is real, you take it into account every time you add voltage drop across any PN Junction, you just don't care about it's value. This is turning into a good old chicken-egg problem :)

I respectfully submit you are a bit out of your element when you start discussing ICs.

I am out of element, I am not IC designer, I work in power electronics. Most books I read on the subjects were published in 70-80, where it's still written that ECL is our greatest achievement that will pave the way towards fast computing, and that CMOS is way too slow. Anyhow, thanks for the correction.

 

[analogdesign]

I am out of element, I am not IC designer, I work in power electronics. Most books I read on the subjects were published in 70-80, where it's still written that ECL is our greatest achievement that will pave the way towards fast computing, and that CMOS is way too slow. Anyhow, thanks for the correction.

You may find it interesting that Current-Mode Logic (CML) which is the CMOS version of ECL is getting more and more important in practice. CML is essentially ECL without the emitter followers. Anyway, I'm sure you're aware of the trend towards high-speed serial digital communications to reduce pin count. It turns out those serializers need to run at the maximum speed possible in a given fabrication process. These current-mode gates are physically large, aren't amenable to automatic place-and-route, and burn an enormous amount of power compared to CMOS, but man oh man do they scream like banshees. That's how we can have data, video, and memory interfaces at 10Gb/s or even 40 Gb/s and faster in a process that can only support CMOS logic at 4 GHz or so.

One of the most intriguing thing about electronics is how old ideas (like ECL) hibernate and then are resurrected decade later. One of the many things I love about the field.