5
\$\begingroup\$

Why do we use BJTs in discrete applications, but prefer to use CMOS in integrated circuits?

\$\endgroup\$
4
  • 11
    \$\begingroup\$ We don't; it's much leveller than what your question implies. \$\endgroup\$
    – Andy aka
    Oct 11, 2022 at 16:19
  • 2
    \$\begingroup\$ There is no general preference. Each choice may depend on hardware system requirements. Learn about systems engineering. \$\endgroup\$
    – Amit M
    Oct 11, 2022 at 16:25
  • \$\begingroup\$ Why CMOS but not pMOS or nMOS? \$\endgroup\$
    – user263983
    Oct 11, 2022 at 17:21
  • \$\begingroup\$ CMOS is "Complementary"-MOS. It combines both NMOS and PMOS as symmetrical pairs. It's 'easier' to design things like simple logic gates using CMOS. With CMOS you end up with smaller circuits which is great because we can fit more logic gates into our designs which may be quite small. \$\endgroup\$
    – BobaJFET
    Oct 14, 2022 at 12:45

6 Answers 6

13
\$\begingroup\$

Not necessarily, MOSFETs are very useful components in discrete applications, too. For example MOSFETs have lower source-drain voltage drop than BJTs have collector-emitter voltage drop if you have reasonable (not too high) currents. Low voltage switched mode power supplies usually use MOSFETs. Also MOSFETs don't have a wasted gate current like BJTs have a wasted base current.

BJTs are good for class B and AB amplifiers so you usually see BJTs in those cases.

The reason digital devices are all CMOS is because of the low power consumption. Neither NMOS nor PMOS nor any kind of BJT logic allows such a low power consumption. CMOS ideally uses only power when switching states, although with modern processes the gates are becoming so thin there may be constant leakage too -- but in cases where low power really matters, you wouldn't use the latest and greatest process anyway.

\$\endgroup\$
2
  • 1
    \$\begingroup\$ However with the newer processes and a design shrink you usually lower the voltage, keeping leakage down. \$\endgroup\$
    – mow
    Oct 13, 2022 at 9:45
  • 2
    \$\begingroup\$ "but in cases where low power really matters, you wouldn't use the latest and greatest process anyway." I work in a field where power consumption is important (smartphone RF transceivers) and we currently use TSMC’s 7nm node. Leakage is a problem but only really at higher temperatures. We have a low supply voltage, we use bigger, slower low-leakage cells (and memories) where possible and we split the chip into several power domains. \$\endgroup\$
    – Michael
    Oct 13, 2022 at 13:23
13
\$\begingroup\$

CMOS transistors on an integrated circuit can be much much smaller (1000's of times) than BJTs. That, together with the ease of driving them (no base current; no need for current limiting resistors) makes them easier to use.

BJTs have more precision and specialized uses in some analog circuits, but modern CMOS processes can approach their performance.

Generally a CMOS (analog) design achieves performance (accuracy, bandwidth, circuit functions) by designing with different sized transistors (length and width) to achieve specific performance. This is not a common in BJT-based designs and usually resistors are used to change circuit performance and parametrics (e.g. gain, noise etc.).

Discrete small MOSFETs would be very delicate and sensitive to ESD; therefore they are not comonly available or used. In addition, it would take a huge number of different devices (length, width) to give the range of attributes that are needed in a circuit design.

\$\endgroup\$
2
  • 3
    \$\begingroup\$ It is worth noting that while there are physically tiny discrete MOSFETs on the market, they are huge compared to the ones used inside a microchip. The smallest gate capacitance discrete MOSFETs are in range of 1-10 pF, while the ones on a microchip are less than 0.1 pF for internal logic. The MOSFETs used for IO pins are much larger to handle ESD and higher currents. \$\endgroup\$
    – jpa
    Oct 12, 2022 at 6:26
  • 2
    \$\begingroup\$ MOSFETs on a chip can have gate capacitances of femtofarads. Of course this is not practical or useful in a discrete device \$\endgroup\$
    – jp314
    Oct 12, 2022 at 15:10
8
\$\begingroup\$

Ease of use.

Lone BJTs have higher gain, lower VBE, and tighter VBE matching, making even current mirrors feasible, to a limited extent (low accuracy say 5-20%, and preferably at low voltage or power dissipation; note the latter can be extended with a cascode e.g. Wilson current mirror). And they depend critically on matching characteristics, so it's no small feat that this is possible (well, kind of it is, but it's nice in any case).

Whereas FETs vary by whole volts between random parts of a given type, and Vgs(th) is higher and gain is lower. So you get less value per part, and they can be harder to use. For example, the VBE drop of a BJT emitter follower can be quite tolerable in some applications (consider the Sziklai switch in the LM2596 and similar bipolar switching regulators; the minimum ~0.7V drop is pretty modest compared to say a 24V supply, where these parts shine the brightest), though annoying in others (conversely, check the same chip's efficiency at low input voltage.. eep!)

The fact that the base also acts as a diode, is sometimes helpful and sometimes a hindrance, but overall more useful I think I would say. For example, you don't have to worry about saturating VBE, just pull it up until it's dropping enough current and BJT physics handles the rest. Whereas pulling up a MOSFET gate (from an unspecified voltage, let's say it could be a high voltage--) could simply blow it up. So maybe you need an extra diode or resistor divider or something to deal with that.

On the other hand, MOSFETs don't require drive current, so if you have a switching application, and a fixed gate voltage source, it's no problem; and it can be much more efficient than for BJTs, not just from the lack of DC gate current, but the switching is intrinsically faster and the voltage drop can be lower (BJTs generally have a minimum (granted, typically small) VCE(sat), while MOSFETs simply drop voltage proportional to load current and Rds(on)).

In ICs, there are several reasons: one, the matching issues that plague discrete applications are completely absent, making MOS current mirrors for example trivial; indeed they have unlimited fanout so an entire amplifier chain can be biased with minimal additional bias current. CMOS circuits can be much smaller than BJTs (at least, digital ones*). Two: well, I mean, both BJTs and MOSFETs are very flexible in IC format of course (width, length, degree of interdigitation, etc.), but arguably there's a stronger effect on MOSFETs (short channel effects, etc.?), making for more interesting variation and flexibility. (Or maybe not, I'm not an ASIC designer.) And three -- CMOS processes are just so dominant these days, run a few wafers of something else, still using standard digital process levels even, but it happens to be analog circuits -- who cares right? Pump out huge numbers of 'em no problem.

Overall, I, think, or get the impression that, BJTs would generally be better for op-amps and such -- but cost, and availability of processes, makes CMOS a big winner even if its performance might be inferior.

And that's not at all the complete story -- there are HBTs available (SiGe BJTs), which significantly outperform both (conventional Si transistors). These are widely used for RF purposes, but one, good luck finding any discrete (other than ~60GHz fT parts for MMIC amp or transmitter application!), and two, they can be put to great effect in ICs, including for analog purposes, even at fairly pedestrian levels...

There was an, I think ~20MHz GBW op-amp that Analog made (makes?), on their XF3? SiGe process, that's... just an absolutely ordinary op-amp, modest fT, normal Vos, normal/low Iib, normal Vcc range... just with suspiciously better Iq. Which is the real problem with op-amps: you can get as much speed as you like, but it'll cost you in current consumption. And within Si alone, no one can do any better, even with the cleverest circuits. But put SiGe to work, and you can suddenly knock off a whole factor of 2 or 3 from that figure of merit.

*Peak analog performance in CMOS is somewhere around the 60-200nm process node, as I recall. IIRC, BJTs are effective at similar scales, but such fine BJT processes might not be... very common? Or, if nothing else -- some of the most common BJT circuits today, take LM393 for example, it substantially predated such processes. So, that's kind of an odd argument to make, but by quantity, what with so many legacy designs still proving useful, maybe there is higher density overall for CMOS analog, just because (or in part) so many old designs pull down the average? Heh, I wonder.

\$\endgroup\$
6
\$\begingroup\$

In general, I would say that the types of MOSFET that you can buy as discrete components are better at switching than BJTs, and the types of BJT you can buy discrete are better at linear operation.

It's not impossible to use MOSFETs in linear regions (many ICs do internally), but the ones sold as discrete components (almost always a type called VDMOS) aren't ideal for it. Likewise, it's not impossible to use BJTs for switching (the earliest switching regulators did), but MOSFETs are far superior for that purpose due to their lower voltage drop when on.


I'd also like to note that your assumption of BJTs in discrete circuits and MOSFETs in ICs is an oversimplification at best, just plain wrong at worst. Discrete MOSFETs are used all the time when you need to switch power to something. If I need to turn something on and off, I'm definitely reaching for an Si2318 before I'd look to a MMBT4401.

It is true that BJTs are only rarely used in ICs, but rarely is not the same as never. I'm not experienced in IC design, however, so that's all I can say on that topic.

\$\endgroup\$
5
  • \$\begingroup\$ And then, you have a combination of the two in form of an IGBT ... \$\endgroup\$
    – mow
    Oct 13, 2022 at 9:52
  • \$\begingroup\$ @mow Yes, though those are mostly a compromise to get higher voltage ratings. I wouldn't be surprised if IGBTs become obsolete old technology if and when SiC MOSFETs (or something else that can achieve those current and voltage ratings) become more affordable. \$\endgroup\$
    – Hearth
    Oct 13, 2022 at 14:44
  • \$\begingroup\$ Even disregarding voltage, IGBTs are at advantage when the loss caused by the transistor's implicit diode is lower than that caused by the FET's channel resistance. And they don't have the need to account for the body diode in design. \$\endgroup\$
    – mow
    Oct 14, 2022 at 17:06
  • 1
    \$\begingroup\$ @mow However, when you can get SiC MOSFETs with a channel resistance in the single-digit milliohms (although they are currently rather expensive), I think you won't hit that break point until the currents get rather unreasonable for a single device to handle either way. \$\endgroup\$
    – Hearth
    Oct 14, 2022 at 17:10
  • \$\begingroup\$ @mow The main advantage of IGBTs right now is economic. They're far cheaper than the fancy new technology that is SiC (and potentially also GaN, though currently those aren't made--just lower-voltage, higher-current devices called HEMTs) MOSFETs. That advantage may well go away as wide bandgap semiconductors get cheaper--but it remains to be seen how much cheaper they will get. \$\endgroup\$
    – Hearth
    Oct 14, 2022 at 17:11
5
\$\begingroup\$

BJTs consume current while idle. That makes them unsuitable for high levels of integration because there is only so much power you can afford per component on a multimillion-transistor die. MOSFETs have wide parameter variation which makes them more amicable to digital than analog applications and/or requires larger support circuitry around them which makes them more suitable for integrated circuits.

Also for discrete MOSFET applications, the body diode often ends up a real nuisance, and it is almost universally not externally available but tied to the source which can be a real nuisance for a number of applications.

I recently made a switchable phantom power supply to an existing amp and did use a dual MOSFET. The wanted behavior essentially makes this a no-brainer: it's binary on/off, a low on-resistance is desirable, acting on a voltage is appropriate (in this case, the voltage after a pull-up resistor lighting an LED or not, depending on a switch).

And even in this simple application, the body diode made for a lot of nuisance to avoid having it come into play: if the inputs (on a combo XLR/TRS) socket are separated with capacitors, the XLR input where the phantom power may end up has to have (even if high-value) pull-down resistors for draining the capacitors if phantom power is off. If phantom power is switched on without significant DC load, signal amplitude must not exceed the body diode voltage.

It's clearly the most suitable component to use here, but it comes with a bunch of nuisances in the discrete application that go away in most integrated circuits where the substrate is not tied to the FET source but is a separate die layer that will get tied to one of the supply voltages.

Add to that that you have to fear the part going to meet its maker until you managed to solder the gate into some voltage domain shared with source/drain because really small charge differences will blow out the gate when handling.

\$\endgroup\$
1
\$\begingroup\$

Because, CMOS have following advantages compare to BJT

  1. high switching speed
  2. low power consumption
  3. low output noise etc
\$\endgroup\$
1
  • 2
    \$\begingroup\$ Your answer could be improved with additional supporting information. Please edit to add further details, such as citations or documentation, so that others can confirm that your answer is correct. You can find more information on how to write good answers in the help center. \$\endgroup\$
    – Community Bot
    Oct 13, 2022 at 4:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.