Author Topic: MOSFETs versus BJT in discrete designs  (Read 3652 times)

0 Members and 1 Guest are viewing this topic.

Offline DMartensTopic starter

  • Contributor
  • Posts: 30
  • Country: ph
MOSFETs versus BJT in discrete designs
« on: November 20, 2023, 04:23:06 am »
When it comes to integrated circuits, MOSFET technology have overtaken BJT technology for a long time now. Most new ICs are implemented using some form of FET technology, whether (A)LS, HCT, etc.
However, I notice that when it comes to designs with discrete components, most electronics engineers will still go for BJTs instead of MOSFETs. Why is that? I doubt price / availability is the answer, nor speed / performance / choice etc. So what are some of the reasons?
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: MOSFETs versus BJT in discrete designs
« Reply #1 on: November 20, 2023, 09:21:29 am »
Well, address confirmation bias first.

From what era are you mostly seeing these designs?

Hint: even if it's current, posted on the web, etc., a vast number of designs have heritage back to the 70s, 60s even.  Almost anywhere you see 2N2222(A) (partial points for PN2222A), you can bet someone is either copying directly, mutating with minor changes, or effectively remixing such old content.  (Blindly reusing a long-irrelevant part like 2N2222, I would argue counts as remixing: it's the meme among BJTs.)

There is a strong heritage, or bias, for BJTs in audio designs as well, which is I suspect better justified overall, but audio is also one of those subsets where practitioners regularly go out of their way to do things differently.  Often "different" as in "wrong", as judged in strict engineering terms.  There are a number of designs with MOSFETs involved (JFETs too, even), but this can be as often as not read in terms of that forced difference, as it is real engineering merit.

And then, "everything else" probably includes what, largely power and digital?  And BJTs have been largely out of power electronics for a very long time, with a few surviving niches (mainly extremely cost-sensitive, modest efficiency, medium to low power applications, mostly lighting and small power adaptors), while logic I would say uses a fairly even mix of MOSFETs and BJTs, usually I would say depending on the user's experience with them, and knowledge of the circuit they're working on.

For that matter, TVs, CRTs, used BJTs for a very long time as well, even as MOSFETs became better generally; this is only partially by accident, as MOSFETs didn't improve quite as quickly as needed, by the time CRTs were obsoleted (namely, that horizontal output transistors require quite high voltages (typically 1500Vcbo), and quite high currents for the largest high-res monitors (10A+ peak), which wasn't feasible with planar MOSFETs, but is with modern SuperJunction types, but they came along, eh, 5-10 years too late to offer, eh, merely marginal improvements to CRT operation, if any?).  Meanwhile, CRTs, especially high resolution, Trinitrons and such, have quite complex circuitry where every quirk of every component matters, and I would expect it would be challenging to preserve the highly-optimized linearity of a BJT line output circuit while migrating it to use a SuperJunction MOSFET.  (That said, it's not that they didn't touch them at all: S-correction was a typical application for modestly-rated MOSFETs.)

Probably some things about automotive too; Darlington transistors probably stuck around a long time?  Internally protected MOSFETs are all the rage these days, but I don't know what a real market breakdown is like these days.

In any case, LED control remains a useful application; BJTs are easy to bias, fine at dissipating power at up to modest voltages (10s, maybe low 100s), and so are good for CCSs (indication or lighting).

BJTs also excellent for certain linear, switching and oscillator applications.  ZVS "Royer" (actually Baxandall) oscillator is particularly appealing with BJTs, I mean it's quite good with MOSFETs too, but BJTs are (were?) a common sight for CCFL supplies for example.

As for actual engineering merits, independent of whether any particular designer is aware of them, or using them in furtherance of an actual spec (like bandwidth or efficiency) or for optimization (whether a specified parameter or yak-shaving): BJTs hold the distinction of a higher linear-mode figure-of-merit, defined as gm / (Cin + Cout) for example.  That is, the input takes less voltage, and the output has more gain, puff for puff, than MOSFETs do.

There is also a practical side, in that small MOSFETs simply don't exist; RF types disappeared long ago, and I mean, good luck finding a BFR92 today too (which, looks like Infineon is still making some variant thereof and a few relatives, but everyone else's is obsolete; but then, I dare you go looking for the complement BFT92..!), but among MOSFETs the similarly sized parts simply haven't existed for decades.  2N7002 remains a staple despite its age, but its ~30pF junction is massive in comparison to the ~4pF of a MMBT3904, on top of a 3-5V working Vgs(on) range.

Speaking of complements, MOSFETs there simply is no such thing, only hand-wavey use-alikes.  BSS84 is often quoted as a complement to 2N7002, but it doesn't really mean much; PMOS is either ~2.5 times more capacitance (same Rds(on)), 2.5 times more Rds(on) (same Ciss etc.), or somewhere inbetween; and with Vgs(th) individually varying a volt or so, matching pairs doesn't really mean anything in terms of that parameter either.  Whereas BJTs, PNP and NPN are remarkably similar, PNP generally being about 10% poorer performance, something like that, and there are some notable differences in curves especially hFE at low Ic, and fT, but complementary pairs are definitely a meaningful thing among them.

Finally, if you want to build out a modest-scale discrete circuit, and want to push the performance envelope say in terms of functionality and efficiency per supply current consumption, you will be using BJTs, you will be taking advantage of complementary pairs, and other tricks (as well as you can, given the limited matching, especially thermally so, of discrete devices), to reduce component count (layout space, assembly cost, additional load capacitance, increased supply consumption?), and you'll still end up with a laughably inferior circuit to even a fairly simple IC -- if one existed for your particular application, which I'm guessing given the effort required of this hypothetical scenario, doesn't exist.

An example would be this current-limiting electronic fuse design I made a bit ago,
https://www.seventransistorlabs.com/Images/LimitingFuseSch.png
which boasts pretty low supply consumption in most any state, and a little more during active limiting (peak current up to a few mA during gate voltage slewing).  Speed could be improved with RF transistor types, but nothing's going to come anywhere close to the price of BC847/57 pairs, and the limit is mostly diff pair + VAS bias + gate capacitance, and is fast enough (10-20µs response time from load shorting to nominal current limiting).

Incidentally, the MOSFET is both optimal for steady-state efficiency (Rds(on) performs better than Vce(sat); no Ib requirement), and the modest voltage swing makes it easy to compensate (it's a voltage gain node so C5 doesn't need to be large to do its job).  A BJT circuit would have to resolve not only saturation (a sort of Baker-clamped Darlington to maintain low Vce(sat) while being miserly with Ib?) but probably need the error amp and compensation done in a current-mode scheme (since the voltage change in Vbe will be quite small), or an additional stage to get a transconductance characteristic, either way taking up many more transistors.  (Which can of course be justified, and readily integrated, on an IC, hence all the bipolar LDOs with good specs available today, particularly low Iq vs. Iout and over Vin (including in dropout), and compensation into ceramic capacitors.)

Why bother?  There are E-fuse parts readily available today; but they're largely (entirely??) integrated, no external switch.  There are some load switch, hot-swap, wired-OR, etc. controllers that use an external MOSFET, but few with active current limiting AND adjustable SOA.  And, noteworthy, no such circuit can be made of IC building blocks -- even an op-amp of comparable GBW (100s kHz) would take as much Icc, let alone the comparator(s) or timer or whatever to handle the SOA limit function, and the latch on/off and autorestart functions.

And despite all this, it was still just a design exercise; I didn't have (and still don't) an intended application or use-case for this, and haven't produced and sold any units (I probably should just make a few some day, and put them up somewhere and see how that goes, but, meh?).  I don't expect that I can sell many at a price point that would be in any way economical for me to do so, and it's just not so fancy a product to justify such a price.

But if you have additional restrictions, or limitations, like an import-restricted country where only basic BJTs and whatnot are available, well, gotta do what you gotta do of course, and that will include compromises in assembly cost, performance, etc.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Online coppercone2

  • Super Contributor
  • ***
  • Posts: 11342
  • Country: us
  • $
Re: MOSFETs versus BJT in discrete designs
« Reply #2 on: November 20, 2023, 09:34:54 am »
well mosfet are creepy transistors if you probe them it might change states for a while, it feels like dealing with a sticky icky card house or like duct flaps etc (the gate between your dryer and the outside that might get stuck)



creepy icky transisticy
« Last Edit: November 20, 2023, 09:40:36 am by coppercone2 »
 

Online SiliconWizard

  • Super Contributor
  • ***
  • Posts: 15801
  • Country: fr
Re: MOSFETs versus BJT in discrete designs
« Reply #3 on: November 21, 2023, 04:01:48 am »
However, I notice that when it comes to designs with discrete components, most electronics engineers will still go for BJTs instead of MOSFETs.

First off, "most", I don't know?

Other than that, many (I would not say "most") have a harder time understanding the behavior of MOSFETs, in particular when dealing with their linear region. BJTs look simpler in comparison. On the surface.
Of course, there are applications for which BJTs are more appropriate than MOSFETs. But when that's not the case, it's often either due to design habits, or a lack of understanding. So, the comfort zone.

As an illustration, to stay on the surface of things (getting deeper can get hairy fast), how many can give the simplified equation giving the collector current as a function of the base current of a BJT? Now, how many can give the simplified equation of the drain current vs. Vgs and Vds of a MOSFET? What about the saturation region?

So while I wouldn't say "most", as it sounds kinda bold, IME outside of engineers having specialized in microelectronics, relatively few properly understand how MOSFETs work, and when they use them, they tend to stick to using them as switches and at most care about driving them with enough voltage and with enough peak current to overcome the gate capacitance. Beyond that, you'll probably look at 20% of EEs or something like this.
 
The following users thanked this post: Someone

Offline magic

  • Super Contributor
  • ***
  • Posts: 7454
  • Country: pl
Re: MOSFETs versus BJT in discrete designs
« Reply #4 on: November 21, 2023, 09:53:40 am »
Discrete MOSFETs have some characteristics which make them at least a little annoying in analog design.

Higher Vgs than typical Vbe of bipolars, which can be problematic in low voltage circuits needing to work close to rails.
Higher production spread of Vgs, making matching harder.
Plentiful 1/f noise (this also applies to CMOS ICs, although some of the newer ones are not as bad as some of the older ones).
Very, very rarely rated for SOA at DC.


OTOH, the lack of base current is a welcome property which makes FETs useful in applications like precision current sources or active loads.
« Last Edit: November 21, 2023, 09:56:39 am by magic »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: MOSFETs versus BJT in discrete designs
« Reply #5 on: November 21, 2023, 10:08:47 am »
Very, very rarely rated for SOA at DC.

All good points; just curious about this one.  "Very very rare" sounds to me like, oh I don't know, maybe 1/1000?  1/10,000 or less?  Like, catching that shiny new Pokemon is just merely "rare", or maybe "very", at, whatever it is, 1/400ish say?  People grind for hours just to get one, but I certainly don't have such bad luck finding such datasheets.

I haven't exactly been keeping count of course, but it feels like the 10-30% range, or at least order of magnitude, for datasheets where I've seen DC SOA.  So I'm just wondering what your meaning of "very very" is here, or if we've seen very very different samples of datasheets in our travels?(..!)

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 
The following users thanked this post: Someone

Offline dmills

  • Super Contributor
  • ***
  • Posts: 2093
  • Country: gb
Re: MOSFETs versus BJT in discrete designs
« Reply #6 on: November 21, 2023, 01:09:21 pm »
For modest currents at modest voltage in anything that looks like switching, mosfets are generally better then BJTs, but this is far less clear when operating in an application that requires linear operation (Really exponential or square law).

BJTs tend to have far more transconductance then the equivalent mosfet, which sometimes matters.
Mosfets don't tend to hit the second breakdown limit on power handling anything like as easily, but have poorly defined threshold voltage.

Different devices for different applications really, but the mosfet does tend to rule supreme for all but the highest power switching applications, where the bipolar transistor comes back in the form of the IGBT because eventually I^2 * Rds(on) exceeds I * Vce(sat), so at a big enough current the efficiency advantage swings to the BJT structure as long as you can deal with the glacial speed.

Given that lots of things are at least superficially digital these days, the mosfet is predictably popular in that space. 

Jfets major use is perhaps analogue switching and very high impedance preamps for sensors, lowish transconductance however, but when you want Giga ohm inputs... 
 

Offline the_cake_is_a_lie

  • Contributor
  • Posts: 32
  • Country: us
Re: MOSFETs versus BJT in discrete designs
« Reply #7 on: November 22, 2023, 12:49:06 am »
Lots of good answers but I didn't see two points. Bigger point is BJTs are more beginner-friendly and most people in electronics are and will remain beginners. BJTs are more ESD resistant and you may get away with leaving unused inputs/outputs floating on their logic gates and opamps. Easier to bias been said. Intro to transistor info online is a good 10x more plentiful for BJTs by extension. Then internet beginners go on to use BJTs in their designs, even when FETs would be vastly superior, because they never learned them. The cycle repeats.

Lesser point is Vgs(th) can vary enough in FETs, particularly JFETs, to require hand-matching in (beginner-friendly) low voltage circuits. J109 Vgs is -2V to -6V and -0.5V to -7.5V for BF256B.

Yes, it's easier to match NPN to PNP than NMOS to PMOS but MOSFETs arrays with an N and a P intended to match exist. Can invert phase on the other input to use 2x NMOS as well.
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 17429
  • Country: us
  • DavidH
Re: MOSFETs versus BJT in discrete designs
« Reply #8 on: November 22, 2023, 06:50:42 am »
I will add that historically bipolar transistors support a higher current density and scale their area at less than the square of their voltage rating, making them less expensive because a given voltage and current rating will require less die area.  (1) This may have changed with "SuperJunction" MOSFETs, and economy of scale may have largely eliminated this advantage for bipolar transistors.

For power management integrated circuits where the linear or switching pass element takes up significant area, bipolar processes are still competitive if not better than MOSFET processes because of their area advantage.

Complementary high frequencies designs could rely on RF PNP transistors, although as of recently that is no longer the case.  Fast p-channel MOSFETs have not been available for a long time; I think Signetics sold them.

(1) This is what makes IGBTs more economical than MOSFETs at high voltages despite being slower.

BJTs hold the distinction of a higher linear-mode figure-of-merit, defined as gm / (Cin + Cout) for example.  That is, the input takes less voltage, and the output has more gain, puff for puff, than MOSFETs do.

An underappreciated side effect is that bipolar transistors are faster; higher transconductance means more current to charge and discharge their smaller parasitic capacitance.
« Last Edit: November 22, 2023, 07:02:20 am by David Hess »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: MOSFETs versus BJT in discrete designs
« Reply #9 on: November 22, 2023, 09:29:11 am »
Yeah, SJ FETs do very well; despite the complexity in producing them, they are cheap and plentiful and well represented among Chinese brands even (though I'm not clear on the balance between Taiwanese vs. mainland parts, I don't have that much experience with the market over there); 400-500V parts have even mostly disappeared in favor of 600, 650, 700V parts, which presumably perform a little bit worse than 400 and 500V parts would, but not by so much that they're worth tuning an entire process for.

Curiously (or maybe not), SJ FETs are, AFAIK, basically planar devices as usual, on the top layer. Like old school HEXFET: lateral channel with self-aligned gate on top, well, not hex patterned but stripe as usual, but the same cross-section.  With the source diffusions aligned to the N-pillars, they are also the preferred avalanche site, just as was the case for ye olde planar (but the gate oxide is close enough / not well enough shielded from avalanche-generated hot carriers, that repetitive avalanche ratings are obtained, at least not regularly from what I've seen).

They could just as well do trench, I would think, but I'd guessing the areal density is limited more by the pillar thickness and doping density, and right now it just doesn't matter what they do on top.  Possibly consequent to this is the limited Id(max), i.e. it doesn't matter if Vgs(on) is much over 8V or so, Id just doesn't go up past a point; at least on a few types I've tested.  (I'm not sure offhand if that's a consequence of the parasitic JFET, or more fundamental from the area density.)

The biggest downside is actually the SJ structure itself; it has dielectric loss, because depletion charge must flow along the thin webs of the pillars, which have very high resistance in comparison to the massive capacitance.  At low rates (dV/dt), it looks like a battery (Qoss charge is stored at a nearly constant voltage), but at realistic switching rates, it causes "drool" in turn-off current (oddly, an effect similar to IGBT switching, for a very different reason!).  Switching loss is still improved, since, I mean hell, Coss is a good like five times better than planar ever were, it's the difference between being able to do something at all, versus do it at high efficiency.

As for power BJTs, it's a shame they never achieved much speed, ironically given that they generally perform better for small signal, and maybe switching, applications.  Part of this is the transit time across the junction, which is a limiting factor in MOSFET operation too.  (Compare the 600V+ RF MOSFETs from Microsemi, versus the usual 40-80V ones; granted, it's anyone's guess which directions they're optimized in, or how thoroughly, but they're in any case a far shot from the GHz+ LDMOS commonly available.)  I suppose with minority carrier dynamics, and mandatory low doping levels, there's just not much you can do about it, even with perforated-emitter or whatever techniques.  Not sure if anyone's making a SJ BJT yet, or tried, but that would be one way to keep doping up, and also extend voltage ratings (since you can just make the pillars ever taller, but a graded junction has breakdown ~inverse to doping level).

I wonder if that's of interest to super high voltage Si parts.  Can SJ even be made consistent enough to work over a whole die, could SJ SCRs (or GCTs or IGBTs) be practical?  A 6kV+ IGBT with sub-us switching times would be fascinating.

SiC by the way are essentially planar ~100V Si designs, reformulated; the junctions are far too thin to employ SJ yet (which also limits its application for <= 200V Si FETs*).  Which means SiC performs doubly well (SiC advantage plus no SJ dielectric loss).

*But, that said, TI's NexFET design is a peculiar, sort of, lateral dual gate cascode structure (with the 2nd gate just hard grounded to give excellent Crss shielding), made on a fine enough process pitch that its current density remains competitive with trench.  A key part of the structure is a vertical finger up to the lateral channel, both of which become depleted under Vds bias; thus Coss(Vds(max)) / Coss(0) is quite small.  The Coss(Vds) curve is not quite as sharp as SJ, but steeper than planar.
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline tszaboo

  • Super Contributor
  • ***
  • Posts: 8218
  • Country: nl
  • Current job: ATEX product design
Re: MOSFETs versus BJT in discrete designs
« Reply #10 on: November 22, 2023, 11:23:57 am »
Always? I use a mix of P and N, PNP and NPN. Turning on a MOSFET on a 1.8V MCU is more difficult. An always on NPN will use more power than an NFET. A PNP transistor could be more suitable to dissipate some power because the power pad of it is connected to the ground and not the series resistor. There are different requirements, different design goals, and I don't know why would a good engineer, who has an understanding on how these 4 devices work would just use one type.
 

Offline Alex Nikitin

  • Super Contributor
  • ***
  • Posts: 1330
  • Country: gb
  • Femtoampnut and Tapehead.
    • A.N.T. Audio
Re: MOSFETs versus BJT in discrete designs
« Reply #11 on: November 22, 2023, 02:35:42 pm »
I like FETs and made many FET-only and even MOSFET-only analogue circuits, mostly for audio applications, from phono stages to power amplifiers. The lack of proper P-ch complements is not necessarily an obstacle. I've designed several amplifiers with N-ch only MOSFET output devices, these amplifiers were manufactured for many years and thousands of units were produced, some receiving top awards, including Stereophile Class A recommendation. I would say that every type of an active device has some advantages and some problems, and it is up to us to use them wisely.

Cheers

Alex 
 

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 4120
  • Country: us
Re: MOSFETs versus BJT in discrete designs
« Reply #12 on: November 22, 2023, 04:29:09 pm »
Having a career that spanned over half a century we've been involved 1st hand from discrete Bipolar, MOS, SiGe, GaAs and GaN design and ICs in CMOS, BiCMOS, SiGe and InP design.

Simple fact is the semiconductor technology is driven by Digital CMOS, nobody is building new fabs for anything other than Digital CMOS (pushing feature size limits), outside a few maybe for GaN or SiC, and boutique fabs for exotic types like InP.

We've utilized the early bipolar from Fairchild, first BiCMOS analog from RCA, their 4000 series CMOS (we actually used this for analog as well), National's bipolar and JFET and so on and discovered the advantages and disadvantages of each technology for analog use.

A fundamental fact about comparing MOS with Bipolar is, Bipolar has more transconductance (for given area), better matching and temperature tracking than MOS, and somewhat faster to a limit (3nm CMOS is really really fast, but not as fast a bipolar devices in InP), MOS has lower Ron (for given area) and no saturation voltage and obviously no gate current (other than leakage).

However the advantages of bipolar have been displaced over time by digital techniques and folks being forced to design in a Digital Friendly CMOS process as things became more integrated with Mixed Signal over time.

Chopper Stabilized and Commutating Auto-Zero techniques allowed pure CMOS to surpass bipolar designs in offset and 1/f noise regions and fully integrable with Digital CMOS. Almost all new ADCs are processed in Digital CMOS, same for DACs and some of the best performing Op-Amps are pure CMOS.

The MOS switch parameters can't be emphasized enough, they are so powerful and allow digital signal processing techniques to be directly applied for analog use. Even displacing the last "Analog" holdouts, the RF and Microwave worlds, with things like Software Defined Radios, pushing the high speed CMOS ADCs right up to the antenna and with Dr Joe Mitola's Cognitive Radio concepts from the 90s, and the PPM or N-Path Mixer, where traditional analog techniques are yielding to more digital like with MOS switches and producing results not possible with pure analog old school techniques (like ~1dB NF passive microwave mixers).

Ironically the much touted BiCMOS technology in it's day originated from the Digital world, as the need for higher output drive capability (weak PMOS) which the bipolar emitter follower buffered CMOS provided until the smaller feature size CMOS eliminated the need for the bipolar buffer and replaced with a pure CMOS buffer.

Even SiGe bipolar has it's modern origins based upon Digital CMOS, IBM was developing SiGe bipolar technology for a new faster ECL/CML logic for the next generation IBM360 mainframe. During the SiGe development IBM decided on pure CMOS multi-core chip sets instead of a fast single core SiGe bipolar design, but allowed the SiGe development to continue with analog emphasis which ended up in a SiGe BiCMOS process. 

Anyway (can't believe I'm saying this being an old analog type), but Digital CMOS has actually been highly beneficial to the analog world and likely ending the era of the bipolar transistor and designs with such. No one is designing new chips or circuits based upon bipolar technology today (maybe a few small areas like voltage references), and no new bipolar fabs are being built, everything is moving or already has to CMOS.

Hopefully we'll still be able to get 2N3904 and 2N3906s in the future as some ancient bipolar fab in someone's garage will still be cranking these out, maybe we all should get a stash now while we can just in case ???

Best,
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 

Online coppercone2

  • Super Contributor
  • ***
  • Posts: 11342
  • Country: us
  • $
Re: MOSFETs versus BJT in discrete designs
« Reply #13 on: November 22, 2023, 04:42:11 pm »
what we need is cheaper fabs so we can get what we want instead of relying on piggyback crap

things may not be as grim as you imagine if fab tech gets economized or new construction methods for transistors occur.

Some kind of new manufacturing machine. The 3d printer was amazing, maybe there is something that can be made for larger transistors. Years ago I heard the same thing about plastic parts that you will never be able to get high quality complicated plastic parts without a tool and die/mold guy, its only kind of true, even for high quality metal parts.

FAB are vastly overpriced because of the secrecy too. I know that. even adding useless steps to throw off the competition or obsfurate manufacturing technique. and its pretty easy to find someone that says they were ignored about process improvements. All the fab stuff is so tight budget its all based on probability of projections coming through to put ANY money into improving ANYTHING. This is mostly based on hearsay amongst businessmen.... you might consider alot of tech being left behind at a expensive legacy level simply because the bosses think demand might decrease making it unprofitable based on the company history not the cost of the capital (true for any technology outside of the say middle 25% of the bell curve in terms of actual demand and projections). That is, the line runs with the upmost priority being lowest possible severance costs (meaning absolutely bare minimum input towards upkeep and zero input for improvement, for instance something as simple as replacing some manual pressure gauges and valves with a PLC that automates part of some process at costs under 10k dollars when its not even the cost of a months return). this kind of stuff exists! It sometimes even happen because there is no growth expected, even if the demand is not expected to change, if the business strategy chosen happens to be "only put money into things that are likely to grow! we are not even interested in reducing day to day and manufacturing cost because we might grow elsewhere, its what the board wants this quarter!!!!!" this results in something being unbelievably bad but IT EXISTS!!!!!! and if someone wants to do a external industry-info cost study of such a thing, you will get a fucking stone wall bullshitter that makes you think its all running at peak fucking efficiency and its been studied to death, your talkin north korea levels of honesty here

and then there is the geopolitical side of things. that can make it extra stupid. like mega. when you involve washington, and the pentagon, into costs, the ramifications are only the limits of peoples imaginations.
« Last Edit: November 22, 2023, 05:07:05 pm by coppercone2 »
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 17429
  • Country: us
  • DavidH
Re: MOSFETs versus BJT in discrete designs
« Reply #14 on: November 22, 2023, 04:53:44 pm »
Chopper Stabilized and Commutating Auto-Zero techniques allowed pure CMOS to surpass bipolar designs in offset and 1/f noise regions and fully integrable with Digital CMOS.

But only for low source impedances and low bandwidths where their high current noise from charge injection is not a problem.  Maybe they now make some parts which solve this, but I have not found them yet.
 

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 4120
  • Country: us
Re: MOSFETs versus BJT in discrete designs
« Reply #15 on: November 22, 2023, 05:06:33 pm »
what we need is cheaper fabs so we can get what we want instead of relying on piggyback crap

things may not be as grim as you imagine if fab tech gets economized or new construction methods for transistors occur.

Why we mentioned the garage bipolar fab based upon leftover old processing equipment.

There is/was a discrete fab that employed leftover higher resolution processing equipment and designed discrete transistors (think MOS and Bipolar) with smaller feature size capability for improved performance, recall Zetex Semi.

Best,
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 

Online coppercone2

  • Super Contributor
  • ***
  • Posts: 11342
  • Country: us
  • $
Re: MOSFETs versus BJT in discrete designs
« Reply #16 on: November 22, 2023, 05:07:56 pm »
nah that's not an improvement thought I would not be surprised if someone figured out how to do it twice as cheap and twice as fast if they had the freedom of a garage available with that equipment. i mean some kind of a new machine that someone might make, a different process. there has got to be more creative solutions.

but even any reasonably agile company acquiring this stuff from the old IBMesque places is probobly gonna figure at least 10 awesome things out  :-DD

https://youtube.com/shorts/mGU0pHwNWLk?si=iRjUk-E4eRiOuJJ9
« Last Edit: November 22, 2023, 05:15:24 pm by coppercone2 »
 

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 4120
  • Country: us
Re: MOSFETs versus BJT in discrete designs
« Reply #17 on: November 22, 2023, 05:14:13 pm »
Chopper Stabilized and Commutating Auto-Zero techniques allowed pure CMOS to surpass bipolar designs in offset and 1/f noise regions and fully integrable with Digital CMOS.

But only for low source impedances and low bandwidths where their high current noise from charge injection is not a problem.  Maybe they now make some parts which solve this, but I have not found them yet.

Yeah that's always been an issue, even the CMOS ADCs have this "feature" and why one must pay close attention to the driving source impedance/dynamics of the source driving said ADC. Some Op-Amps were/are specifically designed to drive these Hi Speed ADC inputs without significant induced "ringing" for the dynamic charge injection. Heck even DMMs can do this, we had one long ago (can't remember name tho) that corrupted anything active it measured, a swift hammer correct this potential future problem ;)

Best,
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 

Online Someone

  • Super Contributor
  • ***
  • Posts: 5156
  • Country: au
    • send complaints here
Re: MOSFETs versus BJT in discrete designs
« Reply #18 on: November 22, 2023, 10:27:56 pm »
Chopper Stabilized and Commutating Auto-Zero techniques allowed pure CMOS to surpass bipolar designs in offset and 1/f noise regions and fully integrable with Digital CMOS.
But only for low source impedances and low bandwidths where their high current noise from charge injection is not a problem.  Maybe they now make some parts which solve this, but I have not found them yet.
There are chopper parts coming into the market where the manufacturer is confident in use at "high" impedances. OPAx189 is pretty well proven at this point.
 

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 4120
  • Country: us
Re: MOSFETs versus BJT in discrete designs
« Reply #19 on: November 23, 2023, 12:54:56 am »
what we need is cheaper fabs so we can get what we want instead of relying on piggyback crap

things may not be as grim as you imagine if fab tech gets economized or new construction methods for transistors occur.

Some kind of new manufacturing machine. The 3d printer was amazing, maybe there is something that can be made for larger transistors. Years ago I heard the same thing about plastic parts that you will never be able to get high quality complicated plastic parts without a tool and die/mold guy, its only kind of true, even for high quality metal parts.

FAB are vastly overpriced because of the secrecy too. I know that. even adding useless steps to throw off the competition or obsfurate manufacturing technique. and its pretty easy to find someone that says they were ignored about process improvements. All the fab stuff is so tight budget its all based on probability of projections coming through to put ANY money into improving ANYTHING. This is mostly based on hearsay amongst businessmen.... you might consider alot of tech being left behind at a expensive legacy level simply because the bosses think demand might decrease making it unprofitable based on the company history not the cost of the capital (true for any technology outside of the say middle 25% of the bell curve in terms of actual demand and projections). That is, the line runs with the upmost priority being lowest possible severance costs (meaning absolutely bare minimum input towards upkeep and zero input for improvement, for instance something as simple as replacing some manual pressure gauges and valves with a PLC that automates part of some process at costs under 10k dollars when its not even the cost of a months return). this kind of stuff exists! It sometimes even happen because there is no growth expected, even if the demand is not expected to change, if the business strategy chosen happens to be "only put money into things that are likely to grow! we are not even interested in reducing day to day and manufacturing cost because we might grow elsewhere, its what the board wants this quarter!!!!!" this results in something being unbelievably bad but IT EXISTS!!!!!! and if someone wants to do a external industry-info cost study of such a thing, you will get a fucking stone wall bullshitter that makes you think its all running at peak fucking efficiency and its been studied to death, your talkin north korea levels of honesty here

and then there is the geopolitical side of things. that can make it extra stupid. like mega. when you involve washington, and the pentagon, into costs, the ramifications are only the limits of peoples imaginations.

Obviously you don't have a semiconductor background, nor bother to study up on things related. Today a new SOTA CMOS fab costs north of $25B, takes over 5 years to come on line and produce the first penny of revenue, and requires continual investments per year well over $1B just to keep up!! TSMC is leading the CMOS charge followed by Samsung, and Intel trying to play catch up!! The USG is trying help the US catch up to what they let out of the bag long ago, thanks to both DC parties involved, and now eventually trying to help US semiconductor technology play catch up.

TSMC with some USG $ assistance, is building 2 SOTA CMOS fabs in US (Az), but coming on-line being delayed because of lack of high level US semiconductor talent, embarrassing indeed!!!

Apple's new M3 chip sets are in TSMC 3nm CMOS process and the most advanced chip has 25 Billion MOS transistors!! If a transistor costs just 1 penny, this chip would cost $250,000,000.00 and yet you can buy a complete Apple computer today for considerable less!! Good example of the massive semiconductor investments paying off in the long run, hopefully someday Apple and Nividia chips will be fabricated on shore!!

We still have a few shinning spots in US advanced technology, semiconductors used to be at the top of that list, but sadly has faded over the years, hopefully the future will bring some of this technology back home.

BTW I do recall something about plastic transistors from way back, and some guy making chips/transistors in his garage, but can't remember details. Funny, that made me think of as a kid about 10~11 getting a couple diodes and connecting them to make a transistor :o

Best
« Last Edit: November 23, 2023, 01:15:15 am by mawyatt »
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 17429
  • Country: us
  • DavidH
Re: MOSFETs versus BJT in discrete designs
« Reply #20 on: November 23, 2023, 02:10:14 am »
Yeah that's always been an issue, even the CMOS ADCs have this "feature" and why one must pay close attention to the driving source impedance/dynamics of the source driving said ADC. Some Op-Amps were/are specifically designed to drive these Hi Speed ADC inputs without significant induced "ringing" for the dynamic charge injection. Heck even DMMs can do this, we had one long ago (can't remember name tho) that corrupted anything active it measured, a swift hammer correct this potential future problem ;)

I keep my oldest Tektronix bench DMMs in repair because their linear differential JFET input stage produces no charge injection.  They can make some difficult measurements that modern DMMs cannot always be relied on to make accurately.

There are chopper parts coming into the market where the manufacturer is confident in use at "high" impedances. OPAx189 is pretty well proven at this point.

I have heard that before from marketing.  In the past it meant they found a new way to obfuscate input current noise specifications in their datasheets.

I will have to get some OPAx189s to test, and Analog devices has a similar part, however from the OPAx189 datasheet:

Zero-drift amplifiers such as the OPAx189 use switching on the inputs to correct for the intrinsic offset and drift
of the amplifier. Charge injection from the integrated switches on the inputs can introduce short transients in
the input bias current of the amplifier. The extremely short duration of these pulses prevents the pulses from
amplifying, however the pulses may be coupled to the output of the amplifier through the feedback network.
The most effective method to prevent transients in the input bias current from producing additional noise at the
amplifier output is to use a low-pass filter such as an RC network.

 

Online coppercone2

  • Super Contributor
  • ***
  • Posts: 11342
  • Country: us
  • $
Re: MOSFETs versus BJT in discrete designs
« Reply #21 on: November 23, 2023, 03:06:39 am »
what we need is cheaper fabs so we can get what we want instead of relying on piggyback crap

things may not be as grim as you imagine if fab tech gets economized or new construction methods for transistors occur.

Some kind of new manufacturing machine. The 3d printer was amazing, maybe there is something that can be made for larger transistors. Years ago I heard the same thing about plastic parts that you will never be able to get high quality complicated plastic parts without a tool and die/mold guy, its only kind of true, even for high quality metal parts.

FAB are vastly overpriced because of the secrecy too. I know that. even adding useless steps to throw off the competition or obsfurate manufacturing technique. and its pretty easy to find someone that says they were ignored about process improvements. All the fab stuff is so tight budget its all based on probability of projections coming through to put ANY money into improving ANYTHING. This is mostly based on hearsay amongst businessmen.... you might consider alot of tech being left behind at a expensive legacy level simply because the bosses think demand might decrease making it unprofitable based on the company history not the cost of the capital (true for any technology outside of the say middle 25% of the bell curve in terms of actual demand and projections). That is, the line runs with the upmost priority being lowest possible severance costs (meaning absolutely bare minimum input towards upkeep and zero input for improvement, for instance something as simple as replacing some manual pressure gauges and valves with a PLC that automates part of some process at costs under 10k dollars when its not even the cost of a months return). this kind of stuff exists! It sometimes even happen because there is no growth expected, even if the demand is not expected to change, if the business strategy chosen happens to be "only put money into things that are likely to grow! we are not even interested in reducing day to day and manufacturing cost because we might grow elsewhere, its what the board wants this quarter!!!!!" this results in something being unbelievably bad but IT EXISTS!!!!!! and if someone wants to do a external industry-info cost study of such a thing, you will get a fucking stone wall bullshitter that makes you think its all running at peak fucking efficiency and its been studied to death, your talkin north korea levels of honesty here

and then there is the geopolitical side of things. that can make it extra stupid. like mega. when you involve washington, and the pentagon, into costs, the ramifications are only the limits of peoples imaginations.

Obviously you don't have a semiconductor background, nor bother to study up on things related. Today a new SOTA CMOS fab costs north of $25B, takes over 5 years to come on line and produce the first penny of revenue, and requires continual investments per year well over $1B just to keep up!! TSMC is leading the CMOS charge followed by Samsung, and Intel trying to play catch up!! The USG is trying help the US catch up to what they let out of the bag long ago, thanks to both DC parties involved, and now eventually trying to help US semiconductor technology play catch up.

TSMC with some USG $ assistance, is building 2 SOTA CMOS fabs in US (Az), but coming on-line being delayed because of lack of high level US semiconductor talent, embarrassing indeed!!!

Apple's new M3 chip sets are in TSMC 3nm CMOS process and the most advanced chip has 25 Billion MOS transistors!! If a transistor costs just 1 penny, this chip would cost $250,000,000.00 and yet you can buy a complete Apple computer today for considerable less!! Good example of the massive semiconductor investments paying off in the long run, hopefully someday Apple and Nividia chips will be fabricated on shore!!

We still have a few shinning spots in US advanced technology, semiconductors used to be at the top of that list, but sadly has faded over the years, hopefully the future will bring some of this technology back home.

BTW I do recall something about plastic transistors from way back, and some guy making chips/transistors in his garage, but can't remember details. Funny, that made me think of as a kid about 10~11 getting a couple diodes and connecting them to make a transistor :o

Best

the ones you mention are the hot '25 percent' of the semiconductor manufacturing world. I mean idk the real percentage, but those are the hot looked at ones right now. what I am saying is that some esoteric old op amp process might not have had so much care and optimization and modernization / honest cost analysis done on it, making it possible to improve and not quite so expensive as originally quoted, if bright people are working on it that don't have really strong corporate 'guidelines' holding them back. like my gut feeling from what I have seen and heard is that probobly they don't incorporate as much new technology or optimize the process as much as we think or even spent time researching how to improve or bypass hard parts of the process because of it simply being uninteresting to the R&D elements of a company if they are told to only work on GROWTH areas not STEADY INCOME areas. They often seem to want to focus on things that have MASSIVE potential not incremental improvements. Like they need to do something to look like the promise they made about the GROWTH of the company, which attracts investors, is being honored. What gets people interested is 'we are a 150 million dollar company looking at being a 5 billion dollar company in 3 years' not '300 million'. They just know the market well enough to know that despite the fact there might be good money to be made, its not the kind of money that keeps people interested in supporting them!

that is like talking about 5th avenue and forgetting about the bronx


like it goes hand in hand with moores law. investors love moores law. it means their money has geometric growth potential. the company is looking at ways to at least look like its trying to grow geometrically or whatever the hell you call moores law curve. if their not doing that then its often considered a bad company. like think big shoot for the stars type stuff. because someone is gonna make that promise elsewhere and their gonna take their money else where. so its like 'nah we don't waste our time on that old shit just keep it running titrated to the nearest cent'. But behind the scenes there might be enough data and insight gained to make pretty massive improvements (in terms of available industry research, and company know how, and realizations about exactly what you need to pass QC), just its not agreeing with moores law, so its kinda put to the side, even if its a mainstay. And the contracts I imagine don't make it easy to change nothing, like lawyers get involved, you could pin a delay on someone 'messing' with the production... exposure...
« Last Edit: November 23, 2023, 03:18:38 am by coppercone2 »
 

Offline PartialDischarge

  • Super Contributor
  • ***
  • Posts: 1625
  • Country: 00
Re: MOSFETs versus BJT in discrete designs
« Reply #22 on: November 23, 2023, 03:17:41 am »
I’ll leave this article here, it’s old but with some interesting thoughts, the author is an expert in power electronics and very smart guy
« Last Edit: November 23, 2023, 05:13:12 am by PartialDischarge »
 

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 4120
  • Country: us
Re: MOSFETs versus BJT in discrete designs
« Reply #23 on: November 23, 2023, 04:19:51 am »
Interesting read, thanks, but as you mentioned quite old (1993).

His conclusion hints at an improved Si bipolar transistor structure for high speed and power applications, which we don't recall seeing. At that time Power MOS was beginning to made some inroads into SMPS use with the Siliconix VMOS and IR HexFET technologies which he fails to mention.

Today believe most new SMPS designs are utilizing some form of MOSFET, especially new SMPS chips. For higher Power Density which implies smaller magnetics and thus higher switching frequencies the GaN FET is beginning to win new designs for the main power switching device.

Anyway, the power sector of semiconductors is rather special case compared to the massive digital CMOS world, but also seems to be yielding to CMOS for the lower to moderate power levels, with some special CMOS technologies for higher voltage use with direct mains SMPS now available. If you look at the popular mains USB power modules, the new designs seem to all be CMOS based.

Edit: Remember being on a panel session with Dr. Tom Lee from Stanford in the early renditions of what later became the RFIC Symposium at the IEEE MTT back in 90s. Believed that SiGe bipolar technology would continue into the future and grow due to the many advantages of SiGe bipolar over CMOS, Tom thought otherwise that CMOS would eventually dominate in spite of the SiGe advantages, well guess who was right ???

Best, 
« Last Edit: November 23, 2023, 04:33:40 am by mawyatt »
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf