What's 50 fF?

[ejeffrey]

How big is the structure you are measuring?  What is it's self capacitance?  If that is over 50 fF then there isn't much you can do.

How broad band do you need the match?  If you are OK with narrowband performance you can fix almost anything.

Do you have access to HFSS?  Honestly that's probably the best way to figure it out.  Simulate a structure you think you can build and see what the capacitance is and how close you are to what you need and whether you can get there by tweaking or if you need to find a different transistor.

[ezalys]

Yeah we have HFSS, maxwell, sonnet, all the fun stuff. Sometimes you just want some intuition before you open the software.

[MK14]

The problem with very high speed packaged transistors is the package almost dictates the ultimate performance. Here's where an IC solution allows much better performance because of the physical dimensions involved.

I see, that makes a lot of sense. I assume, one of the reasons why high frequency MCUs/FPGAs (e.g. 500MHz internal clock), have much lower external frequencies, such as (perhaps) 50MHz external port frequency (maximum port toggle rates), and maybe 16MHz external Quartz clock frequencies. With appropriate PLLs and internal clock dividers, to compensate. It probably also reduces the power consumption, as well.

One could indeed make a 555 with some fast transistors, however the performance would be limited by the package, and with other methods to achieve nanosecond or faster pulses likely not of much practical use. However, this would be an interesting project of a EE student!!

Also, unlike a normal NE555/LMC555, which can be designed with, even by beginners of electronics. Just needing a few passives and simple connection techniques. A 2GHz version, would probably need specialist external parts, highly specialized PCB design/construction, and very significant electronics design skill indeed, unless using the standard application note circuits (which would probably exist, if such a part is possible/practicable).
Some parts of it, could probably use techniques, a bit like in my paragraph above. E.g. consider the timing rate adjusting resistor (very rough approximation of how a 555 works). It might be possible for it to simply adjust an internal high quality/performance constant current generator, so externally the timing control resistor is running at almost DC (frequency), but internally, the constant current generator, could be connected to 2GHz sections. Solutions around NOT having a high frequency external capacitor are also probably possible. Such as an internal switchable selection of a few different values, and binary divider modes, for slower frequencies.
It might be possible (when it is an oscillator), to use PLL and/or DDS techniques (internally), to make it controllable externally via very low frequncies, potentially more stable, and other possible benefits
Such modules or ICs already exist, such as the AD9956, which seems to claim 2.7GHz. It is NOT a 555 as such, but maybe could be modified to be more like it.
https://www.analog.com/media/en/technical-documentation/data-sheets/AD9956.pdf
Similarly, perhaps external logic level signals, could change the internal configuration, between monostable, One-Shot or retriggerable, oscillator, PWM and perhaps other modes. But the final output (possibly differential, which could/maybe help), would still be problematic, and perhaps cause big current/power consumption.

Yes ECL and CML (also SCL) is a pain to work with, unless you really need the speed. CMOS 74AC and ACT logic is pretty easy to use for moderately high speeds and a good replacement for the older 74LS and S Logic.

True. I incorrectly (when first reading your post), thought CML was another description of ECL. But no, Current Mode Logic, is something different, again. Interesting as well, as it can give very low power consumption. Despite very high maximum speeds.

An interesting story behind the development of Silicon Germanium HBT bipolar technology goes back to IBM in the 80s, they wanted to get a faster IBM360 Mainframe which was done with ECL and CML, thus wanted a faster bipolar device. Drs Bernard Meyerson and David Harame were tasked with developing a faster bipolar transistor, and they developed the UHVCD technique for depositing the Ge implants in a Si process, thus creating a SiGe HBT. These SiGe devices were considerably faster than equivalent Si devices, but IBM had decided to go with a parallel CMOS CPU solution rather than a single fast CPU and thus looked for another market for these very fast HBT devices. That market turned out to be in the RF/MW/MMW and other very fast analog uses, as well as the orginial intended ECL/CML use. Various flavors of the BiCMOS SiGe process evolved over the decades producing HBT devices with ~500GHz Ft.

Thanks, that was very interesting, and really amazing!
So, something originally intended, for high speed (computer) logic, ironically ended up helping make very high performance analogue stuff!
As regards Drs Bernard Meyerson and David Harame. I have heard that much of the integrated circuit development, really comes from excellent quality (inventive/innovative) Process Engineers, who really do at times, most/some of the work. But often get little or no recognition, for their immense work. E.g.  https://en.wikipedia.org/wiki/Bob_Widlar  who created some useful analogue stuff, including op-amps, a long time ago, and is partly well known in electronics circles. But David Talbert, was also a very important player (possibly incorrect information by me and/or subjective), as the process engineer, yet David Talbert, is probably a lot less well known. Also, I've heard similar criticism, because the bosses/leaders are famous for such things, but it is really the people below them that did all the work, and developed the stuff.
It is tricky. In some cases the leaders were the motivation/innovating force, and deserve the credits they got. But in other cases, they were simply the boss, and didn't really participate in the invention(s).

A long while ago, things/people often seemed to say that Gallium devices would replace Silicon parts in computers, and would make Silicon become obsolete. Clearly that hasn't happened (although I'm typing this, with my LED lighting/monitors/TVs/torches which are Gallium (compound) LED powered, so they were not completely wrong. I also seem to remember that some Gallium semiconductor/IC devices, did appear at some points in time (maybe they still exist).

[T3sl4co1l]

Yeah, there was a Cray (Cray-1?) made with GaAs NMOS.  Which like Si NMOS, was an absolute pig on power dissipation.  GaAs CMOS isn't a thing: hole mobility is a tiny fraction of electron mobility, so the Rds(on)*Cg figure of merit is atrocious.  So it's all just NMOS (which perform, what is it, 2-3x better than Si?) with passive pullup resistors.  And Idunno probably some source followers and the other usual trickery that makes NMOS a little better, but this is largely how NMOS is done, just pullups and open-drain NMOS wired up.

Nowadays, GaAs FETs aren't so important anymore I think, and GaN is the way to go?  It's mature enough that power devices up to (over? I forget) 600V are available; still rather fragile to use, not to mention screaming fast, but when you want the absolute best in efficiency and power density, you can go out and do it today with standard products.

And indeed, GaN is in mainstream production for LEDs, quite nice.

There's also InP which, I forget how the properties compare to GaAs and GaN, or why it's so popular for random monolithic builds exactly, but there's a bunch of crazy stuff like THz distributed amplifiers made on it.

Tim

[mawyatt]

GaAs CMOS isn't a thing: hole mobility is a tiny fraction of electron mobility, so the Rds(on)*Cg figure of merit is atrocious.
Tim

Funny you mention that!! I argued that very Solid State Physics fact in 80s about hole/electron mobility against a GaAs CMOS that Honeywell was developing. They wanted to have a process that would produce a Rad-Hard Space Computer that was also very fast, and even got a bunch of USG funding!! I lost this argument on a number of occasions, and basically was told to "go away, you don't know anything about GaAs!!". They did end up getting it to work, but had to make the P channel devices something like 6~10X the size of the N channel, then it had to work only on negative edge logic because of the weaker and high capacitance P channel (too slow rising edge) and it wasn't fast and very power hungry. Recall the Silicon on Insulator CMOS out performed the GaAs CMOS in almost every way, so after dumping ~$1B into the GaAs CMOS fab process & designs it was abandoned, and nobody remembered those earlier discussions about hole mobility in GaAs

One good thing did come out of this doomed endeavor tho, the GaAs N Channel devices were used in a new very high frequency Switchmode power converter employing a new planar transformer that GE had developed. This ended up in various avionics systems to supply the DC power for the electronics, and saved considerable size and weight. Much later these GaAs N Channel devices were replaced with GaN with even better performance.

Best,

[mawyatt]

An interesting story behind the development of Silicon Germanium HBT bipolar technology goes back to IBM in the 80s, they wanted to get a faster IBM360 Mainframe which was done with ECL and CML, thus wanted a faster bipolar device. Drs Bernard Meyerson and David Harame were tasked with developing a faster bipolar transistor, and they developed the UHVCD technique for depositing the Ge implants in a Si process, thus creating a SiGe HBT. These SiGe devices were considerably faster than equivalent Si devices, but IBM had decided to go with a parallel CMOS CPU solution rather than a single fast CPU and thus looked for another market for these very fast HBT devices. That market turned out to be in the RF/MW/MMW and other very fast analog uses, as well as the orginial intended ECL/CML use. Various flavors of the BiCMOS SiGe process evolved over the decades producing HBT devices with ~500GHz Ft.

Thanks, that was very interesting, and really amazing!
So, something originally intended, for high speed (computer) logic, ironically ended up helping make very high performance analogue stuff!

The first we heard of this new IBM SiGe BiCMOS technology was by means of a 12 Bit 1GSPS DAC back in early 90s that I think Analog Devices had designed. IBM made well over 2B SiGe Power Amplifiers for cell phones, the SiGe replaced the GaAs PAs. We ended up first using it in the first RF System on Chip (RFSoC) back in 2000 which had an entire MW Transceiver, a few CPUs, an FPGA and a bunch of memory.

Best,

Edit: BTW almost all advanced CMOS silicon processes today are developed to support digital, analog is just an afterthought. As an analog/RF/MW IC designer in advanced CMOS, you just have to use what the digital folks give you.

[MK14]

Thanks for the interesting post, Tim! (the one I'm replying to, here)

Keep in mind, what's possible is far broader than what's available.

Yes. The Intel 4004 is on record for being the first (1971) commercially available, single chip microprocessor. but it was not only possible, to do these before then. They actually did (for a US fighter jet, it even hard built in hardware floating point, which is quite amazing for the time). But it was both secret (military) and not commercially available, so is not usually considered first, even though it was around and in use, before the Intel 4004. Also, at a much later point, someone claiming (it probably was them) to be the inventor of that military air-craft, rather advanced cpu. Actually complaining on this very forum, about their lack of recognition, for inventing an earlier microprocessor.

https://www.eevblog.com/forum/chat/first-microprocessor-mp-944-and-the-f-14-cadc/msg1600678/#msg1600678
Invented by: Ray Holt  https://en.wikipedia.org/wiki/Ray_Holt

I know, it was just an example of course.  But, that said, it can be interesting to do a design experiment like that -- follow through and see how close you can actually get, off-the-shelf

I'm very tempted, and hoping to have time/motivation, to do it. Initially, via simulation, and see what I can come up with (LTSpice). Simulation will largely eliminate the difficulties of putting together, >2GHz circuits, and save phenomenal amounts of time/cost. I've forgotten the exact price on huge tens of GHz oscilloscopes, but I can remember it had way, way too many zeros, on the right hand side of the asking price!
Similarly, I don't know the price of suitable probes (or whatever is needed) at those kinds of frequencies, but that would also probably be very large as well, I assume.

 

(For my part, I've designed a current-limiting switch using discrete transistors, with about 3 times lower power consumption than any building-block solution -- trouble is, the sheer parts count means it costs more.  It's not that such a function is impossible in IC, it would trivially do 10x better than my version; there just doesn't happen to be a commercial offering of it!)

I sometimes wonder, as surface mount components, get smaller and smaller. Especially when they are so tiny, hand assembly is considered not practicable/possible, by most people. But could that mean, as/when they get small enough, having them assembled onto a suitable PCB. Would result in something having extremely high/fast speed characteristics. I.e. would be like a large features (i.e. old technology), IC process. But without the huge cost, big delays to getting the prototypes made (like months), and other hassles/inconveniences of having prototype ICs made.
For example; One could design an circuit, which is similar to the worlds first microprocessor, the Intel 4004 . Using a 1,000 or 1,500, and more if necessary, of the smallest available SMD transistors, to create an IC like, tiny PCB. For fun, and maybe boasting privileges.
A while ago, someone made a 6502 microprocessor, out of discrete mosfets, on a PCB, which was something like 2 or so feet, in both directions (square). I.e. rather big, but it worked well (but a fair bit slower than the original, which probably goes to show how good, even ancient chips are, at reducing stray capacitance/inductance and other effects, which allows the ICs to perform at much better speeds, than typical physical PCB discrete component designs. Especially since the original 6502s 1MHz clock frequency, sounds so low/slow by today's standards.

Only very simple CPUs could run that fast (sub-100GHz say), and they'd be utterly useless for any kind of real work (an 8051 running this fast, is absolutely no competition against a modern 64-bit machine at 4GHz).

Well, I'm not so sure about that bit. You are basically, completely right. But some big/fast CPUs, have already started including, potentially very fast, tiny/small CPU cores as well as the big, fast and powerful main cores. Examples we seem to see today, are where the I/O block(s), have one or more, very fast, very tiny CPU core or cores. Which allow very short/simple programs, to allow simpler I/O functions, to occur rapidly in real-time, relatively independently of the main CPU core(s). With no main CPU core(s) interrupt latency time needed, either.
Alternatively, they can carry on running at rather low frequencies, while the main CPU core(s) are powered down (sleep mode or similar). Hence using tiny amounts of power, while waiting for a keypress, or sensor signal etc.
Recent examples are ESP32-S3 (these cpus, can be so simple, they are called FSM and/or can be switched between being a Finite State Machine or cpu (small RISCV) like functionality, which the ESP32-S3 allows.
Also the Raspberry PI PICO RP2040, with what it calls its 'PIO' state machines.

N.B. An FSM is NOT a cpu, strictly speaking. But it follows the concept of having a tiny CPU core, inside a big CPU core. Anyway, the ESP32-S3, at least one very small RISCV cpu, which counts, even though it runs at a much lower frequency, currently (for huge power saving benefits, to give it limited software capabilities, while only using relatively tiny amounts of power. E.g. logging sensor history, even while the device is supposedly switched off, such as room temperature trends).

In fairness to yourself, they are not necessarily higher clock frequency than the main CPU. But it might happen in the future. E.g. The Raspberry PI PICO's, PIO feature is so fast and flexible, it can be programmed to be a fast video chip. and continually output images equivalent of the old classic BBC Microcomputer, while leaving the main CPUs, to do other things, as regards processing loads.

[mawyatt]

The speed limitation with todays small featured CMOS processors is not the device speed but thermal limitations, it' been that way for some time now.

Best,

[MK14]

The first we heard of this new IBM SiGe BiCMOS technology was by means of a 12 Bit 1GSPS DAC back in early 90s that I think Analog Devices had designed. IBM made well over 2B SiGe Power Amplifiers for cell phones, the SiGe replaced the GaAs PAs. We ended up first using it in the first RF System on Chip (RFSoC) back in 2000 which had an entire MW Transceiver, a few CPUs, an FPGA and a bunch of memory.

How the heck they can get the 12th (least significant bit), to settle at the right voltage, in such a relatively tiny amount of time (a nanosecond), despite a degree of capacitance. Can hurt my head. You can have 12 resistors, with each one double the next ones value, but the 12th one becomes such a large value, it intuitively will take ages, to settle the capacitance involved voltage, to the correct 12th bit. Off-hand, I'd need to look up R .. 2R ladders, and things like that, to better understand it.
Presumably/maybe they are really just super fast analogue to digital converters, as those can be configured in some circuits (I think it was the art of electronics books, where I found it), to act as a DAC, via the very fast A2D converters. I vaguely remember op-amps, in the middle of such circuits. So I suppose, that makes sense, as the op-amp would force the capacitance involved to become the correct voltage (loop closed by A2D converters), rather than waiting for huge setting times, if it was resistor ladder based.

Edit: BTW almost all advanced CMOS silicon processes today are developed to support digital, analog is just an afterthought. As an analog/RF/MW IC designer in advanced CMOS, you just have to use what the digital folks give you.

Yes, I heard a lot about that. Apparently, Tektronix (oscilloscopes), quite a long time ago. Had huge difficulties, getting hold of companies, who could/would make the necessary analogue ICs, for Tektronix's, continually improving oscilloscopes. So, eventually they decided to build their own integrated circuit, plant. Just to make their very own analogue integrated circuits. As they didn't have to play second Fiddle to digital IC production, and could optimize it just for analogue, to their liking.
I think the wafers, were around 2 to 2.5 inches, at that time.
There were also some Tektronix custom digital ICs, for things like drawing characters on oscilloscope screens. I'm not sure if the custom digital chips, were made in that facility, or farmed out to another company.
To my huge surprise, somewhat early IC production, didn't use (expensive) clean rooms. Clean rooms weren't used until smaller feature sizes, came into play. So, some early IC manufacturing, was not that difficult, or uncommon, a very long time ago.
I wonder if, sooner of later, perhaps in China. Amazingly affordable custom IC manufacturing will become available. FPGAs (in my view), already give what is essentially a full high speed, custom digital IC, to all intenses and purposes, even if it is more expensive, slower, and technically speaking NOT a real custom IC part.

[mawyatt]

Recall the SiGe DAC was a classic R-2R ladder and a switchable current source. If you search the IEEE SSJ, you can find many references on DACs. On particular SiGe DAC was developed by Keysight that employed a classic binary weighted 8 bit DAC with a 6 bit DAC for the upper bits in a 14 bit segmented architecture. This utilized a Dynamic Element Matching technique for the upper bits to achieve -90dBc SFDR at 12GPSPS, they published the predecessor to the DAC in the SSJ around ~2007 I recall. At the time we were briefed (actually saw this at the MTT in 2009) they were well on their way to another DAC that had targets of better than -100dBc SFDR at >20GSPS.

Here's a picture from way back of the 12GSPS DAC called Griffin (actually much faster than that).

Best,

[MK14]

The speed limitation with todays small featured CMOS processors is not the device speed but thermal limitations, it' been that way for some time now.

Best,

Assuming you were referring to my post(s). The upper frequencies, are no higher, than the fastest frequencies already in the cpus I was talking about. But the relatively large number of logic delays needed, in order for the big main processor, to calculate powerful things, such as floating point. Are not needed in a tiny CPU or FSM, with a tiny and very simple instruction set, it needs (potentially) few logic delays, per clock cycle. So can run at far higher frequencies, while still keep the IC transistors at the same frequency, as the main processor(s).
I.e. If each transistor stage (logic gate), takes 100ps, and there are 20 logic gates worth of delay, for the main cpu (per pipeline stage). It can run at 20 x 100ps = 500MHz. Ignoring safety margins and other real life practicalities.
But if a tiny CPU or FSM, only needs 5 logic gates worth of delays, with the same 100ps per transistor/gate stage. Then it can run at 100 x 5 = 500ps = 2GHz.
If I remember correctly, there is some internal trickery, which allows the 2GHz section, to effectively run at 2GHz, despite actually running at a lower physical frequency (1GHz ?). Which is some technique (I've forgotten), like double edge clock triggering, or very advanced, clock phase distribution. Whereby the clock is kept at 500MHz, distributed to where it needs to be, then double edge clocking and/or local PLL and/or phased clock techniques (whatever its called), allow it to effectively perform at 2GHz worth of operations, while not consuming the huge amount of power it might/would, if it reached the true GHz needed for a simple, one phase clock.

[MK14]

Funny you mention that!! I argued that very Solid State Physics fact in 80s about hole/electron mobility against a GaAs CMOS that Honeywell was developing. They wanted to have a process that would produce a Rad-Hard Space Computer that was also very fast, and even got a bunch of USG funding!! I lost this argument on a number of occasions, and basically was told to "go away, you don't know anything about GaAs!!"....
...
....
so after dumping ~$1B into the GaAs CMOS fab process & designs it was abandoned, and nobody remembered those earlier discussions about hole mobility in GaAs

That's sad. A pity they didn't listen to you.

I've encountered that sort of situation myself. What can be even more annoying, is that nobody remembers the original discussions, except the person, who pointed out the serious problems, long before things went horribly wrong.

In all fairness to everyone, it is difficult to accurately predict the future. Sometimes things seem virtually impossible to get right. Nethertheless, it does eventually get sorted out and invented.
For a long time, most people thought computers would never beat humans at Chess, or at least not for another 50 or 100 years. But the huge pace of ever faster CPUs, ended that argument.
Computer Speech recognition, was also considered a hurdle, that would take a very long time to master. Yet we now have Amazons Alexa, and other speech recognition devices.

I suppose the current argument, would be when/if we will have self-driving cars, in a big way, in most road systems round the world. Will it be now, 5, 10, 25, 50 or even 100 years into the future ?

[mawyatt]

author=MK14

Yes. The Intel 4004 is on record for being the first (1971) commercially available, single chip microprocessor. but it was not only possible, to do these before then. They actually did (for a US fighter jet, it even hard built in hardware floating point, which is quite amazing for the time). But it was both secret (military) and not commercially available, so is not usually considered first, even though it was around and in use, before the Intel 4004. Also, at a much later point, someone claiming (it probably was them) to be the inventor of that military air-craft, rather advanced cpu. Actually complaining on this very forum, about their lack of recognition, for inventing an earlier microprocessor.

https://www.eevblog.com/forum/chat/first-microprocessor-mp-944-and-the-f-14-cadc/msg1600678/#msg1600678
Invented by: Ray Holt  https://en.wikipedia.org/wiki/Ray_Holt

Very interesting, didn't know that history. Thanks!!

Best,

Edit: Back in those days the the military and various USG agencies led the advanced silicon developments, but after the advent of the PC the tide began to shift towards commercial and mostly CMOS. After the arrival of cell phone, the acceleration rate in commercial CMOS completely outpaced any government investment and quickly became the flagship technology we see today.

[MK14]

Edit: Back in those days the the military and various USG agencies led the advanced silicon developments, but after the advent of the PC the tide began to shift towards commercial and mostly CMOS. After the arrival of cell phone, the acceleration rate in commercial CMOS completely outpaced any government investment and quickly became the flagship technology we see today.

The big, almost certainly confirmed rumor (expanding on what you just said), is that around the late 1950s, and early 1960s, they (US), were extremely worried about the Cold War, especially if it went nuclear. The electronics of the time, would have meant (for precise navigation and location calculations and things) strapping a massive, multi-cabinet Mainframe computer, on top of a ballistic missile. Maybe not impossible as a payload, but highly impracticable, and would be difficult to power and there would be many other problems, like surviving the G-forces, at times, during the flight.

So, the US Government, poured a ridiculously (normally impracticable) amount of money, into getting, what ended up being the TTL 7400 Integrated Circuits. Which could be turned into, (relatively) much more compact units (computers), for fitting to ballistic missiles. Hence the TTL 7400, helped launch the modern computer technology.
There were also, partly competing (or earlier), logic IC types, such as Resistor Transistor Logic (RTL), and Diode Transistor Logic (DTL). But I recently read, that  Fairchild, had developed, the best Logic Gates of the time (effectively the first good TTL or similar technology).

So, Texas Instruments (TI), realized  how good they were, and tried VERY hard to buy a licence to make them, from Fairchild. But they refused to make a deal (they later acknowledged it was a big mistake on their part), so TI, made their own versions of the chips, anyway (I'm not sure of the right terminology for that time, but it was something like 'clone', 'copy but changed enough to NOT go to court', or similar versions). Hence the (US/TI) TTL 7400 logic devices were born, and the military spec'd/cased versions, were used in the applicable military projects.

The 74181 (ALU), could with suitable surrounding TTL chips, be made into a reasonably compact and powerful, computer. Suitable for going somewhere inside a ballistic missile, and with not too unreasonable power requirements. Perhaps compact suitcase sized, depending on its specifications and the 74 packages used.

Off-hand, I'm not sure of the various internet source(s), I read, somewhat recently about it. But I just had a quick look, and found this:
https://computerhistory.org/blog/the-rise-of-ttl-how-fairchild-won-a-battle-but-lost-the-war/

EDIT: I mentioned UK (Fairchild), but the source doesn't (it says US), so I removed references to the UK from my post.