Analog circuits are fast: But how fast?

[technix]

I have this feeling that the Moore law disproportionately applies to the digital domain, driving the cost of digital circuitry down way faster than analog circuitry. For now even when digitally-generated analog signals are used (for example those LVDS and SerDes systems) it is way too often building redundancy into the digital domain, allow analog problems to happen, and happily mask the problems.

Digital circuitry does have the benefit of wide-open gaps between VIL and VIH allowing the circuits to work with little to no trimming driving yield up.

[MK14]

Sadly, I suspect that these (analogue) design skills are being lost. So if we suddenly need them again (and to an extent we do, as they are still needed), it is not so good to lose them.

I think they are. A while back when I found myself looking for work and toying with going more toward hardware rather than the software work I'd been doing I interviewed at a place that makes automated testing systems. They said they normally hire fresh EE grads but had been disappointed in the lack of analog skills in people coming out of the universities. As a college dropout almost entirely self taught hobbyist engineer I wasn't expecting much and I actually preemptively turned down the position because the commute would have absolutely killed me and I knew going in that it would have meant a pay cut but they told me later I was easily the most qualified person they talked to.

Ultimately I decided to stay in software which tends to pay a lot better, and keep EE as a hobby. It's something I enjoy doing enough that I don't want to turn it into work. Anyway the point of all this is that if my analog abilities which while passable are not expert level are that much better than recent EE grads then analog is in trouble.

My opinion, understanding and beliefs/experiences are that QUANTITY wise, we may well have enough analogue engineers.

But the thing is, quality/ability/experience levels do vary. So, although we hopefully have the quantity, we don't necessarily have many/enough of the really high calibre analogue design engineers.

This might matter. Perhaps because we need them, to hopefully compete well in the west, against the upcoming likes of China (and other countries).
There could even be wars again, one day. Electronics will then be an extremely important part of such wars. Even during World War 2, electronics was extremely vital (Enigma decoding computers in the UK, Anti-aircraft proximity fuzes, anti-ship fuzes, artillery proximity fuzes, RADAR, Radios/Transmitters etc etc).

Even our respective countries economic success or not, could be affected, by the electronics industries and stuff.
So I think it is very important.
But also, Digital Electronics (few know how to do complex FPGA (VHDL, Verilog) designs), software engineers, etc etc.

[james_s]

When I was a kid you could spin the channel knob on the TV as fast as you wanted and still see, ever so briefly, the image/video of each channel for that fraction of a second it was selected. I miss that. 

Yes, channel surfing is a painful experience these days, although between the increased ad load, network logos and vastly reduced quality of the content, TV in general is a painful experience these days so perhaps it doesn't matter.

I'm reminded of when we were kids, my best friend in elementary school figured out he could turn the knob on the cable box in between 18 and 19 and wedge it there with a matchbook and the Playboy channel would come in. His parents always seemed to get home just a few minutes after we got it working though.

[rstofer]

Sadly, I suspect that these (analogue) design skills are being lost. So if we suddenly need them again (and to an extent we do, as they are still needed), it is not so good to lose them.

<snip>
Ultimately I decided to stay in software which tends to pay a lot better, and keep EE as a hobby. It's something I enjoy doing enough that I don't want to turn it into work. Anyway the point of all this is that if my analog abilities which while passable are not expert level are that much better than recent EE grads then analog is in trouble.

Everybody works in their own self-interest.  At the moment, it is a given that software pays more than hardware and that digital hardware pays more than analog hardware.  There are also more jobs available in software than in digital hardware than in analog hardware.  And the math is harder for analog design...

No student that has even looked at www.bls.gov would bother studying analog design and, truth be known, even digital hardware design is questionable.  Job openings and pay scale trumps everything.

So the analog companies have to expect to do a little on-the-job training.  All of the other companies do the same thing.  I doubt that any new-grad grasps the complexity of the latest Intel designs regardless of how well they did in hardware and computer design courses.

[coppercone2]

does an analog designer at least physically move more then a digital one (big perk of the job)?

[joeqsmith]

When I was a kid you could spin the channel knob on the TV as fast as you wanted and still see, ever so briefly, the image/video of each channel for that fraction of a second it was selected. I miss that. 

Yes, channel surfing is a painful experience these days, although between the increased ad load, network logos and vastly reduced quality of the content, TV in general is a painful experience these days so perhaps it doesn't matter.

Ads pushing the latest drugs and ads for attorneys going after the old drugs...  I gave up on TV years ago.  I don't even watch the news (if you want to call it that) anymore.   

With the push for digital I wonder as a public service system, how well it would even work now.  The analog systems were very robust.  The digital systems seem the complete opposite. 

[Zero999]

With through hole transistors that would probably be 100's of square feet. An iPhone out of vacuum tubes would be bigger then LA and consume more power too.

Giant analogue circuits, don't really scale up too well. Because the drift and other effects, could mean that it never works, as engineers are forever re-tuning it and fixing it. (Especially old analogue designs, modern stuff can be pretty decent, and solves many of those problems).
Even old digital (valve/tube) based computers were like that as well.

I think I read that the really old valve/tube, massive computers, could only run for an hour or so (maybe a few hours or so), before another vacuum tube would burn out and need replacing.
I think they spent half their time, being repaired and the other half doing stuff (very approximately).

N.B. I'm talking about digital computers from the 1940s and 1950s.
Later ones became more and more reliable.

Yes, no modern digital IC will perform well if built from discrete components, because the physical distances between the components would make the propagation delays too long for it to work at a reasonable speed.

[tggzzz]

If you want to get down to brass tacks, even digital circuitry operates in an analogue world.

Except in the photon counting and femtoamp domains

But otherwise you are right; "digital" is merely an interpretation placed on an analogue value.

[joeqsmith]

Except in the photon counting and femtoamp domains

But otherwise you are right; "digital" is merely an interpretation placed on an analogue value.

I would assume it was still all analog.   

[tggzzz]

Except in the photon counting and femtoamp domains

But otherwise you are right; "digital" is merely an interpretation placed on an analogue value.

I would assume it was still all analog.   

1fA is 6424 electrons/second. At that rate the electrons can be individually counted therefore they "are digital".

[coppercone2]

their speed sure aint digital, unless you got some idea about plank limits

[james_s]

No student that has even looked at www.bls.gov would bother studying analog design and, truth be known, even digital hardware design is questionable.  Job openings and pay scale trumps everything.

That's not always true. My other half manages a veterinary clinic. She could make roughly twice as much doing the same job managing a human medical clinic but chose the job she has because she loves working with animals. A good friend's wife has a masters in chemical engineering and was making a good living as a software developer, she recently quit to become a teacher because doing something she wanted to do trumped the much larger paycheck. Being the primary earner in my household though it was a fairly easy choice to go with the higher paying job, I can do the hardware as a hobby and probably enjoy it more than I would if I were doing it for a living anyway.

[BrianHG]

When I was a kid you could spin the channel knob on the TV as fast as you wanted and still see, ever so briefly, the image/video of each channel for that fraction of a second it was selected. I miss that. 

This was my favorite CATV converter, even kept it after the numeric display wireless remote ones became available.
You can slide through all 42 channels, TWICE!, in 1 second!
Jump from any 1 channel to any other in under 0.25 seconds without numeric entry on a remote...
Excellent for jumping channels and avoiding commercials.

[radioactive]

Shortest distance between two points is a straight line, so I think analog might win this one with quantum tunneling... 

Some physicists have claimed that it is possible for spin-zero particles to travel faster than the speed of light when tunnelling.

[BrianHG]

Screw digital, this analog vacuum tube runs circles around digital ICs...

https://www.extremetech.com/extreme/185027-the-vacuum-tube-strikes-back-nasas-tiny-460ghz-vacuum-transistor-that-could-one-day-replace-silicon-fets

Build some integrated structures in it to create something like an op-amp or mixer, and you can compare/add/sub in the linear analog domain much faster than digital silicon.

[Brumby]

If you want to get down to brass tacks, even digital circuitry operates in an analogue world.

Except in the photon counting and femtoamp domains

I was waiting for someone to go there.   

Congratulations!

[tggzzz]

If you want to get down to brass tacks, even digital circuitry operates in an analogue world.

Except in the photon counting and femtoamp domains

I was waiting for someone to go there.   

Congratulations!

Yeah well, it had to be me, didn't it

Sometimes I like to enliven fairly ill-defined debates with points that stretch people's imagination and understanding

[petert]

If you want to get down to brass tacks, even digital circuitry operates in an analogue world.

Except in the photon counting and femtoamp domains

But otherwise you are right; "digital" is merely an interpretation placed on an analogue value.

Interesting. Could you elaborate on that, considering I am not an expert? Would that affect aliasing/filters?
Would be great if these effects could be simulated on a discrete element basis (discrete time/charge carriers).

[Zero999]

If you want to get down to brass tacks, even digital circuitry operates in an analogue world.

Except in the photon counting and femtoamp domains

But otherwise you are right; "digital" is merely an interpretation placed on an analogue value.

Interesting. Could you elaborate on that, considering I am not an expert? Would that affect aliasing/filters?
Would be great if these effects could be simulated on a discrete element basis (discrete time/charge carriers).

Because electrons are discrete particles, one could argue that their presence or absence is binary, i.e. digital, but it's still analogue, because the number of electrons over a certain time period isn't just a 1 or 0 value. Of course one could then argue that at very low currents or short time periods, the answer will then be 0 or 1 but at that point it's just noise. Longer time periods are required to get a meaningful current reading.

Compare this to counting the number vehicles per hour in a single lane of traffic. When there's not much traffic or over short periods of time, the answer will always be either zero or one, tending towards zero for shorter time periods and quite periods. In the rush hours, it may be acceptable to count the number of vehicles over a 10 minute period, multiply it by six and get a reasonably accurate estimate at the number of vehicles per hour, but counting over a shorter periods or when it's less busy, the result will become progressively less accurate.

This above is an example of shot noise and which is why a long exposure time is required to get a good image, at lower light intensities and it's impossible to measure tiny currents, at high frequencies, with any decent accuracy.

[tggzzz]

If you want to get down to brass tacks, even digital circuitry operates in an analogue world.

Except in the photon counting and femtoamp domains

But otherwise you are right; "digital" is merely an interpretation placed on an analogue value.

Interesting. Could you elaborate on that, considering I am not an expert? Would that affect aliasing/filters?
Would be great if these effects could be simulated on a discrete element basis (discrete time/charge carriers).

Realise that all digital information is transmitted by analogue voltages (or currents, but I'll ignore that for simplicity). The voltages depend on the logic family, with 0-5V, 0-3.3V, 0-1.8V being typical limits.

Examples...

A TTL (not CMOS) input will consider any input voltage above 2.0V or below 0.8V to be valid logic levels. If you look at a TTL high output on a scope, you can see it vary between (say) 2.6 and 3.6V depending on the state of all the devices inputs - but that doesn't matter provided the receiver "sees" a voltage above 2.0V. The receiver regards/interprets/thinks anything above 2.0V as a logic high.

Many digital systems are not binary, but have multiple levels and phases. That's how you can transmit, say, 56kb/s down a telephone line with an 8kHz bandwidth; the baud rate is <<56kHz!

Have a look at transmission line theory for modern fast risetime digital logic families. The voltages seen "on the wire" can have several different values as the signal moves from transmitter to receiver, and those are very much dependent on the source and receiver termination strategies used.

[rstofer]

No student that has even looked at www.bls.gov would bother studying analog design and, truth be known, even digital hardware design is questionable.  Job openings and pay scale trumps everything.

That's not always true. My other half manages a veterinary clinic. She could make roughly twice as much doing the same job managing a human medical clinic but chose the job she has because she loves working with animals. A good friend's wife has a masters in chemical engineering and was making a good living as a software developer, she recently quit to become a teacher because doing something she wanted to do trumped the much larger paycheck. Being the primary earner in my household though it was a fairly easy choice to go with the higher paying job, I can do the hardware as a hobby and probably enjoy it more than I would if I were doing it for a living anyway.

As long as the total household income is sufficient, it is possible to walk away from the highest paying opportunity.  But there is no point in working your way into poverty.  There are easier ways to get there.

I always kept electronics as a hobby and never did work in the field.  I made a decent living working essentially in project management.

[petert]

Have a look at transmission line theory for modern fast risetime digital logic families. The voltages seen "on the wire" can have several different values as the signal moves from transmitter to receiver, and those are very much dependent on the source and receiver termination strategies used.

Thanks, informative as well. Though, I was thinking of an answer more along the lines of Hero999.
I wonder how the transition between discrete/digitial and continous/analog happens.

Because electrons are discrete particles, one could argue that their presence or absence is binary, i.e. digital, but it's still analogue, because the number of electrons over a certain time period isn't just a 1 or 0 value. Of course one could then argue that at very low currents or short time periods, the answer will then be 0 or 1 but at that point it's just noise. Longer time periods are required to get a meaningful current reading.

Why would you get noisy results, isn't it possible to count individual photons, for example?
Assuming a noisy reading, I see we are talking about average currents/voltages, so measuring over a longer/reasonable time period to average out fluctuations.

But if in theory your sampling frequency would be high enough, you'd get discrete electron counts, right?

Assuming that is possible, is then analog really nothing more than averaging over many discrete measurements (either deliberately by later mathematical averaging, or implicitly by measuring over longer time periods)?

[Zero999]

Have a look at transmission line theory for modern fast risetime digital logic families. The voltages seen "on the wire" can have several different values as the signal moves from transmitter to receiver, and those are very much dependent on the source and receiver termination strategies used.

Thanks, informative as well. Though, I was thinking of an answer more along the lines of Hero999.
I wonder how the transition between discrete/digitial and continous/analog happens.

Because electrons are discrete particles, one could argue that their presence or absence is binary, i.e. digital, but it's still analogue, because the number of electrons over a certain time period isn't just a 1 or 0 value. Of course one could then argue that at very low currents or short time periods, the answer will then be 0 or 1 but at that point it's just noise. Longer time periods are required to get a meaningful current reading.

Why would you get noisy results, isn't it possible to count individual photons, for example?
Assuming a noisy reading, I see we are talking about average currents/voltages, so measuring over a longer/reasonable time period to average out fluctuations.

But if in theory your sampling frequency would be high enough, you'd get discrete electron counts, right?

Assuming that is possible, is then analog really nothing more than averaging over many discrete measurements (either deliberately by later mathematical averaging, or implicitly by measuring over longer time periods)?

Yes, at high sample rates you'd get electron counts, the same for photographs at high shutter speeds getting discrete photon counts.

It's noisy because of the discrete nature of the electron/photon. 1fA is 6242 electrons per second, so suppose we measure exactly 1fA flowing in a piece of wire over a period 1ms, there's no way we can get any level of accuracy beyond 1 in 6.242. If we kept repeating the measurement, the reading would fluctuate considerably, seemingly at random. Only when thousands of samples are taken would be know for any certainty we have exactly 1fA flowing, rather than  0.9fA or 1.1fA. Now that's fine at DC, but if the signal is AC and non-repeating, there's no way to separate the information from the random fluctuation in the number of electrons passing per ms.

This is why photographs taken with too shorter shutter speed, when there isn't enough light appear grainy and our vision in low light is blurry. A long exposure time is required, at low light level, with a very large lens, to get a high resolution image.

[BrianHG]

Problem: as transistor size decreases, system voltage decreasing, wont we reach the near bottom voltage switch point where the so-called digital switches are acting at their edge of on-off state, sort of like being in the analog domain.  Or will quantum tunneling make such digital IC too noisy to function at all?

[tggzzz]

Have a look at transmission line theory for modern fast risetime digital logic families. The voltages seen "on the wire" can have several different values as the signal moves from transmitter to receiver, and those are very much dependent on the source and receiver termination strategies used.

Thanks, informative as well. Though, I was thinking of an answer more along the lines of Hero999.
I wonder how the transition between discrete/digitial and continous/analog happens.

There is no transition The discrete/digital domain is merely an interpretation of the analogue domain.

However, it is possible to define analogue circuits that mainly stay within separate bounds and not in the "gap" between the bounds. But they sometimes fail to do that; a classic example is a flip-flop in a metastable state.

Consider a couple of bipolar transistors connected as a "long-tailed pair". Depending on the voltages applied to the transistor bases, the LTP can act as a linear differential amplifier, a mixer, an ECL logic gate, a multiplier, a current switch, etc...