What is more difficult, analog or digital design?

[rbola35618]

I just wanted a point of view from anybody that has done both digital and analog design. I am a power supply designer and generally have never had any interest in digital design; I find that with analog design, I can be more creative in how I design a circuit and I get a great satisfaction in creating a circuit. What discipline requires more knowledge and more skills?

Also, at work, it seems to me that digital designers are viewed with more respect than us power/analog guys. Does anybody have the same situation. To me it seems that most non electrical people view power supply design as trivial and easy; just a collection of transformers, diodes and some filtering. When in contrary, power supplies and analog design are in reality non trivial.

On the other hand, is digital design a hot skill set? Is it considered to be a more favorable skills sets because it is what most universities teach?

I would like to know what is more challenging. I do not want to start a war or argument between digital or analog people. Like I mentioned before, I am ignorant when it comes to digital stuffs and would like to be informed what skills set are required.
 

[Rerouter]

both are valid skill sets, that over time have formed there own distinct roles, with people even at the design level treating them as 2 different things and isolating one from the other as much as possible,

the thing to remeber is digital is just analog set up in a particular way, building upon an abstraction layer if you like, e.g. an opto-isolater is to most begineers veiwed as digital, but it can easily function as analog if you delve deeper,

digital gets more difficult to me personally, because you have to start thinking about timing, delays and propigation delays, with analog everything is always in flux, it doesnt exactly matter if your transistor biases 12ns later than the signal in a power supply, even getting into switchmode territory, but in digital when something has fixed time slots, it can make or break a design,

in short, both have there areas that outweigh the other, most veiw digital with slightly more respect only because they are used to thinking of analog as obsolete, and think of really complex digital system for digital,

[rbola35618]

Hi Reroute,

Taking the timing issue that you mentioned, how do the digital people fix that? Do they fix thru writing the code or do they fixed the hardware.

I design switching power supplies. When I find a problem that I was not anticipating, my fixes always require hardware changes. Some of our power supplies are High Voltage steppers stepping from 0 V to 10000 V and are interfaced to instruments such microchannel plates. So when we have a problem, fixing them takes a significant amount of time to debug the system.

My perception have been that digital is mostly code writing. Is this perception wrong?  As I mentioned, I don't have a feel how the digital work and would like to have an understanding how they fix their problems

[NiHaoMike]

I started with analog and only really started working with digital a few years ago, so I would say that digital is trickier. Tiffany Yep, who works with Verilog and C code every day but very rarely solders anything, would definitely say that analog is trickier. (Of course, everything in the real world is technically analog...)

[Rerouter]

modern digital is dipping its toes furthur and furthur into software with microcontrollers, but raw logic, timing of signals even RF still are the core of the digital domain, as signals get faster, the latter 2 begin to matter more and more, while for slower signals, the first is the specialy,

for example, you have a device that reads the state of 8 pins say a parrellel device sending a byte, if you couldn't tightly route all your traces together the delay of the longer routing of 1 or more traces might be such that the signal hasnt arrived by the time the current time slot has ended, e.g. the device will have whichever value was on the line during the time slot instead of what you wanted,

this may be correctable by delaying the other signals enough to match up, but usually doesnt quite work when you get into higher speed digital systems requiring hardware revisions,

[Rerouter]

on the opposite side of the fence, with analog, if your stepper motor driver had to operate fast enough and you had say a 1m direct run for 1 pole, and for unknown reasons the rest had to run around 20m, the delay could be enough to upset the capability of your stepper, as the one mismatched pole would overlap where it shouldnt with the others, killing off a bit of torque

though this is very unrealistic, its just a concept to try and better convey that analog and digital arent that different, only the terms and general signal speeds differ,

[EEVblog]

My perception have been that digital is mostly code writing. Is this perception wrong?

Not really.
More and more "digital design" would be FPGA and/or microcontroller based, and SOC based design in the ASIC world, as opposed to the fixed discrete hardware of several decades back that required re-spinning of boards to makes changes. In the high end digital world, it's mostly simulation and/or verification work.
But remember, high speed digital design IS analog 

Dave.

[gregariz]

My teachers used to always say that Analog was quite a bit harder than digital and I must say that I believe that. Particularly when you go up in frequency, into the microwave range things can become impossible. The problem is really that in the Analog domain there are a 1000 versions of working. Getting it to work is not enough. Whereas my experience in the digital world, and I do a bit of chip design every now and then its almost exclusively VHDL these days, and the compilers and simulators are so good you really never have to touch hardware. Then when you do have a chip design ready after crunching your test vectors you have to hand it off to fab house, and you magically get a working device back. This is quite different to the microwave world where you may be shaving fractions of a mm off a filter structure in order to try to get one of a number of competing parameters to work. There are simulators that are very good, but they really only get you in the ballpark, not right on the money. There always seems to be an incredible amount of experimental work to dial parameters in.

Case in point would be that Marconi Sig Gen that Dave pulled apart the other day. It may look very modular and what not but that was not done for convenience. Everything in that box is shielded because everything is coupling to everything else. That sig gen is specced to a minimum power output of -137dBm. Try getting a noise floor down that low. That type of stuff can easily take a dozen spins of the board to get it working to spec.

Digital can become very difficult when you are devoloping chips for crunching algorithms, but usually this has nothing to do with the hardware design per se, rather it becomes a math problem ie DSP. Just my 2c

[mikeselectricstuff]

There is no such thing as digital. Everything is analogue. 

[TerminalJack505]

Here's what Bob Widlar has to say about analog vs. digital...

[alm]

There is no such thing as digital. Everything is analogue. 

There's no such thing as EE either, it's all physics.

[T4P]

There is no such thing as digital. Everything is analogue. 

There's no such thing as EE either, it's all physics.

Physics is the world of non-ideal . EE is the real world .

[jahonen]

There is no such thing as digital. Everything is analogue. 

There's no such thing as EE either, it's all physics.

I'd say that digital electronics implementation is just strongly non-linear high-bandwidth analog stuff.

But then, smallest unit of electricity is the charge of an electron, so that means that analog signals have some kind of discrete levels too. For example, current of 1 fA is equivalent to about 6242 electron charges/s. But steps are of course, very small for any practical purpose.

Regards,
Janne

[rbola35618]

Wow!, that is a wonderfull picture of Widlar. Thanks for posting it!

[ejeffrey]

Obviously the hardest thing is mixed signal   You have all the standard problems of analog and digital, plus crosstalk.  Plus you have a beautiful analog signal you want to keep perfectly linear while being subject to the absurdly non-linear process of quantization.   Clock jitter that would be inconsequential for any other circuit can cause nasty spectral artifacts.

[saturation]

They both require different set of skills, so I think one needs both at different degrees of intensity.   

Since analog theory uses more math than digital work, analog has a edge is being more demanding of your skills at first, but the working equations for most tasks are far less complicated than what's taught in schools. 

Digital logic is easier to understand at first, but 2 state combinations happening many times at high speed gets complicated it its own way.   Then there is the software aspect one needs to have control off too, i.e., firmware. 

While quantum states of matter may be discrete, its doesn't follow binary logic, so interaction with the natural world still requires interfacing via analog tranducers.  The entire EM spectra [ from audio to gamma rays], or power [ AC or DC], temperature, humidity, etc., require control of physical properties which are all analog.

http://en.wikipedia.org/wiki/Analog_electronics#Analogue_vs._digital_electronics

The trick is getting the analog data into the digital domain and out again, if needed, with minimum loss of fidelity to allow for the richness of digital processing.   

To address the OPs original question, without analog skills to harness mains power to create the PSU, there is no power to apply digital controls to, and worse case we can still use analog controls.  So, between both, the analog issue comes first. 

[westfw]

Analog is "harder", but digital circuits tend to be much more complex (that's the advantage of being digital, isn't it?  You frequently constrain the result of an analog circuit to be either 0 or 1, and you forgive a lot of error.)
OTOH, even a 70's era CPU would be composed of 3000+ gates, while a 3000-transistor analog circuit is essentially unheard of.  (because, you know, it would be REALLY difficult!)

[T4P]

Analog is "harder", but digital circuits tend to be much more complex (that's the advantage of being digital, isn't it?  You frequently constrain the result of an analog circuit to be either 0 or 1, and you forgive a lot of error.)
OTOH, even a 70's era CPU would be composed of 3000+ gates, while a 3000-transistor analog circuit is essentially unheard of.  (because, you know, it would be REALLY difficult!)

And yet today a 700million gate CPU is rather weak .
Enter the AMD 7970 GPU , it has 4.31 billion gates .
BUT GPU's are a little different

[Blofeld]

From the EE113 Course Notes, Stanford, page 4
http://www.stanford.edu/class/ee122/Handouts/EE113_Course_Notes_Rev0.pdf

Analog Boot Camp Drill Routine
by G. Kovacs

(The words are first barked out by the professor,
then shouted back by students marching in
formation.)

Analog circuits sure are fine,
Just can’t get ‘em off my mind.

Digital circuits ain’t my kind,
Zeros and ones for simple minds.

I guess NAND gates aren’t all that bad,’
’Cause I need them for circuit CAD.

One, two, three, four,
Gain and bandwidth, we want more.

Five, six, seven, eight,
We don’t want to oscillate.

Widlar, Wilson, Brokaw too,
They’ve got circuits, how ‘bout you?

(repeat)

[siliconmix]

digital chips with an analog role and vise versa

[hans]

Analog is "harder", but digital circuits tend to be much more complex (that's the advantage of being digital, isn't it?  You frequently constrain the result of an analog circuit to be either 0 or 1, and you forgive a lot of error.)
OTOH, even a 70's era CPU would be composed of 3000+ gates, while a 3000-transistor analog circuit is essentially unheard of.  (because, you know, it would be REALLY difficult!)

In my view digital circuits are viewed like 'easy expandable', where as this means they grow.. grow.. grow.. and in the end you're having 10 IC's on 1 SPI, I2C or one-wire bus and getting timing or other weird behaviour. Where is the problem? Dive in that pile of software, logic analyser/scope and figure out..
Of course, using the right software tools you can also isolate parts of circuits and eventually find out what's going on.

With an analog setup certain parts are often separate so they don't interfere or unbalance each other. Problems can get more isolated with that. However.. analog always seems very permanent. If you designed an amplifier or power supply supposed to offer a noise or ripple of a certain level, and it doesn't meet it well.. likelihood some wires or fixes are in place (unless you designed some test pads before hand).
With software you can patch some things much more 'under the hood' in my idea, and is also much more flexible in changing stuff. As long as the IC's can talk efficiently with each other.

I'd say analog design is harder. You can't predict what exactly it is going to do, and if you've designed a PCB it never gets prettier if an amplifier or PSU needs to be modified. The worst mistakes I've made on digital circuits is mixing up a pin-out or port of a chip.. like some PICs had input-only pins(multifunction shared with oscillator, unable to drive outputs).. but they are reasonably/easy quick to fix.

[saturation]

Given the costs of these development platforms, I wonder how long it will take for small labs to afford to design on these settings?

I do dislike them as they stand today.  Its usually one or a very small set of vendors,  making you quite dependent on them. 

Repairing devices built on such platforms will be futile if the software no longer runs on current PCs or OSs, just as repairing older 1990s firmware is relegated to finding binary copies of working firmware, replacing the older memory chip and hope for the best.   Having to re-learn an old platform, that could be majorly obtuse, just to repair a limited set of devices may not be even worth it.

One thing PC based development takes away are your visualizing the design, and adding your touch and smell.  Not that it happens often, but you can feel parts overheat, smell them if they burn and visualize the schematic by looking at how the components are connected.  On a black box, everything is based on the PC screen, so you are dependent on your eyes, memory and worse, the software's performance putting the designer far more at the mercy of a machine.

My perception have been that digital is mostly code writing. Is this perception wrong?

Not really.
More and more "digital design" would be FPGA and/or microcontroller based, and SOC based design in the ASIC world, as opposed to the fixed discrete hardware of several decades back that required re-spinning of boards to makes changes. In the high end digital world, it's mostly simulation and/or verification work.
But remember, high speed digital design IS analog 

Dave.

[Neilm]

I have yet to see a digital circuit that radically changes performance (i.e. totally stops working) when you put a scope probe on it.

Neil

[alm]

Wasn't there a myth about a computer being shipped with a scope probe clipped to the circuit because they were unable to figure out how to make it working without the probe attached? Unlikely to be true, but a funny story nonetheless.

Does a clock crystal qualify as a digital circuit? It's not too hard to find a crystal circuit that stops oscillating due to the extra 10-20 pF of a scope probe. I imagine fast digital circuits with tight noise and power budgets may also have trouble with the capacitance of a standard passive hi-Z probe, but we don't notice since nobody uses slow hi-Z passive probes with them.

[T4P]

Wasn't there a myth about a computer being shipped with a scope probe clipped to the circuit because they were unable to figure out how to make it working without the probe attached? Unlikely to be true, but a funny story nonetheless.

Does a clock crystal qualify as a digital circuit? It's not too hard to find a crystal circuit that stops oscillating due to the extra 10-20 pF of a scope probe. I imagine fast digital circuits with tight noise and power budgets may also have trouble with the capacitance of a standard passive hi-Z probe, but we don't notice since nobody uses slow hi-Z passive probes with them.

But the most likely reason is because of logic analyzers .