Hand analysis vs real simulation

[Blilix]

Hi,
I have a question about the hand calculation and real simulation result in analog design. I use Level 1 model in hand calculation but in the simulation, the simulator uses higher level model (around 49 I quite forget). How big difference between the result from hand analysis and simulation in terms of DC-bias, performance? I know that my question is too general but I want to know how people start their analog design for example CMOS op-amp.

Thanks in advance.

 

[antonio_eda]

Good modeling is very important in analog design. Sometimes even latest computer models are bad for some applications. Each model has its validity, limits. For example, MOSFET models are nonquasistatic so they are not so good in analysis of distributed circuits. Level1 is good enough for hand calculations --- we have to start from something --- but for real simulation it is too simple.

Validity of a model depends on applications.

 

[mengcy]

Hand calculations is very important,if you consider the body effect and channel length modulation effect,hand calculation results can almost be similar to the simulation results.

 

[Sadegh.j]

What i usually do is to run a very basic hand calculation, with an many approximations as possible, to get a basic idea of how to design a circuit and what each component does. Also this gives me a basic W/L for the transistors. Do NOT try to be picky here, you you will have to change them in simulations, as your first design never works.

Then go to Hspice / Cadence / ADS, plug in your calculation values and simulate your circuit. Then start playing around with the Ws of the transistors and monitor the output specs to get a feeling of how each parameter affects each spec. (Write a quick summary), then try to design your circuit from there. In analog designs with about 10-15 transistors, this usually take few hours, but it will help you later in the design stage.

Cheers

 

[jgruszynski]

Just repeating what others have said.

In general, every model without exception is an approximation of reality and thus in any absolute sense it "lies" about reality. The common sense strategy is to know when the model starts to "lie" and when it comes to computer simulations that requires doing a hand calculation beforehand.

A classic example I had in grad school when I was TAing a senior analog design lab. A guy was having problems with his amplifier design as it wasn't giving the expected gain. I took a look and the problem was obvious: he had "designed" for voltage gain of 100, was using a 10V Vcc and had 1V input level, yet he was surprised to get a gain of less than 10! He was quite upset and insistent that "reality" was violating the results the model said were true! I never got through to him. Clearly he never understood the difference between the small-signal model and large-signal model (or much else like KVL apparently)!

Computers don't really understand circuits any better than this guy did. They're just using the model they were programmed with. Using a computer to simulate reality with a model simply means it can "lie to your faster".

This is why it's critical to hand calculate and to hand verify simulations to some extent when looking at the results. Does the answer the simulation gave make sense physically or based on your hand calculations?

In short, don't expect 100V out of a circuit powered by 10V even though the small signal model (or simulation) "says" you should get 100V.

 

[nl5]

I already mentioned in several posts my simulator NL5, which would be a good tool to be used instead of (or along with) what you call "hand calculations".
The idea of this simulator is that it is using very simple "generic" models: ideal diodes and transistors, OpAmps, switches, R,L,C, V and I sources, and much more. At the very beginning of the project, when you just estimating how schematic works, you don't need specific models yet, or may be you even don't know what real components exist at all, so you just use a "diode" with two states: On and Off, OpAmps with an infinite gain and bandwidth, etc. This is exactly what you do in your head or on the paper, but all the calculations are more accurate, plus you have all the graphs, you can easily modify schematic and change parameters, etc.
When you're done with estimation, you can switch to simulators with real component models (if you trust the models...) such a LTSpice, Tina, MicroCap, etc., to do detailed analysis of the schematic.
Thanks.

 

[flatulent]

Hand analysis is infinitely important. It lets you have a real feel for what the circuit will do. The computer simulation is only for second order effects. The hand alnalysis will let you know what to change for the better. The CAD program will only let you know what exists and not what is possible.

 

[rfsystem]

A good complement to hand calculation is a practive which Barrie Gilbert called

"Foundation Design"

It uses spice and a model library where you could activate specific physical effects in the model. So starting the circuit design by topology selection, hand calculation, dimensioning and then simulation which the most simple, ideal device model. If add or switch on different effects in the model you can estimate the circuit effect.

This practice could be done with MOS and bipolar. Barrie's course papers use only bipolar but the practice is useful for passive also.


Fab's do not want to deliver this kind of productivity improvement with kits because it allow to compare technologies and place them in risk. Better spinning many masks and tune finally on the tester than lost a customer. It the same story everywhere not only finance.


Course Description from Mead Education

Designing ICs for Manufacturing Robustness
Barrie Gilbert, Analog Devices

Monolithic design has, from the earliest days, placed dependence on the high probability of like devices (transistors, resistors, capacitors, etc.) matching very closely to ensure accurate operation of the overall circuit. While this is well-known, there are many other issues that need attention if an IC product is to be manufacturable in high volumes. This module discusses these issues at some length, beginning with the concept of "Foundation Design", which is step-by-step procedure for exploring individual device sensitivities.