Op-amp Spice macro-models article from Intersil
June 16, 2010
Former EDN analog editor Bill Schweber has published a good article from Tamara Schmitz and Jian Wong about developing Spice macromodels for voltage-feedback op-amps. Part 1 (pdf), and part 2 (pdf). All the youngsters like to use Spice for op-amp circuit design but I am more like Bob Pease and Jim Williams, you have to build the circuit to know what is going on. I will never forget being perfectly happy with a Spice run, until I built the circuit and realized that the quad op amp was running way too hot. I did not notice the power consumption of each of the amps was about ¼ W. That was a newbee mistake, sure, but even if Spice does not lie, it is the product of digital and software minds, so rather than flashing a big red sign that warns you that you are going to burn up the quad op amp, they just require you to define a power variable and display it and then print the result in the same tiny test and the blizzard of other information. Then software people smile and fold their arms and tell us everything is our fault, since the information was right there if only we asked for it. The one thing about analog is that is has a sense of importance. That’s why steering wheels and shift levers are big and prominent and radio treble controls are tiny little buttons. If software people designed cars everything would be a tiny little icon and the crash warnings would be in 10-point text.
So anyway, Spice does not necessarily lie like Bob Pease says, but I guarantee you that if you give it poor models it will give you the wrong answer. This is the big hassle with op amp models. Some of them, like the old National Semi Comlinear models (pdf) published by Mike Steffes before he left for Burr Brown and now Intersil were essentially transistor-level models. An IC designer could infer the design of the part from them. Mike told me that he knew that, but it was just so important to give an accurate model that he felt he had to release those great models. If someone wanted to copy the circuits, well, they had a lot more work to do-anyone can de-cap an op amp and reverse engineer it in a day. That still does not give you the process or the testing regime or the design secrets and tricks.
That is why this Intersil article is so important. Anything that helps you make good models is important in a world where kid engineers trust a computer rather than a breadboard. The article give some history of op amp models and that will tip you off as to what you can expect from a simulation. If the model you use does not model for 1/f noise, and most vendor models do not, you cannot get a meaningful simulation of low-frequency noise performance of the circuit. If the model does include flat-band noise and you are designing and ac-coupled video circuit, well that is fine for your needs. I have yet to see a Spice op-amp model that accurately tells you what happens if you bang the output into the rails and saturate the transistors. I will ask Mike Steffes if his old Comlinear models would do that, and leave a comment.
So anyway, Spice does not necessarily lie like Bob Pease says, but I guarantee you that if you give it poor models it will give you the wrong answer. This is the big hassle with op amp models. Some of them, like the old National Semi Comlinear models (pdf) published by Mike Steffes before he left for Burr Brown and now Intersil were essentially transistor-level models. An IC designer could infer the design of the part from them. Mike told me that he knew that, but it was just so important to give an accurate model that he felt he had to release those great models. If someone wanted to copy the circuits, well, they had a lot more work to do-anyone can de-cap an op amp and reverse engineer it in a day. That still does not give you the process or the testing regime or the design secrets and tricks.
That is why this Intersil article is so important. Anything that helps you make good models is important in a world where kid engineers trust a computer rather than a breadboard. The article give some history of op amp models and that will tip you off as to what you can expect from a simulation. If the model you use does not model for 1/f noise, and most vendor models do not, you cannot get a meaningful simulation of low-frequency noise performance of the circuit. If the model does include flat-band noise and you are designing and ac-coupled video circuit, well that is fine for your needs. I have yet to see a Spice op-amp model that accurately tells you what happens if you bang the output into the rails and saturate the transistors. I will ask Mike Steffes if his old Comlinear models would do that, and leave a comment.
