opamp circuit offsets

I don't have enough influence to insist on that.

An offset distribution would be helpful. I have no idea how IC makers arrive at their data sheet specs, or how much they guardband their worst case values. Sometimes that matters; sometimes an assumption of absolute-max worst-case stackup makes a product not viable.

What's a better way to do it?

Except that the simulation adds nothing to finding a solution to the problem. Once you do the high school algebra to find the max offset at

no value in estimating the distribution.

dash-dash-space should be on a line all by itself, followed by the sig ...Jim Thompson

No need to insist - it is there on page 12 fig 34.

Looks to me like they are quite pessimistic about the offset voltage absolute maximum. According to the datasheet fig 34 the mean is -0.025mV with a std dev of 0.25mV and outliers outside +/-1mV rare.

Depends if you can avoid having worst case scenarios occur or figure out some neat way of nulling out the potential offset problems.

Algebra, and in this case the published distribution (which you may want to spot check in case they have over egged it in the datasheet).

I agree that highschool algebra is a far better approach.

The true probability distribution of likely offsets can probably be approximated to engineering tolerances as sin(a.wt)^(2n+1) with n>2

Ricky ragged me for not writing an equation, and followed up by writing a wrong equation. Engineers need to be very careful no matter what tools they use. It's so easy to make mistakes.

Simulation presents the entire circuit as a flat structure. You can, and should, click on all the interesting nodes, check voltages and currents, and sanity check the intermediate terms. Hand-calculated numerical solutions also present intermediate terms that can be sanity checked to some extent. A classic symbolic solution is a black box, basically a procedural language, where there is no way to crosscheck the steps in the solution; so one has to be extra careful when doing a symbolic-math solution.

Once things go nonlinear, symbolic math solutions are generally impossible.

One mistake I tend to make in Spice is to short things by duplicating net names. Cutting and pasting circuit bits will automate that mistake for you.

The only cred that I seriously want is large purchase orders.

Sounds good to me! Whats it do, anyway?

Cool, I missed that. I can safely change my sinewave amplitudes to 1 mV.

The customer is insisting on 1 mV max output offset, which makes no sense. They are already doing "dark" photodiode measurements and subtracting that out from the pulse power measurements, which takes out dc errors, including photodiode leakage. They also insist that the design include no trimpots. Without a zero pot, I'd need to use a tricky autozero loop, or a nonvolatile DPOT, neither of which will fit on the board. So, I'll fight for a pot. A pot with +-3 or 4 mV of trim range wouldn't be a settability or stability issue, with a 1 mV target.

In this simple case, a symbolic solution is reasonable. In a more complex circuit, not so easy. I thought the sine wave thing was fun and made cool patterns.

Spice is easy and (as illustrated elsewhere in the thread) less prone to errors than equations. We save Spice sims in a project folder, which documents things nicely and supports changes easily.

We'll make 8 first-articles, which is at least some statistics.

I don't know why JL can't learn enough math to solve such a simple problem. He should have also known that transposing the symbol numbers had no impact on the equation since all the offset values are the same. But he just likes to whine when he can't figure things out. It is pretty amazing that he has spent so much time solving such a simple problem. Too bad he is too lazy to learn a little math.

Why would this be a reasonable approximation?

"The customer is insisting on 1 mV max output offset". Does that mean relative to _each_ input, or just the _net_ output? ...Jim Thompson

At the differential output, with the input open, 50 ohms to ground internally. The open input is one reason to use that fet-input amp. The current-mode amps have large input bias currents of arbitrary polarity, which wrecks their DC performance.

I have several other impossible specs: noise, bandwidth, slew rate, size. Makes for fun and profit.

The part where they send a check against the invoice and it clears is my favorite.

--sp

It's a photodiode amplifier for measuring the energy of deep UV light pulses.

I don't know a better way. One could test all op amps, and put 'em in bins. That sucks.

One could get a distribution curve by paying the manufacturer for a special run. That causes delay, confusion, adds cost. That sucks.

One could do 100% tests after assembliy. That costs extra, and if you have a criterion for pass/fail, it means rejecting some units. If you just label according to accuracy, the customer has to decide if he needs the A or the B or the C grade. That sucks.

You can trim it all out, with tiny little trimpots, using tiny little screwdrivers. That sucks.

Allowing the errors, max and min... solves problems.

When there is a published distribution, it's not really as good a guideline. Because, there's a production line that has good days and bad days. I once had to heat-sink an op amp after a dozen samples, from different batches, ALL drew quiescent current at the high end of the range (not a 'typical' one could be found). At least I knew it was something that had to be checked.

Are they going to buy a lot of them? Could you test each opamp for offset before installing, and charge accordingly? I guess that is hard on pick and place machines....

George H.

One trimpot would trim the output offset. If the adjust range was, say, +-5 mV, hitting a 1 mV target would be easy.

There is a lot of prejudice against trimpots, but thay can be great little gadgets.

Alternates are a nonvolatile DPOT or an autozero servo using a chopamp, which is an interesting design but takes too much engineering and board area.

John, You've schematically represented your offset errors, but what are the polarities and balances of the "normal" signals you want to process? ...Jim Thompson