INL simulation issue for temperature sensor

To make it a bit easier to understand what the input data looks like, can you post a plot of VS("/vot2") ?

Andrew.

Hello Andrew,

Thank you for your prompt response.

I read the documentation on viva. I couldn't figure out how to keep the horizontal axis to temperature. 

Here is the plot of VS("/vot2").

I'm not entirely sure what you're trying to plot, but the INL function is plotting the deviation of the output voltage from the ideal output voltage. Just because the x-axis of the input signal is temperature, i'm not sure why you would expect the y-axis (or x-axis) of the INL function to be temperature - it would be in the units of the Y-axis.

I could imagine that you could swap the x and y axis and use INL then (maybe) but I'm not entirely sure what that even would mean, since you'd then have swapped what was the dependent and independent variables.

Andrew.

Hello Andrew,

I would like to plot the deviation of temperature with respect to temperature to determine the temperature error.

The temperature sensor’s INL is then the maximum absolute value of the temperature error over the temperature range determined by the two ending-points fit line.

Is it possible to produce Temperature in the Y-axis from result browser and then use INL?

I saw this post from one month ago about the INL end point. He wrote a skill code for it. What I want looks similar but I want my plot at only TT process corner. 

community.cadence.com/.../calculating-the-std-deviation-and-mean-along-the-y-axis-on-a-family-of-waveform

Thank you so much.

Maybe something like this:

waveVsWave(?x waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2")))  ?y waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2")))-abFirstLastLine(waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2")))))

For my made up example, this came up with this curve:

This is using abFirstLastLine from this post. Bit of converting back and forth to change the x-axis, but hopefully it makes some kind of sense.

My made up netlist (I used v("vot2" ?result "dc-dc") rather than VS("/vot2") since I ran from a netlist):

//

R1 (vot2 0) resistor r=1.2125k tc1=-0.0025 tc2=2u
I1 (vot2 0) isource dc=-1m

dc dc param=temp start=-20 stop=100 step=1

Andrew

Hello Andrew,

Thank you very much indeed. It works. 

I have one question. Is it possible to get the temperature deviation from 0? As shown in the following example.

Integral Nonlinearity Error (INL) error based upon an end-point
fit line over the temperature range of [-20C, 100C] In the attached plot, over the temperature range of -20C to 100C, the  temperature error
is bounded within ±0.4C over process corners.

Mine looks like this:

Will it be possible to include the fitting capability in viva calculator in the future releases?

Thank you again.

Dear wgtkan,

in the examples you provided, the deviation does not appear to be "from 0". Is that really what you intended to request? I think, instead, you might want the plot to have symmetrical maximum and minimum INL values for each curve in order that you might easily compare the maximum INL value each case provides. This is what I do anyway. If so, you need to use a linear regression for each curve to find the slope and intercept that minimizes the squared error for that data set and compute the INL of each curve with respect to its respective slope and intercept. This is a different calculation than provided by Andrew's abFirstLast() function.

Andrew has posted a linear regression function which may be used for this purpose.

Shawn

The approach I provided does compute it based on the end points, but that doesn't mean that 0 will be in the centre. That depends on whether your non-linear response is warped in one direction (as it appears to be in both your data and my simple example), or whether it deviates in both directions away from the ideal straight line. So that's a property of the deviation away from the straight line, not how it's calculated.

Yes you could use a best-fit line (as Shawn suggested, but that wasn't what you asked for - the best fit line curve is in the same thread that you found the abFirstLastLine code), but even that wouldn't guarantee to give the same of curve you wanted, plus I doubt the end points would be at zero in that case (unless they happened to be on the best fit line).

I don't believe there's any plan to add the abFirstLastLine function as standard to the calculator, but the whole point of the "fx" button in the function panel is to add functions such as this and then they are as useful as if they were built-in defaults. You'll see that we do have a library of such useful functions here: Custom IC Calculator SKILL Function Library (including abFirstLastLine). 

Andrew.

Hello Shawn,

Thank you for your response. What want to plot is the INL temperature error. A waveform that looks like this:

Andrew has posted a linear regression function that may be used for this purpose. So you mean to use something like this:

waveVsWave(?x waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))) ?y (waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))) - abBestFit(waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))))))

I changed the bFirstLastLine  to abBestFit and got somethig that looks like this:

Dear wgtkan,

When one compute's the INL from a data set and a linear curve fit to the data set, there is no guarantee that one's INL computation will appear symmetric. In other words, there is no guarantee that the minimum and maximum values of the resulting INL computation will be equal in magnitude but opposite in sign. I have a C routine that takes a dataset (x[i[,data[i]) and iterates on the intercept to minimize the overall variation with respect to 0 - or in essence chooses and intercept that will provide maximum and minimum INL values equal in value but opposite in sign.

Alternately, an algorithm to do this which you might consider follows. Hopefully, it makes some sense to you. Let me know if it does not!

1. Find INL from initial curve fit to data set x[i],data[i] with i = 1,N for an N point data set

  a. Compute best fit values of slope_best_fit, intercept_best_fit  b. Compute INL_pass_1[i]

INL_pass_1[i] = data[i] - (slope_best_fit*x[i] + intercept_best_fit)

2. Re-compute INL using a new intercept modified by the average of the maximum and minimun INL value from INL_pass1[i]

a. INL_symmetric[i] = data[i] - (slope_best_fit*x[i] + intercept_best_fit + 0.50*(max(INL_pass_1[i]) + min(INL_pass_1[i])))    or    INL_symmetric[i] = INL_pass_1[i] - 0.50*(max(INL_pass_1[i]) + min(INL_pass_1[i]))

To use this with Andrew's function abBestFit() and the expression you provided, I believe you need to include an additional term that is the average maximum and minimum value of your expression  outermost waveVsWave() . In essence the ymax() and ymin() vanes of the expression contained in the outermost waveVsWave(). If I manage to re-type it correctly, I think it will be something like:

waveVsWave(?x waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))) ?y (waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))) - abBestFit(waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2")))))) - 0.50*(ymax(waveVsWave(?x waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))) ?y (waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))) - abBestFit(waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))))))) + ymin(waveVsWave(?x waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))) ?y (waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))) - abBestFit(waveVsWave(?x VS("/vot2") ?y xval(VS("/vot2"))))))))

I hope this helps...

Shawn