ISF Function Extraction in Cadence Virtuoso

Hi there ShawnLogan

I am new to the cadence environment. And I have a related question to ppv / ISF.

When you say that ppv gets saved, is there a separate PPV file for each node of circuit? Like separate for current source? Perhaps for passives as well?

Also could you provide a link to some pdf or paper that explains the relationship between ppv and ISF?

because the ones I found are too mathematically rigorous and do not offer practical insight. 

Thanks a ton

Dear Maryem,

Maryem said:

When you say that ppv gets saved, is there a separate PPV file for each node of circuit?

No. One loads the ppv analysis file in ViVA (as shown above) and then one can plot it for the nodes you have selected to save in the analysis. There is not a separate "ppv" database for each node/current you have saved.

Maryem said:

Like separate for current source? Perhaps for passives as well?

I don't understand your comments about there being a separate file for "current course" or "passives". The ppv analysis is designed for studying the impulse response function (ISF) which is defined for an oscillator - an active, nonlinear circuit - not a single circuit component such as a current source or a passive element. Maybe I don't understand your comment correctly?

Maryem said:

Also could you provide a link to some pdf or paper that explains the relationship between ppv and ISF?

The ISF and ppv results are related by the charge used to disturb the phase of the VCO. The ppv normalizes the phase shift by to the charge. Hence, the overall shape of the two are essentially identical - and the shape is usually of greatest interest in studying phase noise sensitivity.

A relatively recent paper that details this without much mathematical complexity is:

S. Galeone and M. P. Kennedy, "A comparison of simulation strategies for estimating phase noise in oscillators," 2017 13th Conference on Ph.D. Research in Microelectronics and Electronics (PRIME), Giardini Naxos - Taormina, Italy, 2017, pp. 213-216, doi: 10.1109/PRIME.2017.7974145.

You will need an IEEE digital library account, as an example, for access. I am not at liberty to upload the copyrighted material due to the following statement: "Authorized licensed use limited to:..."

Shawn

Dear Shawn

Thanks so much for the detailed reply. 

What I meant by "current source ISF" is for example we see in Hajmiri's paper ""Design Issues in CMOS Differential LC Oscillators"" that there is a separate ISF for the active transistors, and separate ISF with respect to Tail Transistor (current source for biasing). 

With a simple NMOS cross coupled oscillator biased with a current source, there will be two ISF plots. One could be measured at the Drain terminal of NMOS transistors, and another would be connected at the Source of NMOS/ Drain of Current Source. 

Thus, the Oscillator sensitivity of phase is dependent on two injection currents in a different way because of the location in the circuit.

This is what I meant by "nodes" in the circuit that does ViVa allows us to view these different ISFs with respect to different nodes? Or do we have to explicitly select the nodes first, the ones we are interested to see ISF for?

Thanks for sharing the titles of PDFs. 

Dear Maryem,

Maryem said:

What I meant by "current source ISF" is for example we see in Hajmiri's paper ""Design Issues in CMOS Differential LC Oscillators"" that there is a separate ISF for the active transistors, and separate ISF with respect to Tail Transistor (current source for biasing). 

Maryem said:

This is what I meant by "nodes" in the circuit that does ViVa allows us to view these different ISFs with respect to different nodes? Or do we have to explicitly select the nodes first, the ones we are interested to see ISF for?

I think you possibly may not fully appreciate the ppv that is computed by Spectre RF. Basically, the objective of an oscillator pss analysis followed by a pnoise analysis is to determine an estimate of the oscillator phase noise characteristic in light of its various device random noise sources and their frequency translation as a result of the non-linear action of the oscillator's large signal steady-state response. As such, there is not a "single" noise perturbation that is applied, but rather an ensemble of noise sources. Hence, the ISF or ppv provided represents the output phase perturbations due to ALL these noise sources - not simply a single impulse of charge as might be used when performing a set of conventional transient analysis as one varies the phase of the injected charge across the oscillator's period. This is discussed in the paper 

S. Levantino and P. Maffezzoni, "Computing the Perturbation Projection Vector of Oscillators via Frequency Domain Analysis," in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 31, no. 10, pp. 1499-1507, Oct. 2012, doi: 10.1109/TCAD.2012.2194493.

which is referenced in section "V. NOISE UPCONVERSION WITH Gammappv"

of the paper entitled "A comparison of simulation strategies for estimating phase noise in oscillators" I referenced earlier.

If you are interested in computing the ppv of a specific charge injection point, you will need to perform the far more intensive and time consuming set of conventional transient analyses in lieu of the time efficient analyses provided by spectreRF  - not to mention the fact that end result may not truly represent the dominant noise sources contributing to oscillator phase noise.

Does this help and did I answer your question with enough clarity (or did I not fully understand your inquiry)?

Shawn

Thanks again. 

So with spectre PPV plot,  i cannot know how much each of the nodes contributed towards it. Its just the net total of all nodes' noise sources. 

I agree with you that Transient analysis is very much time consuming & prone to error. But with this, one can have a separate ISF calculated. As shown in the paper

"Efficient Calculation of the Impulse Sensitivity Function in Oscillators" by Salvatore Levantino in which they use both Transient Method and PXF analysis method with some post processing. At the end they obtain "two separate ISF plots" from each of the methods.  So, for their oscillator design one knows where the exact contribution is coming from. 

So my Question is:

If Spectre ppv plot gives the net result, I am not sure how it combines the different ISFs. Based on what algorithm.?

Because ISF/ppv is essentially just sensitivity of output phase to injected current. And this is always a function of "where" the current is injected.

Or maybe it calculates ISF by default only for nodes that were selected in PSS analysis as Oscillator node+ and Oscillator node-. And ignores the other nodes. (perhaps?)

p.s

the paper that you mentioned ""Computing the Perturbation Projection Vector of Oscillators via Frequency Domain Analysis" is also by the same author as I suggested. But in this paper he has analysis for just one injected current source for LC oscillator while in the paper that I suggested he has made analysis for tow different current sources used at two nodes by injecting current. 

Thanks

Dear MaryEm,

Maryem said:

If Spectre ppv plot gives the net result, I am not sure how it combines the different ISFs. Based on what algorithm.?

I'm not sure I was clear, the ISF (or equivalently ppv) is the phase variation that is produced is a result of the all the individual noise components in the steady-state solution computed in a large-signal sense (i.e., including folding noise). There is only a single ISF.

Maryem said:

"Efficient Calculation of the Impulse Sensitivity Function in Oscillators" by Salvatore Levantino in which they use both Transient Method and PXF analysis method with some post processing. At the end they obtain "two separate ISF plots" from each of the methods.  So, for their oscillator design one knows where the exact contribution is coming from. 

The purpose of the Levantino paper is to show a number of computational methods provide the same basic result. It is totally unrealistic to assume the output noise of an oscillator is due to a single source of charge injection. The spectre pnoise analysis is designed to study the phase noise of an actual VCO or oscillator and hence the ppv represents this capability.

Maryem said:

If Spectre ppv plot gives the net result, I am not sure how it combines the different ISFs. Based on what algorithm.?

It does not as I mentioned above. It computes the ppv based on the total output noise produced by all device noises and their folding due to the large signal steady-state response of the nodes you specify as the outputs in the pss/pnoise GUI dialog boxes.

Maryem, I might suggest you please refer to the Cadence documentation of the pss and pnoise analyses. I get the impression you may not understand these two analyses.

Shawn

Dear Shawn

Thanks for the suggestions, I will look into the documentation.

I just meant that in addition to comparing methods, he has computed Two ISFs with each of the methods for the two oscillators. 

But Anyhow. 

Thanks again for all the feedback