Integrated noise for opamp is not converged.

Spectre (as do all circuit simulators) compute the total output noise - that's the primary output. They do this by adding a number of noise sources around each device, where the magnitude of the noise at the source is dependent upon the size/value of the device and the operating point (depending on the type of noise). It also computes the transfer function from this noise source to the output of the circuit. It multiplies the noise at the source by this transfer function, and then sums up the total noise power at the output of the circuit. This can then be represented in either V^2/Hz or V/sqrt(Hz). If you plot the waveform representing the noise at the output, you'll get one of these - noise versus frequency. You can then integrated it over a band, e.g. using the rmsNoise function in the calculator, or using the noise summary form - it does a straightforward integration to give the noise in a certain bandwidth.

Input referred noise is computed by simply dividing the total output noise by the transfer function from the designated input source to the output. If there's more gain, there's less input referred noise (at a particular frequency) and vice versa. So whether it "converges" is dependent upon the bandwidth of the circuit. Note that input referred noise is not terribly meaningful - it's only useful with knowledge of the transfer function of the circuit - and is just a way of representing the output noise in a different (and sometimes more convenient) manner to aid hand calculations and noise budgeting in your design.

Now, I can't comment on your circuit (without having seen it), but the first step would be to look at the noise summary in ADE (Results->Print->Noise Summary) which allows you to filter the devices and identify which are causing the biggest noise contribution. It's very unlikely that it is wrong unless you have some fundamental modelling error or the devices are not the size you thought they were. Note that all the noise contributors are in terms of the output referred noise they produce - so you can compare their impact.

Regards,

Andrew.

Hello Andrew,

     Thank you very much for explaining how Spectre calculates the noise. I am clear now.

     I think maybe I can use the total integrated output noise to be divided by DC gain of the opamp to calculate the input referred noise by my self. If I remember correctly, in designer guide, there is a post about the calculation of input referred noise, it is

total integrated output noise \ DC gain.

    Am I right?

    Thank you.

    Jacki.

Jacki,

No, it would not be divided by the DC gain. The noise at each frequency should be divided by the gain at that frequency (which is what spectre does). Dividing by the DC gain is not going to be terribly meaningful if there is any kind of gain variation within the noise bandwidth.

If you're simulating a switch capacitor circuit, you should use PSS and PNOISE (there's a paper on this on the Designer's Guide too), because otherwise you're only finding the noise transfer function at a single bias point, rather than giving you the time-averaged noise over the clock period.

Regards,

Andrew.

Hello Andrew,

     Thank you very much to correct my misunderstanding.

     I will do PSS and PNoise in SpectreRF.

     Best Regards!

     Jacki

Jacki,

Note that if the part of the circuit you're analysing is continuous time, then you wouldn't need to use the periodic analyses, but if it's switch cap, then almost certainly you would.

Good luck!

Andrew.

Hello Andrew,

    Can I ask a question not related Cadence Spectre usage?

    For the switched-capacitor amplifier, if I want to realize the accuracy is 10 bits, and I assume only the noise influences the accuracy, other mismatches, distortions are ignored. The amplification of the SC-amplifier is 2 (magnitude) with the input peak to peak signal 100mV, and therefore the output signal peak to peak is 200mV. 

    Now if I want to realize the 10-bit accuracy, I think I have to calculate the integrated noise, then referred to the input. Am I right? Actually I am confused about calculate the accuracy of SC-amplifier. Can you give me some comments, hints?

    Thank you!

    Jacki

Hi Jacki,

Not sure why that would imply you have to do the noise calculation referred to the input. Presumably you could calculate the magnitude of an LSB at the output of the circuit (if you're telling me that a full scale output is 200mV pk-pk then an LSB would be roughly 2mV pk-pk. It's whether the noise at the output is large enough to mask the output signal. Input referring it doesn't particularly tell you any more; after all, the noise is really appearing at the output, so a comparison there would seem sensible to me.

Anyway, there are almost certainly better places to find this information (it's been a while since I designed switch capacitor and DAC/ADCs, so I'm a little rusty!) - numerous books, or the design forums on the Designer's Guide

Best Regards,

Andrew.

Hello Andrew,

     Thank you very much for your help. I will check the output. The reason I am checking the input referred noise is my boss tell me for the SC-amplifier, the input sampling noise is around KT/C, so I want to add the amplifier's noise together, then I can estimate the accuracy of SC-amplifier.

    Best Regards!

    Jacki

Hi jacki,

its enough to consider the noise at the output (over the band that you're interested in). Anything at the input of the amp will be amplified by the gain and therefore referring the noise back to the input doesnt really help you. (m*A)/(n*A) and m/n have the same result after all.