10 Common Device Noise Analysis Mistakes – Part 2

This is part 2 of the “10 Common Device Noise Analysis Mistakes” Hot Topic dealing with transient noise analysis. Part 1 covers periodic noise analysis.  

Transient noise analysis (TN) is a statistical time-based technique that applies to every type of circuit. TN injects random noise for each device noise source during a transient simulation to produce output waveforms that include device noise effects. It is then possible to post-process the resulting waveforms to obtain useful frequency-domain measurements. TN is the only device noise analysis applicable to non-periodic circuits and when used properly it should produce results within 1dB to 2dB of silicon measurements.

Common Mistake #7: Not updating TN every device every timestep

Transient noise analysis users may be compromising TN accuracy without knowing. Traditional transient noise techniques trade off accuracy for performance by updating the random device noise injection only at a fixed time interval (noisetmin) and without using instantaneous device bias (noiseupdate). The AFS TN engine delivers results with nanometer SPICE accuracy by injecting random device noise (white and/or flicker) for every noise source at every time step based on each device’s instantaneous bias.

Common Mistake #8: Setting TN noisefmax too low

The noisefmax setting has direct impact on simulation runtime and accuracy. A smaller noisefmax setting will result in a shorter runtime at the expense of accuracy. Traditional TN truncates the device noise spectrum at noisefmax thereby under computing the device noise impact. To compensate for this truncation, users must use a noisefmax that is 2-3X higher than intended, and even then users have no way to know if they have over-compensated. AFS TN ensures nanometer SPICE accuracy and eliminates any guesswork by using the full noise spectrum from noisefmin to noisefmax without truncating the device noise spectrum at noisefmax.

Common Mistake #9: Setting TN tstop too short

Transient noise results are inherently statistical.  For accurate results, it is critical to ensure that the simulation runs a sufficient number of cycles. Selecting a short tstop to reduce runtime, introduces undesirable statistical uncertainty in the result. In such cases the reported result may be significantly higher or lower than the expected actual device noise impact.  AFS TN recommends a tstop value that ensures sufficient cycles to achieve the desired statistical confidence level.

Common Mistake #10: Post-Processing Mistakes

Transient noise generates results in the time domain. For many circuits, the measurement of interest is in the frequency domain requiring FFT based post-processing (e.g. ADC power spectral density and PLL phase noise measurements). There are several pitfalls in post-processing that compromise the frequency domain result’s accuracy including:

• Excessive spectral leakage beyond 2.5 FFT bins

• Signal frequencies that are not exactly centered

• An FFT window that does not minimize spectral leakage

• MATLAB default FFT windows

Using AFS CalcPad to post-process AFS TN waveforms ensures accurate FFT-based results. AFS CalcPad, for example, uses statistical averaging appropriate for stochastic waveforms to minimize deterministic and random errors in the frequency domain.

See additional Hot Topics here.

For more information, contact your BDA application engineer or click here for a web request.