Rapid
New Member
Offline
Posts: 4
Sweden
|
Hi Ken,
When I was simulating my continuous-time Delta-Sigma with transient noise in Spectre, I found that the switch transistors in the current-steering DAC1 were having a major impact on the SNDR. I expected the tail current source to dominate noise performance but the switch pair has about 4 times more noise power than the tail! To make a long story (~two weeks) short, I've simplified the circuit by looking at only the DAC unit current cell: NMOS Diff pair, ideal noise-free tail current source and two noisefree resistors to bias to Vdd.
I've found that when comparing the PSD (A^2/Hz) of the switching diff pair for PSS+Pnoise and transientnoise+FFT, the transient noise shows a lot more noise (especially low frequency noise). At 10MHz, the PSD is about 1000e-27 A^2/Hz and at 300MHz about 100e-27. Pnoise shows a very small noise from the switch transistors of ~20e-27 A^2/Hz at both 10MHz and 300MHz, which I believe is correct.
(When having a larger risetime, the noise at 300MHz matches better but the 10MHz noise (which is in the band of interest) is approx the same.)
I've attached the two testbenches: PSS and tran noise. The tran noise does FFT in Cadence and a script to get the PSD is attached in the Ocean view. (Normally I do this in Matlab but to simplify debugging, I did it in Cadence). I've tested UMC130 and ST65 PDK and the results are similar.
For stationary noise (e.g. resistors), PSS and tran noise match. I would be very happy if somebody could try this and confirm my problems.
Best regards, Mattias
|