Simulation reports extremely low (9e-18 amps) leakage through pfet_g5v0d10v5 when VGS is positive. Think this is accurate at all? (PDK docs say to keep VGS negative. Raw data indeed seems to always have VGS negative. Any intuition I can use? This is for a low-leakage switched capacitor.)

It's not accurate. Gate-induced drain leakage (GIDL) will lead to higher numbers in reality. As a rule of thumb, look at the leakage for VGS~0. Making VGS more positive for a PMOS will still give you the same current (or slightly higher).

Much appreciated. Is there any easy/systematic way to learn such things?

I don't know abut an easy way, but taking a course on MOS device physics would do.

Actually, the simulation also reports attoamp currents when everything is in range. Is this also erroneous?

Most likely, yes. You saw our other thread where we talked about gm/ID being off for the PMOS devices. These two issues are links, since gm/ID is just the log-slope of the current. Try the NMOS, it may give more realistic numbers.

Hmm still quite tiny for pretty reasonable voltages.

You can see some realistic numbers here: https://github.com/IHP-GmbH/IHP-Open-PDK/blob/main/ihp-sg13g2/libs.doc/doc/SG13G2_os_process_spec.pdf.

Thanks I don't understand why it's A/um. Length and width are fixed at 10,0.13?

10^-10 A/um = 100pA/um. The right column just documents the device that was used in the measurement. For that 10um device, the leakage current would be 1nA.

Yes you multiply by width, and length doesn't matter for off-leakage? Please excuse my ignorance

Leakage depends on L. But they measured it for L=0.13um and it was 100pA per width in microns. You cannot simply throw L into that calculation; that dependency is complex and would require another measurement in this case.

@Boris Murmann, just as sanity check: Does G[ate]I[nduced]D[rain]L[eakage] affect Ids, or could it possibly cause a bulk-to-drain current? I got seriously stuck with some simulations where I got an unexpected exponential current across a reverse biased bulk diode. • Is there any chance at all that this is real and not just bad modeling? • If it's a bad model, which parameters would I want to tweak to get rid of the bulk current?

@Christoph Maier Could be real, due to SCBE?

Bernhard Boser's presentation goes way back …