@Christoph Maier the B-source in xschem is called devices/bsource.sym. You can give it any expression (wrap inside double quotes): the syntax for the FUNC attribute that represents the function that is passed to ngspice is:

FUNC="{expression}"

the braces are ngspice syntax. You can also use single quotes:

FUNC="'expression'"

the example shown in image produces this line;

B1 IOUT 0 V = {OFFSET + AMPLITUDE/2*(tanh(V(IPLUS,IMINUS)*GAIN*2/AMPLITUDE))}

During the time it took you to get back to me, I found not only one, but two solutions. First, I remembered https://github.com/MastellaM/sky130_TAC3/blob/tatzelbranch/sensormodels/b_source_magic.sch, lines 33-35. but I also found

devices/asrc.sym

and used it here: https://github.com/tatzelbrumm/sky130_cm_ip__biasgen/blame/main/xschem/test_isweep.sch, line 84. The problem here is not that I don't know how to do B source magic, cf. https://github.com/tatzelbrumm/teachingskilldemo/blame/master/LTspice/TimeVariableCoupling.asc, lines 69 and 72 (sorry, too busy right now trying to meet an arbitrary and capricious chipalooza deadline that could be used as chokepoint to enforce the Welch Matrix, cf. , if there is a serious inconsistency between the values of Tim Edwards at ISSCC1998 in Monterey [https://link.springer.com/chapter/10.1007/978-1-4757-6414-7_5] and today's efabless company to fire up LTSpice and endanger the very fragile xschem/ngspice tool chain sharing the same computer to take a screenshot) but that its implementation in

xschem

is inconsistent and subject to change without notice. That makes the various and sundry, inconsistent shotgun parsers [https://langsec.org/brucon/ShotgunParsersBruCON.pdf] in your tool chain (including, prominently, xschem) a chokepoint (by accident or by design, cf. https://chokepointcapitalism.com/) for any reliable, predictable design based on your open source tool chain. Nearly everything that matters is a side effect. — Meredith L. Patterson