openjellybean — MIT-licensed jelly-bean SPICE models

One include file, jellybean.lib, with original models that behave like the everyday parts. Every model was written from scratch and its parameters fitted in ngspice to typical values from public datasheets — no manufacturer model file was copied or adapted — so the whole set is under the MIT license.

Names end in _OJ so they never collide with vendor models.

.include jellybean.lib
Q1  c b e  BC547B_OJ          ; BJT:      C B E  (SPICE order, not TO-92 pin order)
M1  d g s  2N7000_OJ          ; VDMOS:    D G S  (body diode built in)
J1  d g s  2N5457_OJ          ; JFET:     D G S  (SPICE order, not TO-92 pin order)
D1  a k    1N4148_OJ          ; diode:    A K
X1  a k    BZX79C5V1_OJ       ; zener:    A K
X2  a k    ZENER_OJ params: VZ=8.2 IZT=5m RZ=10 NK=1   ; any voltage
XU1 inp inn vcc vee out TL07X_OJ   ; op-amp: +IN -IN V+ V- OUT
XU2 o1 n1 p1 vee p2 n2 o2 vcc TL072_OJ   ; complete dual: pins 1-8 in order
XU3 o1 n1 p1 vcc p2 n2 o2 o3 n3 p3 vee p4 n4 o4 TL074_OJ   ; complete quad: pins 1-14 in order
XU4 o1 n1 p1 gnd p2 n2 o2 vcc LM358_DUAL_OJ   ; complete dual: pins 1-8 in order
XU5 o1 n1 p1 vcc p2 n2 o2 o3 n3 p3 gnd p4 n4 o4 LM324_QUAD_OJ   ; complete quad: pins 1-14 in order
D2  a k   LED_RED_OJ            ; LED: A K (also LED_GREEN_OJ, LED_BLUE_OJ)
XR1 a w b POT_OJ params: R=10k POS=0.25   ; pot: A WIPER B; POS 0 = at A, 1 = at B

Contents and verified behaviour (ngspice 47, 27 °C)

Rerun every check with python tests/validate.py, or name models to check only those. It needs ngspice 47 or later and Python 3.13 or later, with nothing to install from pip. If ngspice isn't on your PATH, set the NGSPICE environment variable to its location.

The targets and tolerances live in tests/models.toml. See CONTRIBUTING.md to add a model.

POT_OJ is a generic linear potentiometer. Set the total resistance with R and the wiper position with POS; its values are exact by construction.

Use as a Python dependency

pip install openjellybean

or, in your project's pyproject.toml:

dependencies = ["openjellybean>=0.6"]

from openjellybean import library_path
print(library_path())  # path to the bundled jellybean.lib

The path is a real file, so it can go straight into a netlist, for example .include /path/to/jellybean.lib. To pin to an unreleased commit instead, install from git: pip install "git+https://github.com/xdylanm/openjellybean@<tag-or-commit>".

The package version is read from the * Version X.Y.Z line in the header of jellybean.lib. Releasing means bumping that line, tagging vX.Y.Z, and publishing a GitHub release, which uploads to PyPI.

Limits — read before trusting a result

• These are typical devices. Real parts spread widely (BC547 hFE spans 110–800 across groups; 2N7000 VGS(th) spans 0.8–3 V). Run corners by editing BF, VTO etc.
• BJTs: no breakdown or quasi-saturation modelled. Temperature uses standard SPICE coefficients (not fitted).
• MOSFETs use the VDMOS model (ngspice ≥ 47, LTspice). For PSpice you would need a subcircuit rewrite.
• JFETs use the standard square-law model: no gate breakdown, no conduction tail below cutoff, and gate leakage is set by the simulator (about 30 pA at 15 V, against 5 pA typ on J112/J113). Noise is not fitted. Datasheets give wide or no typical IDSS and VGS(off) for most of these parts, so each model is one point inside the limits, and real parts spread far more: run corners by editing VTO and BETA.
• 2N5457_OJ-2N5459_OJ follow the datasheet's typical IDSS and VGS, which give about half the gfs of its typical curves at 3 mA (1.9 mS, about 3.5 mS). J112, J113, PN4392 and PN4393 use LAMBDA=0.035 to approach the published gos, so their drain current climbs more with VDS than a real part at high current (J112: 30 mA at 5 V, 37 mA at 15 V).
• Zeners: VZ temperature coefficient and noise not modelled; knee shape is approximate.
• Op-amps: no input common-mode range limits or phase reversal, no noise, CMRR/PSRR, offset drift, or LM358 crossover distortion. VOS defaults to 0 — use OPAMP_OJ params: VOS=3m ... to add offset.
• These typical values are not matched, because fixing them would mean changing the model structure or trading off other fits: 1N5819 leakage is 1.6 µA at 40 V (about 20 µA on the datasheet curve), 1N4148 leakage is 2.9 nA at 20 V (about 9 nA), and LM358 short-circuit current is 27 mA (40 mA typ at a 10 V supply). BAT54 leakage is 14 nA at 25 V (0.2 µA typ), and B140 leakage is 0.2 µA at 40 V (about 1.7 µA).
• TL07X_OJ matches the slower published TL07x figures, 3 MHz and 13 V/µs; some current versions reach 5.25 MHz and 20 V/µs. BC557B_OJ also uses a slower figure, fT = 150 MHz; some datasheets list up to 320 MHz.
• The LED models are electrical only, with no light output. They represent one typical chemistry per color: green LEDs made with InGaN run near 3.2 V instead of 2.0 V, and red varies with chemistry too. Reverse breakdown is not fitted, and only the green capacitance is checked against a datasheet.
• LM358_OJ used as one LM324 section is slower and draws twice the supply current of a real LM324. Use LM324_QUAD_OJ for the LM324's own figures.
• POT_OJ is linear only, with no log taper, power limit or temperature coefficient.
• Tested in ngspice 47 only; LTspice should accept the syntax but hasn't been run.