GUI tutorial: two solar cells

Created by Claude Opus 5 (Anthropic), 2026-08-17.

This tutorial builds two complete devices in the Simudo GUI, runs them, and looks at the results. It is a guided path rather than a manual: every panel is described where you first need it, and there is a short reference at the end.

You will build:

1. A silicon p-n diode — in the dark first, then under AM1.5 illumination, and finally a run that stops at a checkpoint and resumes.

2. An intermediate-band solar cell with a heterojunction — p-Al₀.₃Ga₀.₇As | IB | n-GaAs, illuminated.

Starting the GUI

For most users, the GUI runs on your own machine and submits the actual simulation elsewhere — to a Docker container or over SSH — so you do not need FEniCS locally, only the GUI’s own dependencies. The installation instructions cover this in full; in brief, you want to make an environment with the GUI’s dependencies:

conda create -n simudo_gui python=3.11 \
  panel param numpy scipy matplotlib bokeh pyyaml -c conda-forge
conda activate simudo_gui

and then install Simudo itself without its dependencies, since you want the GUI code only rather than the solver:

pip install --no-deps simudo

Then start the GUI, from whatever directory you keep your project files in:

cd ~/my-simudo-projects        # wherever you want your project files to live
python -m simudo.gui --show --port 5006

Launch it from your own project directory rather than anywhere else: the GUI’s Open and Save dialogs start in the directory you launched from.

To open a particular project at startup, give the path to its yaml file as an argument:

python -m simudo.gui my_device.yaml --show --port 5006

Without one, the GUI reopens your most recent project — or, on a first run, the bundled example. Any other arguments are passed on to panel serve, so its options (--port, --address, and so on) work as usual.

The GUI should open in your web browser.

If it stops with ModuleNotFoundError: No module named 'simudo', then Simudo is not importable from the active environment — most often the pip install --no-deps simudo step above went into a different environment from the one that is now activated.

Note: Working from a development clone rather than a pip install? Make the clone importable first. From its top directory:

pip install -e . --no-deps

then launch as above.

---

Part 1 — A silicon p-n diode

1.1 Layers and materials

Simudo in general can treat 2D devices, but the GUI is designed for 1D structures.

In the GUI, a device is a stack of layers. Each layer has physical properties: thickness and material parameters, and a mesh specification. The physical properties of a layer can be specified with material files or given manually, which we will show below.

A new project starts with an empty stack:

Choose a save location for your project using the Set save location button at the top. Then add two layers with + Add layer, and set their names and thicknesses. Leave the Material column alone for now — the project has no materials in it yet, so there is nothing to choose:

Layer	Thickness
p	1 µm
n	99 µm

Next, the doping. Click the coloured dot at the left of a layer row — circled above — to open that layer’s Properties on the right. Under DOPING, pick n-type or p-type and type the Concentration as a positive number. You never type a sign:

Layer	Type	Concentration
p	p-type	1e17 cm⁻³
n	n-type	1e17 cm⁻³

Underneath, the GUI stores this as the single signed charge density Simudo actually solves with, poisson/static_rho — negative for acceptors, positive for donors. That is the name you will see in the saved project file, and in the “parameters not set in any region” list.

Now give the layers a material. Go to the Materials panel, find SiliconMaterial in the library and click Add to project. A library material supplies band gap, effective masses, mobilities, thermal velocities, SRH lifetimes, permittivity — and, for silicon, a measured absorption spectrum that §1.5 will use.

Return to the Layers panel and set the Material column of both layers to Silicon, which is now offered in the dropdown:

Compare this with the two-layer figure above. There, a parameters not set in any region panel listed four band parameters — CB/energy_level, VB/energy_level and the two effective densities of states. All four came with the material, so the panel has nothing left to report and disappears.

One parameter was there from the start and is worth knowing about: temperature, on the domain overlay, at 300 K. No material supplies it — it is a property of the experiment rather than of silicon — so a new project starts with the default. Click the dot on the domain row under Overlay regions to see it, and change it there if you are not working at room temperature.

Warning: Take the “parameters not set in any region” list seriously. An unset parameter does not fall back to something sensible; it is zero, and a zero mobility or a zero density of states will fail in the solver rather than at the point where you forgot it. The list is the GUI telling you what the run is missing, before you spend the time on it.

Note: Any property you set manually on the Layers panel overrides the material’s value, because material rules have the lowest priority. So you can load in a material and manually override any parameters you want to change, without changing the material file.

1.2 Bands

A silicon diode needs the two ordinary bands. Both are Boltzmann bands, which is the right choice unless you are working at degeneracy:

Band	Type	Carriers	Extent
CB	Boltzmann	Electrons	Domain (all layers)
VB	Boltzmann	Holes	Domain (all layers)

Carriers is an Electrons/Holes toggle, not a number: it says which sign of charge the band carries, and the GUI turns that into the −1 or +1 you would see in the saved project file. Extent says where the band exists; Domain (all layers) is the whole device. Part 2 uses a band that exists in only one layer.

1.3 Recombination

Processes move carriers between bands. For a dark diode you need one: SRHRecombination, Shockley–Read–Hall recombination through midgap traps.

Field	Value
Name	SRH
Class	SRHRecombination
Lower E band	VB
Higher E band	CB

Simudo names the two bands of a transition by energy: the lower energy one is the source band, the higher energy one the destination (with source and destination so-named for modeling the associate generation process).

The silicon material already supplies SRH/CB/tau and SRH/VB/tau (1 ms) and the trap energy level, so this device would run as it stands. We will override the lifetimes anyway, because setting parameters may be something you will do frequently, and this is the clearest place to learn it.

Go back to the Layers panel and click the dot on the domain row under Overlay regions. The Properties pane opens on the right. In the Add property box, choose SRH/CB/tau, type 1e-6 with unit s, and click Add. Repeat for SRH/VB/tau. Both bands now have a 1 µs lifetime everywhere in the device.

Now set a different lifetime in one layer. Click the dot on the p layer row, add SRH/CB/tau there, and give it 5e-6 s. The p region now has a 5 µs electron lifetime while the rest of the device keeps 1 µs:

A parameter set on a layer beats the same parameter set on domain. The priority order is named overlay region > layer > domain > material, so the more specific statement about a region wins. That is the mechanism you use to give one layer different properties from its neighbours. Note where domain sits: it is the device-wide default, so a layer overrides it, while a named overlay region — the kind you will build in §2.1 — overrides everything.

Note: The 5 µs override on the p layer barely moves this device’s answer — Jsc changes in the fourth decimal place. The p layer is 1 µm thick and the electron diffusion length there is tens of microns, so almost every carrier reaches the junction whatever the lifetime. It is here to show the mechanism. The lifetime that does matter for this cell is the one in the 99 µm base — see Absorbing is not collecting.

Warning: A parameter you set in one region is not set everywhere. Unset parameters default to zero in the regions you did not set them in, and the Layer panel’s “unset parameters” list stops warning you as soon as the parameter is set somewhere. Set parameters on domain when you mean them to apply device-wide.

1.4 Contacts, and the first run

Go to the Boundary Cond. panel. By default, both contacts are ohmic on both bands. One contact is swept (the voltage is applied there) and the other is the reference. You can instead make zero current or surface-recombination velocity (SRV) boundary conditions for each band at each contact.

In the Simulation panel, choose “Dark J-V” for the mode and list the voltages 0 to 0.6 V in 0.1 V steps. You can use the linspace generator to help make the voltage points; in this case, choose 7 points from 0 to 0.6 and click “Apply.” You can choose whether to save a checkpoint (from which you can restart the simulation) at the end of the voltage sweep.

Set the maximum mesh spacing before you run. It is the Max mesh spacing box in the same panel, and it defaults to 0.02 µm — sensible for a micron-scale cell, far too fine for this one. Left alone it puts about 5000 points into a 100 µm device and the run drags on for many minutes. Set it to 1 µm: the mesh becomes 174 points and the run takes under half a minute. Each layer also has its own starting spacing and expansion factor, on the Layers panel; the defaults (0.005 µm, growing by 1.2×) keep the mesh fine at the junction and at the contacts, which is where it needs to be, and the maximum spacing caps how coarse it is allowed to get in between.

1 µm is not a compromise here. Refining this device from 129 points to 476 changes every point of the J–V curve by less than 0.01 %, and even the most delicate quantity in §1.5 — the near-cancellation of generation against recombination in the middle of the base — moves by about 1 %.

By default, XDMF output will be produced. If you want to make spatial plots in the GUI, this option is required. You can also choose to get csv outputs of all spatial quantities.

The execution profile at the bottom of that panel is where you say where the simulation runs — a local process, a Docker container, or a machine over SSH. Set that up (see the installation instructions) and choose “Set as Default” to avoid needing to do so again. Then press ▶ Run Simulation and watch the log.

This device takes about half a minute on an M2 Mac Mini once the JIT cache is warm; the log scrolls as each bias point converges. The very first run of your life will be slower, because FEniCS compiles the forms.

What you should get

Open the Output panel. It already points at the run that just finished, and the J–V curve is drawn; switch the scale to Log|J| to see the exponential over several decades. Use 📁 Browse… and ↻ Refresh when you want to look at an earlier run instead.

The dark diode gives a textbook exponential:

V (V)	J (mA/cm²)
0.0	−1.7e−19
0.2	1.8e−4
0.4	0.34
0.6	738

The current rises by a factor of about 46.5 per 100 mV. Since n = (q/kT)·dV/d(ln J), that is an ideality factor of 1.008 — an almost perfectly ideal diode, which is what this device should be.

1.5 Turning the light on

To study this device under illumination, we need to make the material absorb light and add the light itself.

The absorber. Back on the Processes panel, add a second process:

Field	Value
Name	opt_cv
Class	BeerLambert
Lower E band	VB
Higher E band	CB

Note: Name it exactly opt_cv. A material supplies its absorption coefficient under a key that includes the process name — opt_cv/alpha_function — so a process with a different name will not find silicon’s absorption data. If the absorption coefficient is missing you get a warning in the log and a device that generates nothing.

Silicon’s measured α(E) (M. A. Green, 2008) comes from the material file, so there is nothing to enter in the optical absorption coefficient section — the panel says so, in green: ✓ provided by a material (opt_cv/alpha_function). If you want an artificial absorption instead, you can add a top-hat in the process card. A top-hat absorption is an energy-independent absorption coefficient between cutoff energies E_min and E_max. If you add one, it sits on one or more layers, and layer rules outrank material rules, so it will override the α(E) from the material file.

The light. In Opt. Fields, add a field set:

Field	Value
Name prefix	sun
Direction	+x
Bins	6
E min / E max	1.12 eV / 4.0 eV
Spectrum	am15g
Concentration	1.0

Simudo divides the spectrum into bins of equal photon flux and solves Beer–Lambert propagation for each. There is no point including photons below the lowest energy your device can absorb — silicon’s gap is 1.12 eV.

Finally, in Simulation, change the Mode to Illuminated J-V. The optical ramp walks the illumination up from zero to full intensity before sweeping voltage.

Add a few voltages while you are there. Under illumination the interesting part of the curve is the knee near open circuit, and 0.1 V steps step straight over it: the current is −7.8 mA/cm² at 0.5 V and +714 at 0.6 V, so the fill factor comes out of thin air. Use 0, 0.1, 0.2, 0.3, 0.4, 0.45, 0.5, 0.55, 0.6. Keeping 0.6 V is worth it as a check on the whole calculation — see the note below.

The illuminated run takes two to three minutes — the intensity ramp and the optical solve at each bias point cost more than the dark sweep did.

Looking at the results

The Output panel reads one run at a time. Point it at the illuminated run’s folder and you get the illuminated J–V; point it at the dark run’s folder and you get the dark one. Flipping between the two folders is how you compare them.

The curve now passes through the fourth quadrant: negative current at zero bias (the device delivers power) crossing zero at the open-circuit voltage. The figure above uses Log|J|, which is the readable scale for this sweep: on a linear axis the 0.6 V point, at 714 mA/cm², is thirty times the short-circuit current and flattens everything else onto the zero line. On the log axis the plateau at 23.8 mA/cm², the dive through open circuit near 0.5 V, and the exponential above it are all visible at once. Reading off this run:

Short-circuit current	≈ 23.8 mA/cm²
Open-circuit voltage	≈ 0.504 V
Fill factor	≈ 0.80
Efficiency	≈ 9.6 %

The first two are visible directly on the plot — J at V = 0, and where the curve crosses zero.

Note: The 0.6 V point is a free check on the whole calculation. An ideal solar cell obeys superposition: the illuminated curve is the dark curve shifted down by the photocurrent. Here the dark run gave 737.8 mA/cm² at 0.6 V and the short-circuit current is 23.8 mA/cm², so superposition predicts 714.0 — and the illuminated run gives 714.3. Agreement to 0.04 % says the two runs are describing the same device. It is worth doing on a device you do not already trust; where superposition fails, the reason is usually physical (an injection-dependent lifetime, series resistance) rather than numerical.

A band diagram. Make sure that you are looking at the illuminated folder. Below the J–V section, choose a (V,I) point from the dropdown and press ⟳ Load Spatial Data. I is usually 0 (dark) or 1 (fully illuminated, using the fields set in the Opt. Fields panel.)

You get the band edges and quasi-Fermi levels across the device at that bias. The figure shows (0.4, 1). Try (0, 1) as well: at short circuit the bands are more steeply tilted, and by 0.4 V — most of the way to the 0.504 V open circuit — the tilt across the base has largely gone and the two quasi-Fermi levels sit about 0.4 eV apart, flat across the whole 99 µm. Flat quasi-Fermi levels mean no net driving force: the current at that bias is being carried by diffusion down a concentration gradient too small to see on this scale.

Looking inside the device

The J–V curve says what came out; the spatial plots say why.

Important: Note: Spatial plots read the solution fields themselves, not the summary CSV, so the run has to have saved them. In the Simulation panel, under Output, turn on the XDMF mesh output before running. Without it, Load Spatial Data stops with “No sim_V=*_full.xdmf or sim_I=*_full.xdmf files found”. Extraction also needs dolfin, so it runs through the execution profile — the same machine that ran the simulation.

In the Output panel, scroll to SPATIAL PLOTS, press ⟳ Load Spatial Data, then ＋ Add Plot. Each plot takes one or more lines, and each line picks a quantity and a point in the sweep.

Important: To reproduce the plots shown here, choose the (V, I) point (0, 1) — zero volts at full illumination. The list is labelled (V, I), and (0, 0) is the equilibrium starting point, where generation and current are both essentially zero and both plots would be essentially flat. For spatial plots, you can also choose to display all of the (V,I) curves on the same axes.

Plot 1 — generation. One line, quantity g_CB, scale Log. This is the net rate at which electrons are generated in the conduction band per unit volume: optical generation minus recombination. It falls by four orders of magnitude over the first 60 µm, then turns around and climbs again toward the back contact.

That turn-around is worth understanding, because it is not a change in the light. Absorption decays smoothly across the whole device; what changes is how much of it survives locally. All three rates below are in cm⁻³·s⁻¹:

Depth	optical generation	SRH recombination	net
0.5 µm	6.4e20	−2.2e16	6.4e20
30 µm	8.7e18	−4.8e18	3.9e18
60 µm	3.0e18	−3.0e18	9.2e16
99.5 µm	1.4e18	−4.3e16	1.4e18

In the middle of the base, recombination cancels almost everything the light creates — 97% of it at 60 µm. Right at the back contact the ohmic boundary holds the carrier densities at their equilibrium values, so there is no excess to recombine and the net rate returns to the optical rate. You can see both components separately by plotting g_opt_cv_CB and g_SRH_CB on the same axes.

Plot 2 — components of the current. Two lines, j_CB and j_VB, both at (V, I) = (0, 1), on a linear scale. The hole current is large just past the junction and decays away; the electron current grows to meet it and then flattens. Both are current densities in mA/cm²:

Depth	j_CB	j_VB
1 µm	−4.9	−18.9
5 µm	−15.7	−8.1
20 µm	−23.4	−0.4
50 µm	−25.0	+1.3
99 µm	−25.4	+1.6

Both curves are flat beyond about 30 µm. The hole current is essentially zero deep inside the n-type layer. Minority carriers produced deep inside the device are not contributing to the current.

Note that j_tot is the total current, which is spatially uniform in a 1D device and equal to the current density shown in the J(V) curves above. It is only interesting to plot when checking whether simulations have converged.

Note: The GUI’s plots are for looking, not for publishing. Anything more involved is better done outside it. Two J–V curves can be compared by reading the sim_V.csv files written with each run — one row per bias point, one column per contact and band. Band diagrams and other spatially resolved quantities are best handled with the tools in simudo/example/fourlayer/sweep_extraction.py, which load the XDMF output directly and give you the fields as arrays.

Absorbing is not collecting

Every photon above silicon’s 1.12 eV gap in the AM1.5 spectrum, if collected, would give 43.8 mA/cm². This cell delivers 23.8 — a little over half — even though it is thick enough to absorb nearly all of that light.

The two plots explain the gap. The minority-carrier diffusion length in the base is

for the 1 µs lifetime and silicon’s hole mobility. The base is 99 µm deep. Carriers generated further than a diffusion length from the junction recombine before they reach it. The cell absorbs the light but cannot collect the holes.

The lifetime that governs it is the one in the base — the n layer here — so it is a good exercise to set SRH/VB/tau on the n layer, as you did for the p layer earlier, and watch Jsc move.

1.6 Checkpoints and resuming

Long sweeps are worth checkpointing. In Simulation, set the checkpoint directory and ask for a checkpoint at the end of the voltage sweep.

To continue, open the project, set resume from to the checkpoint file — ../c/checkpoints/sim_V=0.6.yaml, relative to the new run’s output folder — and extend the voltage list to include 0.7 and 0.8 V. The run picks up where it left off, which you can confirm in the log — it prints a line like:

Resuming from checkpoint: V=0.6  (stage=up_v, .../checkpoints/sim_V=0.6.yaml)

and then continues from 0.6 V rather than starting over at zero. Adding two new voltage points takes less than 20 seconds, and the Output panel now lets you visualize the device under all of the voltages.

You can also resume from a checkpoint using the Checkpoint selector at the top of all panels in the GUI.

---

Part 2 — An intermediate-band cell with a heterojunction

An intermediate band (IB) is a narrow band inside the standard semiconductor band gap. A photon too weak to cross the whole gap can promote a carrier from the valence band to the IB, and a second photon can take it from the IB to the conduction band. Two sub-gap photons therefore do the work of one above-gap photon, and the devices in principle can have the voltage from a larger band gap while still maintaining high current.

Here we demonstrate both IB solar cells and heterojunction devices by making an artificial IB solar cell from an AlGaAs/GaAs heterojunction. This is not a good IB cell, just one to demonstrate how to simulate them.

Start a new project for this one — File ▸ New, then set a save location as you did in §1.1. temperature is already on domain at 300 K.

Go into the Materials panel to add the GaAs and AlGaAs materials to the project. In the Layers panel, construct a 3-layer device:

Layer	Thickness	Material	Doping	Type
p	0.5 µm	AlGaAs (x = 0.3)	1e17 cm⁻³	p-type
ib	2 µm	AlGaAs (x = 0.3) + an IB	1.25e17 cm⁻³	n-type
n	2 µm	GaAs	1e17 cm⁻³	n-type

The first two layers are the same alloy, so there is no heterojunction between them. The ib | n interface is a real one: AlGaAs and GaAs differ by 0.29 eV in the conduction band and 0.10 eV in the valence band.

The mesh needs attention. Leave the maximum spacing (on the Simulation panel) at its default 0.02 µm — in section §1.4 you raised it to 100 µm for the silicon cell. Here the whole device is 4.5 µm, so the default is already the right scale. What matters instead is the starting spacing at each layer’s edges. Click each layer’s dot on the Layers panel, expand MESH OPTIONS, and set Starting mesh spacing = 1e-5 µm — the box will show it as 0.00001 — leaving the expansion factor at 1.2. Do it for all three layers.

That is five hundred times finer than the 0.005 µm default, and it matters:

starting spacing	mesh points	Jsc (mA/cm²)	efficiency
0.005 µm (the default)	247	22.97	19.9 %
1e-5 µm	424	21.40	18.5 %
1e-6 µm	493	21.40	18.5 %

The default overstates the short-circuit current by 7.4 %. Refining a further decade to 1e-6 changes it by 0.03 %, which is how we know 1e-5 is enough — and it costs about three minutes more to run. Voc and fill factor are barely affected; it is the current that moves. The device is graded either side of the ib layer: a heterojunction at ib | n and the IB’s own extent boundary at p | ib, and both need to be resolved.

2.1 Overlay regions

AluminumGalliumArsenideAlloy needs to be told its composition (the aluminum fraction), and that is a good excuse to meet the third way of specifying a parameter.

Every parameter so far has been set either device-wide (on domain) or on a single layer. An overlay region sits between the two: it spans a chosen set of layers, and properties set on an overlay region take priority over both domain and layers.

This device has a natural one — the two AlGaAs layers. In the Layers panel, under Overlay regions, click + Add overlay, name it algaas, and set its extent to the layers p and ib. Then click its coloured dot on the left and add three properties in the Properties pane:

Property	Value
MoleFractionX	0.3
SRH/CB/tau	3e-7 s
SRH/VB/tau	3e-7 s

All three properties now apply across the p and ib layers together and nowhere else. The composition is a property of the alloy, so it belongs exactly where the alloy is; the two lifetimes override what the AlGaAs material supplies, while the GaAs layer keeps its own material’s values.

Overlays are the tidy way to say this property belongs to this part of the device, instead of repeating it layer by layer.

The full priority order is:

named overlay region  >  layer  >  domain  >  material

domain is an overlay region too, but it is the one that means everywhere, so it sits below layers in priority: it states the default, and anything more specific overrides it.

2.2 The intermediate band

In the Bands panel, add a third band:

Band	Type	Carriers	Extent
IB	Sharp Intermediate Band	Electrons	the ib layer only

Choosing anything other than Domain (all layers) for the extent opens a region picker; add the ib layer to it. The panel then shows the boundary conditions at the band’s edges, which default to zero_current — carriers cannot leak out of the end of a band that does not exist beyond it.

The extent matters. We are going to consider a case where the IB physically exists only in the middle layer, and setting the extent correctly enforces that the IB ends.

In the Layers panel, on the ib layer, set the IB’s own parameters:

Parameter	Value	Unit to type
IB/energy_level	0.611	eV
IB/number_of_states	2.5e17	cm^-3
IB/mobility0	500	cm^2/V/s

Type that last unit exactly as shown — cm^2/V/s, with carets, not superscripts. The GUI suggests a unit for the parameters it recognises, but IB/mobility0 is one you type yourself.

IB/mobility0 is the mobility of the empty band. The mobility Simudo actually uses falls as the band fills, because a carrier needs an empty state to move into; at half filling it is about a third of IB/mobility0.

You already set the ib layer’s doping to n-type 1.25e17 cm⁻³. That is half the IB’s density of states, which leaves the band half full — the condition where it can both give up and accept carriers.

With these numbers the energy levels in the ib layer are:

	Energy above the valence band
CB	1.826 eV
IB	0.611 eV
VB	0

2.3 The heterojunction

To configure the interface, in the Layers panel, click the narrow blue bar between ib and n in the schematic at the top of the Layer Stack panel. Add a ThermionicHeterojunction on both CB and VB, and leave the enhancement factor at its default.

The enhancement factor scales how strongly the interface condition is imposed on the solver. Leave it at 1. On this device raising it changes nothing measurable — 1e3 and 1e6 give results identical to 1.

2.4 Processes

This device needs six processes:

Name	Class	Lower E	Higher E	Trap
SRH	SRHRecombination	VB	CB	—
nr_ci	NonRadiativeTrap	IB	CB	IB
nr_iv	NonRadiativeTrap	VB	IB	IB
opt_cv	BeerLambert	VB	CB	—
opt_ci	IBBeerLambert	IB	CB	IB
opt_iv	IBBeerLambert	VB	IB	IB

The two NonRadiativeTrap processes are thermal capture and emission between the IB and each ordinary band. After adding these processes, a yellow dot in the left sidebar tells you to go back to the Layers panel to set parameters. Go to the ib layer and give nr_ci and nr_iv cross sections sigma_th of 5e-18 cm². The required thermal velocity for those processes comes from the band, via the material.

The three optical processes need associated optical absorption information. The GaAs material class provides opt_cv/alpha_function, but the alloy class does not give alpha(E) for AlGaAs. We will add top-hat alpha(E) in the algaas region and top-hat absorption cross sections sigma(E) for the opt_ci and opt_iv transitions that include the intermediate band. Energy conservation is not enforced automatically, but to be physical a transition cannot absorb a photon with lower energy than its own energy separation. Go into the Processes panel and press “+ add top-hat region” for each of those optical processes. Use these thresholds:

Process	Transition	Threshold	alpha/sigma
opt_iv	IB ← VB	0.611 eV	sigma 3e-14 cm^2
opt_ci	CB ← IB	1.215 eV	sigma 3e-14 cm^2
opt_cv	CB ← VB	1.826 eV	alpha 1e4 cm^-1

Add the opt_ci and opt_iv only to the ib region, but add the opt_cv top-hat to both p and ib regions; currently those have to be done separately, but we should soon support adding them to the algaas overlay region directly. Note that 0.611 + 1.215 = 1.826: the two sub-gap steps add up to the host gap, which is required for a sharp (only one energy) IB.

Scroll down the Processes panel and the three optical processes should look like this:

Leave the E< boxes unticked, which means the top hat has no upper cutoff.

Set Include radiative recomb. to No on all three optical processes. §2.5 explains why.

Note: The cross-sections used here — 3e-14 cm² for both IB transitions — are rather large. Half filling puts 1.25e17 cm⁻³ into each transition, so each one contributes , or 0.375 µm⁻¹.

How much light is absorbed therefore depends on the photon energy, because the two transitions overlap rather than dividing the spectrum between them:

Photon energy	Absorbing transitions	Sub-gap α	Absorbed in 2 µm
0.611 – 1.215 eV	opt_iv only	3750 cm⁻¹	53 %
1.215 – 1.826 eV	opt_iv and opt_ci	7500 cm⁻¹	78 %

Above 1.826 eV the band-to-band process opt_cv joins in as well. Raising IB/number_of_states, the optical cross-section, or the ib layer thickness all increase the absorption.

2.5 Radiative recombination, and why it is off here

We have just turned radiative recombination off for all three optical processes, which is not the usual recommendation.

Spontaneous emission is always physically present, and Simudo models it with a Shockley–van Roosbroeck term: every process that absorbs also emits. Simudo does not model photon recycling — a photon emitted inside the device is gone, rather than potentially being reabsorbed somewhere else. So radiative recombination here is a pure loss, and the simulated device is pessimistic by however much recycling a real one would achieve.

For most devices that overestimate is small and you should leave the radiative emission in. For this one it is not small, due to the large optical cross sections. With purely lossy radiative emission on, this cell manages about 2% efficiency; with it off, about 18%. Much of the difference is an artifact of the missing recycling, so the version worth looking at is the one with emission off.

If you did want it on, the extra step is one parameter: the Shockley–van Roosbroeck integral needs a refractive index. Set refractive_index on the domain region in the Layers panel — 3.5 is reasonable for these alloys — unless your material files already supply it, in which case it is picked up automatically. Without it the run stops before solving, with a message naming the process that wanted it.

2.6 Light, and the run

In the Opt. Fields panel, add an AM1.5g field set again, this time from 0.611 eV — the lowest energy anything in this device can absorb — up to 4.0 eV, in 10 bins. Then, in the Simulation panel, set the mode to Illuminated J-V and sweep 0 to 1.0 V — this cell’s open-circuit voltage is near 0.98 V.

Turn on self-consistent optics, for the intensity ramp and the voltage sweep both — the checkbox beside each stage in the Simulation panel. Optical processes through an intermediate band need it. Both IBBeerLambert processes take their absorption from the IB’s occupancy — opt_iv needs empty states in the IB to absorb into, opt_ci needs filled ones to absorb from — so absorption depends on the electrical solution and the electrical solution depends on absorption. With the box ticked the optical problem is re-solved after every Newton iteration instead of once per bias point. Device 1’s BeerLambert had a fixed α(E) from the material, with nothing to be self-consistent about.

Note: On this device it makes almost no difference, and costs almost nothing. In this device, IB stays near half filling here, so absorption does not change significantly during Newton steps. Self-consistency is more important when the occupancy really does move: concentrated light, a low-density IB, or a cell driven hard enough to bleach the sub-gap transitions.

Two step sizes are worth setting, in the same panel. They are the boxes labelled Step size under Intensity Ramp and Voltage Sweep, and they control the solver’s first step, not the spacing of your output.

	default	use	why
Intensity ramp	1e-25	1e-20	the ramp climbs from darkness to one sun; the first 5 decades are wasted
Voltage sweep	1e-4	0.1	at V = 0 the device is already solved, so the first step can be a whole point

Both were measured on this device, not guessed. For the ramp, 1e-20 is the largest first step that succeeds. For the voltage sweep, the solver takes a 0.3 V first step happily; 0.1 is simply the largest value your voltage list can use, since we’re asking for output at V = 0.1 V. Together these step sizes cut about four minutes off this run. Leave them blank and the defaults still work — you just wait longer.

The Simulation panel should look like this:

This run takes about fourteen minutes on an M2 Mac Mini: three optical processes over ten spectral bins on a 424-point mesh. It is a good moment to note that the execution profile exists precisely so Simudo can run on a bigger machine than the one serving the GUI, and that the checkpoint you asked for at the end of the intensity ramp means a second sweep — a different voltage list, say — costs seconds rather than repeating the ramp.

What you should get

Jsc	21.4 mA/cm²
Voc	0.99 V
Fill factor	0.87
Efficiency	18.5 %

Read Voc off the plot with some caution. The curve is exponential where it crosses zero, so interpolating linearly between the two bracketing points underestimates it by 10–20 mV. Add points near the crossing if you need more precision.

Things worth trying

The device is built now, so it is cheap to ask it questions. A few that change the answer in instructive ways:

• Thin the ib layer. Sub-gap absorption falls with thickness — 53 % of the 0.611–1.215 eV window in 2 µm, about 14 % in 0.4 µm — but so does the volume in which carriers are captured back into the IB. Which effect wins is not obvious in advance.

• Turn radiative recombination back on, remembering refractive_index from §2.5, and watch the efficiency collapse. That is the photon-recycling question of §2.5 made concrete.

• Change nr_ci/sigma_th and nr_iv/sigma_th. These set how strongly the IB is thermally coupled to the ordinary bands.

---

Reference

Where the output goes

Each run writes to the folder named in the Output panel. If that folder already exists, the runner increments the last character rather than overwriting: out/2026Aug17/a becomes out/2026Aug17/b. This system is designed to work well if you name output folders like <project>/out/<date>/a.

Inside you will find sim_V.csv (the J–V curve), a spatial profile per bias point, the log, a copy of the project (a yaml file that can be reloaded back into the GUI), and any checkpoints.

The panels

Panel	What it holds
Layers	The layer stack, thicknesses, materials, per-layer and per-region parameters, and the interface editor
Materials	The material library, and derived materials that override library values
Bands	Band list: type, carrier sign, extent
Processes	Recombination and optical processes, and their parameters
Opt. Fields	Illumination: spectra, energy ranges, bin counts, concentration
Boundary Cond.	Contact roles and per-band contact conditions
Simulation	Intensity ramp, voltage sweep, checkpoints, mesh, output, and where the run executes
Output	J–V curves, band diagrams, and spatial plots from a finished run

Process classes

Class	Use it for
SRHRecombination	Shockley–Read–Hall through midgap traps, no explicit trap band
NonRadiativeTrap	Thermal capture/emission between an ordinary band and a sharp IB
ShockleyReadBand2BandTrap	Shockley–Read-like transfer between two dispersing bands
ShockleyReadTrap2Trap	Transfer between two sharp trap bands
BeerLambert	Absorption between two ordinary bands, arbitrary α(E)
IBBeerLambert	Absorption involving a sharp IB, where α follows the band’s filling
NonOverlappingTopHatBeerLambert	As BeerLambert, assuming constant α(E) and that only one process absorbs each photon energy
NonOverlappingTopHatBeerLambertIB	The IB version of the same
StaticGeneration	Not usable from the GUI — see its hint in the Processes panel

Selecting a class in the Processes panel shows a short usage note beneath the selector.