A band gap is the range of energies in a solid where no electron states exist — the separation between the top of the highest filled band and the bottom of the lowest empty band. Its size determines whether a material behaves as a conductor, semiconductor, or insulator, and it is the starting point for designing transistors, solar cells, LEDs, and virtually every modern electronic device.
From Atomic Orbitals to Energy Bands
In an isolated atom, electrons occupy discrete, well-defined energy levels. When atoms are brought together to form a crystal, their orbitals overlap and interact. Because of the Pauli exclusion principle, the sharp atomic levels split into a huge number of closely spaced levels — one per atom in the solid. With \(\sim 10^{23}\) atoms in a macroscopic crystal, these levels are so dense they form effectively continuous energy bands.
Between these bands lie ranges of energy that no electron can occupy — the band gaps. Allowed bands are filled from the bottom up according to how many electrons the material has. The two bands that matter most:
- The valence band: the highest band that is completely (or nearly completely) filled at absolute zero.
- The conduction band: the next band up, normally empty at absolute zero.
In a crystal, energies are labeled by crystal momentum \(\mathbf{k}\). Plotting \(E(\mathbf{k})\) along high-symmetry lines yields the band structure from which the gap, its direct or indirect character, and the effective masses all follow.
Defining the Band Gap Energy
The band gap energy, usually written as \(E_g\), is the difference between the bottom of the conduction band and the top of the valence band:
$$ E_g = E_c - E_v $$where \(E_c\) is the conduction band minimum (CBM) and \(E_v\) is the valence band maximum (VBM). Band gaps are quoted in electronvolts (eV). Silicon, the workhorse of the semiconductor industry, has a band gap of about 1.12 eV at room temperature.
Only electrons in the conduction band (and the empty states, or holes, they leave in the valence band) carry electrical current. If the gap is large, thermal energy at room temperature (\(k_B T \approx 0.026\,\mathrm{eV}\)) is nowhere near enough to promote electrons. The intrinsic carrier concentration scales as:
$$ n_i \propto T^{3/2} \exp\left(-\frac{E_g}{2 k_B T}\right) $$so even a few tenths of an eV change in \(E_g\) shifts conductivity by orders of magnitude.
Direct vs Indirect Band Gaps
Band structure is more than a single number — it depends on the electron’s crystal momentum \(\mathbf{k}\). When we plot energy versus \(\mathbf{k}\), the valence band maximum and conduction band minimum may or may not occur at the same value of \(\mathbf{k}\).
- Direct band gap: the VBM and CBM sit at the same \(\mathbf{k}\)-point. An electron can jump across the gap by absorbing or emitting a photon alone. Photons carry negligible crystal momentum, so vertical transitions in the Brillouin zone are allowed and strong.
- Indirect band gap: the VBM and CBM sit at different \(\mathbf{k}\)-points. A transition across the gap must also involve a phonon (a lattice vibration) to conserve momentum. That second-order process is much less probable, so absorption near the edge is weak and radiative recombination is inefficient.
Silicon vs gallium arsenide
Silicon is the textbook indirect-gap semiconductor. Its VBM is at \(\Gamma\); the CBM lies near \(X\) along \(\Delta\). Room-temperature \(E_g \approx 1.12\,\mathrm{eV}\). Phonon-assisted optical transitions make bulk Si a poor light emitter — silicon photonics still relies on hybrid integration or nanostructures. Yet Si dominates digital electronics: mature CMOS, a high-quality native oxide, and adequate mobility outweigh the optical handicap.
Gallium arsenide is the textbook direct-gap III–V. Both VBM and CBM sit at \(\Gamma\), with \(E_g \approx 1.42\,\mathrm{eV}\) at 300 K. Electrons and holes recombine radiatively with high efficiency, so GaAs and related alloys power LEDs, laser diodes, and high-speed optoelectronics. Higher electron mobility than Si helped early RF devices, but cost keeps GaAs out of mass digital logic.
| Material | Band gap (eV, ~300 K) | Type | Common use |
|---|---|---|---|
| Silicon (Si) | 1.12 | Indirect | CPUs, solar cells |
| Germanium (Ge) | 0.66 | Indirect | Photodetectors, SiGe |
| Gallium arsenide (GaAs) | 1.42 | Direct | LEDs, laser diodes |
| Gallium nitride (GaN) | 3.4 | Direct | Blue LEDs, power electronics |
| Silicon carbide (4H-SiC) | 3.26 | Indirect | Power devices |
| Diamond (C) | 5.5 | Indirect | Insulator, UV optics |
Indirect does not mean useless for absorption: Si solar cells work because thickness and light trapping manage the absorption length. It does mean spontaneous emission is weak — bulk Si is not a laser gain medium.
Optical vs Electronic Band Gaps
Experiment and theory often report different “gaps,” and conflating them is a common source of confusion.
- The electronic (fundamental) gap is the difference between ionization energy and electron affinity — the true quasiparticle gap between adding and removing an electron. Many-body methods such as GW aim at this quantity.
- The optical gap is the onset of strong light absorption. In many semiconductors it is smaller than the electronic gap by the exciton binding energy \(E_b\), because the photoexcited electron and hole form a bound exciton:
In bulk GaAs, \(E_b\) is only a few meV, so optical and electronic gaps nearly coincide. In 2D materials, oxides, and organics, \(E_b\) can be hundreds of meV. For indirect materials, the optical edge can also sit slightly above the electronic gap because phonon-assisted processes need extra energy. A Tauc plot of \((\alpha h\nu)^n\) versus photon energy extracts an optical gap, with \(n = 2\) for allowed direct and \(n = 1/2\) for allowed indirect transitions — a practical rule of thumb, not a theorem.
When comparing DFT to experiment, ask: electronic or optical? Zero or finite \(T\)? Which functional or many-body method?
Classifying Conductors, Semiconductors, and Insulators
The band gap gives a clean way to classify materials by their electrical behavior. The key question is how the bands are filled and how large the gap is.
flowchart TD
A[Electronic structure] --> B{Partially filled band<br/>or band overlap?}
B -->|Yes| C[Conductor / metal<br/>no gap]
B -->|No| D{Band gap E_g}
D -->|0.1–3 eV| E[Semiconductor]
D -->|greater than ~4 eV| F[Insulator]
E --> G{VBM and CBM<br/>same k?}
G -->|Yes| H[Direct gap<br/>LEDs, lasers]
G -->|No| I[Indirect gap<br/>Si, Ge]| Class | Band gap | Band filling | Conductivity |
|---|---|---|---|
| Conductor (metal) | No gap | Partially filled band | Very high |
| Semiconductor | ~0.1–3 eV | Filled valence, empty conduction | Moderate, tunable |
| Insulator | > ~4 eV | Filled valence, empty conduction | Negligible |
In a conductor, the highest occupied band is only partially filled, or valence and conduction bands overlap — electrons move into empty states with almost no energy cost. In a semiconductor, the gap is small enough that thermal energy or deliberate doping promotes a useful number of carriers; that tunability is the foundation of electronics. In an insulator, essentially no electrons are thermally excited across the gap at ordinary temperatures. The distinction between a large-gap semiconductor and an insulator is one of degree: diamond (\(E_g \approx 5.5\,\mathrm{eV}\)) is an electronic insulator but a wide-gap semiconductor in the materials sense.
Why the Band Gap Matters in Real Devices
The band gap is not an abstract quantity — it directly sets the performance limits of technologies:
- Transistors and logic: the gap enables switching between conducting and insulating states. Off-state leakage scales with a Boltzmann factor involving \(E_g\). Si’s 1.12 eV is large enough for room-temperature digital logic; much smaller gaps force cryogenic operation or soft off-states.
- Solar cells: free carriers require photons above \(E_g\). The Shockley–Queisser limit for a single junction under AM1.5 peaks near 1.3–1.4 eV, which is why Si and GaAs remain strong choices. Too small a gap wastes energy as heat; too large wastes the infrared.
- LEDs and lasers: for direct-gap materials, photon energy tracks \(E_g\), so the gap sets color. GaN/InGaN enable blue and white lighting; AlGaAs and InGaAsP cover red through near-IR telecom.
- Power electronics: wide-gap GaN and 4H-SiC support higher breakdown fields (empirically \(\sim E_g^{2}\)–\(E_g^{2.5}\)), higher temperatures, and lower losses in high-voltage converters.
Gap and free-space wavelength:
$$ E_g = \frac{h c}{\lambda} $$so 1.42 eV is roughly 870 nm (near-IR) and 3.4 eV is about 365 nm (near-UV). Designers often pick \(\lambda\), then alloy until \(E_g\) matches.
Temperature and Composition Dependence
The band gap typically shrinks as temperature rises — thermal expansion and electron–phonon coupling shift the band edges. The empirical Varshni form is widely used:
$$ E_g(T) = E_g(0) - \frac{\alpha T^2}{T + \beta} $$where \(\alpha\) and \(\beta\) are material constants. Silicon drops from about 1.17 eV near 0 K to 1.12 eV at 300 K; GaAs from roughly 1.52 eV at low \(T\) to 1.42 eV at room temperature. Always quote temperature with optical or transport gaps.
The gap can also be engineered by alloying. In \(\mathrm{Al}_x\mathrm{Ga}_{1-x}\mathrm{As}\), increasing Al widens \(E_g\) and can switch the lowest gap from direct (\(\Gamma\)) to indirect (\(X\)) near \(x \approx 0.4\) — classic band-gap engineering for quantum wells, lasers, and HEMTs. Similar tuning exists in \(\mathrm{In}_x\mathrm{Ga}_{1-x}\mathrm{N}\) and \(\mathrm{Si}_{1-x}\mathrm{Ge}_x\), subject to miscibility gaps and strain. Strain and quantum confinement further shift effective gaps in wells, nanowires, and 2D materials, so bulk \(E_g\) is only the starting point for nanoscale devices.
Computing Band Gaps with DFT
Computationally, band gaps come from density functional theory (DFT) with plane-wave codes such as Quantum ESPRESSO. DFT yields \(E(\mathbf{k})\) on a Brillouin-zone path; VBM, CBM, gap size, and direct/indirect character follow from the eigenvalues.
Standard DFT with local or semi-local functionals (LDA, PBE) systematically underestimates band gaps, often by 30–50% or more. Silicon’s experimental 1.12 eV may come out near 0.5–0.7 eV with PBE; GaAs is similarly too small. This band-gap problem stems from the missing derivative discontinuity of approximate exchange–correlation functionals and self-interaction error — Kohn–Sham eigenvalues are not formal quasiparticle energies.
Common remedies, ordered roughly by cost:
| Method | What it does | Typical use |
|---|---|---|
| DFT+U | Hubbard \(U\) on localized \(d\)/\(f\) states | Transition-metal oxides |
| HSE06 (hybrid) | Screened exact exchange fraction | Solid-state gaps, moderate cost |
| GW | Many-body quasiparticle correction | Benchmark electronic gaps |
| BSE (on GW) | Bethe–Salpeter for excitons | Optical spectra, \(E_b\) |
HSE06 often recovers Si and GaAs within a few tenths of an eV of experiment at far lower cost than full GW. G0W0 (or partially self-consistent GW) remains the gold standard for quantitative electronic gaps. None replace careful convergence of basis, \(k\)-mesh, and pseudopotentials — a well-converged PBE gap beats a sloppy hybrid run. Always check which method produced a reported gap; see our computational engines overview and blog for workflow context.
Extracting the Gap from Quantum ESPRESSO Bands
A standard Quantum ESPRESSO workflow for \(E_g\):
- Structural relaxation (
pw.x,calculation='vc-relax'or'relax') — gaps are sensitive to lattice constant. - Self-consistent field (
calculation='scf') with a dense enough \(k\)-mesh for the charge density. - Non-self-consistent bands (
calculation='bands') along a high-symmetry path (e.g. \(\Gamma\)–\(X\)–\(W\)–\(L\)–\(\Gamma\)–\(K\) for FCC) viaK_POINTS crystal_b. - Post-process with
bands.x→bands.dat, then identify VBM and CBM from the eigenvalues.
Align the VBM to zero for gap size; confirm direct vs indirect by whether the highest occupied and lowest unoccupied eigenvalues share \(\mathbf{k}\) (Si: CBM not at \(\Gamma\); GaAs: both at \(\Gamma\)). Converge plane-wave cutoff, \(k\)-mesh, and smearing (or fixed occupations for insulators) until \(E_g\) is stable to ~0.01–0.05 eV for trends. Hybrids (HSE) cost more SCF time — many workflows use a PBE density, then a hybrid or GW correction. Checklist for any reported QE gap: functional, same-functional relaxation, path density, and spin–orbit (important for heavy III–Vs).
Run it on Simatra
Simatra computes band gaps, band structures, and full electronic densities of states using DFT on GPU-accelerated clusters built for materials modeling. Running on our GPU-Opt-V2 instances, you get up to 5x faster convergence and can tackle supercells of up to ~2,000 atoms — enough for alloys, defects, and realistic heterostructures. Choose between the open-source Quantum ESPRESSO engine and Simatra’s native KRONOS engine (a C++20, GPL-3.0 DFT code with CUDA/HIP/Metal backends); see our computational engines page for details. Start a free trial with $100 in credits at app.simatra.io and compute your first band gap today.