PX432 Functional Properties of Solids Part I: Electrical properties Term 2, 2015 Lecturer: Dr James Lloyd-Hughes, Room: MAS3.06, Email: J.Lloyd-Hughes@warwick.ac.uk 2 Contents 1 Course description 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Textbooks/resources . . . . . . . . . . . . . . . . . . . . . 1.5 Handout/notes . . . . . . . . . . . . . . . . . . . . . . . . 1.6 Commitment . . . . . . . . . . . . . . . . . . . . . . . . . 1.7 Assessment for PX432 . . . . . . . . . . . . . . . . . . . . 1.8 Outline Syllabus . . . . . . . . . . . . . . . . . . . . . . . 1.8.1 Bandstructure: theory and experiment (3 lectures) 1.8.2 Transport in intrinsic and extrinsic semiconductors 1.8.3 Semiconductor optics (1 lecture) . . . . . . . . . . 1.8.4 Semiconductor devices (1 lecture) . . . . . . . . . . 1.8.5 Terahertz optoelectronics (3 lectures) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 5 5 5 5 6 6 6 6 6 7 7 7 7 2 Electrons in crystals 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Free electron model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Sommerfeld-Drude theory of metals . . . . . . . . . . . . . . . . . . . . . 2.2.2 Density of states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.3 Heat capacity of the Fermi gas . . . . . . . . . . . . . . . . . . . . . . . . 2.2.4 Transport properties: dc electrical conductivity . . . . . . . . . . . . . . . 2.2.5 Wiedemann Franz ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.6 Hall effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.7 Response to electromagnetic waves: ac conductivity . . . . . . . . . . . . 2.2.8 Experimental verification / success and failure of the free electron model . 2.3 The nearly free electron model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Electrons in a periodic potential . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Bloch’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3 The 1D empty lattice approximation . . . . . . . . . . . . . . . . . . . . . 2.3.4 Nearly free electron gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.5 Zone boundaries, origin of energy gap, physical interpretation . . . . . . . 2.3.6 2D: band overlaps, density of states and semimetals . . . . . . . . . . . . 2.3.7 Electrical classification of crystalline solids . . . . . . . . . . . . . . . . . . 2.4 Alternative approaches to bandstructure calculation . . . . . . . . . . . . . . . . 2.4.1 Pseudopotential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Kronig-Penney model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 The tight-binding approach . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.4 Density functional theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Carrier dynamics and collisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 9 9 10 12 13 14 16 16 18 19 20 20 20 22 23 24 27 29 30 30 30 31 33 34 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2 lectures) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 34 35 36 38 39 3 Intrinsic and Extrinsic Semiconductors 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Direct and indirect band gaps . . . . . . . . . . . . 3.1.2 Examples . . . . . . . . . . . . . . . . . . . . . . . 3.2 Intrinsic semiconductors . . . . . . . . . . . . . . . . . . . 3.2.1 Electron concentration . . . . . . . . . . . . . . . . 3.2.2 Hole concentration . . . . . . . . . . . . . . . . . . 3.2.3 Mass action law . . . . . . . . . . . . . . . . . . . 3.2.4 Intrinsic carrier concentrations . . . . . . . . . . . 3.2.5 Conductivity . . . . . . . . . . . . . . . . . . . . . 3.3 Extrinsic semiconductors . . . . . . . . . . . . . . . . . . . 3.3.1 Hydrogenic levels: donor ionisation energies . . . . 3.3.2 Extrinsic carrier concentrations . . . . . . . . . . . 3.3.3 Temperature and dopant dependence of n, p, µ & σ 3.3.4 Conductivity [revisited] . . . . . . . . . . . . . . . 3.3.5 The Hall effect [revisited] . . . . . . . . . . . . . . 3.3.6 Carrier dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 41 41 43 44 44 45 46 46 46 49 50 51 54 54 55 55 4 Semiconductor devices 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 p − n junction . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 The p − n junction under applied bias . . . . . . . 4.2.2 Solar cells . . . . . . . . . . . . . . . . . . . . . . . 4.2.3 LEDs/Lasers . . . . . . . . . . . . . . . . . . . . . 4.2.4 High electron mobility transistor [non-examinable] 4.3 MOSFET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 57 58 62 65 66 66 67 2.6 2.5.1 Conservation of energy and crystal momentum 2.5.2 Scattering: classical versus quantum pictures . 2.5.3 Scattering mechanisms and Mattheissen’s rule . 2.5.4 Effective mass model . . . . . . . . . . . . . . . 2.5.5 Holes . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Terahertz optoelectronics 69 5.1 Terahertz time-domain spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 5.1.1 Pulsed terahertz generation from biased semiconductors . . . . . . . . . . . . . . . . . 70 4 Chapter 1 Course description 1.1 Introduction This is one third of the PX432 ‘Functional Properties of Solids’ module, it introduces the fundamental ideas of how the electrical properties of materials are determined. The course will draw on concepts introduced in PX262 ‘Quantum mechanics and its applications’, PX394 ‘Electrons in solids’, and PX393 ‘Crystal physics’. Functional devices based on semiconductors will be described. Advanced experimental techniques that investigate the electrical properties of solids will be outlined, such as angle-resolved photoemission spectroscopy and terahertz time-domain spectroscopy. The final lectures will discuss optoelectronic devices and spectroscopy using terahertz (far-infrared) radiation. 1.2 Aims • to provide an overview of theoretical ideas that underpin electronic band structure and conduction in solids. • to introduce the fundamental properties of intrinsic and extrinsic semiconductorss. • to provide a basis for the understanding of modern solid-state electronics. • to highlight advanced experimental techniques, materials and devices at the forefront of contemporary research into electronic materials. 1.3 Objectives At the end of the module you should • understand the underlying band theory of how materials are classified as metals, insulators, semiconductors and semi-metals. • understand and appreciate experimental methods that investigate the electronic properties of solids. • be able to apply basic ideas of quantum mechanics and statistical mechanics to situations in condensed matter (metals & semiconductors). • calculate carrier densities in doped and undoped semiconductors under a variety of conditions. 1.4 Textbooks/resources Handouts are available on the course website at go.warwick.ac.uk/PX432. Recommended text books 5 • Hook and Hall - The recommended textbook you are used to from PX394 ‘Electrons in solids’, which I will try to refer to where possible. • J. Singleton, ‘Band Theory and Electronic Properties of Solids’ - A modern and excellent discussion of bandstructure and electron transport. • M. Fox, ‘Optical Properties of Solids’ - Accessible descriptions of the optics of semiconductors and heterostructures Additional resources: • H.P. Myers ‘Introductory Solid State Physics’. • S.M. Sze ‘Physics of semiconductor devices’ - Comprehensive, giving plenty of detail on semiconductor devices beyond that needed for this course. • L. Solymar and D. Walsh, ‘Electrical properties of materials’ - Written for engineers, covers the basics in the context of modern devices • www.NanoHub.org is an excellent website containing a lot of tutorials and applets related to semiconductor physics. 1.5 Handout/notes This handout is based on the handout provided by the course’s previous lecturer (Dr. N. Wilson). It contains, for ease of reference, some material that you will be familiar with from previous courses (e.g. PX394 ‘Electrons in solids’), and additional background material. The concepts listed in the outline syllabus (below) and the material discussed in lectures form the basis of the examinable material for this course. 1.6 Commitment 10 lectures, 2 examples sheets. 1.7 Assessment for PX432 2 hour examination, in which you will be required to answer 3 questions. There will be one compulsory question in Section A covering the whole course (from all three lecturers). You will then have a choice of two from three questions in Section B. 50 % of the marks will be for Section A and 50 % for Section B. This exam structure was adopted in 2014. 1.8 1.8.1 Outline Syllabus Bandstructure: theory and experiment (3 lectures) • Electrons in a periodic potential - Tight-binding approach (linear combination of atomic orbitals) - Nearly-free electron model • Indirect and direct measurements of bandstucture • Effective mass m* - arbitrary band dispersion - results for parabolic and non-parabolic bands (including linear bands, e.g. graphene) - measuring m*: cyclotron resonance • Angle-resolved photoemission spectroscopy 6 1.8.2 Transport in intrinsic and extrinsic semiconductors (2 lectures) • Intrinsic carriers; mass action law • Fermi energy and chemical potential • Extrinsic semiconductors: - Donor and acceptor ionisation - effective Hamiltonian - Extrinsic carrier statistics (donor ionisation regime, exhaustion regime, intrinsic regime) • Temperature and dopant concentration dependence of chemical potential • Transport in semiconductors: - Drude model for conductivity; frequency dependence - Current density for energy bands - Conductivity and mobility - Conductivity of extrinsic semiconductors 1.8.3 Semiconductor optics (1 lecture) • Inorganic semiconductors: - Direct gap/indirect gap semiconductors - Conservation of crystal momentum and energy - Matrix elements - Absorption coefficient - Photoluminescence • Excitons 1.8.4 Semiconductor devices (1 lecture) • Conductive devices: - p-n junction - Transistors • Optoelectronic devices: - LEDs, lasers - Photovoltaics, photodiodes 1.8.5 Terahertz optoelectronics (3 lectures) • THz time-domain spectroscopy - Pulsed terahertz generation from biased semiconductors - Ultrafast spectroscopy: pump-probe techniques - Time-resolved THz conductivity of bulk materials • Quasiparticles in condensed matter - Electrons in nanoscale semiconductors - Plasmons, excitons, polarons • Intersubband devices - Intersubband transitions - Quantum cascade lasers 7 8 Chapter 2 Electrons in crystals 2.1 Introduction In this course we will be looking at functional electrical properties of crystalline solids. We will start by considering conduction in metals which to a ‘zeroth order’ approximation can be treated as a ‘free electron gas’ in a box. The ‘first order’ approximation introduces the effect of the background lattice as a periodic potential which acts to perturb the free electron gas. The insight this model brings will then be applied to understand semiconductors - the bulk of this course. In the final lectures we will introduce terahertz (farinfrared) optoelectronics. The justification for this course is simple; the impact of semiconductor technology on life in the developed world is unrivalled. You should understand the basic science behind the technology. Figure 2.1: Photo copyright Alcatel-Lucent. 2.2 Free electron model The theoretical approach to understanding conduction in metals was first introduced by Paul Drude around 1900, only three years after the discovery of the electron by J.J Thomson. Drude applied the kinetic theory of gases to try to understand electrical and thermal transport properties of metals. For our purposes the key assumptions are: 1. Free and independent electron approximation: between collisions the electrons do not interact with either the ions or the other electrons, but instead move according to Newton’s laws subject to any external fields that may apply. The independent electron approximation (electrons do not interact 9 with one another) is bewilderingly good in many situations, however as we shall see the free electron approximation (electrons do not interact with the ions) is not so useful. 2. Scattering: Collisions are the only form of interaction of the electron with the rest of the material. They are instantaneous, and afterwards the electron moves at a velocity not related to its original one, but instead dependent on the temperature of the material at that point. The probability of scattering is τ −1 per unit time. Drude attributed the scattering events to be due to collisions with the ion cores, we shall see that this is not the case. However the mechanism of scattering does not effect the predictions. This is an entirely classical picture. The resultant velocity distributions of the particles are given by the Maxwell Boltzmann distribution. Unfortunately the predictions from it are wrong, in particular the heat capacity predicted by this model should have a contribution of 32 kB per electron which is not observed. Correcting this required the application of quantum mechanics, a further 25 year wait. 2.2.1 Sommerfeld-Drude theory of metals In this semi-classical approach first the Pauli exclusion principle is applied to determine the velocity distribution of electrons in the metal, and then this velocity distribution is applied classically in the manner developed by Drude. We will derive some of the basic results here. We can write the Hamiltonian for free electrons as HΨ Ĥ = Eψ ~2 2 p2 =− ∇ = 2m 2m (2.1) The solution to this is plane wave eigenstates, i.e. eigenstates of the momentum operator. Ψk = ceik̄.r̄ p~Ψk = ⇒ Ek = −i~∇Ψk = ~kΨk ~2 k 2 2m (2.2) For a finite crystal we impose periodic boundary conditions so that ~ Ψ(~r + L) ~ i~ k.L ⇒e ⇒ ~k = Ψ(~r) = 1 2π (nx , ny , nz ) L = (2.3) i.e. there is one state per (2π/L)3 of ~k-space. We also need to normalise the wavefunction Z Ψ∗k Ψk d3 r = 1 = c2 V 1 ⇒c = √ V 1 i~k.~r ⇒ Ψk = √ e V (2.4) Fermi statistics must be obeyed, i.e. the Pauli exclusion principle excludes two electrons from occupying the same state. Electrons are spin 12 so the wavefunctions are anti-symmetrised and each ~k corresponds to 2 states. The states are filled in order of increasing energy so that at T = 0K all states are filled up to the ’Fermi Energy’ 10 Figure 2.2: Sketches of dispersion relation with filled and empty states, and 2D Fermi surface (circle) - a sphere in 3D. What are EF and kF ? We can calculate the volume of k space that lies in |k| < kF : N = X 2 1 |k|<kF Z 2V d3 k (2π)3 |k|<kF 2V 4π 3 = k (2π)3 3 f 2 31 3π N = V = ⇒ kF (2.5) Here we have imposed the periodic boundary conditions, which quantised the states in k space. These are then filled up to a total number of N states. We can estimate the Fermi wavevector kF kF−1 ∼ V N 13 ∼a Where a is the typical distance between electrons. Hence in a metal kF−1 ∼ few Å. Having derived an equation for kF we can use this to find the Fermi energy, EF . EF ~2 kF2 2me ∼ 10eV = (2.6) ∼ 105 K The energy scale of the Fermi energy is far higher than room temperature or any real thermal excitations, i.e. the Pauli exclusion principle is the dominant effect governing behaviour in metals. The low temperature 11 approximation of T = 0K will be quite valid in many cases. We can also calculate the Fermi velocity vF = = ~kF m ' 108 cms−1 (2.7) Thus far we have assumed that all states up to the Fermi energy are filled, and any state above the Fermi energy is empty. However at finite temperature some of the electrons will be excited into higher states. Roughly we expect that states within kB T either side of EF will have a non-zero probability of being occupied. Let’s use nk as the probability that the state k is occupied, this is equal to the Fermi distribution function as defined by 1 nk = f (Ek ) = (E −µ)/k T (2.8) B k e +1 where µ is the (electronic) chemical potential. The chemical potential at T = 0K is the Fermi energy, µ = EF . For finite temperature µ is determined by N =2 X k 1 e(Ek −µ)/kB T + 1 (2.9) The Fermi function is sketched at zero temperature and finite temperature in Fig. 2.3 (dashed lines). Figure 2.3: Sketch of Fermi distribution function and EF . 2.2.2 Density of states Critical to many physical properties is the density of states, N (E)dE: the number of states with energy E < Ek < E + dE. The density of states is V 2 4πk 2 dk (2π)3 We can rewrite dk dE = dk = dE r 2m 1 √ dE ~2 2 E Putting this together we have V 2mE N (E) = 2 2 π ~ 12 r 2m 1 √ ~2 2 E (2.10) This can be simplified to give the important result N (E) = V 2π 2 2m ~2 3/2 √ E (2.11) The density of states for a parabolic band is sketched in Fig. 2.3, where it is labelled gc (E) 2.2.3 Heat capacity of the Fermi gas The internal energy can be calculated from the energy of the states and the probability that they are occupied X U =2 Ek nk (2.12) k Switching to a continuous representation Z U= N (E) E f (E) dE (2.13) However what we are really interested in here is the density of states at EF : N (EF ) = = = dN dEF dN kF dk F F kF dE dkF 3N 2EF Where we have used the following relations N ∝ kF 3 EF ∝ kF 2 ⇒N ∝ EF 2 3 From this density of states we can make a rough estimate of the heat capacity by considering only the number of ’active’ electrons within kB T of the Fermi energy. Each electron that is thermally excited is gaining ∼ kB T , and the number that are doing so is ∼ kB T N (EF ). So the increase in internal energy due to temperature can be approximated by U (T ) − U (T = 0) ' N (EF )(kB T )2 which gives the heat capacity ∼ kB N (EF )kB T kB T ∼ N kB EF C (2.14) A more thorough and careful calculation gives Cv = π2 kB T N kB 2 EF (2.15) Note that this is substantially less than the 32 kB that was predicted classically, and is now temperature dependent. Recalling from the structure course that the heat capacity due to phonons ∝ T 3 we can write the heat capacity of a metal as C = γT + αT 3 (2.16) where γ is a measure of the density of states. 13 Figure 2.4: Heat capacity of Au from Myers p.150 2.2.4 Transport properties: dc electrical conductivity The current density is related to the electric field via the conductivity (tensor): ~ J~ = σ E (2.17) We will now try to understand the magnitude of the conductivity, starting with a classical model applied to the free electron model. Drude model We start by applying Newton’s second law to the motion of an electron F dv ⇒ dt = ma eE = − m We include damping due to collisions through adding a term with the momentum relaxation rate τ dv eE 1 =− − v dt m τ In the simple case of zero electric field the relaxation of the electron speeds is then given by v(t) = v(0)e −t τ In steady state the speed distribution must be constant, dv dt ⇒v = 0 = − eEτ m The electron speed is related to the current density via the charge on the electron, and the electron density, n: J = −nev (2.18) Combining these results gives J= ne2 τ E m 14 (2.19) By comparison with 2.17 we obtain the Drude formula for the conductivity σ= ne2 τ m (2.20) Note that this predicts Ohm’s Law. A quantum perspective: the moving Fermi sphere When subjected to an accelerating electric field in the x-direction all electrons will feel a force in the −xdirection (force F = qEx = −eEx ), their x-momentum will change by an amount ~∆k, and the Fermi sphere will thus shift [see Figure 2.5]. Figure 2.5: Schematic of shift of Fermi sphere from Myers p.155 We can rewrite Newton’s second law dp dt dk = ~ dt F = mv τ ~k = −eE − τ = −eE − In steady state we can equate the movement of the Fermi sphere ~∆k = −eEτ i.e. the average momentum increase is given by the electric force applied multiplied by the time it is applied for (the time τ before a relaxation process occurs). Again we can write the current density as J = = −nev ~k −ne m Combining these results gives once more ne2 τ E (2.21) m The small shift in the Fermi surface illustrates that the current is effectively carried by only a very small fraction of the electrons, and that these are the electrons at the Fermi energy. J= 15 2.2.5 Wiedemann Franz ratio Kinetic theory predicts the thermal conductivity of a gas of electrons to be 1 l C v vF 3 (2.22) π2 kB T N kB 2 EF (2.23) K= Recall Cv from 2.15 Cv = The Drude formula for the electrical conductivity can be written σ= ne2 τ ne2 l = m m vF (2.24) So we can use EF = 12 mvF2 to find the ratio K σ = L = 2 π 2 kB T = LT 3e2 2 π 2 kB = 2.4 × 10−8 Js−1 ΩK −2 2 3e (2.25) This result explains the Wiedemann-Franz law which states that for metals at not too low temperature the ratio of the thermal to electrical conductivities is directly proportional to temperature. The Lorenz number, L, is the constant of proportionality and as calculated in the free electron approximation above only consists of fundamental constants. This is a surprisingly good prediction for most metals at room temperature. It relies on the scattering processes that determine the electrical and thermal conductivities being the same, and on the electrons dominating the thermal conductivity rather than the phonons. 2.2.6 Hall effect Let’s consider a current flowing in the presence of a magnetic field. The dynamics will be governed by the Lorentz force d~v e ~ ~ − 1 ~v =− (2.26) E + ~v × B dt m τ The final term is the damping or momentum relaxation term. We define the field to be in the z-direction, ~ = B ẑ, and solve for the steady state solution. i.e. B 0 0 0 e (Ex + vy B) − m e = − (Ey − vx B) − m e 1 = − E z − vz m τ = − 1 vx τ 1 vy τ (2.27) Suppose that vy = 0, and Jx = −nevx ⇒ Ex = − = m m vx = 2 Jx eτ ne τ 1 Jx σ 16 Figure 2.6: Schematic of Hall geometry Kittel p.165 The electric field in the y-direction is related to the magnetic field in the z-direction and the current in the x-direction ωc τ Ey = vx B = − Jx (2.28) σ where we have defined the cyclotron frequency ωc = eB m (2.29) In this situation the product ωc τ is the fraction of the cyclotron period completed before scattering. Hall angle: The angle between the current and the applied field tan ΘH = Ey = −ωc τ Ex (2.30) Hall coefficient: This is the usually quoted result, and is defined by RH = 17 VH BI (2.31) where the Hall voltage VH is the voltage induced by the magnetic field. Using the results above we see that the Hall coefficient is given by 1 RH = (2.32) nq where n is the number of charge carriers per unit volume and q is the charge on them. The Hall effect is small in metals due to the large carrier concentration, however as we shall see the carrier concentration in semiconductors is much lower making the effect much larger. The Hall effect can be used to measure the carrier concentration in semiconductors, alternatively a semiconductor with a known carrier concentration can be applied in a Hall probe to measure the magnetic field. The experimentally measured quantity is the Hall voltage B (2.33) VH = − I ne whilst the magnetic field B and current I are controlled. Importantly the Hall effect measures not only the magnitude of the carrier concentration, but also the sign of the charge carriers. 2.2.7 Response to electromagnetic waves: ac conductivity Let us consider the effect of an ac electric field of angular frequency ω h i −iωt ~ ~ E(t) = Re E(ω)e (2.34) ~ where the amplitude of the electric field at ω is E(ω). The resultant velocity of the electrons will be of the form ~v (t) = Re ~v (ω)e−iωt (2.35) and will be governed by the equation d~v ~v e ~ = − − E(t) dt τ m (2.36) 1 e ~ −iω~v (ω) = − ~v (ω) − E(ω) τ m (2.37) Combining these we have ~ which gives us ~v (ω) in terms of E(ω). We can rewrite the current density as ~ J(ω) = −ne~v (ω) (2.38) for the moment let us just consider the magnitudes as the directions are the same. Now we can rewrite the current density in terms of the electric field using 2.37 ⇒ J(ω) = = ne2 τ m E(ω) 1 − iωτ σ0 E(ω) 1 − iωτ (2.39) where we have written the dc (Drude) conductivity as σ0 . From this we can ascertain the (complex) ac conductivity J(ω) = σ(ω)E(ω) σ0 σ(ω) = 1 − iωτ 18 (2.40) The zero frequency limit of this expression is clearly the dc conductivity. At high frequencies the oscillation of the electric field becomes too fast for the electrons to follow, and there is zero conductivity. The ac conductivity is a complex quantity, the complex conductivity σ e(ω) = σ1 + iσ2 , and can be measured directly using terahertz time-domain spectroscopy [Section 5.1]. It is worth noting that the real part of the conductivity gives the in phase response, and the imaginary part the out of phase response. The imaginary part can be interpreted as showing that the current lags behind the driving electric field, due to the finite time taken to accelerate electrons. 2.2.8 Experimental verification / success and failure of the free electron model The free electron model works well for properties dependent on the density of states alone, for example correctly predicting Ohm’s Law and the Lorenz number. However there are a number of places where it falls down1 . For example: The Hall Coefficient The free electron model cannot even account for the positive sign of the Hall Coefficient for some materials, let alone its field and temperature dependence. The Wiedemann-Franz Law Why does this break down at low temperatures? ac conductivity Although the free electron model predicts some aspects of this accurately, it cannot predict for example the colour of copper and gold. There are more fundamental problems as well, such as: What determines the number of conduction electrons? Why are some elements nonmetals? The role of the metallic ions Drude’s original idea that scattering was from the ion cores gives scattering times that are far too short. 1 For more information see Ashcroft and Mermin, Chapter 3, or Singleton, Section 1.4 19 2.3 2.3.1 The nearly free electron model Electrons in a periodic potential To extend the free electron model we must take some account of the background lattice of atomic nuclei. The simplest approximation is to add a periodic potential which represents these nuclei and in an approximate manner the core electron states. To ruin the surprise, the amazing result is that much of solid state physics can be explained in terms of a nearly free electron gas. i.e. that the electrons are effectively independent of one another except during collisions. We can write the Schrodinger equation as before HΨ H = Eψ ~2 2 ∇ + V (~r) = − 2m (2.41) Recalling from the structure course that we can write the translation vectors of a 3D crystal lattice as ~ = n1~a1 + n2~a2 + n3~a3 R (2.42) then we demand that the crystal potential V (~r) has the periodicity of the lattice, i.e. ~ = V (~r) V (~r + R) (2.43) Figure 2.7: Sketch of lattice potentials, Kittel p.178 2.3.2 Bloch’s theorem Bloch’s theorem states that if the potential is periodic as defined above then eigenstates of the Hamiltonian may be chosen such that ~ = ei~k.R~ Ψ(~r) Ψ(~r + R) (2.44) for some ~k. In words this says that if you translate by a lattice vector you change the wavefunction only through a change in phase. An equivalent statement is that eigenstates may be written as ~ Ψ(~r) = eik.~r unk (~r) ~ unk (~r + R) = unk (~r) The quantity ~~k is the crystal momentum, and ~k the wavevector, of the eigenstate. 20 (2.45) (2.46) Proof of Bloch’s theorem With the periodic potential the Hamiltonian has discrete translational symmetry. Defining the translational operator ~ T̂R~ Ψ(~r) = Ψ(~r + R) (2.47) ~ is a lattice vector then operating T̂ ~ on the Hamiltonian gives if R R T̂R~ ĤΨ(~r) = − ~2 2 ~ Ψ(~r + R) ~ ∇ + V (~r + R) 2m ~2 2 ~ ∇ + V (~r)Ψ(~r + R) 2m = Ĥ T̂R~ Ψ(~r) = − (2.48) This implies that Ĥ and T̂R~ commute, which in turn implies that simultaneous eigenstates of the Hamiltonian and translation operator exist, i.e. ĤΨ = Eψ T̂R~ Ψ = cR Ψ (2.49) The order in which translations are made does not matter, so translation operators commute, i.e. [T̂R1 , T̂R2 ] = 0 (2.50) ~ has been dropped. But since the order of translations does not matter where the vector notation for R T̂R1 +R2 Ψ = cR1 +R2 Ψ = T̂R1 T̂R2 Ψ = cR1 cR2 Ψ (2.51) this gives a relation between eigenvalues of the translational operator cR1 +R2 = cR1 cR2 (2.52) From this we can deduce that the eigenvalues must be of the form cR = eik·R (2.53) Putting this together we arrive at Bloch’s theorem T̂R Ψ(~r) = eik·R Ψ(~r) = Ψ(~r + R) (2.54) Properties of Bloch states Momentum: Bloch states for a non-constant potential are not eigenstates of the momentum operator. In the ~ free electron model the wavefunction Ψ(r) = ceik·~r is an eigenstate of the momentum operator (p̂ = −i~∇) with eigenvalue ~~k. However this is not generally the case in the nearly free electron model. If we consider the representation of the Bloch states given in 2.45 above then ~ ~ (−i~∇)Ψnk = (−i~∇)eik.~r unk (~r) = ~~kΨnk − i~eik.~r ∇unk (~r) (2.55) i.e. Ψnk is not an eigenstate of the momentum operator. The quantity ~~k is instead referred to as the crystal momentum. 21 Crystal momentum and the Brillouin zone: The significance of the crystal momentum will hopefully emerge ~ to ~k. Recalling from the structure in later lectures, here let us consider adding a reciprocal lattice vector, G course that the dot product of a reciprocal lattice vector and a lattice vector is a multiple of 2π we have ~ ~ ~ ~ ei(k+G).~r = eik.~r+i2πn = eik.~r (2.56) The phase is unchanged by the addition of the reciprocal lattice vector, or in other words the wavevector ~ We can choose the k with the lowest modulus; this is equivalent to choosing can be written as any ~k + nG. ~k to be within the first Brillouin zone. 2 If we take all solutions of the Schrodinger equation associated with the nearly free electron model we find that the energies of the states form a continuous function of ~k which are referred to as energy bands. We also find that at each ~k there is an infinite number of energy states (due to equivalence under addition of reciprocal lattice vectors). We can label each state with wavevector ~k with a second index n, called the band index. By convention the lowest energy state at a given ~k is given by n = 1 the next by n = 2 etc. To specify a Bloch state we must specify n and ~k, hence we wrote Ψnk . The reduced zone scheme: Label each state by a wavevector within the first Brillouin zone. The extended zone scheme: Label each state by a single wavevector which may be larger than the first Brillouin zone. The repeated or periodic zone scheme: Use all possible wavevectors to label a single state. We can illustrate these ideas by considering the empty lattice approximation. 2.3.3 The 1D empty lattice approximation We can apply the ideas above to the free electron model in 1D. For free electrons V (r) = 0. We can ascribe an arbitrary lattice constant of a, so trivially V (r + a) = V (r). The solutions to the Schrodinger equation will be as in the free electron model, i.e. − ~2 d2 Ψ(x) 2m dx2 = Ψ(x) = ε = εΨ(x) 1 √ eiq.x V 2 2 ~ q 2m (2.57) The energy dispersion diagram is parabolic as discussed before. We can now imagine that the system is periodic with lattice parameter a. Let q = k + G, where the reciprocal lattice vectore G = n 2π a and − πa < k < πa . That is to say we define an arbitrary Brillouin zone and demand that the wavevector is within it. Now we can write the wavefunction and energy as Ψnk (x) = εnk = 1 1 √ eik.x eiGn .x = √ eik.x un (x) V V ~2 (k + Gn )2 2m (2.58) 2 Recall from the structure course that the first Brillouin zone is the Wigner-Seitz cell of the reciprocal lattice. It can be constructed by taking all points closer to the origin of reciprocal lattice space than the nearest reciprocal lattice vectors, its boundaries are thus constructed from the perpendicular bisectors between the origin and the nearest reciprocal lattice points. 22 2.3.4 Nearly free electron gas In this approximation the atomic lattice and core electrons are included via a weak perturbative, periodic potential acting on an otherwise free electron gas. We will use perturbation theory to analyse the effect of this potential, so first we’ll briefly review perturbation theory. Perturbation theory We consider adding a small perturbation Ĥ I to a Hamiltonian Ĥ 0 for which the solutions are known. The total Hamiltonian is Ĥ = Ĥ 0 + Ĥ I (2.59) and we know the solutions Ĥ 0 Ψ0n = En0 Ψ0n (2.60) Perturbation theory then tells us that the eigenstates of the full Hamiltonian are given by (to first order) |Ψn i = |Ψ0n i + X hΨ0 |Ĥ I |Ψ0 i m n |Ψ0m i 0 En0 − Em (2.61) m6=n and the corresponding eigenvalues or energies are given by En = En0 + hΨ0n |Ĥ I |Ψ0n i + X |hΨ0 |Ĥ I |Ψ0 i|2 n m 0 En0 − Em (2.62) m6=n Perturbing the free electron gas Here we start from the free electron Hamiltonian in one dimension, which has the now familiar eigenstates and energies Ψ0q (x) = Eq0 = 1 √ eiq.x V ~2 q 2 2m (2.63) We consider first the simplest possible periodic perturbation X Ĥ I = V (x) = VG eiG.x (2.64) G where VG is small. Let’s consider the first matrix element Z X 0 1 hΨq |V (x)|Ψq0 i = dx ei(q −q).x VG eiG.x V G ( ⇒ hΨq |V (x)|Ψq0 i = 6 G for any G if q − q 0 = 0 if q − q = G 0 VG (2.65) This gives the eigenstates and energies Ψq = Ψ0q + X Eq0 + X G Eq = VG Ψ0q+G 0 Eq0 − Eq+G Eq0 G 23 |VG |2 0 − Eq+G (2.66) Letting q = k + Gm we can rewrite this as Ψmk Ψ0mk + X 0 = Emk + X = Gn Emk Gn 2.3.5 VGn Ψ0(m+n)k 0 − E0 Emk (m+n)k |VGn |2 0 − E0 Emk (m+n)k (2.67) Zone boundaries, origin of energy gap, physical interpretation From this we can see that the situation becomes interesting where the states are degenerate, i.e. at the zone 0 boundaries where Bragg reflection occurs. At these positions Eq0 = Eq+G , and the second term diverges. Near these points we must apply instead degenerate perturbation theory; we start by writing the eigenstate as a linear combination of the two degenerate states Ψ = cq Ψ0q + cq+G Ψ0q+G The Hamiltonian can be expressed in this two state basis as 0 VG Eq Ĥ = 0 V−G Eq+G and we must solve Ĥ cq cq+G =E cq (2.68) (2.69) cq+G (2.70) For simplicity we consider first the situation at the degeneracy point where ε0q = ε0q+G = ε0 then the solutions of the Hamiltonian are given by 0 ε −E VG =0 (2.71) VG ε0 − E This can trivially be solved ⇒ E = ε0 ± VG (2.72) v !2 u 0 u Eq0 − E 0 Eq0 + Eq+G q+G t ± + |VG |2 E= 2 2 (2.73) In general The effect of the periodic potential is to open up a gap in the density states at the zone boundaries, breaking the degeneracy of the states. 24 Figure 2.8: Sketch of nearly free electron model in periodic, extended and reduced zone scheme, Ashcroft and Mermin p. 160, c.f. Myers p.183 Why does the band gap appear? We can also solve for the wavefunction of the two states at the zone boundaries, i.e. where k = ± πa . At the zone boundaries the energies are E = ±|VG | + ε0 , so to find the eigenvectors we must solve c+ πa ∓|VG | VG =0 c− πa V−G ∓|VG | (2.74) This can be trivially solved to give the eigenvectors c+ πa c− πa 1 = 1 25 or 1 −1 (2.75) Explicitly we can write these two solutions as Ψ− = Ψ+ = π x a π sin x a cos (2.76) In the absence of a potential these states have identical energy. In the presence of the periodic potential Ψ+ has a higher charge density in the low points of the potential, and hence a lower potential energy. The low points of the potential correspond to where the atoms in the lattice would be. Figure 2.9: Sketch of wavefunctions and background potential, Kittel p.178 How many states are in a band? x In a finite crystal only discrete wavevectors are allowed corresponding to kx = ± 2πn Lx . There is only one 3 (2π) allowed ~k state per volume V in ~k-space. The total number of electron states in a band is no. of states = 2 V (2π)3 = 2N (2π)3 Vcell (2.77) where N is the number of unit cells in the crystal. The factor of 2 comes from spin degeneracy. At the simplest level we can then see that if there are an odd number of electrons per unit cell then a half filled band will result, whilst an even number of electrons will result in a filled band. This gives a hint to why some materials are metallic, and some materials are insulators. A material with a filled band can still be metallic if there is band overlap, as we now discuss. 26 2.3.6 2D: band overlaps, density of states and semimetals Let’s consider the next most complicated case after the 1D lattice, the 2D square lattice. The direction of the k vector in the energy dispersion diagram must be considered. This can result in band overlap. First let us consider the bandstructure looking in different directions Figure 2.10: Sketch of bandstructure in 2D, Needs 4.12, Myers p.186 Alternatively we can draw a map of equal energy contours in reciprocal space Figure 2.11: Sketch of equal energy contours in reciprocal space, Needs 4.12, Myers p.187 Or we can plot out the density of states Figure 2.12: Sketch of the density of states, Needs 4.12, Myers p.189 27 Example: band structure of Al We can compare the bandstructure of Al (face centred cubic) derived from detailed calculations with that predicted by an empty lattice model. 3 Figure 2.13: A comparison of the empty lattice model and detailed calculations for the bandstructure of Al, Myers p.197 There is a clear similarity between the free electron model and the detailed calculation. Note that the effective size of the Fourier components of the potential can be extracted from a comparison of the two. As discussed before these will be significantly weaker than those one would estimate from the atomic potentials, and instead reflect the magnitude of the appropriate pseudopotential [see Sec. 2.4.1]. 3 The Fermi surface can be probed experimentally by e.g. ARPES (see Lecture 3) and the de Haas Van Alphen effect – see Myers Chapter 9. 28 2.3.7 Electrical classification of crystalline solids Metals: finite density of states at the Fermi level, partially filled bands or overlapping bands Insulators: filled band with a large energy gap Eg to the next empty band, Eg & 2 − 3eV Semi-conductors: filled band with a small energy gap Eg to the next empty band, Eg . 2 − 3eV Semi-metals: filled band but with touching or small overlap of bands at the Fermi level Figure 2.14: Sketch of band structures of metals, semi-metals, semi-conductors, insulators, c.f. Kittel p.194 29 2.4 2.4.1 Alternative approaches to bandstructure calculation Pseudopotential In reality the potential experienced by electrons in a solid is hugely complicated with contributions from the lattice and the other electrons, both core and valence. We have reduced all this to a simple periodic potential, so it is worth considering how valid an exercise this is. For some metals this simple picture is surprisingly accurate. The overlapping atomic potentials result in a strong and fast varying potential, far too strong for the nearly-free electron model to apply. However when considering the valence electrons we must also consider the effect of the core electrons which will be more tightly localised to the atomic lattice. The valence electron states must be orthogonal to these core states. We can construct an effective potential for the valence states which includes the ‘Pauli repulsion effect’ forcing this orthogonality. The real potential is strongly attractive near each atom, but the Pauli exclusion repels the valence electrons from the atom cores so that they see a less attractive potential. This ‘pseudopotential’ is weaker and the nearly free electron model becomes applicable. The wavefunction contains fewer nodes (see figure), meaning far fewer Fourier components are needed to describe the wavefunction – plane wave basis sets then become more appropriate. Figure 2.15: Sketch of real potential and pseudopotential, (Wikimedia commons) The nearly free electron model seems such an outrageous simplification that even using the pseudopotential justification it is still not compelling evidence on its own that crystalline solids will form bandstructures with allowed energy bands and gaps between them. However this is generically true. We can look at two further relatively similar models which demonstrate this. 2.4.2 Kronig-Penney model A useful model for investigating how/why energy bands are formed in solids was introduced in the 1930s by Kronig and Penney. It is treated in reasonable detail in Kittel (pp. 180-182), here the results are reviewed without a full derivation. Consider a 1-D square-well periodic potential such as that shown in the figure below. The time-independent Schrodinger equation must be obeyed in all space − ~2 d2 Ψ + U (x)Ψ = EΨ 2m dx2 (2.78) This can be solved by considering the form of the function within and outside a given barrier, and then constructing Bloch states from it. Let us consider limits of the situation. U0 small: this is equivalent to the nearly free electron model we described before, so energy bands will be created with small gaps between them. 30 Figure 2.16: Square-well periodic potential as introduced by Kronig and Penney, see e.g. Kittel p. 182 U0 large: each well is isolated from its neighbours, and so we have an array of one-dimensional particle in a box problems. The resultant wavefunctions are Ψn = sin( nπ ) a with energy levels ~2 n 2 π 2 2m a2 We can create Bloch states from these by writing them as X Ψkn = eikaj Ψnj En = (2.79) j This is equivalent to flat energy ’bands’ where the allowed states are the atomic energy levels. moderate U0 : in between these situation we get large energy gaps and relatively flat bands, with the band width dependent on the depth of the potential. The large U0 atomic-like case suggests that the bands can be linked to the atomic orbitals, a link which is not apparent in the nearly free electron model. Exercise 2.1 Log-in to NanoHub and use the Periodic Potential Lab to experiment with the Kronig-Penney model. Try changing the barrier heights and comparing the resultant dispersions to the free electron model. 2.4.3 The tight-binding approach Although the nearly-free electron model makes some conceptual sense for metals such as aluminium, where a traditional viewpoint would be of a sea of electrons around a lattice of positive nuclei, it does not fit at all with covalently bound solids such as diamond or graphite. Here the electrons are bound tightly to the atoms, and a more reasonable approach would seem to be to start with atomic orbitals and perturb them, rather than starting with a free electron and perturbing that. This is the origin of the ‘tight-binding’ approach. The tight-binding model is an approximate method for calculating the electronic band structure of a solid. It works on an approximation to the full Hamiltonian of a structure (e.g. a crystalline structure) analogous to the linear combination of atomic orbitals used for molecular structure calculations. The atoms are treated first as independent to find the atomic orbitals. In the crystal there will be some overlap between atomic orbitals so that they are not true eigenfunctions of the full crystal Hamiltonian. However, the tightbinding approximation assumes that the overlap is small and treats it as a perturbation to the atomic wavefunction - often only nearest neighbour interactions are considered. Starting from the atomic orbitals, Bloch wavefunctions are then constructed and the resultant dispersion relation is calculated dependent on the overlap of the atomic states, and on the crystal structure. The tight-binding model is discussed in Hook and Hall, and at a more advanced level in Ashcroft and Mermin. 31 The tight-binding model is discussed in lectures [1D case] and in the Examples sheet for a real system [graphene]. Here for illustration purposes let’s look at a tight-binding calculation for a cubic lattice with one atom per unit cell, with each atom having only one valence orbital, φ(~r), (e.g. an s-state). We can make a Bloch state by writing the wavefunction 1 X i~k·R~ m ~ m) e φ(~r − R Ψk (~r) = √ N m (2.80) ~ m is the position of the mth atom in the lattice. where R Exercise 2.2 Confirm that this is a Bloch function by applying a translation operator (T̂ ). The expectation energy of the Hamiltonian is given by 1 X i~k·(R~ n −R~ m ) h~k|Ĥ|~ki = e hφm |Ĥ|φn i N m,n (2.81) ~ m ). The term hφm |Ĥ|φn i is the overlap between states on atom n and where |~ki = Ψk (~r) and φm = φ(~r − R atom m. We are assuming that the states are tightly bound, i.e. localised on the atoms, and so the overlap will be large if n = m but will decrease rapidly as the separation increases. Here we consider only nearest neighbour interactions and so write hφn |Ĥ|φn i = −α, hφm |Ĥ|φn i = −γ hφm |Ĥ|φn i = 0 (2.82) if n and m are nearest neighbours, otherwise. (2.83) (2.84) With this assumption we have E~k = h~k|Ĥ|~ki = −α − γ X ~ ~ eik.Rneighbour (2.85) n The first term −α = hφn |Ĥ|φn i Z ~ n )Ĥφ(~r − R ~ n ) d~r = φ∗ (~r − R Z = φ∗ (~r)Ĥφ(~r) d~r (2.86) (2.87) (2.88) reflects the binding energy of the atom (note that there may be some corrections to the Hamiltonian as compared to an isolated atom in free space). The second term depends on −γ = hφm |Ĥ|φn i Z ~ m )Ĥφ(~r − R ~ n ) d~r = φ∗ (~r − R Z ~ neighbour )Ĥφ(~r) d~r = φ∗ (~r − R (2.89) (2.90) (2.91) i.e. the overlap between orbitals on neighbouring atoms. This is often called the overlap integral, and its value will depend on the type of orbitals, orientation of the bonds etc. The tight-binding model can be implemented at different levels, from the semi-empirical where α and γ are found from fits to experiments, to the ab-initio where they calculated explicitly from calculated wavefunctions and derived Hamiltonians. 32 ~ neighbour are the The other part of the second term involves the structure of the system more explicitly (R vectors joining each atom to its nearest neighbours). For the example we are considering here of a simple cubic lattice with one s-state ~ neighbour } = {(a, 0, 0), (−a, 0, 0), (0, a, 0), (0, −a, 0), (0, 0, a), (0, 0, −a)} {R (2.92) and the overlap integral is the same in each direction, giving E~k = −α − 2γ(cos(kx a) + cos(ky a) + cos(kz a)) (2.93) This is periodic, but as before only wavevectors within the first Brillouin Zone are unique so we can sketch it in the reduced, or periodic zone schemes. The ’band width’ is the energy difference between the top and bottom of the band, which here is 12γ, i.e. proportional to the degree of overlap. For |k| |π/a| the dispersion is approximately E ∝ k 2 , i.e. free electron like. Other bands of states can be constructed from different atomic orbitals. The result is a series of bands with energy gaps similar to the result derived for the nearly free electron case. When bands arise from p and d states the additional degeneracy associated with the allowed angular momentum of the bands must be included, and the relative orientations and shapes of the bands must be considered. The accuracy of the tight-binding model can be increased by including next-nearest neighbour interactions and further, and by adding overlaps between different orbitals. Exercise 2.3 Log-on to NanoHub and look at the bandstructure lab - this calculates the bandstructure of various materials using the tight-binding approach. 2.4.4 Density functional theory As the number of atoms increases the number of wavefunctions and interactions scales exponentially, so that for large systems the tight-binding approach becomes computationally excessive. Instead approaches based instead on the electron density rather than the individual wavefunctions are used: these are the basis for density functional theory calculations - see e.g. PX441. 33 2.5 2.5.1 Carrier dynamics and collisions Conservation of energy and crystal momentum Collisions between particles underlie many solid state phenomena, for example collisions between the electrons and nuclei which form the crystal, between photons and electrons, or neutrons and nuclei. The normal rules of conservation of energy and momentum apply to the system as a whole, however as we pointed out earlier the individual electron states in the presence of a periodic potential do not have a defined momentum. Instead we introduced the concept of the crystal momentum, ~~k to describe the electron states. The lattice vibrations, or phonons, also have a defined crystal momentum. For spectroscopic purposes incident beams of photons, electrons and neutrons with well defined momentum can be created. It is worthwhile to consider scattering processes involving particles with defined crystal momentum. The law of conservation of crystal momentum asserts that 4 X X ~ ~~ki = ~~kf + ~G (2.94) i f ~ is any reciprocal lattice vector. where i labels the initial crystal momenta of the particles, f the final, and G This makes sense as the wave vector is only defined to within a reciprocal lattice vector anyway. We should be clear however that crystal momentum is only conserved for periodic systems. Let’s now consider the implications of this for two important scattering processes, the interaction of an electron and a photon, and between an electron and a phonon. Electronic transition with absorption of a photon What is the momentum of a 1 eV photon? For the photon ~ω = ~cq so q = ω c ' 0.5 × 107 m−1 . What is a typical crystal momentum of an electron? Typical lattice constant is a ∼ 4 Å, so the Brillouin 10 −1 . zone width is 2π a ' 1.6 × 10 m The photon wavevector is thus less than a thousandth the width of the Brillouin zone, whilst its energy is typical for a semiconductor band gap. So the photon induced transition is essentially vertical when plotted on the bandstructure. Electronic transition with absorption of a phonon What is the momentum of a typical phonon? The crystal momenta of the phonon similarly to the photon is in the range q < πa . What is the energy of a typical phonon? The largest frequency will be ωmax ' vs πa , where vs is the velocity of sound (typically vs = 103 ms−1 ). The energy is then given by ~ωmax ' 0.5 × 10−2 eV . The phonon crystal momentum is thus in the range of the Brillouin zone, but the energy scales are small compared with most transitions of interest. The phonon induced transition is essentially horizontal when plotted on the bandstructure. You should check the numbers estimated above. 2.5.2 Scattering: classical versus quantum pictures In Drude’s original model (Sec. 2.2.4) the positive ions were the scattering points, and were regarded as randomising the direction of the charge carrier’s momentum every time τ . In this classical picture, scattering 4 For a derivation of this see Ashcroft and Mermin Appendix M 34 is thus detrimental to electrical conduction. However, in the quantum picture (of electrons as quasiparticles/wavepackets, with defined energy bands) scattering is actually required for a material to be conductive! This surprising (and somewhat counterintuitive) conclusion can be reached from the following argument:5 . 1. Consider an electron at the bottom of a single (e.g. tight-binding) band, k = 0. If a constant electric field is applied in the x−direction, the electron accelerates under this force (in the −x−direction), and gains momentum. 2. As the electron gains momentum the effective mass begins to increase, and therefore the acceleration decreases. 3. Eventually the effective mass flips in sign, because the curvature of the band is negative at higher energies. In this case the electron is now accelerated in the opposite () direction! It loses momentum, returning towards k = 0. 4. As it approaches k = 0 the curvature returns to positive, and the electron is accelerated back in the original (−x) direction. 5. This process will repeat - the electron therefore only oscillates in real space (and reciprocal space), and there is no current flow, i.e. zero conductivity. Scattering creates conduction as follows: the action of an electric field gives the electron a small velocity for a short time. The electron then scatters, randomising it’s momentum, and ‘reseting’ the electron to an average value of k = 0. Each short period of restricted velocity gain (before scattering) gives electrons a finite drift velocity, and a current flows. 2.5.3 Scattering mechanisms and Mattheissen’s rule In Drude’s model, and our previous discussion, no real mention of the possible mechanisms that contribute to electron scattering was made. In the quantum picture of electrons moving in a periodic potential, anything that interrupts the crystal’s perfection can act as a scattering centre. Thus, phonons (displacements of atoms from their equilibrium positions) will contribute at non-zero temperatures, as will vacancies (missing atoms) and impurities (such as donors or acceptors). Quantum mechanical scattering rates can be derived for these different electron scattering mechanisms,6 but here we restrict our discussion to their temperature dependence, which can often identify the dominant scattering mechanism. Controlling τ , and thus the mobility and conductivity, is obviously of key interest in optimising semiconductor materials for efficient devices. 1.) Impurities: Scattering from impurities is roughly independent of temperature, and will depend on the number of impurities present 1 ∝ concentration of impurities τimp (2.95) 2.) electron-phonon scattering: Electron-phonon scattering scattering will be temperature dependent as the rate of scattering will depend on the number of phonons which at high temperature is proportional to temperature (c.f. structure course) 1 ⇒ ∝T (2.96) τph At high temperature 3.) electron-electron scattering: This process cannot contribute to the momentum relaxation as momentum must be conserved in the collisions. 5 See 6 See Singleton, Sec. 9.1 for a fuller discussion e.g. Yu and Cardona’s ‘Fundamentals of Semiconductors’ for a full treatment. 35 Mattheissen’s Rule The scattering rates add, i.e. 1 1 1 = + τ τph τimp (2.97) m 1 1 = 2 σ ne τ (2.98) Rewriting the conductivity as a resistivity ρ= For instance, for a metal such as Cu and Al (see Fig. 2.17) the resistivity increases linearly with temperature at room temperature, suggesting that electron-phonon scattering dominates. Figure 2.17: Sketch of temperature dependence of conductivity for a good metal, Myers p.154 For a typical metal at room temperature the mean free path is l ∼ 10 − 100 Å corresponding to a momentum relaxation rate of τ ∼ 10−14 − 10−15 s. For carefully prepared samples at low temperature the mean free path can be increased up to about the µm length scales. 2.5.4 Effective mass model Let us consider the conductivity of a metal in the free electron approximation. Although we might expect the lattice to strongly scatter the electrons and result in a significantly higher resistivity, this is already included in the Bloch functions. The electrons propagate through the periodic potential without attenuation due to constructive interference of the waves scattered from the individual ions. A perfect metallic crystal with static ions would have no electrical resistance. In a real metal scattering is due to phonons and defects as discussed before in section 2.2.4. ~ The mean velocity of a free electron is ~v = ~mk . The mean velocity of a Bloch electron in the state Ψnk with 7 energy εnk is 1 ∂εnk vnk = (2.99) ~ ∂~k You should check that for the free electron case this gives the expected result. We can think of the electrical current being carried by a wavepacket of Bloch states, the actual form of which is not important. X Φn (~r, ~k, t) = g(~k, ~k 0 )Ψnk0 (~r)e−iεnk0 t/~ (2.100) ~ k0 7 see Appendix E of Ashcroft and Mermin for a proof of this result 36 For the moment let’s consider a Gaussian of width ∆~k. The wavepacket is also peaked in real space, and will move with the group velocity 2.99. We expect the spatial extent of the wavepacket to be intermediate between the unit cell of the crystal and the wavelength of the applied field (∆~k will be small compared to the width of the Brillouin zone so it is much bigger than the unit cell). Figure 2.18: Sketch of wavepacket and unit cell, p.217 Ashcroft and Mermin Since the external fields vary only slowly over the wavepacket the dynamics of the wavepacket can be described classically, whilst the effect of the ions on the electrons is treated quantum mechanically. This is called the semiclassical model. Effective mass Classically, Newton’s Laws state that the rate of change of momentum is equal to the force, F = dp/dt. Here the rate of change of crystal momentum is equal to the external force8 . F =~ d~k d~v = m∗ dt dt (2.101) Note that the right-hand equality is only true if dm∗ /dt = 0, which is only the case close to band extrema. The more general definition of m∗ = p/(dE/dp) was given in lectures. Recalling 2.99 ! ∂~k ∂~vnk ∂~v ∂~k 1 ∂ 2 εnk ~ = = 2 (2.102) ∂t ~ ∂~k 2 ∂t ∂~k ∂t From this we can define an effective mass, m∗ ⇒ 1 1 ∂ 2 εnk = 2 ∗ m ~ ∂~k 2 (2.103) When acted upon by an external field the electron will react as though it were a free particle with mass m∗ . The effect of the force due to the periodic lattice is incorporated into the definition of the effective mass, under certain conditions the effective mass can even be negative such that the electron accelerates in the “wrong” direction. Newton’s Laws are still being obeyed, but the whole system must be considered. The effective mass gives a more convenient approach. In general semiconductors are anisotropic, and the effective mass is a tensor. 1 ∂ 2 εnk 1 = 2 ∗ mij ~ ∂~ki~kj 8 This result was derived in lectures, see also Kittel chapter 8 37 (2.104) Motion in a uniform electric field In an external electric field the semiclassical equation of motion is ~ d~k dt ~k(t) ~ = −eE ~ eEt = ~k(0) − ~ (2.105) i.e. the wavevector changes with a rate proportional to the applied field, but independent of ~k. In a full band the electrons move out of the first Brillouin zone in one direction and in from the other side. There is no net current flow and as a result the material is an insulator. In real space the electrons are oscillating about fixed points, known as Bloch oscillations. In reality this is very difficult to observe experimentally due to scattering, which we have temporarily ignored. 2.5.5 Holes The current density from electrons in a band (isotropic, and in 3D) is given by ~j = (−e)2 Z ~vk d3~k (2π)3 (2.106) occupied where the factor of two arises from spin degeneracy (two spins per k-point). If all the states in the band are full the net current will be ~j = 0. So let’s consider an almost full band Z ~vk ~j = (−e) 2 d3~k (2π)3 occupied Z Z ~vk ~vk 3~ d d3~k = (−e) 2 k − (−e) 2 3 (2π) (2π)3 unocc. f ull Z ~vk d3~k (2.107) = 0 + (+e) 2 (2π)3 unocc. i.e. we can consider the current to be carried not by the electrons, but by the unoccupied states known as holes. This leads to two sources of current R ~ vk 3~ ~j = (−e) 2 electrons in the conduction band (2π)3 d k unocc. ~j = 0 + (+e) 2 R unocc. ~ vk (2π)3 d3~k holes in the valence band (2.108) Properties of holes Let’s look at the properties of a single hole in the valence band, compared to those of the corresponding electron.9 Charge ofPa hole, qh : Consider P taking one electron out of a full band. The total charge is given by Q = full (−e) − (−e) = full (−e) + qh . So qh = e 9 See Kittel chapter 8, p.206-208 38 (2.109) Energy of the hole state, εh : A similar argument gives εh = −εe (2.110) k~h = −~ke (2.111) Wavevector of the hole state, ~kh Velocity of the hole state, ~vh Consider the current due to the full band less one electron: X X ~j = (−e)~v − (−e)~ve = (−e)~v + (+e)~vh full full As a result we see that ~vh = ~ve (2.112) Effective mass, m∗h since ~vh = ~ve but k~h = −~ke then we see that10 m∗h = −m∗e (2.113) One way to look at this is to consider a hole band constructed by inverting the hole band Figure 2.19: Inverted hole band, Kittel p.208 In short the hole behaves like a positively charged particle, and thus provides an explanation for the anomalous positive values of the Hall coefficient. 2.6 Conclusions First we looked at the free electron model where electrons are independent of one another and do not interact with the background lattice. We saw that this couldn’t account for experimental results such as positive values of the Hall coefficient for some metals. The next step accounted for the atomic lattice through the introduction of a periodic potential, resulting in Bloch states, the reduced zone scheme and band gaps. This accounted for crystals behaving as metals, semi-metals, semi-conductors and insulators. The implications of 10 Note that at the top of the valence band me is negative so the hole effective mass is positive 39 the Bloch states led to the introduction of important concepts such as the crystal momentum, group velocity, and the effective mass. The effective mass tensor is dependent on the state, for physical quantities average values of the effective mass can be used. For example the dc conductivity is due to electrons near the Fermi energy, defining an average effective mass the conductivity can be written σ= ne2 τ m∗ (2.114) Finally we introduced the concept of a hole; an empty electron state which follows the equations of motion of a positively charged particle. 40 Chapter 3 Intrinsic and Extrinsic Semiconductors 3.1 Introduction We have shown that a periodic lattice causes band gaps and briefly discussed how this can result in metals, semiconductors and insulators. In this section we will look in more detail at the physical properties of semiconductors whose properties are both interesting and technologically very important. Crystal Copper Silicon Diamond Resistivity at room temperature (Ωm) 10−8 103 1014 Clearly the conductivity of silicon whose band gap is (Eg = 1.1eV ) is intermediate between that of metallic copper and insulating diamond (Eg = 5eV ). As we shall see, conduction in semiconductors is dominated by carrier concentrations. 3.1.1 Direct and indirect band gaps Typical band gaps of semiconductors are in the range 0.2 − 3eV . For comparison light with energy 1eV has a wavelength of 1200nm (i.e. in the infrared), 2eV is 600nm (orange), and blue light (475nm) is 2.6eV . In other words optical absorption is a good way of probing the band gap of semiconductors. Experimentally two types of result are observed: Direct band gaps The top of the valence band is directly below the bottom of the conduction band, the threshold for optical absorption is ~ω = Eg .1 Indirect band gaps The bottom of the conduction band is separated from the top of the conduction band in k space. The optical absorption threshold is still at ~ω = Eg , but it requires the simultaneous absorption of a phonon which is correspondingly much less probable. A second threshold is reached at the direct gap. 1 recall that the relative crystal momentum of a photon is negligible so the transition is vertical 41 Figure 3.1: Schematics of direct and indirect band gaps. (Kittel Ch. 8 p.202) 42 Figure 3.2: Optical absorption of InSb (Kittel Ch. 8 p.203) 3.1.2 Examples Crystal Si Ge GaAs InSb group IV IV III V III V type i i d d Band gap at 300K (eV ) 1.1 0.7 1.4 0.2 Figure 3.3: (Left) structure of silicon, (Right) structure of gallium arsenide, NanoHub Crystal viewer tool. A second important distinction to make is between intrinsic and extrinsic semiconductors. The properties of a perfect crystal of a pure element or perfectly stoichiometric compound are called intrinsic properties, whereas the influence of added impurities or defects give rise to extrinsic properties.2 2 Definition from Myers Ch. 10 p.275 43 3.2 Intrinsic semiconductors For intrinsic semiconductors the carrier concentrations are determined by thermal excitation of carriers across the band gap. At T = 0K the valence band will be full, and the conduction band empty. The chemical potential is midway through the band gap, and there would be no conductance. As the temperature increases electrons are thermally excited across the band gap, as a result the concentration of electrons in the conduction band (n = p) is equal to the concentration of holes in the valence band. These electrons and holes are free to move through the crystal and can carry an electric current and so are known as carriers. Figure 3.4: Sketch of density of states and Fermi function, Needs 7.3 3.2.1 Electron concentration The electron density is given by Z∞ 1 no.electrons = n= V V dE Nc (E) f (E) (3.1) Ec Recall from 2.11 that the density of states for electrons in the conduction band is V Nc (E) = 2 π 2m∗c ~2 3/2 p E − Ec (3.2) where we have used the effective mass for electrons in the conduction band, m∗c . We introduced the Fermi function in 2.8. Here we will make the assumption that the semiconductor is non-degenerate. We will explain in more detail what this means when we discuss extrinsic semiconductors, but for the moment we assume that kB T Eg , and that µ is away from the band gap edges. f (E) = 1 ' e−(E−µ)/kB T e(E−µ)/kB T + 1 (3.3) Putting these together we find n = Z∞ 1 V dE Nc (E) f (E) Ec = = 1 π2 2m∗c ~2 n0 (T )e 3/2 Z∞ dE p E − Ec e−(E−µ)/kB T Ec −(Ec −µ)/kB T 44 (3.4) where we have defined 1 n0 (T ) = 2π 2 2m∗c ~2 3/2 Z∞ dE 0 √ E 0 e−E 0 /kB T (3.5) 0 We can simplify this using the relation Z∞ √ √ −x xe π 2 dx = 0 giving n = n0 (T )e−(Ec −µ)/kB T n0 (T ) = 2 m∗c kB T 2π~2 3/2 (3.6) This defines the electron concentration as a function of the chemical potential for non-degenerate semiconductors. Note that the exponential will dominate the temperature dependence. 3.2.2 Hole concentration We can construct a similar argument to calculate the hole density in the valence band. ZEv no.”missing” − e 1 p= = V V dE Nv (E) (1 − f (E)) (3.7) −∞ The density of states in the valence band is Nv (E) = V π2 2m∗v ~2 3/2 p Ev − E (3.8) Again we make the assumption that the semiconductor is non-degenerate i.e. that kB T Eg , and that µ is away from the band gap edges. 1 − f (E) = 1 − 1 e(E−µ)/kB T +1 = e(E−µ)/kB T ' e(E−µ)/kB T +1 e(E−µ)/kB T (3.9) Putting these together we find p = 1 V ZEv dE Nv (E) f (E) ∞ = 1 2π 2 2m∗v ~2 3/2 ZEv dE p Ev − E e(E−µ)/kB T −∞ (Ev −µ)/kB T = p0 (T )e (3.10) simplifying as above we find p = p0 (T )e(Ev −µ)/kB T 45 p0 (T ) = 2 m∗v kB T 2π~2 3/2 (3.11) This defines the hole concentration as a function of the chemical potential for non-degenerate semiconductors. 3.2.3 Mass action law If we look at the product of the electron and hole concentrations np = n0 (T )e−(Ec −µ)/kB T p0 (T )e(Ev −µ)/kB T = n0 (T ) p0 (T ) e(Ev −Ec )/kB T (3.12) This gives the ”Mass action law” np = n0 (T ) p0 (T ) e−Eg /kB T (3.13) which is true in general for non-degenerate semiconductors. For the case of intrinsic semiconductors we have the added constraint that ni = pi so that p √ ni = pi = np = n0 (T ) p0 (T )e−Eg /2kB T 3/2 kB T 3/4 (m∗c m∗v ) e−Eg /2kB T (3.14) = 2 2π~2 This defines the intrinsic carrier concentration as a function of temperature for a nondegenerate semiconductor. 3.2.4 Intrinsic carrier concentrations For intrinsic semiconductors the mass action law applies, and n = p. At T=300K, and for m∗ = m then n0 = p0 = 2.5 × 1025 m−3 . So we can estimate the carrier concentration for a semiconductor of band gap 1eV at room temperature. n ' 2.5 × 1025 ∗ e−20 ' 1017 m−3 . Using the fact that n = p it is easy to calculate the chemical potential as a function of temperature ∗ Eg mh 3 + kB T ln (3.15) µ= 2 4 m∗e i.e. close to the centre of the band gap. We can also calculate the carrier concentrations for a given intrinsic material and compare them to metals and semi-metals (see over page). Let’s quickly illustrate why these carrier concentrations are so important. 3.2.5 Conductivity The conductivity of a semiconductor can be written as σ= ne2 τe pe2 τh + m∗e m∗h (3.16) which looks very similar to the equation for a metal. However here the temperature dependence is dominated by the carrier concentrations as opposed to the scattering mechanisms (as we found for metals). We can rewrite the above as σ = (n(T ) e)µe + (p(T ) e)µh 46 (3.17) Figure 3.5: Carrier concentrations at room temperature, Kittel p.198. where the mobilities, µ, are ( µe = µh = eτe m∗ e eτh m∗ h electron mobility hole mobility (3.18) The conductivity is then determined by the mobility times the carrier concentration, the mobility will be higher in materials with low rates of scattering. Typical mobilities at room temperature are Crystal Si Ge GaAs InSb electron mobility (cm2 (V s)−1 ) 1350 3600 8000 800 47 hole mobility (cm2 (V s)−1 ) 480 1800 300 450 48 3.3 Extrinsic semiconductors The tremendous importance of semiconductors is reliant on the ability to control the carrier concentrations through the incorporation of impurity atoms. By doping pure semiconductors with small concentrations of impurities we can increase the electron concentration or the hole concentration independently. Let’s consider silicon, a group IV element with 4 valence electrons (3s2 3p2 ). If some of the silicon atoms are replaced by a group V element like phosphorus they will contribute an extra electron and so act as a donor. The potential of the group V element is otherwise not too different from the silicon, so as long as the concentration of phosphorus is low the conduction and valence bands will be basically unchanged, and the extra electron will occupy a state near the bottom of the conduction band. If instead some of the silicon was replaced by a group III atom such as boron they will contribute one fewer electron (i.e. a hole) and so act as an acceptor and create an energy level for a hole near the top of the valence band. III B Al Ga In IV C Si Ge Sn V N P As Sb A donor atom will have an extra positive charge compared to the lattice, this will attract an electron. Ignoring the lattice this looks very similar to a hydrogen atom (see figure over page). Figure 3.6: Schematic of arsenic atom in germanium lattice (A&M p.577) 49 3.3.1 Hydrogenic levels: donor ionisation energies Recall the Hamiltonian for a hydrogen atom H= p2 e2 − 2m 4π0 r (3.19) The ground state is symmetric with a characteristic size of the Bohr radius (a0 ) and energy of the Rydberg Energy (E0 ) a0 = E0 = 4π0 ~2 = 0.5Å me2 e2 = 13.6eV 8π0 a0 (3.20) (3.21) In the semiconductor the Coulomb attraction is reduced (screened) by the electronic polarisation of the medium, and of course the mass should be replaced by the effective mass. This results in a Hamiltonian Hef f = e2 p2 − 2m∗ 4π0 r (3.22) This is valid for a donor with charge Z = +e. By analogy with the hydrogen atom we can see that the ground state will have characteristic size ad and energy Ed ad = Ed = 4π0 ~2 m = ∗ a0 ∗ 2 m e m e2 m∗ 1 E0 = 8π0 ad m 2 (3.23) (3.24) Note that the ionisation energies depend only on the medium (i.e. the semiconductor) and not on the donor atom (for donors with the same charge). Obviously a similar argument can be made for acceptor levels to estimate the acceptor ionisation energy Ea . Exercise 3.1 Find the donor energy for a Group V element in silicon. The effective mass for silicon is small, m∗e ' 0.2m, and the dielectric constant is large: = 12. [You should find that Ed = 0.02 eV and ad = 3 nm]. The simple approach just outline is correct to the order of magnitude, as a comparison between the result of the exercise and the table below reveals. Importantly, Ed Eg for dopants [and similarly for acceptors], so we have energy states as shown in Fig. 3.7 that are close to either the conduction band edge (for donors) or valence band edge (for holes). Note that the degeneracy of a band can influence the donor/acceptor energies – degeneracy here in the sense of multiple levels at certain points in k-space, such as the heavy-hole and light-hole bands in GaAs. Degenerate bands have several effective masses.3 The discrepancy between the predicted donor ionisation energy for silicon and the observed can be largely accounted for by the degeneracy of the conduction band of silicon so that the effective mass should be much higher. Ed (meV) Si Ge P 45 12 As 49 13 Sb 39 10 Ea (meV) Si Ge B 45 10.4 Al 57 10.2 Ga 65 10.8 Tables of (left) donor and (right) acceptor ionisation energies in meV, values from Kittel p.224-5 3 See for example Kittel, Singleton or Ashcroft and Mermin for more details. 50 Figure 3.7: Myers p.286 Occupation of donor and acceptor levels Note that the ionisation energies are similar to thermal energies at room temperature; thermal ionization of the donors is an important contribution to the conductivity at room temperature. By comparison Eg kB T300K so thermal excitation across the band gap at room temperature is much less likely. Ignoring electron-electron interactions, the donor level could be empty, singly occupied by an electron of either spin, or doubly occupied. However Coulomb repulsion between the electrons energetically prohibits the doubly occupied state. The average occupation can then be calculated using the grand canonical ensemble P −(Ei −Ni µ)/kB T i Ni e fd = P (3.25) −(Ei −Ni µ)/kB T ie with the corresponding states i 1 2 3 Ni 0 1 (spin up) 1 (spin down) Ei 0 εd εd which gives fd = 1 1 (εd −µ)/kB T 2e +1 (3.26) The unusual factor of 1/2 comes from the prohibition of the doubly occupied state. The occupation of the acceptor state can be similarly calculated. For both the situation is further complicated by the degeneracy of the conduction and valence bands which changes the 1/2 to 1/g where g is the degeneracy. 3.3.2 Extrinsic carrier concentrations Many physical properties are dominated by the electron concentration in the conduction band (n) and the hole concentration in the valence band (p). These come from ionization of the donors and acceptors, and 51 excitations from the valence to the conduction band. As long as the semiconductor is nondegenerate the mass action law 3.13 will hold independent of whether impurities are present. The effect of the impurities is to alter the chemical potential, they remove the constraint n = p. If the number of ionised donors (Nd+ ) is much greater than the number of ionised acceptors (Na− ), i.e. Nd Na (if they have comparable binding energy), then n p and the material is called n-type: the majority carriers are electrons. If p n then the material is called p-type: the majority carriers are holes. The mass action law controls the product np; by controlling the impurities we can control the total number of carriers n + p, with the minimum being the undoped (intrinsic) case where n = p. In reality it is the values of n and p that we want to know. In a practical situation we may know or be able to measure Nd , Na , εd , εa , m∗e , m∗h , gd , ga , & T . The unknowns are Nd+ , Na− , n, p, & µ, and these are determined by the 4 equations we have already derived 3/2 ∗ mc kB T e−(Ec −µ)/kB T n = 2 2π~2 ∗ 3/2 m v kB T p = 2 e(Ev −µ)/kB T 2π~2 Nd − Nd+ Nd = Na − Na− Na = 1 1 (εd −µ)/kB T gd e +1 1 1 −(εa −µ)/kB T ga e +1 (3.27) along with the final condition of overall charge neutrality Nd+ − Na− + n − p = 0 (3.28) These can be solved numerically, but it is also worth looking at some limiting cases at specific temperatures. We will consider systems with Nd Na (and we will neglect Na ). In Fig. 3.8 the temperature dependence of n is shown for lightly-doped silicon, and reference to this figure should be made with respect to the three following regimes/ranges. Extrinsic regime/freeze-out range (low temperature) At T = 0 donors are all singly occupied, the chemical potential is between the donor levels and the conduction band, and n = p = 0. As the temperature increases the donors become ionised, it can be shown in this regime (0 < kB T Ed ) that the electron concentration is given by4 p n = n0 Nd e−Ed /2kB T (3.29) This equation predicts that ln n ∝ −1/kB T , as seen in Fig. 3.8. The excitation of electrons into the conduction band only requires thermal excitation over the relatively low barrier of Ed . In this regime the chemical potential is close to the majority carrier level. Exhaustion regime /saturation range (intermediate temperatures) As the temperature becomes greater than the donor ionisation energy the system enters the exhaustion regime (saturation range in Fig. 3.8). The donors are ionised but the probability of thermal excitation over the full band gap is still small. In this case n ' Nd (3.30) and the majority (electron) carrier concentration is roughly independent of temperature. Recall that the concentration of electrons is given by n = n0 (T )e−(Ec −µ)/kB T 4 See Myers for a derivation 52 In the exhaustion regime we have µ = Ec − kB T ln N0 Nd (3.31) i.e. the chemical potential is in the upper half of the band gap, it will move downwards towards the centre of the gap as the minority carrier concentration increases in line with the mass action law. Intrinsic regime/range (high temperature) As the temperature increases further the intrinsic carrier concentration (as defined by the mass action law 3.13) exceeds the donor concentration (ni Nd ) and the electron concentration becomes p (3.32) n = n0 (T ) p0 (T )e−Eg /2kB T in this regime n ' p ' ni Nd and the chemical found before. i.e. Eg µ= + 2 potential will be close to the middle of the band gap as ∗ 3 mh (3.33) kB T ln 4 m∗e Figure 3.8: ln(n) vs 1000/T , From Singleton p61. Degenerate and nondegenerate semiconductors All our analysis so far has been based on the assumption that the semiconductor is non-degenerate. We stated that this required kB T Eg , and that µ is away from the band gap edges. f (E) = 1 ' e−(E−µ)/kB T e(E−µ)/kB T + 1 53 (3.34) A degenerate semiconductor has high carrier concentrations, although still lower than in conventional metals. Usually this will be due to high dopant concentrations. If the dopant concentrations are too high they form an impurity band and the material no longer behaves as a semiconductor (e.g. the conductivity no longer increases with temperature). We estimated the characteristic size of the donor states earlier, from this the degenerate dopant concentrations can be predicted. Exercise 3.2 Predict the dopant concentration at which an impurity band is formed in Si. 3.3.3 Temperature and dopant dependence of n, p, µ & σ You should login to NanoHub and experiment with the simulation ‘Carrier statistics lab’ to see the effects of doping and temperature for yourself. However here we shall sketch out some of the dependencies. First let’s consider the effect of dopant concentration on the chemical potential at fixed temperature. We will consider again an n-type semiconductor with donor concentrations Nd and donor energy levels Ed . Figure 3.9: Sketch of µ vs Nd − Na Figure 3.10: Sketch of µ vs T , from Myers p.289. 3.3.4 Conductivity [revisited] Recall our previous discussion of the conductivity of materials with electrons and holes where we found the conductivity to be (3.17) σ = (n(T ) e)µe + (p(T ) e)µh (3.35) The mobilities are only weakly dependent on temperature in comparison to the carrier concentrations. Recall that the mobilities are ( e µe = eτ electron mobility m∗ e (3.36) eτh µh = m∗ hole mobility h and their temperature dependence is dominated by the scattering times τe and τh . The main scattering mechanisms are scattering by phonons which increase with increasing temperature (τph ∝ T −3/2 ), and scattering by ionized impurities which decreases with temperature so that (τi ∝ T 3/2 ). To find the total scattering time we add the inverse scattering rates. 54 For intrinsic materials, or materials demonstrating intrinsic behaviour, the conductivity will increase exponentially with temperature in line with the carrier concentration and hence gives a measure of the band gap. Doped materials will show a temperature dependence dominated by the carrier concentration, but with some contribution from the temperature dependence of the mobility. 3.3.5 The Hall effect [revisited] Let’s consider a heavily doped sample in the exhaustion regime, where p n and can be ignored. A Hall effect measurement will give us 1 RH = (3.37) nq i.e. the type of the carriers (RH negative implying electrons, as q = −e), and the carrier concentration. Measuring also the conductivity gives us σ = n eµe (3.38) so that combining with the Hall measurement we can extract the mobility. Under the assumption that the transport properties are dominated by the impurities we then have experimentally found the dominant carrier type (electrons), the donor concentration (n), and the mobility (µe ). Now consider the more complicated case where both electrons and holes are contributing to the transport. The Hall coefficient at moderate magnetic fields then becomes (you should derive this yourselves)5 RH = p µ2h − n µ2e e(p µh + n µe )2 (3.39) where e is the magnitude of the electron charge. In the intrinsic regime, where n = p >> Nd or Na then the Hall coefficient is dominated by the electrons, as µe >> µh . A question in the Problem Sheets explores the Hall coefficient for doped InSb, where both holes and electrons contribute. The conductivity is now σ = (n e)µe + (p e)µh (3.40) implying that care must be taken in extracting carrier concentrations, mobilities etc. from Hall effect and conductivity data. 3.3.6 Carrier dynamics In considering the conductivity we looked at the drift of charge carriers induced by an electric field, i.e. situations in which the carrier concentrations are effectively uniform in space. When looking at device structures we will see that this is not always the case. When the concentration is not constant diffusion of charge carriers must be included as well as drift. Other important effects include carrier generation and recombination which may create or destroy charge carriers. For example photoexcitation of charge carriers across the band gap will generate carriers and create a system out of thermal equilibrium. We will consider these effects in more detail when we look at semiconductor devices in Chapter 4. Non-equilibrium carrier dynamics after photoexcitation will be explored in detail in Chapter 5. Beforehand, however, let’s look at the light-matter interaction in semiconductors in more detail. 5 If you get stuck then it is in Hook and Hall section 5.5.2, or Singleton Section 10.2.2. 55 56 Chapter 4 Semiconductor devices 4.1 Introduction We have shown that we can control the electronic properties of semiconductors through band gap engineering and doping. Here we’ll get a taste of why that is so important by looking at a few of the basic semiconductor structures, and how these are used to facilitate our everyday life. In the process we will also learn a little more about charge transport processes in semiconductor structures. Most of the interesting physics and technological applications come from the interfaces. We will look at the interface between n-type and p-type materials, between semiconductors and metals, between semiconductors and insulators, and between two different semiconductors. We will start with perhaps the most important structure the p − n junction. This is not covered well in Myers or Kittel, but Hook and Hall has a good account, as does Ashcroft and Mermin, and it is presented in far greater detail in many books such as Sze. First however we must introduce the concept of a semiconductor heterostructure. We have seen that semiconductors can be doped to be n-type, or p-type, that the charge carrier concentration and conductivity can be controlled through the dopant concentrations, and that different materials have different band gaps. Much of the technological importance of semiconductors comes from the added ability to create heterostructures: structures where the composition of the material is spatially varied with near atomic precision. We will see in a moment what happens when an n-type semiconductor is in contact with a p-type semiconductor, however in reality it is not enough to just touch the two materials together. In that situation surface states and contamination will dominate the properties. Semiconductor research has thus developed many techniques to fabricate materials with well defined changes in composition. In particular many of the structures are planar, and so can be fabricated by layer by layer growth of the constituent parts. Examples of these epitaxial growth techniques are molecular beam epitaxy, (plasma enhanced) chemical vapour deposition, and pulsed laser deposition. These are all areas in which the physics department at Warwick has active research interests (see for example the ASR and Surface Science groups). Characterising these materials is also an active area of research, and this involves many groups and facilities within the department and at central facilities; such as the ASP (SIMS group), X-Ray and crystallography, Surface science, and Microscopy groups. You can find plenty of details about these techniques from the relevant research groups, we’ll just quickly consider molecular beam epitaxy (MBE). In MBE separate sources are used to sublime beams of atoms which then condense on a single crystal substrate. The process is in ultra-high vacuum and the beams are at low rate so that they do not interact before the surface. Once on the surface they adopt the structure and orientation of the substrate (at least initially!). Composition can be changed by changing the intensity of each beam allowing the fabrication of homo-epitaxial layers of the same semiconductor at different dopant types and concentrations, or hetero-epitaxial growth of different semiconductors. 57 4.2 p − n junction p-n junctions are very important, they are used in rectifiers, semiconductor lasers, light emitting diodes, solar cells and detectors, and are a key feature of the bipolar transistor. Let’s consider a planar junction between n-type and p-type material. First let’s look at the energy levels of the two materials separately Figure 4.1: Sketch of energy levels in (left) p-type and (right) n-type. The chemical potentials within the materials must be the same, charge carriers flow from an area of higher to lower chemical potential until this is the case. So we can draw the energy level of the junction Figure 4.2: Sketch of energy levels in p-n junction. 58 The flow of charge which equilibrates the chemical potential results in an electric dipole at the junction. The dipole is formed by the removal of charge carriers leaving the ionised donors as excess charge (can think of the electrons and holes as annihilating each other near the junction). This layer where the charge carrier concentration has been reduced is called the depletion layer, and you can see that it must exist by looking at the position of the chemical potential relative to the band edges at the junction. The chemical potential in the n-type material away from the junction is: n0 (4.1) µn = Ec − kB T ln Nd whilst in the p-type region it is given by µp = Ev + kB T ln p0 Na The difference between the two is equal to the contact potential of the junction Na Nd e∆φ = (µn − µp ) = Eg + kB T ln n0 p0 Na Nd = kB T ln n2i (4.2) (4.3) (4.4) where we have used the Mass Action Law. To simplify matters we will assume that the junction is abrupt, and that the depletion layer is also abrupt. This gives Figure 4.3: Sketch of p-n junction, charge density and carrier concentrations. Conservation of charge implies that if the width of the depletion region on the n-type side is wn , and on the p-type wp , then Na wp = Nd wn (4.5) 59 The electrostatic potential is related to the charge density by Poisson’s equation ρ(x) d2 φ =− dx2 0 (4.6) where x is the distance from the junction. Integrating this gives the field within the depletion region ( −wp < x < 0 − Na e (x + wp ) (4.7) E = Nde0 0 < x < wn 0 (x − wn ) This fulfills the boundary conditions that the field is zero away from the depletion region, and continuous at the junction. We can integrate the field to get the potential ( Na e (x + wp )2 −wp < x < 0 φ(x) = 20 (4.8) Nd e 2 ∆φ0 − 20 (x − wn ) 0 < x < wn In this case the boundary conditions were that the potential in the p-type region outside of the depletion layer is zero, whilst in the n-type region it is ∆φ0 . We can now sketch out the potential and field at the junction Figure 4.4: Sketch of p-n junction, potential and field. The potential must be continuous at the junction e (Na wp2 + Nd wn2 ) ⇒ ∆φ0 = 20 (4.9) Using the charge neutrality condition above we can now solve for the widths of the depletion layers 1/2 1/2 20 Na ∆φ0 20 Nd ∆φ0 wn = , wp = (4.10) eNd (Nd + Na ) eNa (Na + Nd ) 60 As we might have expected the depletion layer is narrower in highly doped materials. To put it in perspective the depletion layer is wn + wp ∼ 1µm for Na = Nd ∼ 1015 cm−3 , and wn + wp ∼ 0.1µm for Na = Nd ∼ 1017 cm−3 . We assumed the junction was abrupt, for this to be true the doping must change from n to p over lengthscales much less than the depletion width, well within the capabilities of modern methods. Finally we should consider the approximation that the depletion layer is abrupt - the carrier concentrations will drop off exponentially as µ moves away from the band edges. The linear change in potential across the junction thus implies an exponential change in carrier concentration, meaning our original approximation was valid. We can finish by sketching out what we now understand about the p − n junction at equilibrium. Figure 4.5: Sketch of p-n junction, charge density, carrier concentrations, field and potential. 61 4.2.1 The p − n junction under applied bias We’ve now been through the equilibrium properties of the p − n junction, but for use we need to consider the application of a bias. The p − n junction acts like a diode, rectifying the current flow. As the depletion layer is highly resistive the potential will be dropped across it. If a voltage V is applied to the n side of the device the effective contact potential becomes ∆φ = ∆φ0 − V (4.11) Figure 4.6: Sketch of potential in device and applied potential. If the potential is positive the effective barrier is decreased and the device is ’forward biased’, negative bias increases the effective barrier the device is under ’reverse bias’. The potential shifts the chemical potential on either side of the junction so we can draw it in forward bias as Figure 4.7: p − n junction under forward bias. The applied potential changes the width of the depletion layer. wn = 20 Na (∆φ0 − V ) eNd (Nd + Na ) 1/2 , wp = 20 Nd (∆φ0 − V ) eNa (Na + Nd ) 1/2 (4.12) To predict the form of the current-voltage curve across the p − n junction we must first consider the charge transport processes which setup and maintain the equilibrium. We will first consider the electron current. 62 Generation current Thermal activation of an electron in the p-type region near the junction results in the electron being swept across the junction by the internal field. The limiting step is the thermal activation across the band gap, so the current (Jegen ) will be independent of the bias voltage. Figure 4.8: Generation current in a p − n junction. Recombination current An electron from the n type region flows against the internal field, and recombines with a hole in the p-type region. Figure 4.9: Recombination current in a p − n junction. The limiting step is the flow of the electron over the potential barrier at the junction, this will be thermally activated and so the current density (Jerec (V )) will be exponentially dependent on the height of the barrier. As a result it will be exponentially dependent on the applied bias, i.e. Jerec (V ) = Jerec (0)eeV /kB T (4.13) In the absence of the applied bias the system is in equilibrium so Jegen + Jerec (0) = 0 63 (4.14) Hence the applied bias creates a total electron current of Je (V ) = je (eeV /kB T − 1) (4.15) A similar argument can be made for the hole current. Without calculating the prefactor we can then predict the I − V response of the junction I = Is (eeV /kB T − 1) (4.16) This is the diode equation. Note that understanding the details of the p − n junction is complicated and beyond the scope of this course, you need to consider not just drift currents but also diffusion currents (non-uniform carrier concentrations results in diffusion of carriers). For a more thorough discussion of the p − n junction see S.M. Sze. Figure 4.10: I − V response of the p − n junction, Ashcroft and Mermin p600. 64 4.2.2 Solar cells A photon of sufficient incident on a semiconductor can result in the excitation of an electron-hole pair. If this occurs at the junction, or within the diffusion distance of the electrons/holes, then the field of the junction will separate the electron hole pair. The photo-current flows in the direction of the reverse current of the diode. Figure 4.11: Photocurrent in a p − n junction. Figure 4.12: Electric field and carrier movement at the p − n junction. The solar cell consists of a large area unbiased p−n junction. The energy of the electron-hole pair excited by the photon is equal to the band gap (excess energy is dissipated as heat), so it is important to match the band gap to the spectrum of incident light. Multiple junction thin film devices can be used to do this more effectively (an example of Third Generation solar cells). Figure 4.13: Schematic of solar cell taken from Molecular ExpressionsTM 65 4.2.3 LEDs/Lasers A p − n junction under large forward bias creates a population inversion at the junction, which may result in radiative recombination of the electrons and holes and hence light emission with wavelength equal to the band gap. Stimulated emission gives amplification, and if placed in a suitable waveguide/cavity (e.g. cleave sharp facets at the end of a ridge structure) laser action can result. Figure 4.14: Light emission from a p − n junction, Hook and Hall p.183. The carriers are injected as minority carriers from both sides of the depletion regions. They will combine by radiative or non-radiative transitions. The junction must be designed to favour radiative recombination, a starting point for which is using direct band gap semiconductors and avoiding trap states at the junction. For description of p − n junction use in LED lasers see p.304 in Myers. 4.2.4 High electron mobility transistor [non-examinable] Another type of semiconductor junction is between materials of different band gaps. For example GaAs has a very high mobility and so is attractive for use in high-frequency devices. The introduction of ionised donors can be used to adjust the chemical potential, but in the process it reduces the mobility due to increased scattering from the donor ions. This problem can be overcome by the use of a hetero-junction structure whereby a thin layer of n-type Ga1−x Alx As is sandwiched between a metal gate and pure GaAs. An n-type layer is formed in the GaAs at the junction with the Ga1−x Alx As. As the GaAs itself is pure this results in a high mobility electron gas. The electrons are from the Ga1−x Alx As layer, and their concentration can be controlled by the gate electrode (see section below on MOSFETs). Figure 4.15: High electron mobility transistor (the GaAs is weakly p-doped rather than pure), Hook and Hall p194 66 4.3 MOSFET The metal oxide semiconductor field effect transistor. There are a huge number of different types of field effect transistor (e.g. MISFET, MESFET, IGFET, JFET, JUGFET ...). A field effect transistor is in essence a three terminal device. Typically the current is carried by the majority carriers and the device is unipolar. The current flow through two of the terminals is controlled by the voltage on the third. FETs are particularly suited to switching applications and to high density fabrication on the chips. In a FET a channel is formed through which electrons flow from the source to the drain contact. The conductivity of the channel is controlled by the potential on the third, gate, electrode. Let’s consider first the structure of a MOS capacitor Figure 4.16: Bandstructure at the MOS capacitor, Hook and Hall p.189. So by applying a potential to the gate we can create an n-type channel (inversion layer) at the semiconductoroxide junction. The conductivity of the channel can either be enhanced (enhancement mode), or depleted by the gate potential, effectively turning the channel on or off. 67 Figure 4.17: MOSFET device structure, Hook and Hall p.190. Detailed operation of the MOSFET is again beyond the scope of this course, but also again is covered in plenty of detail in S.M. Sze. The modern version of a MOSFET has optimised source drain and dielectric components (i.e. the acronym MOSFET is no longer strictly accurate!), and lateral dimensions < 50nm. It is the fundamental building block of the Intel chip, illustrating the amazing technological contribution of twentieth century Solid State Physics. Figure 4.18: Intel 45 nm transistor. 68 Chapter 5 Terahertz optoelectronics The terahertz (far-infrared) spectral range covers frequencies from 0.3 THz to 10 THz, where it is hard to generate and detect light. As 1 THz corresponds to 4.1 meV energy or 1 ps timescales, THz radiation is a fantastic probe of low-energy electron dynamics in semiconductors. For instance the Drude conductivity is a strong function of frequency in the THz range, allowing N and τ to be determined from spectroscopic (transmission or reflection) measurements. Figure 5.1: The electromagnetic spectrum. 5.1 Terahertz time-domain spectroscopy Single-cycle pulses of terahertz radiation are generated by converting a femtosecond infrared pulse into a pulse in the THz range. This can be done by a number of methods including photoconductive excitation, as discussed in lectures, and optical rectification (the mixing of different IR wavelengths within a non-linear medium). Terahertz time-domain spectroscopy (THz-TDS) uses IR pulses from an ultrafast laser to generate and detect the electric field amplitude of the THz pulse ETHz . Since ETHz is detected as a function of time, instead of intensity I ∝ |ETHz |2 , the phase information contained in the electric field is preserved. This means that both the refractive index n and extinction coefficient κ (i.e. linked to the absorption coefficient α) can be determined, for instance by comparing the detected THz electric field pulse with and without the sample. 69 1 10 (a) |E(ν)| /Vm−1 E(t) /Vm−1 1000 500 0 −500 −5 0 0 10 −1 10 5 10 15 20 Time t /ps 0 (b) 1 2 3 4 5 Frequency ν /THz Figure 5.2: Transmitted THz pulse through vacuum (grey, thin lines) and an undoped Si substrate (red, thick lines). 5.1.1 Pulsed terahertz generation from biased semiconductors 6 5 4 3 2 1 0100 Amplitude (arb. units) Intensity (arb. units) To generate a pulse of THz radiation, a biased semiconductor device can be used. An infrared (800 nm) pulse with < 100fs duration (Fig. 5.3a) photoexcites a biased semiconductor, causing a transient pulse of current to flow. Because the photoexcitation causes a rise in the carrier density on similar timescales to the pulse duration, i.e. since the pulse envelope contains frequencies around 1 THz (see Fig. 5.3b), the current density will rapidly increase. For a dipole in the far-field the emitted electric field on-axis is E ∝ d2 p/dt2 ∝ dJ/dt, where J ∝ dp/dt = edx/dt using the definition of the dipole p = ex. 50 0 Time (fs) 50 100 (a) 600 500 400 300 200 100 0 100 101 102 Frequency (THz) 103 (b) Figure 5.3: Autocorrelation of a 50 fs, 800 nm laser pulse. (a) Time-domain. One optical cycle at 800 nm is 2.67 fs. (b) Fourier spectrum of the autocorrelation in (a). The envelope contains components at THz frequencies, while the frequency corresponding to 800 nm is 375 THz. Various geometries for photoconductive emitter exist (Fig. 5.4). A wide-area emitter (panel a) has a large gap, requiring a high voltage source. The infrared spot size is about 100 µm, i.e. comparable to the THz wavelength, meaning that the THz beam divergence is not too high. Narrow-gap (dipole) antennae (panel b) use lower voltages for the same applied field, but have greater beam divergence as a consequence of their sub-wavelength size. Interdigitated contacts (panel c) provide a compromise with the best of both designs. 70 (a) (b) (c) ~300 m ~10 m ~300 m w + + + + + + h + + + + + h w e e + + + + + + + + Figure 5.4: Pulsed THz generation from photoconductive emitters. 71