Arxiv:1602.01655V2 [Cond-Mat.Mes-Hall] 25 Apr 2016 Which Is Often Smaller Than 1

(Pat. pending, to appear in Comptes Rendus) . Linear and Nonlinear Mesoscopic Thermoelectric Transport with Coupling to Heat Baths

Jian-Hua Jiang1, ∗ and Yoseph Imry2, † 1School of Physical Science and Technology, Soochow University, 1 Shizi Street, Suzhou, Jiangsu, China 2Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot 76100, Israel (Dated: September 25, 2018) Decades of research on thermoelectrics stimulated by the fact that nano- and meso-scale thermoelectric transport could yield higher energy conversion efficiency and output power has recently uncovered a new direction on inelastic thermoelectric effects. We introduce the history, motivation, and perspectives on mesoscopic inelastic thermoelectric effects.

I. INTRODUCTION found lately to be less effective as, for example, the output power is suppressed in the zero band width limit[3] Research on the science and technology of thermo- due to, e,g., suppressed electrical conductivity[4]. electric phenomenon has a long history since its dis- Alongside with those developments, the scientific com- covery almost two centuries ago[1]. The modern the- munity was pursuing a better understanding of (charge) ory for thermoelectric transport in bulk semiconductors transport in nano- and meso-scale systems, localization was established, with the help of energy band theory and phenomena, and strongly interacting electron systems. semi-classical transport theory, in the middle of the last These studies activated research on thermoelectric trans- century[1]. The key concepts such as the figure of merit port in non-standard (including bulk) semiconductors, and the power factor were introduced then, which facili- which is still ongoing[5]. The most influential studies are tated and stimulated many theoretical and experimental based on two seminal works by Hicks and Dresselhaus[6], studies. It was found that the figure of merit (a measure who found that in reduced-dimension semiconductors, of the optimal thermoelectric efficiency in a material), such as quantum wells (2D), quantum wires (1D) and quantum dots (0D) heterostructures, the interrelation σS2T between the electrical conductivity, the Seebeck coeffi- ZT = (1) κ cient, and the thermal conductivity can be partially de- coupled. This is mainly because the density of states is limited by the following interrelated transport quanti- can be effectively enhanced and modulated (at differ- ties: the electrical conductivity σ, the Seebeck coefficient ent energies) when the spatial confinement along any di- S, and the thermal conductivity κ. The latter has contri- rection is small enough to induce significant quantum- butions from the electronic transport κe and other mech- confinement effects. Opportunities for enhancing the anisms (mainly from phonons in semiconducting materi- thermoelectric figure of merit ZT and the power fac- 2 als, κp), i.e., κ = κe + κp. Increasing S usually leads to tor S σ by engineering nanostructures and nanomate- reduced electrical conductivity σ. In addition, σ and κe rials thus emerge[7–11]. In addition, when low dimen- are interrelated. In metals, these two quantities mostly sional structures are packed up (densely) to macroscopic follow the Wiedemann-Franz law structures, the abundant interfaces effectively reduce the phonon heat conductivity[7, 8]. If the phonon heat con- κ = LT σ, (2) ductvity is reduced more significantly (e.g., by introduc- ing effective phonon scattering centers, or packing mate- 2 where L ≡ aL(kB/e) is the Lorenz number (aL ∼ 1, de- rials with mismatched phonon impedance) then the elec- pending on the material and the temperature). In semi- trical conductivity, the thermoelectric figure of merit can conductors, similar relations are usually approximately be improved[9]. This was demonstrated in BiTe quantum confirmed. When the Wiedeman-Franz law holds, the well superlattices and PbTe quantum dot superlattices[8]. 2 2 2 thermoelectric figure of merit is approximately S e /kB, In the latter a high density of quantum dots forming a arXiv:1602.01655v2 [cond-mat.mes-hall] 25 Apr 2016 which is often smaller than 1. regular array induces a large electronic density of states Mahan and Sofo proposed to improve the figure of in the PbSeTe alloy layer. Other methods, such as en- merit by using conductors with very narrow energy ergy filtering and semimetal-semiconductor transition, in bands, so that the thermal conductivity κe is reduced engineering the electronic density of states and the en- from the Wiedemann-Franz law[2]. However, this was ergy dependent conductivity are also introduced[10, 11]. The underlying physics is manifested in the Mott-Cutler formula[12, 13]

∗Electronic address: [email protected] hE − µi Var(E − µ) † S = , κ = σ (3) Electronic address: [email protected] eT e2T 2 where Var(E−µ) = h(E−µ)2i−h(E−µ)i2 is the variance II. BOUNDS AND QUANTIZED VALUES OF of the transport electronic energy; and the averages of THERMAL-ELECTRIC TRANSPORT powers of the electronic energy are defined as COEFFICIENTS (THIS SECTION IS DEDICATED TO JACOB BEKENSTEIN, A 1 Z HIGHLY ESTEEMED COLLEAGUE, SOME OF h(E − µ)ni = dEσ(E)(E − µ)n[−∂ f (E)], WHOSE EARLY WORK IS REVIEWED HERE) σ E F n = 1, 2, (4a) In this section, we discuss bounds (upper limits) on Z electric, thermal and thermoelectric transport coeffi- σ = dEσ(E)[−∂ f (E)]. (4b) E F cients. For the first two, these bounds can be reached in ideal conducting channels and give rise to “quantized” Here σ(E) is the energy dependent conductivity and values of these coefficients. E−µ It makes physical sense that maximal values of currents fF (E) = 1/[exp( ) + 1] is the Fermi distribution kB T function. The energy dependence of the conductivity carried in a pipe are proportional to the cross sectional σ(E) is the key quantity for engineering the thermoelec- area of that pipe. The same is valid for the transport of tric transport properties. It can be written as σ(E) = charge, energy and heat through a conducting wire. This is of interest by itself and because, as found by Whitney ekBT ρ(E)b(E) where ρ(E) is the density of states (DOS) and b(E) is the energy-dependent mobility. [14] these bounds also give the scales on which the maximal efficiency is reduced from Carnot. Nanostructures and nanocomposites have been demon- We start with the simplest case of charge transport, strated as effective ways to tune the carrier density of which is not the first example to have been discussed [18], states and the energy dependence of carrier scattering, as but the easiest to understand [19]. Consider a 1D wire well as to reduce the phonon thermal conductivity. These carrying electrical current, at a temperature T E methods lead to high thermoelectric efficiency and power F with EF being the Fermi energy. We imagine, following density for applications. Nevertheless, several nontrivial Landauer[20], that the strictly 1D wire (so narrow that aspects of mesoscopic electron transport have not been the lowest transverse excitation energy is larger than E exploited, which might give a chance to further enhance F plus several kBT , so only one transverse state is relevant) thermoelectric efficiency and power. These are: (1) non- connecting two reservoirs each at equilibrium, but with linear effects, (2) inelastic effects (due to strong carrier- slightly different electrochemical potentials µL and µR carrier interaction and carrier-phonon interaction), (3) and, at this stage, identical temperatures, T . The current thermoelectric transport with spatially separated electri- flowing between L and R is: cal and thermal currents. The main purpose of this review is to emphasize those aspects and their potential val- Z ues for improving thermoelectric performance, as well as I = e dE(fL − fR)v(E)ρ(E), (5) realization of thermoelectric diodes and transistors. We mainly focus on the inelastic effects and show that they v(E) and ρ(E) being the velocity and the density of states can offer an alternative (new) route toward high perfor- at energy E. fL and fR are the Fermi functions of L and mance thermoelectric structures that may reduce the ne- R. Since the 1D single-particle DOS, spin included, is cessity for novel functional materials. We further discuss given by nonlinear effects and the spatial separation of heat and 1 electrical currents in thermoelectric transport through in- ρ(E) = , (6) elastic thermoelectric transport. We will illustrate how π~v(E) these aspects may lead to better thermoelectrics. Several the velocity factors cancel. For linear response fL −fR = example systems are used to demonstrate the principles. 0 0 −eV f with f ≡ ∂Ef (f is the equilibrium Fermi distri- In the next section we consider bounds (which, when 2 bution), the integral trivially gives I = e V . Therefore, reached, are sometimes referred to as “quantized values”) π~ the conductance is given by of thermoelectric response coefficients. These bounds de- termine sometimes [14] the scales over which the maxi- e2 mal efficiency is reduced from Carnot. In section III, G = I/V = (7) π we consider linear and nonlinear transport above a bar- ~ rier. The latter is crossed by thermal activation, the This inconspicuous result is amazingly deep! The con- energy for which is taken from the (usually phonons, or ductance of an ideal 1D (“single channel”) wire is uni- the other electrons) associated heat baths. In section IV versal and totally independent on material and proper- we discuss a simple, quite generic, model [15, 16] for in- ties of the wire! No single-channel wire can have a larger elastic transport using a heat bath. Synergistic heat and conductance than an ideal one. So, this value is both electrical rectification and transistor effects in systems an upper limit and a so called “quantized value” of this with pronounced inelastic thermoelectric transport[17] is conductance. The upper limit is realized in the ideal reviewed in section V. We conclude and summarize in conductor (no scattering) limit. We shall soon see that a section VI. similar phenomenon occurs for the thermal conductance, 3

2 but before that we discuss the effect of a finite thickness phonon mode is K = (π/6~)kBT . This was first observed wire, comprising several channels. experimentally by Schwab et at [25]. If the wire has n transverse states below the Fermi Finally, we turn to the upper bound on the thermo- energy, there will be a continuum of longitudinal (par- electric coefficient of 1D systems. A systematic way to e2 get that would be to write the thermopower in terms of allel to the wire) states for each, which will create a π conductance. Thus, the total low-temperature conduc-~ the energy-dependent conductance G(E) and vary the e2 latter to maximize the former. However, there are Phys- tance will be n π , Similarly, if the temperature is higher than the separation,~ ∆, to a higher transverse state, its ical limitations on the behavior of G(E), which are not channel will be populated and it too will contribute an- obvious and need more work. A simpler approach is to 2 2 appeal to the requirement of stability – positive definite- other conductance quantum. Since ∆ ∼ ~ /ma where ness of the thermoelectric transport (Onsager) matrix Lˆ, a is the relevant dimension of the wire, the condition 2 i.e., L ≤ L11L22, using the above bounds on the diag- T ∆ (still, with T EF ) is equivalent to a λT . 12 −1 onal elements. This gives, (Similarly, ∆ EF can also by described by a kF ). Thus, the number of channels, n, in a wire is on the p e −2 2 L12 ≤ L11L22 = √ kBT, (10) order of the number of elementary kF or λT areas at 3~ zero or higher temperatures, respectively. At the higher as the upper bound for the coefficient L for spin- 1 1D temperatures, the measured conductance, related to the 12 2 electron systems. As for the thermopower, it is given number of channels, is a quantum property, dependent by L /(T L ) and it can in principle be as large as we on ! This has been mentioned by Whitney [14] in the 12 11 ~ please, for small enough electrical conductance L . As context of thermal conductivity, which is the subject we 11 Eq. (3) shows, however, when a large enough energy is shall discuss shortly. transferred by each electron, S is large. But then the out- The quantized conductance has been observed experi- put power is (usually, exponentially) small as well (due mentally, concurrently by two groups [21], in suitable bal- to a small electrical conductance). listic quantum point contacts in GaAs, within a percent accuracy, about two years after the prediction. More recently, it became a standard item in mesoscopic Physics. III. ACTIVATED TRANSPORT ABOVE A Now we discuss the electronic thermal conductance, BARRIER to that end we assume that the two Landauer reservoirs have the same electrochemical potentials and different A. linear transport temperatures TL and TR. (in linear response, the effects of TL − TR and µL − µR are simply additive). We want [22] the heat current between L and R. Thermodynamics Here we consider a very simple system where the elec- tells that an electron of energy transferring from L to R tronic transport is effectively funneled to a narrow band. carries heat given by [23, 24] E − µ = TS. Thus, the We imagine an ordinary barrier in one dimension (1D, R generalized later), depicted in Fig. 1. The barrier is cho- heat current, dE(fL − fR)(E − µ)v(E)ρ(E) between L and R, is given by sen so that its height W (measured from the averaged chemical potential µ) and thickness, d, satisfy 2 √ Z 0 (E − µ) J = (T − T ) dE(−f ) v(E)ρ(E), (8) W kBT ; d ~/ 2mW , 1/kF . (11) Q L R T The second inequality is more strongly obeyed than the 0 (E−µ) first, so that the dominant transport is via thermal acti- where we used fL − fR = (−f ) T (TL − TR). We now R ∞ vation above the barrier and not by quantum-mechanical use the Sommerfeld expansion (see, e.g. [23]) 0 dE(E− 2 0 2 2 tunneling. We assume that the barrier is tapered (see µ) (−f ) = (π /3)(kBT ) , to find Fig. 1) so that the transmission through it changes rather quickly from 0 to 1 when the electron energy E increases K = (π/3 )k2 T (9) ~ B through W . This is certainly the case in the 1D clean This is the upper bound on the electronic thermal tunnel junction of the type discussed here, or in the quan- conductance per channel, or the “quantum of ther- tum point contact [19, 26]. Its validity in a high dimen- mal conductance” (for two spin directions). Not sur- sional system will be confirmed later on. When some prisingly (both based on the Sommerfeld expansion), disorder exists, rendering the electron motion diffusive, it satisfies the Wiedemann-Franz relationship, K = it makes sense that the transmission still changes from 0 2 2 e2 to 1 when the electron energy increases through W . This (π /3)(kB/e) TG, with the conductance quantum . π~ increase may become slower than in clean systems, but Somewhat more surprising is the situation with the that should not change the qualitative behavior. phonon thermal conductivity. The two differences are A significant feature of our setup is that the electronic that here heat and energy are identical (µ = 0) and that heat conductance can be made to be very small while the the Bose function should be used, instead of the Fermi thermopower stays finite one. As noticed already by Bekenstein [18] the result 0 2 2 0 is nevertheless the same. The conductance quantum per Ke = Ke − GS T Ke , (12) 4 where G is the conductance, S is the thermopower, Ke currents are is the electronic thermal conductance (defined at vanish- ∞ 0 2e Z ing electrical current) while Ke can be termed a “bare” Ie = dET (E)[fL(E) − fR(E)], thermal conductance (defined at vanishing electrochemi- h 0 Z ∞ cal potential difference). As mentioned above, according e 2 IQ = dEET (E)[fL(E) − fR(E)], (14) to Ref. [2], the largest two-terminal figure of merit is h 0 2 2 achieved in systems with the smallest Ke/(GS T ). Here this ratio is very small, and then ZT is mainly limited and hence by the phonon heat conductivity Kp between the two 2 Z ∞ 2e −1 metallic contacts [see Eq. (17) below]. Kp can be small G ' dE(kBT ) f(E)[1 − f(E)], in nanosystems[7, 8]. Our system is then expected to pos- h W Z ∞ sess a high figure of merit. Another way to understand 2e −1 L1 ' EdE(kBT ) f(E)[1 − f(E)], the situation (shown by the formulae below) is that S h W is the average energy transferred by an electron, divided 2 Z ∞ 0 0 2 −1 by eT , while Ke /G is the average of the square of that Ke ' E dE(kBT ) f(E)[1 − f(E)], (15) 2 h W energy, divided by e . Therefore Ke/G is proportional to the variance of that energy. The latter obviously van- with f(E) = 1/[exp( E ) + 1] being the equilibrium ishes for a very narrow transmission band. In this case kB T Fermi distribution. It reduces for W kBT to the the transmission band is the range of a few kBT above Boltzmann distribution and then f(E)[1 − f(E)] ∝ W . Thus, it is not surprising that Ke is of the order of E 2 exp(− k T ). Therefore, (kBT ) . B 2 2 2 hEi = kBT (xw + 1), hE i − hEi = (kBT ) , (16)

with xw ≡ W/(kBT ). From the transport coeﬃcients one readily obtains the electronic ﬁgure of merit:

Ke ZT = (ZT )e , (17) Ke + Kp 2 2 hEi 2 W (ZT )e = 2 2 = (xw + 1) = + 1 . FIG. 1: (Color online) The suggested device: a long (d hE i − hEi kBT characteristic tunneling length) and high (W kB T ) bar- (18) rier separating two electron gases. The transferred electron gets an energy W ± O(kB T ) from the LHS thermal bath and In two and three dimensions (d = 2, 3), the calculation deposits it in the RHS one. proceeds similarly. The energy is the sum of the longitudinal part which goes over the barrier and the transverse part which should be integrated upon. The latter has the The thermoelectric linear transport problem is fully usual density of states in d−1 dimensions. This is E−1/2 characterized, for this two-terminal situation, by: for d = 2 and constant for d = 3. The overall factors do not matter for ZT . The final result differs from Eq. (18) just by numerical factors. At d = 2, the result is: Ie GL1 δµ/e = 0 . (13) Ie 0 2 Q L1 Ke δT/T 2 W 3 (ZT )e = + (19) 3 kBT 2 where I is the charge current and Ie is the heat cur- e Q The result for d = 3 is rent, δT = TL − TR, and δµ/e ≡ (µL − µR)/e ≡ V is the voltage between the left and right terminals. The 2 × 2 1 W 2 matrix contains the regular conductance G, the bare elec- (ZT )e = + 2 . (20) 0 2 kBT tronic thermal conductance Ke , and the (off-diagonal) 0 thermoelectric coefficients L1 and L1. That the latter The large order of magnitude of ZT remains. The ad- two are equal is the celebrated Onsager relation (valid vantage of the last, d = 3, case is twofold: it is easier to for time-reversal symmetric systems). We remind the make (two thick layers of the conducting material, sep- readers that S = L1/(TG). All currents and transport arated by an appropriate barrier) and the total current coefficients in Eq. (13) are given in 1D in terms of the for given W , and hence the power, is proportional to the energy-dependent transmission of the barrier, which we cross-section of the device. take as T (E) ' Θ(E − W ). We measure all energies A vacuum junction has too high a barrier for most ap- from the common chemical potential µ (i.e., µ ≡ 0). The plications [27]. The ballistic quantum point contact is a 5 very effective realization of the model, when biased in the has to cross (setting e ≡ 1 in this section). This is rather pinch-off regime and in the region where activated con- high, especially for such a simple device. The cooling duction is dominant. It requires, however, rather high power is IW , I being the current. Ways to reduce the technology and can handle only small powers. We be- effect of the phonon thermal conductance are suggested. lieve that the two more realistic straightforward ways to effectively achieve the requirements of the model, are:

1. A metal-semiconductor-metal junction, with a properly chosen diﬀerence between work functions

and electron aﬃnities (an example might be Au−p- W Si−Au). A large area will increase the power of the

device. V 2. A superlattice [28] separating the two metallic elec- trodes, where the Fermi level is inside the gap between the “valence” and a “conduction” mini-bands FIG. 2: Schematics of a barrier (W) with a large bias (V) of the superlattice. Electron-hole symmetry should be strongly broken either by intrinsic lack of band We consider a wide (to neglect tunneling) solid-state or vacuum, electronic barrier, of height W , bridging two symmetry or by the placement of EF away from the middle of the gap. Obviously, highly-doped semiconductors. It is idealized, and taken as biased by a volt- conductors can be substituted for the metallic elec- age V , where W & V kBT , such that all electronic trodes. conduction is from the left- to the right- terminal. We have in mind W ∼ eV . Now, even if we keep the LHS The last remark brings us to the issue of semiconduct- (RHS) at temperatures Tc (Th), where Tc < Th, the de- ing systems. Imagine first an intrinsic small-gap semi- vice may transfer heat between the cool left terminal and conductor, such as BiTe, the current work-horse of ap- the warmer right one; i.e we have a (thermoelectric) re- plied thermoelectricity. The first thought would be that frigerator. That heat per transported electron is given in since conductivity is thermally activated, with effectively the ideal case roughly by W − V ∼ W & V [27]. This is a Boltzmann distribution at temperatures much below because an electron has to borrow an energy of about W the gap, it is a simple realization of our model. This to cross the barrier from the LHS heat bath, it then re- would be the case only if electron-hole symmetry will turns it to the RHS heat bath. Thus, the cooling power is be strongly broken. If that symmetry prevails, the elec- IW and the invested power is IV , where I is the current. tron and hole thermopowers will cancel. That symme- Therefore, the efficiency and power are try breaking can be achieved by judicious selective dop- ing and alloying, and seems to dominate the art of the η = W/V, P = IW (21) present manipulations of BiTe and its derivatives. How- where we neglected the “parasitic” phonon thermal con- ever, what is proposed here is a different approach: let a duction back from right to left and took the energy price larger band-gap semiconductor, even Si, Ge, graphite, or per transferred electron with the bias as V . We re- a member of the GaAs family, bridge two layers of metal mark that this is not a linear response (small V and so chosen that their Fermi level is significantly closer to δT ≡ Th − Tc) calculation. It is valid for finite driving either the conduction or the valence band. In a large forces, as long as V is the self-consistent voltage. temperature regime this will, as discussed in the above, By essentially reversing the process (i.e., employing realize our activated junction model. We add, finally, temperature differences to generate electrical energy), that a yet another way to realize the model is via such a one may use this device for energy harvesting as well. semiconductor, e.g., n-doped so that the Fermi level is, Here we give a proof, within the linear transport ap- say, (3 ∼ 10)kBT below the conduction band and much proximation, that the W/V efficiency does not exceed the further from the valence band. The conduction band e Carnot one. Write the heat current as IQ = −KδT/T + should then play the role of our activation barrier. Some LV , where L = SGT with S being the Seebeck coeffi- experimentation should show which of these schemes is cient, K and G are effective thermal and electrical con- superior. ductances. The “working condition” is that to refriger- ate, V > (K/L)δT/T ⇒ V/W > K/(SGW T )δT/T = K/(S2GT 2)δT/T (Remembering that ST is the average B. Nonlinear activated transport above a barrier heat transferred by an electron, S = W/T ). Now we use the stability of the Onsager matrix: K/G > (ST )2, to Here we show that a relatively high barrier, W, in the get W/V < T/δT = ηc . ηc is the Carnot efficiency (“fig- nonlinear regime, W & V kBT , is a rather efficient ure of merit”) for a refrigerator. Thus ηc is indeed the thermoelectric device. Neglecting return currents and upper limit on the efficiency. phonon parasitic thermal conductance, the efficiency is The efficiency obtained is quite high. Even pushing the found to reach W/V , where W is the barrier the electron voltage up to, say, W/2, we get 2 in the ideal case. If the 6 idealizing factor of negligible phonon heat conductance Wohlman, Imry, and Aharony[29] for molecular junc- decreases the efficiency by a factor of two, a value of 1 is tions, and later by Sánchez and Büttiker[30] for capaci- still respectable for such a simple device. It is comparable tively coupled double quantum dots. For the former, the to the values for current thermoelectric devices having additional energy is provided by a phononic heat bath; the parameter ZT =∼ 1. Of course, we have the usual for the latter, the additional energy is provided by a ca- conundrum between power and efficiency. pacitively coupled electronic reservoir. Entin-Wohlman, As far as the negative effect of the phonon thermal con- Imry, and Aharony[29] pointed out, based on thermody- ductance, besides decreasing its relative influence using a namic arguments, that the full description of thermoelec- larger V (decreasing the phonons’ relative contribution, tric transport in such three-terminal system must include see below) it is suggested to use a soft metal on one side two heat currents (one is the electronic heat current from of the barrier and a hard semiconductor or metal on the source to drain and the other is the heat current from the other side, to reduce the phonon thermal conductance third terminal), coupled with a single electrical current. via acoustic impedance mismatch. We also reemphasize Such a property of three-terminal mesoscopic system im- that the phonon conductance is largely independent of mediately leads to two correlated thermopowers (i.e., the the applied voltage, so that larger values of the latter are electrical current can be induced by two temperature dif- favorable also in that respect. ferences that contribute additively in the linear-response regime). The merit of such an effect has not been fully appreciated until recently in Refs. [16, 35]. Another im- IV. THE INELASTIC THERMOELECTRIC portant property is that the direction of heat current TRANSPORT ASSISTED BY A HEAT BATH: from the third terminal is different from the direction LINEAR TRANSPORT of electrical current. This means spatial separation of electrical and thermal currents[15, 29–32], which allows In true inelastic transport processes energy has to be independent tuning of electrical and thermal conduction, absorbed/emitted from heat baths to compensate for the and achieve high thermoelectric figure of merit and power energy difference between the initial and final states. factor[16, 35]. This energy can come in the form of phonons, plas- As in many situations, the above findings share some mons, charge (spin) fluctuations or other collective ex- similarities with some earlier works. In 1993, Edwards citations. In bulk materials, those collective excitations et al.[33] proposed to use inelastic transport to cool an usually thermalize quickly with electrons. Therefore, the island of electrons, which can be useful for sub-Kelvin local temperature of those collective excitations is nearly cooling of the electron gas following conventional cooling the same as that of the electronic ensemble, unless the method[34]. Nevertheless, the physics uncovered in re- system is out of the linear-response regime. Interaction cent studies of mesoscopic inelastic thermoelectric effect with phonons and other collective excitations may mod- goes far beyond a novel method of cooling[15, 17, 35–40]. ify thermoelectric transport coefficients via, e.g., phonon The remaining of this paper will present a pedagogi- drag effect, which is here an unessential quantitative cor- cal review of the essential aspects using a simple picture rection. which is depicted in Fig. 3. In this case electrons are In mesoscopic regimes, the abundance of interfaces and transported, under certain bias, from source to drain. disorder may significantly reduce the thermalization pro- The dominant path for such transport is jumping elas- cesses. Substantial temperature or electrochemical po- tically to a quantum dot (or a localized energy level) tential differences can be maintained between adjacent with energy E1 and then hopping inelastically to another nanoscale regions. Thus inelastic and nonlinear trans- quantum dot with energy E2, and finally tunneling into ports become significant and prevail. Moreover, geo- the drain. Because the energies E1 and E2, do not match, metric configurations that support inelastic transport are electrons have to exchange energy with a heat bath. This richer. It turns out that the latter opens an important heat bath can have a temperature Tp differing from the direction for improving thermoelectric performance that temperatures of the source (TL) and drain (TR). has not been fully explored so far. The fact that meso- In this picture, there are three reservoirs: the source, scopic thermoelectric transport can have significant spa- the drain, and the heat bath. Their states are charac- tial separation of electrical and thermal currents, partic- terized by their temperatures and chemical potentials. ularly in inelastic transport processes, enables disentan- The heat bath can be electronic or phononic (or consist- glement of electrical and thermal conductivity and pro- ing of another type of bosons that carry energy). With- vides new opportunities for improving the thermoelectric out loss of the essential physics, we focus on the situa- figure of merit. tion that there is no electrical charge flowing out of the The simplest nontrivial geometry configuration for in- heat bath. Under this assumption we can treat the heat elastic transport processes is the three-terminal set-up bath as bosonic. In this regime, the relevant thermody- where energy absorbed/emitted by the electron comes namic variable of the heat bath is just its temperature from a third heat bath differing from the source and Tp. The elementary excitation that provides energy to drain [see Fig. 3]. Inelastic thermoelectric transport in electron transport can be phonons, charge fluctuations three-terminal geometry was first examined by Entin- (i.e., charge density waves or plasmons)[36, 41, 42], spin 7

fluctuations (i.e., spin density waves or magnons[43]), For the double quantum dots system, when inelas- photons[44], etc. This energy is sometimes termed as tic transport dominates, the transport coefficients in energy gain in the literature[30]. Eq. (27) can be written as, The energy and charge flows into the reservoirs are 2 defined as the time derivatives of their total energy E ω E and charge of the reservoirs, E˙ (i = L, R, p) and eN˙ L1 = G ,L2 = G ,K1 = G , i i e e e2 (i = L, R) where L/R/p stand for source/drain/heat Eω ω2 bath, respectively. Ni is the total electron number of K12 = G ,K2 = G , the i reservoir and e < 0 is the charge of an electron. To- e2 e2 tal energy and charge conservation gives P E˙ = 0 and E1 + E2 i i E ≡ , ω = E2 − E1. (29) P ˙ 2 i Ni = 0. Therefore, there are only three independent currents, which can be organized into two heat currents e2 The conductance is G = Γ1→2 where Γ1→2 is and one electrical current, kB T the inelastic transition rate between the two quantum 1 dots[15, 17]. We have assumed here that the coupling Ie ≡ (Q˙ − Q˙ ),Ip ≡ −Q˙ ,I ≡ eN˙ , (22) Q 2 R L Q p e R between the QD 1 and the source as well as that between the QD 2 the drain is much stronger than the coupling where Q˙ i = E˙ i − µiN˙ i for i = L, R and Q˙ p = E˙ p. between the two QDs[15, 17]. The thermodynamic affinities conjugated to those three currents satisfy the following relation[45],

˙ e p Source Drain Stot = IeA1 + IQA2 + IQA3. (23) E2 We found that[17] these conjugated affinities are µ µ L E1 R T T µL − µR 1 1 1 1 L R A1 = ( + ),A2 = − , e 2TL 2TR TR TL 1 1 1 A3 = + − . (24) 2TL 2TR Tp Heat bath Tp The phenomenological Onsager transport equation is then, FIG. 3: Schematic of three-terminal inelastic thermoelectric X I = M A + O(A2), (25) mesoscopic systems. Electrons transport from source to drain i ij j have to absorb energy from the heat bath. There are three j reservoirs: source, drain, and the heat bath. They can have different temperatures. where Mˆ is the 3 × 3 thermoelectric response matrix. In the linear response regime, The conventional thermoelectric effect is described by µ − µ δT Eq. (25) for A3 = 0, i.e., the temperature of the heat A ' L R ,A ' , δT ≡ T − T , 1 eT 2 T 2 L R bath is equal to the average temperature of the source ∆T 1 and drain [more precisely, when Tp = 2TLTR/(TL +TR)]. A3 ' , ∆T ≡ Tp − (TL + TR), (26) Thus, S is the conventional thermopower. In contrast, T 2 2 Sp is closely related to the inelastic thermoelectric trans- where T is the equilibrium temperature. And port: the temperature of the heat bath comes into play only for the inelastic thermoelectric effect. Moreover, the   p GL1 L2 direction of heat flow out of the heat bath, IQ, can be Mˆ = T L K K (27) e  1 1 12  different from the heat flow IQ (we shall term this as L2 K12 K2 “electronic heat current”). In the conventional thermoelectric effect, the electrical where G is the conductance, K , K , and K are the 1 2 12 current flows parallel to the heat current, which leads diagonal and off-diagonal heat conductances. The two to strong correlation between electrical conductivity and thermopowers are thermal conductivity. This has been found to be a key obstacle that limits the figure of merit[4]. More explicitly, L1 L2 S = ,Sp = . (28) the conventional thermoelectric figure of merit here is TG TG given by Particularly, Sp enables the possibility of cooling of the heat bath by an electrical current[39], or the cooling of GS2T ZT = 2 . (30) the drain by a hot heat bath[40]. K1/T − GS T 8

For a slab of material with area A and thickness l, the In our system, besides the conventional thermoelectric electrical conductivity is σ = Gl/A and the thermal con- effect, there is an additional thermopower due to inelas- 2 ductivity is κ = (K1/T − GS T )l/A. Hence the above tic transport, Sp. Using this effect, waste heat from the equation recovers the well-known Ioffe’s figure of merit heat bath can be converted to useful electrical power (re- ZT = σS2T/κ. versely, electrical power can be used to cool the heat bath The correlation between different transport coefficients without passing electricity to it). The figure of merit of can be revealed by considering many parallel transport such thermoelectric energy conversion was first derived channels. For the present model, these are many double by Jiang, Entin-Wohlman, and Imry[15, 35]: quantum dots systems that conduct electricity and heat between the source and the drain. That is, GS2T 2 2 0 ˜ p hωi e K2 ZT = 2 = , ∆2 = . Z Z K2/T − GSp T Var(ω) + ∆2 G G = ρ(Ei)dEi ρ(Ef )dEf G(Ei,Ef ), (35) The advantages of the inelastic thermoelectric effect are L1 hE − µi L2 hωi S = = ,Sp = = , manifested in the above equation: (1) High figure of merit TG eT TG eT demands small variance of ω (i.e., the energy of collective Var(E − µ) Var(ω) excitation) in the heat bath, which can be disentangled K = G + K0 ,K = G + K0 , 1 e2 1 2 e2 2 from electronic band structures! (2) Hence the conduc- h(E − µ)ωi tivity G can be uncorrelated with the variance of ω (This K = G + K0 , (31) 12 e2 12 also opens the opportunity for high output power if G is large). (3) Sharp DOS peaks in the spectrum of collective where Ei is the energy of the electron when it leaves the excitations can reduce thermal conductivity due to such source, Ef is its energy when it enters the drain, excitations (i.e., reduce the denominator in (35)). This also promotes the inelastic scattering rates and hence the 1 E = (Ei + Ef ), ω = Ef − Ei, (32) electrical conductivity and the output power. (4) Spatial 2 separation of electrical and heat currents enables engi-

ρ(E) is the density of states, G(Ei,Ef ) is the (initial and neering of electrical and heat transport more indepen- final) energy-dependent conductivity, and the average is dently. For the inelastic transport to give considerable defined as output power and to dominate over the elastic trans- Z Z port, the coupling to the bosons has to be very strong −1 hOi = G ρ(Ei)dEi ρ(Ef )dEf G(Ei,Ef )O. (33) and the elastic transport have to be suppressed by an energy barrier. The heat conductivities from non-electronic processes are The advantages of the inelastic thermoelectric effect 0 0 are explored in Ref. [35] by considering a realistic p-n also included in the above. They are denoted as K1, K2, 0 junction thermoelectric device. As shown in Fig. 4, a and K12, for diagonal and off-diagonal heat conduction. We can then express the figure of merit as, pair of an electron and a hole is generated by absorbing a phonon from the hot phonon reservoir. The de- 2 2 0 hE − µi e K1 vice has similarities with a solar cell, except that it ex- ZT = , ∆1 = . (34) ploits heat energy in the form of phonons. The same Var(E − µ) + ∆1 G paradigm can be applied for heat in the form of other The above equation, although derived from our model, bosons, as long as the coupling between the electron and 0 also applies to normal thermoelectric materials where K1 the bosons is strong. For the case of phonons, the cou- comes from the parasitic phonon thermal conductivity. pling between electrons and optical phonons or accoustic Mahan and Sofo have argued that the figure of merit can phonons near van Hove singularities is very strong[35, 46]. be significantly improved in materials with very small Studies in Ref. [35] indicate that, beside material engi- electronic bandwidth W . In that limit, Var(E − µ) ∼ neering, thermoelectric efficiency and power can be sig- 2 2 W (kBT ) , the figure of merit is expected to be much nificantly improved via structure and geometry engineer- 2 2 larger than unity if hE − µi ∼ (kBT ) . However, in the ing. Later, inelastic thermoelectric effects based on quan- W → 0 limit, the output power vanishes[3]. In a recent tum dots[16, 37] and quantum wells[38] have been stud- study[4], it is argued that for narrow electronic band- ied. Considerably large output power and optimal effi- width materials, electron mobility is suppressed by the ciency have been found with realistic quantum dot/well large effective mass and strong back scattering. The sup- parameters[16, 37, 38]. In Ref. [16] the phonon ther- pressed electrical conductivity G then makes the factor mal conductivity is included using experimentally mea- ∆1 dominate the denominator. In such a regime the fig- sured values. It was found in Ref. [16] that, although ure of merit is suppressed, rather than promoted. Real- the phonon thermal conductivity reduces both figures of istic optimization of the figure of merit needs to balance merit, ZT and ZT˜ , the inelastic thermoelectric figure of the variance and the factor ∆1[4]. The correlation be- merit ZT˜ is more robust than the elastic one ZT . In tween electrical and thermal conductivity limits further addition, ZT˜ is usually considerably larger than ZT for improvement of the figure of merit[4]. the quantum dots array system considered in Ref. [16]. 9

In Ref. [17] it has been shown that combining thermal and electrical degrees of freedom together, several new nonlinear functions can be realized where heat and electricity have synergistic effects. For example, charge rectification can be induced by a temperature gradient, heat rectification can be realized by voltage bias. To describe those phenomena, we expand the currents to second order in affinities,

X X 3 Ii = MijAj + LijkAjAk + O(A ). (36) j jk The first term on the right hand side describes the linear responses, whereas the second term gives the lowest order nonlinear responses. The functionalities represented by various second order coefficients are summarized in Table I[17]. Important examples include charge rectification induced by the temperature difference ∆T , i.e., the difference between phonon and electronic temperature, as well as phonon-thermoelectric transistor effect, i.e., tuning the Ie-V or the Ie-δT curve by the phonon temperature. The Mij and Lijk coefficients for a double quantum-dots system are calculated using realistic material parameters. The results in Ref. [17] indicate that FIG. 4: (a) Structure of p-n junction three terminal thermoelectric engine. (b) Energy diagram: Absorbing a hot phonon the nonlinear effects are considerable and can be mea- generates a pair of electron and hole which drift along the sured using the state-of-the-art experimental techniques. built-in electric field of the junction to induce a current, simi- These observations indicate that synergy of electronics lar to photo-voltaic effects. From Jiang, Entin-Wohlman, and and phononics can offer new platforms and opportuni- Imry[35]. ties for high performance information nanotechnologies.

TABLE I: Functionality of second-order coefficients Moreover, the output power for the inelastic thermoelectric effect is significantly greater than that of the elastic Terms (Lijk) Diode or Transistor effect thermoelectric effect[16]. The figure of merit can reach L111 charge rectification a decent value of ZT˜ ∼ 1 using quantum dot arrays em- L222, L333 electronic and phononic heat rectification bedded in polymer thin films[16] or using HgCdTe p-n L233, L322 off-diagonal heat rectification junctions[35]. These results confirm the advantages of L122, L133 charge rectification by δT or ∆T the inelastic thermoelectric effect. L211, L311 heat rectification by voltage

L113, L123 phonon-thermoelectric transistor L , L other nonlinear thermoelectric effects V. RECTIFICATION AND TRANSISTOR 212 112 EFFECTS IN THE NONLINEAR REGIME FOR INELASTIC TRANSPORT Beyond those observations, it is found that the inelastic three-terminal transport can induce thermal transistor effect in the linear-response regime without relying on In this section we show how inelastic three-terminal the onset of negative differential thermal conductance[17]. transport enhances nonlinear devices like diodes and The underlying physics can be revealed using thermody- transistors. Diodes and transistors are crucial elements namic arguments, without an explicit microscopic model. for electronics and conventional information technology. Consider the restrictions on thermal conductance in There is increasing interest to invent diodes and tran- the linear-response regime imposed by the second law of sistors of heat for information technology. Using heat thermodynamics in a three-terminal set-up for thermal and electricity together to achieve information process- transistor functionality. If the thermal transport mecha- ing may have energy advantages[47]. This idea was first nisms involve only two terminals (reservoirs) in each mi- explored in spin caloritronics which tried to exploit heat croscopic process, then we can write the thermal trans- to manipulate magnetic domain walls or induce spin cur- port equations as rents, magnon currents, that may help to reduce the energy cost compared to using only electrical currents IL→R = KLR(TL − TR)/T, (37a) to achieve the same goal[47, 48]. A goal of pursuing I = K (T − T )/T, (37b) the use of heat alone for information technology, namely L→p Lp L p “phononics”, was also explored in the past decade[49]. IR→p = KRp(TR − Tp)/T, (37c) 10

where the subscripts L, R, p denote the source, drain, When |E1 − µ| > |E2 − E1| (e.g., E1 > E2 > µ) then α and heat bath, respectively, and T is the equilibrium tem- can be greater than 1 and there can be heat transistor perature. The second law of thermodynamics requires eﬀect in the linear response regime. that VI. CONCLUSIONS

KLR ≥ 0,KLp ≥ 0,KRp ≥ 0. (38)

However, due to energy conservation, there are only Although the development of bulk semiconductor ther- two independent heat currents, which we can choose as moelectric materials may be limited by material parameters, the study of mesoscopic thermoelectric effects is L p IQ = IL→R + IL→p,IQ = Ip→L + Ip→R. (39) still far from from maturity. It has given significant in- sights for improving thermoelectric efficiency and power Then we can cast the thermal transport equation into in nanostructured thermoelectric materials (one of the the form, main directions in thermoelectric research). We point out ! ! ! several aspects that mesoscopic systems may help with IL K K TL−TR improving thermoelectric performances: inelastic trans- Q = L o T . (40) p Tp−TR port, nonlinear effects, and spatial separation of ther- IQ Ko Kp T mal and electrical currents in thermoelectric transport. where We also discuss bounds and quantization of thermal and thermoelectric transport coefficients in 1D single chan- KL = KLR + KLp, nel conductors. Channel number and the dependence on it are considered as well. We then discuss activated Kp = KLp + KRp,Ko = −KLp. (41) transport above a barrier in both the linear and nonlin- The transistor current amplification factor is defined as ear regimes and show their relevance for thermoelectric applications. We emphasize, particularly, inelastic ther- L ∂T I moelectric effects in three-terminal geometry, which has α ≡ p Q (42) ∂ Ip attracted a lot of research interest recently. The physics Tp Q and merits of the three-terminal inelastic thermoelectric effect are discussed and demonstrated for both linear and In the linear-response regime one usually finds nonlinear transport. This paper attempts to give a peda- gogical review of the research frontiers of thermoelectric Ko α = < 1. (43) inelastic transport effects in mesoscopic physics. Recent K p experimental progress[41, 42] is pushing the field forward. Thus no thermal transistor effect exists in the linear- response regime. However, the above analysis missed an important type of thermal transport mechanism that involves simulta- neously three terminals (reservoirs) in each microscopic process. A typical example is the inelastic transport process illustrated in Fig. 3. For this type of transport, the thermal transport equation can also be written in the form of Eq. (40). However, the second law of thermodynamics only requires that ACKNOWLEDGEMENTS 2 KL ≥ 0,Kp ≥ 0,KLKp ≥ Ko , (44) which does not forbid α = Ko > 1. A realistic example Kp We thank Hamutal Bary-Soroker, Ora Entin- that achieves α > 1 in the linear-response regime can be Wohlman, Zvi Ovadyahu, Dan Shahar, Moty Heiblum, found in Ref. [17]. This can also be illustrated by con- Eli Zeldov and Dorri Halbertal for useful discussions. sidering the double quantum dot model, using Eq. (29). L e 1 p JHJ acknowledges support from the faculty start-up Keeping in mind that IQ = IQ − 2 IQ [using Eq. (22) for funding of Soochow University. YI acknowledges support µL = µR], one has from the Israeli Science Foundation (ISF) and the US- Israel Binational Science Foundation (BSF). We thank |E − µ| α = 1 . (45) the referee and Robert Whitney for pointing out an er- |E2 − E1| ror in the previous version of this paper. 11

[1] T. C. Harman and J. M. Honig, Thermoelectric and ther- (out of print). 2nd edition, 2002). momagnetic effects and applications (McGraw-Hill, New- [20] R. Landauer, Phil. Mag. 21, 863 (1970). York, 1967); H. J. Goldsmid, Introduction to Thermo- [21] B. J. van Wees, H. van Houten, C. W. J. Beenakker, J. electricity (Springer, Heidelberg, 2009). G. Williamson, L. P. Kouwenhoven, D. van der Marel, [2] G. D. Mahan and J. O. Sofo, Proc. Natl. Acad. Sci. and C. T. Foxon, Phys. Rev. Lett., 60, 848 (1988). (USA) 93, 7436 (1996). D. A. Wharam, T. J. Thornton, R. Newbury, M. Pepper, [3] O. Entin-Wohlman, J.-H. Jiang, and Y. Imry, Phys. Rev. H. Ahmed, J. E. F. Frost, D. G. Hasko, D. C. Peacock, E 89, 012123 (2014). D. A. Ritchie, and G. A. C. Jones, J. Phys. C: Solid State [4] J. Zhou, R. Yang, G. Chen, and M. S. Dresselhaus, Phys. Phys. 21, L209 (1988). Rev. Lett. 107, 226601 (2011); C. Jeong, R. Kim, and [22] The thermal conductance is defined under the condition M. Lundstrom, arXiv:1103.1274. of zero current, not zero voltage. This gives a (usually [5] E.g., J.-H. Jiang, O. Entin-Wohlman, and Y. Imry, Phys. small) thermoelectric correction to the “bare” thermal Rev. B 87, 205420 (2013); R. Bosisio, C. Gorini, G. conductance K0 that we calculate. We shall ignore this Fleury, and J.-L. Pichard, New J. Phys. 16, 095005 correction in the present section. (2014); R. Bosisio, C. Gorini, G. Fleury, and J.-L. [23] J. M. Ziman Principles of the Theory of Solids, Chap. 7, Pichard, Phys. Rev. Applied 3, 054002 (2015). cambridge (1969). [6] L. D. Hicks and M. S. Dresselhaus, Phys. Rev. B 47, [24] Y. Imry and A. Amir, The localization transition at fi- 12727 (1993); ibid., 47, 16631 (1993). nite temperatures: electric and thermal transport, in: 50 [7] R. Venkatasubramanian, Phys. Rev. B 61, 3091 (2000); Years of Anderson Localization, E. Abrahams, ed., chap- J.-K. Yu, S. Mitrovic, D. Tham, J. Varghese, and J. R. ter 9, pp 191-213 (World Scientific, Singapore, 2010), or Heath, Nat. Nanotechnol. 5, 718 (2010); R. Venkatasub- arXiv:1004.0966. ramanian, E. Siivola, T. Colpitts, and B. O’Quinn, Na- [25] K. Schwab, E. A. Henriksen, J. M. Worlock, and M. L. ture 413, 597 (2001). Roukes, Nature (London) 404, 974 (2000). [8] A. I. Boukai et al., Nature 451, 168 (2008); B. Poudel [26] N. Nakpathomkun, H. Q. Xu, and H. Linke, Phys. Rev. et al., Science 320, 634 (2008); P. Pichanusakorn and B 82, 235428 (2010). P. Bandaru, Mater. Sci. Eng. R-Rep. 67, 19 (2010); A. [27] G. D. Mahan, J. Appl. Phys. 76, 4363 (1994); G. N. J. Minnich, M. S. Dresselhaus, Z. F. Ren, and G. Chen, Hatsopoulos and E. P. Gyftopoulos, Thermionic Energy Energy Environ. Sci. 2, 466 (2009); C. J. Vineis, A. Shak- Conversion (MIT, Cambridge, 1973), Vol. 1. ouri, A. Majumdar, and M. G. Kanatzidis, Adv. Mater. [28] D. Vashaee and A. Shakouri, Phys. Rev. Lett. 92, 106103 22, 3970 (2010); Z.-G. Chen, G. Han, L. Yang, L. Cheng, (2004). and J. Zou, Prog. Nat, Prog. Nat. Sci. 22, 535 (2012); [29] O. Entin-Wohlman, Y. Imry, and A. Aharony, Phys. Rev. J.-F. Li, W.-S. Liu, L.-D. Zhao, and M. Zhou, NPG Asia B 82, 115314 (2010). Mater. 2, 152 (2010). [30] R. Sánchez and M. Büttiker, Phys. Rev. B 83, 085428 [9] Kanishka Biswas, Jiaqing He, Ivan D. Blum, Chun-I (2011). Wu, Timothy P. Hogan, David N. Seidman, Vinayak P. [31] R. Sánchez, B. Sothmann, A. N. Jordan, and M. Dravid, and Mercouri G. Kanatzidis, Nature 489, 414 Büttiker, New J. Phys. 15, 125001 (2013). (2012). [32] B. Sothmann, R. Sánchez, and A. N. Jordan, Nanotech- [10] L. D. Hicks, T. C. Harman, M. S. Dresselhaus, Appl. nology 26, 032001 (2015). Phys. Lett. 63, 3230 (1993). [33] H. L. Edwards , Q. Niu , and A. L. de Lozanne , Appl. [11] M. S. Dresselhaus, G. Chen, M. Y. Tang, R. Yang, H. Phys. Lett. 63, 1815 (1993); H. L. Edwards , Q. Niu , G. Lee, D. Wang, Z. Ren, J.-P. Fleurial, and P. Gogna, Adv. A. Georgakis , and A. L. de Lozanne , Phys. Rev. B 52, Mater. 19, 1043 (2007). 5714 (1995). [12] M. Cutler and N. F. Mott, Phys. Rev. 181,1336 (1969). [34] J. R. Prance, C. G. Smith, J. P. Griffiths, S. J. Chorley, [13] U. Sivan and Y. Imry, Phys. Rev. B 33, 551 (1986). D. Anderson, G. A. C. Jones, I. Farrer, and D. A. Ritchie, [14] R. Whitney, Phys. Rev. Lett. 112, 130601 (2014). Phys. Rev. Lett. 102, 146602 (2009). [15] J.-H. Jiang, O. Entin-Wohlman, and Y. Imry, Phys. Rev. [35] J.-H. Jiang, O. Entin-Wohlman, and Y. Imry, New J. B 85, 075412 (2012). Phys. 15, 075021 (2013). [16] J.-H. Jiang, J. Appl. Phys. 116, 194303 (2014). [36] B. Sothmann, R. Sánchez, A. N. Jordan, and M. [17] J.-H. Jiang, M. Kulkarni, D. Segal, and Y. Imry, Phys. Büttiker, Phys. Rev. B 85, 205301 (2012). Rev. B 92, 045309 (2015). [37] A. N. Jordan, B. Sothmann, R. Sánchez, and M. [18] H. G. Bremmerman, Proc. Fifth Berkeley Symp. on Büttiker, Phys. Rev. B 87, 075312 (2013). Math. Statistics and Probability, L. M, LeCam and J. [38] B. Sothmann, R. Sánchez, A. N. Jordan, and M. Neyman, eds. (Univ. of Cal., Berkeley, 1966); J. D. Büttiker, New J. Phys. 15 095021 (2013). Bekenstein, Phys. Rev. Lett. 46, 623 (1981); P. A. Be- [39] L. Simine and D. Segal, Phys. Chem. Chem. Phys. 14, nioff, Int. J. Theor. Phys. 21, 177 (1982); J. B. Pendry, 13820 (2012); ibid., J. Chem. Phys. 141, 014704 (2014); J. Phys. A: Math. and Gen. 16, 2161 (1983). L. Arrachea, N. Bode, and F. von Oppen, Phys. Rev. B [19] Y. Imry, Physics of mesoscopic systems. In: Directions in 90, 125450 (2014). Condensed Matter, Memorial volume in honor of Profes- [40] B. Cleuren, B. Rutten, and C. Van den Broeck, Phys. sor Shang-keng Ma. Eds. G. Grinstein and G. Mazenko Rev. Lett. 108, 120603 (2012). (World Scientific, Singapore, 1986); Y. Imry, Introduction [41] B. Roche, P. Roulleau, T. Jullien, Y. Jompol, I. Far- to Mesoscopic Physics (Oxford University Press, 1997 rer, D.A. Ritchie, and D.C. Glattli, Nat. Comm. 6, 6738 12

(2015). cal and Electronic Properties of Semiconductors, 4th ed. [42] H. Thierschmann et al. Nature Nanotech. 10, 854 (2015). (World Scientific, Singapore, 2004). [43] B. Sothmann and M. Büttiker, Europhys. Lett. 99 27001 [47] J. H. Jiang, Phys. Rev. E 90, 042126 (2014). (2012). [48] G. E. W. Bauer, E. Saitoh, and B. J. van Wees, Nat. [44] B. Rutten, M. Esposito, and B. Cleuren, Phys. Rev. B Mater. 11, 391 (2012). 80, 235122 (2009); T. Ruokola and T. Ojanen, Phys. Rev. [49] N. Li, J. Ren, L. Wang, G. Zhang, P. Hänggi,and B. Li, B 86, 035454 (2012). Rev. Mod. Phys. 84, 1045 (2012). [45] L. Onsager, Phys. Rev. 37, 405 (1931); 38, 2265 (1931). [46] H. Haug and S. W. Koch, Quantum Theory of the Opti-