Gsph∆ Ttp= ∆T. (4)Gsph+ ktpa+ kair( A − a)The open circuit voltage is given by:α ∆T GsphV =, (5)( Gsph+ kairA)+ a ( ktp− kair)where α is Seebeck coefficient. Expressing <strong>the</strong> electricalresistance <strong>of</strong> a <strong>the</strong>rmopile through its resistivity ρ, <strong>the</strong> poweron <strong>the</strong> matched electrical load can be written as:222spα ∆TG haPel=4 ⋅, (6)ρ2[( Gsph+ K air A)+ a(ktp− k air )]At its derivative dP el /da = 0, <strong>the</strong> maximal powercorresponds to a = a opt = (G sp h + k air A) / ( k tp – k air ), sothat:2α ∆TG hPmax=. (7)216ρ( ktp− kair)Substituting a opt into Eq. (4), <strong>the</strong> temperature difference∆T tp , corresponding to <strong>the</strong> maximal power, is:tp,opt22sp∆TG∆ T =, (8)sp⋅2 Gsp+ GTEG,0where G TEG,0 = k air A / h denotes <strong>the</strong> <strong>the</strong>rmal conductance <strong>of</strong><strong>the</strong> same TEG, but <strong>with</strong> no <strong>the</strong>rmocouples in between <strong>the</strong>plates, i.e., <strong>of</strong> <strong>the</strong> empty TEG; this is <strong>the</strong> parasitic <strong>the</strong>rmalconductance <strong>of</strong> a TEG. Eq. (8) shows that if <strong>the</strong> <strong>the</strong>rmalconductance <strong>of</strong> <strong>the</strong> air in empty TEG G TEG,0 > Ramb,0(13) and RTEG 0R amb , opt,>> . (14)In <strong>the</strong> optimized device, Inequalities (13) and (14) shouldhold, at least in a weak form (<strong>with</strong> “much more” replaced<strong>with</strong> “more”), fur<strong>the</strong>rmore, in typical situations <strong>of</strong> <strong>the</strong> energyscavengers, <strong>the</strong> ambient resistance does not vary greatly<strong>with</strong> temperature. For <strong>the</strong>se reasons, Eq. (12) instead <strong>of</strong> Eq.(11) can be usually used as a condition for optimizing <strong>the</strong>device.Considering that∆T TEG,0 = R TEG,0 W TEG,0 and ∆T TEG,opt = R TEG,opt W TEG,opt , (15)<strong>the</strong> condition <strong>of</strong> Eq. (12) can be rewritten as:R TEG,opt W TEG,opt = R TEG,0 W TEG,0 /2 . (16)In those cases, where <strong>the</strong> <strong>the</strong>rmal resistance <strong>of</strong> <strong>the</strong> ambiencedominates, <strong>the</strong> heat flow does not depend on <strong>the</strong> <strong>the</strong>rmalresistance <strong>of</strong> a TEG, and Eq. (16) simplifies to:R TEG,opt = R TEG,0 /2 , (17)<strong>the</strong>reby stating that <strong>the</strong> <strong>the</strong>rmal resistance <strong>of</strong> <strong>the</strong>rmocouplesand <strong>of</strong> <strong>the</strong> air are equal to each o<strong>the</strong>r. This condition iswidely used in designing <strong>the</strong> TEGs. As <strong>the</strong> goal <strong>of</strong> <strong>the</strong>optimization is to make <strong>the</strong> <strong>the</strong>rmal resistance <strong>of</strong> <strong>the</strong> TEGcomparable or larger than <strong>the</strong> one <strong>of</strong> <strong>the</strong> ambience, in mostcases, Eq. (17) cannot be used and must be replaced <strong>with</strong>Eq. (12) or, even better, <strong>with</strong> Eq. (11).We proceed now <strong>with</strong> <strong>the</strong> optimization <strong>of</strong> <strong>the</strong> TEG. First,we replace ∆T TEG,opt and ∆T TEG,0 in Eq. (11) <strong>with</strong> Eqs. (15).Then we eliminate <strong>the</strong> heat flows using:∆TWTEG,0= , (18)R + RWTEG,optamb,0amb,optTEG,0∆T= . (19)R + RTEG,optAfter such replacements, we solve Eq. (11) for R TEG,opt andobtain:RTEG,optRamb,optRTEG,0= . (20)2Ramb,opt+ RTEG,0Eq. (20) can be solved by iterations. In <strong>the</strong> beginning, <strong>the</strong>value <strong>of</strong> R amb,0 can be used instead <strong>of</strong> R amb,opt . Uponobtaining <strong>the</strong> first approximation value <strong>of</strong> R TEG,opt , <strong>the</strong> values<strong>of</strong> W TEG,opt and R amb,opt can be recalculated more accurately;<strong>the</strong> latter <strong>the</strong>n can be used in <strong>the</strong> next iteration. Only severaliterations are usually required for excellent accuracy.
As far as only two parallel <strong>the</strong>rmal resistors compose <strong>the</strong>TEG, <strong>the</strong> required <strong>the</strong>rmal resistance <strong>of</strong> <strong>the</strong> <strong>the</strong>rmopile R tp,optcan be easily obtained from <strong>the</strong> value <strong>of</strong> ∆T TEG,opt . Theoptimal area <strong>of</strong> <strong>the</strong>rmoelectric material a in <strong>the</strong> TEG is:a opt h / ktpRtp,opt= . (21)The minimal number <strong>of</strong> <strong>the</strong>rmocouple legs, while satisfying<strong>the</strong> requirement for <strong>the</strong> output voltage on <strong>the</strong> matched loadV m , is given by:2Vmn = . (22)α ∆TTEG,optFinally, <strong>the</strong> required cross section s <strong>of</strong> <strong>the</strong>rmopile legsshould not exceed s = a / n .<strong>Thermal</strong> matching: <strong>the</strong> application on manAs an example <strong>of</strong> application <strong>of</strong> <strong>the</strong> method described,let us consider a <strong>the</strong>rmopile <strong>with</strong> <strong>the</strong> same minimal lateralleg dimensions as in [1] <strong>with</strong> <strong>the</strong>ir size <strong>of</strong> 80 µm × 80 µm ×600 µm. For calculations, we assume that <strong>the</strong> hot plate has a<strong>the</strong>rmal contact <strong>with</strong> <strong>the</strong> skin in <strong>the</strong> outer side <strong>of</strong> <strong>the</strong> wristover a circular area <strong>of</strong> 3.14 cm 2 , while <strong>the</strong> radiating area is 7cm 2 , so that <strong>the</strong> device body resembles a watch. The hot andcold plates are separated by 1.3 mm, which is <strong>the</strong> thickness(1), however, this is because in our device <strong>the</strong> Inequalities(13) and (14), even in <strong>the</strong>ir weak form, are not satisfied.Small-size <strong>the</strong>rmopiles available on <strong>the</strong> market do not fit<strong>the</strong> requirements for <strong>the</strong> <strong>the</strong>rmopile legs coming out <strong>of</strong> <strong>the</strong>modeling <strong>of</strong> an optimal <strong>the</strong>rmopile because <strong>the</strong>ir aspect ratiois much smaller than needed. An appropriate aspect ratio<strong>the</strong>n can be obtained by stacking <strong>the</strong>rmopiles on top <strong>of</strong> eacho<strong>the</strong>r [2]. This increases <strong>the</strong> R TEG and hence <strong>the</strong> outputpower. For example, in case <strong>of</strong> a 10-stage <strong>the</strong>rmopile, <strong>the</strong>power increases in 5.7 times. Larger number <strong>of</strong> stages couldfur<strong>the</strong>r increase <strong>the</strong> power, but <strong>the</strong> device would be too thickand <strong>the</strong>refore inconvenient for <strong>the</strong> users.The normalized power for a 10-stage TEG, Fig. 2,coincides <strong>with</strong> <strong>the</strong> similar curve for a one-stage device. Theheat flow is, however, different. Inequality (14), even in itsweak form, is not yet satisfied in <strong>the</strong> 10-stage TEG; <strong>the</strong>R TEG,0 is still 84% <strong>of</strong> R amb,opt , however, <strong>the</strong> change in heatflow is already 44% at <strong>the</strong> matching point (1). It fur<strong>the</strong>rFigure 2. <strong>Thermal</strong> matching <strong>of</strong> a TEG to <strong>the</strong> ambience:normalized power (solid line) and normalized heat flow(dashed lines). The arrows show <strong>the</strong> heat flowscorresponding to three marked points.<strong>of</strong> <strong>the</strong> <strong>the</strong>rmopile including <strong>the</strong> ceramic plates. The powerproduced depends on <strong>the</strong> number <strong>of</strong> <strong>the</strong>rmocouple legs,<strong>the</strong>refore, at a certain number, corresponding to <strong>the</strong> optimalcross area <strong>of</strong> <strong>the</strong>rmoelectric material, <strong>the</strong> matching condition<strong>of</strong> Eq. (11) is satisfied. The normalized power computednumerically for an air temperature <strong>of</strong> 22 °C is shown in Fig.2 versus <strong>the</strong> ∆T TEG /∆T TEG,0 ratio. Each point <strong>of</strong> <strong>the</strong> curvecorresponds to a different value <strong>of</strong> a. The maximal power,point (1), is observed at ∆T TEG = ∆T TEG,0 /2.The analysis <strong>of</strong> numerical simulations shows that this isbecause <strong>the</strong> ambient resistance is only weakly dependent ontemperature. In addition, <strong>the</strong> <strong>the</strong>rmal resistance <strong>of</strong> <strong>the</strong>ambience is larger than <strong>the</strong> one <strong>of</strong> <strong>the</strong> generator. So, <strong>the</strong>device is in a condition, where <strong>the</strong> heat flow is nearlyindependent <strong>of</strong> R TEG , Fig. 2, so that Eq. (17) approximatelyholds. Using Eq. (17) for determining <strong>the</strong> number <strong>of</strong><strong>the</strong>rmocouples corresponding to <strong>the</strong> optimal power wouldhave generated a small error giving <strong>the</strong> point (2), Fig. 2. Onemay see that it is not very far from <strong>the</strong> true matching pointFigure 3. The ratios <strong>of</strong> <strong>the</strong> <strong>the</strong>rmal resistance <strong>of</strong> <strong>the</strong><strong>the</strong>rmopile and <strong>of</strong> <strong>the</strong> TEG to <strong>the</strong> one <strong>of</strong> <strong>the</strong> ambience versus<strong>the</strong> number <strong>of</strong> stages at <strong>the</strong>rmal matching, i.e., at a=a opt .increases after satisfying Inequality (14) in its weak form.This confirms that heat flow cannot be assumed as constantwhile modeling and that Eq. (17) cannot be used foroptimization: its use would have given <strong>the</strong> point indicated as(3) in Fig. 2, far away from real maximum.The ratios R TEG,opt /R amb,opt and R tp,opt /R amb,opt are shown inFig. 3 versus <strong>the</strong> number <strong>of</strong> stages. One can notice that <strong>the</strong><strong>the</strong>rmal resistance <strong>of</strong> a TEG does not halve at <strong>the</strong> matchingpoint (i.e., R TEG,opt ≠ R tp,opt /2), reflecting <strong>the</strong> difference <strong>with</strong>both parallel and serial matching.Some aspects <strong>of</strong> <strong>the</strong>rmal matching <strong>of</strong> TEGs on humansAs shown above, multi-stage <strong>the</strong>rmopiles allow <strong>the</strong><strong>the</strong>rmal matching, but it occurs at a very high effectiveaspect ratio, which is <strong>the</strong> ratio <strong>of</strong> <strong>the</strong> length <strong>of</strong> all <strong>the</strong><strong>the</strong>rmocouple legs on top <strong>of</strong> each o<strong>the</strong>r to <strong>the</strong>ir width. Thereare, however, some o<strong>the</strong>r helpful practical ways <strong>of</strong> easilytuning <strong>the</strong> <strong>the</strong>rmal resistance <strong>of</strong> <strong>the</strong> TEG components t<strong>of</strong>ulfill <strong>the</strong> <strong>the</strong>rmal matching requirements at smaller aspectratio or at smaller number <strong>of</strong> <strong>the</strong> stages.First, when applying <strong>the</strong>rmal matching conditions to <strong>the</strong>TEG on human skin, Inequality (14) in its weak formdemands to provide a <strong>the</strong>rmal plate-to-plate isolation <strong>of</strong> atleast 2 000 cm 2 K/W, which can be done only at <strong>the</strong>irdistance <strong>of</strong> about 1 cm from each o<strong>the</strong>r. Anyway, this spaceis required for a multi-stage <strong>the</strong>rmopile. On <strong>the</strong> o<strong>the</strong>r hand,<strong>the</strong> free/forced convection layer near <strong>the</strong> body, e.g., around