Simultaneous Extraction of Charge Density Dependent Mobility and Variable Contact Resistance from Thin Film Transistors Riccardo Di Pietro1*, Deepak Venkateshvaran2, Andreas Klug3, Emil J.W. ListKratochvil3,4, Antonio Facchetti5, Henning Sirringhaus2 and Dieter Neher1 Supporting information 1 Universität Potsdam, Institut für Physik und Astronomie, Karl-Liebknecht-Str. 24-25, 14476 Potsdam, Germany 2 Cambridge University, Cavendish Laboratory, J. J. Thomson Avenue, CB3 0HE Cambridge, United Kingdom 3 NanoTecCenter Weiz Forschungsgesellschaft mbH, Franz-Pichler-Straße 32, A-8160 Weiz, Austria 4 Institute of Solid State Physics, Graz University of Technology, Petersgasse 32, A-8010 Graz, Austria 5 Polyera Corporation, 8045 Lamon Ave, STE 140, Skokie, IL 60077-5318, USA * dipietro@uni-potsdam.de 1 - Full derivation of the saturation mobility equation In the following derivation the onset voltage for charge accumulation in the channel is assumed to be zero. The results can be easily extended by substituting ππΊ with πΜπΊ = ππΊ − πππ , and taking into account that all derivatives with respect to the gate voltage are not affected by the constant value of πππ (i.e. ΜπΊ ) ππ(π Μ πππΊ = ΜπΊ ) ππ(π ). πππΊ The equations can be used both for p-type (ππ· , ππΊ − πππ < 0) and n-type (ππ· , ππΊ − πππ > 0) transistors. The charge sheet model together with the gradual channel approximation allows the charge density per unit area q at position x along the channel to be defined as: π(ππΊ − π(π₯)) = πΆπ β (ππΊ − π(π₯)) (1) where |ππΊ − πππ | ≥ |ππ· |. In order to consider the effects of gate voltage dependent mobility, we define an effective mobility (gate voltage dependent) as the average of the charge density dependent mobility u(V) over the accumulated charge density at position x along the channel: π ππππ (ππΊ − π(π₯)) = ∫0 π(π)ππ π = π −π(π₯) π(π£)ππ£ ∫0 πΊ (2) ππΊ −π(π₯) We note that the function is defined for the whole range where equation (1) is defined. Using equation (1) the charge accumulated in a strip parallel to the source drain contacts of width dx can be written as: ππ = ππΆπ β (ππΊ − π(π₯))ππ₯ (3) Combined with the effective mobility (the average mobility at which charges move) it can be used to write the drift current equation: πΌ= ππ ππ‘ = ππΆπ β (ππΊ − π(π₯)) β ππππ (ππΊ − π(π₯)) ππ(π₯) ππ₯ π −π(π₯) = ππΆπ (∫0 πΊ π(π£)ππ£ ) ππ(π₯) ππ₯ (4) By separating the variables in equation (4) it is possible to write the general equation for charge transport in field-effect transistors: πΌ= π ∗ π −π(π₯) π πΆ ∫ π· (∫0 πΊ π(π£)ππ£ ) ππ(π₯) πΏ π ππ∗ (5) In the saturation regime it is possible to assume ππ·∗ = ππΊ , therefore equation (5) can be written as: πΌ= π π πΆ ∫ πΊ πΏ π ππ∗ π −π(π₯) (∫0 πΊ π(π£)ππ£ ) ππ(π₯) (6) In order to extract the mobility from equation (6) we consider the following function: π −π(π₯) π(ππΊ − π(π₯)) = ∫0 πΊ π(π£)ππ£ (7) And expand it around ππΊ in a Taylor series: π(ππΊ − π(π₯)) = ∑∞ π=0 π (π) (ππΊ ) (−π(π₯))π π! (8) 2 From which it is possible to write the generalized integral of π(ππΊ − π(π₯)) as: ∫ π(ππΊ − π(π₯))ππ(π₯) = − ∑∞ π=0 π (π) (ππΊ ) (−π(π₯))π+1 (π+1)! (9) The simplest way to obtain π(π) is to perform a second derivative of the saturation current: π2 πΌ πππΊ 2 ≅ π πΆ π(π) πΏ π (10) where the terms containing ππ ∗ have been neglected. However, performing a second derivative on a transfer curve introduces a high noise in the result, thus requiring longer integration times and/or more data points to obtain a reasonable signal to noise ratio. In order to perform a single derivation also in the saturation regime, it is sufficient to calculate the following derivative: 2 π √πΌ ππΌ 2 ) √πΌ πππΊ 1 (ππ ) = (2 πΊ 1 2 ππΌ = 4|πΌ| (ππ ) (11) πΊ Replacing the integral in equation (6) with equation (8), the first derivative can be expressed as: ππΌ πππΊ π π = ππ { πΏ πΆπ [− ∑∞ π=0 πΊ ππΊ π (π) (ππΊ ) π+1 (−π(π₯)) ] } (π+1)! ππ∗ ∑∞ π=0 π (π) (ππΊ ) (−ππ ∗ )π+1 ] (π+1)! ∑∞ π=0 π (π) (ππΊ ) (−ππΊ )π (π)! + ∑∞ π=1 ∑∞ π=1 π (π) (ππΊ ) (−ππ ∗ )π (π)! − ∑∞ π=0 π πΆ πΏ π = [− ∑∞ π=1 = π π π (π) (ππΊ ) ∑∞ (−ππΊ )π+1 πΆ [− π π=0 (π+1)! πΏ πππΊ π (π) (ππΊ ) (−ππΊ )π (π)! π (π) (ππΊ ) (−ππ ∗ )π (π)! − ∑∞ π=0 + + ππ ∗ π (π) (ππΊ ) (−ππ ∗ )π π ] (π)! πππΊ = π πΆ πΏ π ππ ∗ π (π) (ππΊ ) (−ππ ∗ )π π ] (π)! πππΊ [π(ππΊ ) + (12) While the current can be written as: πΌ= π π πΆ ∫ πΊ πΏ π ππ∗ π(ππΊ − π(π₯))ππ(π₯) = π πΆ πΏ π [− ∑∞ π=0 π (π) (ππΊ ) ((−ππΊ )π+1 (π+1)! π πΆ πΏ π [− ∑∞ π=0 ππΊ π (π) (ππΊ ) (−π(π₯))π+1 ] (π+1)! ππ∗ = − (−ππ ∗ )π+1 )] (13) Combining equations (11), (12) and (13), the following equation is obtained: 2 π √πΌ ) πππΊ ( = π πΆ 4πΏ π 2 (π) π π πππ∗ π(π) (ππΊ ) ∞ π (ππΊ ) ∗ −∑ ∗ β (−π ) (−π ) ] π=0 (π)! π π (π)! πππΊ [π(ππΊ )+∑∞ π=1 π(π) (π ) πΊ π+1 −(−π ∗ )π+1 )| |− ∑∞ π π=0 (π+1)! ((−ππΊ ) (14) In order to write equation (13) in a more useful form, it is necessary to perform some simplifications. First it is assumed that πππ∗ πππΊ βͺ 1 so that we can remove it from the numerator of the fraction. Furthermore, keeping in mind that the conductivity of the depleted channel is negligible, i.e. π(0) = 0, it is possible to write: π(0) = ∑∞ π=0 π (π) (ππΊ ) (−ππΊ )π π! ⇒ π(ππΊ ) = − ∑∞ π=1 π (π) (ππΊ ) (−ππΊ )π π! (15) 3 Then equation (14) can be written as: π √πΌ 2 (ππ ) π πΆ 4πΏ π = πΊ π 4πΏ = 2 π(π) (π ) πΊ π+1 −(−π ∗ )π+1 )| |− ∑∞ π π=0 (π+1)! ((−ππΊ ) π π(π) (ππΊ ) π(π) (ππΊ ) (−ππΊ )π +∑∞ (−ππ∗ ) ] π=1 π! π! 2 [− ∑∞ π=1 π 4πΏ = π π(π) (ππΊ ) π(π) (ππΊ ) (−ππΊ )π +∑∞ (−ππ∗ ) ] π=1 π! π! [− ∑∞ π=1 (16) π(π) (π ) πΊ π+1 −(−π ∗ )π+1 )| |π(ππΊ )ππΊπ∗ −∑∞ π π=1 (π+1)! ((−ππΊ ) π π(π) (ππΊ ) π(π) (ππΊ ) (−ππΊ )π +∑∞ (−ππ∗ ) ] π=1 π! π! 2 [− ∑∞ π=1 |− ∑∞ π=1 π(π) (ππΊ ) π(π) (ππΊ ) π+1 −(−π ∗ )π+1 )| (−ππΊ )π ππΊπ∗ −∑∞ π π=1 (π+1)! ((−ππΊ ) π! By separating the contribution of the first derivative and including the rest of the terms of the Taylor series expansion in a parameter π(ππΊ ), it is possible to rewrite equation (16) as: 2 π √πΌ ) πππΊ ( π ′ (ππΊ )2 ππΊπ∗ 2 π β ′ 4πΏ |π (ππΊ )ππΊπ∗ ππΊ −1π ′ (ππΊ )((ππΊ )2 −(ππ∗ )2 )| = π(ππΊ ) = 2 π π ′ (ππΊ ) 4πΏ |1(ππΊ −ππ∗ )| = 2 ππΊπ∗ π π ′ (ππΊ ) π 4πΏ | πΊ −1(ππΊ +ππ∗ )| ππΊπ∗ 2 π ππΊπ∗ β π(ππΊ ) (17) β π(ππΊ ) = 2πΏ π′(ππΊ ) β π(ππΊ ) where π(ππΊ ) depends on the particular kind of charge density dependence of the mobility. In the simpler case of a power law dependence of the mobility: π(π£) = πΌ(π£)π½ (18) it can be easily seen how equation (17) is reduced to: 2 π √πΌ ) πππΊ ( = π πΆ π(ππΊ ) β 2πΏ π π, π= π½+2 2(π½+1) (19) and the value of π becomes gate voltage independent (for the full derivation see below). Most importantly, it can be seen that: 0 ≤ π½ ≤ +∞ ⇒ −1 < π½ < 0 1 ≥ π ≥ 0.5 (20) ⇒ +∞ > π > 1 In the physically relevant case of a mobility increasing with charge density, the correction factor π(ππΊ ) will therefore be limited between 1 and 0.5 although it will not be constant throughout the π √πΌ 2 curve. If it is possible to fit the function (ππ ) with the power law dependence in the range used for πΊ the measurement, the extracted value of π½ can then be used to approximate the value of k to a constant value, and the mobility in the saturation regime can be obtained as: 2πΏ π √πΌ π πΊ 2 π(ππΊ ) = ππΆ βπ (ππ ) (21) without the need to perform a second derivative of the transfer characteristics. This approximation holds also when the approximation with a power law dependence is not perfect due to the relatively little dependence of the value of k with respect to the value of π½. The value of πΆ(ππΊ ) can be easily 4 obtained either by calculating the geometric capacitance of the dielectric layer or by measuring the capacitance-voltage characteristics of the actual device. The comparison of corrected and uncorrected mobility with the second derivative method and the experimental methods is shown in figure S1. The same approach is followed in the main text, showing in that case an even better agreement with the four point probe measurements. Mobility (cm2/Vs) 0.3 Uncorrected mobility Corrected mobility Saturation mobility (II derivative) power law fit (ο’=0.76, k=0.78) 0.2 0.1 N2200 0.0 0 20 40 60 Gate Voltage fpp mobility VD=5V 0.3 fpp mobility VD=10V Mobility (cm2/Vs) Saturation mobility Saturation mobility (corrected) 0.2 0.1 N2200 0.0 0 20 40 60 Gate Voltage Figure S1. (a) Comparison of the different mobility extraction methods, the uncorrected mobility (k=1), the corrected mobility (equation (18)) obtained with a power law approximation (red trace) and the second derivative method (equation (9)) with a 15 point Savitzky-Golay filtering (third order polynomial). (b) Four point probe mobility (for VDS=5V and VDS=10V) and the corrected and uncorrected mobility showing the higher agreement with the correction. 5 2 - Full derivation of the saturation mobility equation for a power law dependent mobility Starting from the following function (where the onset voltage πππ is explicitly taken into account): π(π) = πΌ(π − πππ )π½ { 0 π ≥ πππ π < πππ (22) It is possible to calculate the current-voltage characteristics of the device using the following equation (the inner integral lower limit needs to be shifted to πππ , as function (22) is only defined for ππΊ ≥ πππ and is identically 0 below that value): π π −π(π₯) πΌπΏ = ππΆπ ∫0 πΊ ∫π πΊ ππ πΌ(π£ − πππ )π½ ππ£ ππ(π₯) (23) Keeping ππΊ as a parameter, the double integration can be performed by taking into account that the generalized integral is zero for lower integration parameter: π π −π(π₯) ∫0 πΊ ∫π πΊ ππ π −πππ 1 πΆ πΌ(ππΊ π½+1 π πΆπ πΌ(π£ − πππ )π½ ππ£ ππ(π₯) = ∫0 πΊ 1 πΆ πΌ(ππΊ (π½+1)(π½+2) π − π(π₯) − πππ )π½+1 ππ(π₯) = 1 − πππ )π½+2 = (π½+1)(π½+2) πΆπ πΌ(ππΊ − πππ )π½+2 (24) Where the limit on the outer integral is shifted always to keep into account the presence of the onset voltage (channel is pinched-off, that is charge density at the upper voltage limit is 0, if onset voltage is not taken into account it would become negative) Therefore the current can be written as: 1 πΌ = (π½+1)(π½+2) π πΆ πΌ(ππΊ πΏ π − πππ )π½+2 (25) And the saturation mobility can be extracted from the following derivation: 2 π √πΌ (ππ ) πΊ π½+2 = π 1 π(π −π ) 2 πΆ πΌ ( πΊ ππππ πΏ (π½+1)(π½+2) π πΊ (π½+2) π 2 ) = (π½+2)2 π πΆ πΌ ((ππΊ πΏ 4(π½+1)(π½+2) π π½ 2 π − πππ ) 2 ) = 2πΏ πΆπ πΌ(ππΊ − (π½+2) πππ )π½ 2(π½+1) = 2πΏ πΆπ π(ππΊ ) 2(π½+1) (26) (π½+2) With a correction factor now gate voltage independent and equal to 2(π½+1) . 6 3 – Transconductance estimation in the linear regime In the linear regime ππ ∗ , ππ·∗ βͺ ππΊ , therefore Equation (5) can therefore be rewritten as: πΌ= π πΆ πΏ π [− ∑∞ π=0 π π·∗ π (π) (ππΊ ) π+1 (−π(π₯)) ] (π+1)! ππ∗ (27) And now the derivative with respect to the gate voltage can be easily calculated, and if the series is limited to the first order (which is not a strong assumption since ππ ∗ , ππ·∗ βͺ ππΊ , therefore the Taylor series is always evaluated close to ππΊ ), and neglect terms containing πππ·∗π∗ πππΊ for the same reasons, the first derivative of the current in the linear regime can be written as: ππΌ πππΊ ≅ π πΆ π(ππΊ )ππ·∗ π ∗ πΏ π (28) 7 4 - Example of correction factor approximation As an example of how the correction factor can be estimated for an arbitrary function, the transfer characteristics have been simulated starting from the following polynomial function: πΆπ π(π) = 1.17 β π(π) β (0.1 β π 0.5 + 2 β 10−5 β π 2.5 ) (29) In figure S2(a) the chosen mobility function is shown alongside the two power law components, responsible for the low voltage and high voltage part of the plot. A simple power law fit is shown in a dashed line. Using equation (19) mobility is extracted from the calculated transfer characteristics (figure S2(b)), without any correction factor (in red) and using the approximated correction factor obtained from the power law fit (in blue). The procedure returns a much more accurate mobility value despite the rather poor agreement between the original curve and the fit. The reason for this is shown in figure S2(c), where the mobility ratio (equal to 1/π) is calculated for the full curve (black trace), for the two power components of equation (29) (red and green traces) and for the power law fit of the chosen function. As shown in the mathematical derivation, the correction factor for an arbitrary function is in general gate voltage dependent, but in any case is within the values of the exponents of the different monomers which are used in the polynomial function. The rough approximation resulting from the power law fit returns a value that is an average of the complete gate voltage dependent correction factor. Although it does not fully recover the initial mobility, it greatly improves the accuracy of the mobility extraction, which is in turn fundamental for the contact resistance extraction. 8 2.5 ο‘V0.5+ο’V2.5 power law fit (ο§V1.08) ο‘V0.5 ο’V2.5 Mobility (a.u.) 2.0 1.5 1.0 0.5 0.0 0 20 40 60 80 Gate voltage 2.5 Simulated Extracted Corrected (k=0.74) Mobility (a.u.) 2.0 1.5 1.0 0.5 0.0 0 20 40 60 80 Gate voltage 1/k 1.5 1.0 ο‘V0.5+ο’V2.5 power law fit (k=0.74) ο‘V0.5 (k=0.83) ο’V2.5 (k=0.64) 0.5 0 20 40 60 80 Gate voltage Figure S2. (a) A model polynomial mobility curve, used for the calculation of the transfer characteristics (black trace) is shown together with the two power law components (red and green trace) and a power law fit (dashed trace). (b) The model mobility curve (black trace), together with the uncorrected mobility extracted from the calculated transfer characteristics (red trace) and the saturation mobility corrected using the scaling factor extracted from the power law fit (blue trace). (c) Calculated correction factors for the full polynomial mobility (black trace, gate voltage dependent), for the two different components of the polynomial mobility (red and green) demonstrating how the real correction factor is always in between the two extreme values, and the correction factor obtained from the power law fit (averaging the variable correction factor of the initial function. 9 5 - Extended experimental details All the devices were prepared in a bottom-contact top-gate geometry. A hall bar structure with patterned semiconductor [1] has been used for the four point probe measurement (W=200µm, L=100µm, probe at ¼ and ¾ of the channel length), while a pre-patterned substrate with different channel length transistors (W=2.85mm, L=5/12/50µm) has been used for the TLM measurements. Devices for the OSC thickness dependent measurement have been performed on interdigitated source-drain electrodes prepared by shadow mask evaporation (W=149mm, L=100µm, contact thickness 20nm). The effective contact length, required for the extraction of contact resistivity for the interdigitated pattern is defined as half the width of the contact finger (πΏπΆ = 50 ππ). P(NDI2OD-T2) and PBTTT, an n- and a p-type organic semiconductor respectively, were dissolved in dichlorobenzene for the four point probe devices, while P(NDI2OD-T2) dissolved in a 1:1 mixture of chloronaphthalene:xylene was used for the TLM devices. All solutions had an 8 g/l concentration (except for the variable thickness measurements where different concentrations from 5 to 20 g/l were used) and were deposited via spin-coating at 1000 rpm for 60 seconds and annealed at 200°C and 180°C for P(NDI2OD-T2) and PBTTT, respectively. A dielectric layer made of EG-PMMA (Polymer Source P8509P-MMA) with a concentration of 60 g/l was subsequently spin-cast at 1500 rpm for 90 seconds and dried overnight at 80°C. Aluminum gate electrodes were deposited through shadow mask evaporation. All measurements have been performed at room temperature in inert atmosphere using an Agilent 4155C semiconductor parameter analyzer and a needle probe station, in all cases performing forward and backwards voltage sweeps. The capacitance was determined from the measured thickness of the PMMA layer (roughly 550 nm for all devices). All thickness measurements were performed using a Dektak3ST stylus profilometer. [1] J.-F. Chang, M. C. Gwinner, M. Caironi, T. Sakanoue, and H. Sirringhaus, Adv. Funct. Mater. 20, 2825 (2010). 10 6 - N2200 four-point probe mobility extraction 0.30 N2200 Mobility (cm2/Vs) 0.25 0.20 0.15 0.10 linear (Vd=5V) 0.05 saturation (Vd=60V) 0.00 0 20 40 60 Gate Voltage (V) Figure S3. Extracted mobility from the linear and saturation transfer curves of an N2200 four point probe transistor, using equation (9) and (18) respectively. Figure S3 shows the extracted mobility of an N2200 transistor. At low gate voltages the saturation mobility becomes lower than the linear one, thus preventing the calculation of any contact resistance. The unexpected behavior is possibly due to the additional preparation steps required for pattering the semiconductor, as the same effect is not present in any other N2200 device. This could lead to an invalidation of the assumption πππ∗ πππΊ βͺ 1, necessary for the extraction of accurate values of the mobility in the saturation regime. However, at higher gate voltages the correct dependence is recovered and the model can be used to correctly estimate the contact resistance (as reported in the main text). 11 7 - Uncertainty estimations The measurement accuracy for the used equipment (Agilent 4155C) is dependent on many factors including measurement integration time, applied voltage and measurement range. For the voltages and ranges of interest in this work, it’s safe to assume that the relative error on a current measurement is constant and equal to ππ = 10−3. The numerical derivation for a generic data point n is performed according to the following formula: ππΌ πΌ 1 −πΌ πΌ −πΌ πΎπ = πππ· | = 2 (ππ· π −ππ· π−1 + ππ· π+1 −ππ· π ) πΊ πΊπ π πΊ π−1 πΊ π+1 (29) πΊπ With the relative error that can be calculated as: ππΎπ πΎπ ≅πΌ ππ βπΌπ· π (30) π· π −πΌπ· π−1 And for the saturation regime it is possible to derive an analogous formula: πππ ππ ≅ 2(πΌ ππ βπΌπ· π (31) π· π −πΌπ· π−1 ) with ππ = π √πΌπ· πππΊ | . π Starting from the equations above, it is possible to determine the uncertainty for mobility and contact resistance using a simple error propagation rule for uncorrelated values: πππ ππ‘ | = ππ ππ‘ π ππ πΆ | ≅ π πΆ π πππππ | ≅πΌ ππ ππ‘ π πππ·′π′ ππ βπΌπ· π (32) π· π −πΌπ· π−1 π βπΌπ ππ‘ π | ≅ √(πΌπ ππ‘ −πΌπ·π ππ‘π ππ·′π′ π π· π π· π−1 2 ) +( πππ ππ βπΌπ· π πππ −πΌ πππ πΌπ· π π· π−1 2 ) (33) The result of this analysis is shown in figure S4 for the four point probe measurement, where the difference between the measured and calculated values (relative to the calculated values) is compared against the accuracy of the calculated values (taken as twice the standard deviation). 12 Figure S4. The difference between the calculated and experimental values reported in figure 2 (main article) relative to the calculated values is plotted: (a) pBTTT mobility, (b) P(NDI2OD-T2) mobility, (c) pBTTT contact resistance, (d) P(NDI2OD-T2) contact resistance. These are compared against the accuracy of the calculated values, displayed as the black lines. Values in the gray shaded area are not significant since differences are below the accuracy of the calculated values. Only the calculation on the backwards traces are shown for clarity. 13 8 - N2200 four-point probe mobility extraction Figure S5 shows the total device resistance at different source-gate voltages, obtained by fitting the output curve between ππ·π = 1π and ππ·π = 4π. Figure S5. Total device resistance of the TLM devices plotted against channel length, for different gate voltages. The linear fits used for the calculation of contact and channel resistance are also shown in the plot for each data set. 14