A Pulse-Biasing Small-Signal Measurement Technique Enabling 40 MHz Operation of Vertical Organic Transistors

Organic/polymer transistors can enable the fabrication of large-area flexible circuits. However, these devices are inherently temperature sensitive due to the strong temperature dependence of charge carrier mobility, suffer from low thermal conductivity of plastic substrates, and are slow due to the low mobility and long channel length (L). Here we report a new, advanced characterization circuit that within around ten microseconds simultaneously applies an accurate large-signal pulse bias and a small-signal sinusoidal excitation to the transistor and measures many high-frequency parameters. This significantly reduces the self-heating and therefore provides data at a known junction temperature more accurate for fitting model parameters to the results, enables small-signal characterization over >10 times wider bias I–V range, with ~105 times less bias-stress effects. Fully thermally-evaporated vertical permeable-base transistors with physical L = 200 nm fabricated using C60 fullerene semiconductor are characterized. Intrinsic gain up to 35 dB, and record transit frequency (unity current-gain cutoff frequency, fT) of 40 MHz at 8.6 V are achieved. Interestingly, no saturation in fT − I and transconductance (gm − I) is observed at high currents. This paves the way for the integration of high-frequency functionalities into organic circuits, such as long-distance wireless communication and switching power converters.

gate capacitance at high frequencies and using this value for extracting the mobility from DC, i.e. near 0 Hz, I-V measurements brings an additional error if there is a strong gate capacitance frequency dependence 13 .
Secondly, there are also time-dependent processes happening in organic transistors such as self-heating and bias-stress so that the dynamic behavior of organic FETs is insufficiently described by state-of-the-art DC parameter models.
Self-heating is an important physical effect which occurs during the device characterization at high DC I× V points and has two major drawbacks. First, it results in an unknown device junction temperature if the junction-to-substrate thermal resistance is not known, making the measured data unsuitable for fitting any compact models into it. Second, junction burning or breakdown limits the maximum I-V point at which the device can be characterized, whereas in some real circuits such as digital logic gates and super regenerative oscillators only a very high transient pulse current is passing through the device during the switching or on time.
Self-heating is an issue in all semiconductor-based devices, but organic transistors are inherently more sensitive to it because their charge carrier mobility strongly increases with temperature 14,15 , and their source/drain metal electrode to polymer channel contact resistance rapidly decreases with it 15 . In addition, these devices are mostly intended for flexible plastic substrates, which are in general very poor thermal conductors resulting in large junction to substrate thermal resistance. This problem is partly alleviated by low current densities in organic devices. However, these devices generally work instead at relatively higher voltage levels, and the current density will increase by development of higher mobility organic semiconductors in the future.
Device measurement can be divided into two general domains of large-signal and small-signal measurements. For the large-signal I-V characterization, the self-heating can be suppressed by applying short pulses instead of long-time DC biasing 14 . This can be easily done using existing commercial equipment such as the Keysight Precision Source/Measure Unit B2912A. However, for the case of small-signal measurements this is absolutely not straightforward and has not been done so far. All existing small-signal measurement techniques work based on the principle of long-time DC biasing of the device at a known I-V point, and then superimposing a known small-signal voltage/current on this DC bias at one input terminal of the device, and measuring the output small-signal voltage/current at another terminal of the device. For example, this approach has been used in our previous works 13,16 for measuring the transconductance (g m ), intrinsic gain (A v0 ), gate/base impedance, and current gain (h 21 ); for S-parameter measurement 17 ; and for transit frequency (f T ) measurement [18][19][20] .
In this context, DC biasing of the device also has a third drawback. Most organic transistors still suffer from the bias-stress effect, which is due to the trapping of charge carries, and causes dynamic shift of the threshold voltage as well as the electric field inside the device 21,22 . Therefore, applying a DC bias during long-time measurements changes the characteristics of the device and reduces the accuracy and repeatability of the results.
In order to overcome the above-mentioned problems, we present in this paper for the first time an advanced pulse-biasing characterization circuit, which can turn on the organic transistor and apply an accurate bias I-V to it in less than ten microseconds, then apply a small sinusoidal signal to the device and measure several small-signal parameters such as h 21 , f T , g m and A v0 , and then quickly turn off the device again. This approach significantly reduces the junction temperature increase, and allows measurements at much higher bias currents. In addition, the new setup can be used for measuring the temporal evolution of stress and self-heating effects over a μs to ms time scale.
As a case study, we characterized the fully-thermally-evaporated n-type vertical Organic Permeable-Base Transistor (OPBT) 23 , shown in Fig. 1. The OPBT is a promising low-voltage high-current high-speed device, fabricated solely using low-resolution low-cost shadow masks. This transistor resembles triodes and bipolar junction transistors, but here we have a thin metal base layer that contains naturally occurring pores. The native aluminum oxide of this electrode leads to a channel formation around it and the fact that the base potential controls the electrons flowing from emitter to collector. Space charge limited current (SCLC) in the undoped C 60 fullerene layers is the main limitation of this transistor 24 , whereas electron injection is very efficient due to the thin n-doped C 60 layer. Details of the device fabrication and operation mechanism, DC I-V characteristics, SCLC and device simulation can be found in other publications 23,24 , as well as the methods section.
This manuscript is organized in the following way. Firstly, we demonstrate the pulse-biasing measurement results of an OPBT device including the small-signal performance, and the temporal evolution of the charge distribution and self-heating effects occurring in the device. Secondly, we focus on the characterization circuit and small-signal parameter extraction techniques. Finally, we discuss the consequences of our findings on the context of real circuit applications.

Case Study Measurement Results
Measurements are performed at a controlled room temperature of ~25 °C, on the samples shown in Fig. 1, with an active area of Aact = 200 μm × 200 μm, at three low/moderate/high collector-to-base voltages of Transfer large-signal pulse I-V characteristics of the OPBT are shown in Fig. 1(c), reaching a peak current of I E = 200 mA at a total voltage of V CE = V CB + V BE = 8.6 V. This corresponds to a current density of 5 μA/μm 2 in the A act . At V CE = 1.0 V, the device can still drive 47 nA/μm 2 .
Transit Frequency (f T ). The small-signal current gain (h 21 ) is defined as the ratio of the small-signal collector current (i c ) over the small-signal base current (i b ) measured when both emitter and collector are small-signal AC ground.
An example of the magnitude of h 21 as a function of frequency at a pulse bias I E = 1.0 mA at three different V CB is shown in Fig. 2(a). The h 21 is decreasing ~20 dB per decade (i.e. proportional to 1/f) as known for conventional transistors. f T is defined as the frequency at which the extrapolation from the low frequency part of h 21 falls to unity. It is an important figure of merit of a transistor, indicating the frequency range in which the device can amplify the input current signal.
Measured f T versus pulse bias emitter current is shown in Fig. 2  for Pentacene at 25 V were previously reported for planar transistors on glass substrate 19 , fabricated using high-resolution patterning techniques such as photolithography and lift-off process. f T = 20 MHz at 30 V was achieved using laser sintering of high resolution electrodes on glass 18 . In vertical structures, f T = 1.5 MHz was reported for step edge devices 25 , and f T = 20 MHz at 15 V for a 3D transistor structure 20 . Fig. 2

(c) and
is calculated as the ratio of the i c to the small-signal base-emitter voltage; measured below the frequency of f T /10. Interestingly, no saturation of the g m-eff at high current densities is observed. This indicates that the parasitic emitter resistance R E is small, otherwise g m-eff = g m /(1 + g m × R E ) would eventually saturate to 1/R E at high currents. The small R E means that the electrode resistance is small and the n-C 60 doped layer is making low impedance interfaces to both Cr and intrinsic-C 60 layers.
Considering the equation f T ≈ g m /(2π × (C be + C bc )), increasing the V CB bias from 0.1 V to 3.3 V at fixed I E in Fig. 2(c) improves the transconductance, however, the amount of the corresponding f T or h 21 improvement at the same I E in Fig. 2(b or a) is always higher, e.g. X2 is 89% more than X1. This is because of the fact that increasing the V CB depletes the collector-base C 60 layer from the charge carriers, and this also reduces the C bc and therefore further improves the f T .

Intrinsic Gain (A v0 ).
Intrinsic gain is the maximum small-signal voltage gain that a device can provide, and is equal to g m × r out , where r out = ∂V CE /∂I C is the output resistance of the device. Basically a transistor with A v0 less than one, i.e. 0 dB, is a useless device, because it cannot perform any amplification.
In general, A v0 decreases with scaling down the channel length. However, as shown in Fig 16 .
A diffusion-driven organic transistor was recently reported 26 , which can even provide 57 dB of intrinsic gain at W/L = 100 μm/12.5 μm. However, this transistor can drive less than 200 nA at tens of volt, and therefore is only suitable for very low-current low-speed applications.
Temporal Evolution of Charge Distribution, Self-Heating and Bias-Stress. The charge carrier mobility in organic semiconductors is known to improve with temperature. In addition, polymeric materials as used for plastic substrates are generally weak thermal conductors. For these reasons, gradual self-heating of organic devices at high I × V points has a large impact on the device characteristics.
As will be explained in the next section, the developed pulse-biasing setup can also accurately monitor the variations of V BE , i.e. ΔV BE after applying a fixed I E to the device. Measuring the temporal evolution of this ΔV BE allows to study the effect of self-heating with the highest sensitivity, because in case of self-heating, this ΔV BE would be proportional to the increase of the device temperature, i.e. ΔT, and therefore to the power dissipated in the device. This fact is because organic semiconductors have usually strongly increasing mobility with temperature 14,15 . Further, the contact resistance strongly decreases simultaneously 15 , resulting in a lower required V BE , i.e. negative ΔV BE , at fixed I E . This ΔV BE would be linearly proportional to ΔT when ΔT is small, but in general it is a nonlinear function.
For the purposes of comparison and verification of this method, ΔV BE of a general purpose, silicon Bipolar Junction Transistor (BJT) is measured at two different I E × V CE products of 100 mW and 200 mW, and is shown in Fig. 3(a). Conventional BJT is also a vertical device, but the silicon substrate conducts heat around 10 times better than the glass. In this experiment, tracking of the V BE variation is started 90 μs after turning on the device. The V BE variation during the initial 90 μs was smaller than the accuracy of the measurement setup. 200 mW is applied one time by doubling the current and one time by doubling the voltage. As expected, in both cases ΔV BE is nearly two times that of the 100 mW. This confirms that here we only have self-heating, but no stress effects.
Similar experiments are performed on the OPBT at two low and high current levels as shown in Fig. 3(b and c). The V BE and V CE given in this figure are average values over the measurement time. In Fig. 3(b), although we have a considerable amount of ΔV BE at 1 mA × 1.6 V, surprisingly, doubling the V CE does not affect the ΔV BE , whereas doubling the current increases it by ~60%. This proves that in this case the observed effect is not self-heating.
As shown in Fig. 1(a), although the emitter window is 200 μm × 200 μm, the underneath base and collector electrodes are wider. Therefore we speculate that this effect is induced by the lateral diffusion of some of the electrons accumulated around the base oxide towards the outside of this window. This diffusion gradually increases the effective active area of the channel, and therefore decreases the required V BE at the fixed I E = 1 mA by the amount of ~9% after 200 ms in Fig. 3(b). Increasing the collector voltage has a small impact on the charge in the C 60 layer of the emitter side, whereas increasing the I E largely affects this charge density and the required V BE .
At a 15 times higher I E × V CE shown in Fig. 3(c), doubling the current increases the ΔV BE ~95%, but here doubling the voltage also increases it ~66%. This indicates that in this case in addition to the lateral diffusion of the charge carriers, self-heating is also occurring and is important. Since self-heating has a considerable impact after hundreds of μs at 10 mA, obviously at 100 mA range it would already have an impact after tens of μs. In order to circumvent this effect, the pulse-biasing circuit developed here can turn on and apply an accurate bias to the device under test within few μs, and immediately start doing small-signal analysis afterwards. Figure 3(d) shows a long-time DC stress measurement, performed using the Keysight B2912A Precision SMU, at a very low current of 100 μA at which self-heating would be negligible. The ΔV BE tracking is started ~200 ms after turning on the device. Here, we clearly see different mechanisms occurring in different directions. The ΔV BE initially goes negative due to the lateral charge diffusion, but finally starts increasing due to the bias-stress effect. Surprisingly, the lateral charge diffusion seems to be the dominant effect for an incredibly long time of ~1000 s. This might really be the case, because far away from the A act there is no lateral electric field, and the undoped C 60 SCIEnTIfIC REPoRTS | (2018) 8:7643 | DOI:10.1038/s41598-018-26008-0 layer could be extremely resistive causing very slow diffusion of electrons. However, other unknown mechanisms might also be the reason, such as very slow changes in the morphology of C 60 or base-oxide under the electric field. We cannot provide a concrete explanation for the observed behavior. Anyway, it is a very slow and small effect in the mV range. Figure 4 shows the simplified schematic of the new measurement setup. Examples of the waveforms are illustrated in Fig. 5. An accurate 1 V reference is generated using the LT1004 voltage reference, and is used for defining the pulse I E passing through R E1 = 1 V/I E as shown in Fig. 5(b). DC voltage at the base electrode is zero. ME3 forces V E = 0 V when there is no V G pulse, keeping the OPBT off. The 10 V low-duty-cycle gating pulse V G turns on ME2 on its rising edge to start the pulse bias current I E , and turns off ME3. The small-signal excitation comes from the sinusoidal source 2 × V in* , while C E1 and C C2 keep both emitter and collector at small-signal AC ground. Applying and stabilizing of the bias point at different nodes is done over the time interval of 0 to ~6 μs. Circuit parameters such as V D1 , V E1 and R C1 are manually tuned to minimize this initialization time. The small-signal measurements are carried out afterwards.

Pulse-Biasing Small-Signal Characterization Circuit
At the base side, we have R B4 ≫ R B1 ≫ 50 Ω, and the input base impedance |Z b | ≈ 1/C b ω<<R B1 at the measurement frequency. The diode D B1 has less than 0.3 pF parasitic capacitance and is completely off during the small-signal measurement time. h 21 and g m Measurement Circuitry. The small-signal sinusoidal excitation comes from a differential signal generator. The invert of V in arrives at the oscilloscope, and then V in is calculated from it. The Cables 1-4 are 50 Ω, i.e. V in and V inB are slightly delayed versions of V in* . Since R B1 ≫ 50 Ω we have |V inB | ≈ |V in* | = |V in |. i c /i b /i e are small-signal currents into the collector/base/emitter, respectively. As long as R B1 > 7/C b ω, |i b | can be estimated as |V in |/R B1 with less than 1% error. However, by measuring the signal amplitude at V B using a high-impedance probe, as shown in Fig. 5(g), we can calculate the input base capacitance C b and then calculate |i b | even more precisely.
At the collector side, the large-bandwidth op-amp O 1 forces n3 to be ground and the large coupling capacitor C C2 keeps the collector to be AC ground. L C1 provides a path for the pulse I E bias current, and L C1 ω ≫ 1/C C2 ω, i.e. most of the i c passes through C C2 , but C C2 is so large that the small-signal voltage amplitude at the collector is much smaller than the signal at the base. It is possible to calculate the exact value of |i c | using the equation given in Fig. 4. Since I E is a pulse, it is important to make sure that L C1 -C C2 do not cause additional ringing on the V out . This is assured by tuning R C1 ≈ 2 × sqrt(L C1 /C C2 ). As shown in Fig. 5(i), when I E pulse is applied but V in is zero, no ringing appears on the V out after Time = 6 μs. A 50 pF capacitor is added to the feedback loop around O 1 to improve the stability.  Cable1 is intentionally twice the length of Cables 2 and 3; so that the signal delay from V in* to V inB plus the signal delay from n4 at the output of the op-amp O1 to V out would be equal to the delay along Cable1. In this way, we could also measure the phase of h 21 by measuring the phase difference between V in and V out on the oscilloscope. However, only the magnitude of h 21 is needed to extract the f T , and |h 21 | = |i c |/|i b |.
At the emitter side, a very large C E1 , in the μF range, is needed to keep the emitter at AC ground during the small-signal measurement after Time = 6 μs. On the other hand, a time much larger than T 0 = V BE × C E1 /I E would be needed for V E to reach its required value; with the parameter values given in Fig. 5, this would be T 0 = 57 μs. In order to significantly speed up this initialization time, the power switch ME4 is added. During the long off time, i.e. V G = 0 V, C E2 is charged to a tunable voltage V D1 in the range of 0-30 V, and then on the rising edge of V G , a very large transient current up to V D1 /1.5 Ω is sunk through C E2 and as shown in Fig. 5(c) this current quickly charges C E1 to a V BE that enables an OPBT current exactly equal to the I E . However, this is true only if after this initialization time, no part of the I E passes through C E1,2,4 . In other words, the gradual slope of V E should become zero.
To accurately monitor the gradual slope and variations of V E after its sharp falling edge, V G turns on ME5 for the first 1-3 μs, and during this time the small capacitor C E4 is charged to the initial emitter voltage. Then the kΩ resistor R E2 quickly pulls n1 to 0 V, turning off ME5, resulting in a >4 GΩ resistance across ME5,6. Because of the large time constant of C E4 × R DS-ME5,6 , any further mV variations at V E will be directly tracked at n2 and V S . V D1 is manually tuned to have ~0 mV/μs gradual slope at V S over the small-signal measurement interval. Comparing Fig. 5(c to d), the V S circuitry is providing ~25X zoom on the V E for accurate tuning of V D1 . ME6 just forces n2 to ground on the falling edge of V G .
On the falling edge of V E , a large R B1 × C b time constant can slow down the charging of the base-emitter capacitance and increase the required initialization time. To make this faster, on the rising edge of V G , a certain amount of charge is injected into the base through C B1 -R B2 -D B1 . V E1 is a negative DC voltage that controls the amount of this charge, and turns off the D B1 afterwards. V E1 is manually tuned to keep V B around 0 V. DC leakage current into the base, I B , is usually negligible. However, in case of a large I B , R B4 can be added, and V D2 controls the DC current injected into the base.
Intrinsic Gain Measurement Method. For measuring the intrinsic gain, V E is accurately measured at two slightly different collector voltages, but with equal pulse I E . Then we would have g m × ΔV BE = ΔV CE /r out . Therefore A v0 = g m × r out = ΔV CE /ΔV BE is obtained. For this measurement V B is grounded, C E1,2 are not needed, and V C is shorted to V DD . Measurement Limits. The low-leakage BS170 MOSFET used as ME1,2 current source can drive up to ~300 mA into the OPBT. Larger MOSFETs with higher current drive capabilities, and accurate resistors down to 100 mΩ range for R E1 are widely available in the market. Therefore, accurate I E pulses up to 10 A should be feasible with the proposed circuit, but it was not needed in our case study. At I E < 1 μA range, the gate leakage through ME1,2 can cause bias inaccuracies; therefore ultra-low leakage MOSFETs would be needed for this case. The maximum bias voltages V CB and V BE are basically controlled by the V DD and V EE ; there is no specific limit in this regard.
The shortest measurement pulse width is mainly limited by the required initialization time for reaching the steady state (~6 μs in Fig. 5); afterwards, the small-signal information can be extracted in principle even from one sinusoidal signal cycle. As can be seen in Fig. 5(b), the I E pulse settles in less than 2 μs. The charge injector circuitry ME4-C E2 has a very short R-C time constant of <160 ns. However, the OPBT capacitances and the frequency of the sinusoidal signal impose the main limitations on the initialization time. Although the charge injector path C B1 -R B2 -D B1 significantly speeds up the charging of the base capacitance C b , as can be seen in Fig. 5(g), still some time (~3 μs) proportional to R B1 × C b is needed for the V B to reach the steady state after D B1 turns off. On the collector side, C C2 should be more than 20 × 20 times larger than C b to keep the small-signal voltage at V C twenty times smaller than V B at the frequency where |h 21 | = 20 is measured. We recommend L C1 C C2 ω 2 > 10 so that the inaccuracy of the L-C components does not make large error in the i c equation in Fig. 4. Taking all these points into account, practically we found that by optimizing the circuit parameters, the initialization time can be minimized to 5-7 periods of the sinusoidal signal.

Further Discussions and Conclusion
The proposed pulse-biasing small-signal measurement circuit enables analog characterization of the device over >10 times wider I-V range comparing to the simple DC biasing approach 13,23 , and provides a much better control on the junction temperature by adjusting the pulse duration. It can also precisely track the ΔV BE at different power densities for better understanding of the self-heating effects.
For the ~12 μs measurement shown in Fig. 5, if we repeat the pulse every 1.2 s, the waveforms can still be well monitored on the oscilloscope, while the bias-stress rate in the device decreases by a factor of ~10 5 comparing to the DC biasing method. Therefore, many more measurements can be performed on the same fresh device during several days, providing reliable data for device modeling.
The OPBT currently suffers from a low charge carrier mobility in the vertical direction ~0.06 cm 2 /V.s 23 . However, the study presented in this work proved that despite this very low mobility, it can already reach the record f T of 40 MHz and the intrinsic gain of 35 dB at V CE = 8.6 V thanks to the short channel length. There is certainly a large room for further improvement of this speed by developing organic semiconductors which can provide higher mobility in the vertical direction. This can be better understood by considering the fact that in FETs, above the threshold region g m and therefore f T are proportional to sqrt(μ × I E ). This g m − I E trend can be seen in Fig. 2(c), and more importantly, no g m or f T saturation is observed at high currents, confirming the room for reaching higher speeds. A higher mobility will also result in a lower bias voltage, power and self-heating for reaching the same I E . In planar organic FETs, mobility in the range of 3-30 cm 2 V −1 s −1 has been reported by several groups 12,27-29 . Device simulations also predict sub-nano-second intrinsic switching delay for OPBTs with a mobility of 10 cm 2 /V.s 30 .
SCIEnTIfIC REPoRTS | (2018) 8:7643 | DOI:10.1038/s41598-018-26008-0 A f T ≥ 40 MHz can pave the way for integrating new functionalities into organic circuits and systems. Long-distance wireless communication is the first example, because in this case the key challenge is the extremely large size of the antenna required to have electromagnetic wave radiation in MHz regime. For example, an electrically-small single-turn loop antenna has a radiation resistance, and therefore efficiency, roughly proportional to frequency 4 × D 4 , where D is the diameter of the loop; an antenna with D = 1 m radiates with an efficiency of 18% at 10.1 MHz 31 . Therefore, we can expect ~7% RF power radiation at 40 MHz with D = 20 cm which is a reasonable size for integration onto human body or cloth, or toys. Wireless communication can also be done in pulse mode. For example, super regenerative receivers, which are in use since 1940s 32 , turn on an oscillator just for a few μs, receive the data, and then turn it off again for a long time, similar to how we operated the device in this work.
As the second example, inductor-based switching power converters can be mentioned. Power supply is an essential part of any electronic system, and inductor-based switching power converters are the most energy efficient circuits widely used in up/down DC-DC converters and battery chargers. Key elements of such circuits are FET switches, diodes, inductors and capacitors. A typical switching circuit working with <1 μs pulses would require spiral inductors in the 10 μH range which can be easily fabricated by inkjet printing or evaporating a single metal layer on a plastic substrate with outer diameter <10 cm. Organic rectifying diodes already can work at tens of MHz 33,34 . Printed flexible 10 μF range multi-layer capacitors for power conversion applications have been demonstrated 35 and show a high relative dielectric constant of 15-22 up to 10 MHz. In addition to the high f T , the high current drive capability of the OPBT is also an advantage for compatibility with this application. In this context, it also worth to mention that flexible batteries and organic solar cells are also already developed and even commercialized.
The pulse measurement results are also relevant for digital logic circuits in which we only have transient currents in the transistors.

Methods
The OPBT presented here is built in a single chamber UHV-tool. The glass substrate is previously cleaned with N-Methylpyrrolidone, distilled water, ethanol, and an Ultra Violet Ozone Cleaning System. By using thermal vapor deposition at high vacuum conditions (p < 10 −7 mbar), the OPBT stack layers are deposited through laser-cut, stainless steel shadow masks. The layer stacks, evaporation rates, and treatments are: Al 100 nm (1.0 Å/s); Cr 10 nm (0.1 Å/s); i-C 60 100 nm (1.0 Å/s); Al 15 nm (1 Å/s); 15 min oxidation at air; i-C 60 100 nm; 2 times (perpendicular to each other) SiO 100 nm with a free stripe of 200 μm (1.0 Å/s); n-C 60 20 nm (0.4 Å/s) co-evaporating C 60 with W 2 (hpp) 4 (purchased from Novaled GmbH, Dresden) with 1 wt%; Cr 10 nm (0.1 Å/s); Al 100 nm (1.0 Å/s); encapsulation in a nitrogen atmosphere using UV cured epoxy glue without UV exposure on the active area; annealing for 2 h at 150 °C in a nitrogen glove-box on a heat plate, for positively affecting both current-transmission through the base and regeneration of the air-exposed C 60 36,37 .
The 50 pF capacitor across the op-amp O 1 is a mica capacitor, but other capacitors in the circuit are polypropylene or polyester film capacitors. For C E1,2,4 polypropylene is preferred. Figure 1(a) has been taken using the Nikon microscope ECLIPSE LV 100 ND. Dot lines are added around the base electrode and the active area. This photo is not recolored; base and collector look colorful because of the materials deposited on top of them; emitter electrode looks white because it is on the top. Data availability. All data generated or analysed during this study are included in this published article.