Fabrication and Characterization of Single GaN Microwire Radiation Sensors: Assessment of the detection capabilities and radiation resistance

UNIVERSIDADE DE LISBOA INSTITUTO SUPERIOR TÉCNICO Fabrication and Characterization of Single GaN Microwire Radiation Sensors: Assessment of the detection capabilities and radiation resistance Dirkjan Verheij Supervisor:Doctor Katharina Lorenz Co-Supervisors:Doctor Susana Isabel Pinheiro Cardoso de Freitas Doctor Jorge Manuel Dos Santos Ribeiro Fernandes Thesis approved in public session to obtain the PhD Degree in Technological Physics Engineering Jury final classification: Pass with Distinction and Honour 2023 UNIVERSIDADE DE LISBOA INSTITUTO SUPERIOR TÉCNICO Fabrication and Characterization of Single GaN Microwire Radiation Sensors: Assessment of the detection capabilities and radiation resistance Dirkjan Verheij Supervisor:Doctor Katharina Lorenz Co-Supervisors:Doctor Susana Isabel Pinheiro Cardoso de Freitas Doctor Jorge Manuel Dos Santos Ribeiro Fernandes Thesis approved in public session to obtain the PhD Degree in Technological Physics Engineering Jury final classification: Pass with Distinction and Honour Jury Chairperson: Doctor Maria Teresa Haderer de la Pe˜ na Stadler, Instituto Superior T´ ecnico, Universidade de Lisboa Members of the Committee: Doctor Jo¨ el Eymery, CEA Grenoble, France Doctor Katharina Lorenz, Instituto Superior T´ ecnico, Universidade de Lisboa Doctor Henrique Leonel Gomes, Faculdade de Ciˆ encias e Tecnologia, Universidade de Coimbra Doctor Lu´ ıs Humberto Viseu Melo, Instituto Superior T´ ecnico, Universidade de Lisboa Doctor Marco Ant´ onio Baptista Peres, Instituto Superior T´ ecnico, Universidade de Lisboa Funding Institutions Fundação para a Ciência e a Tecnologia 2023 ii Acknowledgments Naturally, the work carried out in the scope of this thesis would not have been possible without the support and contributions of many people. First of all, a sincere and heartfelt thank you is in place for my supervisor, Dr. Katharina Lorenz. I joined her research group when I started working on my Master thesis and, as far as supervising students goes, I do not think it can be done in a much better way. Her inexhaustible patience to discuss experimental results and theoretical details, as well as her trust and support were fundamental for the successful conclusion of this thesis. Besides this, her kindness and optimism creates a very pleasant working environment in the laboratory. I would also like to express my genuine gratitude to my other supervisors, prof. Susana Cardoso, who introduced me into the world of microfabrication and nanotechnology, and was always available to discuss details related with the fabrication process, and to prof. Jorge Fernandes, who always showed interest in the work and was able to provide valuable insights. I am also extremely grateful to Lu´ ıs Cerqueira Alves without whom all measurements carried out using the nuclear microprobe at LATR would not have been possible. Despite the many, and sometimes frustrating, hours spend in the small room where the proton irradiations are carried out, Lu´ ıs never rejected a challenge or said no whenever I asked if we could do just one more experiment or one more measurement. The gratitude also applies to Marco Peres, responsible for designing and building most of the experimental setups used for the experimental work. His energy and motivation are truly unmatched. Also fundamental for this work were Dr. Jo¨ el Eymery and Dr. Christophe Durand, from the Universit´ e Grenoble Alps, who kindly provided the microwires that were used for the development of the radiation detectors in this thesis. Furthermore, I would like to thank Dr. Milko Jakˇ si´ c and Milan Vi´ centijevi´ c, from the Ruder Boˇ skovi´ c Institute, for carrying out the IBIC measurements and Dr. Gw´ enol´ e Jacopin for making the EBIC measurements. I am also grateful to Dr. Reinhard Schwarz for allowing me to use his equipment and helping me to carry out the photocurrent spectroscopy measurements. I also would like to thank all the process engineers and researchers working at INESC MN who in some way helped me during the microfabrication of the detectors. Since this work would not have been possible without financial support, I would like to acknowledge the portuguese Fundac¸ ˜ ao para a Ciˆ encia e Tecnologia, who financed the work through the AIM doctoral program and the NASIB and DEOFET projects, as well as the financial support from the Radiate project to perform the IBIC measurements. Finally, I want to thank all my friends, family and colleagues who in some way contributed to make my life a very enjoyable experience, even at times during my PhD when this was not straightforward. I especially thank Ana, for always encouraging me and reminding me of the joy and fulfillment that comes with pursuing my passion, but most importantly, for being the person that is always and unconditionally there for me. iii iv Resumo Microfios de nitreto de g´ alio s˜ ao estruturas muito interessantes devido ` as suas caracter´ ısticas intr´ ınsecas e, durante a ´ ultima d´ ecada, diversos dispositivos optoeletr´ onicos, baseados em microfios de GaN foram demonstrados. Sendo um semicondutor com uma elevada resistˆ encia ` a radiac¸ ˜ ao ionizante e est´ avel ` a altas temperaturas, os microfios de GaN tamb´ em apresentam um potencial para serem aplicados como sensores de radiac¸ ˜ ao em ambientes extremos. Os microfios tˆ em uma altura at´ e 30 µm e um diˆ ametro de 2 µm e, devido ` a geometria e dimens˜ ao reduzida, a sua estrutura cristalina ´ e praticamente livre de defeitos. Adicionalmente, na sua configurac¸ ˜ ao casca-n´ ucleo, os microfios apresentam uma junc¸ao p-n na direc¸ ˜ ao radial, o que promove a detec¸ ˜ ao de carga. Neste trabalho, sensores de radiac¸ ˜ ao baseados num ´ unico microfio foram fabricados usando diversas t´ ecnicas de microfabricac¸ ˜ ao e caracterizados por diferentes t´ ecnicas experimentais. Quando irradiados com prot˜ oes com uma energia de 2 MeV, a corrente el´ etrica aumenta at´ e quatro ordens de grandeza relativamente ` a corrente escura com tempos de resposta inferiores ` a 20 ms. Para al´ em disto, experiˆ encias de carga induzida por feixes de i˜ oes (IBIC) permitiram estudar a eficiˆ encia de colec¸ ˜ ao de carga (CCE). Os resultados indicaram de CCE valores m´ aximos na ordem dos 30%. Durante a irradiac¸ ˜ ao com prot˜ oes de energia igual a 1 MeV e 2 MeV, verificou-se que at´ e uma fluˆ encia de 1x1014 prot˜ oes/cm2a modificac¸ ˜ ao das propriedades el´ etricas dos sensores ´ e negligenci´ avel. Para fluˆ encias mais elevadas a corrente comec¸a a diminuir, com uma queda mais abrupta para fluˆ encias entre 5x1014 prot˜ oes/cm2e 1x1015 prot˜ oes/cm2. No entanto, mesmo ap´ os irradiar os dispositivos com uma fluˆ encia igual a 5x1015 prot˜ oes/cm2, os sensores continuam a mostrar uma resposta ` a radiac¸ ˜ ao incidente. Tendo em conta as vantagens inerentes dos microfios, nomeadamente no que toca a possibilidade de detetar radiac¸ ˜ ao ionizante e boa estabilidade em ambientes extremos, verifica-se o seu potencial na aplicac¸ ˜ ao em diversos campos onde se requer sensores com uma s´ olida resistˆ encia ` a radiac¸ ˜ ao ionizante. Palavras-chave: Sensores de radiac¸ ˜ ao, GaN, microfios, efeitos de radiac¸ ˜ ao em semicondutores, induc¸ ˜ ao de carga por feixes de i˜ oes v vi Abstract Gallium nitride microwires are very interesting structures due to their intrinsic characteristics and, over the last decade, several optoelectronic devices based on GaN microwires have been demonstrated. Being a semiconductor with a high resistance to ionizing radiation and stable at high temperatures, GaN microwires also have a potential to be applied as radiation sensors in extreme environments. The microwires have a height of up to 30 µm and a diameter of 2 µm, and due to their specific growth kinetics, their crystal structure is practically free of dislocations. Additionally, in their core-shell configuration they feature a p-n junction in the radial direction, where created charges are readily separated and contribute to the generated signal. In this work, radiation sensors based on single microwires were fabricated, microfabrication processes were developed and devices characterized by different experimental techniques. When irradiated with protons with an energy of 2 MeV, the electric current increases up to four orders of magnitude with respect to the dark current, while presenting response times below 20 ms. In addition, ion beam induced charge (IBIC) experiments allowed the charge collection efficiency (CCE) to be determined. The results indicated maximum values in the range of 30%. During irradiation with 1 MeV and 2 MeV protons, it was found that up to a fluence of 1x1014 protons/cm2the modification of the electrical properties of the sensors is negligible. For higher fluences the current starts to decrease, with a more abrupt drop between 5x1014 protons/cm2and 1x1015 protons/cm2. However, even after irradiating the devices at a fluence of 5x1015 protons/cm2, the sensors still show a response to the incident radiation. Considering the inherent advantages of microwires, namely in terms of the possibility to detect ionizing radiation and good stability, this thesis paves the way for potential applications in several fields where sensors with a solid resistance to ionizing radiation are required. Keywords: Radiation sensors, GaN, microwires, radiation effects in semiconductors, ion beam induced charge vii xiv List of Figures 1.1 Crystal structure of wurtzite GaN and respective lattice constants . . . . . . . . . . . . . . 5 2.1 Band diagrams of the p-n junction in equilibrium (a), reverse bias (b) and forward bias (c). 17 2.2 a) Equivalent circuit of the p-n junction including a series and a shunt resistance; (b-f) Simulation of the I-V curves obtained by solving eq. 2.11 when giving different values I0, n,RSand RSh........................................... 19 2.3 Schematics of the mechanisms responsible for leakage currents in p-n junctions: SRH generation and recombination (a); Tunneling assisted by traps located in the bandgap (b). 21 2.4 Ideal band diagram of a metal/n-type semiconductor junction (a) and a metal/p-type semiconductor junction (b). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.5 Schematic of the carrier transport mechanisms for metal/n-type semiconductor junctions with different doping concentrations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.6 Schematic of a simple photoconductor together with the band diagram representing the process of photoexcitation from the valence to the conduction band. . . . . . . . . . . . . 26 2.7 Schematics of the band diagrams of a p-n junction diode and the Schottky diodes where the movement of the generated charge carriers is indicated . . . . . . . . . . . . . . . . . 27 2.8 Simulation of the I-V curves obtained by solving eq. 2.27 when giving different values to Ipc,I0,n,RSand RSh ...................................... 29 3.1 Plots showing the ideal currents flowing through the semiconductor after the creation of electron-hole pairs by the incident radiation and the collected charge as a function of the time. In the right plot, t1and t2represent the time after which the faster and slower carriers are collected, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 4.1 Schematic and SEM images of the n-type microwires (a,b) and the p-n core-shell microwires (c,d)). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 4.2 (a) Multibeam TEM image of a microwire along its axis; (b,c) Bright field TEM images of the sapphire/GaN interface, with the bent dislocations indicated by the arrows. (d-f) Schematic representation and dark field TEM images of a GaN shell grown on top and around the GaN:Si core. Reprinted from [154], with the permission from AIP Publishing . 45 xv 4.3 SEM images showing the p-GaN shell for microwires from batch T2766 (a) and T2360 (b). In the latter we can see that the shell is relatively short and has significant voids and discontinuities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 4.4 Schematics of the main steps of the microfabrication process of single microwire detectors with a p-n junction geometry. The process used for n-type detectors is the same but only requires one metal contact deposition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 4.5 a) Map that was used for the lithography of six inch wafers; b) Mask that was used for the lithography of the alignment markers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 4.6 Optical microscope image of the alignment markers after the lithography. The dark areas indicate the presence of photoresist. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 4.7 a) Six inch Si/SiO2wafer after the etching and PR strip steps; b) Optical microscope images showing the final alignment markers. . . . . . . . . . . . . . . . . . . . . . . . . . 50 4.8 a) Photograph showing how the droplet dispersion is done; Optical microscope images of the substrate after dispersion (b) and after dispersion and cleaning (c). We can see that after cleaning the distribution of wires is uniform and that no agglomerations of wires are formed. .............................................. 50 4.9 Visual workflow of the WireFinder code. The optical microscope image is given as input and through image processing, the code identifies the alignment crosses and wires that are present in the sample site. It then chooses one of the available wires and defines the coordinates of the vertices of the contact paths. These are written to a script that results in the shown CAD drawing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.10 Optical microscope image after the second lithography (a) and the metal-lift after the first contact deposition (b) for the n-GaN contact. . . . . . . . . . . . . . . . . . . . . . . . . . 53 4.11 Optical microscope image of a single wire device during processing, namely after the second lithography (a), the first lift-off and third lithography (b) and the second lift-off (c). . 54 4.12 Optical microscope image of as-deposited TiWN/Al/TiWN contacts (a) and annealed TiWN/Al/TiWN contacts (b). We can see formation of irregularities on the metal film, when using a higher magnification the bubble-like defects become very evident. . . . . . . . . . . . . . . . . . 55 4.13 SEM images of an n-type wire devices (a) and a core-shell wire device (b). An electron beam with an acceleration of 10 kV and apperture of 10 µm were used. The images were taken with an angle of 45◦between the beam and the substrate (colors are only for representation). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.14 SEM images showing the contact regions of an n-type wire sample. Here we can clearly see the existence of ”rabbit-ear” structures delineating the contact paths. Furthermore, no cracks in the contact film are visible. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.15 Optical microscope images of a representative sample taken at the end of the microfabrication process. In this specific case the n-GaN and p-GaN contacts are Ti/Au and Ni/Au respectively............................................. 56 xvi 4.16 Microscope image of a wire-bonded sample (a) and photographs of the final devices fabricated on Si/SiO2and sapphire substrates before and after integration onto a chip-carrier (b) ................................................. 57 5.1 Map of the accelerator infrastructure at LATR with the accelerators and irradiation lines that were used in this work highlighted in blue. . . . . . . . . . . . . . . . . . . . . . . . . 59 5.2 Picture and schematic of the nuclear microprobe experimental setup at CTN. The schematic indicates the position of the sample with respect to the beam as well as the different components of the experimental setup (not in scale). . . . . . . . . . . . . . . . . . . . . . . . 60 5.3 PIXE map obtained when irradiating a 2000 mesh copper grid with 2 MeV protons. The total scan area is equal to 53x53 µm2............................. 61 5.4 Alignment procedure of the sample with respect to the beam during measurement with the microprobe. The first alignment is done optically and (a) represents the first scan with an area of 264 by 264 µm2; (b) represents the second scan with an area of 264 by 264 µm2after vertical alignment; (c) represents the third scan with an area of 106 by 106 µm2; (d) represents the final alignment scan with an area of 106 by 106 µm2after horizontal alignment. Note that the yellow dots correspond to Au counts and the red dots to Ga counts. 63 5.5 Photographs of the Tandem accelerator at CTN and schematic of the experimental setup with the relevant components (not in scale). . . . . . . . . . . . . . . . . . . . . . . . . . . 64 5.6 Map of the ion beam laboratory facility at RBI and photograph showing the interior of the chamber where the IBIC experiments are carried out. . . . . . . . . . . . . . . . . . . . . 65 5.7 Nuclear and electronic stopping as a function of the ion energy for impinging H (a) and Si (b) ions in GaN as obtained via SRIM simulations . . . . . . . . . . . . . . . . . . . . . . . 66 5.8 Range of Si ions and protons in GaN as a function of the ion energy as obtained via SRIM simulations. The table in the inset indicates the obtained ranges for the relevant energies within the scope of this work. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 5.9 Simulated ionization profiles for 2 MeV hydrogen ions in GaN (a) and 1 MeV Si ions in GaN (b), obtained via SRIM simulations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 5.10 Simulated vacancy profiles for 2 MeV and 1 MeV hydrogen ions in GaN (a) and 1 MeV and 750 keV Si ions in GaN (b), obtained via SRIM simulations. . . . . . . . . . . . . . . . 68 5.11 Simulated NIEL profile in GaN irradiated with 1 MeV and 2 MeV protons (a) and 750 keV and 1 MeV Si ions (b), obtained through SRIM simulations. . . . . . . . . . . . . . . . . . 69 xvii 5.12 Photocurrent measured with a commercial Si p-i-n photodiode (Thorlabs FDS010) while illuminating it with the UV LED (365 nm, Thorlabs M365D1). The inset on the left indicates the parameteres extracted from the linear regression of the data as well as the R-square parameter, which indicates the good linearity of the photocurrent with the LED input current. The inset on the right represents the equation that is used to convert the measured photocurrent into the optical power. Apd and Rpd are the active area and the responsivity of the photodiode at the considered wavelength, respectively. In this case, Apd = 0.8mm2 and Rpd = 0.051 A/W....................................... 70 5.13 Experimental setup, available at the laboratory for photoconductive measurements at the Physics department IST, used for the photocurrent spectroscopy measurements that use a lock-in amplifier to amplify the output signal. . . . . . . . . . . . . . . . . . . . . . . . . . 71 5.14 IBIC maps made while scanning the copper grid and using different scan sizes. Also indicated are the bar and hole width of the grid. . . . . . . . . . . . . . . . . . . . . . . . 73 5.15 Energy calibration data for the IBIC experiments . . . . . . . . . . . . . . . . . . . . . . . 74 6.1 I-V curves obtained when measuring as-fabricated detectors based on GaN n-type microwires with Cr/Au contacts. As shown, devices fabricated in the same way can yield different I-V characteristics, more specifically, we have non-linear asymmetric (a); nonlinear almost symmetric (b); linear (c) and rectifying devices (d). . . . . . . . . . . . . . . 76 6.2 Comparison between the dark I-V curves before and after annealing for n-type microwire samples with Ti/Au contacts (a); Cr/Au contacts (b) and TiWN/Al/TiWN contacts (c). The annealing conditions are specified in the inset of each graph. We can see that the annealing lowered the overall conductivity of the samples with Ti/Au and Cr/Au contacts indicating enhancements of the Schottky barriers but transformed the TiWN/Al/TiWN into ohmic contacts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 6.3 Comparison between the dark current and photocurrent for an n-type detector with a linear I-V characteristic (a) and an asymmetric I-V characteristic (b). We can observe a photocurrent in both situations but whereas in (a) the photocurrent is the same for both negative and positive bias, in (b) the photocurrent has a larger magnitude for positive bias. 79 6.4 Signal-to-noise ratio plotted as a function of the applied bias for an n-type microwire detector with linear contacts (a) and asymmetric contacts (a). We can see that for the linear sample the signal-to-noise ratio is also independent of bias. For the asymmetric sample this is not the case and when the more rectifying contact in reverse bias the signal-tonoise ratio is larger. The highest signal-to-noise ratio is achieved at low positive voltages however, with a maximum value close to 2. I note that the discontinuities at 0 V are due to the division by zero and that the signal-to-noise ratio of 1 for the linear sample close to 2 V and -2 V are due to the current compliance that forces the currents to be equal. . . . 79 xviii 6.5 Transient I-t curves at a bias of 2 V and -2 V for an n-type microwire detector with linear contacts (a) and asymmetric contacts (b). The response times are similar except for the photocurrent of the asymmetric detector measured at negative bias, which presents a significant increase in rise and decay time. . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 6.6 (a) Photocurrent of an n-type microwire detector as a function of the LED input current and results obtained through fitting the data to the power law function; (b) Photocurrent spectroscopy data at 1 V and 3 V. We can see that the spectra are independent of the bias. The dashed line represent the bandgap wavelength/energy of GaN. . . . . . . . . . 81 6.7 Dark I-V curves for a representative p-n junction microwire detector with Ni/Au contacts on the p-GaN extremity, in logarithmic scale (a) and linear scale (b). We can clearly observe a rectifying behavior although significant deviations exists with respect to the ideal p-n junction diode equation given by eq. 2.10. . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 6.8 a) Measured forward current as a function of bias where three distinct regions can be identified. Region I, III and IV have linear slopes in the semilog plot and the ideality factor can be extracted; b) Schematic representing the equivalent circuit of a p-n junction wire detector. The circuit contains three diodes, corresponding to both metal-semiconductor interfaces and the p-n junction. Furthermore, shunt resistances for each diode as well as the resistances of the neutral regions were included. . . . . . . . . . . . . . . . . . . . . . 85 6.9 Reverse bias dark current measured for different p-n junction microwire samples in linear (a) and logarithmic (b) scale. Although the magnitude differs, the dependence of the current on bias is similar for all samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 6.10 Reverse bias current of two different p-n junction microwire detectors, presented in a double logarithmic plot. The linearity indicates that the current as a function of the applied bias follows a power law. For the two represented samples we obtain u-values of 4.06 and 3.75 and, in general, this value is larger than 3. . . . . . . . . . . . . . . . . . . . . . 89 6.11 Plots showing IR/VRas a function of p(Vbi +VR)for three p-n junction microwire samples. We can observe linearity of samples A and B between 0.5 and 1 V-1/2 while the linearity of the curve corresponding to sample B is slightly worse. The linear nature of the curves indicates that the data follows the Zener model described by eq. 2.16 . . . . . . . 90 6.12 Fits of the reverse bias current using the Zener tunneling model including a series resistance, given by eq. 6.5 for three p-n junction microwire samples. The left and right graphs show the current in linear and logarithmic scale respectively. The obtained fit parameters are indicated for each sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 6.13 Dark I-V curves measured before and after annealing the Ni/Au contacts in air at 500 ◦C for 120 seconds for two representative p-n junction microwire samples. In graph a) we observe the fast exponential increase of the current around 3.7 V that does not occur in b). 92 xix 6.14 Dark I-V curves of three different p-n junction microwire samples, fabricated with annealed Ni/Au contacts on the p-GaN extremity in linear (a) and logarithmic (b) scale. We can observe that for each sample the fast exponential increase activates at a different forward voltage. .............................................. 92 6.15 Dark I-V curves measured before and after annealing for a p-n junction microwire sample fabricated using Cr/Au contacts (a) and a p-n junction microwire sample fabricated with Ni/Au contacts (b). The sample fabricated with Cr/Au contacts was firstly annealed in N2 and afterwards in air whereas the sample fabricated with Ni/Au contact was first annealed in air and then in N2. All annealing treatments were carried out at 500 ◦C for 60 seconds. 93 6.16 I-V curves measured in the dark and while illuminating a representative p-n junction microwire sample with the UV LED in logarithmic scale (a) and linear scale (b). . . . . . . . 94 6.17 a) SEM image of a p-n junction sample indicating the region used to estimate the active area of the detector; b) Short-circuit current of different samples normalized to the irradiance as a function of the active area. The linear dependence of Isc on the active area shows that the estimation of the latter is in good agreement with the real value. . . . . . . 95 6.18 Dark and photocurrent for the bias range in which the p-n junction microwire detector works in the photovoltaic regime. (a) Measurement made on a sample that was not annealed while (b) and (c) show measurements made on samples that were annealed. The insets show the dark I-V curves measured over the entire bias range. In the right graph we can see that there is no trace of an ”s” shape in the photocurrent I-V curve whereas we can clearly see the kink in the center graph. . . . . . . . . . . . . . . . . . . . . . . . . 95 6.19 Dark and photocurrent I-V curves when illuminating the p-n junction microwire sample with different irradiance (a). By extracting the value of the photocurrent at a certain bias we can also plot the photocurrent as a function of the optical power, here the markers represent the experimental data and the line the fit to the power law function (data is shown in log-log scale and in linear scale in the top and bottom graph respectively) (b). Extracted βparameters obtained by fitting the photocurrent using the power law function (c).................................................. 98 6.20 Photocurrent fits at -0.5 V, 0 V and 0.5 V when only using the data points measured when using the diffuser lens (a), when using the normal lens (b) and when using all data points. 99 6.21 Photocurrent and responsivity as a function of the irradiance for a representative p-n junction microwire detector while applying different voltages to the sample. . . . . . . . . . . . 100 6.22 Responsivity of a representative p-n junction microwire detector as a function of the bias for different values of irradiance in linear (a) and logarithmic (b) scale. We can see that between 0.5 V and -0.75 V the responsivity does not vary with the irradiance but that at higher reverse bias, the responsivity increases at a higher rate for lower values of irradiance.101 xx 6.23 Transient photocurrent measured using a p-n junction microwire detector, shown for different applied biases. In (a) we can see that in self-powered mode the rise and decay times of the sensor are faster than the sampling rate of the measurement system; When we increase the reverse voltage a small persistent component appears for -2 V that becomes stronger at -4 V. In forward bias the response time increases dramatically as a consequence of the reverse biased Schottky contact and stronger relative influence of the photoconductive mechanism. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 6.24 Spectral response as a function of the photon energy at zero bias for two distinct p-n junction microwire samples (a) and two different applied biases (b). Also represented in the right graph is the Urbach tail fit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 7.1 Transient I-t curves measured while irradiating the n-type microwire detector using a scan size equal to 264 by 264 µm2and while applying a bias of 2 V to the detector (a); Optical microscope image and PIXE map measured during the scan where the yellow dots corresponds to the Au signal and the red dots to Ga signal (the scale refers the optical microscope image) (b); Graphs showing the shape of an individual peak (c) and of the persistent ionocurrent decay (d) in more detail. The inset in (d) shows the measured I-V curve of the detector. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 7.2 Transient I-t curves measured while irradiating an n-type microwire detector with linear contacts with 2 MeV protons and using a scan size equal to 53 by 53 µm2, when applying different voltages to the terminals of the detector. The initial dark current was subtracted from the total current for each measurement to allow comparison between the data measured at different applied voltages despite the loss in conductivity. I note that measurements were carried out using the following order: V= 1 V, V=−1V, V=−2V, V= 2 V............................................... 107 7.3 Ionocurrent signals measured during different line scans for an n-type microwire detector with asymmetric contacts. Optical microscope image of the device as well as the PIXE map with the lines that were scanned indicated in yellow (a); Ionocurrent obtained when applying a bias of 1 V and -1 V to the right contact (b); The I-V curve of the device, measured before the first line scan, is shown in the inset of the right graph. . . . . . . . . 109 7.4 Transient I-t curves measured while scanning a p-n junction microwire detector using a an area of 106x106 µm2(a). The same data but zoomed in on two of the peaks (b); PIXE maps obtained while irradiating the detector using the scan size (c). We can see the contribution from the different materials that compose the wire and the metal contacts. 110 7.5 Transient I-t curves measured while irradiating a p-n junction microwire detector when using different scan sizes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 xxi 7.6 (a) Static I-V curves measured when irradiating specific areas of a p-n junction wire, namely the core-shell region (top graph) and the neutral n-GaN region (bottom). The areas that were irradiated are shown in the insets; (b) Transient I-t curves measured while irradiating the core-shell region indicated in the inset of the top graph if (a), and while applying different voltages to the detector, in logarithmic (top graph) and linear (bottom graph)scale. ........................................... 112 7.7 Transient I-t curves measured while irradiating different sections of a p-n junction microwire detector. The areas that were irradiated are indicated in the PIXE map and the right and left contacts corresponds to the n-GaN contact and p-GaN contacts, respectively. In all cases we applied a bias of 0 V. We can see that the ionocurrent signal is high when irradiating the core-shell region and gradually decreases as we move towards the neutral n-GaN region. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 7.8 Transient I-t curve measured while irradiating a p-n junction microwire detector using a static beam. The inset shows a PIXE map measured before the irradiation with the static beam, where the white dots represent the specific locations that were irradiated, including five points on top of the core-shell region and four points far away from the detector. We can see a very large iono-to-dark current ratio for the first static irradiation. In the subsequent periods the beam moves to the core-shell region, the signal decreases due to the created irradiation damage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 7.9 (a) PIXE map obtained during the ionocurrent map experiments, including the points of the irradiated region which are marked in green. The right graph shows the raw transient data measured during the experiment. The detected signal is shown in orange and the voltage signal that controls the beam shutter is shown in blue. In the inset the full time range is shown. (b-d) Current maps showing the measured data. Each pixel in the map corresponds to a point in the PIXE map. The applied bias is equal to 0 V (b), -1 V (c) and -2V(d)............................................... 115 7.10 SEM image and IBIC maps of a p-n junction microwire sample irradiated with 750 keV Si ions while applying different voltages to the p-GaN terminal of the detector. In the SEM image, the relevant dimensions are indicated and the p-GaN contact corresponds to the bottom electrode. In the IBIC maps we can see that, when the applied reverse bias increases, the measured CCE also increases. We also measure a non-zero CCE when applying a positive voltage, indicating that we are not completely cancelling out the SCR of the p-n junction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 7.11 SEM image and IBIC maps of a p-n junction microwire detectors irradiated with 1 MeV Si ions while applying different reverse bias voltages to the p-GaN contact. The bottom contact corresponds to the p-GaN contact. . . . . . . . . . . . . . . . . . . . . . . . . . . 117 7.12 Histogram data, showing the number of ions that originate a certain charge collection as a function of the energy, corresponding to the IBIC maps shown in figure 7.10 . . . . . . . 118 xxii 7.13 Histogram data, showing the number of ions that originate a certain charge collection as a function of the energy, corresponding to the IBIC maps shown in figure 7.11 . . . . . . . 119 7.14 Schematic showing the cross-section of the core-shell wire and the difference in terms of ion trajectory in the SCR when the Si ion hits the central part of the wire and when the ion hits the border of the wire (a); Impact of this difference indicated in the IBIC map. Also identified is the contact region (b). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 7.15 Projection of the CCE at different applied bias along the diameter of the wire for a p-n junction microwire detector irradiated with 750 keV Si ions (a-c) and for a detector irradiated with 1 MeV Si ions (d-f). The projections were carried out on the maps shown in figures 7.10 and 7.11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 7.16 (a) Comparison between the average CCE at an applied bias of -1 V along the microwire diameter for the different ion energies; (b) Schematic showing how the penetration depth may affect the magnitude of the CCE that is measured due to the hexagonal core-shell geometry.............................................. 122 7.17 IBIC maps of the same p-n junction microwire detector measured without applying a bias at the beginning of the experiment and after performing several irradiation measurements using 1 MeV Si ions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 7.18 IBIC maps and corresponding CCE profiles along the length of a p-n junction microwire detector at different applied biases. Both the average CCE of the left and top facets are presented. Maps were obtained while irradiating the detector with 1 MeV Si ions. The p-GaN contact is located in the bottom region of the IBIC map which corresponds to L between 2.5 µm and 7.5 µm. .................................. 124 7.19 IBIC maps and corresponding CCE profiles along the length a p-n junction microwire detector at different applied biases. Both the average CCE of the left and top facet are presented. Maps were obtained while irradiating the detector with 750 keV Si ions. The p-GaN contact is located in the bottom region of the IBIC map which corresponds to L between 10 µm and 15 µm. ................................... 125 7.20 Average CCE of a p-n junction microwire detector considering the different sections in the microwire as a function of the applied bias for 1 MeV Si ions (a) and 750 keV Si ions (b). Also shown is the comparison between the total average CCE for both irradiation energies (c).................................................. 126 8.1 Dark I-V curves measured in-situ after irradiating a representative p-n junction microwire sample with 2 MeV protons (a) and 1 MeV protons (b). We can observe that the conductivity of the device decreases as the irradiation fluence increases. It also possible to see that the shape of the I-V curve is modified indicating a possible modification of the transport mechanisms that define the I-V characteristics. . . . . . . . . . . . . . . . . . . . 132 xxiii SCR Space-charge Region. SEM Scanning Electron Microscope. SPA Semiconductor Parameter Analyzer. Space-charge Limited Current SCLC. SRH Shockley-Read-Hall. SRIM Stopping Range of Ions in Matter. UV Ultraviolet. VRH Variable Range Hopping. WBGS Wide Bandgap Semiconductor. xxx Nomenclature Ion beam parameters ΦBeam fluence ΦABeam fluence per area scan θbeam Beam flux Abeam Beam area Ascan Beam scan area DdDisplacement damage dose Ibeam Beam current Detector figures of merit βParameter of linearity of the photocurrent with optical power ηeExternal quantum efficiency τdDecay time τrRise time CCE Charge Collection Efficiency EUUrbach energy Gpc Photocurrent gain RResponsivity Material properties εeElectronic stopping power εiIonization energy εnNuclear stopping power εsPermittivity xxxi Aabs Optical absorption EdDisplacement energy ETTrap energy Ieconstant dependent on the mean excitation and ionization energies of the material ZAtomic number Physical constants cSpeed of light hν Photon energy kbBoltzmann constant meElectron mass qElementary charge Semiconductor properties λnElectron drift length λpHole drift length µnElectron mobility µpHole mobility σConductivity τCarrier lifetime DnElectron diffusion coefficient DpHole diffusion coefficient nElectron concentration NAAcceptor concentration NDDonor concentration niIntrinsic carrier concentration pHole concentration xxxii Chapter 1 Introduction Our capability to sense the environment around us through, for example, our sight or touch, is fundamental for our survival. It allows us to identify danger and situations in which some kind of action, like touching a very hot surface, can have a negative impact on our well-being. However, our senses are not perfect and there are forms of radiation that do not trigger a warning message in our brains yet can severely damage our DNA and the cells that compose our organs. In some cases, such as sunburns, the effects of the created damage appears relatively fast but in other cases, the consequences of exposure to radiation can take years to develop and may originate serious health problems. One of the most dangerous forms of radiation for living creatures is ionizing radiation, which is defined as radiation that carries enough energy to ionize atoms or molecules. Examples of ionizing radiation are x-rays, gamma rays or particles such as protons, neutrons or electrons. Although at low doses the negative effects of ionizing radiation are negligible and it can even be used for the benefit of human kind, exposure to high doses of radiation is very dangerous. For this reason, it is of great importance to develop systems that monitor the radiation environment that surrounds us, so that we can take advantage of the benefits of ionizing radiation and explore regions characterized by extreme radiation environments, without compromising our safety. The main component of these systems is the sensing component. A structure that in some way converts incident radiation into a signal that we can read and interpret. Historically, perhaps the most famous kind of radiation detector is the Geiger-M¨ uller counter. This device counts the number of incident particles and transmits the detection through the famous ticking noise. Nevertheless, this device is not perfect and presents several limitations, mainly in terms of measuring high radiation doses and radiation energy. Additionally, it is a large device, especially when compared with today’s standards regarding the size of electronics and devices [1]. Nowadays, radiation detecting systems are mostly based on semiconductors. In the experiments that are part of the Large Hadron Collider at CERN, for instance, practically all detection systems are based on silicon [?]. The reason for this is, as will become evident over the course of this thesis, that semiconductors are materials that present several distinct advantages that make them useful for radiation detection. Nevertheless, there are also some downsides to semiconductors, the main one being that semicon1 ductors can also be degraded by the incident radiation which, over time, leads to a loss of their detection capacity. Consequently, there is a large research interest in materials that present a high resistance to ionizing radiation. As will be shown in this thesis, one of these materials is gallium nitride (GaN). Since the beginning of the century, besides new materials, novel ways to grow, shape and modify semiconductors have also become increasingly popular. Nowadays, it is possible to create structures in many different configurations with feature sizes down to the scale of the atoms themselves. Naturally, as the scale of the structures decreases, the physical mechanisms that define properties of the devices fabricated with them also change, which can lead to an increase in sensitivity and efficiency. Nanowires, microwires, quantum dots, nanoparticles, nanotubes, nanoribbons are just some examples of the large amount of geometries that are nowadays commonly used in nanotechnology. The advances in semiconductor and nanotechnology have established the semiconductor as the cornerstone of the current technological revolution. Currently, radiation detectors and radiation resistant electronics are, in part, responsible for allowing astronauts to travel and stay safely in outer space, for enabling the generation of nuclear energy in power plants and for allowing the conduction of high energy radiation physics experiments. Nevertheless, there are still many questions that need to be answered regarding the behavior and properties of semiconductors when exposed to different types of radiation, as well as engineering challenges that need to be solved before semiconductor radiation detectors can be fully used in specific applications. An example of such a challenge is the real-time and in-vivo measurement of the radiation dose to which a patient is exposed during proton or brachytherapy. During these treatments a proton beam or radioactive seeds are used respectively. The main advantage of these therapies is that they provide a better spatial control, allowing a better selectivity in terms of the irradiation of healthy and unhealthy tissue. However, misalignment or deviations from the simulated dose of the treatment can have negative consequences for the patient, which justifies the need of better control systems [2]. In the remaining pages of this chapter, a broader introduction about semiconductors, GaN, radiation detection systems based on GaN and the particularities of GaN nanoand microwires will be given. In the last section the thesis scope and outline will be presented. 1.1 Historical view of semiconductor devices It was on a cold Tuesday afternoon in December that John Bardeen and Walter House Brattain, researchers at Bell Telephone Laboratories under the guidance of William Shockley, started what we now know to be as one of the most exciting and influential technological revolutions in human history. They succeeded in creating a working transistor [3]. The concept of their transistor, for today’s standards at least, is relatively simple. The so called point-contact transistor consisted of a slab of germanium on a metal base with two very closely spaced gold contacts against it. The separation between the contacts was notably achieved by sliding a razor blade along the edge of the wedge that was used to press the contacts against the germanium. The main requirement for the transistor effect to exist is germanium being a semiconductor. This term, introduced in 1911 by the german physicist Josef Weiss in his PhD 2 thesis [4] (he used the german word Halbleiter), classifies materials that are neither metals nor insulators, but something in between. The observation of the transistor effect was not the first time physicists took note of the immense potential of semiconductors. Already in the 18th century Alessandro Volta studied what he called cattivi conduttori [5] (bad conductors in english) and in the 19th century both the photovoltaic effect [6] and Hall effect [7] were discovered. However, at that time a theory of solid-state physics was still lacking and good quality semiconductor materials were hard to obtain, making the measurements and observations erratic and non-reproducible. Only in the early 1900s, with the momentum brought by the revolutionary formulation of quantum mechanics, did the properties of semiconductors become comprehensible. Fundamental for the advancement in knowledge were the establishment of the band theory of semiconductors by Alan Wilson and the theory of movement of electrons through the atomic lattice by Felix Bloch [4]. What makes semiconductors special is that their conductivity can be controlled by external factors, such as heat or light, as well as by the introduction of impurity atoms in their crystal lattice. One of the fundamental properties of a semiconductor that enables this control is their bandgap. Simply put, the bandgap can be seen as the separation between the conduction band and the valence band of a material. Metals do not have a bandgap, allowing electrons and holes to flow freely through the lattice whereas the bandgap of insulators is so large that all the electrons are stuck in the completely filled valence band and cannot move. The bandgap of semiconductors is small enough to allow electrons to jump from the valence band to the conduction band by, for example, absorption of a photon. When this happens, the electron is free to move in the material and starts contributing to the conductivity of the semiconductor. The opposite process, the recombination of an electron in the conduction band with a hole in the valence band, is also possible and is usually associated with the emission of a photon, containing an energy equal to the bandgap energy. Furthermore, when impurities, also called dopants, are introduced in a semiconductor, they create intermediate energy levels in the bandgap which modulate the electrical and optical properties. This feature, better known as doping, led for instance to the invention of the silicon p-n junction in 1939 by Russel Ohl [8] and to the already mentioned development of the transistor [3]. Among semiconductor materials, there is one that clearly stands out if not for giving the name to one of the most famous valleys in the world: silicon. All electronic devices that we use today contain silicon in all sort of shapes and forms. Its name was given by Humphry Davy (also known as the inventor of laughing gas) and it is the eighth most abundant element in the universe and the second most abundant element in the earth’s crust. This abundance, together with the fact that it can be easily doped to create both p-type and n-type semiconductors and its compatibility with complementary metaloxide-semiconductor technology, are some of the reasons for its unique success story. Nevertheless, the semiconductor boom also led to the development of other materials and among these are the so called Wide Bandgap Semiconductors (WBGSs). Silicon has a bandgap of 1.1 eV, meaning that at least an infrared photon with a wavelength around 1127 nm would be necessary to excite one electron from the valence band to the conduction band. As the name already reveals, WBGS materials have a larger bandgap. Although there is no official value 3 from which a semiconductor becomes a member of the WBGS club, typically materials with a bandgap above 2.5 eV are allowed to join. These members include, for example, diamond (EG= 5.47 eV), zinc oxide (ZnO, EG= 3.37 eV), silicon carbide (SiC, EG= 2.3-3.3 eV) and boron nitride (BN, E G= 5.96-6.36 eV). There is, however, one WBGS material that has a great success story of its own, gallium nitride (GaN) and related compounds with indium nitride (InN) and aliminum nitride (AlN). Despite its amazing potential, there are some applications where silicon does not shine as bright. Silicon has what is called an indirect bandgap, which means that the conduction band minimum and the valence band maximum k-vectors are different. As a consequence, when an electron decays from the conduction band back to the valence band the law of conservation requires the emission of a phonon, making the emission of photons that occur during the decay very inefficient. In direct bandgap materials, the conduction band minimum and valence band maximum k-vectors are aligned, hence, when electrons transit from one band to the other, energy and momentum conservation can be satisfied solely by the absorption or emission of a photon. This makes direct bandgap materials much more interesting for applications related with light emission. After the discovery of the electroluminescence effect, which essentially consists of the emission of photons when a bias is applied to a semiconductor, by Henry Round in 1907 [9], the first solid-state Light-Emitting Diodes (LEDs) were build using GaAs, SiC and InP [10, 11]. Nevertheless, besides only emitting photons in the non-visible spectrum their efficiencies were quite low. In the 1960s the first visible LEDs started to be produced and became commercially available however they still could only provide a small light output and their applications were limited to small screens and indicators. Development of new semiconductor materials led to the improvement of the efficiency and brightness of the LEDs with light emission mainly in the infrared and red region of the light spectrum [12]. Although light emission in other colors was eventually achieved, efficiencies were still low. This changed when GaN entered the scene, and a true revolution in the lighting industry was looming. GaN not only has a wide, but also a direct bandgap, making it the ideal candidate for efficient light emission. The first proof of concept was demonstrated by Maruska and Tietjen who achieved emission of blue-violet light using magnesium doped GaN films grown on sapphire [13]. What followed where almost 25 years of research, including several breakthroughs in terms of material quality and processing. Key advances where the ability to grow high quality GaN through Metal-Organic Vapor Phase Epitaxy (MOVPE) using a low-temperature AlN buffer layer in 1985 by Hiroshi Amano [14] and the realization of p-type GaN through magnesium acceptor activation by means of an electron beam in 1989 by Isamu Akasaki and Hiroshi Amano [15], which allowed the fabrication of the first GaN p-n junction. Furthermore, in 1991 Shuji Nakamura showed that epitaxial growth of high quality GaN films on sapphire was also possible using a GaN buffer layer [16] and obtained p-type GaN by annealing Mg doped GaN with temperatures above 400 ◦C in an H2 free environment [17]. The final piece to the puzzle of creating highly efficient blue LEDs consisted of the p-doping and growth of AlGaN and InGaN [18, 19] and in 1994 Nakamura et al. produced the first bright blue LED using a double heterojunction InGaN/AlGaN structure [20]. The achievements stated in the previous paragraph had an enormous impact on the lighting industry. The high efficiency and low power consumption of the LEDs make them a much more cost effective op4 tion than the traditional light bulbs and fluorescent tubes and, nowadays, GaN based LEDs can be found at almost every street corner, office and personal home. For their research on GaN, Isamu Akasaki, Hiroshi Amano and Shuji Nakamura were awarded with the 2014 Nobel prize [21]. Furthermore, due to its unique properties, physicists started to consider using GaN for more than just lighting up the world. 1.2 Properties of GaN There are different epitaxial techniques that can be used to grow GaN but, whatever the technique, it crystallizes either in a wurtzite or in a zincblende structure. Between both, the former is thermodynamically more stable and therefore more often studied and used. This is also true in our research and as such we will focus our attention to the wurtzite structure. Figure 1.1 shows a schematic of the unit cell that consists of hexagonal planes stacked as ABAB in the c direction. The intrinsic lattice parameters of relaxed GaN are also given in the figure [22]. In this configuration, GaN has a direct bandgap with an energy equal to 3.4 eV at room temperature [23], which corresponds to a wavelength of 364 nm and means that photons in the visible range do not carry enough energy to excite an electron from the valence to the conduction band. In essence, this means that GaN is intrinsically visible blind and, consequently, a very good material for UV photodetection purposes [24]. Another advantage of the wide bandgap is that it allows operation at room temperature, contrary to, for example germanium, which needs to be cooled down to low temperature. N-type doping in GaN is most commonly a result of the introduction of Si or O atoms into the crystal lattice. When grown through epitaxial methods, in most cases unintentional n-type doping occurs as a by product of the reaction, giving the GaN an n-type conductivity. we already saw that p-type doping is achieved by introducing Mg impurities and annealing them at high temperature in an H2free environment. One of the main drawbacks is that, albeit possible, efficient p-type doping is still hard to achieve due to the high activation energies that are necessary. Another property of GaN is its relatively high thermal conductivity and melting point, giving GaN a very good thermal stability. Additionally, it can endure high electric fields due to its high breakdown voltage. The wide bandgap does also come with some costs mainly in terms of mobilities. Maximum Figure 1.1: Crystal structure of wurtzite GaN and respective lattice constants 5 electron mobility values in GaN are of the order of 103cm2V-1s-1 while the hole mobility is typically a decade smaller [25]. Compared to, for instance GaAs, these values are smaller [26]. However, the main drawback that is still preventing GaN (and other WBGSs) from being widely used in all branches of the semiconductor market is its crystal quality. Heteroepitaxial growth of GaN is only feasible on substrates with a large lattice mismatch which leads to films with a high density of dislocations and grain boundaries. Although some strategies, such as the ones where a buffer layer is used found by Akasaki, Amano and Nakamura, allow to decrease the amount of dislocations, their density is still much larger than in silicon. This has negative effects on the properties of devices, such as the increase of the leakage currents, poor visible-to-UV light rejection ratio and the appearance of persistent effects. High quality GaN can be grown through homoepitaxial methods, however, the lack of low-cost high quality bulk GaN substrates makes this an expensive choice. Still, the exceptional properties of GaN make it worth to accept the problems associated with the poor crystalline quality, especially when considering high-power and high-temperature applications. In the field of power electronics, the development of GaN based devices is posed to generate a similar sort of shock wave as it created in the lighting industry. Examples of such devices are High-Electron Mobility Transistors (HEMTs) and power converters [27]. 1.3 GaN as a radiation hard material Nevertheless, in the scope of this work, the most fundamental characteristic that GaN presents is its high resistance to ionizing radiation. The reason GaN presents a superior hardness when exposed to high energy radiation can be traced back to its structure. The energy that is required to displace one N or Ga atom from its equilibrium position in the crystal lattice and leave behind a vacancy is relatively large. The average values of this energy, better known as the displacement energy (Ed), calculated through molecular dynamic simulations are equal to 109 eV for N atoms and 45 eV for Ga atoms [28]. GaAs, on the other hand, presents an average Edvalue of 13 eV [29]. Practically, this means that in order to create the same number of point defects in GaN when compared to other semiconductors, a much higher irradiation fluence is necessary. When characterizing, for example, the electrical properties of HEMTs irradiated under similar conditions, it was shown that for the GaAs based devices they degrade at fluences about two orders of magnitude lower than for GaN based devices [30, 31]. As a consequence, GaN devices began to be considered for applications where extreme environments are a reality, like in space, nuclear facilities or high-energy physics laboratories. Take the Large Hadron Collider (LHC) at CERN for example, a future upgrade envisions to boost up the luminosity of the accelerator, i.e. the number of collisions that are detected in a certain period of time and along a certain cross-section, to levels close to 5x1034 cm-2s-1 [32]. It is therefore of utmost importance that the radiation detection and read-out systems are capable of handling these immense radiation levels without compromising the measurements or being often replaced. The RD50 collaboration, which joins over 60 research institutes around the world, was created to answer the challenges and one of the spearpoints is the development of detectors based on WBGSs, GaN among them [33]. Furthermore, GaN based 6 HEMTs are already being used in several satellites orbiting the earth. The first one was the Proba-V [34] and recently the BIOMASS [35] satelite was also launched by ESA. Although the displacement energy magnitudes are good indications for the grade of radiation resistance of a material, a good understanding of the damage creation mechanisms are vital. Studying damage effects due to irradiation is not straightforward, different types of radiation can create different defects, defects can recombine or form complexes, they can interact with the impurity atoms in doped semiconductors and when the energy of the incident radiation is high, damage cascades can originate extended defect regions. During the first two decades of the 21st century, many research groups tried to tackle relevant questions regarding the interaction of radiation with GaN and uncover the related secrets. In the scope of this work, we will mainly focus on proton and ion irradiation. In general, proton irradiation creates simple point defects in the GaN lattice. These defects can exist in the form of vacancies and interstitials, forming what are known as Frenkel pairs. According to theoretical calculations, nitrogen vacancies (VN) and gallium interstitials (Gai) create donor like centers whereas nitrogen interstitials (Ni) and gallium vacancies (VGa) are predicted to form acceptor levels in the forbidden bandgap [36]. Nevertheless, when taking into account that the defects can recombine, form complexes with each other and dopants, reality is often much more complex. Still, for proton irradiation, the main effects that occur as a consequence of the irradiation damage in GaN devices are the reduction of the free carrier concentration due to trapping of carriers into the induced defects and a decrease of the carrier mobility due to scattering of carriers at defect sites [37]. Irradiation of GaN with ions has been extensively investigated in the scope of ion implantation studies. Ion implantation is an extremely powerful tool to selectively modify the electrical and optical properties of semiconductors and understanding the dynamics of defect creation and recovery is therefore very important for both research and industrial applications. One of the first studies on the damage accumulation by ion irradiation revealed that, besides the strong bonding of the atoms in GaN, there is another property that has a strong impact on the radiation resistance of GaN. This effect is its dynamic annealing at room temperature. In layman’s terms, this means that the radiation induced defects in GaN are mobile, meaning that a displaced atom can return to its lattice site and undo the vacancy that was created. One of the consequences of the strong dynamic annealing effects is that relatively high fluences are needed to amorphize GaN by ion irradiation [38]. Furthermore, Catarino et al. suggested that the resistance to implantation damage is different for a-plane and c-plane GaN [39]. This was confirmed by Lorenz at al. who implanted c-, mand a-plance GaN films with 300 keV Ar ions at 15 K to fluences from 2x1014 to 4x1016 ions/cm2and saw that the damage build-up for a-plane GaN is different than for the other configurations [40]. This is especially relevant for nanoand microwires where the major part of the surface is composed of m-plane GaN [41]. Recently, the effect of swift heavy ion irradiation on GaN has also been studied combining molecular dynamic simulations and experimental techniques [42]. The very good agreement between results obtained through the model and the experiments showed that recrystallization effects within an ion track are the key mechanism behind the observed radiation resistance of GaN. Due to their interest for space applications, a big part of the research regarding irradiation defects in 7 14 Chapter 2 Brief description of semiconductor detector theory In this chapter a brief theoretical description about the theory of semiconductor detectors will be given. Naturally, covering the entire semiconductor theory would require a large number of pages and information, and is not within the scope of this work. However, to fully comprehend the concepts described in this work, a proper understanding of the mechanisms that define semiconductor detectors is fundamental. First, a brief introduction of the doping and the transport mechanisms in semiconductors will be given to then present the main basic semiconductor building blocks that are relevant in semiconductor detector technology. Afterwards, the different semiconductor detector configurations will be presented, namely, the photoconductor, the Schottky diode and the p-n junction diode. I note that the presented concepts and equations were mostly taken from references [92] and [93]. 2.1 Carrier concentration and transport mechanisms One of the most important properties of a semiconductor is that it can be doped with different types and concentration of impurities which change the properties of the semiconductor. Depending on the type of impurity, the semiconductor can be n-type or p-type doped, where in the former case the crystal is doped with donor atoms and in the latter case with acceptor atoms. Physically, this means that donor impurities contain a loosely bound electron while acceptor impurities contain a loosely bound hole. When the impurities are ionized, the respective charge carriers become free, and increase the carrier concentration of the semiconductor. When this happens, a positively charged donor is left in case of n-type doping, and a negatively charged acceptor is left in case of p-type doping. In other words, we can also say that a ’n’ (negative) type semiconductor contains an excess of electrons and that a ’p’ (positive) type semiconductor contains an excess of holes. The free carriers that are introduced in the semiconductor can move through the crystal lattice via two main mechanisms which are defined as carrier drift and carrier diffusion. The drift current originates as a consequence of the existence of an electric field which applies a force on the electrons and holes 15 so that they will experience a net acceleration. The total drift current density is given by Jdrf =q(µnn+µpp)E, (2.1) where qis the elementary charge, µnthe electron mobility and µpthe hole mobility, nand pthe electron and hole concentrations respectively, and Ethe applied electric field. Furthermore, we can also write q(µnn+µpp) = σ, (2.2) where σis the conductivity of the semiconductor. From this equation it becomes clear that the conductivity is thus dependent on both the hole and electron concentrations and on the respective mobilities. The reciprocal of the conductivity is defined as the resistivity of a semiconductor. Taking into account that either p≫nor n≫pwhen the semiconductor is p-type doped or n-type doped respectively, we can see that the conductivity and resistivity are primarily a function of the majority carrier parameters (here majority refers to the type of carrier that exists in highest concentration). The diffusion current is a consequence of carrier gradients inside the semiconductor and appears due to the movement of carriers from a high concentration region to a low concentration region. Unless the material is at a temperature of absolute zero, the carriers will have some random motion defined by their mean free path (the average distance traveled by an electron before a collision occurs). If there is a carrier gradient inside the semiconductor, despite the random nature of the thermal motion, on average more carriers will move towards the region with lower concentration. This net movement of carriers originates the diffusion current. In a single dimension, the total diffusion current density is given by Jdif =qDn dn dx −Dp dp dx(2.3) , where Dnand Dpare the diffusion coefficients of the electrons and holes respectively. If nor pare constant in the semiconductor, dp/dx =dn/dx = 0 and the diffusion current density is zero. Joining the drift and diffusion of carriers, we can write the total current density as J= (µnn+µpp)E+qDn dn dx −Dp dp dx(2.4) 2.2 The p-n junction diode The p-n junction can be described as the interface between an n-type region and a p-type region inside a single semiconductor. What happens when an n-type semiconductor is merged with a p-type semiconductor is that electrons (holes) diffuse into the p-region (n-region) whereas the positively charged donor ions (negatively charged acceptor ions) are not mobile. Hence a negative space charge originates in the p-region and a positive space charge in the n-region, originating a so called built-in potential Vbi: Vbi =kbT qln NAND n2 i, (2.5) 16 n-region p-region E + + + + + + + + - - - - - - - - SCR E p n EF di ✁ usion di ✁ usion drift drift Conduction band Valence band holes electrons qVbi (a) n-region p-region E SCR + + + + + + + + - - - - - - - - + - E p n di ✁ usion di ✁ usion drift drift q(Vbi + VR) (b) n-region p-region E + + + + + + + + - - - - - - - - E p n di ✁ usion drift drift + - di ✁ usion q(Vbi - VF) (c) Figure 2.1: Band diagrams of the p-n junction in equilibrium (a), reverse bias (b) and forward bias (c). where NAis the acceptor concentration, NDthe donor concentration and nithe intrinsic carrier concentration of the semiconductor, kbthe Boltzmann constant and Tthe temperature. The region depleted of mobile charges is also known as the depletion region or Space-charge Region (SCR). In thermal equilibrium and steady-state conditions, the width of the SCR is given by W=s2εs qNA+ND NANDVbi, (2.6) where εsis the permittivity of the material. If either NA≫NDor ND≫NAwe obtain a one-sided junction. This name was given because in this situation the SCR extends almost exclusively into the lower doped side. If we consider, for instance, the latter relation we obtain W≈s2εsVbi qNA (2.7) When we apply a bias, the band structure of the p-n junction is modified. Figure 2.1 shows a schematic of the band diagram in equilibrium and when applying a positive and negative voltage to the p-side of the junction relative to the n-side of the junction. When we apply a positive voltage to the p-side of the semiconductor with respect to the n-side, we are in forward bias and the potential, Vbi, decreases by VF. As a consequence the electric field accross the junction and the depletion width also decrease. On the other hand, when we apply a negative voltage to the p-side with respect to the n-side the opposite happens. Now the built-in potential is enhanced by VR, the electric field grows and the depletion width increases. This state is called reverse bias. Considering the applied bias we can rewrite eq. 2.7 to W≈s2ϵS(Vbi −V) qNA (2.8) where Vis positive for forward bias and negative for reverse bias. 17 Current-voltage characteristics The disruption of the equilibrium state of the p-n junction when a bias is applied affects the current that moves through it. In forward bias, more electrons (holes) in the n-region (p-region) will be able to diffuse into the p-region (n-region), giving rise to minority carrier injections. In reverse bias, the opposite happens and diffusion will be greatly reduced. The interplay between the applied voltage and the current can be described by the diode equation: ID=I0exp qV nkBT−1(2.9) When Vis positive, i.e. we are in forward bias, IDincreases exponentially whereas when Vis negative the exponential term tends to 0 and the IDreduces to -I0, defined as the saturation current. I note that in forward bias, the increase in current becomes fast when the forward voltage exceeds the so called cut-in voltage. Ideally, this value is typically slightly lower than the bandgap of the semiconductor divided by the elementary charge. The parameter nis called the ideality factor and is an empirical value that is introduced to take recombination currents into account. In the ideal case the ideality factor is unity and this corresponds to the situation where the forward bias current density is solely a result of diffusion of carriers through the barrier. However, carrier recombination through intermediate bandgap energy levels can also contribute to the current and when recombination is the dominant process, the ideality factor will have a value of 2. Then, we also need to take into account the impact of a series resistance when we have a high current injection. This series resistance appears as a consequence of the non-zero resistivity of a semiconductor. At low injection, the voltage drop due to the series resistance is typically low but for high currents, the factor IRSexceeds kbT/q which reduces the bias across the SCR and, consequently, the rate at which the current increases. Finally, when considering the equivalent circuit of a device composed by a p-n junction diode, it is also common to include a so called shunt resistance in parallel with the diode. This resistance typically represents the presence of alternative current paths that are responsible for leakage currents travelling through the p-n junction. When we take into account the series and shunt resistance we can draw the equivalent circuit diagram shown in figure 2.2 a) and the total current of this equivalent circuit can be written as I=I0exp q(V−IRS) nkBT−1+V−IRS RSh , (2.10) where RSand RSh are the series and shunt resistance, respectively. This equation can be solved for I(V)by writing it in the following form: I(V) = V Rs−I0 1 + Rs Rsh +nkT qRs WLam   qI0Rs nkT 1 + Rs Rsh exp   qV nkT 1−Rs Rs+Rsh +qI0Rs nkT 1 + Rs Rsh   , (2.11) where WLam is the Lambert function. Figure 2.2 b) shows the I-V curve for an ideal diode with n= 1, 18 Rseries Rshunt p-n junction (a) Ideal diode I-V curve (b) −4 −3 −2 −1 0 1 2 3 4 Bias (V) 10 −17 10 −15 10 −13 10 −11 10 −9 10 −7 I (A) n = 4 R S = 5 × 10 6 Ω R Sh = 1 × 10 15 Ω I 0 = 1 × 10 −13 A I 0 = 1 × 10 −14 A I 0 = 1 × 10 −15 A I 0 = 1 × 10 −16 A (c) −4 −3 −2 −1 0 1 2 3 4 Bias (V) 10 −13 10 −12 10 −11 10 −10 10 −9 10 −8 10 −7 10 −6 I (A) I 0 = 1 × 10 −15 A R S = 5 × 10 6 Ω R Sh = 1 × 10 11 Ω n = 1 n = 2 n = 4 n = 8 (d) −4 −3 −2 −1 0 1 2 3 4 Bias (V) 10 −13 10 −12 10 −11 10 −10 10 −9 10 −8 10 −7 10 −6 I (A) I 0 = 1 × 10 −15 A n = 4 R Sh = 1 × 10 11 Ω R s = 2 × 10 6 Ω R s = 1 × 10 7 Ω R s = 5 × 10 7 Ω R s = 1 × 10 9 Ω (e) −4 −3 −2 −1 0 1 2 3 4 Bias (V) 10 −13 10 −12 10 −11 10 −10 10 −9 10 −8 10 −7 10 −6 I (A) I 0 = 1 × 10 −15 A n = 4 R S = 5 × 10 6 Ω R sh = 1 × 10 10 Ω R sh = 5 × 10 10 Ω R sh = 1 × 10 11 Ω R sh = 5 × 10 11 Ω (f) Figure 2.2: a) Equivalent circuit of the p-n junction including a series and a shunt resistance; (b-f) Simulation of the I-V curves obtained by solving eq. 2.11 when giving different values I0,n,RSand RSh. a saturation current equal to 1x10-13 and with RS= 0 and RSh =∞. The other graphs shows some simulated curves when varying some of the parameters included in eq. 2.10. When lowering I0we can see that the cut-in voltage decreases because a higher voltage is necessary to overcome the saturation current. Naturally, the reverse bias also decreases, however, if the saturation current is lowered to very small values, the reverse bias current becomes governed by the shunt resistance instead. In the graph shown in figure 2.2 c) this is the case for I0<1×10−15 A and this is also visible in figure 2.2 f), where we can see that for a low shunt resistance the current in reverse bias is no longer defined by I0but rather follows V/RSh. Furthermore, we can see that for a high series resistance, the current increase in forward bias becomes linear. Regarding the ideality factor, it is important to note that this value impacts 19 the rate of the exponential increase. When plotting the current in logarithmic scale this is translated into a variation of the slope of the linear section of the curve. In fact, we can relate the slope with nthrough the following equation dlog(I) dV =q nkT (2.12) Reverse bias current mechanisms As we will see later, when used for detection purposes, a p-n junction is mainly used in reverse bias. The main reason for this is that normally maximizing the width of the depletion region is beneficial. As such, one of the most important parameters of a photodiode is the reverse bias dark current magnitude, where the goal is always to keep it as low as possible. In eq. 2.10, the reverse bias current results from the saturation current and the inclusion of a shunt resistance in parallel with the p-n junction diode. However, sometimes this model fails to adequately describe real-life scenarios where the reverse bias current is neither constant nor linear. For a starter, the saturation current given in eq. 2.10 represents the diffusion currents from the neutral regions but does not take the contribution from the generation currents into account. Similarly to the recombination currents that affect the forward bias by originating an increase in the ideality parameter, the generation current is a consequence of the Shockley-Read-Hall (SRH) mechanism. This mechanism describes the generation and recombination of carriers through localized energy states in the bandgap, also called recombination centers, which can include the donor and acceptor levels, defect state energy levels or surface state energy levels. In essence, these energy levels can be seen as steps from a ladder that facilitate movement of carriers through the bandgap. The four basic transitions that occur in the SRH model are illustrated in figure 2.3 a). The reverse bias current density that results as a consequence of the generation in the depletion region can be given by Jgen =qniW τg (2.13) where τgis defined as the generation lifetime, a parameter that depends on the density of defects. Due to the depletion width in the numerator of the equation, the generation current density has a square root dependence on the applied reverse bias. Hence, in this case the total reverse bias current is composed of a term independent of bias representing the diffusion current and a term that varies with the square root of the applied voltage as follows Irev ∼ =I0,diff +I0,gen(pVR)(2.14) Another possible source of leakage are tunneling currents. When the SCR region is narrow, the possibility of carriers moving across the bandgap through quantum mechanical tunneling becomes important. This is thus especially relevant in p-n junctions where the doping concentrations are large as this will decrease the width of the depletion region. When the tunneling occurs directly from one band to 20 Ec Et Ev Recombination Center (a) trap assisted tunneling trap assisted tunneling drift di ✁ usion drift di ✁ usion (b) Figure 2.3: Schematics of the mechanisms responsible for leakage currents in p-n junctions: SRH generation and recombination (a); Tunneling assisted by traps located in the bandgap (b). the other it is called band-to-band tunneling and this tunneling mechanism is the basic working principle of the tunnel diode developed firstly by Esaki et al. [94]. However, when defect densities are large, tunneling through traps is also a possibility. The trap-assisted tunneling mechanism is illustrated in figure 2.3 b). Both band-to-band and trap-assisted tunneling depend strongly on the electric field that is applied to the p-n junction and impact both the reverse and forward biased current. For forward bias, the tunneling current has an exponential dependence, similar to the exponential dependence of the diffusion current of the ideal diode, but where the term inside the exponential is characterized by the trap energy ET. Following [95], we can write IT,fwd =IT0exp eVF ET−1, (2.15) where IT0is the tunneling saturation current. In reverse bias, the tunneling current can be described by the empirical relation based on the Zener tunneling model [96] IT,rev =aVRexp −U √Vbi +VR(2.16) If a multistep tunneling process is assumed as the one illustrated in figure 2.3 b), the parameter ais dependent on the density of available states in the depletion region while Uis defined by the number of steps that are required to traverse the entire bandgap and the tunneling energy barrier for each step. Therefore, if the electrical I-V curves are well modeled by eq. 2.16 it is possible to estimate the number of traps present in the semiconductor [96]. There are other current leakage mechanisms such as Variable Range Hopping (VRH) [97], SCLC (Space-charge Limited Current) [98] that can affect the reverse and forward bias current of a p-n junction. The voltage dependence of each mechanism as well as the configuration of the device under study usually allows to identify which are the main ones impacting the transport mechanisms but, on the other hand, different regimes can become active depending on the magnitude of the applied bias which often makes it difficult to perform a solid analysis about all the leakage mechanisms that are present. Finally, 21 it is also worth mentioning that for high enough biases, impact ionization becomes relevant [92]. During this process, an electron moving through the conduction band gains enough energy to ionize atoms and promote other electrons to the conduction band, which in turn do the same. This creates an avalanche of electrons contributing to the current density and when this happens the reverse bias current increases dramatically. For this reason it is said that the p-n junction enters breakdown. Despite the name, this process is not necessarily irreversible. In fact, avalanche photodiodes are specifically engineered to operate in this regime because it generates a large a photocurrent gain [99]. 2.3 Metal-Semiconductor junction Another building block of semiconductor devices that is very relevant is the so called metal-semiconductor junction. This junction essentially acts as the metal contact that is necessary to include the semiconductor device in an external circuit. However, the properties of this junction can also define the working mechanisms of the device. When a metal film is deposited on top of a semiconductor surface, carriers (electrons in case of an n-type semiconductor and holes in case of a p-type semiconductor) move into the metal to establish a thermal equilibrium in the system. As a consequence, a volume depleted of mobile carriers originates in the semiconductor region near the interface, in similar fashion to what occurs in a p-n junction. Likewise, an internal electric field is generated and a potential barrier is formed at the interface. This barrier, called the Schottky barrier, limits the further injection of carriers from the semiconductor into the metal. The height and thickness of this barrier depend on the material properties, surface states and on the distribution and concentration of doping impurities in the semiconductor. Schematic representations of the ideal band diagrams of the metal-semiconductor junction for an n-type and a p-type semiconductor are given in figure 2.4. In the schematics, we can see that for an ideal metal-semiconductor junction, where the semiconductor is n-type doped, the band height, ϕb, is given by the difference between the metal work-function, ϕm, and the electron affinity, χSC, of the semiconductor. The built-in voltage can be seen as the barrier found by the electrons in the conduction band that try to move into the metal and is given by the difference between the conduction band energy and the Fermi energy, subtracted from the barrier height. If a positive voltage is applied to the semiconductor EF EC EV WSCR ✁ b= ✁ m- ✂ SC Vbi Metal n-type SC (a) EF EC EV WSCR ✁ b= Eg - ( ✁ m- ✂ SC)Vbi Metal p-type SC (b) Figure 2.4: Ideal band diagram of a metal/n-type semiconductor junction (a) and a metal/p-type semiconductor junction (b). 22 with respect to the metal, the barrier seen by the electrons increases to Vbi+VR, where the Rsubscript is indicative of the reverse bias condition. The opposite happens when we apply a negative voltage to the semiconductor with respect to the metal. In this case the barrier decreases in height to Vbi-VFand we are in forward bias. For p-type semiconductors the barrier height is given by the difference between the metal work-function and the electron affinity of the semiconductor, subtracted from the bandgap. Furthermore, regarding the polarization we have the inverse situation. A forward bias corresponds to a negative voltage applied to the semiconductor with respect to metal and a reverse bias to a positive voltage applied to the semiconductor with respect to metal. The width of the SCR of the Schottky junction, for an n-type semiconductor, is given by W=s2εS(Vbi +VR) eNd , (2.17) hence the thickness of the barrier is not only dependent on the applied bias but also on the doping concentration of the semiconductor. This consideration is important because the current densities as a function of bias are strongly dependent on the shape of the Schottky barrier. Broadly speaking, a junction can operate in one of the following three regimes: thermionic emission, thermionic field emission or field emission. The former regime dominates when the barrier height and width are such that tunneling of carriers through the barrier is negligible and, to contribute to the current, carriers need to move over the barrier. The opposite is true for the latter regime, in this case the barrier is very thin and tunneling of carriers through the barrier is strongly enhanced, resulting in high current densities even when a reverse voltage is applied. Finally, thermionic field emission is a combination of both these regimes. Since the width of the SCR is proportional to the inverse square root of the doping concentration, the type of carrier transport that dominates is strongly correlated with the doping level. Typically, thermionic emission rules when the doping concentration is below 1017 cm-3. For doping concentration above 1019 cm-3, field emission is prevalent and in between the current transport is best modeled by thermionic field emission. The three regimes are schematically represented for an n-type semiconductor in figure 2.5. Taking into account the nature of the carrier transport in each regime, it is clear that the highest specific contact resistivities, i.e the resistance charge carriers feel when moving through the metal-semiconductir junction, are obtained for junctions that operate through thermionic emission, and the lowest specific contacts resistivities for junctions that operate through field emission. EC EV WSCR Metal n-type SC Thermionic emission ND < 1x1017 cm-3 EC EV WSCR Metal n-type SC Thermionic Field emission 1x1017 cm-3 < ND < 1x1019 cm-3 EC EV WSCR Metal n-type SC Field emission 1x1019 cm-3 < ND Figure 2.5: Schematic of the carrier transport mechanisms for metal/n-type semiconductor junctions with different doping concentrations. 23 Type of detector Advantages Disadvantages Photoconductor High gain Low cost Simple fabrication Low speed Low signal-to-noise ratios Non-linear photocurrent with optical power p-n junction diode Low dark current High signal-to-noise ratios Fast response times Linear photocurrent with optical power Low noise Low sensitivity Complex fabrication Schottky diode Fast response Low capacitance Low noise Simple fabrication Require high reverse voltages High reverse leakage currents Table 2.1: Advantages and disadvantages of the different types of semiconductor detectors. WBGS and proper ohmic contacts can be hard to achieve. Furthermore, since the current in a Schottky diode is due to majority carrier diffusion, the time response is typically better than in p-n junction diodes where the minority carrier diffusion transient limits the time capabilities. Furthermore, since the SCR is located very close to the surface, Schottky diode detectors are particularly useful to detect light with small wavelengths since, for these wavelengths, the photon absorption coefficients are relatively high in semiconductors. Table 2.1 summarizes some of the advantages and disadvantages for each type of detector covered in the previous sections. 2.5 GaN based photodetectors The earliest detectors based on GaN and related ternaries were photoconductors and it was found that they possessed some distinctive characteristics namely a strong sub-bandgap response [104], very high photoconductive gains [105], slow non-exponential photocurrent decays [101] and highly non-linear photocurrent dependencies on the incident optical power [106]. Photoconductors often present internal gain mechanisms which typically have a negative impact on the response time of the photoconductor. Although there can be different origins, the most common model to describe the gain is based on conductivity modulation through carrier generation. However, Garrido et al. showed that for GaN detectors, a model that considers the variation of the conductive volume due to the presence of excess charge carriers also helps to explain the behavior of GaN photoconductor devices [107]. This model states that the presence of negatively charged surface states generates a SCR region which decreases in thickness when the device is illuminated and thus produces a variation in the conductive volume of the detector. Hence, the photocurrent can be seen as the result of two terms, one due to photogenerated carriers and the other due to the light induced modulation of the conduction volume. The simultaneous influence of both mechanisms leads to very high gains. The development of GaN Schottky photodiodes started in the same period as the development of photoconductors. Metals with high work function, such as Pt, Ni, Pd or Au are commonly used to obtain good Schottky contacts [108]. Different geometries were developed including back and front-side 30 illuminated devices and it was also found that by introducing a certain amount of Al content in the GaN lattice, allowing the fabrication of AlxGa1-x N Schottky photodiodes, barrier properties were improved [24]. One particular observation that raised the interest of the scientific community was the appearance of gain and a slow decay component, similar to those detected with the photoconductors, when operating the Schottky photodiode in reverse bias. Ideally, these phenomena should not occur but were attributed to trapping of minority carriers at the metal-semiconductor junction which induces a reduction of the barrier height [109, 110]. Using the MSM configuration, it was possible to drastically reduce the dark current leakage values [111]. The MSM photodiodes can be particularly useful for high-speed applications since the low capacitance associated with the structures leads to the response times being typically limited by the carrier transit times [112]. Single wire GaN photodetectors based on two metal-semiconductor contacts were also demonstrated and it was found that the slow transient responses [113] as well as the non-linear characteristics of the photocurrent as a function of the irradiance [114], with similar properties to those observed in planar Schottky based devices were measured. The first GaN based photodetector containing a p-n junction structure was fabricated by Chen et al. [115] and not much later a MOVPE grown GaN p-n junction photodetector with a peak responsivity of 0.09 A/W [116] and an MBE grown p-i-n photodetector with a peak responsivity of 0.11 A/W [117] were reported. At that time, the p-n junction allowed huge improvements in response times and better spectral selectivity compared to the photoconductors and Schottky diodes and through improved growth mechanisms as well as the use of slightly different configurations the FOMs could be further improved [118, 119, 120, 121]. Nowadays, GaN based p-n junction photodetectors typically present linear responses as a function of the optical over several orders of magnitude and responsivities lie in the 0.10.15 A/W range [122]. The success of the p-n junctions in terms of photodetection can also be identified when considering devices that use wire geometries. One of the first reports on GaN wire based p-n junction photodetectors was published in 2006 [123]. Since then, many results regarding both microand nanowire in single or ensemble configuration have been published [87]. Using single wires grown by a similar process as the ones used in this work, Zhang et al. not only showed they possessed peak responsivities above 0.15 A/W but also studied the frequency response, showing that it was faster than for photoconductive wire devices. Nevertheless, a small PPC was measured for a reverse bias larger than -1 V, attributed to the presence of photoconductive mechanisms [124, 85]. The interplay between photodiode and photoconductive processes was also studied in MBE grown nanowire ensemble devices [125]. Furthermore, the flexibility of the microwire structures was demonstrated by Tchernyeva et al., who reported the fabrication of an integrated photonic platform where both the emitter and detector were realized using GaN wires. Finally, flexible photodiodes, in which the GaN microwires were integrated in a PDMS matrix, are an indication that scale-up potential exists [82]. 31 32 Chapter 3 Radiation detection with semiconductors In this chapter, the physical processes that allow the detection of ionizing radiation using semiconductors will be presented. When radiation interacts with a material, different interactions may occur that depend both on the type of radiation and on the material. The concepts that allow us to predict which interactions dominate will be introduced. Furthermore, the detection of radiation can be achieved using different methods and the most relevant ones for this work will be explained. Finally, a brief description and literature revision about the impact of the radiation damage in GaN based devices will be given. 3.1 Detection of radiation Detection of ionizing radiation is carried out using a radiation sensitive material. When particles or high energy photons interact with the atoms or molecules of this material, information about the nature of the ionizing radiation, such as the energy or radiation flux, can be extracted. There are different types of ionizing radiation and how each of them interacts with the same material can be quite different. Electrons, protons, neutrons, gamma rays or x-rays are all considered ionizing radiation but have very different properties. Contrary to electrons and protons, neutrons for instance, do not carry any charge and therefore do not interact directly with the electrons. Nevertheless, they can create secondary particles via nuclear reactions or elastic collisions and these secondary particles can ionize the atoms of the material. A consequence of this is that a detector that efficiently detects one type of radiation may not produce any signal when irradiated with some other type of radiation. Another consequence of the non-identical interaction of different radiation types with a material is that, depending on the detector, the signals that are generated can be categorized differently. A so called counter detector counts the number of interactions that occur within the sensitive volume whereas a spectrometer measures the deposited energy in the detector. Additionally we have the dosimeters which, as the name indicates, measure the absorbed radiation dose by the sensitive volume. There are also different methods to detect the radiation. The detection can, for instance, be direct 33 or indirect. In the latter case, the radiation interacts with the volume of the detector and a different kind of emission, e.g. light emission, is generated, which is then detected by means of another sensor. Examples of detectors that detect ionizing radiation indirectly are scintillation detectors, which in the case of GaN are especially interesting for the detection of neutrons and x-rays [63, 126]. When the detection of radiation is direct, this means that the incident radiation creates excess charge carriers in the semiconductor, which induce a current at the electrodes that is read by some acquisition system. The mechanism is thus comparable to what happens when we illuminate a detector with UV photons. In general, semiconductor radiation detector devices are also based on the same configurations as photodetectors, namely photoconductors or diodes. The big difference between ionizing radiation and photodetectors is naturally the energy of the incident radiation and the way it interacts with the material. Finally, the detection of radiation can be carried out in real-time or after the irradiation has taken place. For real-time detection typically scintillators and semiconductor detectors are used. A posteriori measurement of the impact of ionizing radiation is often used in dosimeters, where the effect of the ionizing radiation on the material modifies its properties and this modification can be correlated with the total deposited dose. 3.1.1 Interaction of ions with matter In this work, one of the main objectives is to use fabricated microwire sensors to detect charged particles, namely protons and heavier ions. When charged particles cross a semiconductor, they lose energy mainly through ionization and excitation of the atoms and molecules of the medium, and through nuclear interactions with the nuclei. Generally speaking, when an ion travels through a material it loses energy either by inelastic interactions with the electrons of the atoms of the material or by elastic collisions with the nuclei of these atoms. Other processes, such as nuclear reactions, photo-emission or Cherenkov radiation may occur but are not relevant for the irradiation conditions in this work. To account for the interaction of the ions with the electrons and nuclei of the material we can define the so called electronic stopping power, εe, and nuclear stopping power, εn, respectively. The stopping power quantifies the rate of energy loss per unit length in a material and the total stopping power can be defined by the sum of the electronic and nuclear stopping power. In the former process, the ion loses its energy mainly through ionization and excitation whereas in the latter, the energy is lost to the creation of phonons and atom displacements [127]. For radiation detection we are interested in the process of ionization and excitation, as it is through this process that free charge carriers will be generated. The transfer of kinetic energy to the electron by a charged particle per unit length is well described by the Bethe-Bloch equation [128, 129]: −dE dx =4πq4z2 mev2NZ ln 2mev2 Ie−ln(1 −β2)−β2(3.1) Here, vis the velocity and zthe charge of a charged particle, Nthe number of atoms per cm and Zthe atomic number of the material, Ieis a material constant dependent on the mean excitation and ionization energies of the absorber, meis the electron mass and β=v/c is known as the relativistic parameter. 34 When the velocity of the particle is much lower than the speed of light, the relativistic parameter can be neglected and we obtain −dE dx =4πq4z2 mev2NZ ln 2mev2 Ie (3.2) For high velocities, the term inside the square brackets varies slowly with energy and dE/dx essentially scales with 1/v2. In other words, the energy loss is inversely proportional to the particle’s energy. The reason for this is that, when the particle has a higher velocity, it spends less time in the vicinity of the electron and consequently the cross section of the interaction is reduced. For two different particles with the same velocity, particles with a larger charge lose energy at a greater rate due to the term z2 in the numerator. If we plot the energy loss as a function of the penetration distance in a material, we obtain the so called Bragg curve. As the charged particles penetrates deeper inside the material their velocity is reduced and hence dE/dx increases. Near the end of the track, the maximum of energy loss is reached. This maximum is called the Bragg peak. After the Bragg peak the dE/dx curve falls off due to electron pickup by the charged particle which reduces its total charge and consequently lowers the energy loss. The total energy that is deposited inside a material with a thickness dcan be obtained by integrating the stopping power over the thickness of the material. 3.1.2 Charge collection As said previously, the basic structures that are used to detect the deposited energy by a charged particle are based on conductive or diode devices, where the latter are much more popular nowadays. Of course, the exact configuration that is more appropriate depends on the specific application in mind. If one needs spectroscopic abilities it is important that the active region of the device is large enough so that the particle will deposit the entirety of its energy in the active region of the detector. On the other hand, when the aim is to measure the radiation flux, detectors with a thin SCR can be more useful depending on the penetration depth of the incident particles. The working principle of radiation detectors based on conductive mechanisms is similar to that of photoconductors in the sense that the radiation creates excess charges that drift to the electrodes by influence of an externally applied electric field. Therefore, the identified limitations when photoconductors were discussed also apply to ionoconductors. When considering detectors based on p-n junctions there are some different notions to take into account however. The main reason for this is that when a photon interacts with a semiconductor, only one electron-hole pair is created whereas an ion creates a very large number of electron-hole pairs. Detectors that contain SCRs are often denominated as solid-state ionization chambers when used to detect radiation with energies much higher than the bandgap energy of the semiconductor due to their similarity, in terms of working principle, with the traditional ionization chamber. However, one of the main advantages of semiconductors over, for instance, gas filled detectors is their much lower ionization energy. For GaN this energy is 8.9 eV [130] whereas in a typical gas-filled detector 30 eV are required to generate a pair [131]. This means that radiation with the same energy will create much more charge carriers in GaN than in a gas-filled detector. 35 If we consider a simple planar detector geometry where the semiconductor is fully depleted and sandwiched between a top and bottom electrode, the number of generated electron-hole pairs, neh, by a charged particle is directly proportional to the deposited energy Edep by the ionizing radiation: neh =Edep εi (3.3) where εiis the ionization energy. Both the presence of the internal and externally applied electric fields will separate the created charge carriers, with the holes drifting towards p-electrode and the electrons towards the n-electrode. This motion will induce a charge at the electrodes and consequently generate an electric current that persists until the charge carriers are fully collected by the electrode. Note that this last statement implies that the electrical signal starts to appear as soon as the charge carriers are created. In other words, the movement of the electrons and holes in the semiconductor generate a current so that the signal begins to form immediately after energy deposition. Since the position where the electron-hole pair is created is not necessarily equidistant from both electrodes, the drift distances will be different. Additionally, carrier mobilities are also different which means that either the hole or electron current will persist for a longer time than the other. If we integrate the current pulses generated by the detector using a measuring circuit, we can obtain the induced charge that is created by the incident ions and collected by the electrodes. Operation modes Generally speaking, solid-state radiation detectors that are based on charge detection can work in different operation modes which are defined by the method the charge is measured. In the context of this thesis, I will briefly refer to current mode detection and to pulse mode detection. The latter is the most common operation mode but there are certain application for which detectors that work in current mode are more appropriate. In both cases, the signals are a result of the charge that is collected by the electrodes. The difference between both modes is that in pulse mode, the detected signal corresponds to the total charge induced by a single particle or photon when it interacts with the semiconductor whereas in current mode, the signal is composed of several pulses generated by more than one particle or photon. Detecting radiation using the current mode is usually required when the rate of events, i.e the number of particles or photons that create charge carriers per unit time, is very high and the time between successive events becomes to short. In this case, if we assume that the measurement time, Tm, of the system is constant, the current generated by a sequence of events will be given by I(t) = 1 TmZt t−T i(t′)dt′(3.4) It is important to note that Tmis typically quite long when compared with the average time between individual current pulses which may not all have the same magnitude, hence the recorded current can be seen as an average that depends on the interaction rate and the average charge per interaction. Furthermore, the arrival of the radiation quanta is a random phenomenon and consequently, the time 36 intervals between successive events are also randomly distributed which introduces a statistical variation in the total measured current. Having a large Tmreduces the statistical variation but makes the system less sensitive to variation in rate and induced average charge. Nevertheless, most applications require the detector to work in pulse mode since information regarding the amplitude and timing of individual events is required. To detect pulses generated by a single charged particle or photon, the detector is typically connected to a preamplifier that converts the charge to a time dependent voltage signal. Using the correct calibration, this voltage is then converted to energy which is related with the energy that is deposited in the detector. All detectors that are applied in radiation spectroscopy or particle counting measurements are required to operate in pulse mode since information about the interaction of a single incident ion is required. For a detector with an arbitrary geometry, different methods have been developed to evaluate the instantaneous induced current by moving charges. Once again, it is important to note that the signal in the detector arises as a consequence of the motion of electron and holes inside the semiconductor after they are formed by the incident radiation. As soon as the last charge carrier is collected by its respective electrode the pulse is fully formed. A common way to calculate the shape of a pulse is using the Shockley-Ramo theorem [132, 133]. The Shockley-Ramo theorem makes use of the concept of a weighting potential that can be defined as the potential that would exist in the detector with the collecting electrode at unit potential while maintaining the other at zero potential. This theory was extended to account for the presence of a stationary space charge to be able to model the charge pulse formation in depleted regions of semiconductor devices [134]. The extended theory takes advantage of the Gunn theorem [135], which states that the current induced by a moving charge at the sensing electrode, isis equal to is=−q−→ v .∂−→ F ∂Vs (3.5) where ∂−→ F ∂Vsis the Gunn’s weighting field, defined as the partial derivative of the actual field, −→ F, with respect to the bias voltage, Vs, applied to the sensing electrode while maintaining the others at zero potential, and −→ vis the velocity vector of the moving charge. Note that the weighting field is not the actual potential in the detector but serves as a convenient way to determine the induced charge at an electrode by taking the differences in the weighting potential at the start and end of the carrier motion. If we assume that all charge carriers are created at a single point, equidistant from both electrodes Carrier current time time Q t1t2 faster charge carrier slower charge carrier Figure 3.1: Plots showing the ideal currents flowing through the semiconductor after the creation of electron-hole pairs by the incident radiation and the collected charge as a function of the time. In the right plot, t1and t2represent the time after which the faster and slower carriers are collected, respectively. 37 in a fully depleted planar detector, we can represent the resulting electron and hole current by the plot in figure 3.1. The different collection times are a consequence of the lower mobility of holes. Also presented in the figure is the measured induced charge by a circuit with an integration time constant that is longer than the collection time. As the charge carriers move towards the electrodes the induced charge increases. When the electrons are collected, they no longer contribute to the signal hence the slope decreases. The maximum induced charge is reached when the hole is also collected. This simple example assumes that all created charges are collected and does not take into account charges created outside the SCR. These charges will yield a smaller and slower signal since part of the generated carriers will recombine before reaching the electrodes and the path they need to cross to reach them is much larger. Charge Collection Efficiency One very important parameter of radiation detectors is their CCE. It is defined as the ratio between the total charge that is created in the semiconductor by an incident ion Q0and the total charge collected at the electrodes Q: CCE =Q Q0 (3.6) If the entirety of the charge that is created is collected, Q=Q0, and the CCE is 100%. This would be the case for a fully depleted detector where all the energy is deposited in the SCR and in absence of trapping of charge carriers. However, in any semiconductor there are defects that lead to the capture of electrons and holes and recombination through intermediate bandgap states that lead to a collected charge Q < Q0. Furthermore, in case the detector is not fully depleted, which is often the case, charge carriers generated outside the depletion region suffer from an increased recombination which also impacts the CCE. If we assume a planar structure of thickness d, with infinitely large non-segmented electrodes and a uniform electric field across the device we can write the CCE as [136] CCE =Q Q0 =1 dλe1−exp −x0 λe+λp1−exp −d−x0 λp (3.7) This equation is known as the generalized Hecht’s equation, where x0is the distance between the interaction position and the anode and λeand λpare the electron and hole drift lengths, which are assumed to be constant. Accordingly, besides being a parameter that defines the efficiency of radiation detection of a detector, the measurement of the CCE can also provide valuable information about the charge transport and recombination mechanisms due to its dependence on the carrier mobilities and lifetimes. A very common technique used for the measurement of the CCE of a detector is Ion Beam Induced Charge (IBIC) microscopy. 38 3.2 Radiation damage in GaN As was already mentioned in the introductory chapter of this thesis, one of the main properties that makes GaN stand out relative to other semiconductor materials is its high resistance to radiation. In a simple way, this means that higher doses of radiation are necessary in order to create enough damage in the crystalline structure of the material so that the properties will be notably degraded. One parameter that is directly related with the radiation resistance of a material is the threshold displacement energy Ed. This energy is defined as the energy that is required to remove one of the atoms of the crystal structure from its lattice site and replace it by a vacancy or substitutional atom. Molecular dynamic simulations were carried out to calculate the displacement energies for GaN and although minimum values of 18 eV for the Ga atoms and 22 eV for the N atoms were found, the average values were much higher, namely, 45 eV for Ga and 109 eV for N [28]. These values are relatively high when compared to other semiconductors, such as GaAs (13 eV) and Si (20 eV) [29, 137]. This does not mean that GaN is immune to irradiation damage but indicates that higher irradiation fluences and energies are required to create the same amount of defects in GaN when compared to these semiconductors. Another important parameter that affects the radiation resistance of a material are the dynamic annealing capabilities of the defects. Essentially, this property measures the rate of recombination of the created defects, which may occur even below room temperature. GaN is well known for its strong dynamic annealing. Hence, many of the created point defects recombine and do not contribute to the decay of the electrical and optical properties of the semiconductor. This property was studied through implantation experiments and it was shown that the defects in GaN are mobile even at 15 K and that defect recombination at 250 K decreases the defect concentration by 30% [138]. Although more prevalent for irradiation with heavier ion species due to the much higher amount of defects that are introduced, at high fluences it may also impact the defects created by proton irradiation and enhance the amount of defects that can be recovered by high temperature annealing of the device after irradiation. As said, the properties of the defects that are created depend heavily on the type, energy and fluence of the irradiation. Besides this, the geometry of the device or material will also impact defect creation. For instance, Rutherford Backscattering Spectrometry/Channeling (RBS/C) measurements carried out on GaN implanted with Ar ions showed that the damage build-up was different for a-plane GaN relative to c-plane and m-plane GaN [39, 40]. On the other hand, the density of vacancies that are created in the crystal lattice depends on the depth, therefore the thicknesses of the active layers of the device are also important considerations. Taking this into account, a lot of research has been carried out to better understand the type of defects and corresponding energy levels of the traps that are created in GaN when irradiated. It has been found that proton and electron radiation mostly produce simple point type defects, with the creation of Frenkel pairs being the most common [139]. Heavy ion and neutron irradiation are associated with more complex regions of extended defects [36]. Extended reviews regarding the damage accumulation processes in GaN can be found in references [36, 139, 37]. 39 2 µm 500 nm 500 nm (a) 5 µm 2 µm (b) Figure 4.3: SEM images showing the p-GaN shell for microwires from batch T2766 (a) and T2360 (b). In the latter we can see that the shell is relatively short and has significant voids and discontinuities. first 200 nm of the microwires, while the remainder of the wire stays essentially dislocation free, as shown in the TEM images in figure 4.2 [154]. Furthermore, the crystalline quality has also been demonstrated by nanoprobe coherent x-ray diffraction, an experimental method that is very sensitive to all structural defects. The measurement also revealed that the main defects are inversion domain boundaries i.e. the coexistence of domains of different polarity [156]. p-GaN shell quality One important characteristic of the core-shell microwires is the quality of the pGaN shell. In the SEM images we can easily identify the shell and as such, assess its quality. This becomes especially relevant because we identified that the microwires of one of the two growth batches (identified as T2360) that we had available to develop the sensors, presented shells that were relatively short and/or contained voids. Figure 4.3 presents the SEM images where this is visible and allows comparison with microwires of the other growth batch (identified as T2766), where the shells show better features. We noticed that this heavily impacts the operation of the devices since, when fabricating samples using batch T2360, the voids in the shell would allow the metal contacts deposited on the p-GaN extremity to touch the n-GaN core. As a consequence, no rectifying characteristic would be present. On average, in a fabrication run where 36 sensors are fabricated, if using batch T2360, only 4 sensors will present good electrical characteristics whereas when using batch T2766 this will be the case for almost all samples. No direct comparison was made between working samples fabricated with each batch however, and the majority of the results presented in this document were obtained using batch T2766. Nevertheless, some measurements were only made with samples from batch T2360. Whenever this is the case, it will be identified. 4.2 Fabrication of single nanowire devices In this section the microfabrication procedure that was followed to fabricate the radiation sensors will be covered. Similar steps were followed for the n-type and core-shell samples, the main difference existing in the metals that were used to deposit the contacts and subsequent annealing. Figure 4.4 shows a schematic of the microfabrication process containing all the important steps. Overall, the fabrication 46 (a) Initial substrate (b) TiWN (150 ˚ A) depositions (c) Marker definition (1st lithography) (d) TiWN etch (e) Microwire dispersion (f) 2nd lithography and Ti/Au (300/4000 ˚ A) deposition (g) Ti/Au lift-off (h) 3rd lithography and Ni/Au (300/4000 ˚ A) deposition (i) Ni/Au lift-off and final device Figure 4.4: Schematics of the main steps of the microfabrication process of single microwire detectors with a p-n junction geometry. The process used for n-type detectors is the same but only requires one metal contact deposition. process can be divided in three stages: substrate preparation, wire dispersion and contact deposition, and integration into a chip-carrier. In the first stage a metal layer is deposited and afterwards etched to define the alignment markers that are used to identify the position of the dispersed microwires. Then, two lithographies, two metal depositions and two metal lift-off steps are carried out to define the contacts of the device. The entire microfabrication process was carried out at the cleanroom installations of INESC MN, except for the contact annealing. For the microfabrication runsheets, where all the process step conditions are indicated, I refer to appendix A. 4.2.1 Substrate preparation The substrate preparation (figure 4.4 a-d) is very important in the fabrication process since the steps followed during this stage will enable the definition of the position of the dispersed microwires according to a fixed reference. Alignment markers are defined at specific positions which will act as reference 47 points to determine the coordinates of the wires, once these are placed on the substrate. Along the execution of this work, three different base substrates were used: 6 ′′ Si wafers covered with a 300 nm thick SiO2layer grown by dry thermal oxidation, 8 ′′ Si wafers covered with a 2 µm thick SiO2layer grown by wet thermal oxidation and 2 ′′ c-oriented sapphire wafers. I note that full wafers were processed in this stage except the eight inch wafers, which were divided in quarters, and that the processing was done in the same way for each substrate. The final output of this stage of the microfabrication are smaller square substrates with a lateral size of 20.2 mm, obtained by dicing the initial substrates. These are then used to perform the microwire dispersion, which will be explained in the next section. Alignment marker definition The alignment markers consist of small metal crosses, deposited across the wafers in square matrices. As said, the objective is to create a system through which we are able to determine the position of the microwires which is necessary since their positions on the substrate after dispersion are random. The first step consists in depositing a 150 ˚ A thick layer of TiWN (TiW is deposited in an N2rich environment) via magnetron sputtering using a Nordiko 7000 machine (figure 4.4 b). The choice of metal is not critical since the only requirement is a good contrast with the substrate. After the TiWN deposition, follows the optical lithography, which is carried out using a Direct Write Laser (DWL) II machine from Heidelberg Instruments. This equipment uses a HeCd laser with a wavelength of 442 nm and is capable of drawing structures with feature sizes down to 0.8 µm. Furthermore, the system makes use of softmasks, i.e. lithography masks that are defined through software instead of hardware, allowing flexibility in terms of mask design. Before exposing the wafers to the laser, they need to be coated with Photoresist (PR). To improve the adhesion of the PR to the wafer, the wafers are placed in a vapor prime oven where they are dehydrated and primed with hexamethyldisilizane (HDMS). They are then placed on the coating track (Silicon Valley Group (SVG) 88 coater and developer track) where positive PR is deposited through spin coating. The spinning velocity is set to 2500 rpm and lasts for 30 seconds, which results in a 1.45 µm thick layer of PR. The coating is concluded with a softbake step (85◦C for 60 seconds) to evaporate the solvent and make the PR more solid. The lithography map and mask must also be defined. The former delineates the configuration of the dies on the wafer and the latter represents the drawing that is made on each die. Figure 4.5 shows the mask and the map (for the six inch Si/SiO2wafer) that were used for this process. As we can see, the mask is comprised of four smaller squares, each representing a sample, and within these squares a square matrix of 12 by 12 alignment crosses are defined. We can see that each set of four alignment crosses are identified with a number (also visible in figure 4.6), representing a so called sample site. In each of these sites there is space for one radiation detector. Hence, the maximum number of detectors that can be build on a single sample is 36. From the map we can furthermore see that from a six inch wafer, we obtain a total of 28 smaller sample substrates. With the exposure map and mask defined, we proceed with the lithography. Taking into account that the subsequent step is the etching step we want to perform the lithography in such a way that, after 48 (a) (b) Figure 4.5: a) Map that was used for the lithography of six inch wafers; b) Mask that was used for the lithography of the alignment markers. the process, the structures defined in the mask are still covered with PR, while the PR outside of the structures is removed. Since we are using positive PR this means that the latter regions are exposed to the laser. As soon as the exposure is finished we place the wafer on the development track. Firstly, the wafer is baked at 110 ◦C for 60 seconds to stop any residual processes that might still be active and afterwards it is developed between 60 and 120 seconds. After cleaning and drying the wafer we are left with PR covering only the structures defined in the mask as can be seen from the optical microscope images in figure 4.6. The next step is the ion beam etching of the TiWN areas that are not protected by the PR. For this we use a Nordiko 3600 machine. This tool uses an ion beam to remove the unwanted metal, allowing very precise control of the etched thickness and uniformity by setting the correct etch time and angle. After the etching, the wafer is submerged into a beaker with microstrip, a chemical that will remove the PR that is still protecting the metal, and placed in a hot bath (T=65 ◦C) with ultrasounds. Figure 4.7 shows 200 µm 200 µm 12.5 µm 12.5 µm Figure 4.6: Optical microscope image of the alignment markers after the lithography. The dark areas indicate the presence of photoresist. 49 (a) (b) Figure 4.7: a) Six inch Si/SiO2wafer after the etching and PR strip steps; b) Optical microscope images showing the final alignment markers. a photograph of a six inch Si/SiO2wafer after completing the etching and resist strip steps as well as microscope images showing the alignment crosses. The final step that concluded this stage of the microfabrication process is the dicing of the wafer in smaller substrates. A Disco DAD 321 dicing saw is used for this purpose which cuts the larger substrates into the aforementioned smaller square substrates, with a lateral dimension of 20.2 mm. 4.2.2 Microwire dispersion The first step of the second microfabrication stage is the microwire dispersion, which is done by a droplet deposition process. To achieve this, first of all it is necessary to remove the wires from their growth substrate. A small piece of the growth substrate (approximately 0.25 cm2) is placed inside an eppendorf tube together with 50 ml of Isopropyl Alcohol (IPA). The tube is then placed inside an ultrasound bath that will detach the wires from the growth substrate, creating a suspension of microwires in the IPA. This (a) (b) (c) Figure 4.8: a) Photograph showing how the droplet dispersion is done; Optical microscope images of the substrate after dispersion (b) and after dispersion and cleaning (c). We can see that after cleaning the distribution of wires is uniform and that no agglomerations of wires are formed. 50 suspension is then transferred to a syringe which is used to deposit droplets on the device substrates as shown in figure 4.8 a). After some 30 seconds the IPA evaporates, leaving only the wires on the substrate. The wires present a uniform distribution across the substrate without creating agglomerates. Typically, around five droplets are necessary to achieve a full occupation of the sample sites (i.e. each sample site contains at least one wire). Since the dispersion leaves residues on the substrate, it is afterwards cleaned with acetone, IPA and DI water. Figures 4.8 b) and c) show optical microscope images of substrates after wire dispersion, before and after cleaning. WireFinder code With the wires on the substrate the next step is to define the contacts. This is achieved by drawing the contact paths using optical lithography, followed by a metal deposition and a metal lift-off. However, defining the contact paths is not straightforward since the distribution of the wires on the substrate is random. To avoid the tiresome task of drawing contact paths for each wire manually, this process was automatized using image processing and script writing. The code, written in python, takes microscope images of individual sample sites (these still need to be taken manually) as input and yields a script containing the contact paths, that can be run in CAD applications, as output. In a simplified way, the code analyzes the wires that are present in each sample site and chooses the best one (i.e. the one with the largest diameter and length), saving the coordinates of the wire extremities with respect to the four alignment crosses that define the sample site. Then it also defines the coordinates of the vertices of the contact paths towards these extremities. Figure 4.9 shows the output generated by the code when using the indicated microscope image as the input. The obtained CAD mask can then be used for the lithography. For a more detailed description of the WireFinder program I refer to appendix B. 4.2.3 Contact deposition Among the most important steps in the microfabrication process are the contact depositions. Depending on the type of junction that is created at the metal-semiconductor interface, the electrical behavior Figure 4.9: Visual workflow of the WireFinder code. The optical microscope image is given as input and through image processing, the code identifies the alignment crosses and wires that are present in the sample site. It then chooses one of the available wires and defines the coordinates of the vertices of the contact paths. These are written to a script that results in the shown CAD drawing. 51 of the sensor will be completely different. As explained before in chapter 2, in general two types of metal-semiconductor contacts exist: ohmic contacts and Schottky contacts. The former can be characterized by its linear current-voltage behavior, meaning it simply behaves as a resistance whereas when a Schottky contact is formed a potential barrier is generated at the metal-semiconductor interface and the contact behaves more like a diode. As we saw, the properties of the Schottky barrier depend on the metal and on the semiconductor properties. More specifically, we can state that the conditions that favor formation of ohmic contacts are metals with a low work function, in order to minimize the barrier height, as well as semiconductors with high doping concentrations to minimize the width of the SCR. Nevertheless, due to its wide bandgap, low Schottky barrier heights are difficult to achieve in GaN. For n-GaN, experimental studies showed that Ti, Al and Cr yield Schottky barrier heights of around 0.5 eV [108] whereas values around 1.5 eV were achieved for p-GaN using metals such as Ni, Au and Pt [157]. For n-GaN, the most important results regarding ohmic contacts have been reported for Ti based systems and the most successful configurations consist of a Ti/Al stack, covered with a barrier layer (most commonly Ni, Ti, Pd or Mo) and a Au capping layer, annealed at temperatures between 650 ◦C and 850 ◦C in innert environments. This multilayer stack has been found to yield stable contacts with specific resistances below 10-5 Ωcm2[158]. Cr/Au based contacts were also shown to yield good linear characteristics although they have been found to give larger specific resistivities [159]. In our case, the relatively high doping concentration of the n-GaN, especially the heavily doped bottom section, favors the creation of an ohmic contact. Consequently, even when a Schottky barrier is formed, a relatively small specific resistivity is expected. For p-GaN, the best results have been obtained using a Ni/Au stacks followed by an annealing step in an oxygen rich environment, although the obtained values for the specific resistance are about two orders of magnitude larger than those found for n-GaN [158]. The annealing step in an oxidizing environment is fundamental to reduce the contact resistivity since it enables the formation of NiO [160]. What occurs is that, during annealing, the order of the metal layers is inverted. TEM studies showed that the Ni diffuses through the gold to the surface where it forms a continuous layer of NiO while the Au gets into contact with the GaN. Although the role of this inversion in the reduction of the specific resistivity is still under debate, Ishikawa et al. pointed out that it enables the total or partial removal of the thin contamination layer present on the GaN surface formed due to the exposure to air [161]. Another possibility is related with the presence of residual H in p-GaN, that reduces the effective acceptor concentration NA[162] by forming Mg-H complexes. Annealing could induce the dissociation of these complexes and decrease the H concentration. As a consequence NAincreases and the specific resistivity lowers. The 3D geometry of the microwires also complicates the formation of a uniform film covering the GaN. To enhance uniformity and avoid cracks in the contact along the sidewalls of the wires, a planarization step is often used prior to contact deposition [113, 85]. In this work, this was not done and instead thicker layers were deposited to ensure that the entire wire is properly covered. Additionally, the fact that different metals are required to obtain ohmic contacts on p-GaN and n-GaN means that, for the core-shell wires we cannot do their depositions simultaneously. Instead, we need to do a lithography, a deposition and the subsequently lift-off for each contact. For fabrication of n-type 52 wire devices both contacts can be deposited at the same time. Optical lithography Before depositing the metals, the contact paths and patterns need to be defined on the substrate, which is achieved through optical lithography. Considering the dimensions of the wires and the minimum feature sizes we can define with the laser (0.8 µm), no e-beam lithography is needed. The procedure is the same as explained before with the exception of an extra step that is performed before exposing the sample. This is a pre-development step of the PR layer, during which we place developer on the substrate for 30 seconds, that is done to facilitate the lift-off of the metal after it is deposited. In this case, we need to remove the PR in the regions where we want to deposit the metal. Therefore, taking into account that a positive PR is used, the area inside these regions is exposed. The lithography masks that were used can be consulted in appendix A and figure 4.10 shows an optical microscope image of a small section of the resulting pattern after the PR is developed. Metal deposition The contacts of the devices are deposited by magnetron sputtering using an Alcatel SCM450 machine or a Nordiko 7000 machine. As mentioned before, the thickness of the deposited metal layers should be large enough to cover the wire adequately and avoid cracked contacts. Over the course of this work, three different n-contact configurations were used, namely Cr/Au, Ti/Au and TiWN/Al/TiWN. In the former two cases the deposition time was set to yield a 300 ˚ A thick layer of Cr or Ti and a 4000 ˚ A thick layer of Au, whereas in the latter case the bottom TiWN thickness was varied between 150 ˚ A and 300 ˚ A, the Al layer thickness was equal to 4000 ˚ A and the top TiWN was equal to 150 ˚ A. I note that the Cr/Au and Ti/Au contacts are deposited with the Alcatel 450SCM machine while the TiWN/Al/TiWN contacts were deposited with the Nordiko 7000 machine. The contacts deposited on the p-GaN shell were either Cr/Au or Ni/Au, deposited with the Alcatel, the first layer with a thickness of 300 ˚ A and the gold layer with a thickness of 4000 ˚ A. In all cases, the deposition was carried out at room temperature and with a normal 20 µm 60 µm (a) Ti/Au (300/4000 A) (b) Figure 4.10: Optical microscope image after the second lithography (a) and the metal-lift after the first contact deposition (b) for the n-GaN contact. 53 (a) (b) (c) Figure 4.11: Optical microscope image of a single wire device during processing, namely after the second lithography (a), the first lift-off and third lithography (b) and the second lift-off (c). incidence. Metal lift-off After contact deposition the excess metal is removed using a lift-off process. During this step, the substrate is submerged into a beaker with microstrip and placed for 3-5 hours in a hot bath at a temperature of 65 ◦C. Normally, ultrasounds are used to aid in the removal of the metals but in this case this is not viable since this will also remove the wires from the substrate. Figure 4.10 b) shows optical microscope images of the deposited contacts, in this case Ti/Au deposited on the n-GaN extremity of a core-shell wire after lift-off. Furthermore, in figure 4.11 we can observe optical microscope images of a single wire device after the second lithography and first lift-off of the p-GaN contact and after the second lift-off of the n-GaN contact. Contact annealing A critical step to obtain ohmic contacts is the thermal annealing performed after contact deposition. The annealing is done using a AnnealSys Rapid Thermal Annealer (RTA). This machine uses halogen lamps to heat the substrates and can reach temperatures up to 1500 ◦C with ramps up to 20 ◦C/s. Furthermore, the annealing can be done in air, vacuum, nitrogen or argon. Table 4.1 indicates the annealing conditions that were used for each type of contact in this work. Depending on the metal stack and the annealing temperature, some morphological modifications can be noticed when comparing the samples before and after annealing. This effect is most noticeable when annealing the TiWN/Al/TiWN contacts. Whereas Contact Metals Temperature Time Environment p-GaN contact Ni/Au 500 ◦C 120 s air n-GaN contact Ti/Au 500 ◦C 60 s N2 Cr/Au 500 ◦C 60 s N2 TiWN/Al/TiWN 600 ◦C 60 s N2 Table 4.1: Annealing conditions used for the thermal treatments of the contacts 54 40 µm (a) 40 µm 8 µm (b) Figure 4.12: Optical microscope image of as-deposited TiWN/Al/TiWN contacts (a) and annealed TiWN/Al/TiWN contacts (b). We can see formation of irregularities on the metal film, when using a higher magnification the bubble-like defects become very evident. before annealing the metal surface is smooth, after annealing we observe the formation of bubblelike defects across the contact area as demonstrated in figure 4.12. This indicates degradation of the contacts most likely due to Al oxidation processes. However, since the contact area with the GaN microwire is relatively small, this degradation may in some cases not affect the quality of the metalsemiconductor junction. For the gold based contacts, in some cases the roughness increases and the reflectivity seems to decrease locally, which may be due to the Ni diffusion to the surface and subsequent oxidation. 2 µm Ti/Au (a) 4 µm Ti/Au Ni/Au (b) Figure 4.13: SEM images of an n-type wire devices (a) and a core-shell wire device (b). An electron beam with an acceleration of 10 kV and apperture of 10 µm were used. The images were taken with an angle of 45◦between the beam and the substrate (colors are only for representation). 55 scanning an area it is more useful to consider the fluence as a function of the scan area, ΦA, rather than the fluence as a function of the beam area. To obtain this parameter we can use ΦA=Ibeam qAscan tpixelNpixel (5.2) where Npixel is the total number of pixels in the raster and tpixel is the irradiation time of each pixel, and we assume that Ibeam is constant during the scanning. When performing multiple scans we can than simply multiply ΦAby the number of full scans that are carried out. If we have a beam current of 100 pA, a scan area of 106x106 µm2and tpixel equal to 110 µs (standard conditions used during the experiments) we obtain FA= 4 ×1013 protons/cm2for one scan. I note that this is already a relatively high fluence and we found that it is easy to reach a fluence at which we already see some irradiation damage affecting the detectors that are being studied. As such, it is rather important to minimize the beam current to the lowest values possible, which for the present system are around 50 pA. We could reduce the fluence by increasing the scanning area, however since our samples have dimensions in the micrometer range, this would lead to a big loss in terms of resolution and is therefore not a viable option. Alignment procedure Besides requiring small scan areas, the reduced size of the samples also makes it very hard to precisely align the detectors with respect to the beam using only the magnification of the available lens system. Instead, we manage to place the sample in the correct position using PIXE maps. In other words, we are using the irradiation and the x-rays that are emitted from the elements present in the sample to locate the samples and move them to the right spot. This procedure consists of a short irradiation using a large scan area (e.g. 256x256 µm2). The reduced time and large area do not allow to detect any signal coming from the microwire but yield enough counts from the gold to show the deposited contact paths (yellow dots in figure 5.4). Through comparison with the optical microscope image it is possible to identify the location that is being irradiated. With this reference, we move the sample holder to the correct position and repeat the irradiation procedure with a smaller area. If the area is small enough, we already obtain some counts corresponding to the Ga from the microwire (N atoms are too light to be detected by PIXE) and we can place the wire in the center of the PIXE map with higher precision to conclude the alignment. The obtained PIXE maps during the alignment as well as optical microscope images are shown in figure 5.4. I note that one of the downsides of using this method to align the device is that we are irradiating the microwire before starting the actual experiment and, consequently, some irradiation effects may already exist. 5.1.2 1 MeV proton irradiation The experimental setup that uses the Tandem accelerator to generate the proton beam is different from the nuclear microprobe. First of all, the protons are accelerated to an energy of 1 MeV instead of 2 MeV. Furthermore, in this case the beam is not focused to a micron scale but rather to a centimeter scale. The advantage of this is that it allows us to irradiate several samples at the same time with 62 Figure 5.4: Alignment procedure of the sample with respect to the beam during measurement with the microprobe. The first alignment is done optically and (a) represents the first scan with an area of 264 by 264 µm2; (b) represents the second scan with an area of 264 by 264 µm2after vertical alignment; (c) represents the third scan with an area of 106 by 106 µm2; (d) represents the final alignment scan with an area of 106 by 106 µm2after horizontal alignment. Note that the yellow dots correspond to Au counts and the red dots to Ga counts. 63 Figure 5.5: Photographs of the Tandem accelerator at CTN and schematic of the experimental setup with the relevant components (not in scale). the same beam conditions but mostly that the beam flux is much lower. For a beam current of 50 nA (typical beam current used in the experiments) and a beam area of 1 cm2, the flux, calculated using eq. 5.1, is equal to 3x1011 protons/(cm2s), about four orders of magnitude lower than the point flux in the microprobe setup. Another advantage of the experimental setup is the larger irradiation chamber, which allows to place a UV LED inside the chamber. Consequently, we can monitor both the dark and photocurrent during the experiments. Figure 5.5 shows photographs of the Tandem accelerator and the chamber where the experiments are carried out, as well as a schematic of the experimental setup with the relevant components. Taking into account the beam size, the alignment of the beam can be carried out without using any special procedures. A glass piece with a scale is placed on the shutter that is directly in front of the sample. By manipulating the steering and focus of the beam, it is set to the correct dimensions and location by inspecting the light emitted by the glass when it is irradiated by the ion beam. The irradiation time can then be controlled by opening and closing the shutter. 5.1.3 Si irradiation Si ion irradiation was done at the Ion Beam Laboratory of the Ruder Boˇ skovi´ c Institute using a 1 MV Tandem accelerator. The top schematic of figure 5.6 shows the beamlines overview at the RBI accelerator facility. The working principle of the ion acceleration is the same as the one explained for the 1 MeV proton irradiation, with the difference that in this case a sputtering target is used as ion source. The experiments carried out in this work use Si2+ ions accelerated to an energy of 1 MeV and 750 keV. Similarly to the 2 MeV protons accelerated with the van de Graaff accelerator, the Si ions are also directed to a nuclear microprobe setup and the characteristics of the experimental setups are similar, except for a slightly different setup of the quadrupole lenses. The microprobe chamber at RBI uses two cylindrical poles and one conical pole, located closest to the irradiation chamber, with its top vertex extending into the chamber, as can be seen in the bottom picture of figure 5.6. The main advantage here is that this reduces the distance between the sample and the final quadrupole lens which in turn allows 64 Figure 5.6: Map of the ion beam laboratory facility at RBI and photograph showing the interior of the chamber where the IBIC experiments are carried out. a better focusing of the ion beam. As such, in this system it is possible to focus the beam down to lateral sizes of around 200 nm, significantly increasing the spatial resolution of the system. Besides this, the accelerator also allows much lower beam currents to be used, depending on the ion species it can be set to values in the range of 0.1-100 fA. Consequently, the beam flux and fluence are significantly lower when compared to the ion beams at LATR. The low beam size and currents make this experimental setup ideal for IBIC measurements, which is the main goal of the Si irradiation. Besides being equipped with additional detectors, the sample holder also contains a charge sensitive pre-amplifier that is used to collect the charges induced by the ion beam. For a full description of the IBIC technique I refer to section 5.3.4. The bottom photograph of figure 5.6 shows the interior of the irradiation chamber where the IBIC measurements are carried out, indicating the relevant components necessary for the the IBIC measurements. 65 5.2 SRIM simulations The interaction of radiation with matter is strongly dependent on the mass and energy of the incident ion, as well as on the irradiated material. As discussed before in chapter 3, the rate of energy loss per unit length is also called the stopping power and taking into account the process of energy loss it can be described by the electronic stopping power, εe, and the nuclear stopping power ,εn. The values of the stopping powers are strongly dependent on the mass and energy of the incident ion. Figure 5.7 shows the magnitude of the nuclear and electronic stopping powers as a function of the ion energy for both protons and Si ions in GaN. We can see that the obtained profiles are very distinct. For protons (or hydrogen ions) in GaN, the nuclear stopping is essentially negligible when compared with the electronic stopping over the entire energy range shown in the graph. For Si ions this is also true at energies above 500 keV however, at lower energies, nuclear stopping increases and eventually becomes dominant. The reason for this is that, when the ion energy is lower so is its velocity. Consequently ions spend a longer time near the nuclei and the cross-section of the ion-nuclei interaction becomes larger. We do not observe this for protons because their atomic number is much smaller. The values presented in figure 5.7 were calculated using the Stopping Range of Ions in Matter (SRIM) code developed by Ziegler et al. [164]. This is one of the most commonly used codes to simulate the transport of ions through a material and allows to calculate the range, and induced damage and ionization profiles. Obtaining this information is extremely relevant within the scope of this work since the response of the detectors is strongly dependent on how the ions lose their energy in the GaN microwires. As such simulations of the transport of Si ions and protons with the aforementioned energies in GaN were carried out. First of all, it is important to have a notion of the range of the particles in GaN, which are shown in figure 5.8. We can see that the ranges of protons and Si ions in GaN are also very different. This is essentially because, due to their atomic number, the stopping power of the Si ions is significantly larger and they lose their energy faster. The inset of figure 5.8 shows the projected ranges for the relevant energies in the scope of this work. We can see that the range of both 1 MeV and 2 MeV protons is much larger than the typical dimension of the diameter of the wire. Consequently, during the irradiation of the detector, they will only deposit a small part of their total energy in the GaN microwire. Hydrogen in GaN (a) Silicon in GaN (b) Figure 5.7: Nuclear and electronic stopping as a function of the ion energy for impinging H (a) and Si (b) ions in GaN as obtained via SRIM simulations 66 H (2 MeV) ~ 24 µm H (1 MeV) ~ 9 µm Si (1 MeV) ~ 0.71 µm Si (0.75 MeV) ~ 0.55 µm Figure 5.8: Range of Si ions and protons in GaN as a function of the ion energy as obtained via SRIM simulations. The table in the inset indicates the obtained ranges for the relevant energies within the scope of this work. The Si ions, on the other hand, come to a full stop at a distance below 1 µm, hence they do deposit the entirety of their energy in the microwires. Besides affecting the range, the different stopping powers also affect the ionization and damage profiles created by the Si ions and protons. Intuitively, one can already imagine that the silicon will have a larger impact on the atoms that compose the crystalline structure than the protons. It is a little bit like comparing firing a small bullet and a large canon ball against a wall. Figure 5.9 shows the ionization profile for both 2 MeV protons and 1 MeV Si ions obtained through the SRIM simulations and, as expected, the ionization created by the Si ions is much higher. The data shows that for Si ions, the ionization is maximum right at the surface and diminishes as the ion penetrates deeper into the GaN. This is a consequence of the decreasing electronic stopping power. At the same time the nuclear stopping power increases as the energy of the ion decreases. The ionization becomes zero at the depth at which the ion has lost all of its energy. Also shown is the ionization profile for a bi-layer configuration where the GaN is covered by a 400 nm thick layer of gold. This configuration simulates the contact region of the devices. First of all, we can see that the ionization goes to zero at a lower depth due to the energy the ion has lost in the gold layer. Nevertheless, according to the simulations we still expect some energy depositions in the active regions of the microwire. In case of proton irradiation, 0 250 500 750 1000 1250 1500 1750 2000 Depth (nm) 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0 Ionization ( eV Å . ion ) 2 MeV protons in GaN 2 MeV protons in Au(400 nm)/GaN (a) 0 250 500 750 1000 1250 1500 1750 2000 Depth (nm) 0 50 100 150 200 Ionization ( eV Å . ion ) 1 MeV Si in GaN 1 MeV Si in Au(400 nm)/GaN (b) Figure 5.9: Simulated ionization profiles for 2 MeV hydrogen ions in GaN (a) and 1 MeV Si ions in GaN (b), obtained via SRIM simulations. 67 0 250 500 750 1000 1250 1500 1750 2000 Depth (nm) 0 1 2 3 4 5 6 Vacancies/( Å -ion) 1e −5 1 MeV protons 2 MeV protons (a) 0 250 500 750 1000 1250 1500 1750 2000 Depth (nm) 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Vacancies/( Å -ion) 1 MeV Si 750 keV Si (b) Figure 5.10: Simulated vacancy profiles for 2 MeV and 1 MeV hydrogen ions in GaN (a) and 1 MeV and 750 keV Si ions in GaN (b), obtained via SRIM simulations. the ionization profile is almost constant across the diameter of the wire, increasing only slightly as the proton penetrates deeper into the material. The situation is similar for the Au-GaN configuration where the ionization is only slightly larger when compared to the previous case due to the energy loss of the protons in the gold layer. Similar assessments can be made about the created damage profiles by both types irradiation. Figure 5.10 shows the number of vacancies that are created per ion per unit length for both 1 MeV and 2 MeV protons, and 750 keV and 1 MeV Si ions. Once again, the number of vacancies created by the Si ions is much larger, almost four orders of magnitude, than the number of vacancies created by the protons. But we can observe a clear peak in the vacancy profile of the Si irradiated GaN, which was not present in the ionization profile in the considered energy range. This peak is called the Bragg peak and occurs right before the particle comes to rest in the material. It is a consequence of the increase in the nuclear stopping power when the energy of the ion becomes low. Naturally, since the protons do not come to a stop in GaN for the plotted range we do not observe the Bragg peak. I note that one limitation of the SRIM software is that it only allows simple layer geometries, in other words, the three-dimensional microwire structure cannot be modeled. Furthermore, SRIM simulations do not take effects such as dynamic annealing or ion channeling into account, as such the simulated number of created vacancies is an overestimation of the real value. Nevertheless, especially when using them for relative comparisons and range estimations it provides very valuable information. Non-ionizing Energy Loss (NIEL) Taking into account that we are using different types of particles and energies in the radiation damage studies, it is useful to describe the damage effects using the displacement damage dose Dd. This parameter is defined as the product of the average NIEL and the irradiation fluence. The NIEL is essentially the rate at which energy is lost to non-ionizing events and we can obtain the value of the NIEL through the SRIM simulations [165]. The NIEL as a function of the irradiation depth is shown in figure 5.11 and we can see that the profile is very similar to the vacancy profile shown in figure 5.10. The average NIEL values are equal to 0.05 MeVcm2/g and 0.12 MeVcm2/g for 2 MeV and 1 MeV protons respectively. 68 0 250 500 750 1000 1250 1500 1750 2000 Depth (nm) 0.00 0.05 0.10 0.15 0.20 0.25 NIEL (MeVcm 2 /g) Av. NIEL 1 MeV = 0.12 MeVcm 2 /g Av. NIEL 2 MeV = 0.05 MeVcm 2 /g 1 MeV protons 2 MeV protons (a) 0 250 500 750 1000 1250 1500 1750 2000 Depth (nm) 0 200 400 600 800 1000 1200 NIEL (MeVcm 2 /g) 750 keV Si 1 MeV protons (b) Figure 5.11: Simulated NIEL profile in GaN irradiated with 1 MeV and 2 MeV protons (a) and 750 keV and 1 MeV Si ions (b), obtained through SRIM simulations. Since the Si ions stop inside the active region of the devices, we cannot simply use the average NIEL values. Nevertheless, we can obtain the so-called adjusted NIEL [166]. For the considered energies of 1 MeV and 750 keV the obtained values are, respectively, 306.1 MeVcm2/g and 342.5 MeVcm2/g. Once we have the NIEL we can calculate Ddthrough the following equation Dd= ΦNIELAv (5.3) In semiconductors exposed to different types of radiation, good agreement was found between the NIEL of gamma, electron and proton irradiation [167] and proton irradiation studies performed in GaAs solar cells showed that the degradation of the maximum power output curves as a function of the irradiation fluence for different proton energies collapsed into a single curve when plotting the data as a function of the displacement dose. Only for low proton energies at which the ion would stop in the active layer of the device did the curves show deviations [166]. Although there are also cases in which analyzing the irradiation damage by means of the displacement dose is contested, since it does not take into account the influence of dynamic annealing or the impact of the ionization on the damage creation [168], the NIEL concept is generally very successful in correlating the effects of the induced damage in semiconductor devices [169]. 5.3 Device characterization 5.3.1 Electrical characterization The electrical characterization is carried out using an Agilent B1500A Semiconductor Parameter Analyzer (SPA). The chip carrier containing the wirebonded samples is inserted into a PCB with connectors for the cables that lead the signal to the measurement boards of the SPA. Alternatively, tungsten tips can be placed directly on the metal contacts of the sample with micropositioners. Normally, two channels are used, one terminal is set to a static zero voltage (GND channel) while the other channel applies a bias 69 to the device. In order to avoid high currents that can damage the sensors, the current that is allowed to cross the circuit can be limited by setting the right compliance. The system can be used to measure both static I-V curves and transient I-t curves. In the former, the applied bias is varied while the current is measured. During transient measurements, the bias is fixed and the current is measured over time, this is useful to perform transient photoor ionoconductivity measurements. For high resolution current measurements, i.e. with noise levels of the order of 1 pA, the minimum measurement rate is around 20 ms per point. This can be lowered to about 1 ms per point using the high speed measurement mode but this will lower the current resolution to nA values. 5.3.2 Photoconductivity measurements During photoconductivity measurements the detectors are illuminated with a light source which induces a variation in the conductivity of the detector. This change in the electrical properties can be measured by probing the current at the the terminals of the detector. There are many different methods to study the photocurrent in a detector and each of them can provide different information. In this work, the photoconductivity of the GaN microwires is studied mainly through static I-V and transient I-t measurements when irradiating the sample with a UV LED (365 nm, Thorlabs M365D1). The wavelength of the light source is important and needs to be smaller than the wavelength that characterizes the bandgap, since otherwise the photons do not carry enough energy to excite the electrons from the valence to the conduction band. for the static I-V measurements the light source is turned on before starting the measurements and turned off when the measurement is finished. This allows to study the photocurrent dependence on bias and compare the obtained I-V curve with the one measured in dark conditions. In transient measurements, the light source is turned on and off while performing the measurement. Like this, the transient response when changing from dark to an illuminated environment and vice-verse can be analyzed. Another important experiment is the measurement of the photocurrent as a function of the incident Figure 5.12: Photocurrent measured with a commercial Si p-i-n photodiode (Thorlabs FDS010) while illuminating it with the UV LED (365 nm, Thorlabs M365D1). The inset on the left indicates the parameteres extracted from the linear regression of the data as well as the R-square parameter, which indicates the good linearity of the photocurrent with the LED input current. The inset on the right represents the equation that is used to convert the measured photocurrent into the optical power. Apd and Rpd are the active area and the responsivity of the photodiode at the considered wavelength, respectively. In this case, Apd = 0.8mm2and Rpd = 0.051 A/W 70 optical power (or irradiance) of the detector. Since the LED is powered by a power supply that has a variable input current we can control the optical output of the LED. To correlate the optical output of the LED with the irradiance incident on the detector, we use a commercial Si photodiode (Thorlabs FDS010) with known responsivity at the relevant wavelength for calibration purposes. Figure 5.12 shows the photocurrent as measured by the photodiode as a function of the input current provided by the power supply. Using the data points we can obtain the optical power as a function of the input current by dividing the photocurrent data by the sensitive area of the detector and the detector responsivity, followed by a linear regression. It is important to note that the photocurrent changes linearly with the input. Assuming that the photodiode response is linear with optical power, this shows that the optical power also varies linearly with the input current of the LED. Secondly, the calibration measurements are very sensitive to the position of the sensor relative to the light source, hence, it is fundamental to carry out a new calibration for each measurement series. In addition to the static I-V and transient I-t measurements, photocurrent spectroscopy measurements are also carried out. Although ideally the photocurrent response for photons with an energy below the bandgap should be zero, in reality this is often not the case. Photocurrent spectroscopy allows to measure the magnitude of the photocurrent as a function of the wavelength of the incident photons. In this case, a light source with a broad emission spectrum needs to be used to cover the largest possible interval of wavelengths. In this work, setups that use a Xe lamp or a combination of a halogen and deuterium lamp were used. The spectra of both light sources are such that the lowest emitted wavelength is well below the GaN bandgap wavelength of 364 nm. To disperse the light into individual wavelengths, a monochromator is used. Naturally, this also impacts the intensity of the light and in some cases the photocurrent induced by the low intensity light is too small to be measured without using any amplification of the signal. To overcome this, a lock-in amplifier, together with a chopper to chop the light with the adequate frequency, available at the laboratory for photoconductive measurements at IST, led by prof. Figure 5.13: Experimental setup, available at the laboratory for photoconductive measurements at the Physics department IST, used for the photocurrent spectroscopy measurements that use a lock-in amplifier to amplify the output signal. 71 with Cr/Au and Ti/Au contacts see their resistance increase respectively from 565 Ωto 1087 Ωand from 714 Ωto 769 Ωafter annealing at 500 ◦C, whereas for the samples with TiWN/Al/TiWN contacts, it drops from 1333 Ωto 1299 Ωafter annealing at 600 ◦C. The fact that the variation of resistance is rather small for the Ti based contacts and that the major modifications of the I-V curve occur when applying a reverse bias, indicates that the changes occur mainly at the junction deposited at the moderately doped extremity of the wire. The bigger increase of the resistance of the sample with Cr/Au contacts is possibly due to the formation of CrN structures at the interface, however no experimental studies were done to confirm this. Additionally, for the TiWN/Al/TiWN contacts the annealing time was also optimized by studying the linearity of the contacts as a function of the annealing time through extraction of the R2parameter from the linear regression, the optimal annealing time was found to be 60 seconds (the experimental data is presented in appendix C). It is important to note that the fact that we obtain ohmic contacts with the TiWN/Al/TiWN stack does not necessarily mean they are the better choice for all device processing. Contacts based on Cr/Au and Ti/Au tend to have a lower contact resistivity due to the fact that the resistivity of TiWN is rather high. Add to this that the morphology of annealed TiWN/Al/TiWN contacts is quite bad as was shown in figure 4.12 in chapter 4 and the choice for an annealed TiWN/Al/TiWN stack, albeit yielding linear contacts is not obvious. This is especially true when considering the fabrication of the p-n junction detectors, since in this case the n-type contact will always be deposited on the heavily doped section, and we already saw that the ohmic characteristics of Cr/Au and Ti/Au contacts when deposited on this extremity is quite good. On the other hand, when interested in studying the photoconductive mechanisms of the GaN microwires, linear contacts are desired. 6.1.2 Photoconductivity measurements Studying the photoconductive properties of the microwires is exactly one of the objectives and, as such, samples fabricated with TiWN/Al/TiWN contacts were studied using the different experimental photoconductivity methods presented in chapter 5. At the same time, Schottky diode devices with Cr/Au contacts and an almost rectifying I-V curve were also measured. Figure 6.3 shows the photocurrent I-V −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Bias (V) −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 I (mA) Metals: Ti/Au Thickness: 15/400 nm Annealing conditions: 500 ∘ C, 60 s, N 2 Not annealed Annealed (a) −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Bias (V) −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 I (mA) Metals: Cr/Au Thickness: 15/400 nm Annealing conditions: 500 ∘ C, 60 s, N 2 Not annealed Annealed (b) −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Bias (V) −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 I (mA) Metals: TiW/Al/TiW Thickness: 15/400/15 nm Annealing conditions: 600 ∘ C, 60 s, N 2 Not annealed Annealed (c) Figure 6.2: Comparison between the dark I-V curves before and after annealing for n-type microwire samples with Ti/Au contacts (a); Cr/Au contacts (b) and TiWN/Al/TiWN contacts (c). The annealing conditions are specified in the inset of each graph. We can see that the annealing lowered the overall conductivity of the samples with Ti/Au and Cr/Au contacts indicating enhancements of the Schottky barriers but transformed the TiWN/Al/TiWN into ohmic contacts. 78 −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Bias (V) −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 I (mA) Dark UV (a) −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Bias (V) −0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 I (mA) Dark UV (b) Figure 6.3: Comparison between the dark current and photocurrent for an n-type detector with a linear I-V characteristic (a) and an asymmetric I-V characteristic (b). We can observe a photocurrent in both situations but whereas in (a) the photocurrent is the same for both negative and positive bias, in (b) the photocurrent has a larger magnitude for positive bias. curve together with the electrical characteristics measured with the detector in dark conditions, for two representative samples. As can be seen, the increase in current due to photoexcitation, albeit small, is clearly observed for both positive bias and negative bias. Nevertheless, whereas for the linear device the photocurrent has more or less the same magnitude in both directions, the asymmetric device clearly shows a higher absolute photocurrent for positive voltages in comparison with negative voltages. More specifically, for the linear device the total induced current (i.e. the photocurrent minus the dark current) is equal to 69 µA and 64 µA at 1 V and -1 V, respectively. For the asymmetric device the values are 66 µA and 12 µA, respectively. On the other hand, if we consider the relative change in the current, i.e. the amount by which the photocurrent increases with respect to the dark current, we obtain a percentage of 16% at 1 V and of 42% at -1 V for the asymmetric detector. In other words, the signal-to-noise ratio is higher when the more rectifying contact is in reverse bias. To get a better idea about the evolution −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Bias (V) 1.00 1.02 1.04 1.06 1.08 Signal-to-dark ratio (a) Linear sample −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Bias (V) 1.0 1.2 1.4 1.6 1.8 2.0 2.2 Signal-to-dark ratio (b) Asymmetric sample Figure 6.4: Signal-to-noise ratio plotted as a function of the applied bias for an n-type microwire detector with linear contacts (a) and asymmetric contacts (a). We can see that for the linear sample the signalto-noise ratio is also independent of bias. For the asymmetric sample this is not the case and when the more rectifying contact in reverse bias the signal-to-noise ratio is larger. The highest signal-to-noise ratio is achieved at low positive voltages however, with a maximum value close to 2. I note that the discontinuities at 0 V are due to the division by zero and that the signal-to-noise ratio of 1 for the linear sample close to 2 V and -2 V are due to the current compliance that forces the currents to be equal. 79 2.225 2.250 2.275 2.300 I (mA) Bias = 2 V 0 1000 2000 3000 4000 5000 Time (s) 2.06 2.08 2.10 2.12 I (mA) Bias = -2 V (a) 1.20 1.25 1.30 I (mA) Bias = 2 V 0 200 400 600 800 1000 1200 1400 Time (s) 0.325 0.350 0.375 0.400 I (mA) Bias = -2 V (b) Figure 6.5: Transient I-t curves at a bias of 2 V and -2 V for an n-type microwire detector with linear contacts (a) and asymmetric contacts (b). The response times are similar except for the photocurrent of the asymmetric detector measured at negative bias, which presents a significant increase in rise and decay time. of the signal-to-noise ratio in terms of the applied bias the data is plotted as a function of the voltage in figure 6.4. A very interesting behavior emerges, first of all we can indeed see that the asymmetric profile is also visible here and overall, the signal-to-noise ratio values are larger for negative voltages. Nevertheless, the maximum signal-to-noise ratio is obtained at low positive voltage, more specifically at 0.14 V with a value of approximately 2. For higher positive voltage a sharp drop occurs after which a slower decay with bias follows. The drop at negative voltages also exists but is much more subtle. The fact that the maximum signal-to-noise ratio occurs at low positive voltage seems to indicate that the less rectifying contact, although being almost linear, still plays a role in the photocurrent generation when the bias is small enough. If we look at the signal-to-noise ratio of the linear device, the profile is constant and independent of bias. Furthermore, overall the values are smaller which allows us to confirm that the different behavior is in fact caused by the contacts. Whether the more rectifying or more ohmic contact is in reverse bias also has a significant impact on the transient behavior of the photodetector. Figure 6.5 shows the transient I-t curves at 2 V and -2 V, once again for the linear and asymmetric device. The measurements were carried out by fixing the bias and measuring the current induced when turning the UV LED ON and OFF after specific time intervals. The sampling rate, τs, of the system was set to 50 ms and in all measurements the input current of the LED was set to its highest value (700 mA). The first observation is that we can observe a persistent photocurrent (PPC) decay in all measurements. However, whereas the decay seems to be the same for positive and negative bias for the linear device, the persistent decay in the curve measured with the asymmetric device at negative voltage is significantly more pronounced when compared to positive voltage. Furthermore, the rise profile of the photocurrent is also much slower. Albeit the observation is clear, the reasons we observe it are not necessarily. Intuitively, one would think that when a reverse bias is applied to the more rectifying contact this would promote not only higher relative values of photocurrent but also faster response times. To better explain this, it is important to understand the different mechanisms that can be responsible for persistent photocurrent decays in GaN. Nevertheless, before doing this it is worth to complete the photoconductivity picture of the n-type 80 Fit parameters: = 0.164 ± 0.01 = 0.350 ± 0.06 = 0.220 ± 0.01 (a) 300 320 340 360 380 400 420 440 Wavelength (nm) I PC (a.u.) Bias = 1 V Bias = 3 V 4 3.6 3.4 3.2 3 2.8 Energy (eV) (b) Figure 6.6: (a) Photocurrent of an n-type microwire detector as a function of the LED input current and results obtained through fitting the data to the power law function; (b) Photocurrent spectroscopy data at 1 V and 3 V. We can see that the spectra are independent of the bias. The dashed line represent the bandgap wavelength/energy of GaN. microwires. As written before, one of the important FOMs of a detector is its linearity with radiation intensity. To evaluate this, the input current of the LED was varied while measuring the current in transient mode at a bias of 2 V for a linear sample. The photocurrent values for each LED current input where then extracted and plotted as a function of the LED current as shown in figure 6.6. Additionally, the experimental data was fitted using the power law function (eq. 2.22) described in chapter 2. We can see that that the photocurrent is not linear in this case and that βgives a value of 0.35. This is not entirely surprising since non-linear photocurrent behavior is common in photoconductors. The other graph in figure 6.6 shows the photocurrent spectroscopy measured using the deuterium-halogen lamp. The photocurrent spectra were measured while applying a voltage of 1 V and 3 V to the detector. We can see a clear cut-off of the photocurrent at the wavelength that corresponds to the GaN bandgap. A small tail is visible up to a wavelength of around 380 nm. This tail can be attributed to the existence of a large number of states in the vicinity of the conduction band as a result of the high doping concentrations [174, 175]. PPC and gain mechanisms in GaN As soon as GaN emerged as a candidate material for photodetection application, it became evident that the photodetectors were characterized by very slow non-exponential photocurrent decays [101]. Furthermore, very high photocurrent gains as well as strongly non-linear photocurrents as a function of the optical irradiance were observed. The presence of deep levels and trapping states located within the bandgap was readily pointed out as the culprit for the presence of these effects [106]. Trapping of carriers inevitably leads to an increase of the recombination time and different defect levels were proposed as candidates to explain the PPC behavior in GaN [176]. But, motivated by the lack of a physical model explaining all features of GaN based photodetectors, Garrido et al. developed a model based not on the modulation of the conductivity, but on the modulation of the conductance volume [107]. They proposed that SCRs, present inside the semiconductor due the existence of grain-boundaries, metalsemiconductor interfaces or surface states, decrease in size in the presence of light. When excess 81 carriers are generated they are swept by the electric field in the SCR and become trapped at, for example, the surface states. This decreases the surface charge density and leads to a reduction of the SCR width, which, in turn, expands the conduction volume. By considering the expansion of the conduction volume, the experimentally observed photoconductive gain could be adequately explained. Besides this, the model also provides an insight into the dynamic behavior, namely the non-exponential decay. Essentially, as the light source is turned off, carrier generation is stopped and the carrier recapture process begins. As carriers that are trapped at the surface states recombine, the SCR width starts to recover. Due to this recovery, the probability of carriers recombining decreases because the potential barrier they have to surpass grows with each recombination. The lowering of the carrier capture rate results thus in very slow and non-exponential decay times. Additionally, the model also predicts a non-linear dependence of the photocurrent with optical power that is in good agreement with experimental data. A limitation of the model was that it could not completely explain the existence of gain in both reverse and forward bias of GaN based Schotty detectors. To explain this, Katz et al. proposed a similar mechanism, based on the lowering of the Schottky barrier due to the trapping of the minority carriers at the metal-semiconductor interface [109]. This model was equally successful in explaining the slow response times [110]. If we consider the devices we are studying, the Schottky barrier reduction and recovery dynamics are a prime candidate to explain our results. Being a consequence of the lowering of the potential barrier at the metal-semiconductor junction, it is natural that the contact that is more rectifying is more affected by this mechanism. This does not mean that it does not occur when the more ohmic-like contact is in reverse bias but, due to the high probability of electron tunneling, its effect is suppressed. Although the impact of light on the Schottky barrier height explains the asymmetry in the measured decay times, the surface states may also contribute. In fact, taking into account the large surface-tovolume ratio of the wires its impact can be significant. Essentially, the surface states pin the Fermi level which induces an upward bending of the valence and conduction bands. As a consequence, holes will tend to move towards the sidewalls while the electrons will prefer to stay in the center. Due to this separation, the recombination rate of non-equilibrium carriers is reduced. The dependence of the photoconductivity on the size of GaN wires was first reported by Calarco et al. [177]. They showed that the height of the potential barrier created at the surface depends on the diameter of the wire. For nanowires with a small enough diameter, the SCR extends all the way into the nanowire core, leading to complete depletion and less band curvature. In this case, because the potential barrier is lower, the recovery of the photocurrent was shown to be fast since it is easier for the electrons to overcome the barrier and recombine with a hole. For wires with a diameter larger than a critical diameter the wires are only partially depleted and slow transients were measured. The diameter of the microwires used in this study is too large to cause complete depletion of the microwire core, hence the surface band bending will also contribute to slow transients. The Fermi level pinning and band bending can also explain the appearance of the bandgap tail measured by the photoconductive spectroscopy measurements through the Franz-Keldysh effect, which gives rise to an absorption tail in the presence of an electric field. Cavallini et al. identified this as a possible mechanism for sub-bandgap absorption by showing 82 that the band tail widths varied with the diameter of the GaN nanowires [178]. It is thus clear that there are many factors that can reduce the recombination rates when the light source is turned off during the transient I-t measurements. Especially, it became evident that the metalsemiconductor interface plays an important role, considering the much slower decay obtained when the more rectifying contact is set to reverse bias. Overall, although detectors based on n-type microwires present some interesting properties, namely, the high absolute photocurrent values owed to the existing gain mechanisms, they also have some strong limitations. The slow decay times, non-linear photocurrents and low signal-to-noise ratios limit their applicability in a market where speed and signal strength are fundamental. 6.2 Characterization of single GaN core-shell p-n junction microwires In the current section of this chapter I will describe the results obtained through the characterization of devices based on the single core-shell p-n junction microwires using electrical and optoelectrical measurements. In a first stage I will focus on the electrical characteristics without excitation to then present the photoconductivity results. Most of the presented results were obtained using samples with Ni/Au contacts on the p-GaN extremity and Ti/Au contacts on the n-GaN extremity. When this is not the case this will be indicated in the text. 6.2.1 Electrical characterization In this case, the electrical characterization is carried out with the channel connected to the n-GaN contact set to a static zero voltage (GND channel) while the other channel, connected to the p-GaN contact, applies a bias to the device. In this configuration the diode is in reverse mode when the applied bias is negative and in forward mode when the applied bias is positive. The voltage is swept from -4 V to 4 V and the current compliance is limited to 1 µA. Furthermore, to enable current measurements with −4 −3 −2 −1 0 1 2 3 4 Bias (V) 10 −13 10 −12 10 −11 10 −10 10 −9 10 −8 10 −7 10 −6 I (A) (a) −4 −3 −2 −1 0 1 2 3 4 Bias (V) 0 200 400 600 800 1000 I (nA) (b) Figure 6.7: Dark I-V curves for a representative p-n junction microwire detector with Ni/Au contacts on the p-GaN extremity, in logarithmic scale (a) and linear scale (b). We can clearly observe a rectifying behavior although significant deviations exists with respect to the ideal p-n junction diode equation given by eq. 2.10. 83 a noise level below 1 pA we set the integration factor to 1 power line cycles (PLCs). This essentially means that we increase the integration time of the measurement to a value such that the measurement error caused by noise from the AC supply voltage is eliminated. Figure 6.7 shows the dark current-voltage (I-V) characteristic of a representative p-n junction microwire sample in linear and logarithmic scale. The rectifying behavior of the p-n junction is clearly evident, when applying a reverse bias, the current has much lower values compared to the forward bias current and, when the forward voltage is higher than 3 V, the current increases exponentially. Nevertheless, we can also see clear deviations from the ideal p-n junction diode I-V characteristic. Firstly, in reverse bias we observe a significant leakage current, that follows neither a linear nor a constant value, indicating that it is not governed by a low shunt resistance or high saturation current. For low forward voltages we observe what looks like an exponential increase of the current, recognizable from the linear section in the semilog plot. The rate of increase lowers for V > 1V and the growth tends to become more linear, as predicted by the electrical model including a series resistance explained in chapter 2. However, for a forward bias superior than 2 V, the linear behavior turns once again into an exponential behavior but this time with a very low slope. Only when the voltage exceeds a certain value, the exponential growth of the current becomes faster once again. The turn-on voltage for this latter increase is around 3 V which is close to the expected value of the built-in voltage of a GaN p-n junction but, as will be shown, this voltage value is also dependent on the specific contact properties. This brief qualitative analysis indicates that the transport model cannot be fully described by the diode model that includes a series and shunt resistance, but is composed of a more complex set of physical mechanisms. To better understand these mechanisms, the forward and reverse bias I-V curves will be analyzed separately in the following subsections. Forward bias current As mentioned before, in an ideal diode the forward bias current described by eq. 2.9 yields an exponential growth for increasing bias. If we include the existence of a series resistance, the increase becomes linear when the voltage drop across this resistance is no longer negligible. The I-V curve shown in the previous figures does not follow this pattern. We do observe an initial exponential increase between 0.5 V and 1 V, which then slows down and after which the curve presents a small linear section (between 1 V and 2 V) but between 2 V and 3 V it starts growing exponentially once more. Although initially this growth has a much lower slope, when the bias exceeds 3.5 V the exponential increase becomes steeper again and the current grows rapidly until reaching the set current compliance. Figure 6.8 a) shows the ideality factors that were extracted from the linear sections in the semilogrithmic plot using eq. 2.12 and we can thus identify four regions: •Region I: 0.5 V <V<1 V - Exponential increase of the current with n= 3.71 •Region II: 1.5 V <V<2 V - Almost linear increase of the current •Region III: 2 V <V<2.75 V - Exponential increase of the current with n= 34.78 84 (a) (b) Figure 6.8: a) Measured forward current as a function of bias where three distinct regions can be identified. Region I, III and IV have linear slopes in the semilog plot and the ideality factor can be extracted; b) Schematic representing the equivalent circuit of a p-n junction wire detector. The circuit contains three diodes, corresponding to both metal-semiconductor interfaces and the p-n junction. Furthermore, shunt resistances for each diode as well as the resistances of the neutral regions were included. •Region IV: 3 V <V<3.2 V - Exponential increase of the current with n= 5.36 The first observation to be made is that whereas the ideality parameters of the first and fourth region are within reasonable distance with typical values, this is not the case for region III. In fact the distinction between region II and III is not evident in all devices and the exact current dependence on the applied bias varies slightly. Nevertheless, most samples follow the generic behavior described by the four regions so in the upcoming analysis these will be considered. In this discussion, which will be mostly qualitative, two mechanisms that are most likely impacting the forward bias current are presented. The first one takes into account current modulation by the Schottky barriers formed at the metal-semiconductor interfaces and the second one is based on current tunneling of carriers through the SCR. To deconstruct the I-V curve and understand the behavior of the devices it is important to consider the different junctions present that can influence the current. Figure 6.8 b) shows a simplified schematic of the equivalent circuit composed of the p-n junction diode, the two metal-semiconductor interfaces and the neutral regions. Each diode is modeled including a shunt resistance to account for possible alternative current paths and two resistances, RPand RNare included to account for the transport across the neutral regions. From the schematic we can see that the diodes corresponding to the metalsemiconductor interfaces have an opposite polarity with respect to the p-n junction diode. This means that when when we apply a positive bias to the p-GaN, the latter is in forward bias but the Schottky diodes are in reverse bias. Following the reasoning carried out when discussing the metal-semiconductor interfaces in chapter 2 and the n-type microwire I-V characteristics, we can assume that the contact on the n-GaN simply acts like a resistance due to the high doping concentration at this wire extremity. In this case we are left with three main components, the p-GaN Schottky diode, the p-n junction diode and a series resistance which includes the resistance of the neutral regions and that of the n-GaN contact. The existence of other junctions, with opposite polarization to the p-n junction, can impact the IV curves mainly when these junctions are reversely biased. It was shown by Shah et al. that the ideality factor extracted from the I-V curves is no longer solely dependent on the dominating transport mechanisms of carriers through the p-n junction barrier. Instead, they developed a model that also 85 took into account the contribution from other junctions, in their case a Schottky barrier contact and a heterojunction formed by GaN and AlGaN[179]. Since the current densities of these junctions also follow an exponential dependence with their own characteristic slope, the total current density of the system can be written in the form ln I(V) = qV PinikbT+Piniln I0,i Pini , (6.1) where it is assumed that V≫kbT/q so that exp(qV/nkbT)≫1. Here I0,i represents the saturation current and nithe ideality parameter of each junction. The slope of this curve can thus be written as dln I dV =q kbTX i ni(6.2) indicating that the externally extracted ideality factor is composed of the sum of the individual contribution of each junction: n=X i ni(6.3) . This result makes it clear that values of nhigher than 2 are viable when additional junctions are present in the device. Due to the difficulty of depositing a contact with low resistance on p-GaN, this is often the case, especially when considering nanoand microwire devices, hence they many times suffer from high ideality factors [180, 125]. In our case we observe that at relatively low forward bias (V < 1) we have n= 3.71 but that in region III we obtain n∼30. The change in the ideality factor can thus be explained by the increasing influence of the Schottky barrier as we raise the forward voltage and, when the bias is around 2 V, the major voltage drop occurs at the p-GaN contact with the current being blocked by the potential barrier. A similar I-V behavior was observed by Aspitarte et al. using carbon nanotubes based p-n junctions in series with Schottky barriers. They observed plateau like behavior when the current was limited by one of the Schottky barriers and exponential growth of the current when defined by the p-n junction [181]. The transition that occurs between the first plateau and the second exponential growth was explained as being a consequence of the decrease of the n-region width of the device due to band bending induced by the Schottky contact. At high enough forward bias, the bending is strong enough so that effective length of the n-region becomes smaller than the diffusion length of the holes. As a consequence, holes will be swept directly into the electrode on the n-doped side by the electric field of the Schottky barrier and turn the diode ON [182]. Considering the microwire structures this explanation does not apply to justify the transition from region III to the the final region where the current resumes its exponential increase with an ideality factor of 5.36. An alternative explanation is based on the breakdown of the Schottky barrier deposited on the p-doped GaN that would allow the flow of electrons from the semiconductor into the metal. In this case, the metal-semiconductor junction assumes a resistive behavior and the current is once again defined by the p-n junction. One argument in favor of this explanation is the fact that the voltage at which this breakdown occurs, is related with how rectifying the contact is, as will be shown further below. Contacts that are less rectifying, also show a lower breakdown voltage. 86 However, we cannot exclude the presence of other mechanisms acting at the same time. Although the high ideality factor is often attributed to the poor quality of the p-GaN contact it has also been attributed to the presence of heterojunctions between two layers with different composition [179], poor carrier transport properties of quantum wells inside the active region of a LED [183], asymmetric doping profiles of the p-n junction [184] and the existence of tunneling of carriers assisted by deep-level traps [95]. The first three mechanisms are not applicable in our case but tunneling currents may play a significant role in the shaping of the I-V curves. As explained before, the tunneling of carriers can occur directly from the valence to the conduction band or via trap-assisted mechanisms, and it was found that the electric field dependence of the forward tunneling current follows an exponential of the same form as the diode equation [95]. If the trap energy ETin eq. 2.15 has a value such that dlog(I)/dV is larger than 2 this leads to n > 2. The presence of tunneling currents have been observed in several GaN based LED structures and is often associated with the high level of dislocation defects that exist in epitaxial GaN. When comparing the I-V curves of two LEDs, Cao et al. saw, for instance, that the LED that presented a lower ideality factor also had a lower reverse bias leakage. The higher magnitude ideality factor and reverse bias current of the other LED was attributed to the poorer crystalline quality [95]. Other authors also identified defectassisted tunneling as the cause for the existence of forward leakage currents in GaN based LEDS when applying low voltages, and ideality factors between 2 and 6.5 were reported [185, 186]. Forward bias current behavior studied using temperature dependent measurements in GaN nanowires also showed agreement with tunneling mechanisms [187]. The fact that an additional exponential governs the current may lead to the appearance of exponential growths with different slopes. With this in mind, it is also possible that the current in region I of the forward bias I-V curve is a consequence of tunneling rather than being defined by the diffusion of carrier through the SCR. Even very large ideality factors can be caused by tunneling when several types of deep traps are involved in the process [188] which could explain the I-V behavior in regions II and III. Eventually, at high enough bias, the diffusion of carriers across the p-n junction barrier becomes dominant causing the exponential increase in region IV. Reverse bias current Considering the specific detection application in mind, understanding the mechanisms ruling the reverse bias current is arguably even more important. When we apply a reverse bias to the detector, the current stays below the noise level of the measurement system only for some hundreds of mV and rapidly starts increasing to higher values. In the I-V curve shown in figure 6.7 the current at -1 V is already equal to 118 pA and at -4 V it increased to 18.9 nA. Although some variation in the magnitude of the reverse bias current exists between samples, in general the observation of the bias dependence is very reproducible as can be seen from the several I-V curves shown in figure 6.9. First of all, it is important to note that we certified that the leakage does not come from the substrate but that it occurs as a consequence of the physical mechanisms defining the transport of carriers through the device. Secondly, in ideal diodes the leakage current in reverse bias is defined by the saturation current or, if included, by the shunt resistance. The former is independent on bias while the latter yields 87 the sometimes two order of magnitude larger reverse bias leakage currents, generally samples with Ni/Au contacts are preferred. 6.2.2 Photoconductivity measurements In this section, the photoconductivity data measured with the p-n junction microwires will be presented. Figure 6.16 shows the obtained I-V curves in the dark and when illuminating a representative detector with the UV LED, in linear and logarithmic scale. We can immediately observe the existence of a significant photocurrent that is present in both reverse and forward bias. This specific sample yields a short-circuit current of 10.2 nA while the open-circuit voltage is 1.0 V. Regarding the Voc, the average value obtained when considering 38 samples is 0.95 ±0.13 V whereas Isc typically has a value between 1 nA and 20 nA. I note that it is important to keep in mind that these values depend on the irradiance as was shown when carrying out the photocurrent simulations in chapter 2. In both cases the values were obtained using the LED at the highest input current (0.7 mA) but some variation in the incident optical power can occur due to unavoidable differences in the spatial alignment of the LED with respect to the measured microwire. Nevertheless, these differences are not significant enough to explain the large variations in the Isc of the measured sensors. To account for this we also need to consider the active area of the device. In this case, the active area was estimated to be equal to the region of the surface of the core-shell region, not covered by the metal contact. As such, we can approximate the active area as a rectangle where the width is equal to the diameter of the shell and the length equal to the uncovered p-GaN region, as shown in the SEM image in figure 6.17 a). Figure 6.17 shows the measured Isc for several samples, normalized by the irradiance, as a function of the active area. The fact that we obtain a linear correlation between the calculated area and the Isc indicates that this is a good estimation. Although the photocurrent generated by the p-n junction is evident, the non-linear growth of the photocurrent at larger reverse biases and the existence of a signal when applying a forward bias are clear indications that deviations from the ideal case also exist in this case. So, in accordance with the dark current, the photocurrent is also defined by a complex set of conduction and photocurrent −4 −3 −2 −1 0 1 2 3 4 Bias (V) 10 −13 10 −12 10 −11 10 −10 10 −9 10 −8 10 −7 10 −6 I (A) Dark UV (a) −4 −3 −2 −1 0 1 2 3 4 Bias (V) −50 0 50 100 150 200 I (nA) Dark UV (b) Figure 6.16: I-V curves measured in the dark and while illuminating a representative p-n junction microwire sample with the UV LED in logarithmic scale (a) and linear scale (b). 94 2 µm Active area (a) 5.0 7.5 10.0 12.5 15.0 17.5 20.0 Active area ( μ m 2 ) 4 6 8 10 I sc /Irradiance (A/W) (b) Figure 6.17: a) SEM image of a p-n junction sample indicating the region used to estimate the active area of the detector; b) Short-circuit current of different samples normalized to the irradiance as a function of the active area. The linear dependence of Isc on the active area shows that the estimation of the latter is in good agreement with the real value. generation mechanisms. To facilitate the qualitative and quantitative analysis we can differentiate four regions: the photovoltaic regime (0 V < V < Voc), low reverse bias (-1 V < V < 0 V), high reverse bias (V < -2 V) and forward bias (V > Voc). In the photovoltaic regime, the photodetector essentially acts as a solar-cell. This regime is very interesting in terms of energy efficiency because when operating in this bias range, we are not injecting any power into the system and the detector works in self-powered mode. From the analysis carried out before, we saw that the shape of the curve is dependent on the parameters defined in eq. 2.27, however, not taken into account was the influence of the contacts. Figure 6.18 shows the I-V curves for three different samples, one sample of which the contacts were not annealed after concluding the fabrication process and two samples that were annealed at 500 ◦C in air for 60 seconds but that show a different electrical characteristic in dark conditions. First of all, we see that the I-V curve of the detector that was not annealed looks very different from the annealed samples. Not only are Isc and Voc lower, but we also observe that the photocurrent decreases continuously from 0 V to Voc. Between the samples that were annealed we also see a significant difference, namely the existence/absence of a kink or ”s” shape 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Bias (V) −3 −2 −1 0 1 2 I (nA) Dark UV (a) (b) (c) Figure 6.18: Dark and photocurrent for the bias range in which the p-n junction microwire detector works in the photovoltaic regime. (a) Measurement made on a sample that was not annealed while (b) and (c) show measurements made on samples that were annealed. The insets show the dark I-V curves measured over the entire bias range. In the right graph we can see that there is no trace of an ”s” shape in the photocurrent I-V curve whereas we can clearly see the kink in the center graph. 95 in the photocurrent I-V curve. I note that the processing of both samples was the same. The presence of such a kink in the photocurrent was observed previously and was attributed to different origins. In InGaN MQW solar cells the ”s” shape was attributed to the suppression of the carrier collection due to the existence of a large potential barrier within the InGaN MQWs [199]. There are no MQWs present in our microwires however, so this reasoning does not apply to our case. Nevertheless, the phenomenon can also be explained by the presence of a potential barrier that prevents the extraction of chargecarriers. In organic solar-cells, for instance, it was observed that the shape of the photocurrent I-V curve is strongly dependent on the metal used for the electrodes and the interface between the material and the cathode [200]. In fact, simulations of the photocurrent transport mechanisms show that a reduction of the conductivity close to one of the electrodes is sufficient to induce an ”s” shape in the photocurrent I-V curve [201]. Messanvi et al. extensively studied the photovoltaic properties of single core-shell GaN/InGaN wires. The structures that were used are very similar to the core-shell microwires used for the radiation detectors but for the presence of a continuous layer of InGaN or InGaN MQWs grown between the p-GaN shell and the n-GaN core [202]. They attributed the presence of the ”s” shape to the Schottky barrier that is formed by the junction bewteen the Ni/Au metal contact and the p-GaN shell. Taking into account that the electric field of the Schottky diode is in the opposite direction of the electric field induced by the p-n junction, we essentially have two competing photocurrent contributions. The rectifying contact generates a positive photocurrent in this regime whereas the p-n junction generates a negative photocurrent. After optimizing the Ni/Au contact through annealing, Messanvi et al. observed that the kink in the photocurrent I-V curve vanished, which was accompanied by an improvement of the short-circuit current, the open-circuit voltage as well as other solar device parameters. The same occurs in our devices, whereas before annealing the contribution of the Schottky barrier is large, after annealing it becomes much smaller which is reflected in the enhancement of the observed photocurrent. We can thus associate the ”s” shape with an imperfect ohmic contact on the p-GaN shell. The fact that we do not entirely erase the ”s” shape in all samples means that the annealing does not always modifies the contact enough to completely undo the influence of the Schottky barrier at the interface, as we already saw before when analyzing the dark I-V curves. It is therefore interesting to see the correlation between the span of the kink and the onset of the identified region IV in the dark I-V curves, shown in the inset of figures 6.18 b) and c). Generally, for detectors where the exponential increase starts at lower voltages, the ”s” shape is barely visible whereas for detectors were region IV occurs at V > 4V it is clearly present. This shows that the metal-semiconductor contact does in fact have a strong influence on the dark and photocurrent mechanisms when the applied voltage is positive. Detectors where the p-GaN contact has a smaller Schottky barrier a small or no kink is visible in the photocurrent I-V curve and the onset of the exponential growth of the dark current occurs between 3 V and 4 V. On the other hand, for detectors that have a more significant Schottky barrier, the kink becomes larger and the exponential growth of the current is normally not present for V < 4V. The value of Voc was already mentioned before, but it is worth to comment it and to compare it with values reported in previous literature. Firstly, taking into account the wide bandgap of GaN, in the ideal case we would expect a value located between 3 V and 3.4 V [203]. Nevertheless, there are many 96 factors that may reduce the open-circuit voltage. We already saw that potential barriers at the contact interface impact the value of Voc but recombination may also plays an important role in reducing Voc. Using planar GaN based structures Voc values between 1.5 V and 2.1 V have been achieved [204, 205] which are high compared to the results we obtained. On the other hand, through comparison with results obtained using other microwire detectors based on GaN p-n junctions, the average Voc we obtained is on the high-end [202, 124, 85]. This latter observation is possibly related with the absence of InGaN MQWs in our structures as it has been shown that these can introduce defects which are directly related to the lower Voc values [202]. One important remark that has to be made when comparing Voc is that it depends on the light power intensity. Since this parameter is not always provided in the literature it may also be the cause different magnitudes are obtained. Looking now at the case where we apply a low reverse bias voltage to the detector, this is the region where we obtain the largest photo-to-dark current ratio, mainly because of the low dark current values. When using the UV LED at maximum input current, the ratio is normally above 104. The photocurrent in this bias range shows a small linear variation, from 0 V to -1 V it increases by approximately 2 nA. This increase can be seen as a consequence of the widening of the depletion region. However, when the reverse bias voltage goes beyond -1 V, the growth of the photocurrent becomes non-linear. The bias at which this change occurs varies slightly from sample to sample but is usually the same as the bias at which the dark leakage current ceases to be negligible. I note that the rate of increase of the photocurrent is larger than for the dark current, as can be seen from the linear graph in figure 6.16, so even when subtracting the dark current, Ipc maintains its non-linear profile. This abnormal photocurrent increase cannot be explained by the p-n junction diode theory and instead we must look at other photocurrent generation mechanisms. As we saw before, the potential barrier located at the interface between the semiconductor and the metal creates a potential barrier that affects the photocurrent generation. However, when the p-n junction is reversely biased, the Schottky diodes are in forward bias, hence, it is not plausible that the excess photocurrent exists due to the contacts. The more likely explanation is that when increasing the bias, charge carriers that are created in the neutral regions become increasingly relevant and a photoconductive mechanism is induced. Similarly, when looking at the photocurrent when applying a forward bias to the detector we also see a significant photocurrent that is beyond the p-n junction diode model presented in eq. 2.27. When the bias exceeds Voc the current rapidly surpasses the dark current whereas according to the model of the p-n junction diode the photocurrent should be approximately equal to the dark current. In this case, the Schottky barriers are reversely biased so we need to take the contribution from the contacts into account. At low positive bias, we already saw that in some cases this contribution leads to the introduction of a kink in photocurrent I-V curve but in general the relative contribution of the p-n junction is stronger. As the applied bias grows beyond Voc the SCR width of the Schottky barriers becomes large enough to overcome the photocurrent generated by the p-n junction and the signal becomes positive. Additionally, the photoconductive mechanism also starts to play a role. The interplay between the photoconductive and photodiode mechanisms was previously observed in UV photodiodes fabricated with similar microwires [85] and more thoroughly studied by Jacopin et al. [125]. They fabricated photodetec97 tors based on an ensemble of axial p-i-n junction GaN nanowires and saw that for |V|>2V, the device operated in photoconductive mode, presenting a large gain and high values of responsivity. When studying the n-type microwires we saw that non-linear photocurrents and slow photocurrent transients were also characteristic of the photoconductivity in our samples. For this reason, it will be interesting to see how the different applied biases will affect these parameters in the p-n junction microwires. Linearity of the photocurrent and responsivity In order to study the linearity of the photocurrent with irradiance of our detectors, we varied the input current of the UV LED and measured the resulting photocurrent I-V curves. To guarantee that the irradiance of the LED scales linearly with the input current as well as to calibrate the irradiance, a commercial Si PIN photodiode was used. Furthermore, to obtain irradiance values over a larger range we used both a condenser lens with and without diffuser. Figure 6.19 a) shows the I-V curves that were measured while illuminating the sensor with UV light with different irradiance. Naturally, when the incident light power decreases the photocurrent decreases as well. To study the linearity we can plot the photocurrent as a function of the optical power for different applied biases as shown in figure 6.19 Decreasing irradiance (a) −4.0 −3.5 −3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0 0.5 Bias (V) 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 β Sample: 7112 Sample: 7113 Sample: 7321 Sample: 7323 Sample: 7416 (c) (b) Figure 6.19: Dark and photocurrent I-V curves when illuminating the p-n junction microwire sample with different irradiance (a). By extracting the value of the photocurrent at a certain bias we can also plot the photocurrent as a function of the optical power, here the markers represent the experimental data and the line the fit to the power law function (data is shown in log-log scale and in linear scale in the top and bottom graph respectively) (b). Extracted βparameters obtained by fitting the photocurrent using the power law function (c). 98 b). At the same time, we can fit the data using eq. 2.22 and the obtained values for the βparameter are indicated in the graph. It is very interesting to see that the photocurrent at low reverse bias scales nearly linearly with the optical power whereas when the bias increases, the curves become sublinear. To get a better idea of the dependence of the βparameter on the applied bias, figure 6.19 c) plots the obtained βvalues between -4 V and 0.5 V for several samples. The results confirm that the βparameter is only close to unity in the low reverse bias regime whereas for increasingly negative voltages βgradually decreases. In some cases the curves become supra-linear with β > 1when the detector operates in the photovoltaic regime. This is unexpected and to better understand it, we can compare fits performed only to the data measured at low irradiance (measured with the diffuser lens), to the data at high irradiance (measured with normal lens) and to the entire data set. The results are shown in figure 6.20 and here we see that at lower irradiance we have β < 1and at higher irradiance we have β > 1. This may be caused by experimental errors since we obtain the irradiance values indirectly and the alignment may not always be exactly the same when measuring the photocurrent with both lenses. Therefore we can conclude that, within the experimental uncertainties, the detectors are linear in the low reverse bias regime. The lower βvalues at higher reverse voltages essentially mean that, at a fixed bias, the increase rate of the photocurrent with optical power becomes smaller for higher irradiance. In other words, we can also say that the responsivity is higher for lower values of irradiance than for higher values of irradiance. The responsivity can be calculated using eq. 2.19 and to visualize the change in responsivity as a function of the irradiance, the evolution of this parameter is shown together with the photocurrent at different bias values in figure 6.21. We can see that the responsivity is approximately constant between -1 V and 0.5 V but that, when we increase the reverse bias voltage, it shows higher values for low irradiance in accordance with the photocurrent behavior. However, we can also see that the values at higher reverse bias are much larger than for voltages close to zero bias. Furthermore, figure 6.22 shows the responsivity as a function of the applied bias for different values of irradiance. First of all, we can once again observe a clear distinction between the constant and variable responsivity regime. If we compare the values of Rat lowest and highest irradiance for -4 V and 0 V we get 10.21 A/W versus 0.030 A/W and 0.162 A/W versus 0.029 A/W, respectively. Although the increase may be partially caused by the widening of the SCR, the high magnitude of the values is clear evidence of the activation of a gain mechanism at higher reverse bias. In fact, for an EQE of 100% and a unit gain, the responsivity 0.01 0.02 0.03 0.04 0.05 0.06 0.07 Irradiance (W/cm 2 ) 0.05 0.10 0.15 0.20 0.25 0.30 0.35 Photocurrent (nA) β = 0.97 β = 0.96 β = 0.89 Bias = -0.5 V Bias = 0.0 V Bias = 0.5 V (a) 0.5 1.0 1.5 2.0 2.5 3.0 3.5 Irradiance (W/cm 2 ) 2 4 6 8 10 12 14 Photocurrent (nA) β = 1.08 β = 1.07 β = 1.07 Bias = -0.5 V Bias = 0.0 V Bias = 0.5 V (b) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 Irradiance (W/cm 2 ) 0 2 4 6 8 10 12 14 Photocurrent (nA) β = 1.09 β = 1.09 β = 1.09 Bias = -0.5 V Bias = 0.0 V Bias = 0.5 V (c) Figure 6.20: Photocurrent fits at -0.5 V, 0 V and 0.5 V when only using the data points measured when using the diffuser lens (a), when using the normal lens (b) and when using all data points. 99 10 −1 10 0 Irradiance (W/cm 2 ) 0 2 4 6 8 10 12 14 Photocurrent (nA) Bias = 0.5V 0.000 0.005 0.010 0.015 0.020 0.025 0.030 Responsivity (A/W) (a) 10 −1 10 0 Irradiance (W/cm 2 ) 0 2 4 6 8 10 12 14 16 Photocurrent (nA) Bias = 0V 0.00 0.01 0.02 0.03 0.04 Responsivity (A/W) (b) 10 −1 10 0 Irradiance (W/cm 2 ) 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 Photocurrent (nA) Bias = -1V 0.00 0.02 0.04 0.06 0.08 Responsivity (A/W) (c) 10 −1 10 0 Irradiance (W/cm 2 ) 5 10 15 20 25 Photocurrent (nA) Bias = -2V 0.0 0.2 0.4 0.6 0.8 1.0 Responsivity (A/W) (d) 10 −1 10 0 Irradiance (W/cm 2 ) 10 15 20 25 30 35 40 45 50 Photocurrent (nA) Bias = -3V 0 1 2 3 4 Responsivity (A/W) (e) 10 −1 10 0 Irradiance (W/cm 2 ) 30 40 50 60 70 80 90 Photocurrent (nA) Bias = -4V 0 2 4 6 8 10 12 Responsivity (A/W) (f) Figure 6.21: Photocurrent and responsivity as a function of the irradiance for a representative p-n junction microwire detector while applying different voltages to the sample. according to eq. 2.21 is equal to 0.29 A/W for light with a wavelength equal to 364 nm. Hence, the large values of Rat high reverse bias can only exist when considering a gain superior to unity. The results presented so far are a clear manifestation of a photoconductive mechanism contributing to the photocurrent for high applied biases and are in good agreement with the results obtained using the n-type microwires and with previous literature [125, 85]. In our case, the photoconductivity processes start to affect the photocurrent for V < −1V, more or less at the same voltage where the dependence of the photocurrent on bias becomes non-linear as shown in figure 6.16 and where the leakage current mechanisms start to dominate. Finally, I note that no data was shown for the photocurrents measured at 100 −4.0 −3.5 −3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0 0.5 Bias (V) 0 2 4 6 8 10 Responsivity (A/W) 0.02 W/cm 2 0.05 W/cm 2 0.09 W/cm 2 1.13 W/cm 2 2.69 W/cm 2 3.67 W/cm 2 (a) −4.0 −3.5 −3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0 0.5 Bias (V) 10 −1 10 0 10 1 Responsivity (A/W) 0.02 W/cm 2 0.05 W/cm 2 0.09 W/cm 2 1.13 W/cm 2 2.69 W/cm 2 3.67 W/cm 2 (b) Figure 6.22: Responsivity of a representative p-n junction microwire detector as a function of the bias for different values of irradiance in linear (a) and logarithmic (b) scale. We can see that between 0.5 V and -0.75 V the responsivity does not vary with the irradiance but that at higher reverse bias, the responsivity increases at a higher rate for lower values of irradiance. higher forward voltages (V > 2V) due to the associated slow transients, which will be shown next and that make measurement of the photocurrent as a function of the irradiance difficult. Transient I-t measurements We saw that gain and non-linear photocurrents are often related with slow photocurrent decay times. Therefore, taking into account the results shown up to this point we expect that the decay times of the photocurrents measured with the p-n junction wires will also have a bias dependence. The transient I-t measurements were carried out by fixing a bias and measuring the current induced when turning the UV LED ON and OFF after specific time intervals. In this case, the sampling rate of the system was set to its minimum value, around 24 ms, and in all measurements the input current of the LED was set to its highest value. Figure 6.23 shows the obtained results for several applied biases and we can easily confirm that indeed the response times strongly depend on the applied bias. Looking first at the curve obtained in short-circuit mode, we can see that in this case the response is immediate. Both the rise and decay time are faster than the sampling rate of the measurement equipment and no persistent photocurrent is observed. At a bias of -2 V this is no longer the case. Although the rise time is equally fast, the photocurrent now presents a small persistent decay tail after turning the LED off. If we apply the definition of decay time as the time it takes to go from 90% to 10% of the maximum photocurrent value, however, we would still obtain τd<24 ms. At -4 V this is no longer the case as now the slow transient dominates the decay response of the detector following the stronger influence of the photoconductive contribution. When we apply a positive voltage, the response times increase dramatically. Taking into account that the Schottky diode at the p-GaN contact is now in reverse bias this in agreement with the observation made when studying the n-type microwires and can thus be seen as a confirmation of the influence of this potential barrier on the photocurrent generation. This behavior was also observed by Zhang et al. when using detectors based on core-shell p-n junction wires containg InGaN MQWs [85] and by Spies et al. when using axial p-n junction nanowires [87]. 101 (a) (b) (c) (d) Figure 6.23: Transient photocurrent measured using a p-n junction microwire detector, shown for different applied biases. In (a) we can see that in self-powered mode the rise and decay times of the sensor are faster than the sampling rate of the measurement system; When we increase the reverse voltage a small persistent component appears for -2 V that becomes stronger at -4 V. In forward bias the response time increases dramatically as a consequence of the reverse biased Schottky contact and stronger relative influence of the photoconductive mechanism. Spectral response The final photocurrent characterization consist on studying the spectral response of the devices. The spectral signature of the detectors was obtained through photocurrent spectroscopy. In this case, the measurements were carried out using a Xe lamp and light dispersion was achieved using a Spex Model 1681 monochromator with a grating of 305 lines/mm to split up the light into different wavelengths. The low responsivity of the sensors, especially at zero bias, required the use of additional amplification of the current. As such, the light was chopped at a frequency of ∼60 Hz and the output signal was demodulated using a lock-in amplifier. A schematic of the experimental setup is shown in figure 5.13 in chapter 5. Figure 6.24 shows the obtained data measured without applying a bias for two different samples that represent, within the experimental error, the behavior of all measured samples. Also shown are the spectra obtained using the same sample but with different applied voltages. We can see that, besides a slight variation in the intensity of the signal, no difference is visible when applying a reverse bias of -2 V or a zero bias. The maximum of the photocurrent occurs at approximately the bandgap energy of GaN and it decreases for both higher and lower energies. Regarding the decrease at higher energies, on the one hand this can be explained by the larger absorption coefficient of GaN 102 2.82.93.03.13.23.33.43.53.6 Energy (eV) 0.0 0.2 0.4 0.6 0.8 1.0 Normalized I pc (a.u.) Sample 30920 Sample 30936 340 360 380 400 420 440 Wavelength (nm) (a) 2.83.03.23.43.63.8 Energy (eV) I pc (a.u.) Bias = 0 V Bias = -2 V Urbach fit 330 360 390 420 450 Wavelength (nm) U E = 69.1 meV U E = 70.8 meV (b) Figure 6.24: Spectral response as a function of the photon energy at zero bias for two distinct p-n junction microwire samples (a) and two different applied biases (b). Also represented in the right graph is the Urbach tail fit. at higher energies [206], which leads to a lower absorption depth and a larger generation close to the surface where carriers can be trapped by surface states [85]. On the other hand, it also reflects the low output efficiency of the Xe lamp in this wavelength range. Although the data is corrected using the lamp spectra, at 320 nm the output of the lamp is already close to zero. For lower energies we first observe a steep decline of the signal up to an energy of approximately 3.1 eV, followed by a slower decrease up to an energy of 3.0 eV. The generation of photocurrent for photons with energies below the bandgap was already commented when the spectroscopy data for the n-type wires was shown and the overall behavior is similar. Taking into account the unique geometry of the microwire however, it can be a consequence of the Franz-Keldysh effect caused by lateral band bending in nanowires [178] but there can also exist some additional effects due to the presence of the p-GaN shell. We already saw that a subbandgap response can be an indication of the presence of defect states close to the edges of the bandgap [207] and, in this case, a possible source could be Mg-related deep levels present in the p-GaN shell [208]. To further analyze the signal decrease below the bandgap we can use Urbach’s rule, defined by equation 2.24 to obtain the Urbach energy EU[103]. I note that to achieve this we assume that the absorption is proportional to the photocurrent. The obtained values for EUat 0 V and at -2 V are 69 meV and 70 meV, respectively. Compared to previously reported Urbach energies for GaN thin films and nanowires, these values are rather high. Values between 30 and 50 meV were reported for MBE grown nanowires [209, 125] while Monroy et al. obtained values close to 20 meV for n-GaN Schottky photodiodes but close to 140 meV for p-GaN Schottky photodiodes [208]. The value is also too large to be solely explained by the Franz-Keldysh effect [178], hence the data seems to indicate that defects and energy levels in the forbidden gap are playing a role in our devices. 6.3 Chapter conclusions In this chapter we have discussed the electrical and electro-optical properties of the detectors based on GaN n-type and core-shell p-n junction microwires. We demonstrate that the detectors based on core-shell p-n junctions have a good signal-to-noise ratio with fast response times and good linearity, 103 surements of radiation hard. Here it is important to note however, that the irradiation fluence that is used in the measurements is large so, as will be shown later, the devices actually present a relatively good radiation resistance. 7.1.2 Ionoconductivity measurements in p-n junction microwires Taking into account the presence of the SCR, the p-n junction microwires have some distinct advantages over the n-type microwires and are naturally more interesting for detection applications. The experimental setup used to irradiate the p-n junction wires and measure the signals is similar to the one used to measure the n-type wires. Figure 7.4 shows the transient I-t curve for a run that consisted of 10 complete scans using a scan area equal to 106×106 µm2without applying any bias to the electrodes. Once again, we only observe ten peaks and the same explanation as before can be provided, namely, that the sampling rate of the measurement equipment is not fast enough to measure the current induced each time the beam hits the wire but instead integrates the signal resulting from all hits for one line in the scan. However, in this case we see a significant improvement of the signal-to-noise ratio as well as fast transient responses. Also shown in figure 7.4 are the PIXE maps obtained for each relevant element present in the detector, namely the gold, nickel and titanium present in the contact, and the Ga from the microwire. 0 10 20 30 40 50 60 70 80 90 Time (s) 0 1 2 3 4 5 I (nA) (a) 48 50 52 54 56 58 60 Time (s) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 I (nA) (b) Au signal Ga signal Ti signal Ni signal (c) Figure 7.4: Transient I-t curves measured while scanning a p-n junction microwire detector using a an area of 106x106 µm2(a). The same data but zoomed in on two of the peaks (b); PIXE maps obtained while irradiating the detector using the scan size (c). We can see the contribution from the different materials that compose the wire and the metal contacts. 110 0 20 40 60 80 Time (s) 0 1 2 3 4 5 6 I (nA) Scan size: 264x264 μ m 2 Scan size: 106x106 μ m 2 Scan size: 53x53 μ m 2 Figure 7.5: Transient I-t curves measured while irradiating a p-n junction microwire detector when using different scan sizes. Ascan (µm2) Average Ic(nA) Ascan×Ic,av 264 0.45 118.8 106 1.07 113.4 53 2.10 111.3 Table 7.1: Average ionocurrent values of the measured peaks as a function of the scan area and respective product between the scan area and ionocurrent values. If we compare the magnitude of the peaks in figure 7.4 (a) we observe some oscillations, which can be explained by variations in the beam current, variation due to the asynchronous nature of integration time and due to loss of signal as a consequence of the induced irradiation damage (this will be discussed in more detail in chapter 8). Regarding the first point, the lack of stability of the beam current is a significant drawback of the experimental setup. It is related with the fact that we require the beam current to be as low as possible in order to minimize the irradiation fluence. From experience, if initially set to 50 pA, it can oscillate by more or less 30 pA both to higher and lower values. Considering the scan area, this translates into a variation in fluence per scan between 0.8×1013 protons/cm2and 3×1013 protons/cm2. Regarding the second point, since the integrated signal includes points where the beam is irradiating the substrate and no charges are created in the wire, the longer the beam is irradiating the substrate, the lower the integrated signal will be. This is well demonstrated when looking at the dependence of the magnitude of the signal on the scan size that is used, illustrated in figure 7.5. On average, a good agreement between the scan size and the average magnitude of the peaks is obtained and the product between the lateral size of the scan area and the average ionocurrent is similar, as shown in table 7.1. This can be seen as an indication that the sensitivity of the detector to the number of incident protons is good. As shown in figure 7.4 (b) we can also see some variations of the current within each peak. Once again, this is partially caused by the varying beam current and by the fact that the beam spends a significant amount of time irradiating the substrate, even when the wire is located in its path. However, a physically more interesting cause for the variation of the current within each peak, is that we also do not necessarily expect a homogeneous signal generation across the entire length of the wire. Since 111 carriers generated far away from the contacts need to drift a longer distance in order to reach the electrodes, they may recombine before completing their journey and the resulting current will be smaller. Although the resolution of the data presented in figures 7.4 (a) and (b) is not good enough to show this, ionoconductivity measurements using smaller scan areas and IBIC measurements presented later will give a better idea about this effect. To study the signal generation in the different regions of the wire we again take advantage of the possibility to define specific irradiation scan areas through the computer software. Like this, we can reduce the time the beam spends irradiating the substrate as well as compare the signal that is generated while irradiating the p-GaN or n-GaN half of the wire independently. An important note here is that a reduction of the region-of-interest does not translate into an increase of the fluence per scan area because the number of pixels that are irradiated is also reduced. In other words, the ratio of Npixel/Ascan is constant when modifying the scan area using the software. Figure 7.6 shows two I-V curves measured in-situ, during proton irradiation and in the dark. The top graph shows the curve obtained when irradiating the p-n junction and the bottom graph the curve obtained when irradiating the n-GaN base. PIXE maps marking the irradiated areas are also shown. It is immediately clear that when we irradiate the p-n junction, the signal is significantly larger when compared to the situation where we irradiate the n-GaN section. This is an indication that the active region does not extend all the way to the n-GaN contact. (a) (b) Figure 7.6: (a) Static I-V curves measured when irradiating specific areas of a p-n junction wire, namely the core-shell region (top graph) and the neutral n-GaN region (bottom). The areas that were irradiated are shown in the insets; (b) Transient I-t curves measured while irradiating the core-shell region indicated in the inset of the top graph if (a), and while applying different voltages to the detector, in logarithmic (top graph) and linear (bottom graph) scale. 112 Additionally, the low ionocurrent signal measured even when applying a large forward bias indicates that the contribution from an eventual Schottky barrier at the interface between the n-GaN and the metal contact is, as expected, negligible. Regarding the ionocurrent signal measured when irradiating the p-n junction we see that, relative to the dark level, it increases by a factor of 102. If we compare this ionoto-dark current ratio with the photo-to-dark current ratio presented in figure 6.16 in chapter 6, which has values between 103and 104, we conclude that it is smaller. The main explanation for this is the signal quenching caused by irradiation damage. When the I-V curves presented in figure 7.6 (a) were measured, they were already irradiated with a fluence close to 1×1015 protons/cm2which, as will be shown later, is already a fluence at which we observe a significant degradation of the signal. Finally, I also note here that the I-V curves presented in figure 7.6 (a), namely the lower reverse bias leakage current, are notably different from those shown when presenting the photocurrent I-V curve of the p-n junction microwire detector (figure 6.16 in chapter 6. This is a consequence of the different contact scheme and annealing conditions used in these samples (see chapter 4 and appendix D), more specifically, in this experiment we used devices with a Cr/Au contact on the p-GaN extremity that were annealed in N2at 500◦C. To take a closer look at the dependence on bias and at the transient characteristics of the ionocurrent, we performed transient I–t measurements at a fixed bias. For each measurement shown in figure 7.6 (b), the fluence of irradiation is ∼2×1015 protons/cm2. Again, we attribute the variation in the ionocurrent to fluctuations in the beam current. Regarding the dependence of the ionocurrent on bias, we can see that the ionocurrent increases by a factor of approximately 2 from 0 V to -2 V and a factor of 3 from 0 V to -4 V, which we attribute to the increase in the depletion region width of the p–n junction. Since we are using detectors with Cr/Au contacts deposited on the p-GaN we expect a lower contribution from ionoconductive mechanisms, as is shown in appendix D. Indeed, the rise and fall times have values of the order of the sampling rate and do not increase when applying a reverse bias, which is also an indication that conductive effects are limited. Finally, although introduced damage in the device already Figure 7.7: Transient I-t curves measured while irradiating different sections of a p-n junction microwire detector. The areas that were irradiated are indicated in the PIXE map and the right and left contacts corresponds to the n-GaN contact and p-GaN contacts, respectively. In all cases we applied a bias of 0 V. We can see that the ionocurrent signal is high when irradiating the core-shell region and gradually decreases as we move towards the neutral n-GaN region. 113 Figure 7.8: Transient I-t curve measured while irradiating a p-n junction microwire detector using a static beam. The inset shows a PIXE map measured before the irradiation with the static beam, where the white dots represent the specific locations that were irradiated, including five points on top of the coreshell region and four points far away from the detector. We can see a very large iono-to-dark current ratio for the first static irradiation. In the subsequent periods the beam moves to the core-shell region, the signal decreases due to the created irradiation damage. led to a strong decrease of the signal-to-noise ratio of the detector, we do not observe any decay of signal in the measurements presented in figure 7.6 (b). This indicates that despite the lower signal when compared with a non-irradiated sample and the high fluences, the detectors are still able to yield a remarkably stable signal. To take full advantage of the precision of the microbeam, measurements along the entire length of the wire were also performed. Figure 7.7 shows the transient signal measured when irradiating the different sections of the detector shown in the corresponding PIXE map. Once again, we observe that the signal decreases as we move farther away from the p-GaN contact, in accordance to what was mentioned above. To ensure that this is not an effect of the irradiation damage, the experiment was also done in the opposite direction and the same results were obtained. So far, the measurements were all performed with the ion beam scanning the full area or a defined region-of-interest. However, the possibility to use a static beam also exists. Figure 7.8 shows signals that were obtained while directing the beam to the positions indicated in the corresponding PIXE map. We observe that we obtain a signal with a very high magnitude, a consequence of the fact that the beam is now continuously irradiating the same position. A downside of this is of course that we reach very high fluences in a very short time period. Taking into account eq. 5.1 for a beam area of 2x3 µm2and a beam current of 50 pA we would only need five seconds to reach a fluence higher than 1x1016 protons/cm2and we would rapidly destroy our detector. To partially, overcome this we can perform ionoconductivity map measurements. This experiment consists in drawing a predefined region that consist of a certain number of points. Through an electrical signal, the beam is directed to each one of the points and irradiates it for a short time period. Between the irradiation of each point the beam is directed to a spot outside of the area to allow the detected signal to go down. Simultaneously, the electrical signal is measured in transient mode. Figure 7.9 shows the PIXE map with the irradiation points in green as well as the raw data obtained from the transient signal measured at 0 V obtained during the experiment. In this case, we defined an area consisting of 15 by 15 points and defined an irradiation time of 0.4 seconds per 114 (a) 0.0 2.5 5.0 7.5 10.0 12.5 x (pixel) 0 2 4 6 8 10 12 14 y (pixel) 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 Ionocurrent (nA) (b) 0.0 2.5 5.0 7.5 10.0 12.5 x (pixel) 0 2 4 6 8 10 12 14 y (pixel) 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 Ionocurrent (nA) (c) 0.0 2.5 5.0 7.5 10.0 12.5 x (pixel) 0 2 4 6 8 10 12 14 y (pixel) 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 Ionocurrent (nA) (d) Figure 7.9: (a) PIXE map obtained during the ionocurrent map experiments, including the points of the irradiated region which are marked in green. The right graph shows the raw transient data measured during the experiment. The detected signal is shown in orange and the voltage signal that controls the beam shutter is shown in blue. In the inset the full time range is shown. (b-d) Current maps showing the measured data. Each pixel in the map corresponds to a point in the PIXE map. The applied bias is equal to 0 V (b), -1 V (c) and -2 V (d). point. In the transient I-t graph, the beam is irradiating one of the points whenever the beam control voltage is high. Regarding the measured signal, we can see that it is approximately zero except at some time instants where we can observe higher peaks. By associating the transient profile of the beam control voltage, the corresponding measured signal and the points in the defined raster we can obtain the maps that are also shown in figure 7.9. The experiment was performed at different biases, namely at 0 V, -1 V and -2 V. Once again, we can see that the signal magnitude increases with bias but we can also see that the region where current generation occurs is concentrated very close to the p-contact and that it does not extend very far into the wire. This region extends somewhat when increasing the bias but not much. The points at which we detect a signal distinguishable from the background noise are also marked on the PIXE map by means of the white rectangle. First of all it is worth to note that these measurements were carried out using the microwires with the low quality p-GaN shell (samples T2360) which are characterized by a relatively small active area. Nevertheless, there are other possible causes for the short extension of the SCR that need to be considered. More specifically, it is likely that the amount of collected charge carriers decreases as a function of the distance of the beam to the p-GaN contact. This was observed when carrying out EBIC measurements using similar microwires and attributed to charge recombination in the p-GaN shell as a consequence of the large resistance 115 [195]. This effect can be enhanced by the created irradiation damage because defects will increase the resistivity and recombination that occurs in the p-GaN shell. Since the IBIC measurements that will be shown in the next section are also relevant in the context of the analysis of this data, I refer the reader to the next section for a more complete description of the mechanisms in play. 7.2 IBIC measurements In the following section the IBIC measurement off the p-n junction microwire detectors will be presented. The experiments were carried out through a collaboration with the Laboratory for Ion Beam Interactions of the Ruder Boˇ skovi´ c Insitute in the framework of the Radiate project. The IBIC measurements were carried out using 750 keV and 1 MeV Si ion beams, where the Si ions had a +2 charge. The samples that were studied using this setup were p-n microwire detectors with Ni/Au contacts deposited on the p-GaN extremity. The main parameter that is extracted from the IBIC data is the CCE. The value of this parameter is obtained by dividing the collected charge by the deposited charge, which in this case is proportional to the Si ion energy. For a full description of the experimental setup please see chapter 5. (a) (b) Bias = -1 V (c) (d) (e) (f) Figure 7.10: SEM image and IBIC maps of a p-n junction microwire sample irradiated with 750 keV Si ions while applying different voltages to the p-GaN terminal of the detector. In the SEM image, the relevant dimensions are indicated and the p-GaN contact corresponds to the bottom electrode. In the IBIC maps we can see that, when the applied reverse bias increases, the measured CCE also increases. We also measure a non-zero CCE when applying a positive voltage, indicating that we are not completely cancelling out the SCR of the p-n junction. 116 7.2.1 IBIC maps Figure 7.10 shows several IBIC maps obtained while scanning the same microwire using a scan area of 27.65 by 27.65 µm2. The different maps show the results obtained when using different applied biases to the devices where the colored points correspond to the measured values of the CCE as a function of the beam position. The rectangular shape of the microwire is easily distinguishable and no signal is measured when the beam is irradiating the substrate of the device. I note that the detectors that were characterized using IBIC are all fabricated on sapphire substrates. The choice of substrate is important in this case, if Si/SiO2substrates with a 300 nm SiO2layer are used, the incident ions also create charges in the Si substrate. These charges are measured by the preamplifier and shadow the signal induced by the microwire. Since sapphire is insulating this does not happen for detectors fabricated on sapphire substrates. Shown as well is the SEM image of the detector. In this case, the p-GaN contact is deposited on the extremity of the microwire and corresponds to the bottom contact in the image. In the IBIC maps we can distinguish the area where the contact is located as being the region where the signal presents some discontinuities as a result of the ions that are stopped and deviated by the gold layer (5 µm< y < 7.5 µm). Dimensions of the IBIC map There are several important features that we can extract from the IBIC maps. But to properly understand the data, it is important to relate the dimensions of the obtained signal with the actual dimensions of the microwire, namely with the length of the p-GaN shell. The shell is clearly visible in the shown SEM image (a) (b) (c) (d) (e) (f) Figure 7.11: SEM image and IBIC maps of a p-n junction microwire detectors irradiated with 1 MeV Si ions while applying different reverse bias voltages to the p-GaN contact. The bottom contact corresponds to the p-GaN contact. 117 and in this specific sample it has a length of 11.66 µm, of which 5.05 µm is covered by Ni/Au, and a diameter of 1.94 µm. From the IBIC map, we extract a sensitive area that has a length of approximately 13 µm and a width of around 2.2 µm. Naturally, the dimension taken from the IBIC map have a larger error associated than the dimensions measured in the SEM images due to the larger beam size and straggling of the ions. Taking into account the comparison between the diameter of the SEM and IBIC data, the dimensions in the IBIC map present an uncertainty between 0.3 and 0.4 µm. The discrepancy between the length in the IBIC map and the SEM image is somewhat larger but this may be due to active region of the detector extending slightly into the neutral n-GaN region, beyond the core-shell region. In any case, we can say that we have a good agreement between the length and diameter of the p-GaN section of the wire extracted from the SEM image and the IBIC map, indicating that, in fact, the active area of the device is essentially limited to the core-shell region. Influence of the applied bias Naturally, the measured signal in the IBIC maps is strongly dependent on the applied bias. In figure 7.10 we see maps measured with a voltage between -2 V and 2 V and in figure 7.11 we show maps for a different detector measured up to a reverse bias of -4 V. As expected, in both cases we see an increase of the CCE with increasing reverse bias as a consequence of the stronger electric field and (a) (b) (c) (d) (e) Figure 7.12: Histogram data, showing the number of ions that originate a certain charge collection as a function of the energy, corresponding to the IBIC maps shown in figure 7.10 118 wider depletion region. This is also visible in the IBIC histograms depicted in figures 7.12 and 7.13. These histograms show the number of ions that originate a certain charge collection as a function of the energy (I recall that the collected energy is proportional to the collected charge). The histograms do not carry any spatial information, meaning that the amount of ions that originate a specific collected energy is counted, independently of where the energy deposition occurs. This is not the case in the IBIC maps and, because the beam scans the total area several times, the collected energy that is obtained at a specific pixel in the IBIC map corresponds to the average of the contribution of all ions at that specific pixel. Hence, due to the different nature of the data representation, we cannot directly relate the data shown in the histograms and the IBIC maps, although the effect of the applied bias is the same. As we can see, the main peak in the histograms shifts to higher energy when increasing the reverse bias. In some cases, an overlap occurs between the noise signal and the energy peak, either when the applied bias is such that the peak shifts to very low energy values (e.g. figure 7.12 (e)) or when the applied bias increases the magnitude of the noise (e.g. figure 7.13 (e)). The fact that both IBIC maps and histograms show a signal at positive voltages is surprising. In figure 7.10 the signals measured at 1 V and 2 V, albeit showing a lower CCE have the same spatial extension as the signals measured at -2 V and the energy peaks are clearly visible in the corresponding histograms. Here it is important to note that the experimental setup, namely the trigger position on the oscilloscope, is set to only transmit the negative (a) (b) (c) (d) (e) Figure 7.13: Histogram data, showing the number of ions that originate a certain charge collection as a function of the energy, corresponding to the IBIC maps shown in figure 7.11 119