Optical characterization of PureGaB germanium-on-silicon (Ge-on-Si) photodiodes was performed for wavelengths between 255 nm and 1550 nm. In PureGaB technology, chemical vapor deposition is used to grow germanium islands in oxide windows to the silicon substrate and then cap them in-situ with nm-thin layers of first gallium and then boron, thus forming nm-shallow p+n diodes. These PureGaB Ge-on-Si photodiodes are CMOS compatible and characterized by low leakage currents. Here they are shown to have high responsivity in the whole ultraviolet (UV) to near infrared (NIR) wavelength range. Particularly, two sets of diodes were studied with respect to possible detrimental effects of the Al metallization/alloying process steps on the responsivity. Al-mediated transport of the Ge and underlying Si was observed if the PureGaB layer, which forms a barrier to metal layers, did not cover all surfaces of the Ge islands. A simulation study was performed confirming that the presence of acceptor traps at the Ge/Si interface could decrease the otherwise high theoretically attainable responsivity of PureGaB Ge-on-Si photodiodes in the whole UV to NIR range. A modification of the device structure is proposed where the PureGaB layer covers not only the top surface of the Ge-islands, but also the sidewalls. It was found that to mitigate premature breakdown, it would be necessary to add p-doped guard rings in Si around the perimeter of Ge islands, but this PureGaB-all-around structure would not compromise the optical performance.
PureB single-photon avalanche diodes (SPADs) were investigated with the aid of a newly developed TCAD-based numerical modeling method with which characteristics related to the avalanching behavior can be simulated. The p+ region forming the anode of the PureB p+n photodiode is extremely shallow, only a few nanometer deep, which is essential for obtaining a high photon detection efficiency (PDE) for near-, vacuum- and extreme-ultraviolet (NUV/VUV/EUV) light detection but when an implicit guard ring (GR) is implemented, the dark count rate (DCR) can, despite the GR, be deteriorated at the very sharp corners of the p+-region where there is a high concentration of the electric-field. By comparing measurements to simulations, the main mechanism dominating the DCR in the PureB SPADs was identified as band-to-band tunneling (BTBT) while trap-assisted-tunneling also plays a role when the perimeter breakdown is low. Increasing the dose of carriers in the enhancement region negatively impacts the total DCR of the device, but also shifts the origin of the dominant DCR contribution from perimeter to the active region. The simulations for optimization of the SPAD geometry predict that a modification of the n-doped epitaxial region of the PureB SPADs could decrease the DCR by almost two orders of magnitude. This is achieved by increasing the n-epi-layer thickness from 1 μm to 3 μm and lowering the doping from 1015 cm-3 to 1014 cm-3. A high electric field at the vertical pn junction in the active region can also be minimized by modifying the implantation parameters of the n-enhancement region thus keeping the BTBT contribution to the DCR sufficiently low.
The light emission from silicon PureB photodiodes was investigated in both forward- and avalanchemode operation and correlated to the presence of process-dependent defects that influence the reverse IV characteristics. As opposed to “defect-free” diodes with low dark currents and abrupt breakdown behavior, the diodes with defects had higher current levels and light-emitting spots appearing at voltages far below the breakdown voltage otherwise set by the implemented doping profiles. The role of such defect-related behavior for the application of the photodiodes as single-photon avalanche diodes (SPADs) and avalanche-mode light-emitting diodes (AMLEDs) is assessed in connection with the recent demonstration of these basic devices as both the light-emitting and light-detecting elements in optocoupler circuits integrated in CMOS for data transmission purposes.
PureB silicon photodiodes have nm-shallow p+n junctions with which photons/electrons with penetration-depths of a few nanometer can be detected. PureB Single-Photon Avalanche Diodes (SPADs) were fabricated and analysed by 2D numerical modeling as an extension to TCAD software. The very shallow p+ -anode has high perimeter curvature that enhances the electric field. In SPADs, noise is quantified by the dark count rate (DCR) that is a measure for the number of false counts triggered by unwanted processes in the non-illuminated device. Just like for desired events, the probability a dark count increases with increasing electric field and the perimeter conditions are critical. In this work, the DCR was studied by two 2D methods of analysis: the “quasi-2D” (Q-2D) method where vertical 1D cross-sections were assumed for calculating the electron/hole avalanche-probabilities, and the “ionization-integral 2D” (II-2D) method where crosssections were placed where the maximum ionization-integrals were calculated. The Q-2D method gave satisfactory results in structures where the peripheral regions had a small contribution to the DCR, such as in devices with conventional deepjunction guard rings (GRs). Otherwise, the II-2D method proved to be much more precise. The results show that the DCR simulation methods are useful for optimizing the compromise between fill-factor and p-/n-doping profile design in SPAD devices. For the experimentally investigated PureB SPADs, excellent agreement of the measured and simulated DCR was achieved. This shows that although an implicit GR is attractively compact, the very shallow pn-junction gives a risk of having such a low breakdown voltage at the perimeter that the DCR of the device may be negatively impacted.