High-responsivity silicon p–i–n mesa-photodiode

Silicon p–i–n mesa-photodiodes have been made. Different variants of masking coating during etching of the mesa-profile have been studied. Comparative characteristic of photodetectors manufacture by planar- and mesa-technology have been carried out. Defect formation on the surface of silicon substrates in different options of technology was investigated. Parameters of photodiodes manufactured by planar and meso-technology were investigated. Photodiodes with a meso-structure have higher responsivity and lower capacitance than samples made by conventional planar technology.


Introduction
Along with the technical improvement of optoelectronic devices and systems, the requirements for parameters of their structural elements, photodetectors, in particular, are increasing. Silicon p-i-n photodiodes (PDs) are widely used. They are employed in both civilian and military spheres: in rangefinders (for measuring distances, in geodesy or construction), in missile technology, etc. The most important PD parameters, subject to the most stringent criteria are: dark current (I d ), current monochromatic pulse responsivity (S pulse ) or modulated flux current responsivity (S Iλ ) and capacitance of responsive elements (REs) (C RE ). These parameters depend on the material and the design of the device and the technology of its manufacture. Dark current depends mainly on the concentration of impurities in the n + -layer and charge * Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. states at Si-SiO 2 interface, capacitance-on the concentration of charge carriers in the i-region, and hence, on the resistivity of the material and the area of responsive elements [1], to a greater extent, on the lifetime of minority charge carriers (MCCs)-τ [2].
Modern PDs are manufactured by diffusion-planar technology [3]. According to the latter, responsive crystals are obtained by sequentially performing the following operations: thermal oxidation to create a passivating coating; photolithography to make crystal topology and etch windows for diffusion; diffusion to the front; diffusion in the opposite direction to make an ohmic contact; photolithography to make contact windows; Cr-Au evaporation onto the front and back sides of the substrate. In our case, p-i-n PDs were made on the basis of high-resistivity p-type conductivity silicon of resistivity ρ = 18-24 kΩ cm and a MCC lifetime τ = 1.8-2.2 ms. Phosphorus diffusion was carried out from planar sources in two stages, a predeposition and a drive-in. Accordingly, the route of commercial production included four thermal operations and allowed to obtain PD with a maximum value of S pulse = 0.48 A W −1 and S Iλ = 0.5 A W −1 [4]. The specified parameters refer to PD15M-01 («Rythm», Ukraine). To further improve these parameters, it is necessary to increase MCC lifetime. This can be implemented in several ways: the first one is to use silicon of maximum τ , the second one is to use the technology that provides minimal degradation of the parameter in the production process. It should be noted that the MCC lifetime depends on the resistivity of silicon, i.e. with increasing ρ, τ will increase. However, as practice shows [5], the use of a material of high resistivity increases the probability of defects through scatter in values or instability of dark currents. As far as ρ of silicon increases, the concentration of defects or impurity ions decreases, and it can cause changes in the inverse characteristics of the devices. Given the above, to preserve τ , the production techniques and procedures should be optimized. This can be made by reducing the number of thermal operations in the manufacturing route. It is known that each such an operation reduces the MCC lifetime. Therefore, it was decided to make a mesa-PD, the structure of which will help to eliminate the first thermal oxidation, in contrast to classical planar structures where oxidation is the main operation [6,7]. The possibility of manufacturing silicon structures without passivation before diffusion processes can reduce the cost of products and shorten technological routes. This is one of the novel aspects of this work.
Many works devoted to the manufacture of mesa-type structures and photodetectors have been published. Thus, in [8] the study of photoelectric characteristics of different area breadboard construction responsive pixels is described. The objects under study were PDs ranging in size from 30 × 30 to 100 × 100 µm based on Cd x Hg 1−x Te/Si mesa-structure at x = 0.235 grown by molecular-beam epitaxy. In [9], parameters of p-i-n PD matrices in the format of 320 × 256 elements with the spacing of 30 µm based on InGaAs of the responsivity in the spectral range of 0.9-1.7 µm were investigated. Matrices of responsive elements were made according to planar and mesa-technology. In [10], the process of manufacturing high-speed germanium p-i-n PD with a mesaprofile for a laser rangefinder of maximum responsivity at a wavelength of 1.54 µm was described. In [11] a technology for manufacture mesa-elements of p-i-n diodes focal matrix based on Al x Ga 1−x As/InP is presented. In [12] a study of voltampere characteristics of arrays of responsive elements based on In x Ga 1−x As/InP, made by mesa-technology, is presented. In [13] a comparative analysis of magnetic properties of SiC thin epitaxial layers grown by atomic substitution on the surface of single-crystal silicon in the framework of planar and mesa-technology has been carried out. However, there are no works on silicon p-i-n mesa-PDs.
The purpose of this article is to create silicon p-i-n mesa-PD responsive at the wavelength of λ op = 1.064 µm, which would have a responsivity higher than that provided for by conventional planar technology.

Manufacture of samples
The research was carried out on silicon four-element pi-n PDs with a guard ring (GR). The starting material was single-crystal dislocation-free p-type silicon grown by the Czochralskyi method with orientation [111], ρ ≈ 18−24 kΩ cm and τ ≈ 1.8-2.2 ms. The thickness of the substrate was 490-500 µm. The first thermal operation in the manufacture of responsive crystals was phosphorus predeposition to create a RE-n + -type GR and responsive areas (figure 1(a)). Duration of predeposition was 1200-2400 s. During this operation a p-n junction of 0.5-1 µm in depth, determined by angle lapping, was formed [14].
The surface resistance of wafers after the process was R s = 3.6-3.8 Ω/□. Then the first photolithography was performed. It was important to use a chemically stable photoresist because it was it to be a masking coating when creating a profile of the mesa-structure. Unlike mesa-technology, masking oxide is usually used in planar technology. The photolithography was followed by etching of mesa-grooves by chemical dynamic polishing (CDP) ( figure 1(b)). As a preliminary step, a slow etchant for CDP was experimentally was chosen and an optimal process mode was established to ensure the required depth of the mesa-profile of 5.5-6 µm. HNO 3 :HF:CH 3 COOH = 7:1:3 solution was selected.
FP-383 and FP-25 (Frast-M) photoresists [15] were used in production of experimental samples. PD made by the use of these photoresists did not show the desired results. The coatings did not show adequate chemical resistance to the etchant. Thus, FP-383 photoresist peeled off during the etching of the mesa-structure, and due to it the entire surface of the wafers, including the RE, was etched. FP-25 and  In view of the results, it was decided to try other options for masking. Photoresists AZ1518 and AZ4533 (Microchemicals) [16,17] were chosen for the process. To ensure the required chemical resistance, the following application and tanning modes were used for AZ1518: (1) the speed of the centrifuge during the deposition v = 4000 rpm, the duration of hardening t hard = 30 min at the temperature of T = 120 • C; (2) the deposition of two layers at v = 3000 rpm and t hard = 30 min at T = 120 • C; (3) the deposition of two layers at v = 3000 rpm, with intermediate drying lasting 20 min at the temperature of 95 • C, t hard = 30 min at T = 120 • C.
Tests of the photoresist were performed on experimental wafers, and an arbitrary topology was formed (in this case, a circle) with the provision of the above modes of chemical and dynamic polishing. None of the photolithography variants using the photoresist met the needs. The coating did not peel off, but the surface of the crystal was still etched in places, forming various kinds of 'grids' of defects, which can be seen in figure 3.
AZ4533 photoresist fully met the requirements for adhesion and chemical resistance (v = 4000 rpm, t hard = 30 min at T = 120 • C). Experimental samples obtained when using the photoresist were characterized by the absence of defects and etches (figure 4(a)). But this coating had several disadvantages: high density, which caused problems during the deposition, and a significant market value.
It was also noticed that when polishing wafers with oxide film in this etchant, silicon is etched faster than SiO 2 . As far as the etchant used contains a small volume fraction of hydrofluoric acid, the etching rate of the SiO 2 oxide film is much lower than that of silicon: 0.05-0.06 µm min −1 versus 4 µm min −1 . Based on the above, it was decided to use silicon oxide as a masking coating in the process of etching the profile of the mesa-structure. But as mentioned above, the purpose of using mesa-technology is to reduce the number of thermal operations. Therefore, a masking oxide of the thickness of about 0.5 µm was applied using a Leybold-Heraeus Z-550 plant for RF cathode sputtering with a power of P = 1500 W, argon pressure P Ar = 1 • 10 −2 mbar and process duration 75 min. First photolithography in this case was performed by any photoresist. Then the mesa-structure profile was etched. Residual silicon oxide was etched in HF to create the same crystal topology as in the case of AZ4533 photoresist for comparison and with provision for technology.
Next, the drive-in of phosphorus (figure 1(c)) was carried out, during which there were a redistribution of the dopant in the bulk of the substrate (x n+−p = 4-5 µm [1]) and growing of an anti-reflecting oxide film (2 in figure 4). The surface resistance of the responsive element was R S = 2.3-2.4 Ω/□. Antireflective silicon oxide was formed in an atmosphere of dry oxygen. The thickness of SiO 2 was 0.18-0.19 µm, which corresponds to the minimum reflection condition [18]: where λ is the working wavelength; n the refractive index of SiO 2 (n = 1.46 [19]); d SiO2 is the thickness of the antireflective film.
To create a p-p + -junction (x p+−p = 1-2 µm [4]) (5 in figure 4) and form an ohmic contact, the next thermal operation was diffusion of boron into the back side of the substrate ( figure 1(d)). After the diffusion of boron, photolithography 2 was performed to create contact pads for the RE (3, 4 in figure 4) and gold (d Au = 0.5-0.8 µm) (7 in figure 4) with a sublayer of chromium (6 in figure 4) was evaporated onto the front and back sides of the wafers (d Cr = 10-20 nm for back side and d Cr = 60-100 nm for front side).
Images of responsive crystals created by using the two described options of mesa-technology (PD mesa -AZ4533, PD mesa -SiO 2 ) and a crystal made by planar technology (PDcommercial) can be seen in figure 5. The topology of crystals PD mesa -AZ4533 and PD mesa -SiO 2 is the same.

Study of crystal surface
Mesa technology was used in order to preserve the lifetime of minor charge carriers, as well as to reduce the number of structural defects and generation-recombination centres on the surface and in the volume of crystals.
To confirm the statement, it was decided to examine the samples obtained by three methods, when etching in a selective etchant (figures 6-8). Since dislocations-free silicon of [111] orientation was used, Sirtl's etchant composed of HF-100 cm 3 , CrO 3 -50 g, H 2 O-120 cm 3 was chosen [20]. The duration of etching was 5 min. Previously, gold, chrome and the oxide were removed from the samples. Before the selective etching, washing was performed in a mixture of Caro and ammonia-peroxide solution. The number of dislocations was calculated by the metallographic method [21].
Characteristic triangular etching pits were found on the surface of the samples, which indicated the presence of dislocations. The crystals had a different number of acquired structural defects. Thus, the PD-commercial surface was covered by a small number of dislocation lines with a relative position at an angle of 60 • , which is typical for orientation   Individual dislocations detected after HF cathode sputtering are misfit dislocations. In contrast to thermally grown oxide film, the deposited one does not correspond to the crystal lattice of silicon.

Dislocation formation factors
It should be noted that in mesa-technology when using a masking photoresist, several key factors that contribute to the formation of dislocations are minimized: these are point defects that are formed during thermal oxidation, and thermomechanical stresses in the wafers that appear during each thermal operation. Dislocations negatively affect the magnitude of dark currents and MCC lifetime, and, accordingly, the responsivity.
Consider these factors in the formation of dislocations in more detail. In planar technology, before the diffusion of phosphorus by thermal oxidation, a masking layer of SiO 2 of the thickness of about 0.7 µm is grown. Accordingly, in the presence of such a thick oxide, in subsequent thermal operations there occur near-surface mechanical stresses in the wafers, due to more than seven times the difference between the coefficients of thermal expansion of Si and SiO 2 . Also, when loading/unloading substrates in/from quartz reactors, the temperature rises/falls rapidly to higher/lower one, resulting in radial (σ R ) and angular (σ θ ) tensile stresses in the wafers, which can be described by the following expressions [22]: where σ R , σ θ is the average value of radial and angular tensile stresses, respectively (distribution across the wafer differs); σ R , σ θ are values of radial and angular tensile stresses at T 0 ; α Si is temperature coefficient of thermal expansion of silicon; E is the Young's modulus; T 0 , T are room temperature and thermal process temperature, respectively. In expressions (5) the '−'sign in the denominator describes the cooling process, '+'-the heatin one. Therefore, when cooled, the compression deformation occurs, when heatedexpansion. Reduction of stresses and their influence on the formation of structural defects is achieved by slow cooling of the wafers after the heat treatment and slow introduction of the cassette with wafers in the high temperature zone of the furnace.
Also during oxidation, excess impurity atoms or excess interstitial silicon atoms are formed, which condense in the form of localized clusters of point defects (so-called Bdefects) in a tightly packed crystal lattice. It is known [22] that during the diffusion of phosphorus in the oxidizing atmosphere, local generation of defects, such as stacking faults and dislocations, occurs in places of localized disturbances of the wafer surface. And given the fact that the wafers before thermal operations undergo chemical and dynamic polishing, which minimizes the presence of surface defects [23,24], point defects formed during oxidation are the main centres of localization of dislocations.

Dark current density
Parameters of PDs made by mesa-technology with two variants of masking coating were compared with products created by conventional planar technology. Diffusion modes were the same for all the product variants. The I-V-characteristics of PDs were measured using a hardware-software complex implemented on the basis of the Arduino platform, an Agilent 34410A digital multimeter and a Siglent SPD3303X programmable power source, which were controlled by a personal computer using software created by the authors in the LabView environment.
The I-V-characteristics of the studied PDs were obtained (figure 8).
From figure 8 you can see a significant difference in parameters of mesa-structure PDs made with different masking coatings. The RE dark current density in PD mesa -SiO 2 is higher than J d PD mesa -AZ4533. It is caused by the presence of surface recombination dislocations centres. Although the difference in the concentration of dislocations in these two cases are relatively insignificant N SiO2 dis = 3 • 10 4 -3.4 • 10 4 Cm −2 and N AZ4533 dis = 1 · 10 3 −1,3 · 10 3 cm −2 , about an order of magnitude, but the size of these structural defects is of great importance. Thus, N SiO2 dis1 = 0.4 · 10 4 cm −2 misfit dislocations of increased size relative to other technology options are on the surface of the RE PD mesa -SiO 2 . This means that these defects are characterized by a larger σ ss capture crosssection, which in its turn affects the surface component of dark currents (I d surf. ), being determined according to [25] as follows: where e is electron charge, σ ss is the cross section of the capture, N ss is the density of surface states, v-is the average relative velocity of thermal charge carriers, A p-n is the area of the p-n junction. PD mesa -AZ4533 had slightly higher values of dark currents than commercial products. This can be explained by the increase in the perimeter of the p-n junction due to the etching slope of the mesa-structure (figure 9) and the corresponding increase in the area of the p-n junction,  A p−n = W i P p−n (4) where W i is the width of the SCR (W i ≈ 490 µm [25]), P p-n is the perimeter of the p-n junction. The mesa-structure angle reached 50 • , respectively, the radius of the responsive element increased by 3-4 µm. In this case, when using mesa-technology, the value of A p-n increased by 8-12 µm 2 .

Responsivity
The responsivity of experimental and serial PDs was also investigated at U bias = 120 V and at the pulse duration τ i = 500 ns. The pulse s responsivity of the samples made by mesa-technology was S pulse = 0.46-0.53 A W −1 of PD mesa -SiO 2 and S pulse = 0.5-0.53 A W −1 of PD mesa -AZ4533. The pulse s responsivity of the samples made by planar technology was S pulse = 0.45-0.47 A W −1 of PD-commercial. Dependence of S pulse on the bias voltage of PDs was also obtained. This can be seen from the ( figure 10).
The pulse s modulated flux current responsivity of the samples made by mesa-technology was S Iλ = 0.52-0.61 A W −1 of PD mesa -SiO 2 and S Iλ = 0.57-0.61 A W −1 of PD mesa -AZ4533. The pulse s responsivity of the samples made by planar technology was S pulse = 0.46-0.5 A W −1 of PDcommercial. When using mesa-technology, it is possible to reduce the number of recombination centres and thermal shocks that reduce to the MCC lifetime by eliminating one thermal operation. Therefore, mesa-PD responsivity is higher than that of the PD made by planar technology, which also follows from formula (5), where the responsivity is directly proportional to τ [26]: where e is the charge of the electron; β is the quantum yield; α is absorption coefficient τ n , τ p is lifetime of electrons and holes, respectively; µ n , µ p is mobility of electrons and holes, respectively. Spectral characteristics of the responsivity of PDcommercial and PD meza -AZ4533 were also obtained (figure 11).
From figure 11, it can be seen that the current responsivity of PD meza -AZ4533 is higher than that of commercial PDs in the entire investigated range of wavelengths. Also, after comparing the level of responsitivity with the theoretically possible, it was seen that in the spectral maximum the responsitivity (S λmax ) of the PD meza -AZ4533 approaches the level of an ideal PD. The theoretically possible responsivity value can be determined by formula (5) [10]: In the studied samples, the maximum of responsivity is at λ = 1.01 µm. According to (5), the theoretically possible value of current responsivity of silicon photodetectors is ∼0.815 A W −1 , and for PD meza -AZ4533 S pulse,max ≈ 0.8 A W −1 .

Capacity of responsive elements
The dependence of the C RE on the bias voltage was obtained ( figure 12). The capacity of the experimental samples was slightly lower than that of the serial ones. This is caused by a higher specific resistance of PD meza compared to planar ones, as a result of better preservation of the resistance of the substrate when using mesa technology, because it is known that the capacity depends on the specific resistance of the base material [1]: where A RE is effective area Of RE; φ k is contact potential difference.

Comparison of the proposed FD with samples from the world market
The parameters of the obtained samples compared to PDs of the same type and size available on the world market (table 1). It can be seen from the table 1 that the proposed PDs manufactured using mesa technology are somewhat inferior in terms of dark currents and pulse responsivity, but are samples with the best capacity and modulated flux current responsivity among those offered on the market.
Also, when using mesa technology, it is possible to significantly reduce the cost of products due to the exclusion of a long high-temperature thermal operation, which allows you to save carrier gases, electricity, and mixtures for chemical processing. So, the total duration of thermal operations in the production of commercial PDs is about 450-500 min. The duration of the first thermal oxidation, which is excluded from the technological route of PD meza , is 160 min. Accordingly, the total duration of thermal operations can be reduced by 32%-35% during the production of PD meza , which allows to reduce the cost of the PD crystal by 7%-10%.

Conclusion
Silicon four-element p-i-n PD with a GR have been made by mesa-technology. Different variants of masking coating for etching the mesa-structure profile have been studied. Of the studied coatings, AZ4533 photoresist and SiO 2 deposited by RF cathode sputtering showed the best results. But the surface of the samples of the second variant was covered with a significant number of misfit dislocations, which contributed to the increase of dark currents. The surface of the samples with AZ4533 photoresist sprayed, was characterized by a reduced number of structural defects. As of one thermal operation-oxidation in comparison with commercial PD was kept out, the number of dislocations was reduced by more than an order of magnitude. A comparative analysis of PD parameters made according to two variants of mesa-technology and planar technology was carried out. The smallest dark currents were characteristic of PD-commercial samples, PD mesa -AZ4533 had slightly higher values of dark currents in consequence of the increase in the area of the p-n junction due to the formation of the etching slope. PD mesa -SiO 2 had the worst dark currents due to an increase in the number of recombination centres.
Mesa-structure samples in both cases had a higher monochromatic responsivity than commercial ones. In the case of PD mesa -SiO 2 , a scatter of responsivity and dark currents were observed in different samples, from what it can be concluded that the distribution of structural defects is unsteady. But the upper limit of the responsivity of PD with mesa-structure both variants reached S pulse = 0.53 A W −1 and S Iλ = 0.61 A W −1 as compared to the value of S pulse = 0.47 A W −1 and S Iλ = 0.5 A W −1 for conventional planar technology.
The proposed PDs manufactured using mesa technology are somewhat inferior in terms of dark currents and pulse responsivity, but are samples with the best capacity and modulated flux current responsivity among those offered on the market.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Conflict of interest
There are no conflicts to declare.