A.

Process for deep UV extension

Processing has been extensively presented in reference [4]. It consists in a hybrid FPA whose range of detection is extended from near UV to deep UV. It is based on 320 x 256 pixels of Schottky photodiodes with a pitch of 30 µm. AlGaN is grown on a silicon substrate instead of sapphire substrate only transparent down to 200 nm. The architecture of the detector is based on an undoped active layer with an aluminum content of 35% and a n doped window layer with an aluminum content of 55%. A n doped gradual alloy prevents any conduction and valence band discontinuity between the window and active layers. The presence of a thick and doped window layer limits the spectral bandwidth to 20-25 nm. Basal AlN / GaN multilayer are incorporated for material improvement. But its final removal doesn’t impact the final optoelectronic properties as long as it isn’t exposed to light and exposed to access resistance variations.

The fabrication of an AlGaN 2D array on silicon is similar to that of an AlGaN 2D array on sapphire. It consists in mesa etching, ohmic and Schottky contact definition, passivation with dielectric, contact pads deposition, dicing and hybridization to readout circuit. As usual with nitride components, a strong mechanical stress arises during the epitaxy and induces a strong bowing. The bowing of AlGaN layers grown on silicon is larger than 45 µm for a 2 inch wafer. It requires a particular attention during processing.

After a flip-chip hybridization, silicon substrate is thinned and finally removed by ICP etching. The use of a honeycomb structure straightens the membrane after hybridization in order to allow the membrane integrity all along processing and stress tests. The results show that the dry etching process doesn’t affect the readout circuit properties. The dark current is negligible and we measured noise due to the large capacitance of the photodiode. We also benefit from 3-5 ICP to continue etching close to the gradual and active layers. A key parameter is the temperature kept under 80°C requested for indium bump compatibility and control of crack expansion across layers. Another advantage is its anisotropy and the capability to create a honeycomb structure in order to limit the expansion of cracks.

B.

First optoelectronic results

We measured a QE of 35% at 300 nm for which the window layer is transparent and absorption takes place in the active layer very close to the Schottky contact. When the wavelength is lower, i.e. 260-300 nm, the absorption takes place in the window layer and closer to the backside surface. The QE decreased down to several percents. It means that carriers were immediately recombined in the highly doped layer.

As it was not possible to begin a new dry etching by ICP due to the high risk of failure, we consequently tested a local wet photochemical etching. Silicon honeycomb was preliminarily etched. Then the highly aluminum and silicon doped AlGaN layer (200 nm window and gradient alloy layers) were selectively eliminated [5].

C.

Optoelectronic properties after chemical etching

The effect of wet etching on the enhancement of QE is obvious on figure 2, left. But several drawbacks take place. First, the contribution corresponding to GaN band-gap, i.e. 360 nm, is due to a lower access resistance when carriers are generated in the GaN based honey-comb. Second, a long response time follows the fast responsivity as shown in figure n°2 (right). Such effects may be due to the versatile access resistance in the honeycomb but also to a high doping level in the active layer prompt to deep level activation. When acquiring signal with integration time shorter than 10 ms, this slow activation is negligible and the only measured signal is the fast component.

Fig.1.

Sketch of the FPA adapted to deep UV before(left) and after wet etching (right).

Fig.2.

Effect of photo-chemical etching on near UV spectra (left) and response time (right)

A readout noise around 340 electrons rms was measured [5]. This value is high compared to the 150 electrons rms noise measured with AlGaN on sapphire camera. Compared to the readout circuit noise measured at 45 electrons with unconnected pixels, the excess noise is suspected to be due to the high capacitance of photodiodes. This high value is achieved by the low desertion layer due to a high doping level. Indeed, n doping is achieved by silane. This gas is present in the growth chamber after the intentionally doped layers growth (window and gradual layers) and induces a large residual doping.

D.

Responsivity in the deep UV range

The QE is estimated from the fast component of the photocurrent. The spectral responsivity of this focal plane array presented a QE from 10% to 20% from 130 nm to 290 nm after the removal of the highly doped contact layer. We observe in figure n°3 a QE around 1% from 310 nm to 360 nm due to the activation of the GaN layer in the honeycomb. These values consist in an optimistic results but need to be clarified. But some strong improvement in terms of rejection of visible and elimination of long response time are required.

A

Fabrication

The epitaxial material is similar to the wafer involved in previous attempts. It consists in a AlN / GaN / AlN / Al_.6Ga_.4N / Al_.4Ga_.6N multilayers. The aluminum content in both active and window layers is increase by 5% in order to approach solar blind devices properties in terms of dark current and noise figure. We paid attention on the solid state doping instead of silane based doping suspected to induce high capacitance and readout noise.

Several honeycomb thicknesses from 7 to 8 µm are tested in order to improve filling factor and robustness. A honeycomb structure is shown in figure 4. The dark layer reveals the AlN layer just below the window layer. A set of 4 FPA was fabricated in order to determine the best stabilized conditions for the next set of 8 FPA. After silicon etching the devices with the narrowest honeycomb and thinnest thickness (i.e. those extracted from the edge of the wafer: R08C07-D3 and R08C06-D2) revealed a high density of cracks. Few cracks appear during honey-comb etching in AlGaN. Such a density of crack is crippling for future use in harsh environment. Consequently, the AlGaN etching was subdivided in steps of 1 minute in order to avoid a dramatic increase of temperature for indium bumps. Then, the next set of 8 representative FPA will be defined with larger honeycomb safer for membrane robustness: 10 µm corresponding to a filling factor of 45%.

Fig. 3.

Spectral responsivity from 130 to 360 nm.

Fig. 4.

Silicon honeycomb after ICP etching. (left). AlGaN / AlN / GaN honeycomb after silicon removal for SEM acquisition

B.

Electrical properties

Some of the electrical properties are summarized in table 1. The cracks are contributing to the average dark current and short cuts showing the necessity to avoid it. Unconnected pixels will have to be solved by planarity concern. Future micro-section analyze will define whether indium bump suffer from two ICP etching in silicon and (Ga,Al)N.

Tab.1.

Electrical features measured on 4 devices dedicated to etching optimization

C.

Spectral responsivity

The near UV spectral properties are presented in figure 6. We observe a rejection of visible wavelengths larger than 4 orders of magnitude limited by internal fluctuations in Schottky contact [6], a sharp cut-off at 300 nm limited by alloy fluctuations in (Ga,Al)N [7]. Future architectures dedicated to larger rejection of visible are under design. On the contrary, the responsivity at wavelength lower than band gap (300 nm with 40% of aluminium content) is particularly flat except below 250 nm for which the calibration with Xenon lamp in the air is difficult. As a first conclusion, the highly doped gradient layer has to be completely removed.

Fig.5.

Cartography and histogram of backthinned AlGaN (R08C07-D3, gain 16µV/e-, integration time: 10 ms)

Fig.6.

Spectral responsivity with logarithmic (left) and linear scale (right)

Photoresponse spectra from 50 eV (25 nm) to 220 eV (6 nm) have been achieved owing to the synchrotron line Metrology X-UV from Soleil. We observe in the next figure a QE larger than 100% (1 e-/ph) with energies larger than 120 nm. This is due to multi-excitation when photon energy is larger than 3 times the band-gap energy. Most of the differences of QE are explained by the membrane’s thickness. 450 nm is too thick to avoid recombination in the gradient layer (R08C07-D3 is an exception). The thinnest membrane of 300 nm gives the best results, but the numerous cracks induce access resistance’s limitations and saturation effects. We observe a slightly better QE with R03C04 (wet etched FPA from chapter II) from 12 nm to 21 nm. This effect has to be considered with caution. An unexplained spectral feature took place. Some signal is added to the diffracted order in the incoming signal. Consequently, 400 nm is the best tradeoff between membrane’s robustness and QE in deep UV.

Then, we define the collection efficiency as the ratio QE/3hν to avoid multi-excitation effects when photon energy exceed the band gap energy. We still observe figure 8 an increase of collection efficiency correlated to the deeper penetration of photon in the AlGaN membrane as shown in the cartography of figure 8 (right).

Fig.7.

Quantum efficiency and efficacy of collection from 50 eV (25 nm) to 220 eV (6 nm)

Fig.8

Left: Collection efficiency; and Right: Cartography of electric field intensity I in AlGaN membrane depth (abscises) versus wavelength (vertical)

Another interesting improvement due to the dry etching is the fast response time observed for all 4 samples and presented figure 9 for several energies with FPA number R08C07-D3. We don’t observe any overshot but a tiny increase of responsivity. When switch off, the signal comes to near zero immediately. A component of few seconds of less than few percents of the total signal is observed for all samples.

Fig.9.

Response time of 2D array to synchrotron signal with photon energy from 50 to 210 eV

Fig.10.

Decrease of responsivity under high flux of 10¹³ ph/s/cm² at 120 nm

In order to check the behaviour of our device in harsh environments, we have measured the responsivity of our device under a large flux from the metrology line just filtered with a Lyman a filter. Quantitative measurements are suspicious due to parasitic features observed in images and spectra. The estimation of 1% is too low compared to the results observed with the first prototype chemically etched. It will be precisely measured by a continuous spectra from visible to 50 nm in the DISCO line in early September. Nevertheless, we observe a slow decrease of signal after a fluence about 10¹⁶ photons/cm². A comparison of behaviour with the state of art silicon photodiodes from AXUV [8] is interesting. Indeed, silicon based photodiodes with a SiO₂ passivation show similar behaviour with equivalent fluence. Some preliminary explanation could be a dead layer, a thin oxide or even localized defects and traps in bulk AlGaN.