A. Comparisons of several T2SL structure designs

Comparisons are first made with 3 monoelement structures that have the same active zone thickness of 1μm, as seen in Fig. 3. They were made by Molecular Beam Epitaxy (MBE) using GaSb as substrate and were designed in order to have the same wavelength cutoff (λ_c=5μm). Sample A is GaSb-rich, with a period made of 10ML of InAs and 19 ML of GaSb (10/19). Sample B is symmetric (10/10). Sample C is InAs-rich (7/4). Quantum Efficiency (QE) measurements have been performed at 0V bias voltage on these 3 structures. First the relative photoresponse was measured using a FTIR spectrometer and then the absolute QE was calibrated using a SR200 blackbody [1].

Fig 3

Right ::Comparison of quantum efficiency of 3 different T2SL photodiodes at 77K. Left : QE as a function of the wavelength, for Vbias=0V; right : QE as a function of the bias, for λ=4.5μm.

The QE results were compared in Fig. 3. At λ=4.5μm, better QE values were obtained by the symmetric and GaSb-rich SL structures while the absorption is higher in the InAs-rich design because of its large electron–holes wave-function overlap.

To explain why the QE of the InAs-rich structure is unexpectedly much lower than the other two, measurements at λ=4.5μm were analysed as a function of the bias voltage. Results are also shown in Fig 3. Increasing the bias voltage improves QE for the InAs-rich structure. It increases linearly from 6% to 17% by changing the bias voltage from 0V to -1V and then saturates. On the contrary, the QE of the GaSb-rich structures is independent of the bias voltage

The photocurrent is created in two steps: photon absorption (which creates the electron/hole pair) and collection of those carriers. The fact that increasing the bias voltage also improves the QE means that there is a problem of transport in the InAs-rich structure.

Simulations using Hovel equations [2][3] have been performed in order to determine the minority carrier diffusion length. QE is the sum of the contributions of three zones: the quasi neutral zone N (QEn), the depletion zone (QEzce) and the quasi neutral zone P (QEp).

With:

Where α is the absorption coefficient, set equal to 2200 cm^-1 [4]; L_e and L_h are the electron and hole diffusion lengths respectively, with L_e set equal to 6μm [5]. R is the reflectance at the air/T2SL interface and set equal to 0.3. x₁ and x_max-x₂ the thickness of the quasi neutral zone N and the quasi neutral zone P and depend on the bias voltage.

The InAs-rich SL structure is n-type residual with hole minority carriers [6]. We determined the hole diffusion length L_h by fitting the equation with the measured QE, using L_h as the only adjustable parameter. Fig 4 shows the QE simulation at V_bias=0 and T =77K for three InAs-rich samples with three different active zone thicknesses equal to 500nm, 1μm and 4μm. The best agreement of the measured QE curves with Hovel equations was obtained with L_h=80nm. This poor value in diffusion length does not allow the hole minority carrirs to reach the collection zone and then penalizes the QE.

Fig 4

Left : comparison between QE measurements and predictions from Hovel equations for three InAs-rich photodiodes with different active zone thicknesses (500 μm, 1 μm and 4 μm), at 77K and 0V bias voltage; Right : schematic view of the front-side illuminated InAs-rich photodiode.

B. Conclusions

In the InAs-rich structure, the active zone features a residual N-type doping, meaning that the minority carriers are holes. As holes have a short diffusion length, the collection process is not optimal. One solution would be to change the minority carriers into electrons, which have a better diffusion length [6]. This can be done by slightly P-doping the active zone, leading to a P⁺P^-N junction as used in [7] to optimize LWIR structures. Fig 5 displays the QE comparison between intentionally P doped and non intentionally doped (nid) N-type residual InAs-rich photodiodes (7/4, 4μm active zone thickness, front side illuminated). We note that the measured QE is enhanced by up to 10 times by P-doping the InAs-rich T2SL (40% QE efficiency at λ=4.5μm for the P doped structure). To complete the comparison, QE measurements as a function of bias voltage are also plotted. In the P doped structure, we can see that the QE does not vary with the bias voltage, staying at 40% for λ=4.5μm, highlighting an optimised collection of electron minority carriers in p-doped InAs-rich T2SL structure.

Fig 5

Left : QE measurements on InAs-rich T2SL photodiodes with non intentionally doped or P doped active zone at 77K and V_bias=0V; Right : schematic view of the front-side illuminated InAs-rich photodiode (the p-n junction is now at the i/n interface, allowing a good overlap between the absorption and the collection zones).

Another solution to improve the QE is to ensure that the holes are created closer to the P/N junction, by backside illuminating the T2SL just like the FPA configuration. Fig 6 represents the QE measurement of an InAs-rich nid MWIR FPA (320x256) made in France [2]. The 42% average QE is due to the fact that this structure is double-pass and that the depletion zone is now betterr overlapping with the absorption zone, which allows more hole minority carriers to be collected.

Fig 6

Left : QE measurements on InAs-rich MWIR FPA (320x256) at 77K on 600 pixels at the center; right : schematic view of the back-side illuminated InAs-rich photodiode (the p-n junction is now at the p/i interface, allowing a good overlap between the absorption and the collection zones).

A. Residual Fixed Pattern Noise

Spatial noise is due to various physical effects that can appear in the photodiode itself, in the readout circuit or during the technological steps of hybridization. It results in the fact that pixels’ response is not absolutely identical in an FPA. By using a gain/offset correction table, it is possible to damp the impact of those non-uniformities.

In order to obtain the gain/offset table, we used a Two Point Correction (TPC). It consists in the use of a blackbody at two fixed fluxes, or rather at two temperatures T_bb with radiances L_bb given by Planck’s law for a blackbodies:

where h is the Planck constant,k is the Boltzmann constant, c is the speed of light and λ is the wavelength. Gain and offset for each pixel are calculated such that all the pixels deliver the same signal when exposed to blackbodies at T1 and T2. As the pixels’ behavior is not rigorously linear, the table calculated will not be absolutely correct at temperature between T1 and T2. This explains the W shape seen on measurements of RFPN, which is the remaining spatial noise after correction.

RFPN has been measured on commercial IRNova T2SL MWIR FPA (320x256 FPAs with 30μm pixel pitch). Measurements have been done using the following routine:

This makes 2 measurements a day, except for the first day when more measurements were done.

For each measurement, gain and offset corrections were applied on each pixel, using the G/O tables determined on day D0.

We consider the correction is still valid as long as RFPN is lower than the temporal noise (TN) measured at a given day.

In the case of the IRNova FPA, results in Fig 7 show RFPN/TN for each measurement. We can see that even after 3 weeks, the correction can be deemed as valid, which means that the stability over time is excellent. The number of flickering pixels is also very low (typically 10 pixels across the full FPA).

This result means that a calibration can be made and does not need to be refreshed for a long time, which ultimately increases the IRFPA uptime (as the IR device has to be disabled during calibration phases),. In some cases, the IR device could even be designed without on-board calibration device, which greatly impacts the Size Weight and Power (SWaP) issue.

Fig 7

RFPN/rms using the same correction on different days

B. Discussion

Results in terms of RFPN and number of flickering pixels seem very promising for T2SL technology, potentially ensuring long-term stability of infrared instruments absolute calibration.

In a more general manner, one may wonder whether RFPN is the most relevant figure of merit to faithfully depict stability over time. Indeed, RFPN highlights the nonlinearity of pixels. It may not be a major concern as long as the behavior is stable, thus correctable. This preliminary work shows that only using RFPN standards will not be enough to state on the “correctability” (or ability to be calibrated on the long-term) of the FPA. For example, flickering pixels (which may bias the “standard” pixel response used for corrections), may be overlooked as the RFPN cannot discriminate them (and apply proper correction to them). We may reach a scenario where RFPN is very low (if no flickering occurred during the calibration) and the resulting image however cannot be considered as stable. Complementary measurements are ongoing at ONERA, with the objective to define a measurement protocol to quantify the “correctability” of FPAs and to compare the merits of different competing technologies.