© The Author(s), published by EDP Sciences, 2026

1 Introduction

In response to the concerning climate changes, such as the steep increase of temperature anomaly w.r.t. 1980 that is approaching 1.5 °C in 2025 [1], actors from both Industry and Academia have increasingly fuelled research on innovative photovoltaic (PV) devices. As a result, global PV capacity is increasing and approaching 2 TW [2].

Research on innovative photovoltaic devices is actively working to increase this value, by making photovoltaic devices more efficient [3, 4] and their power generation more continuous [5]. This research strongly relies on the indoor characterization of PV devices, for the proper assessment under standard, repeatable experimental conditions, to effectively correlate measured electrical properties with device manufacturing steps. The scientific community agreed on Standard Testing Conditions (STCs) for the indoor characterization of PV devices [6, 7], that define the PV cell to be characterized with its temperature kept at 25 °C, under an irradiance of 1000 W/m², in 1.5 air mass condition (AM 1.5, corresponding to a zenith angle of 48.2°). International standards such as IEC 60904 and ISO 9845-1 define procedures for indoor characterization of devices, and provide reference spectra that should be used, that correspond to the irradiance received on an equator-facing 37° tilted PV cell [8]. Indoor Characterization in STCs is carried out through solar simulators that accurately measure current–voltage (I–V) characteristics of the device under test (DUT), under artificial illumination with controlled spectrum and intensity. Solar simulator sources can be either pulsed or continuous-wave and are equipped with measurement circuits to record DUT electrical quantities such open-circuit voltage, short-circuit current, power conversion efficiency, and fill factor. During the measurement, the cell temperature is usually kept constant by means of a (e.g., thermoelectric) cooler.

Pulsed solar simulators are usually required when large currents are to be measured, as for multijunction (MJ) concentrating photovoltaics (CPV) devices, that could heat the experimental device up. Spectral matching to the standard spectrum is ensured through proper absorption filter, and any mismatch is evaluated in the spectral mismatch ratio (SMR) according to the standard IEC 60904-7:2008-11 [9]. A triggered measurement circuit is used to perform fast I-V sweeps (either voltage or current controlled) during the constant region of the flash extinction.

So far, most of literature [10, 11] lacks fast spectral characterization of pulsed solar simulators, and those who employed fast spectrometer setups [12, 13] reported important effects that could not be measured with steady-state spectrometers, such as the illumination spectra shifting from blue rich to red rich during the flash extinction.

This work focused on the spectral characterization of a pulsed solar simulator, which makes use of xenon lamps discharge to produce fast (few milliseconds) flashes with good spectral and intensity uniformity over the illuminated area of the DUT. Solar simulator’s pulsed flashes were extensively characterized in time and wavelength, to evaluate the impact of lamp’s supply voltage on the flash extinction, and therefore on the measured I–V curves of three DUTs: a top single junction (SJ), a bottom SJ, and a silicon photodiode. The measurement setup can be of low impact, as it can be placed next to the DUT without any relevant shadowing, to allow the in-situ characterization of the exact light hitting the device during that specific measurement.

2 Material and methods

In-situ fast spectral characterization was carried out through the setup displayed in Figure 1. We employed a Technoexan Ltd (an innovation company of the Ioffe Institute in St. Petersburg, Russia) CLASS A (in 400–1200 nm) pulsed solar simulator, designed to work with either semi-collimated 1-sun radiation, with the aperture/lens fixture visible in Figure 1b, or in concentration, up to 5000 suns.

Fig. 1 (a) Measurement setup with a triggered fast spectrometer next to the DUT; (b) 1-sun setup, with radiation from one lamp illuminating the DUT through an aperture/lens fixture; (c) schematic of the setup shown in (b), displaying the trigger photodiode and the spectrometer slit both next to the DUT, under the light from the Fresnel lens.

The simulator is designed to ensure a 5% spatial light uniformity across a 2 cm × 2 cm area, a 2% flash-to-flash stability, and can emit a pulse of light once every 15 s.

Current–voltage sweeps are carried out through fast voltage discharges over the 800 μs constant region of the flash extinction, to record light I–V characteristics in the range −5 V/+10 V and up to 10 A.

The solar simulator is equipped with four xenon arc lamps, allegedly identical, each mounted with an absorption filter to include atmospheric absorption and emulate the standard ASTM G173-03 AM1.5 1000 W/m² irradiance spectrum. The measured curves should be independent of lamp, from manufacturer’s specifications.

To characterize the flash occurring during the I–V sweep, which lasts around 800 μs, a microprocessor-based custom-made circuit was built and set to trigger a measurement once input from a photodiode exceeded a set threshold. Spectra were recorded by the commercially available Ocean Optics FX-XR1-ES CCD spectrometer, used without further modification, in free-space, coupled directly with the semi-collimated light of the simulator. This spectrometer was chosen for its speed, since it’s capable to perform one measurement every 221 μs with an integration time of 10 μs, and for its measurement range: 200–1025 nm, with a 0.4 nm resolution (typical). To characterize the whole flash extinction, therefore, 27 consecutive measurements were performed over a 6 ms duration, each containing 2136 points between 200 and 1025 nm. However, only the 350–950 nm range is reported, as it’s the most relevant.

As visible in Figure 1, spectrometer slit and trigger photodiode were put directly in the simulator’s chamber, next to the DUT.

Together with fast spectral characterization of flashes, I–V curves were recorded by the instrument itself, through its custom source-meter unit that performs voltage–current sweeps during the 800 μs measurement window. Three different photovoltaic devices were measured for this purpose: a top SJ manufactured by Fraunhofer and sensitive in the 300–700 nm range (“SJ top”), a bottom SJ manufactured by Fraunhofer and sensitive in the 850–1800 nm (“SJ bot”), and a silicon photodiode manufactured by THORLABS (p/n FDS1010) and sensitive in the 350–1100 nm range (“Si PD”).

3 Results and discussion

3.1 Characterization of flash extinction

Figure 2a depicts a typical flash from one lamp at a given supply voltage, compared to the ASTMG173 1000 W/m² reference spectrum, that the simulator tries to emulate with a ±25% maximum mismatch (class A). Each spectrum is formed by a broad blackbody radiation that peaks around 570 nm and quickly decreases to 50% after 2.5 ms, superimposed with narrow Xe discharge spectral lines (e.g., 764 and 822 nm) that decrease more slowly, and never turn off completely during the 6 ms window. Since different wavelengths exhibit different dynamics, the temporal variation should be considered at each wavelength, such as those visible in the plots in Figures 2b–2c.

Fig. 2 Waterfall plot and surface plot (b) for the flash extinction from one xenon arc lamp, with one spectrum every 221 μs; the grey curve in (a) displays the ASTMG173 1000 W/m2 reference spectrum; the black arrow indicates the direction of extinction, i.e. from red to blue curves; (c) time series for selected wavelengths from data in (a) and (b), normalized over each curve’s maximum.

3.2 Impact of supply voltage on lamp’s flash

Supply voltage is varied to adjust the simulated irradiance until the 1-sun short-circuit current density ($J_{\mathrm{sc}}$) of the DUT matches that of a calibrated reference cell, regardless of the differences in spectral response of the two devices [1], which is taken into account by the spectral mismatch factor (SMM, or M). When adjusting the supply voltage, one expects the whole spectrum to vary coherently, with every wavelength showing the same increase/decrease with increasing/decreasing voltage. If this was the case, spectral irradiance at any wavelength would increase linearly with supply voltage, integrated irradiance would increase linearly with supply voltage, and measured $J_{\mathrm{sc}}$ would increase linearly as well.

Results in Figure 3a, however, report that this occurs only for the blackbody radiation (around 600 nm), and that the wavelengths corresponding to the Xe lines (e.g., 764 and 823 nm) exhibit a slower increase with voltage, marked by the dips visible in Figure 3b. Irradiance does not increase coherently with supply voltage, as visible in Figure 3c, but lines and 764 and 823 nm exhibit slower increases with voltage. This non-ideal behaviour might introduce measurement artifacts, as spectra with higher supply voltage lacks some IR spectral components.

Fig. 3 (a) Irradiance maxima recorded for the same lamp at four different supply voltages; (b) normalized curves from (a) over the 650 V curve (black line); (c) values at specific wavelengths from (b) plotted vs. supply voltage; (d) $\mathit{SM}{R}_{\mathrm{top}/\mathrm{mid}}$ plotted vs. time, for four different supply voltages.

The effect can thus be estimated through the top/middle Spectral Mismatch Ratio plotted in Figure 3d, defined as ${\mathrm{SMR}}_{\mathrm{top}/\mathrm{mid}}=\frac{\int E_{\mathrm{mis}}^{\mathrm{top}}\left(\mathrm{\lambda }\right)\mathrm{d\lambda }}{\int E_{\mathrm{mis}}^{\mathrm{mid}}\left(\mathrm{\lambda }\right)\mathrm{d\lambda }}\frac{\int E_{\mathrm{ref}}^{\mathrm{mid}}\left(\mathrm{\lambda }\right)\mathrm{d\lambda }}{\int E_{\mathrm{ref}}^{\mathrm{top}}\left(\mathrm{\lambda }\right)\mathrm{d\lambda }}$, being $E\left(\mathrm{\lambda }\right)$ the measured/reference spectral irradiance in W/m²nm for the InGaP-top (350–650 nm) or InGaAs-middle (650–950 nm) range. Consistent with results in Figure 2, and with results reported in [2], the $\mathit{SM}{R}_{\mathrm{top}/\mathrm{mid}}$ within the first 5 ms increases with applied voltage and decreases with time as the spectra become richer with IR, since these components decrease faster during the flash extinction.

3.3 Impact of supply voltage on measured I–V curves

Results reported in Section 3.2 suggest that the spectrum at one supply voltage (e.g. 650 V) does not increase coherently with voltage, but discharge at higher voltages (e.g. 800 V) lack of some spectral components, such as 467, 764, 823 nm, etc. Varying supply voltage without evaluation of its impact might lead to an underestimation of the simulator’s irradiance and to an underestimation of the measured $J_{\mathit{sc}}$.

In order to prove this claim, we measured the I-V curves of three different devices, each sensitive to a different spectral range, thus each including a different number of the dips shown in Figure 3b. We expect the more dips are included in the DUT’s range, the stronger underestimation of the $J_{\mathit{sc}}$.

For this purpose, Figure 4 shows, for each device, the $J_{\mathit{sc}}$ at different supply voltage (650, 700, 750, 800 V), normalized over its value at 650 V. Each data point is the result of 5 consecutive measurements.

Fig. 4 Short-circuit current density, normalized over its initial value, versus the integrated irradiance (350–950 nm). For each device, the solid curve represents the measured data, and the dotted curve represents the linear regression of the first two points of the curve.

4 Discussion

Results in Section 3.1 reveal that a typical xenon lamp’s intensity does not decrease coherently during its extinction, but it’s formed by spectral lines that decrease far more slowly than others. This dynamics can be considered coherent enough during the first ~1 ms of the extinction, where voltage–current sweeps are carried out, as shown in the SMR reported in Figure 3d. This assumption was possible thanks to speed of the spectrometer that allowed to characterize the flash dynamics with good temporal resolution.

Results in Section 3.2 report that when supply voltage is varied to adjust irradiance spectra, do not change coherently, but those wavelengths appearing as dips in Figure 3b increase more slowly, and the higher-voltage spectra appear poorer of these spectral components. SMR@1 ms in Figure 3d decreases from 1.069 to 0.985 as supply voltage is decreased from 800 to 650 V, which gives a clear numerical value to quantify the effect.

The impact of supply voltage on measured I–V curves is depicted in Figure 4. These results compare the ideal behaviour where spectra increase coherently with supply voltage (dotted lines), and the real behaviour (solid lines) constituted by the actual measurements. The magnitude of this discrepancy is quantified by the “Max nonlinearity” in Table 1, which is high for the “Si PD” device since it includes a large number of dips in its large conversion range 850–1800 nm, and highest for the “SJ bot” device, which conversion range 850–1800 nm extends further into the IR, beyond the spectrometer’s range. These results suggest that further into IR, although not measurable, even more dips might appear and strongly affect the “SJ bot” current. Contrarily, being most of these dips in the 650–850 nm range, the “SJ top” device is almost unaffected, as its sensitivity range 300–700 nm barely includes any dip.

Comparison between experimental results and those the device would have in ideal conditions, namely if all the spectral components would increase coherently with voltage, leads to the outcome that $J_{\mathrm{sc}}$ is always underestimated, up to −2.75% (“SJ bot”).

5 Conclusion

Results in this work highlighted measurements artifacts that can impact indoor characterization, if not identified and considered. Temporal variation of spectral lines is not the same for every wavelength, but the discrepancy is acceptable for the first ~1 ms of the discharge, where the fast I–V sweeps occur. Supply voltage, surprisingly, introduces 10% inhomogeneity in lamp’s spectral components when increased by just 50 V, and 20% inhomogeneity when increased from 650 to 800 V. Consequently, irradiance does not scale linearly with supply voltage, as some spectral components are modified less than others. This yields to a voltage-dependent short-circuit current density underestimation up to −2.75%. The proposed methodology requires an expensive spectrometer, but offers an easy in-situ implementation, providing spectral measurements for each flash hitting the DUT. Results in this work suggest to extend the measurement range further into the infrared region, as infrared spectra at higher supply voltages could differ even more from the lower supply voltages ones.

Funding

This work was financed by the Research Fund for the Italian Electrical System under the Three-Year Research Plan 2025–2027 (DM MASE n. 388, 06.11.2024), in compliance with the Decree of April 12th, 2024.

Conflicts of interest

All authors have nothing to disclose.

Data availability statement

Data available upon reasonable request.

Author contribution statement

Conceptualization and Methodology, M.V. I., F. T., and A. M; Data Curation, Writing, and Original Draft Preparation, M. V. I.; Review & Editing, Supervision, G. T.

References

All Tables

All Figures

	Fig. 1 (a) Measurement setup with a triggered fast spectrometer next to the DUT; (b) 1-sun setup, with radiation from one lamp illuminating the DUT through an aperture/lens fixture; (c) schematic of the setup shown in (b), displaying the trigger photodiode and the spectrometer slit both next to the DUT, under the light from the Fresnel lens.
In the text

	Fig. 2 Waterfall plot and surface plot (b) for the flash extinction from one xenon arc lamp, with one spectrum every 221 μs; the grey curve in (a) displays the ASTMG173 1000 W/m2 reference spectrum; the black arrow indicates the direction of extinction, i.e. from red to blue curves; (c) time series for selected wavelengths from data in (a) and (b), normalized over each curve’s maximum.
In the text

	Fig. 3 (a) Irradiance maxima recorded for the same lamp at four different supply voltages; (b) normalized curves from (a) over the 650 V curve (black line); (c) values at specific wavelengths from (b) plotted vs. supply voltage; (d) $\mathit{SM}{R}_{\mathrm{top}/\mathrm{mid}}$ plotted vs. time, for four different supply voltages.
In the text

	Fig. 4 Short-circuit current density, normalized over its initial value, versus the integrated irradiance (350–950 nm). For each device, the solid curve represents the measured data, and the dotted curve represents the linear regression of the first two points of the curve.
In the text