Tensile‑Stress Engineering Enhances Photoluminescence in InGaN/GaN Quantum‑Well LEDs on Silicon

Background

InGaN/GaN MQW structures on silicon, rather than conventional sapphire, have attracted growing attention for their potential to enable low‑cost solid‑state lighting, display panels, and silicon photonics [1–5]. The primary challenge lies in the 56 % thermal expansion mismatch between GaN and Si, which induces significant tensile stress and crack formation in the GaN film [6–9]. For LEDs and laser diodes, an Si‑doped n‑type GaN layer is essential, adding further tensile stress. Recent efforts to mitigate these issues include the use of intermediate layers with suitable compressive stress [10–16], δ‑doping for strain relaxation [17,18], and lattice‑matched buffer layers [19,20]. Periodic Si δ‑doping in the n‑type GaN layer has been shown to produce smoother GaN layers, higher crystalline quality, and reduced crack density compared with uniformly doped GaN, primarily by mitigating tensile stress [17]. While several studies have examined surface morphology, dislocation density, and electrical properties of δ‑doped GaN on sapphire or silicon [21–23], few have directly correlated luminescence efficiency with strain release in InGaN/GaN MQWs—an essential link for device optimization. Indirect strain measurements, such as applying mechanical pressure to modulate piezoelectric fields, remain the standard approach to evaluate internal strain because direct measurement without sample damage is challenging [24–27]. In all cases, photoluminescence spectra are indispensable for assessing strain‑related device performance.

In this work, we deposit InGaN/GaN MQW LED structures on crystalline silicon substrates. We compare uniformly Si‑doped n‑type GaN layers with periodic Si δ‑doped layers. Additionally, we introduce thin AlGaN (compressive) or InGaN (tensile) spacer layers on top of the uniformly doped n‑type GaN to further probe the influence of compressive versus tensile stress on device performance. Temperature‑dependent steady‑state (SS) PL and time‑resolved (TR) PL spectra provide the relative PL efficiencies and recombination lifetimes for each sample, enabling a systematic analysis of strain‑controlled recombination mechanisms.

Methods

As illustrated in Figure 1, the epitaxial growth of InGaN/GaN MQWs was performed by metal‑organic chemical vapor deposition (MOCVD) following the protocol reported in Ref. [17]. The growth sequence on a Si (111) substrate comprised a 100 nm AlN nucleation layer, a 660 nm linearly graded AlGaN buffer, and a 200 nm nominally undoped GaN buffer, grown at 1060 °C, 1060 °C, and 1020 °C, respectively. For samples S1, S3, and S4, a 1 µm Si‑uniformly doped n‑type GaN layer (≈10¹⁸ cm⁻³) was deposited on the buffer. In samples S3 and S4, a 20 nm InGaN spacer (≈10 at % In) or a 20 nm AlGaN spacer (≈20 at % Al) was inserted after the n‑type layer. Sample S2 employed 20 periods of Si δ‑doping followed by 50 nm nominally undoped GaN, achieving a total thickness of 1 µm. Subsequently, six periods of InGaN/GaN QWs were grown at 800 °C, with an indium composition of ~22 at % and an average well/barrier thickness of 2.4 nm/10 nm. A 220 nm Mg‑doped p‑type GaN layer was finally grown at 1020 °C.

SSPL measurements utilized a Zolix‑750 system with a 10 mW, 377 nm pulsed laser excitation and an ANDOR Newton CCD (0.09 nm resolution). TRPL decay curves were recorded using a time‑correlated single‑photon counting system over 10–300 K.

Figure 1. Structure of the MQW LED samples on Si substrates. S1, S3, and S4 contain a 1 µm uniformly Si‑doped n‑type GaN layer. S3 incorporates a 20 nm InGaN spacer; S4 incorporates a 20 nm AlGaN spacer. S2 consists of 20 Si δ‑doping planes separated by 50 nm undoped GaN, totaling 1 µm.

Results and Discussion

Figure 2 displays the SSPL spectra of all four MQW samples at 10 K. The spectra exhibit Fabry–Pérot oscillations—typical for GaN LEDs on Si due to the high reflectivity at the buffer/Si interface [28–30]. After deconvolution (Gaussian PL peak multiplied by 1 + A cos(4πnd/λ)), the intrinsic PL peaks are extracted. Sample S1 shows a sharp green peak at 531 nm, consistent with the bandgap of ~22 at % InGaN. The Si δ‑doped sample S2 is red‑shifted to 579 nm, whereas the InGaN‑spacer sample S3 is slightly blue‑shifted to 517 nm, and the AlGaN‑spacer sample S4 is red‑shifted to 545 nm. These shifts confirm that tensile stress on the MQW interfaces blue‑shifts the emission, while strain‑release mechanisms (δ‑doping, AlGaN insertion) red‑shift it. The strain‑release effect of Si δ‑doping is notably stronger than that of a single spacer layer.

Figure 2. SSPL spectra of S1–S4 excited by a 377 nm laser at 10 K. The inset shows the deconvolution of the original signal for S1.

To probe recombination dynamics, TRPL decays were recorded at the peak wavelengths of each sample across 10–300 K. Figure 3 presents the decay curves for S1 at representative temperatures. Although a single exponential model (I(t)=I₀e^−t/τ) is a simplification—multiple recombination centers often necessitate multi‑exponential fits—it yields an average lifetime τ that captures the overall PL dynamics. The resulting lifetimes for S1–S4 are plotted in Figure 4a, and the corresponding recombination rates k=1/τ are shown in Figure 4b. The tensile‑stress‑released samples S2 and S4 exhibit lower k values across all temperatures, indicating reduced total recombination. Moreover, the temperature dependence of k is steeper for these samples, reflecting the dominant role of non‑radiative recombination at elevated temperatures. By combining k with the relative PL efficiency η (defined as the ratio of PL intensity at temperature T to that at 0 K), the radiative (k_r=kη) and non‑radiative (k_nr=k(1−η)) components are extracted, as illustrated for S2 in Figure 5b.

Figure 3. TRPL decay curves for S1 at 10 K, 100 K, and 300 K.

Figure 5a shows the temperature dependence of the relative PL efficiencies. S2 and S4 achieve efficiencies up to 17 %, far exceeding the 2.5 % of the regular S1 and the 1.6 % of the InGaN‑spacer S3. Since PL intensity reflects only radiative recombination, η directly measures the fraction k_r/k. The extracted k_r and k_nr for all samples are summarized in Figures 6a and 6b. At 300 K, the non‑radiative rates for S2 and S4 are (2.5–2.8)×10⁻² s⁻¹, noticeably lower than those for S1 (3.6×10⁻² s⁻¹) and S3 (4.7×10⁻² s⁻¹). This confirms that strain release suppresses defect‑mediated non‑radiative pathways. Conversely, the radiative rates for S2 and S4 remain high and temperature‑stable (≈5.7–5.8×10⁻³ s⁻¹ at 300 K) compared to S1 (9×10⁻⁴ s⁻¹) and S3 (7×10⁻⁴ s⁻¹). The stability of k_r indicates that deep radiative centers within the InGaN wells dominate the recombination, whereas shallow interface states—promoted by tensile stress—are suppressed in the strain‑released structures.

Figure 6. (a) Non‑radiative recombination rates vs. temperature for S1–S4. (b) Radiative recombination rates vs. temperature for S1–S4.

Conclusions

Temperature‑dependent SSPL and TRPL studies of InGaN/GaN MQWs on silicon reveal that incorporating Si δ‑doping or AlGaN spacers significantly reduces total recombination rates and boosts PL efficiency to 17 %, compared with 2.5 % for the regular structure and 1.6 % for the InGaN‑spacer variant. The improvement stems primarily from lower non‑radiative rates—attributed to fewer dislocations and cracks—and from more stable, higher radiative rates, reflecting the dominance of deep radiative centers in the InGaN wells when tensile stress is alleviated. These findings provide a clear pathway for optimizing MQW LED performance on silicon by engineering strain through δ‑doping or compositional spacers.

Abbreviations

IQE

Internal quantum efficiency

LD

Laser diode

LED

Light‑emitting diode

MQW

Multiple quantum well

PL

Photoluminescence

SSPL

Steady‑state photoluminescence

TRPL

Time‑resolved photoluminescence