Lasers, optical fibers and semiconductor photonic devices are generally regarded as the three major inventions in the field of optics in the twentieth century. Among them, semiconductor lasers play an important role in the whole laser field because of their small volume, light weight and easy integration^[1]. High power semiconductor lasers in the red band are widely used for pump sources, laser display, medical devices and laser lighting^{[2, 3]}. However, the catastrophic optical damage (COD) that often occurs at the laser cavity surface is a major problem that hinders its power improvement and long-term service reliability^[4-6]. This damage is due to the overload of laser power density and high cavity surface temperature, which causes the bandgap to shrink, and then promotes itself to absorb more photons, forming a positive feedback process, resulting in the melting and recrystallization of the cavity surface area^[7].

Moreover, the device will fail in the range of tens of nanoseconds. An effective method to prevent this problem is to make a nonabsorbent window (NAW) at the laser cavity surface. As a post-treatment process, impurity induced quantum well hybrid technology has attracted extensive attention because of its advantages, such as easy process, low equipment requirements and low manufacturing cost. This research has been developed since it was found that the diffusion of Si impurities can promote the mutual diffusion of quantum barrier and quantum well components in the 1980s^[8]. In 1985, Kaliski et al.^[9] conducted Si impurity induced quantum well intermixing on AlGaAs/GaAs superlattices. At 750 °C annealing temperature and 144 h annealing time, the blue shift of the emission spectrum peak of quantum well is up to 85 nm. In 2010, Cong et al.^[10] implanted InGaAs/AlAsSb double quantum well structure chip with Si ions, and obtained a blue shift of 160 nm under the condition of rapid thermal annealing (RTA) at 600 °C/1 min. There are only a few studies on Si impurity induced quantum well intermixing in China. In 2007, Peng^[11] of Hebei University of technology conducted Si impurity diffusion and Si ion implantation on GaAs/AlGaAs semiconductor lasers to induce quantum well intermixing. However, the p-type dopant in the epitaxial layer is Zn, which hinders the diffusion of Si, resulting in a small blue shift. In 2020, Liu^[12] performed secondary epitaxy of Si dielectric layer on the surface of InGaAs/AlGaAs quantum well semiconductor laser by means of MOCVD, and the devices displayed a distinct 52 nm blue shift under the condition of RTA 900 °C/180 s. A 93 nm blue shift was also obtained when annealed in tubular furnace at 825 °C/2 h, while the blue shift of the luminescence gain region under various experimental conditions after annealing can reach up to 65 nm. In regards to all of these doping methods, ion implantation^{[13, 14]} methods appear to be inferior because of their high cost and they are incompetent because of their inevitable damage to the lattice, which will irreversibly do harm to the later device fabrications. In addition, when growing Si dielectric layer induced quantum well intermixing^[15], the high temperature or long-time thermal annealing will not only damage the surface morphology of the epitaxial wafer^[16] but will also cause the blue shift of the wavelength of the gain luminescence region^[17]. Therefore, to solve these problems, the quantum well intermixing experiment is redesigned in this paper. The composited Si–Si₃N₄ dielectric layer is used to induce the quantum well intermixing of 680 nm semiconductor laser, through which the surface morphology of epitaxial wafer is not affected by thermal annealing at low temperature of 540−600 °C and at the same time the wavelength of the gain luminescence region basically does not change under the condition of ensuring a certain blue shift in the nonabsorbent window region. This approach will not only save the cost and improve the efficiency but will also effectively protect the sample, realizing a better intermixing affect.

For 680 nm single quantum well GaInP/Al_0.53GaInP red semiconductor laser, Al, Ga and In atoms occupy group III lattice, while P atoms occupy group V lattice. Generally speaking, in III–V compound semiconductors, the two defects with the largest diffusion rate are group III vacancies V_IIIand group III interstitial atoms I_III. The mutual diffusion of Al and Ga atoms is mainly considered in the process of component calculation and simulation because In atoms are large and have a slow diffusion rate. In the process of component interdiffusion caused by impurity induction, it is mainly carried out in the form of Frenkel defect pairs through Ⅴ_III and I_III^[18]:

1VGa↔VGa+(IAl+VAl)↔(VGa+IAl)+VAl↔VAl,

2IGa↔IGa+(VAl+IAl)↔(IGa+VAl)+IAl↔IAl.

The initial concentration of Al component is:

3C={Cw,|z|<d,Cb,|z|⩾d,

where C_w and C_b represent the component concentration of Al in the quantum well and quantum barrier, respectively. Taking the central position of the quantum well as the z-axis origin, the width of the quantum well is 2d.

Setting the diffusion length to L_d and after the completion of quantum well mixing, the expression of Al component concentration in quantum well and quantum barrier is as follows^[19]:

4C(z)=C1+(C2−C1)2[erfc(z+d2Ld)+erfc(−z+d2Ld)].

The width of the quantum well of the semiconductor laser epitaxial wafer used in this experiment is 7 nm, and the Al diffusion distance L_d is set to 0, 1, 2 and 3 nm. The Al component concentration can be calculated according to the diffusion distance after mixing, as shown in Fig. 1.

From this figure it can be seen that when there is no diffusion, the content of Al in the well and barrier is 0 and 0.53, respectively. Meanwhile, the distribution of Al atoms in the well barrier shows an obvious step shape. When the diffusion length is short as L_d = 1 nm for example, Al occupies a relativeley large portion in the well region so the quantum well region and quantum barrier region can maintain good heterojunction morphology. However, with the increase of diffusion length, when L_d = 2 nm, the Al component concentration in the well region gradually increases, and the Al component concentration at the interface between the well region and the barrier region gradually tends to be consistent. When the diffusion length L_d reaches 3 nm, the Al composition in the well and barrier tends to be the same, the obvious interface between the well region and the barrier region has not been observed, and the heterojunction state is difficult to maintain. This figure also shows that the quantum well mixing degree between the well region and the barrier region has been sufficient.

According to the variation of Al component concentration with diffusion distance calculated in the previous section—that is, after the diffusion distances L_d are 0, 1, 2 and 3 nm—, the corresponding values of Al component concentrations in the quantum well are 0, 0.007, 0.115 and 0.217, respectively. The variation of quantum well gain spectrum with Al component concentration is simulated in this section, as shown in Fig. 2.

It can be seen that when the diffusion distance is small—that is, the Al component changes little—, the gain spectrum peak basically has no blue shift and the value of gain has a small variation. However, when the diffusion distance gradually increases, the gain spectrum peak blue shift also increases, while the gain value decreases. A 40 nm of blue shift is observed when the diffusion distance is 2 nm and the gain value is 88.5% of the original value.

In group III–V semiconductors, the point defect concentration is usually much less than the group III atom concentration (about 2 × 10²² cm⁻³); that is, the mutual diffusion coefficient is mainly determined by the point defect concentration. The mutual diffusion coefficient is expressed as^[13]:

5DIII=f1DVIII[VIII]+f2DIIII[IIII],

where D_III is the mutual diffusion coefficient of group III atoms, f₁ and f₂ are constants containing information about crystal structure and group III lattice concentration.

Assuming that there are N lattice points in group III lattice and the vacancy occupation number isn, then the possible arrangement number is

6G=N!n!(N−n)!.

By using the Stirling formula, the Boltzmann entropy S can be written as:

7S=kBlnG=kBlnN!n!(N−n)!=kB[NknN−nlnn−(N−n)ln(N−n)],

and the free energy of the whole system can be expressed as:

8F=E−TS=F0+nEV−kBT[NknN−nlnn−(N−n)ln(N−n)],

where E_v is the energy required to form a vacancy, F₀ is the partial free energy independent of the number of vacancies n, T is the temperature. In this system, the equilibrium is determined by minimizing the free energy and by taking the derivative of F to n we get

9∂F∂n=EV+kBTlnnN−n=0,

10nN−n=exp(−EVkBT).

In this formula, the number of lattice points N is greater than vacancies n in the crystal. This can be simplified as:

11n=Nexp(−EVkBT).

Then, the normalized thermal equilibrium concentrations of group III vacancy and group III interstitial atoms are:

12[VIII]=exp(−EV/kBT),

13[IIII]=exp(−EI/kBT).

By introducing these two formulas, the expression of the mutual diffusion coefficient of group III atoms with temperature can be obtained:

14DIII=f1DVIIIexp(−EV/kBT)+f2DIIIIexp(−EI/kBT).

From the Fig. 3, it can be seen that the value of the relative diffusion coefficient D increases gradually with the increase of temperature. It is proven that the two coefficients are close to the exponential relationship. When the temperature is higher than 800 K, the value of relative diffusion coefficient D increases sharply. This figure also shows that high temperature annealing is a necessary process in quantum well intermixing technology^[20].

The 680 nm quantum well semiconductor laser used in this experiment was prepared on n-GaAs substrate by aix-200/4 system low-voltage MOCVD equipment. As shown in Fig. 4, from the bottom to top, the structure of the epitaxial wafer is: 200 nm GaAs buffer layer, 2000 nm Al_0.7GaInP confinement layer, 120 nm Al_0.53GaInP waveguide layer, 20 nm Al_0.53GaInP barrier, 7 nm Ga_0.2In_0.8P QW layer, 20 nm Al_0.53GaInP barrier layer, 120 nm Al_0.53GaInP waveguide layer, 1800 nm Al_0.7GaInP confinement layer, 100 nm GaInP cladding layer and 150 nm GaAs contact layer.

Before manufacturing the experiment, it is essential to test the PL spectrum and Electrochemical Cyclic Voltammetry (ECV) of the epitaxial wafer, which shows the key information of light intensity, peak wavelength and carrier concentration, and carrier distribution. The test results are shown in Figs. 5 and 6, in which the peak value of PL spectrum is 669 nm and the luminous intensity of the sample is relatively uniform. It can also be seen from the ECV test that the quality of the epitaxial wafer meets our expectations.

The lift-off technique process is adopted, as shown in Fig. 7.

The specific operation process is as follows. First, photolithography is carried out on the epitaxial wafer to form a gain luminescence region and nonabsorbent window region. Then, we sputter a layer of Si with sputtering equipment under the conditions of sputtering pressure of 1.3 Pa, sputtering power of 100 W, argon flow rate of 50 sccm and sputtering time of 30 min. By using the ellipsometry, the thickness and refractive index of Si dielectric layer were 100.85 nm and 3.4607, respectively. A whole epitaxial wafer was cut into 6 × 6 mm² pieces followed by using rtp-500 rapid annealing furnace to handle these pieces. The annealing temperature was set to 540, 560 and 580 °C, while the annealing temperature remained unchanged for 10 min. After annealing, we remove the Si dielectric layer with KOH solution and observe the sample surface under the condition of annealing at 580 °C/10 min with a metallographic microscope, as shown in Fig. 8.

From this figure it can be seen that the surface quality of the sample remained fine after annealing, while the PL spectra peak of Si covered region and non Si covered region were both 668.9 nm at these three temperatures. Taking the error of measurement into consideration, it can be considered that there is no blue shift. Fig. 9 shows the PL spectrum under the annealing condition of 580 °C/10 min.

It can be concluded that under this temperature range, quantum well intermixing could not happen if the condition is only simply thermal annealing^[21] or growth of a single Si dielectric layer.

First prepare the same epitaxial wafer and then use PECVD equipment to grow 100 nm Si₃N₄ after completing the process described in Section 3.2, as shown in Fig. 10.

Similarly, a wafer is cut into 6 × 6 mm² small standard pieces, which were consequently annealed under the condition of 540–600 °C/10 min with the temperature gradient of 20 °C. We found that there was basically no blue shift in the Si₃N₄ covered region, while the PL spectrum of Si–Si₃N₄ covered region is shown in Fig. 11.

It can be observed from this figure that with the gradual increase of annealing temperature, the wave peak of the spectrum gradually blue shifted. At 540 °C, the blue shift was 24 nm, while at 600 °C the blue shift reached 54.8 nm and the luminous intensity of all samples maintained more than 75% of the origin sample. This indicates that nonabsorbent window regions and gain luminescence regions were well realized.

An ECV test was carried out on the sample that was annealed at 580 °C/10 min, the result is shown in Fig. 12.

Compared with the ECV test result of the original sample, near 1.3 μm, the p-type doping concentration decreased while the n-type doping concentration increased. This indicates that Si impurities diffused into the sample and became n-type donors^[22].

To avoid damage caused to the quantum well epitaxial wafer caused by long annealing time or high annealing temperature, we conducted three periods of annealing and the temperature of each period was 580 °C and the annealing time was 10 min. After one annealing period was completed and the temperature was cooled to room temperature, we repeated the same experiment three times. We then tested the PL spectrum for each time, as shown in Fig. 13.

It can be found from this figure that the luminescence intensity did not decrease significantly with the increase of cyclic annealing times, and the blue shifts were 42.9, 45.7 and 48 nm respectively from one to three annealing times. This demonstrates that under the condition of 580 °C/10 min cyclic annealing, even if it can promote the increase of blue shift, the increase was limited.

A thin film stress tester was used to test the stress between dielectric layers, as shown in Table 1.

The stress of 100 nm Si₃N₄ on GaAs substrate is –1204.315 MPa and the stress on Si substrate is –755.056 MPa, which is also relatively large. The surface compressive stress of GaAs will contribute to the generation of more Ga vacancy defects^[23], and then produce a stronger quantum well intermixing effect. Therefore, it is speculated that the stress applied by Si₃N₄ made Si impurities diffuse into the epitaxial wafer during annealing, promoting the mutual diffusion of quantum well and quantum barrier components. Meanwhile, the single Si dielectric layer is difficult to diffuse into the epitaxial wafer due to its small diffusion coefficient in GaAs.

In this paper, we propose the use of Si–Si₃N₄ composited dielectric layers to induce quantum well intermixing of 680 nm red semiconductor lasers. Under the condition of low annealing temperature of 540–600 °C and annealing time of 10 min, the blue shift difference between nonabsorbent window region and gain luminescence region greater than 30 nm is well realized and the luminous intensities are also maintained at more than 75% of the original sample. Moreover, in this temperature range, only simple thermal annealing or growth of single Si dielectric layer will not cause quantum well intermixing, while the large stress exerted by Si₃N₄ will promote the diffusion of Si in GaAs during thermal annealing and induce quantum well intermixing. This method will not only ensure the good lattice quality of the epitaxial wafer surface but will also reduce the process difficulty, which is conducive to the later device fabrication and provides an effective scheme to improve the COD threshold of 680 nm band semiconductor lasers.

This work was supported by the National Natural Science Foundation of China (NNSFC) (Grant No.62174154).