1. INTRODUCTION

Energy-efficient and low-cost silicon photonics technologies for CMOS-compatible optical interconnects are a very attractive solution for high-capacity, high-speed data links in data centers and supercomputers. In this context, quantum dot (QD) laser diodes, already considered the most promising lasers for optical communications [1], are attracting strong interest for direct epitaxial growth on silicon. This monolithic integration [24" target="_self" style="display: inline;">–4] is the ultimate solution to minimize the complexity and cost of the fabrication process and improve the integration density of the silicon photonic integrated circuit. However, the mismatch in the lattice constant and thermal expansion coefficients causes the generation of threading dislocations (TDs), which act as non-radiative recombination centers and degrade the laser performance. Despite the significant progress made recently [5–8" target="_self" style="display: inline;">–8], the achievable TD density is still around 105−106cm−2. At such high densities, quantum well (QW) lasers either show very poor performances or do not even lase [9]. On the contrary, QDs have proved to be more tolerant to TDs, due to 3D carrier localization within the QDs [10,11].

Apart from this feature, QD lasers on Si share several similarities with ordinary QD laser diodes on GaAs. Among these, we cite the double-state lasing from both ground state (GS) and excited state (ES) and the fast hole thermalization in the QD confined states of the valence band [12,13]. A detrimental effect related to double-state emission is the GS power quenching taking place above the ES threshold [1416" target="_self" style="display: inline;">–16]. The hole thermalization, which increases the laser threshold and reduces the differential gain, is usually cured via p-type modulation doping [1719" target="_self" style="display: inline;">–19], which has been recently employed also for QD lasers on silicon [20]. Experimental results have demonstrated that an optimum doping level exists [20], as an excessive doping can degrade the GS optical power and the threshold current. This effect is a matter of debate and has been attributed to different phenomena such as the enhanced free carrier absorption [18] or the increase in the carrier scattering rate and defect-assisted recombination with the doping level [20].

In order to devise a proper strategy to overcome these limitations, modeling can be an instrumental platform to achieve deeper understanding of the device operation. In this context, most of the literature relies on (often ambipolar) lumped rate equations describing the interaction of carriers and photons in each energy state. Even though such approaches provide an appropriate picture of the intersubband interactions, they reduce the model of the separate-confinement heterostructure (SCH) to phenomenological capture terms [11,14], neglecting carrier transport and electrostatic effects. This paper aims to fill this gap by investigating the role of transport and electrostatics in the peculiar features of QD-on-Si lasers, with the help of a quantum-corrected physics-based semiclassical model [2123" target="_self" style="display: inline;">–23]. The model is applied to characterize the steady-state performance of QD-on-Si lasers with InAs/GaAs self-assembled QDs emitting at 1.3 μm, at ambient temperature. Specifically, we study the impact of TDs on laser power versus current characteristics, the role of carriers transport in the competition between GS and ES lasing, and the physics behind the existence of an optimum p-type modulation doping level that minimizes the laser threshold current.

2. METHOD

2.1 A. Device under Study

A schematic drawing of the epitaxial structure of the investigated QD lasers is illustrated in Fig. 1. The structure is similar to others reported in the literature [24,25]. In these devices, the layers grown immediately above the silicon substrate act as buffers and dislocation filters reducing the density of TDs in the n-type cladding and SCH region. The n-type electrical contact is positioned on top of a GaAs n-doped buffer layer, thus avoiding carrier injection through the much more defective layers located below. For this reason, we have not simulated transport through the silicon substrate and buffer and we have assumed that electrons are directly injected through the 500 nm n-doped GaAs layer. The active region includes five layers of InAs QDs, each grown in a 11.5-nm-thick In0.15Ga0.85As QW, thus forming a dot-in-a-well (DWELL) structure. In each of the 37.5-nm-thick GaAs spacers separating a DWELL layer from another, the central 10-nm-thick region can either be intrinsic or p-type modulation doped.

Fig. 1. Schematic representation of the epitaxial structure of the studied QD lasers, similar to those in Refs. [24,25]. The growth direction is from the bottom to the top.

下载图片 查看所有图片

2.2 B. Model

The simulation framework consists of an in-house developed Poisson-drift-diffusion system augmented with carrier and photon rate equations of the localized states. The code requires as input a detailed description of the layer stack (material composition, doping) and QD characteristic parameters listed in Table 1. In this work, these parameters have been taken from our previous work [14] or extracted from literature experimental data representative of QD lasers on silicon.

Figure 2 shows the calculated energy band diagram under thermodynamic equilibrium for the device sketched in Fig. 1. The quantum-corrected carrier continuity equations are formulated within a drift-diffusion-Poisson formalism as where x is the growth direction of Fig. 1; Jn is the electron current density, UB is the net band-to-band recombination rate including Shockley–Read–Hall (SRH) and radiative processes, and N is the net doping. Rn,CAPB→WL,i and Rn,ESCWL,i→B are the capture and escape rates between the barrier and the i-th QD layer (located in the node xi), respectively. For the sake of brevity, drift-diffusion and carrier rate equations are reported here for electrons only, but a similar set of equations is implemented also for holes.

Fig. 2. Band diagram at thermodynamic equilibrium, with the conduction band (blue), the valence band (red), and the Fermi level (dashed, black). The dotted, vertical lines delimit the SCH region.

下载图片 查看所有图片

Figure 3 illustrates the model employed for the dynamics in each QD layer. The rate equations governing the electron dynamics in the ith QD layer are where Uk,i accounts for SRH and radiative recombination and Rstk,i is the stimulated recombination rate. Rn,CAPk,i→k−1,i and Rn,ESCk−1,i→k,i are the capture and escape rates between the kth and (k−1)th states with k=B,WL,ES and k−1=WL,ES,GS, computed as in Refs. [21,27], where WL indicates the wetting layer state including both the 2D well states and the upper continuum of the QDs. The capture rate depends on the capture time (τn,CAPk→k−1) from the state with higher energy (k) to the state with lower energy (k−1), whereas the escape rate is dependent on the escape time that leads the system to equilibrium in absence of any injection current or recombination rate [21]. These equations put in relation the 3D carrier volume densities (n,p) with the electrostatic potential ϕ and the sheet carrier density (nk,i,pk,i) of the kth QD energy level of the ith QD layer.

Fig. 3. Schematic representation of the QD energy states and intersubband transitions.

下载图片 查看所有图片

The net recombination rate Uk,i is given by Uk,i=Uradk,i+USRHk,i. Here, Uradk,i is the net radiative recombination rate modeled as where nk0,i and pk0,i are the electron and hole sheet densities in the kth energy level at thermal equilibrium, respectively, and Bradk is a radiative recombination coefficient related to the radiative time τradk [21]. The net SRH recombination rate is modeled as where τn,SRHk and τp,SRHk are SRH lifetimes for electrons and holes, respectively. The SRH lifetime in the barrier and DWELL (k=B,WL) is reduced by the presence of TDs according to [28,29] with τn(p),SRH0 being the typical (doping-dependent) lifetime of dislocation free GaAs [30], and TDD the TD density. The WL stimulated emission rate is assumed to be negligible. The GS and ES stimulated emission rates are given by where k=ES,GS, Sk is the photon density emitted by the GS or ES (k=GS,ES), vg is the group velocity, and ρn(p)k,i is the occupation probability of electrons (holes) kth energy level of the ith QD layer. The coefficient G0k is the saturated material gain (i.e., the material gain if the electron and hole kth state was fully filled) of the QD layer. Γi is the optical confinement factor of the TE fundamental mode in the ith QD layer and it has been computed from the fundamental TE mode profile of the epitaxial structure with no optical gain. The rate equation for the photon density Sk is governed by where Rspk is the spontaneous emission rate, given by Rspk=∑iUradk,i. Gkmod is the GS (ES) modal gain, βsp is the spontaneous emission factor, and τp is the photon lifetime [31], which accounts for the intrinsic and mirror loss. The GS and ES modal gains are given by with k=ES,GS. The coefficient G0GS was estimated to be ∼433cm−1 through a fitting of the measured modal gain versus current reported in Ref. [24] (see Fig. 4), resulting in G0GS∑iΓi∼52.4cm−1, where Γi∼2%. The coefficient G0ES has been set to 1.8G0GS to account for the ES degeneracy. Unless otherwise specified, the main simulation parameters employed in this work are summarized in Table 1. The separation between GS and ES energy levels is taken from Ref. [14] and corresponds to a difference in GS and ES recombination energy of 55 meV similar to the one reported in Ref. [16]. The QD shape, composition, and size as well as the InGaAs DWELL thickness significantly influence the QD electronic structure and therefore the escape rates out of the QD states. The less carriers escape out of the GS and ES, the less they are captured by the TDs in the WL. Therefore, the DWELL and QD design provides further degrees of freedom for device optimization, e.g., to improve the insensitivity of QDs to threading dislocations. The values of carrier capture and relaxation times are in line with Ref. [26] where ultra-fast hole dynamics and picosecond or subpicosecond electron dynamics have been measured.

Fig. 4. Calculated GS modal gain versus current density for different levels of TDD and experimental gain (circles) from Ref. [24].

下载图片 查看所有图片

All the simulations assume isothermal conditions and ambient temperature.

3. RESULTS AND DISCUSSION

3.1 A. Impact of Threading Dislocations

We investigate the impact of SRH recombination due to TDs, both in the DWELL and barrier layers. Figure 4 shows the modal gain of the device under study for different TD density. In the TDD range 5×106−107cm−2, the calculated gain is similar to the experimental one reported in Ref. [24].

For a qualitative comparison with experimental results, we consider a Fabry–Perot laser with loss of αm=8.6cm−1 (corresponding to a Fabry–Perot cavity length L=1325μm), to guarantee lasing on the GS only [24].

To disentangle the impact of TDs in barrier and QW layers, Figs. 5(a) and 5(c) show simulation results when TDs are present in the barrier only, whereas Figs. 5(b) and 5(d) consider the more realistic situation of similar TD density in barrier and DWELL layers. The GS threshold current density [Fig. 5(a)] and the light–current characteristics [Fig. 5(b)] are marginally affected by TD-induced lifetime degradation in the barrier layers.

Fig. 5. (a) GS threshold current density and (c) optical power as a function of TDDbulk, for fixed DWELL SRH lifetime corresponding to TDDWL=105cm−2. The solid lines are almost overlapped. (b) GS threshold current density and slope efficiency and (d) optical power as a function of TDD in the barrier and DWELL layers (TDDWL=TDDbulk).

下载图片 查看所有图片

On the other hand, TD-mediated recombination in the DWELL affects the laser performance significantly. Figure 5(b)analyzes the threshold current density and slope differential efficiency as a function of TDD, taking into account SRH lifetime reduction according to Eq. (6) both in barrier and DWELL layers. The slope efficiency is defined as (q/ℏω)·(dP/dI), with ω being the GS emission angular frequency, and it has been evaluated just above laser threshold. In this case, as the TD density in the DWELLs increases, the effective injection efficiency into the QDs diminishes, causing increased threshold current and reduced slope efficiency. This is also evident from the light–current characteristics in Fig. 5(d) calculated for different TDD values. Therefore, the TDs in the DWELL layers are those responsible for the degradation of the laser performance. Similar scaling of threshold current with TD density has been observed experimentally in Ref. [32], where devices of different generations having TDD density in the same range of Figs. 5(b) and 5(d), have been compared. These results also support the hypothesis that the degradation mechanisms limiting the long-term reliability could be correlated to a gradual propagation of TDs toward the quantum wells of the device active region [33].

3.2 B. Impact of Carrier Transport on Double-State Lasing and GS Power Quenching

In this section we discuss the influence on the CW performance of the asymmetry in the transport of electrons and holes across the barrier, owing to the different electron and hole low-field mobility in the GaAs layers. We demonstrate that this asymmetry is at the basis of the GS power quenching after the ES lasing threshold. For this purpose, we have set the mirror loss to αm=26.6cm−1 (corresponding to L=428μm). In this case, dual-state emission occurs at a reasonable ES threshold current. All results reported in this section are calculated for TDD=5×106cm−2.

Figure 6(a) shows the optical power emitted on the GS (solid) and ES (dotted) when the mobilities of electrons and holes in the SCH region are set to conventional room temperature values, corresponding to μn=8500cm2/(V·s) and μp=350cm2/(V·s). In this case, the GS power decreases as the current overcomes the ES threshold and it is completely quenched at sufficiently high currents. To highlight the impact of the unbalance of electron and hole mobility, Fig. 6(b) shows the GS (solid) and ES (dotted) optical power by forcing μn=μp=8500cm2/(V·s). In this case, the GS power is not quenched when the ES turns on, demonstrating that the origin of quenching can be explained by the different mobilities. The same conclusions can be drawn forcing μn=μp=350cm2/(V·s) (not reported in this work), which only causes a larger separation between GS and ES threshold currents.

Fig. 6. GS (solid) and ES (dotted) optical power with (a) μn=8500cm2/(V·s) and μp=350cm2/(V·s) and (b) μn=μp=8500cm2/(V·s) in the SCH region.

下载图片 查看所有图片

The asymmetry in the transport of electrons and holes is mirrored into the rate at which carriers are filling the various QD layers. This is emphasized in Fig. 7(a), showing the net capture rate of carriers from the bulk states to the WL with μn=8500cm2/(V·s) and μp=350cm2/(V·s). Each color corresponds to a different QD layer, with layer 1 (5) being the closest to the p-contact (n-contact). Layers located farthest from the p-contact are filled by holes at a smaller rate. On the contrary, if electron and hole mobilities are equal, holes are more evenly distributed and all the QD layers are filled at a similar rate, which increases linearly with the injected current. This is presented in Fig. 7(b), reporting the net capture rate of carriers from the bulk states to the WL with μn=μp=8500cm2/(V·s). This last case is in line with the exciton rate equation approach that, neglecting carrier transport and hole dynamics, was initially used to model double-state lasing in QDs [34]. Indeed the power–current characteristics in Fig. 6(b) well reproduce the results one would get in the exciton approximation [34], which is unable to reproduce the GS quenching. On the other hand, several experiments have shown that double-state lasing is often characterized by the GS quenching, as reproduced in Fig. 6(a) by our transport model with realistic material parameters. This provides a physical background for the results presented in Ref. [14] based on a phenomenological fitting parameter that was ad hoc introduced to mimic the asymmetric electron/hole transport. It is worth noticing that the onset of the GS quenching is characterized by an optical threshold. Varying the mirror loss in the asymmetric transport case, we have found that the reduction of the GS power after the ES threshold is observable only for αm>21cm−1. On the contrary, for equal electron and hole mobilities, the GS power quenching never occurs whatever the value of αm is.

Fig. 7. Net capture rate from the bulk states to the WL with (a) μn=8500cm2/(V·s) and μp=350cm2/(V·s) and (b) μn=μp=8500cm2/(V·s) in the SCH region. Layer 1 (5) is the closest to the p-contact (n-contact).

下载图片 查看所有图片

In order to achieve deeper insight into the dual-state competition, it is convenient to inspect the overall, separate contributions of electrons (GGSmod,n) and holes (GGSmod,p) to the GS modal gain [see Eq. (9)]. For the sake of brevity, we only discuss explicitly the case with μn=8500cm2/(V·s) and μp=350cm2/(V·s). As the injected current grows between the GS and ES threshold currents, GGSmod,n decreases [see Fig. 8(a)] due to the spectral hole burning of GS electrons that are injected from the ES at a slower rate than they recombine. The electron gain compression due to spectral hole burning is compensated by the growth of hole gain to keep the lasing threshold gain. Indeed GGSmod,p increases [see Fig. 8(b)], such that the GS modal gain remains clamped. The dashed curves in Figs. 8(a) and 8(b) quantify how electrons and holes of the various layers contribute to the overall gain. Only the GS holes of the two layers close to the p-side contribute with increasing gain, compensating for the gain reduction in all the other layers. In contrast, the hole gain contribution of layers far from p-side reduces as power increases because the slow hole transport prevents the refurbishment of holes lost due to stimulated emission. On the other hand, in the case of symmetric mobility μn=μp=8500cm2/(V·s), we have found that the hole contribution is always increasing for any current injection. At the ES threshold current, the ES stimulated emission turns on and starts recombining the ES electrons and holes. This mechanism competes with the relaxation rate from the ES to the GS. Now the contribution of the GS holes is not enough to compensate for the electron spectral hole burning. As a consequence, the GS electron population has to increase such that GGSmod,n can maintain the GS threshold condition. Then, the net transition rate of electrons from the ES to the GS decreases due to Fermi blocking; this rate balances the GS stimulated emission rate until the GS optical power is eventually quenched [14]. These trends are also supported by the electron and hole GS occupation probability of the various QD layers: as shown in Figs. 8(c) and 8(d), the GS is almost completely filled with electrons, whereas it is populated scarcely and unevenly by holes.

Fig. 8. Contribution of (a) electrons and (b) holes to the GS modal gain: solid line is the overall contribution, whereas colored dashed lines are the contribution of the different layers (color legend is the same as in Fig. 6). Vertical lines indicate GS and ES threshold currents. (c) GS electrons and (d) holes occupation probability. The mobility of electrons and holes in the SCH region is μn=8500cm2/(V·s) and μp=350cm2/(V·s). Layer 1 (5) is the closest to the p-contact (n-contact).

下载图片 查看所有图片

In summary, the reduction with increasing current of the hole gain contribution in the layers closer to the n-side is caused by the slow hole transport and it is therefore an additional mechanism, besides spectral hole burning, causing gain compression. Spatial hole burning or other gain compression effects are not directly included in the model, but could be taken into account by introducing a phenomenological compression factor parameter [31]. The accurate quantification of the effective gain compression coefficient requires, however, the simulation of the laser intensity modulation response which is out of the scope of this work [35].

3.3 C. Impact of p-Type Modulation Doping

In order to analyze the impact of modulation doping on dual-state emission, we set the mirror loss to αm=26.6cm−1. All results reported in this section are calculated for TDD=5×106cm−2. In this context, the aim of p-type doping is reducing the GS threshold current and increasing the ES one such that the GS power is optimized. To focus on the role of doping in terms of electrostatics and transport, in the analysis we intentionally neglect possible additional loss due to free carrier absorption caused by the extra holes in the barrier. For the sake of the following analysis, it is worth noticing that the SRH lifetimes calculated according to Eq. (6), accounting for the doping dependence in dislocation free GaAs from Ref. [30], are dominated by the effect of the TDs.

Modulation doping is implemented by 10-nm-thick doped layers, with doping density NA, placed as in Fig. 1. Figure 9 shows the optical power emitted on the GS (solid) and ES (dotted) with NA equal to 0, 5×1017cm−3, and 3×1018cm−3, respectively. These doping levels correspond to 0, 10, and 60 extra holes per QD. Adding 10 extra holes per QD is beneficial, as the GS power quenching is cured and the GS threshold current is reduced. However, 60 extra holes turn out to be excessive, as the previous benefits are partially undermined. Figure 10(a) shows the GS (blue) and ES (red) threshold current density, JthGS and JthES respectively, as functions of NA. An optimum density of 5×1017cm−3 exists, as this concentration minimizes JthGS and maximizes JthES.

Fig. 9. GS (solid) and ES (dotted) optical power with (a) no p-type modulation doping and a p-type modulation doping of (b) 5×1017cm−3 and (c) 30×1017cm−3.

下载图片 查看所有图片

Fig. 10. (a) GS (blue) and ES (red) threshold current density as functions of the p-type modulation doping density. (b) Total radiative and SRH recombination rates as functions of p-type modulation doping density calculated at the JthGS values in (a).

下载图片 查看所有图片

These results, showing an optimal doping density to minimize the GS threshold current, are consistent with the experiments in Ref. [20]. We have also verified that such optimum value is almost insensitive to the TDD value. In Ref. [20], it was speculated that the optimum results from a balance between the increase of gain enabled by doping and the increase of doping-induced defects. However, the analysis of the integrated radiative and non-radiative recombination rates as a function of NA reported in Fig. 10(b) demonstrates that the GS threshold minimum is strongly correlated to the doping dependence of the radiative recombination rate, whereas the influence of doping on the SRH recombination rate turns to be marginal. The optimum appears as an intrinsic consequence of a non-monotonic dependence of the modal gain on doping density: the increase of the hole modal gain contribution with NA, due to the larger density of holes, is in fact counteracted at higher NA values by a decrease of the electron modal gain contribution due to electrostatic effects.

We report in Fig. 11(a) the modal gain versus current density for three different doping levels NA and in Fig. 11(b) the separate contributions of electrons and holes to the modal gain. While the hole contribution to the gain increases as expected for p-doping, we observe a decrease of the electron contribution to the GS gain that, to the best of our knowledge, has not been previously reported. The balance between the increased hole and reduced electron contributions makes the doped samples exhibit a higher or lower total gain—at fixed current injection—with respect to the undoped one. Long devices with low threshold gain (e.g., lower than 15cm−1 in this example) could indeed not benefit too much from doping, since their threshold current would be higher in the doped case. This result may explain the anomalous increase of threshold current with doping recently reported in Ref. [36].

Fig. 11. (a) GS modal gain versus current density and (b) holes (GGSmod,p, dashed) and electrons (GGSmod,n, solid) contributions to the modal gain.

下载图片 查看所有图片

Figure 12(a) shows the electron and hole contributions to the GS modal gain, GGSmod,n and GGSmod,p, respectively, as a function of NA, at constant current density J=580A/cm2. Since this value is smaller than the minimum GS threshold current density (see Fig. 10), the GS modal gain is not clamped for any of the considered doping levels. In addition, Fig. 12(b) shows the resulting GS modal gain. At low doping densities, the increase in GGSmod,p outweighs the decrease in GGSmod,n, such that the GS modal gain grows. At NA=5×1017cm−3, the electron and hole contributions balance each other, maximizing the GS modal gain. At higher doping concentrations, the drop in GGSmod,n is steeper compared to the growth in GGSmod,p and the GS modal gain is diminished. Therefore, the existence of an optimum value appears to stem from the impact of p-type modulation doping on the GS modal gain at a given injected current.

Fig. 12. (a) Contribution of electrons (blue) and holes (red) to the GS modal gain at J=580A/cm2 versus p-doping density and (b) corresponding GS modal gain.

下载图片 查看所有图片

To shed further light on this mechanism, we report in Fig. 13(a) the conduction band energy (solid) and electron quasi-Fermi level (dashed) for the bulk states of the SCH region at J=580A/cm2 and for the same NA values in Fig. 9. Notice that the multiple peaks observed in the QD stack region arise because of the electric field modulation caused by the depleted thin doped regions. As the doping concentration increases, the energy barrier which electrons injected from the n-contact have to overcome to reach the QD active layer region steadily rises. Concurrently, the electron quasi-Fermi level is pushed deeper in the energy gap. As a consequence, the bulk electrons in the active region diminish progressively, becoming unavailable to be captured by the QDs. Therefore, GGSmod,n in Fig. 12(a) steadily decreases as the doping increases. On the other hand, the valence band (solid) and hole quasi-Fermi level (dashed) for the bulk states of the SCH region at J=580A/cm2 are shown in Fig. 13(b). The energy barrier experienced by the holes injected from the p-contact is not significantly affected by NA. Therefore, the change of hole density in the active region is directly related to the sole variation in the concentration of p-type dopants, which act as local source of extra holes. However, as doping grows holes almost entirely fill the GS, and the GGSmod,p in Fig. 12(a) gradually saturates.

Fig. 13. (a) Conduction band (solid) and electron quasi-Fermi level (dashed) for the bulk states of the SCH region at J=580A/cm2. (b) Valence band (solid) and hole quasi-Fermi level (dashed) for the bulk states of the SCH region at J=580A/cm2.

下载图片 查看所有图片

To corroborate this interpretation, we report in Fig. 14 for each QD layer the net capture rate of carriers from the bulk states to the WL (denoted by RNETB→WL) at J=580A/cm2. In the undoped case, the farther the QD layer is from the p-contact, the smaller RNETB→WL is. Indeed, while diffusing throughout the epitaxial structure upon injection from the p-contact, holes are progressively captured by the QD layers. As NA grows, the electrostatic deformation of the band diagram reflects on RNETB→WL. For doping levels higher than 1017cm−3, the closer the QD layer is to the n-contact, the larger RNETB→WL is. This signifies that most of the electrons injected from the n-contact are blocked by the energy barrier induced by the doping.

Fig. 14. Net capture rate from the bulk states to the WL at J=580A/cm2 for each layer of QDs.

下载图片 查看所有图片

The influence of doping on carrier injection also has a significant influence on the dependence of the SRH recombination rate on injection conditions. As depicted in Fig. 15, at low voltages doping basically mitigates the (TD dominated) non-radiative recombination [22]. However, as carrier injection increases, non-radiative recombination in the doped samples starts to grow with a higher rate than it does in the undoped sample, owing to the higher hole density available in the DWELLs. The increase is larger at the intermediate doping level than at the highest one, because in the latter case electron injection from the n-contact is hindered by the energy barrier induced by the depleted delta-doped layers. Despite the marked difference in the voltage dependence of the SRH recombination rate, all the three samples present a similar amount of SRH recombination at the lasing thresholds, highlighting that the most relevant factor for the GS threshold minimization is the change of the radiative recombination rate, and therefore of the GS modal gain, with doping density.

Fig. 15. Total SRH recombination rate versus voltage at three different doping levels. The vertical dashed lines indicate the voltage value corresponding to the lasing threshold.

下载图片 查看所有图片

4. CONCLUSIONS

This work has presented an investigation of the mechanisms limiting the performance of QD lasers epitaxially grown on silicon, on the basis of a transport model including QD carriers and photon rate equations. Overcoming the traditional lumped formulations based on rate equations, our approach can simulate the PIV characteristics of a dual-state lasing device starting from the description of materials, doping, and geometry. For bulk materials, consolidated measurement-based properties (e.g., mobility, bandgap, density of states) are available. For the QD material, electronic structure and characteristic carrier dynamics time constants can be extracted from literature devices representative of QD-on-Si technology, as done here, or calculated from first principle approaches. This enables applying the tool in the computer-aided design of the epitaxial structure, as a supporting tool in prototyping campaigns.

It is demonstrated how carrier transport and electrostatic effects are essential for catching and explaining the main static characteristics of these lasers. The analysis of the impact of threading dislocations in the barrier and DWELL layers on the lasing performance pinpoints a dominant effect of the TDs in the DWELL layers, supporting the hypothesis that long-term reliability issues suffered by these layers can be correlated to a thermally activated climbing of the TDs toward the active region. When accounting for transport mechanisms, double-state lasing and GS power reduction naturally emerge as a consequence of the inherently asymmetric transport of electrons and holes, explaining the reason why conventional excitonic models cannot reproduce such an effect, and providing justification for the use in lumped rate equation models of phenomenological fitting parameters mimicking this effect. Finally, the physics behind p-type modulation doping and its effectiveness to minimize the GS threshold and maximize the ES/GS thresholds’ ratio has been thoroughly investigated. The existence of an optimum doping level results from the competition between the doping-induced increase of hole modal gain contribution and the decrease of electron modal gain contribution, and therefore depends also on the threshold gain of the specific sample under study. If non-radiative carrier lifetime is dominated by threading dislocations, the impact of doping on the non-radiative recombination rate turns to be marginal with respect to the impact on the modal gain.

The presented model does not account for self-heating effects. While this could be a worthy extension of the present work, we expect that the overall scope of the presented results, which focus on next-to-threshold operating conditions, remains valid from a qualitative standpoint. In support of this, we remark that two-state lasing and GS quenching have been reported for both pulsating and continuous wave laser operation.