A Review of the Reliability of Integrated IR Laser Diodes for Silicon Photonics

: With this review paper we provide an overview of the main degradation mechanisms that limit the long-term reliability of IR semiconductor lasers for silicon photonics applications. The discussion is focused on two types of laser diodes: heterogeneous III–V lasers bonded onto silicon-on-insulator wafers, and InAs quantum-dot lasers epitaxially grown on silicon. A compre-hensive analysis of the reliability-oriented literature published to date reveals that state-of-the-art heterogeneous laser sources share with conventional laser diodes their major epitaxy-related degradation processes, such as the generation of non-radiative recombination centers or dopant diffusion, while eliminating cleaved facets and exposed mirrors. The lifetime of InAs quantum dot lasers grown on silicon, whose development represents a fundamental step toward a fully epitaxial integration of future photonic integrated circuits, is strongly limited by the density of extended defects, mainly misﬁt dislocations, protruding into the active layer of the devices. The concentration of such defects, along with inefﬁcient carrier injection and excessive carrier overﬂow rates, promote recombination-enhanced degradation mechanisms that reduce the long-term reliability of these sources. The impact of these misﬁts can be largely eliminated with the inclusion of blocking layers. J.E.B., G.M., E.Z. and M.M.; visualization, M.B., C.D.S., J.N., C.S., J.E.B., G.M., E.Z. and M.M.; supervision, M.B., C.D.S., J.N., C.S., J.E.B., G.M., E.Z. and M.M..


Introduction
Silicon represents the most widely employed semiconductor material for the realizations of electronic devices and integrated circuits. Owing to its indirect bandgap, however, its role in optoelectronics is limited to applications that leverage its waveguiding and modulation properties, rather than its inefficient light emission, or that take advantage of it as a cheap and readily available substrate for the epitaxial growth of other alloys. This means that while high performance detectors and solar cells sensitive in the visible-to near-infrared (NIR) range can be manufactured through well-established Si processes adopted for large-scale microelectronics, the fabrication of light emitters, such as laser diodes and LEDs, has to rely on other semiconductor materials systems, depending on the wavelength of interest. This requirement pushed both the scientific community and companies to develop optimized III-V and II-VI processes and devices that allow the realization of a wide variety of optical sources and sensors. The wavelength of operation of these devices can range from the deep UV, by leveraging the AlGaN/GaN semiconductor material system [1], to the far infrared, for devices based on binary semiconductors such as GaAs, InSb, CdTe, GaP, InAs, and their alloys. While this approach represents the optimal path for the manufacturing of stand-alone components to be included in discrete-based systems, it severely limits the direct inclusion of optical emitters into Si-based electronic and photonic circuits [2], which would allow fabricating low-cost fully integrated electrophotonic circuits for a variety of silicon photonics applications. In fact, while efficient and reliable III-N visible LEDs can be grown on silicon substrates despite the presence of lattice mismatch-related extended defects [3][4][5], continuous-wave (CW) laser operation of IR semiconductor devices requires very-low defect densities to be present within the active region of the devices. Several strategies have been studied and implemented to achieve epitaxial integration of III-V materials on silicon [6][7][8], but none of these ensured high enough efficiency and scalability to promote their adoption for large-scale integration into Si-and SOI-based photonic and electronic circuits.
Modern PICs for telecom applications circumvent the intrinsic limitation of Si as an optical emitter either by relying on separate laser sources placed in the same package or fiber coupled to the photonic circuit [9], or by selectively bonding III-V laser sources, grown through a dedicated III-V process, onto the SOI substrate by means of low temperature wafer bonding techniques [10][11][12] (Figure 1). This approach leverages both the high quality of an optical source manufactured through an optimized process and the relatively low manufacturing costs of PICs manufactured through a conventional SOI process [13,14]. While this approach is beneficial from a reliability standpoint, since high semiconductor quality implies lower degradation rates and higher robustness against catastrophic optical damages, it is not ideal in terms of industrialization, due to the additional costs and manufacturing complexity associated with the required wafer bonding steps.
based systems, it severely limits the direct inclusion of optical emitters into Si-base tronic and photonic circuits [2], which would allow fabricating low-cost fully inte electro-photonic circuits for a variety of silicon photonics applications. In fact, whi cient and reliable III-N visible LEDs can be grown on silicon substrates despite th ence of lattice mismatch-related extended defects [3][4][5], continuous-wave (CW) las eration of IR semiconductor devices requires very-low defect densities to be p within the active region of the devices. Several strategies have been studied and mented to achieve epitaxial integration of III-V materials on silicon [6][7][8], but n these ensured high enough efficiency and scalability to promote their adoption for scale integration into Si-and SOI-based photonic and electronic circuits.
Modern PICs for telecom applications circumvent the intrinsic limitation of S optical emitter either by relying on separate laser sources placed in the same pack fiber coupled to the photonic circuit [9], or by selectively bonding III-V laser so grown through a dedicated III-V process, onto the SOI substrate by means of low te ature wafer bonding techniques [10][11][12] (Figure 1). This approach leverages both th quality of an optical source manufactured through an optimized process and the rela low manufacturing costs of PICs manufactured through a conventional SOI p [13,14]. While this approach is beneficial from a reliability standpoint, since high sem ductor quality implies lower degradation rates and higher robustness against catast optical damages, it is not ideal in terms of industrialization, due to the additiona and manufacturing complexity associated with the required wafer bonding steps. The most promising alternative to the heterogeneous integration of III-V las an SOI process is represented by the adoption of laser diodes based on epitaxial g of quantum-dots (QDs), rather than on multi-quantum-wells (MQWs) active region ure 2). Owing to their 3D carrier confinement properties [16], the zero-dimensional structures are more tolerant with respect to the presence of extended defects [17,18 allow the fabrication of reliable and efficient optical sources even in the presence o densities of dislocations, as in the case of the heteroepitaxial growth of III-V ma onto Si substrates. This fundamental property allowed 1.3 μm InAs quantum-dot epitaxially grown on Si with a threading dislocation density (TDD) of 10 8 cm −2 to rea lasing threshold and operate continuously, even at high temperature [19]. Once a degree of maturity is reached, these devices will allow a fully-epitaxial integration optical source into the SOI-based PIC. The most promising alternative to the heterogeneous integration of III-V laser into an SOI process is represented by the adoption of laser diodes based on epitaxial growth of quantum-dots (QDs), rather than on multi-quantum-wells (MQWs) active regions ( Figure 2). Owing to their 3D carrier confinement properties [16], the zero-dimensionality QD structures are more tolerant with respect to the presence of extended defects [17,18], and allow the fabrication of reliable and efficient optical sources even in the presence of high densities of dislocations, as in the case of the heteroepitaxial growth of III-V materials onto Si substrates. This fundamental property allowed 1.3 µm InAs quantum-dot lasers epitaxially grown on Si with a threading dislocation density (TDD) of 10 8 cm −2 to reach the lasing threshold and operate continuously, even at high temperature [19]. Once a higher degree of maturity is reached, these devices will allow a fully-epitaxial integration of the optical source into the SOI-based PIC. Considering that Si-based PICs offer (i) a very high degree of integration, due to the high refractive index contrast between Si and the surrounding dielectric, (ii) the possibility to integrate all the optical signal conditioning functions required by telecom applications, and (iii) an easy integration with standard CMOS processes and microelectronic circuits, the improvements in the reliability of the integrated optical source represents a fundamental step for the future development of this technology. The aim of this review paper is therefore to describe the device-specific degradation modes and mechanisms determining the long-term reliability of laser diodes for silicon photonics communication applications. In particular, our analysis focuses on heterogeneous III-V MQW lasers on SOI, representing current state-of-the-art integrated lasers, and on quantum-dot laser epitaxially grown on silicon, which may become the preferential light source for future photonic devices and systems operating in the 1.31 μm -1.55 μm optical windows.

Reliability of Heterogeneous III-V Lasers on SOI Substrates
The III-V epitaxies adopted for the fabrication of heterogeneous lasers on SOI substrates do not differ from the ones adopted for conventional laser diodes and therefore, in principle, should exhibit similar degradation modes and mechanisms. During the early stages of development of this novel technology, however, this assumption was not valid, since the additional processing steps required to bond the III-V and the SOI wafers were Considering that Si-based PICs offer (i) a very high degree of integration, due to the high refractive index contrast between Si and the surrounding dielectric, (ii) the possibility to integrate all the optical signal conditioning functions required by telecom applications, and (iii) an easy integration with standard CMOS processes and microelectronic circuits, the improvements in the reliability of the integrated optical source represents a fundamental step for the future development of this technology. The aim of this review paper is therefore to describe the device-specific degradation modes and mechanisms determining the long-term reliability of laser diodes for silicon photonics communication applications. In particular, our analysis focuses on heterogeneous III-V MQW lasers on SOI, representing current state-of-the-art integrated lasers, and on quantum-dot laser epitaxially grown on silicon, which may become the preferential light source for future photonic devices and systems operating in the 1.31 µm-1.55 µm optical windows.

Reliability of Heterogeneous III-V Lasers on SOI Substrates
The III-V epitaxies adopted for the fabrication of heterogeneous lasers on SOI substrates do not differ from the ones adopted for conventional laser diodes and therefore, in principle, should exhibit similar degradation modes and mechanisms. During the early stages of development of this novel technology, however, this assumption was not valid, since the additional processing steps required to bond the III-V and the SOI wafers were found to cause the formation of thermally-induced defects within the III-V material in correspondence of the bonding interface, as shown in Figure 3 [20].
Electronics 2021, 10, x FOR PEER REVIEW 4 found to cause the formation of thermally-induced defects within the III-V materi correspondence of the bonding interface, as shown in Error! Reference source not fo [20]. Despite these defects not being associated with the formation of extended de propagating through the epitaxial layers toward the active region, they were identifie be limiting both device performance and reliability [20,21]. To reduce residual la strain and avoid the eventual propagation of point defects during device operation, per-lattice (SL) structure was inserted in the proximity of the bonding interface [22]. addition of a strained super-lattice effectively increased the performance of the dev measured after the bonding procedure and increased yield from 60% to 99% [21]. In a tion, the authors found that devices based on epitaxies featuring an SL layer exhibit nificantly lower premature failure rates during aging performed at temperature and rent density levels up to 85 °C and ~7 kA/cm 2, respectively. Conversely, this optimiza did not seem to impact on the degradation rate of non-prematurely failed devices, p bly indicating that their lifetime was mainly determined by other degradation proce This latter aspect was investigated more in-depth in [15], where the authors evaluated long-term reliability of 1.55 μm distributed Bragg reflector (DBR) heterogeneous l diodes by means of a set of constant-current aging experiments carried out at current sities up to 8.6 kA/cm 2 , at a temperature of 45 °C. In these stress conditions, the heter neous laser diodes exhibited a peculiar set of degradation processes, affecting both optical and electrical characteristics of the devices. In particular, the LDs under inves tion showed a gradual and stress-current dependent increase in threshold current was found to be correlated with the decrease in both turn-on voltage, as in [23], and dependent ideality factor, estimated at low forward voltages. These three processes w found to be consistent with the increase in defect concentration within the active re of the device. Conversely, no significant lowering in the slope-efficiency (SE) of the vices was found to affect their optical characteristics. Considering that the slope effici of a laser diode can be expressed as [24]: Despite these defects not being associated with the formation of extended defects propagating through the epitaxial layers toward the active region, they were identified to be limiting both device performance and reliability [20,21]. To reduce residual lattice strain and avoid the eventual propagation of point defects during device operation, a super-lattice (SL) structure was inserted in the proximity of the bonding interface [22]. The addition of a strained super-lattice effectively increased the performance of the devices measured after the bonding procedure and increased yield from 60% to 99% [21]. In addition, the authors found that devices based on epitaxies featuring an SL layer exhibit significantly lower premature failure rates during aging performed at temperature and current density levels up to 85 • C and~7 kA/cm 2, respectively. Conversely, this optimization did not seem to impact on the degradation rate of non-prematurely failed devices, possibly indicating that their lifetime was mainly determined by other degradation processes. This latter aspect was investigated more in-depth in [15], where the authors evaluated the long-term reliability of 1.55 µm distributed Bragg reflector (DBR) heterogeneous laserdiodes by means of a set of constant-current aging experiments carried out at current densities up to 8.6 kA/cm 2 , at a temperature of 45 • C. In these stress conditions, the heterogeneous laser diodes exhibited a peculiar set of degradation processes, affecting both the optical and electrical characteristics of the devices. In particular, the LDs under investigation showed a gradual and stress-current dependent increase in threshold current that was found to be correlated with the decrease in both turn-on voltage, as in [23], and biasdependent ideality factor, estimated at low forward voltages. These three processes were found to be consistent with the increase in defect concentration within the active region of the device. Conversely, no significant lowering in the slope-efficiency (SE) of the devices was found to affect their optical characteristics. Considering that the slope efficiency of a laser diode can be expressed as [24]: where η inj is the carrier injection efficiency, and α m and α i are respectively the mirror and the modal losses, this finding indicated that the stress procedure did not impact either the injection efficiency or the mirror losses of the laser. This latter point actually represents a peculiar strength of heterogeneous laser diodes, in terms of reliability. In fact, if a proper optical mode transition between the III-V material and the silicon waveguide is achieved [25][26][27], the effective facets of the laser can be moved into the optical waveguide and decoupled from the active semiconductor region. This eliminates the problem of facet damage as one of the major sources of device degradation [21], and improves the robustness against catastrophic and gradual mirror degradation [28]. This is important because it enables lower cost, non-hermetic packaging. The central role of the epitaxy, rather than mirrors, in heterogeneous devices degradation was further demonstrated in [29], where the authors showed loop-mirror LDs featuring the same III-V structure of the lasers analyzed in [15] exhibited similar degradation modes.
By analyzing the variation of the capacitance-voltage (C-V) characteristics during stress, and of the derived apparent charge profile (Figure 4), the authors were able to ascribe the observed gradual increase in threshold current to the relocation of charged defects, or impurities, acting as non-radiative recombination centers (NRRCs). The diffusing species was found to originate from the p-side of the device and to diffuse toward the active region, corresponding in Figure 4 to lower apparent depths (x d ).
where is the carrier injection efficiency, and and are respectively the m and the modal losses, this finding indicated that the stress procedure did not impact the injection efficiency or the mirror losses of the laser. This latter point actually repr a peculiar strength of heterogeneous laser diodes, in terms of reliability. In fact, if a p optical mode transition between the III-V material and the silicon waveguide is ach [25][26][27], the effective facets of the laser can be moved into the optical waveguide an coupled from the active semiconductor region. This eliminates the problem of facet age as one of the major sources of device degradation [21], and improves the robu against catastrophic and gradual mirror degradation [28]. This is important because ables lower cost, non-hermetic packaging. The central role of the epitaxy, rather tha rors, in heterogeneous devices degradation was further demonstrated in [29], whe authors showed loop-mirror LDs featuring the same III-V structure of the lasers ana in [15] exhibited similar degradation modes.
By analyzing the variation of the capacitance-voltage (C-V) characteristics d stress, and of the derived apparent charge profile (Error! Reference source not fo the authors were able to ascribe the observed gradual increase in threshold current relocation of charged defects, or impurities, acting as non-radiative recombination c (NRRCs). The diffusing species was found to originate from the p-side of the devic to diffuse toward the active region, corresponding in Error! Reference source not f to lower apparent depths (xd). In addition to that, by considering that the amount of the observed free charge ation was in the order of 10 17 cm −3 and that only dopants are intentionally introdu similar concentrations within the device, the authors postulated that the diffusion p-dopant Zn may be driving the observed degradation process. Doping-related dif is not something uniquely associated with heterogeneous lasers, since it is inheren lated to the III-V epitaxy. This type of degradation can affect both III-N [30] and [31] emitters, and represents the reason why manufacturers of optoelectronic device to transition to less diffusive p-dopants, like carbon, for large scale manufacturing [3 Generally speaking, the correlation between optical degradation and the generation  In addition to that, by considering that the amount of the observed free charge variation was in the order of 10 17 cm −3 and that only dopants are intentionally introduced in similar concentrations within the device, the authors postulated that the diffusion of the p-dopant Zn may be driving the observed degradation process. Doping-related diffusion is not something uniquely associated with heterogeneous lasers, since it is inherently related to the III-V epitaxy. This type of degradation can affect both III-N [30] and III-As [31] emitters, and represents the reason why manufacturers of optoelectronic devices tend to transition to less diffusive p-dopants, like carbon, for large scale manufacturing [32,33]. Generally speaking, the correlation between optical degradation and the generation/propagation of point defects or impurities suggests that even for modern state-of-the-art devices, this kind of degradation process represents a limiting factor for their long-term reliability.
Heterogeneous laser sources have been commercialized for datacenter products with current production rates of several million transceivers/year [34]. The reliability of these devices is superb in spite of the fact that the transceivers are not hermetically packaged. Figure 5 shows 80 • C aging with negligible aging over 19,000 h [34]. This illustrates the advantage of silicon facets without III-V material at the facet. agation of point defects or impurities suggests that even for modern state-of-the-ar vices, this kind of degradation process represents a limiting factor for their long-term liability.
Heterogeneous laser sources have been commercialized for datacenter products current production rates of several million transceivers/year [34]. The reliability of t devices is superb in spite of the fact that the transceivers are not hermetically packa Error! Reference source not found. shows 80 °C aging with negligible aging over 19 h [34]. This illustrates the advantage of silicon facets without III-V material at the fac

Reliability of InAs Quantum-Dot Laser Epitaxially Grown on Silicon
From a reliability standpoint, the advantages offered by a QD-based active lay heteroepitaxially-grown devices were first demonstrated in [35], where an average time is more than 10 times higher than conventional QW lasers grown on the same G on-silicon was measured. To understand the physical origin of this dramatic impr ment, we must consider that in the presence of a high concentration of extended def as in the case of heteroepitaxial III-As layers on Si [36], non-radiative trap-assisted rec bination processes are favored. The energy released by these processes can induce a increase in lattice temperature, thus favoring the generation of additional defects and moting other non-radiative recombination events [37]. Therefore, the magnitude of recombination-enhanced (RE) defects-generation process drastically increases in the p imity of extended defects, like dislocation sites. This mechanism was also confirme preliminary material-level analyses carried out by means of planar-view scanning T (PVSTEM), which revealed that misfit dislocations (MDs) located in the proximity o active layer acquired a helical component after 600 h RT aging at 2.5 kA/cm 2 . Since th a characteristic behavior of dislocation growth by climb processes [38], this finding cates that the aging procedure induced the re-growth of pre-existent extended def This process was further confirmed by more recent investigations carried out on InAs LDs grown on Si with a lower TDD [39], through which the authors proved that m dislocations in device layers submitted to high levels of carrier injection during aging dergo RE-climb/growth: this effect is demonstrated by the evident protrusions affec the MD in the aged device, compared to the unaged one, as reported in Error! Refer

Reliability of InAs Quantum-Dot Laser Epitaxially Grown on Silicon
From a reliability standpoint, the advantages offered by a QD-based active layer to heteroepitaxially-grown devices were first demonstrated in [35], where an average lifetime is more than 10 times higher than conventional QW lasers grown on the same GaAs-onsilicon was measured. To understand the physical origin of this dramatic improvement, we must consider that in the presence of a high concentration of extended defects, as in the case of heteroepitaxial III-As layers on Si [36], non-radiative trap-assisted recombination processes are favored. The energy released by these processes can induce a local increase in lattice temperature, thus favoring the generation of additional defects and promoting other non-radiative recombination events [37]. Therefore, the magnitude of this recombinationenhanced (RE) defects-generation process drastically increases in the proximity of extended defects, like dislocation sites. This mechanism was also confirmed by preliminary materiallevel analyses carried out by means of planar-view scanning TEM (PVSTEM), which revealed that misfit dislocations (MDs) located in the proximity of the active layer acquired a helical component after 600 h RT aging at 2.5 kA/cm 2 . Since this is a characteristic behavior of dislocation growth by climb processes [38], this finding indicates that the aging procedure induced the re-growth of pre-existent extended defects. This process was further confirmed by more recent investigations carried out on InAs QD LDs grown on Si with a lower TDD [39], through which the authors proved that misfit dislocations in device layers submitted to high levels of carrier injection during aging undergo RE-climb/growth: this effect is demonstrated by the evident protrusions affecting the MD in the aged device, compared to the unaged one, as reported in Figure 6. Since this process is proportional to the number of carriers that are locally captured by the defective sites, the authors explained the huge difference in degradation rate between QD-and QW-based devices by considering that QD structures offer a 3D confinement that spatially localizes injected carrier, thus preventing their capture by nearby extended defective sites (Error! Reference source not found.). On the other hand, carriers that are not injected into, or that escape from the QDs, are free to laterally diffuse, though with some limitations [40], and promote REDR processes in correspondence of dislocation sites.  Since this process is proportional to the number of carriers that are locally captured by the defective sites, the authors explained the huge difference in degradation rate between QD-and QW-based devices by considering that QD structures offer a 3D confinement that spatially localizes injected carrier, thus preventing their capture by nearby extended defective sites (Figure 7). On the other hand, carriers that are not injected into, or that escape from the QDs, are free to laterally diffuse, though with some limitations [40], and promote REDR processes in correspondence of dislocation sites. Since this process is proportional to the number of carriers that are locally captured by the defective sites, the authors explained the huge difference in degradation rate between QD-and QW-based devices by considering that QD structures offer a 3D confinement that spatially localizes injected carrier, thus preventing their capture by nearby extended defective sites (Error! Reference source not found.). On the other hand, carriers that are not injected into, or that escape from the QDs, are free to laterally diffuse, though with some limitations [40], and promote REDR processes in correspondence of dislocation sites.  Despite the partial immunity of QD-based LDs to the growth of climb dislocation networks that would usually be seen in comparable QW lasers, it has been demonstrated from a statistical point of view that the density of dislocations still represents a limiting factor for the long-term reliability of InAs QD LDs grown on Si [35]. In particular, a decrease in the TDD of an order of magnitude can increase the device lifetime of a factor approximately equal to 1000, as evidenced by the plots reported in Figure 8.
Despite the partial immunity of QD-based LDs to the growth of climb d networks that would usually be seen in comparable QW lasers, it has been dem from a statistical point of view that the density of dislocations still represents a factor for the long-term reliability of InAs QD LDs grown on Si [35]. In particu crease in the TDD of an order of magnitude can increase the device lifetime o approximately equal to 1000, as evidenced by the plots reported in Error! R source not found.. The above mentioned findings suggest that lifetime improvements of QD l taxially grown on silicon should likely proceed through the reduction in the conc of extended defects, mainly misfit dislocations [39], or through the optimizati carrier injection dynamics into the InAs QDs [40]. The former goal can either be through the optimization of the growth procedure, of the buffer layer or throug engineered positioning of MD-trapping layers that prevent the formation of l nally extended defects in the proximity of the active region, which may happe the heating-up and cooling-down phases of the growth process [42,43]. The intr of misfit trapping layers, which block over 90% of the misfit dislocations whi otherwise grow along the active region [43,44], allows for a great increase in dev bility, as testified by the higher optical stability exhibited by devices featuring layers within their epitaxy (Error! Reference source not found.9). This misfit gr pears to be the dominant failure mechanism for QD lasers grown on silicon sub The above mentioned findings suggest that lifetime improvements of QD lasers epitaxially grown on silicon should likely proceed through the reduction in the concentration of extended defects, mainly misfit dislocations [39], or through the optimization of the carrier injection dynamics into the InAs QDs [40]. The former goal can either be achieved through the optimization of the growth procedure, of the buffer layer or through a well-engineered positioning of MD-trapping layers that prevent the formation of longitudinally extended defects in the proximity of the active region, which may happen during the heating-up and cooling-down phases of the growth process [42,43]. The introduction of misfit trapping layers, which block over 90% of the misfit dislocations which would otherwise grow along the active region [43,44], allows for a great increase in device reliability, as testified by the higher optical stability exhibited by devices featuring trapping layers within their epitaxy ( Figure 9). This misfit growth appears to be the dominant failure mechanism for QD lasers grown on silicon substrates.
On the other hand, a more efficient injection and stimulated recombination of carriers within the QDs can be achieved by adopting a modulation-doping scheme [45], rather than building a conventional p-i-n active structure. This doping approach can favor device performance by easing hole injection, increasing the high-temperature and dynamic performance of QD LDs, and favoring, under appropriate design rules, GS over ES operation [46]. In addition to that, p-modulation doping can improve device reliability, since the faster holes transfer from the modulation doped layers to the GS of the QDs effectively reduces the amount of free carriers available for defect-assisted non-radiative recombination events and degradation processes [35,41]. Efficient confinement of carriers, and associated device reliability, were also found to exhibit a strong dependence on the operating regime, as evidenced by the experimental results reported in [47]. On the other hand, a more efficient injection and stimulated recombination of within the QDs can be achieved by adopting a modulation-doping scheme [45] than building a conventional p-i-n active structure. This doping approach can favo performance by easing hole injection, increasing the high-temperature and dynam formance of QD LDs, and favoring, under appropriate design rules, GS over ES op [46]. In addition to that, p-modulation doping can improve device reliability, si faster holes transfer from the modulation doped layers to the GS of the QDs eff reduces the amount of free carriers available for defect-assisted non-radiative reco tion events and degradation processes [35,41].
Efficient confinement of carriers, and associated device reliability, were also f exhibit a strong dependence on the operating regime, as evidenced by the exper results reported in [47].
By means of a current step-stress experiment, carried out at T = 35 °C and at densities up to 10 kA/cm 2 , the authors were able to ascribe the optical degradatio QD LDs on Si under test to a defect-related generation process, whose rate was f increase when the devices are driven at current levels above the onset of ES emis this bias regime, corresponding to current levels above 220 mA (Error! Reference not found.), injected carriers tend to occupy QD states at higher energy levels, an fore see an effective escape barrier which is lower compared to the case of GS-on ation, where most of the carrier occupies the GS of the first ES of the QDs. As a quence, the rate of escape and of REDR processes described in [35] and by Error ence source not found. increase, thus increasing the degradation rate. This is a denced by the increase in the slope of the Ith-1/SE curve reported in the right plot o Reference source not found. for stress time roughly corresponding to the onse By means of a current step-stress experiment, carried out at T = 35 • C and at current densities up to 10 kA/cm 2 , the authors were able to ascribe the optical degradation of the QD LDs on Si under test to a defect-related generation process, whose rate was found to increase when the devices are driven at current levels above the onset of ES emission. In this bias regime, corresponding to current levels above 220 mA (Figure 10), injected carriers tend to occupy QD states at higher energy levels, and therefore see an effective escape barrier which is lower compared to the case of GS-only operation, where most of the carrier occupies the GS of the first ES of the QDs. As a consequence, the rate of escape and of REDR processes described in [35] and by Figure 7 increase, thus increasing the degradation rate. This is also evidenced by the increase in the slope of the I th -1/SE curve reported in the right plot of Figure 10 for stress time roughly corresponding to the onset of ES emission. The straight linear correlation between the threshold current increase and the increase in 1/SE also indicated that the degradation process was lowering the injection efficiency, measured under nominal operating conditions. This kind of variation was found to be compatible with the increase in SRH recombination rate within device layers close to the active region: defect-assisted recombination occurring within these layers prevents carriers from being potentially injected into the GS of the QDs, therefore lowering the apparent injection efficiency of the laser. The effect of non-radiative recombination centers placed within the InGaAs well on the optical performance was also confirmed by numerical simulations, performed on models based on similar device structures [48].
Electronics 2021, 10, x FOR PEER REVIEW 10 of 13 centers placed within the InGaAs well on the optical performance was also confirmed by numerical simulations, performed on models based on similar device structures [48]. Further constant-current aging experiments [49,50] confirmed the role of the generation/propagation of non-radiative recombination centers in the gradual optical degradation of QD LDs on Si, evidencing both a strong dependence of this process on stress current density and showing that the observed optical degradation kinetics are compatible with a diffusion process. While this latter finding still requires additional investigations for the determination of the physical origin of the related mechanism, the strong powerlaw dependence of the degradation rate on stress current density, testified by a mean timeto-failure (MTTF) ∝ . , is compatible with a recombination-driven degradation, since the number of escaping carriers is directly related to the rate of injected carriers during stress. The improvements in epitaxial quality and the previously described mitigation strategies allowed to attain devices with very-high MTTF. In particular, if we define this latter parameter as the average time required for a sample operating at a constant 10 mW OP to increase by two times its initial threshold current, QD LDs with lifetimes in excess of 22 years under high temperature aging (T = 80 °C) have recently been demonstrated [44].

Conclusions
With this paper, we have reviewed the most relevant scientific results concerning the degradation mechanisms and the reliability of SWIR laser diodes for silicon photonics applications, with a focus on heterogeneous III-V/Si lasers and on InAs quantum-dot lasers epitaxially grown on silicon.
The few available literature reports dealing with the reliability of heterogeneous IR laser diodes indicate that such devices, nowadays, only suffer from minor reliability issues related to the wafer bonding procedure required to couple the III-V IR source to the rest of the photonic circuit contained within the SOI substrate. Additionally, the decoupling of the laser facets from the active semiconductor region increases the robustness against mirror damage and COMD. Conversely, experimental results indicate that these lasers can still be affected by gradual degradation processes involving the generation and/or Further constant-current aging experiments [49,50] confirmed the role of the generation/propagation of non-radiative recombination centers in the gradual optical degradation of QD LDs on Si, evidencing both a strong dependence of this process on stress current density and showing that the observed optical degradation kinetics are compatible with a diffusion process. While this latter finding still requires additional investigations for the determination of the physical origin of the related mechanism, the strong power-law dependence of the degradation rate on stress current density, testified by a mean time-tofailure (MTTF) ∝ J −3.9 , is compatible with a recombination-driven degradation, since the number of escaping carriers is directly related to the rate of injected carriers during stress.
The improvements in epitaxial quality and the previously described mitigation strategies allowed to attain devices with very-high MTTF. In particular, if we define this latter parameter as the average time required for a sample operating at a constant 10 mW OP to increase by two times its initial threshold current, QD LDs with lifetimes in excess of 22 years under high temperature aging (T = 80 • C) have recently been demonstrated [44].

Conclusions
With this paper, we have reviewed the most relevant scientific results concerning the degradation mechanisms and the reliability of SWIR laser diodes for silicon photonics applications, with a focus on heterogeneous III-V/Si lasers and on InAs quantum-dot lasers epitaxially grown on silicon.
The few available literature reports dealing with the reliability of heterogeneous IR laser diodes indicate that such devices, nowadays, only suffer from minor reliability issues related to the wafer bonding procedure required to couple the III-V IR source to the rest of the photonic circuit contained within the SOI substrate. Additionally, the decoupling of the laser facets from the active semiconductor region increases the robustness against mirror damage and COMD. Conversely, experimental results indicate that these lasers can still be affected by gradual degradation processes involving the generation and/or propagation of impurities and non-radiative recombination centers within the III-V epitaxy. This consideration ultimately shows that point defects may represent a major reliability issue for modern MQW emitters, as far as their long-term reliability is concerned, and suggests that a tighter control of defects, as well as the transition to a less diffusive p-type dopant may be necessary steps for the improvement of the lifetime of these novel sources.
On the other hand, statistical analyses, material-oriented investigation and accurate evaluation of the gradual degradation processes related to QD LDs epitaxially grown on Si evidence that the improvement in the long-term reliability of such devices should proceed mainly through the reduction in dislocation density, mainly misfit-type dislocations, present in correspondence of the active region. The reliability of QD-based lasers is also strongly affected by structural design choices, such as the p-doping approach, that impact on the carrier injection and recombination dynamics within the DWELL region. The optimization of both of these aspects, along with the addition of misfit trapping layers, may allow QD lasers to replace conventional QW-based devices for future silicon photonics applications, and to attain a fully-epitaxial integration of the laser source into the PIC.