Reliability of InAs quantum dot lasers grown on silicon: Physics, Limits and Solutions
Direct crystal growth approaches for III-V/silicon integration are of great technological interest for photonic integrated circuit applications. However, historically, these approaches have been hindered due to the generation of large numbers of interfacial dislocations that form to mediate materials property mismatches between GaAs/InP based films and the silicon substrate. The presence of dislocations at this interface, far from critical device regions, is not inherently problematic. But their ends, termed threading dislocations, rise upward through all the subsequent device layers, facilitating non-radiative recombination losses where they cross the light-emitting ‘active’ region. Further, the energy released in these defect-assisted processes facilitates ongoing dislocation growth during laser operation. As a result, these runaway ‘recombination enhanced dislocation climb’ and ‘recombination enhanced dislocation glide’ (REDC and REDG, respectively) processes culminate in device failure. Thus, laser reliability engineers have traditionally focused on efforts to reduce threading dislocation densities and to develop dislocation tolerant infrared emitters. However, even with decades of research seeking to reduce threading dislocation densities and to develop more dislocation tolerant gain materials such as p-modulation doped InAs quantum dots (QD), device lifetimes at industrially relevant temperatures (near 80°C) remained too short by orders of magnitude.Seeking to better understand the mechanism by which even relatively low densities of threading dislocations led to failure in InAs QD lasers, my colleagues and I identified that the threading dislocations give rise to in plane misfit dislocations lying at the QD layers. This however was puzzling: these layers are designed to be too thin for misfit dislocations to form due to lattice mismatch during growth. We therefore proposed that these specific misfit dislocations formed instead during post-growth cooldown due to (1) thermal expansion mismatch between the (Al)GaAs layers and the silicon and (2) local mechanical hardening effects in the QD layers. To test this hypothesis, we then used metallurgical principles to intentionally widen the mechanically hardened region and thus displace misfit dislocation formation to less critical regions of the device. This was very successful. Placing a single, thin, alloy-hardened ‘misfit dislocation trapping layer’ approximately 100-200 nm to either side of the quantum dot active region reduces the total QD adjacent misfit dislocation length by >10x and trapped misfit dislocations show no evidence of recombination enhanced dislocation climb (REDC) after more than 1000 hours of continuous operation at 60°C. Furthermore, lasers with misfit dislocation trapping layers operating 80°C demonstrate median extrapolated lifetimes significantly greater than 100,000 hours—meeting the requirements for data center operation. Thus, these results simultaneously emphasize both the importance of moving in earnest towards the advanced photonic integration schemes and the importance of continuing fundamental materials studies on III-V semiconducting systems.