(Invited) Integration of InP Heterojunction Bipolar Transistors on Silicon Substrates for 6G Networks

The fast expansion of the global datasphere creates a huge demand on the future wireless communication technologies. Although 5G, the 5 th generation of the cellular network technology, is still being rolled out, the potential of 6G is already intensively explored. The vision of 6G implies not only massive connectivity and higher data rate (up to 100 Gb/s and more) in ultra-reliable low-latency communication, but machine-to-machine interaction will dominate over those involving humans. AI (artificial intelligence) and machine learning will be omnipresent. 6G network will address a frequency spectrum beyond 100 GHz, which provides wider bandwidths and, therefore, potentially simpler modulation schemes and eventually lower power consumption. However, a clear disadvantage of these mm-waves or sub-THz frequencies are the atmospheric absorption and blockage effect by physical obstacles or rain. These challenges must be tackled by a sufficiently high transceiver device density per area together with advanced beam steering and AI empowered networks. The power amplifier (PA) is one of the most critical components of the RF front-end module. Here, InP based heterojunction bipolar transistors (HBT) clearly outperform PAs based on CMOS or SiGe BiCMOS concerning saturated output power, energy efficiency and breakdown voltage. However, the limiting downside of InP HBTs is the missing cost-efficient and scalable fabrication, which hampers its adoption in high-volume 6G applications. In this paper we review various InP HBT fabrication concepts with the aim of lowering the production costs and maintaining high device performance. The best HBT device performance is demonstrated for the lattice-matched deposition on native small-diameter (≤ 150 mm) InP substrates due to perfect III-V crystal quality. Although device fabrication on these wafers cannot profit from the advanced tool park available in 200/300 mm fabs, the main drawbacks are the high cost and brittleness of InP substrates. Realizing InP HBTs on Si substrates allows to benefit from their robustness, low cost, and the access to a mature device fabrication on large-diameter wafers. In wafer reconstitution, numerous diced InP wafer tiles, which already contain the grown HBT layer stack, are bonded to a Si carrier to populate the wafer surface. After planarization, the reconstructed Si wafer can be further fabricated in a CMOS compatible 200/300 mm fab. In transfer printing a comparable approach is followed while only the active device stack is transferred to the Si carrier. Transferring only a thin InP layer to the Si carrier, which serves as a starting surface for the HBT deposition, leads to a so-called engineered substrate. Compared to the previous transfer methods, the III-V device deposition is carried out after bonding. The key advantage of all these heterogenous integration concepts is the access to a CMOS compatible and scalable device fabrication while maintaining the high III-V crystal quality. Nevertheless, the HBT production costs can only be significantly reduced with a certain re-usability of the InP substrates. The monolithic deposition directly on Si substrates would be the most cost-efficient integration approach. Unfortunately, the mismatch in lattice constant and thermal expansion coefficient between Si and III-V materials leads to the formation of relaxation defects, which degrade the HBT device performance. The growth of thick strain relaxation buffers allows to reduce the defect density in the device layers but bears new challenges such as wafer warpage or III-V crack formation. Nano-ridge engineering (NRE) is based on selective area growth (SAG) applying trench-patterned Si wafers. The III-V deposition starts inside very narrow trenches to achieve efficient aspect ratio trapping (ART) of misfit defects. After the trench is filled with III-V material, the growth is continued to engineer a broad and pristine nano-ridge (NR) which holds the device stack. A second oxide pattern prevents sidewall deposition around the NRs during the deposition of the HBT heterolayers. In the first part of this paper, pros, and cons of the most common InP HBT integration scenarios as summarized in figure will be compared in detail. Among all concepts, NRE holds the potential of being the most cost-efficient integration approach due to SAG in line with very limited III-V material deposition. However, the key open question is what NR crystal quality, and hence, device performance can be realized. This question will be addressed in the second part of the paper, reporting about the latest developments at imec. Figure 1