Challenges for Sige Bicmos in Advanced-Node SOI

Abstract

D-band (110-170GHz) spectrum is gaining attention for various applications, including 6G mm-Wave, sub-THz sensing, and radar. These systems require lattice spacing for antenna elements at sub-1mm and a very low loss signal path from antenna to integrated chip. A highly efficient front-end in a very small form factor will be required for these systems. This drives the requirement for a monolithically integrated high-gain, high-efficiency front-end that also leverages the benefits of a high-speed / high-density digital CMOS. Silicon germanium (SiGe) heterojunction bipolar transistors (HBT) integrated along with a high-density CMOS provide such an all-silicon monolithic solution.The US government is fostering the expansion of “unique and differentiated domestic manufacturing” with funding through DARPA’s Technologies for Mixed-mode Ultra Scaled Integrated Circuits (T-MUSIC) [1] program to enable disruptive RF mixed-mode technologies by developing high performance RF analog integrated with advanced digital CMOS. Through the T-MUSIC program, DARPA seeks to: 1) advance RF and mixed-mode devices to support ultra-wideband RF frontends from HF to 100 GHz; 2) integrate those devices with high density digital CMOS electronics at the wafer scale to enable embedded digital intelligence; 3) develop and explore ultra-high resolution broadband mixed-mode circuit building blocks for DoD-relevant applications; 4) explore innovative device topologies and materials to form THz devices in an advanced digital CMOS fabrication platform; and 5) establish a domestic ecosystem that facilitates enduring DoD access to differentiated capabilities for high performance RF mixed-mode SoCs.Under T-MUSIC, GlobalFoundries is demonstrating BiCMOS on 45nm PDSOI, which is the focus of this paper, and 22nm FDSOI CMOS with goals of increasing HBT performance of fT/fMAX from 350/500 GHz to 400/600 GHz and 600/700 GHz. HBTs with fT/fMAX of 380/550GHz GHz have been demonstrated building upon previously published results [2]. This paper will touch on some of the challenges that were encountered in achieving that result and discuss those anticipated in future work.Achieving these results required scaling transistor dimensions. Vertical scaling of the emitter, base and collector layers, with higher doping concentrations, reduces transit time but results in higher current densities and higher electric fields. Lateral scaling of the transistor structure reduces parasitic capacitance and resistance but concentrate the power dissipation in a smaller area. The thermal conductivity of silicon is 148W/m-K whereas that of silicon dioxide is ~1.4W/m-K. Even a thin layer of oxide will significantly increase the self-heating of the HBT. Therefore, we replace the SOI with coplanar epitaxy in regions where the HBTs are formed.The vertical scaling of the HBT requires limiting the thermal cycles that the HBT will experience during processing and suggests forming the HBT as late as possible in the CMOS process. However, the thermal cycles associate with the epitaxy and film depositions to form the HBT impact the CMOS transistors which suggests forming the HBT early in the process. We found a point in the process that offers the best compromise minimizing the impact to the doped-channel PDSOI CMOS while achieving the HBT performance goals. Work is just beginning on integration tradeoffs for FDSOI with metal gate and high-K dielectrics.Advanced-node CMOS processes can form components having smaller dimensions which offers advantages for lateral scaling but also presents challenges for forming the HBT. The contact height in 45nm is significantly less than the height of the HBT structure used in GF’s 9HP process. We changed the formation of the emitter and base so that the emitter and base contacts are almost coplanar in contrast to 9HP where the emitter was almost twice the height of the base. This problem is being further exasperated as we migrate to 22nm.The shrinking of BEOL wiring dimensions, along with the ability of the HBT to drive high currents, presents challenges in designing within limits imposed by electromigration. The use of wider wires is constrained by metal density rules. The use of stacked metal levels and redundant vias impact the parasitic capacitances and resistances of the interconnects.The paper and presentation will review these, and other challenges encountered in achieving BiCMOS integration of SiGe HBTs with fT/fMAX of 380/550GHz GHz [see figure] on a 45nm PDSOI CMOS and touch future work.This research was developed with funding from the Defense Advanced Research Projects Agency (DARPA) and is Approved for Public Release, Distribution Unlimited. The views, opinions and/or findings expressed are those of the author and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government.[1] https://www.darpa.mil/attachments/T-MUSIC_Proposers%20Day_Presentations_Combined.pdf [2] J. Pekarik et al., 2021 IEEE BCICTS, 2021, pp. 1-4, Figure 1