Page 55 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 7 – Terahertz communications

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 7

technologies at 22-nm and 28-nm might also be in the matching network. On the other hand, we
considered for the benefit of digital integration but consider a power target and investigate the
typically do not offer substantially different required size of the transistor. A larger transistor
performance in UmmW bands than 65-nm or CMOS introduces design challenges to distribute the RF
SOI processes and the process parameters are power into and out of the transistor. A large device,
generic. The roadmap of RF-optimized InP HBT relative to the wavelength, typically incurs a drop in
processes has been discussed in [11] and [12]. the potential fmax. At 140 GHz, a device
Current 250-nm InP HBT processes are capable of width/length of more than 200 um would pose
fmax exceeding 600 GHz while being relatively significant conditions to distribute the signal.
mature with commercial applications. Scaling to
130-nm and beyond can yield fmax exceeding 1 THz.
The evolution of SiGe HBTs has also produced
remarkable fmax increases that have reached
similar speeds to InP [13]. SiGe BiCMOS processes
have been optimized for digital and RF
performance [14]. Current processes offer several
differentiated HBTs in a single process optimized
for breakdown and fT/fmax. CMOS SOI processes
based on partially depleted SOI substrates have
evolved from a digital process to RF-optimized
approaches with high-resistivity substrates and RF
back end-of-the-line [15]. CMOS and CMOS SOI offer
similar supply and knee voltage. The 40-nm GaN
HEMT process is described in [16] and offers a
400-GHz fmax. The characteristics of the different Fig. 4 - Comparison of process technology trade-offs under
processes is summarized in Table 1 with an fixed resistance (50 Ohm) and fixed power (20 dBm)
emphasis on commercially available processes with conditions.
the highest fmax. First, to minimize the loss of the impedance

Table 1 - Comparison of UMMW semiconductor device matching to the load line, we might choose the
technologies device geometry (i.e. width) to provide a 50 Ohm

Technology fmax VSUP VK IMAX PRF RLL load line. The top plot in Fig. 4 indicates the output
power that is developed by each device technology.
(GHz) (V) (V) (A/mm) (W/mm) (mm)
For instance, the InP process will produce 15 dBm
InP HBT 600 2.5 0.7 3 1.4 1.2 while the SiGe process will produce roughly 6 dBm.
The GaN HEMT would deliver around 30 dBm
SiGe HBT 450 1.3 0.5 2.2 0.44 0.7
output power. For a 20-dBm target power outlined
CMOS 310 1.1 0.3 1 0.2 1.6 in Section 2, the InP HBT and GaN HEMT are closest
GaN 400 12 2 1.6 4.0 12.5 to the target for a 50-Ohm match.
HEMT
Table 2 - Theoretical efficiency bounds for 20 dBm output
Based on the supply voltage and the knee voltage, power at 140 GHz
the maximum output power can be calculated from InP SiGe CMOS GaN
1 Technology
P RF = (V DD − V )I MAX while the load-line HBT HBT FET HEMT
K
4
resistance is R LL = 2(V DD − V )/I MAX . The RF PAE (Q = 1000) 45% 34% 32% 43%
K
output power and load-line resistance is shown in PAE (Q = 10) 39% 24% 23% 34%
Table 1 when normalized to 1 mm. The high supply
voltage of the GaN HEMT due to breakdown Conduction Angle 203° 222° 260° 232°
characteristics suggests high power density and
load-line resistance relative to the other processes. Second, we might also compare the technologies for
a fixed output power such as 20 dBm in Fig. 4. The
We compare these technologies in two different device presenting a load-line matching condition
ways to understand device selection for high closest to 50 Ohms is the most desirable from the
efficiency. On one hand, we would choose a device standpoint of PAE. Notably, the InP HBT is the best
that offers a load line close to 50 Ohm to avoid loss choice as GaN HEMT presents a large load-line

© International Telecommunication Union, 2021 43