268 Design of CMOS Millimeter-Wave and Terahertz Integrated Circuits
12.3.1 Architecture and System Specification
The de sign of a CMOS heterodyne receiver in THz has to be conducted in a
scenario without any low noise amplifiers (LNA) as illustrated in Figure 12.2,
because hardly any amplifiers can be designed at a frequency close to or above
the f
max
. For example, the f
max
in a typical CMOS 65nm process is around
300 GHz [26 5],
The receiver gain (G
tot
) in such c ase can be calculated as
G
tot
= G
ant
G
mix
G
pga
(12.4)
where G
ant
, G
mix
and G
pga
denote the gain of antenna, mixer and power gain
amplifier (PGA), respectively. Also, the total rece iver noise figure (NF
tot
)
becomes
NF
tot
= NF
mix
+
NF
pga
− 1
G
mix
(12.5)
where NF
mix
and NF
pga
denote the NF of the mixer and VGA, respectively.
Eq. (12.4) denotes that the total receiver gain can be improved from antenna,
mixer and PGA, while (12.5) deno tes that the noise contributed by each sta ge
decreases as the total gain of preceding stages. The noise contributions from
mixer and PGA are no longer negligible witho ut LNA, so the NF of the
receiver has to be improved by increas ing the conversion gain of the mixer
and minimizing the noise fig ure of PGA. In the following sections, the designs
of the mixer and PGA in a THz CMOS heterodyne receiver are introduced.
12.3.2 Down-Conversion Mixer
As the first active building block connected to the antenna, the design of the
down-conversion mixer largely affects the performance of heterodyne receiver,
including conversio n gain and NF. Co nventionally there are two types of mix-
ers that are commonly used in the mm-wave region: Gilbert-cell mixer and
single-ga te mixer [266, 267]. Gilbert-cell mixer [266] has a compact size and
low implementation loss , and it generates the cross-modulation product of LO
and RF signals. However, the co nversion gain of the Gilbert-cell mixer largely
depe nds on the transconductance of transistors in the saturation re gion, which
will be heavily reduced when the signal freque nc y is approaching f
max
. On the
other hand, the single-gate mixer [267] utilized the nonlinea rity of transistors
when biased in the subthreshold region, which is less frequency dependent
compared to the Gilbert-cell mixer, and is able to work at a higher frequency
in THz. Moreover, compared to the subharmonic mixer working with 1/3-LO
[234], the conversion loss could be larg ely reduced when directly mixing the
RF and LO signal in fundamental tones. In this work, a single-gate mixe r is
designed to down-convert the RF signal in 220 ∼ 300 GHz to the baseband
by the fundamental tone of LO signal.
Figure 12.8 shows the schematic of a proposed down-conversion mixer
design. A Wilkinson combiner implemented by coplanar waveguide (CPW)