Low-Power 60-GHz CMOS Radios for Miniature Wireless Sensor Network Applications - Ultra-Wideband and 60 GHz Communications for Biomedical Applications - page 195

difference in voltage magnitudes at these two taps as frequency changes, a monotonic
curve passing through zero will be traced out, with the zero-crossing occurring at the
resonant frequency. Figure 13 b shows three monotonic curves plotting the difference
in magnitude measured at these two taps. The solid line is derived from the lossless
transmission line model, assuming perfect matching over the frequency band. The
dashed line is obtained from Ansoft HFSS full-Electro-Magnetic (EM) simulations
of the three-port antenna reference setup in Fig. 13 c, showing the discrepancy due to
the limited bandwidth of the patch antenna. Measured results from a 60-GHz patch
antenna fabricated in 130-nm CMOS are shown as dots in Fig. 13 b, which follow the
same trend as the simulation. Figure 13 c also shows the S-parameter analysis. S21
and S31 represent the ratio of power delivered onto Z A and Z B , respectively, from the
antenna feed point at Port 1. The two tap nodes are designed to be high-impedance
nodes, so that they do not load the patch antenna significantly, but provide enough
voltage swing for envelope detector circuits below the patch to measure the standing
wave magnitude. Simulation results show that only 2 % of the power delivered onto
the patch antenna is lost on the two tap nodes.
The locations of the two taps along the edge of the antenna depend on several
factors. A wider separation between the two taps provides a larger slope of the
monotonic curve, thus a larger controller gain in the FLL and more accurate resonant
frequency detection. However, the frequency locking range over which the sensed
voltage magnitude difference remains monotonic is inversely proportional to the
separation between taps. Therefore, if the separation is too wide for a target frequency
range, the monotonic characteristic no longer applies, which could result in instability
in the FLL. Furthermore, the two envelope detectors would be placed far away from
each other on the chip, causing mismatch from separation within the single IC [ 34 ],
and thus affecting the accuracy of tracking the resonant frequency. In the final layout
of the antenna reference, the two taps are equally spaced, 110

m away from the
length center, resulting in a calculated locking range of 8.5 GHz. Another factor that
will affect the locking range of the antenna reference is the antenna bandwidth. A
larger bandwidth implies a wider locking range. In this case, the antenna bandwidth
is around 1 GHz, which provides sufficient range for covering process variation of
the patch antennas.
In order to test the antenna resonant frequency and bandwidth, the patch antenna
design with the same dimensions as the one in the FLL was fabricated in 130-
nm CMOS. Twenty of the patch antennas were tested, all from the same wafer.
Figure 14 shows the distribution on measured resonant frequency and bandwidth over
20 replica patch antennas. The mean and standard deviation of the center frequency
is 59.7 GHz and 65.1 MHz, respectively. The mean and standard deviation of the
bandwidth is 1.03 GHz and 67.9 MHz, respectively. This results in a 3 σ variation
in center frequency of 3,270 ppm, which is enough to ensure the transmitter is FCC
compliant (i.e., it remains in the 60-GHz ISM band). In order for the transmitter to
reliably communicate with, e.g., an energy-detection receiver with an identical patch
antenna (not included in this work), process variation should not cause the transmitter
frequency to fall outside the bandwidth of the receiving antenna. Assuming a worst-
case 3 σ variation on all parameters from process variation, the 3,270 ppm variation is

μ