CMOS mm-Wave Transceiver Techniques beyond 50 GHz - IEEE Xplore

3 downloads 47583 Views 1018KB Size Report
1Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 2University of ... technologies and the high levels of process variation further.
CMOS mm-Wave Transceiver Techniques beyond 50 GHz Adrian Tang12 & Nacer Chahat1 1

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 2University of California, Los Angeles

Abstract —Deep sub-micron scaled CMOS offers relatively low RF performance at mm-wave frequencies compared with III-V technologies and the high levels of process variation further exacerbate design margins and lesser performance. Instead of optimizing mm-wave circuits for performance at design time, we instead adopt a “maximum tune-ability” approach in which the mm-wave front-end circuitry is made highly tunable and optimized at run-time by digital control and optimization algorithms to overcome process variation effects.

process variation. The design was fabricated on two different wafer runs spaced 6 months apart and 10 chips were selected from each run for a total of 20 chips. Using a setup similar to the one discussed in [3], the center frequency and output power of each chip was measured at the same supply and bias conditions for each chip.

Index Terms — millimeter-wave calibration, mm-wave optimization

I. INTRODUCTION Recent advances in silicon technology have enabled the possibility of constructing CMOS based Gb/s communication systems beyond the 50 GHz frequency range. Design of RF front-ends at mm-wave carriers (100 GHz) and beyond is a relatively new topic with most emphasis placed on the circuit design aspects. One major challenge of mm-wave front-ends operating at frequencies beyond 100 GHz is that the stage gain for power amplifiers is quite low [1,2,3], typically only 3-6 dB of gain per stage. This low available gain makes the front-end extremely sensitive to the effects of process variation as a small change in the CMOS device parameters may drastically reduce the front-end performance. To illustrate this we first constructed the simple circuit shown in Fig. 1, where a 140 GHz oscillator is coupled directly to a power amplifier, also optimally centered at 140 GHz (D-Band).

Fig. 2. Measured center frequency of 20 D-band oscillator-PA chips from two different fabrication runs.

Figure 2 and 3 plot the output frequency and power measured from the 20 chips respectively. As the frequency spread is quite large (135.9 GHz < f